Jay David Atlas

Logic, Meaning,
and Conversation:
Semantical
Underdeterminacy,
Implicature, and Their
Interface
JAY DAVID ATLAS
OXFORD UNIVERSITY PRESS

Logic, Meaning, and Conversation
This page intentionally left blank
Logic, Meaning,
and Conversation
Semantical Underdeterminacy,
Implicature, and Their Interface
JAY DAVID ATLAS
1
2005
3
Oxford New York
Auckland Bangkok Buenos Aires Cape Town Chennai
Dar es Salaam Delhi Hong Kong Istanbul Karachi Kolkata
Kuala Lumpur Madrid Melbourne Mexico City Mumbai Nairobi
São Paulo Shanghai Taipei Tokyo Toronto
Copyright © 2005 by Jay David Atlas

Published by Oxford University Press, Inc.
198 Madison Avenue, New York, New York 10016
www.oup.com
Oxford is a registered trademark of Oxford University Press
All rights reserved. No part of this publication may be reproduced,
stored in a retrieval system, or transmitted, in any form or by any means,
electronic, mechanical, photocopying, recording, or otherwise,
without the prior permission of Oxford University Press.
Library of Congress Cataloging-in-Publication Data
Atlas, Jay David.
Logic, meaning, and conversation : semantical underdeterminacy,
implicature, and their interface.
p. cm.
Includes bibliographical references and index.
ISBN-13 978-0-19-513300-4
ISBN 0-19-513300-5
1. Pragmatics. 2. Semantics (Philosophy) I. Title.
B831.5 .A75 2000

121'.68—dc21 99-030797
9 8 7 6 5 4 3 2 1
Printed in the United States of America
on acid-free paper
To
Morton G. White, Paul Benacerraf, Rogers Albritton,
philosophers,
to
Stephen C. Levinson, Frans Zwarts, Laurence R. Horn, Jerrold Sadock,
linguists,
to
Sean D. P. Fennessy, John Robert Purvis,
and to
John Francis Walter
PREFACE
T he flight to Frankfurt on the Lufthansa 747-400 passed slowly, as August flights

from the summer afternoon sun of Los Angeles into the European darkness tend to do.
The seat was comfortable enough, even for my six-foot one-inch frame, the cuisine
was German bourgeois respectable, and I managed to sleep. The charm of the trip was
awakening to a large, healthy, hot breakfast, but only after the stretching exercises
were completed. With attention to detail, hygiene, and the physical culture of northern
Germany, some Mephisto at the airline had mandated a wake-up video that taught
passengers a way to stretch their chair-encased muscles while they remained in their
seats. It was a show and practice exercise video, done while strapped in one’s seat, that
loosened and stretched the feet, ankles, calves, upper legs, arms, and neck. So, at the
command of the celluloid instructor, I rolled, waggled, extended, raised, and kneaded
my body back to life, as did most of my fellow passengers, except for some unruly
nonconforming Americans who decided to remain cramped and caged in their seat-
formed postures. By the end of the video, I felt awake and physically alive. “Damn the
Germans,” I thought; “Such a good idea!” I enjoyed the breakfast.
Transferring in Frankfurt for a short flight to Brussels, I then took a train to Leuven
in Belgium, a university city where I was to lecture for a week in August 1990 at the
Second European Summer School on Language, Logic, and Information. It was to be
an advanced course on “Implicature and Logical Form: The Semantics-Pragmatics
Interface” for graduate students in philosophy, logic, linguistics, and computer science
from universities in the European Union. Some six hundred graduate students had as-
sembled for two weeks, 30 July–10 August 1990, at the Katholieke Universiteit Leuven
for this intellectual feast of a summer school, and I was on the menu. “I’m an hors
d’oeuvre,” I thought, and I expected about six students to show up at the lecture hall on
Monday morning for the week’s ten hours of course lectures that I had written. I had
even bought my first laptop computer in May, a Sharp 8088 laptop with two 720K
disk drives, in order to compose the lectures, realizing that without electronic help in
editing I would never write, type, revise, and retype the body of lectures in time.
viii PREFACE
The commuter train was packed with students, bureaucrats, grandmothers, and
me. I barely found a seat and a place to put my small backpack containing my pre-
cious lectures and my bulgy, denier nylon, soft luggage. To call it “luggage” would
be infelicitous; it was a sack with cloth handles, but wonderfully light and commo-
dious, just awkward to stow, and I had bought it on a whim at Crate and Barrel where
my cousin Pamela had been working between her usual six-month walk-abouts: eight
countries in Africa, or five provinces of China, or sixteen blocks of Santa Monica. I
couldn’t really decide, in my jet-lagged state, whether the luggage was properly a
nylon crate or a nylon barrel. Fatigue made me more and more sodden as we crept
away from Brussels on little steel wheels. Then, intruding into this linguistical mus-
ing, came the slightly accented, youthful voice of a university student speaking En-
glish to me. “Are you going to Leuven?” he asked. I looked up into the cheerful,
handsome Flemish face of a student, who sat across the aisle from me, next to his
comely girlfriend. “Yes,” I admitted. “There is a summer school, and I shall be lec-
turing.” “Where are you staying?” he asked, and I told him the name of the hotel.
“It’s not far from the station. I will show you the way.” I thought, he even uses the
first-person ‘will’ form correctly; hardly any of my American students could, but
putting this Henry Higginsish thought aside, I accepted his aid, thanked him for his
kindness, and marveled at his manners.
I registered at the hotel, examined my first room, asked to be moved to the back
of the hotel where there would be no street noise, and enjoyed a hot shower—those
splendid German showerheads again—and a short nap before lunch. At lunch I en-
countered Frans Zwarts of Rijksuniversiteit Groningen, The Netherlands, and his
American wife, Sharon Parry. I had spent five days talking with Zwarts at a week’s
conference in Stuttgart a year before, the last afternoon of which put me and Hans
Kamp on the platform together. My usual luck: playing the other bookend to some-
one like Hans Kamp. I had also had fascinating talks with Jaap Hoepelman,
Hoepelman’s and Zwarts’s doctoral student Peter Blok, and Sjaak de Mey. The Dutch
semanticists, I discovered, had read my published work, particularly my (1988) ar-
ticle on negative existence statements, which showed that the Russellian problem of
the relationship between meaning and ontology had been fundamentally miscon-
ceived. Until then I knew only three people who had liked, been convinced by, or
even read, the essay: Mark Richard; the Waynefleet Professor of Metaphysics in
Oxford, Sir Peter Strawson; and the Most Famous Syntactician in the world at MIT,
who had written me that “It is a convincing piece, and does indeed cut through a
hoary tradition” (Noam Chomsky, personal communication, 12 February 1989).
Everybody else had ignored it as far as I knew, until I met the Dutch semanticists.
But at the end of the Stuttgart conference, Zwarts had asked whether I would
like to give a course in Leuven the following year at the European summer school.
After I asked whether he thought more than five students would show up, I said “Sure.”
Why not? Perhaps it was time to summarize what I thought was right and what was
wrong with Paul Grice’s (1989a) theory of conversational inference, a model of talk
as rational activity in which Grice tries to explain the grounds for the addressee’s
inference from what (he believes) a speaker asserts to what (he believes) the speaker
“implies, suggests, or conveys” by, in, or when making the assertion. Then I’d have
to try to explain my own (Atlas 1974, 1975a,b, 1977b, 1978a,b, 1979) wildly popu-
PREFACE ix
lar view that a radical, nonspecific semantics would be required by any theory of
pragmatic inference, if the theories of the pragmatic module and the semantic mod-
ule of the mind/brain’s language faculty were to provide a consistent, descriptively
adequate and explanatory account of utterance-interpretation (see Lakoff 1977; K.
Bach 1987; Kempson 1988a; Horn 1989; Iten 1998: 63, 67; and Levinson 1997, 2000).
So this book began life in the lecture halls of the Katholieke Universiteit Leuven,
Belgium, during the Second European Summer School in Language, Logic, and In-
formation in August 1990. It turned out that more than one hundred graduate stu-
dents from universities throughout the European Union and a dozen or so members
of the summer school faculty—even once George Bealer—showed up for my lec-
tures. More copies of my course book of readings were sold to the graduate students
for my course than for any other course in the summer school. (These people are
obviously nuts, I thought.)
Out of my Leuven lectures, and my lecture in Stuttgart in 1989, the invitation
for which I am much indebted to Jaap Hoepelman, grew three doctoral dissertations;
those of Ana von Klopp (Edinburgh, 1993) on negation, Peter Blok (Groningen, 1993)
on focus, and Michiel Leezenberg (Amsterdam, 1995) on metaphor. Von Klopp began
her dissertation with the piquant remark, “At the 1990 European Summer School in
Language, Logic, and Information, Jay Atlas told me about the dangers of wasting
one’s youth on negation. I was foolish enough not to listen, and this is the result.”
After the last lecture in the series, Richard Oehrle suggested that the lectures
might serve some intellectual and pedagogical purpose if published, and my then-
editor Angela Blackburn at Oxford University Press, U.K., agreed. Angela wanted
my lectures right then, August 1990. After all, sitting at lunch in Oxford, she had the
notebook containing the typed lectures in her hand. Well, I said, perhaps a little pol-
ishing would be appropriate, as usual barely controlling my desire to rush into print,
contribute papers to the annual meetings of the APA, shower the journals with paper,
and generally festschrift it up.
Eleven years later I have finished this little, chatty, nontechnical book, written
with the easy accessibility that I have made my trademark, supplemented with much
new material but inspired by my 1990 Leuven lectures on the semantics-pragmatics
interface: the relationship between literal meaning, logical form, and interpretative
inference. After remonstrances from John Francis Walter, Stephen Levinson, and
Thomas James Rankin, I did not throw the 1997 version of the manuscript off a cliff;
actually, I was at the Max Planck Institute for Psycho-linguistics in Nijmegen, The
Netherlands, at the time (fall 1997), and there wasn’t a cliff within two hundred
kilometers.
What I realized in Groningen during my visiting research professorship in 1995,
and it was reinforced when I finally read Alberto Coffa’s fascinating 1991 book The
Semantic Tradition from Kant to Carnap, was my commitment to a semantic tradi-
tion originating in the views of the mathematicians Bolzano, Dedekind, Frege, and
Hilbert and in my tutelage at Amherst College by Robert Breusch, himself Zermelo’s
assistant in the editing of Cantor’s collected papers, the semantic tradition that sur-
vived in the Vienna Circle among Schlick, Waismann, and the early Wittgenstein.
My admiration for the writings and teaching of Nelson Goodman, Morton White,
Noam Chomsky, Donald Davidson, Sir Peter Strawson, Jonathan Cohen, Jerrold Katz,
x PREFACE
Paul Benacerraf, Arthur Prior, Hao Wang, and Dana Scott, my formative schoolboy
reading of Morton White, Ernest Nagel, Stephen Toulmin, and Arthur Danto and
Sidney Morgenbesser’s anthology in the philosophy of science during my last year
at Phillips Exeter Academy, and my study of the problems and rigorous arguments
of the semanticists of the fourteenth century, a century that was as catastrophic for
Europe as our own twentieth century (Tuchman 1978), resulted in an antipathy to
behaviorist and epistemologically motivated views about meaning.
These views are adulterated by post-positivist epistemology and have caused
much philosophical misunderstanding of language—for example, see C. Peacocke
(1992). I rejected neo-Kantianism and various forms of verificationism in the theo-
ries of meaning of Neurath, Reichenbach, Carnap, the middle Wittgenstein, and even
those of my teacher Sir Michael Dummett, who provided so much of the philosophi-
cal stimulus to my thinking about negation and presupposition. Their views have
resulted in confusions about meaning that are almost impossible to remedy in our
current intellectual climate. But there are now signs of a reconsideration taking place:
see Chomsky (1995b, 1996b,c) and J. A. Fodor (1998); for his re-thinking of his own
views, see Michael Dummett’s splendid work (1993: 157–61); for discussion of the
role that Dummett’s views can play in actual psycholinguistic theorizing about the
acquisition of language, see Atlas (2001).
This book presents an account of the interface between literal meaning and in-
terpretative inference that purports to be more “descriptively and explanatorily ade-
quate” than Grice’s William James lectures of 1967, but it is not a neo-Kantian essay
on the conceptual possibility of any future pragmatics of language. The linguistic
data are explicable if we hypothesize that the literal meanings of sentence-types are
quite different from either truth conditions or assertibility conditions of sentence-
tokens and that idealized interpreters conform to, and perhaps employ, certain prag-
matic principles of inference. But my hypotheses are not considered by me to be
necessary principles of any possible use of language or constitutive of the rationality
of language use (or any other bits of the neo-Kantian, verificationist, or later
Wittgensteinian framing of questions about language as a practical ability that even
Paul Grice was tempted by). What I do claim to have shown, by actually construct-
ing one rather than as a conclusion of a transcendental argument, is that a theory of
interpretative inference that “saves the phenomena” and a theory of literal meaning
that “saves the phenomena” will be congruent in the ways that I describe in this book:
if there is an interface between the semantic and the pragmatic, between Chomsky’s
Internalist Semantics and the Performance System, it has the character described
herein (see Atlas 1978b, 1979, 1989).
Two applications of my theory of the semantics-pragmatics interface are given
in chapter 5, in my account of the semantics and pragmatics of comparative adjec-
tives and adverbial approximatives, and in chapter 6, in my account of numerical
adjectives. This work was begun at the Institute for Advanced Study, Princeton, New
Jersey, in 1983. Morton White and my colleague Robert Sleigh were wonderfully
supportive and tolerant of this logical inquiry and of my youthful obsessions. The
institute is my idea of heaven. Living in a Marcel Breuer–designed apartment, abut-
ted by the woods of a wildlife sanctuary with trails for solitary or companionable
ramblings, an office to work in that had once been occupied by Sir Isaiah Berlin,
PREFACE xi
surrounded by the most intelligent conversation in the world, and when one had
thought enough about logic, language, and philosophy, conversing about quantum
gravity or string theory with Andrew Strominger and seeing off-off-Broadway plays
with John Walter—that is intellectual heaven.
My first inklings of my ideas on the semantics-pragmatics interface—that is, the
relationship between semantical underdeterminacy (nonspecificity) and interpreta-
tive inference—germinated during the summer of 1973 at the Mathematical and Social
Sciences Board Workshop on the Formal Pragmatics of Natural Language in the
University of Michigan, Ann Arbor, where, at the invitation of George Lakoff and
Lauri Karttunen, I collaborated intensively with Stephen Levinson on thinking through
Grice’s theory of conversational implicature, and I had the epiphany that ‘not’ cre-
ated not ambiguous but, as it turned out, semantically nonspecific sentences (see Atlas
1974, 1975a,b, 1977b, 1978a,b, 1979, 1989). In the spring of 1973 at the University
of Texas, Austin, Conference on Performatives, Presupposition, and Implicature,
Lakoff had introduced me to Stephen Levinson, then a graduate student. That was
the beginning of my intellectual collaboration with Levinson on the pragmatics of
language, a collaboration that has now spanned thirty years; it has been a splendid
intellectual adventure enriched by a warm personal friendship with him, his wife
Penny Brown, and their son Nicholas. As my late father, Jacob Henry Atlas, once
remarked to me, in the persona of Atlas-Brown, Inc., Houston and Fort Worth, Texas,
after dining with a group of my Princeton graduate student friends, “You have such
wonderful friends. How do they put up with you?” How my father put up with me is
an even more unanswerable question; I had only one conversation with him about
my choice of vocation as logician and philosopher and student of language, during
the Christmas vacation of my first year at Amherst College. He: “Have you thought
about what you’d like to do as a vocation?” The seventeen-year-old me: “I’m enjoy-
ing physics, mathematics, and philosophy so much, I think I might continue to study
one of them.” He: “It’s just like you to choose the least-paying profession.” He never
said another word about it, but he paid all the bills for it. My first book was dedicated
to him, and he was able to hold a copy in his hands four years before he died.
The penultimate version of the penultimate version of this book was written at
the Max Planck Institute for Psycholinguistics, Nijmegen, The Netherlands, in the
fall term 1997, after I had spent time in Groningen thinking about the De Morgan
properties of adverbial verb phrases (Atlas 1998) brought to my attention by H. Klein
(1997, 1998), Frans Zwarts, and Sjaak de Mey. The latter inquiry is a logical inves-
tigation that William of Shyreswood, Walter Burleigh, and Peter of Spain, those
worthies of the fourteenth century, would understand the point of.
The penultimate version was written in the spring of 1999 and the final version
was written in the summer of 2001 after the appearance of Levinson’s (2000) Pre-
sumptive Meanings: The Theory of Generalized Conversational Implicature. Stephen,
like Larry Horn in his Natural History of Negation (1989), provided essential stimu-
lus and matter for reflection as I reconsidered neo-Gricean views that in concert with,
and in reaction to, them I had been developing since the mid-1970s. My admiration
for their work is surpassed only by their cosmic patience with my criticizing, not to
say needling, them.
ACKNOWLEDGMENTS
P ortions of this book were written at the Institute for Advanced Study, Princeton,
New Jersey, and I am deeply indebted to Morton G. White and to the faculty of the
School of Historical Studies.
Portions of this book have been delivered as invited lectures or have in earlier
versions appeared in print. I am grateful to the following for permission to reprint
revised material from previously published work: the editors of Linguistics and Phi-
losophy; the editor of Journal of Semantics; D. Reidel Publishing Company; Aca-
demic Press; B.H. Blackwell, Ltd.; Cambridge University Press, and Oxford
University Press.
I am grateful to audiences in Princeton University; University of California, Los
Angeles; University of California, Riverside; University of California, Irvine; Uni-
versity of California, Santa Barbara; University of Southern California; University
College, London; Cambridge University; University of Edinburgh; University of
Salford, Manchester, U.K.; University of Amsterdam; University of Groningen;
University of Utrecht; University of Antwerp; University of Leuven; Centre de
Récherche en Épistemologie Appliquée, École Polytechnique, Paris; and the Max
Planck Institute for Psycholingistics, Nijmegen.
For visiting professorships I am grateful to the faculty in the department of phi-
losophy, University of California, Los Angeles, where I gave graduate seminars in
the philosophy of language in the fall terms of 1991, 1994, and 1995. I am especially
indebted to David Kaplan, Keith Donnellan, Tyler Burge, and Andrew Hsu. I owe a
special debt to the late Rogers Albritton with whom for over twenty-five years, week
in and week out, month in and month out, I discussed metaphysics, epistemology,
philosophy of mind, philosophy of language, and Wittgenstein. These conversations,
and similar ones with Edmund Gettier many years ago, constitute an entire philo-
sophical education and reeducation.
xiv ACKNOWLEDGMENTS
I gave the course “Implicature and Logical Form: The Semantics/ Pragmatics
Interface,” on which this book is based, at the invitation of the organizing committee
and Frans Zwarts, University of Groningen, The Netherlands, for the Second Euro-
pean Summer School on Language, Logic, and Information at the University of
Leuven, Belgium, in August 1990. I began work on this book at the suggestion of
Richard Oehrle, made during a conversation over coffee after my last lecture.
I then had the pleasure of being second promotor to Zwarts on the Groningen
doctoral dissertation of Peter Blok in 1993, followed by enjoying the hospitality of
the University of Groningen’s department of Dutch linguistics and its Institute for
Behavioural, Cognitive, and Neuro-Sciences as a visiting scholar in the spring term
1995. I am grateful to Frans, Hoeksema, de Mey, and Victor Sánchez Valencia, Sharon
Parry, and Simone Zwarts. I am grateful for support from the Nederlandse Organisastie
voor Wetenschappelijk (NWO; the Netherlands Organization for Scientific Research)
and the Institute for Behavioural, Cognitive, and Neuro-Sciences in the University
of Groningen, The Netherlands.
I am particularly grateful for the hospitality of the Max Planck Institute for
Psycho-linguistics, Nijmegen, The Netherlands, and of its then-managing director
Pim Levelt, for providing such inestimable facilities during the writing of this book.
Pomona College, Claremont, has provided travel assistance and other research
support, as well as sabbatical leaves; my thanks to the faculty research committee,
especially for support from a National Endowment for the Humanities sabbatical
grant, to former Associate Dean Fred Greiman, and to Jane Arnal. The President
Emeritus of Pomona College, Peter Stanley, was wonderfully supportive of intellec-
tual life at the college.
During this lengthy project I have been aided by Rogers Albritton, Kent Bach,
Paul Benacerraf, Brandon Birdwell, Diane Blakemore, Peter Blok, Peter Bosch, David
Braun, Penny Brown, Tyler Burge, Noel Burton-Roberts, Robyn Carston, Noam
Chomsky, Peter Cook, Robin Cooper, Mikael Dolfe, Elisabet Engdahl, Sean Diarmuid
Peter Fennessy, Robert Fogelin, Jack Hoeksema, Jaap Hoepelman, Larry Horn, Yan
Huang, Andrew Hsu, the late Jerrold Katz, Ruth Kempson, Henny Klein, Wim
Klooster, Ana von Klopp, Eric Krabbe, Joshua Krist, George Lakoff, Michiel
Leezenberg, Nicholas Levinson, Stephen Levinson, Jason Linder, Brian Maddox,
A. P. Martinich, the late James McCawley, Sjaak de Mey, Jan Nuyts, the late John
Robert Purvis, Thomas Rankin, François Récanati, Mark Richard, David Rosenthal,
Peter Ross, Jerry Sadock, Ivan Sag, the late Victor Sánchez-Valencia, Rob van der
Sandt, Joshua Seifert, Pieter Seuren, Scott Soames, Keith Stenning, Sir Peter Strawson,
Henriette de Swart, Jamie Tappenden, Frank Veltman, Henk Verkuyl, John Francis
Walter, Ton van der Wouden, and Charles Young.
My colleague René Coppieters at Pomona College has been a rich source of
criticism, appreciation, and support; few have been vouchsafed such a well-tempered
and incisive, friendly critic as René has been.
Michiel Leezenberg (University of Amsterdam) has been a source of intellec-
tual stimulation, friendship, and assistance on many occasions.
I am, as always, deeply grateful to the playwright John Francis Walter for his
friendship, his continuing interest in my intellectual work for more than twenty years,
and his intense intellectual and artistic energy.
ACKNOWLEDGMENTS xv
I continue to draw on what I learned many years ago from Dana Scott and Paul
Benacerraf, Donald Davidson, Michael Dummett, Edmund Gettier, the late Joseph
Epstein, the late Robert Breusch, and Morton G. White, and from the late Rogers
Albritton in our conversations.
I am indebted to the Atlas family, especially to my cousins Daniel and
Marlene Bergman; to my Aunt Rose Atlas Bergman Weiss and Sol Weiss; to
Morris Atlas, LL.B, and Rita Atlas; to Joe Atlas, M.D.; to William A. Atlas, M.D.,
and Marnie Atlas, M.D.; to Scott Atlas, LL.B., and the Honorable Nancy Atlas,
LL.B.; to Robert Atlas, Ph.D., Pamela Atlas, and John Atlas; and to Leon T. Atlas,
M.D., who saved my life. Someday I hope to write a book that more of them will
be interested in reading.
The manuscript was typed by me using PC WRITE LITE, Version 1.01, programmed
by Bob Wallace, and LQMATRIX, a font design program by J. David Sapir, and, as Jay
Rosenberg said in a similar context, I’m everlastingly grateful to me for doing it. I
am also grateful to these brilliant programmers for making old-fashioned DOS pro-
grams such effective instruments for writing (see J. D. Atlas, “Do It in DOS,” PC
LapTop Computers Magazine, November 1994, volume 6, number 11: 40–45).
My editors Peter Ohlin and Catharine Carlin at Oxford University Press, New
York, have been wonderfully patient and supportive; I am grateful to them and to
the production staff for dealing gracefully with a complex manuscript.
CONTENTS
1 Semantical Underdeterminacy 3
2 Grice’s Theory of Conversational Inference: A Critical Exposition 45
3 The Rise of Neo-Gricean Pragmatics 80
4 The Post-Gricean Theory of Presupposition 118
5 Assertibility Conditions, Implicature, and

the Question of Semantic Holism: Almost but Not Quite 149
6 The Third Linguistic Turn and the Inscrutability of Literal Sense 185
Appendix 1 On G. E. Moore’s Term ‘Imply’ 225
Appendix 2 On Hitzeman (1992) on ‘Almost’ 231
Appendix 3 The Semantics and Pragmatics of Cleft Sentences 234
Appendix 4 A Note on Notation 248
Bibliography 253
Index 273
Conversation is like playing tennis with a ball made of
Krazy Putty, that keeps coming back over the net in a
different shape. . . . The same axiom, every decoding is
another encoding, applies. . . . In ordinary spoken
discourse the endless cycle of encoding-decoding-
encoding may be terminated by an action, as when for
instance I say, “The door is open” and you say, “Do you
mean you would like me to shut it?” and I say, “If you
don’t mind,” and you shut the door, we may be satisfied
that at a certain level my meaning has been understood.
David Lodge, The Practice of Wriring
We cannot assume that statements (let alone sentences)

have truth-conditions. At most they can have something
more complex: ‘truth indications’ in some sense. . . .
There is no reference-based semantics. There is a rich and
intriguing internalist semantics, really part of syntax, on a
par in this respect with phonology. Both systems provide
‘instructions’ for performance systems, which use them
. . . for articulation, interpretation, inquiry, expression of
thought, and various forms of human interaction.
Noam Chomsky, “Language and Nature”
The linguistic turn was, I think, an uncompleted revolu-

tion; to really turn from theories of knowledge to theories
of meaning, you would have to stop construing content in
epistemological terms. Many analytic philosophers can’t
bear not to construe content in epistemological terms
because they think of philosophy as conceptual analysis,
and of conceptual analysis as displaying a concept’s
possession conditions, and of possession conditions as
characteristically epistemic. If, as I believe, that whole
picture is wrong, a certain kind of analytic philosophy is
ripe for going out of business.
Jerry Fodor, In Critical Condition
Logic, Meaning, and Conversation
1
Semantical Underdeterminacy
1 Metaphor, nonspecific meaning, and utterance

interpretation: Two dogmas of literary modernism
Twentieth-century studies of literary style and early, influential studies in philoso-

phy of language have been conditioned in large part by a dogma.1 It is a belief in
1The first version of this chapter was written in 1975 in response to a question posed to me by Mark
Allen Phillips (aka Kontos), publicly presented in a faculty research lecture series in December 1977
at Pomona College, Claremont, but heretofore unpublished. Its inspiration was an essay by William
Gass (1970), and it is offered to him as a modest gesture of appreciation. I am also indebted to the
Educational Foundation of America for its sponsorship of this research through a Pomona College,
Claremont, research award in 1975–76 to me and Mark Allen Phillips. The anti-Fregeanism in it bears
a family resemblance to Hilary Putnam’s (1975) views on natural kind terms, except Putnam preserves
the thesis that meaning determines reference while rejecting that meaning (including the determina-
tion of reference) is “in the head.” In February 1978 Donald Davidson (1984a: 245–64) read his “What
Metaphors Mean” at the University of Chicago, and in the Hilary Term 1978 I heard him read a ver-
sion in a lecture in the University of London. I was struck by some of the similarities in our views of
metaphor, which I had never heard him discuss in his seminars at Princeton, but I was unsurprised that
I had absorbed from his teaching an approach that brought me close to his position. Yet I discovered
that the position that I had taken in 1975 and continue to defend here was semantically more radical
than his (see my Philosophy without Ambiguity [1989]), and my semantic problem of metaphor, unlike
his, was formulated as a choice between “abstraction” and “homonymy”/“ambiguity” explanations, in
the fashion to be found later in G. Lakoff’s (1977) “Linguistic Gestalts” and in G. Lakoff and M.
Johnson’s (1980: 106–14) Metaphors We Live By. Again there is a superficial resemblance between
my 1975 view and a position taken by Sperber and Wilson (1986a) in their “Loose Talk,” but the prob-
lem for Sperber and Wilson is their notion of “looseness,” not to speak of their explanation of meta-
3
4 LOGIC , MEANING , AND CONVERSATION
some fundamental cleavage between language that is figurative or literary, contain-

ing terms used metaphorically, and language that is standard or ordinary, containing
terms used literally. This dogma, I shall argue, is ill founded. One effect of abandon-
ing it is a revision of Fregean semantics, a departure not only from the view that
meaning determines reference but from the view that literal meaning is determinate.
A second effect is the disappearance of the supposed boundary between linguistic
art and linguistic life.
1.1 The first dogma

The American philosopher and novelist William Gass’s (1970: 60) thesis is that “fic-
tion is life in terms of . . . ,” that fiction “is incurably figurative, and the world the nov-
elist makes is always a metaphorical model of our own.” For Gass, art, metaphor, and
imagination are all linked. To adapt a figure of W. H. Auden’s, Gass should describe
works of literary art like racehorses: Art is out of Metaphor by Imagination. For Gass,
imagination is the writer’s talent for constructing metaphors, including lengthy ones
called ‘novels’. Since, in Gass’s view, metaphorical language is literary language, we
are back to the original Auden figure: Verse is out of Language by Poet.
Gass would claim that Clifford is a mouse and My life is an “Omensetter’s Luck”
are statements of the same logical type, both metaphors, both presentations of imagi-
native transfigurations at the touch of a word. Anyone wishing to understand the
modernist view of the relationship between art and life must understand an early-
twentieth-century view of metaphor.
To take a recent example of the view, James Wood, in reviewing Tom Wolfe’s
novel A Man in Full, contrasts Wolfe’s treatment of character with Charles Dickens’s,
in these words:
Wolfe’s prose always prefers the most ordinary, the most vulgar word. His descrip-
tions are always the most ordinary details, without any capacity for simile or meta-
phor (which is one of the absolute definitions for the literary). But Dickens finds the
unexpected detail, the vivid simile. Think of Joe Gargery in Great Expectations, “With
eyes of such a very undecided blue that they seemed to have somehow got mixed
with their own whites.” Or, in David Copperfield, Dora’s cousin “in the Life-Guards,
with such long legs that he looked like the afternoon shadow of somebody else.” Or
Uriah Heep in the same novel, his mouth “open like a post-office.” Or Mr. Trabb,
who “had sliced his hot roll into three feather beds, and was slipping butter in be-
tween the blankets, and covering it up.” The delight of such wit has little to do, at
phorical utterance. For interesting comment on my view, see Noel Burton-Roberts (1991: 169), who
had read a samizdat copy. Since the essay elaborates the theme of this book in a rather different form,
and uses the concept of semantical nonspecificity in analyzing a part of language that many humanists
and literary critics find interesting, I have included it here as a foil to Grice’s brief comments on the
subject of metaphor. There has been an extraordinary amount of work on metaphor in the last twenty
years—for example, Bergmann (1982), Fogelin (1988), Glucksberg (2001), Johnson (1981), Kittay
(1987), Leezenberg (1995, 2001), Ortony (1993), Sacks (1981), Searle (1979), Stern (1983, 1985, 1991,
2000), and Sperber and Wilson (1986b). None of it makes the point that I want to make here and made
in 1975. For discussion of my view, see Leezenberg (2001: 211–13).
SEMANTICAL UNDERDETERMINACY 5
times, with accuracy; a mouth never really looks like a post-office. The joy, the liter-
ary joy, is in the local fizz of each detail, and in the relation of each detail to the other,
and then in the moral revelation that such similes provide. (Uriah Heep is like a post-
office, that is, he is everyone’s willing courier.) (Wood 1988: 40)
It is notable how, in literary rhetoric, the terms ‘ordinary’ and ‘vulgar’ and the words
‘simile’, ‘metaphor’, and ‘literary’ juxtapose.
Paul Horgan tells the following story:
My neighbor’s very small boy, not quite four years old, came charging across my
garden where I was working on a very hot summer morning. He was pursuing an
imaginary enemy. He wore only a cowboy hat and the briefest of under-trunks, and
he carried a toy shotgun. Suddenly, on becoming aware of me, he was abashed by
his near-nakedness and his imaginary game. He paused in his chase and said an-
grily to me, “I am really a United States Marshal, but sometimes I go around like
this.” I nodded seriously, and, reassured, he ran on. (Horgan 1974: 94)
An imaginary enemy is a nonexistent enemy, which, as Gilbert Ryle (1949) would

have remarked, is not a special kind of enemy; it is an enemy only “in” the child’s
imagination. The child’s activity is an “imaginary game,” in Horgan’s words, but if
that means ‘game played in the imagination’, it is a false description of the activity.
The boy’s was a real game, but a game “out of” the imagination, with pretend vil-
lains and mock beliefs. We might also say that the boy played his game imagina-
tively. When the child announces angrily that he is really a United States marshal,
we credit him with an understanding of storytelling (as contrasted with lying) and
the strength of imagination that energized his pursuit of his imaginary villain. But
shall we say that he was sincere, even though his tone was serious? (Does the four-
year-old really believe that he is a United States marshal? Surely not. Was he really
asserting that he is a United States marshal? Surely not.) But, if, say, Paul Benacerraf
and David Kaplan, dressed in shorts, cowboy hats, and carrying toy shotguns, ran
into my seminar and said to me in angry seriousness, “We are really United States
marshals, but sometimes we go around like this,” should I say they were the pos-
sessors of dramatic and vivid imaginations, or should I take them to be simply de-
luded? Interestingly, one is common-sensically inclined to say that the adults are
childishly deluded while saying that the child is culturally sophisticated. Common
sense is wrong on both counts, as I shall show in what follows.
In describing the craft of the writer, phrases like these tend to come to mind: ‘a
work of the imagination’, and ‘the writer’s task . . . to revive his imagination every
day during his working hours’. ‘Imaginative’ is applied to the writer’s mental state
while composing, to kinds of literary product, and to the ability of a reader to under-
stand the product created. But what do we know of these literary abilities, powers,
motives, and products when we know them all to be imaginative in this sense or these
senses? Have we said anything more than that these literary abilities, powers, mo-
tives, and products are . . . well, literary? The answer, I believe, is “No.” And if imagi-
native art is just art, we most focus on that.
In the opening of the essay “The Medium of Fiction,” Gass writes:
It seems a country-headed thing to say: that literature is language, that stories and
the places and the people in them are merely made of words as chairs are made of
smoothed sticks and sometimes of cloth or metal tubes. Still, we cannot be too simple
at the start, since the obvious is often the unobserved. . . . That novels should be
made of words, and merely words, is shocking really. It’s as though you had dis-
covered that your wife were made of rubber: the bliss of all those years, the fears
. . . from sponge. (Gass 1970: 27)
Modernist writers like Gass keep announcing to us the “obvious,” excusing them-
selves in advance for doing so, because what they are after is not “really” obvious,
and the claim that it is is just a rhetorical trick. They have a philosophical belief about
the literary use of language, a belief shared by the mid-twentieth-century W. H. Auden,
the late-century William Gass, and the early-century Karl Kraus, among others:
Auden It is both the glory and the shame of poetry that its medium is not its pri-
vate property, that a poet cannot invent his words and that words are prod-
ucts, not of nature, but of a human society which uses them for a thousand
different purposes. (1968: 23)
Gass The novelist makes his book from boards which say Ladies and Gents.
Every scrap has been worn, every item handled; most of the pieces are
dented or split. The writer may choose to be heroic—poets often are—he
may strive to purify his diction and achieve an exclusively literary lan-
guage. He may pretend that every syllable he speaks hasn’t been spit, some-
times, in someone else’s mouth. Such poets scrub, they clean, they smoothe,
they polish, until we can scarcely recognize their words on the page. “A
star glide, a single frantic sullenness, a single financial grass greediness,”
wrote Gertrude Stein. . . . The use of language in fiction only mimics its
use in life. (1970: 30–31)
Kraus My language is the universal whore whom I have to make into a virgin.
(cited in Auden 1968: 23)
The distinctions between literary and ordinary, figurative and literal, or poetic
and standard language were systematically employed throughout the 1920s and 1930s
by the Russian formalists and members of the Prague Linguistic Circle. Their intent
was the development of poetics as a rigorous science of “poeticality” and “literariness”
comparable to structural linguistics as a rigorous science of “grammaticality.” In Jan
Muka6ovský’s highly influential essay of 1932, “Standard Language and Poetic
Language,” literary language is asserted to be independent of standard language,
although related, in that “the standard language is the background against which is
reflected the aesthetically intentional distortion of the linguistic components of the
work, in other words, the intentional violation of the norm of the standard” (1970:
42). In poetic language attention is primarily on the words themselves and only sec-
ondarily on their communicative use. This attention is focused by special intonation,
unusual vocabulary, and startling semantic juxtapositions of words.
Of course, there is more to this view than dispassionate linguistic analysis.
Muka6ovský quotes approvingly from a pre–World War I (1913) essay of Ferdinand
Brunot:
Modern art, individualist in essence, cannot always and everywhere be satisfied with
the standard language alone. The laws governing the usual communication of thought
must not, lest it be unbearable tyranny, be categorically imposed upon the poet who,
beyond the bounds of the accepted forms of language, may find personalized forms
of intuitive expression. It is up to him to use them in accord with his creative intu-
ition and without other limits than those imposed by his own inspiration. (cited in
Muka6ovský 1970: 52)
In the view of the early modernists the language least like Muka6ovský’s poetic
language is, of course, the language of natural science. It is here that the literary
aesthetes join forces with the logical empiricists. Both agree that scientific sentences
are true or false and are communicative of thought, whereas poetic sentences are
neither true nor false and are emotive. We find this supposed difference in language
noted in a passage of John Locke’s, who in Essay Concerning Human Understand-
ing writes:
If we would speak of things as they are, we must allow that all the art of rhetoric,
besides order and clearness, all the artificial and figurative application of words
eloquence hath invented, are for nothing else but to insinuate wrong ideas, move
the passions, and thereby mislead the judgment; and so indeed are perfect cheats.
(Locke 1690: 3.10.34)
Speaking of things as they are! How far this seventeenth-century phrase seems from
our self-conscious, post–World War II philosophical rhetoric! Upholders of the lit-
eral/figurative distinction have typically claimed that ordinary language is truthfully
descriptive of the world and clearly communicative of thought, but figurative lan-
guage is neither.
Let us consider Muka6ovský’s arguments in favor of the distinction. He first
observes that literary language makes use of lexical and syntactic resources that
are allegedly unavailable to the standard language—for example, slang in poetry;
a combination of nonstandard dialect in dialogue and standard dialect in narrative
within a novel; and archaic forms, like Locke’s “hath.” Although it is true that these
features are absent from discourse in standard dialect, similar kinds of differences
obtain between the standard and almost any nonstandard dialect. Such nonstand-
ard dialects are just as clearly, cognitively communicative as the standard. One
cannot conclude that such features make literary language uncommunicative of
thought when ordinary language with such features is communicative of thought.
Moreover, that the properties mentioned hold of literary texts is obviously not
sufficient to show the existence of a coherent literary “language,” a genuine liter-
ary dialect.
Other devices mentioned by Muka6ovský include unusual intonation, choice of
words, and uncommon combinations of meanings, but, I would have thought, the
four-year-old’s remark, describing his father’s bald spot, “Daddy has a hole in his
head,” is not literature.
Muka6ovský’s (1970: 53–54) final argument concerns poetic neologisms, which,
he says, are invented for aesthetic purposes, are “unexpected, unusual (in form and
meaning), and unique.” The argument for neologisms of a poetic kind proceeds as
follows. The aesthetic function of a term is incompatible with communicative func-

tion. Terms with standard form and meaning function communicatively. If neolo-
gisms that functioned aesthetically had standard form and meaning, they would
function communicatively. If they functioned communicatively, they would not func-
tion aesthetically. Since they do function aesthetically, they do not have standard form
and meaning. And, hence, they are not part of standard language.
This is a valid argument, but it begs the question. The distinction between po-
etic and standard language is just the distinction between aesthetic and communica-
tive terms. The question of the intelligibility of the former distinction is the same as
that of the latter distinction. Muka6ovský offers no noncircular defense of the literal/
figurative distinction.
In “The Medium of Fiction,” which closely follows Paul Valéry’s (1961) “Po-
etry and Abstract Thought,” Gass looks for the difference between literary and lit-
eral language in the effects that language has on its audience rather than in its syntactic
and semantic properties:
The purpose of a literary work is the capture of consciousness, and the consequent
creation, in you, of an imagined sensibility, so that while you read you are that patient
pool or cataract of concepts which the author has constructed. (Gass 1970: 33)
Fiction and poetry provide the reader with a new self. Valéry supports the same view:
A poem is really a kind of machine for producing the poetic state of mind by means
of words. . . . Poetry is an art of language; certain combinations of words can pro-
duce an emotion that others do not produce, and which we shall call poetic. What
kind of emotion is this? . . . I recognize it in myself by this: . . . that things and be-
ings—or rather the ideas that represent them—somehow change in value. They at-
tract one another, they are connected in ways quite different from the ordinary; they
become . . . musicalized, resonant, and, as it were, harmonically related. (Valéry
1961: 79, 64, 59)
Valéry is more candid than Gass in revealing the logic of this position. Left with Gass’s
description of the reader’s new and imagined sensibility—one characterized by at-
tention to and absorption in the poem, the story, or the novel—it is unclear why this
effect on the mind is characteristic of literary language. Could not a theoretical physi-
cist like Steven Weinberg or a biologist like Stephen Jay Gould, in reading an essay
on general relativity or the theory of evolution, experience the same absorption Gass
describes? Valéry is more explicit. The answer to the question “What can poetic lan-
guage do that ordinary language cannot?” is “Create a poetic state of mind in the
reader.” Insofar as our mental states are individuated by their causal antecedents and
consequents, and by their objects, to be in a poetic state of mind is to be in a state
caused by reading poetry and in a state having as its object a literary text, the poem
read. The thesis now goes: poetic language differs from ordinary language because
the former produces the poetic state of mind and the latter does not.
Of course, the poetic state is just that state caused by and having as its object
poetic language. Thus Valéry’s claim reduces to this: poetic language differs from
ordinary language because poetic language produces the mental state caused by and
having as its object poetic language. Obviously Valéry has not advanced the under-
standing of the difference between poetic and ordinary language; he has merely re-
stated the difference in terms of mental states.
Valéry’s claim is not quite so vacuous as I’ve just made it seem. He character-
izes the receptive state as follows:
Observe the effect of poetry on yourselves. You will find that at each line the meaning
produced within you, far from destroying the musical form communicated to you,
recalls it . . . as though the very sense which is present to your mind can find no
other outlet or expression, no other answer, than the very music which gave it birth.
(Valéry 1961: 72)
What characterizes the poetic state of mind in the reader is the feeling of this inti-
mate union between sound and sense. We shall de-psychologize Valéry’s descrip-
tion and say that in literature, especially poetry, the sound and the sense of a sentence
or of a word are inseparable; in ordinary language, they are not.
To emphasize the consequences of Valéry’s view, I generalize his criterion of
literariness:
A sentence S of English is literary if and only if (a) there is no sentence T of
English, T not identical to S, that “paraphrases” S, and (b) there is no sentence
T* in any natural language L, L not identical to English, that “translates” S.
Our readerly intuitions about poetry support Valéry’s claim, as long as we look none
too closely at the relations of translation and paraphrase. But this calls for a little
example, taken from Robert Frost, who held a view similar to Valéry’s:
It is blue-butterfly day here in spring,

And with these sky-flakes down in flurry on flurry
There is more unmixed color on the wing
Than flowers will show for days unless they hurry.
But there are flowers that fly and all but sing,
And now from having ridden out desire
They lie closed over in the wind and cling
Where wheels have freshly sliced the April mire. (Frost 1979: 225)
Even in as uncomplicated a structure as this, it is evident that paraphrase into prose

loosens the taut connection between sound and sense that makes those words worthy
of attention. How would one flatly begin? Would one say, It’s spring here, and there
are lots of blue butterflies around today?
It is not unfair to take Gass and Valéry to claim that paraphrase or translation
of poetic language into prose language is impossible because any change of word-
ing or grammar changes the sound and so alters whatever relation between sound
and sense obtains in a poetic line. But then it is equally impossible to “translate,”
in the sense of preserving those linguistic features of the phrase, ordinary prose
into different ordinary prose. By the criterion of literariness just mentioned, this
makes a prosaic sentence poetic—and this is far from Valéry’s intent. The only
translation he allows is the homophonic one that paraphrases the clause ‘It is blue-
butterfly day here in spring’ by the sentence ‘It is blue-butterfly day here in spring.’
The employment of this question-begging conception of paraphrase and transla-
tion is not, I believe, defensible in a criterion of literariness that is intended to de-
fend the autonomy of literary language. But even if we adopt a more conventional
notion of translation, as Robert Frost did, the criterion is still in difficulty. Auden
comments illuminatingly:
Frost’s definition of poetry as the untranslatable element in language looks plau-

sible at first sight, but, on closer examination, will not quite do. In the first place,
even in the most rarefied poetry, there are some elements which are translatable.
The sound of the words, their rhythmical relations, and all meanings and associa-
tion of meanings which depend upon sound, like rhymes and puns, are, of course,
untranslatable, but poetry is not, like music, pure sound. Any elements in a poem
which are not based on verbal experience, are, to some degree, translatable into
another tongue, for example, images, similes, and metaphors which are drawn from
sensory experience. (Auden 1968: 23)
Obviously I do not believe the matter is quite as simple as Auden makes it seem, but
it is a more reasonable assessment than Valéry’s, Frost’s, or Gass’s. In turn, Auden’s
account will not quite do. For good translations of literature not only preserve sense
and convey the connotations of the original text, it is possible for them to capture
relations of sound and rhythm. If translation could capture only sense and certain
implications of sense, translations of Lewis Carroll’s “Jabberwocky” from Through
the Looking Glass would be impossible (see Guenthner and Guenthner-Reutter 1978
and Atlas 1980b). You’ll recall the first verse:
’Twas brillig and the slithy toves

Did gyre and gimble in the wabe,
All mimsy were the borogoves,
And the mome raths outgrabe. (Carroll 1963: 191)
Not only are there two Latin “translations,” a French “translation,” and a German
“translation,” but it is possible to judge their relative quality. In my view the German
“translation” by Robert Scott is better than the French one by Frank Warrin:
Il brilgue: les tôves lubricilleux

Se gyrent en vrillant dans le guave,
Enmimés sont les gougesbosqueux,
Et le mômerade horsgrave.
Es brillig war. Die schlichte Toven

Wirrten und wimmelten in Waben;
Und aller-mümsige Burggoven
Die mohmen Räth ausgraben. (Carroll 1963: 193)
By any reasonable standards “Le Jaseroque” and “Der Jammerwoch” are successful
translations of “Jabberwocky.” And Valéry’s criterion notwithstanding, “Jabberwocky”
is still a literary achievement, even if it is nonsense.
Let us review. I have argued that the term ‘imaginative’ in classifying literary
works is nonexplanatory. The classification rests instead on a notion of “literary” as
opposed to “ordinary” uses of language. I have considered several attempts to de-
fend this distinction: Muka6ovský’s classic, 1930s account of the distinction between
poetic and standard language in which poetic language is (a) a violation of linguistic
norms and (b) language used for its own sake; Valéry’s and Gass’s account of the
distinction between poetic and literal language in which poetic language is (c) un-
translatable and (d) creates a poetic state of mind. The arguments put forth to defend
the figurative/literal distinction are circular, presupposing versions of the distinction
itself, but in my discussion I have not confronted directly the central question. Is there
a theoretical distinction to be drawn between metaphorical and ordinary language?
Will this distinction do any explanatory, linguistic work?
Gass writes wonderfully about the opening scene of Hamlet:
Hamlet, Horatio, and Marcellus walk upon the castle platform awaiting midnight
and Hamlet’s father’s ghost. Hamlet says, “The air bites shrewdly; it is very cold,”
and Horatio answers, “It is a nipping and an eager air.” Hamlet and Horatio do not
think of it as cold, simply. The dog of air’s around them, shrewd and eager, run-
ning at heels. The behavior of this dog is wittily precise in their minds. It nags—
shrewishly, wifelike. The air is acidulous, too, like sour wine. Hamlet and Horatio,
furthermore, are aware of the physical quality of their words. Horatio not only de-
velops Hamlet’s implicit figure, he concludes the exchange with the word that began
it, and with sonorous sounds. The nature of the weather is conveyed to us with
marvelous exactitude and ease, in remarks made by the way, far from the center of
action; so that we find ourselves with knowledge of it in just the offhand way we
would if, bent on meeting a king’s ghost, we too went through the sharp wind. Yet
Hamlet’s second clause is useless. “The air bites shrewdly” is the clause that tells
us everything. It is cold. The wind is out. The wind is alive, malevolent with wise
jaws. The two clauses have a very clear relation. The first is metaphorical, the sec-
ond literal [my emphasis]. Both are about the weather, but one is art, the other not
[my emphasis]. . . . We are forced, in what is really a very complicated and very
peculiar manner, to infer [the state of the weather] from logical absurdities, strange
comparisons, and silly riddles. The speed with which we make our inferences should
not deceive us of the fact we make them. The air bites, therefore the air is alive. The
air bites shrewdly, therefore the air is wise. It is eager, so it feels. These deductions,
upon the information that it nips, and the immediate conclusion that it nips as dogs
nip, give us the dog of the air itself. To communicate the nature of the weather,
Shakespeare has introduced an altogether novel set of concepts; novel, that is, with
respect to the idea of weather as such; and it is through these concepts that we un-
derstand the kind of wind and cold we’re in. (Gass 1970: 60–62)
Of course, ‘bites shrewdly’ is not literally true of the air; air is not the sort of
thing that literally bites. But let us agree that ‘bites shrewdly’ is figuratively true
of the air.
The first systematic discussion of metaphor in Western philosophy occurs in

book III of Aristotle’s Rhetoric. He notes what we all intuitively would observe about
an example like ‘The air bites shrewdly’. The predicate ‘bites shrewdly’ has appar-
ently shifted its sense and has been applied to a thing of a different category. As Nelson
Goodman (1976: 69) has metaphorically put it, “a metaphor is an affair between a
predicate with a past and an object that yields while protesting.” For what is special
about the metaphor is that, in context at least, the application of an old term in a new
way is both perfectly intelligible and linguistically odd.
1.2 The second dogma

The oddity has been characterized by some philosophers, following Bertrand Russell
(1956b [1908]) and G. Ryle (1949), as a type-theoretic or “category mistake” (or the
violation of a selectional restriction; Chomsky 1965) in which facts of one logical cate-
gory are described in the vocabulary appropriate only to another category. Air, not being
animate, cannot literally have jaws that bite. Air neither bites nor doesn’t bite.
It is common to believe that many (not all) metaphorical sentences are literally
anomalous category mistakes, that nonetheless they do make sense, that “the mean-
ings”—interpretations—they have may vary from person to person, are probably not
recursively specifiable, and are highly sensitive to the context. It is also usual to believe
that the oddity and intelligibility are in conflict, as in Goodman’s elegant metaphori-
cal formulation: metaphor is an affair between a predicate with a past and an object
that yields while protesting.
It would usually be said that, literally, the sentence ‘The air bites shrewdly’ is
nonsense; it is a category error. It would make as much sense to say that √2— is six feet
tall. Some theories of the understanding of metaphor will claim that the reader or lis-
tener constructs an alternative metaphorical interpretation of the sentence-token (which
is taken to be a metaphorical “sense”) because a literal interpretation is blocked—for
example, by a category error. We would interpret the sentence-token literally if we could,
but since we can’t, we have to search for another interpretation, the “metaphorical”
one, or relinquish the assumption that the speaker intends to be intelligible. The sen-
tence-token is viewed as having a “pseudoambiguity,” possessing both an anomalous
literal and an acceptable, though odd and hard-to-specify, metaphorical reading. Thus
one standard story goes. (On my view, by contrast, the alleged shift in sense is not the
creation of an ambiguity, since in the metaphor the “new” use is a transformation of
the “old” use in a way uncharacteristic of an ambiguous or homonymous term. ‘Bank’
in ‘money bank’ is not a metaphorical use of ‘bank’ in ‘river bank’. A semantical ex-
planation of metaphor will not rest on the notion of ambiguity.)
Even Grice characterized figurative language as floutings—that is, blatant vio-
lations, of Grice’s Conversational Maxims—namely, of the First Maxim of Quality
(“Do not say what you believe to be false”), by category errors:
Examples like You are the cream in my coffee characteristically involve categorial
falsity, so the contradictory of what the speaker has made as if to say will, strictly
speaking, be a truism; so it cannot be that that such a speaker is trying to get across.
The most likely supposition is that the speaker is attributing to his audience some
feature or features in respect of which the audience resembles (more or less fanci-
fully) the mentioned substance.2 (Grice 1989b: 34)
The second dogma of literary modernism is the reduction of metaphoricity to

semantical anomaly.
Most philosophical views recognize the importance of verbal context in guid-
ing the reader or listener to make the correct inference in understanding the “meta-
phorical meaning” of the utterance. Consider what one would say the line meant if
one confronted the single sentence:
The air bites shrewdly.
Contrast this with the sequence of lines:
The air bites shrewdly; it is very cold. It is a nipping and eager air.
The semantic information of those lines obviously interacts in the reader’s mind,
guiding his inferences. The problem is, Why is this special to metaphor?
In the sentence ‘The cold bites shrewdly’, ‘bites shrewdly’ is true of the cold. It
is then possible to explain the figurative application of ‘bites shrewdly’ to the air.
Suppose we antecedently know ‘The cold bites shrewdly’. We read or hear the sen-
tence ‘The air bites shrewdly’. To understand the appropriateness of such an utter-
ance, we construct this argument: if anything cold bites shrewdly, and the air is cold,
it follows that the air bites shrewdly. We infer the premise—viz., that the air is cold,
that we need to deduce the metaphoric utterance we heard—thereby “making sense”
of it by making explicit the deductive relationships in which it figures. (This is an
inference analogous to what C. S. Peirce called ‘abduction’; in Atlas and Levinson
1981 it was called by me, suggested by analogy with G. H. Harman’s 1965 phrase
‘inference to the best explanation’, ‘inference to the best interpretation’.) This premise
that the air is cold is what we understand to be conveyed, or to be implicated as Grice
(1975a) would say, by, in, or when asserting the sentence ‘The air bites shrewdly’.
We fix on this implicatum because of prior knowledge of ‘The cold bites shrewdly’.
We sometimes use metaphors to understand metaphors via their implicata, and thus
inference analogous to abduction is essential. (In fact, the same formal analogy holds
of Practical Syllogisms; see Brown and Levinson 1978, 1987 and Kenny 1966. It is
a mistake, though, to think that an analogy is an identity. Interpretative inference is
not identical to abductive, explanatory inference. Psychological explanation is not
the same as semantic interpretation, I shall claim, contra Grice (1989a) and Hobbs
et al. 1993; see chapter 2, Section 5.)
2Grice fails to consider that the notion of “fanciful resemblance” is parasitic on the metaphorical use
rather than conversely. Occasionally I have mused on Grice’s choice of example You are the cream in
my coffee. The American edition of Sir John Mortimer’s (2000) third volume of autobiography appeared
in his 79th year. He reports that “one of my earliest memories is of a bungalow my parents rented at
some seaside where, each morning, my father would run a Union Jack up an improvised flagpole and
we would ceremoniously burn the contents of the wastepaper-baskets and bury the leavings of the sink
to the accompaniment of a song which went, ‘You’re the cream in my coffee, You’re the sole of my
shoe’, scratchily played on a wind-up gramaphone. At least I learned a respect for rubbish” (Mortimer
2000: 138–39).
Horatio’s remark, that it is a nipping and eager air, illustrates how the use of a
term licenses the use of another term semantically related to the first. ‘Bites shrewdly’
licenses the use of ‘nipping’. Metaphorical understanding is achieved by tapping a
whole system of concepts (Black 1954c; Lakoff and Johnson 1980). Again, the prob-
lem is, Why is this special to metaphor?
1.3 The third use argument

I shall now argue that the modernist distinction between the literal and the figurative
is precisely the wrong conception to use in understanding language and art. I am not
claiming that the distinction is rather one of degree—tight versus loose—than of kind
(Sperber and Wilson 1986a), or that it is hard to know how to make it. This distinc-
tion is not susceptible to conversion from one of kind to one of degree and still be an
intelligible distinction between terms, or of meanings, or of uses of language. I am
claiming that there is no distinction. It is not there to be drawn. And it gives a false
picture of the way language functions.
Consider the negative sentence:
My cousin isn’t a boy any longer.3
One can understand a use of this sentence to make a statement that the speaker’s male
cousin is now an adolescent or an adult. One can also understand a use of this sen-
tence to make a statement that the speaker’s young cousin has had a sex-change
operation. These two understandings of the sentence do not constitute an ambigu-
ity.4 But there is a third, possible, intelligible use of that sentence. One can also under-
stand a use of this sentence to make a statement that the speaker’s cousin, who is a
hermaphrodite, has grown up.
Notice that there is a temptation here to say that the sentence cannot be taken
“literally” and understood in this third way. One might be tempted to say that one
must understand the statement “metaphorically.” Has the predicate ‘isn’t a boy any
longer’ experienced a transfer of meaning from [ISN’T A YOUNG MALE HUMAN ANY
LONGER] to . . . , to what? To [ISN’T A YOUNG HUMAN ANY LONGER]—that is, to [ISN’T A
CHILD ANY LONGER]? Then is the sentence ‘My cousin isn’t a boy any longer’ used to
state figuratively that the speaker’s hermaphroditic cousin isn’t a child any longer?
This might be an example, I suppose, of J. M. Sadock’s (1984) Principle of Loose
Talk, which I will discuss, skeptically, in chapter 6. But the suggested metaphorical
analysis is not plausible: [CHILD] is not the figurative meaning of ‘boy’ in the speaker’s
use of the sentence. He meant [BOY], not [CHILD], but it is true that the predicate ‘isn’t
a boy any longer’ is being applied to an entity that is not in an extension of the
predicate’s stereotypical application—that is, to a hermaphrodite possessing ovaries.
For any literal statement of the sentence might be thought to entail that the speaker’s
3This sentence is first discussed, for different theoretical purposes, in D. Terence Langendoen, “Pre-
supposition and Assertion in the Semantic Analysis of Nouns and Verbs in English,” in Steinberg and
Jakobovits (1971: 343). It was also used by Robert Stalnaker, for his own purposes, in his “Pragmatic
Presuppositions,” published in Munitz and Unger (1974: 204).
4See Langendoen, “Presupposition and Assertion”; Atlas (1975a,b, 1977b, 1989).
cousin was a boy in the past, but, note, one’s hermaphroditic cousin has never been
just a boy—has never been a stereotypical boy (cf. Atlas and Levinson 1981: 40–41,
Atlas 1984a). Of course, the application of the predicate is made easier by its being
applied to an entity that possesses all the commonsense stereotypical properties of
boys (Putnam 1975).
The semantic features of this understanding of this sentence—the apparent shift
in sense to something “odd” and not readily paraphrased and the apparent shift in
the predicate’s application to objects—have been taken to be characteristic features
of metaphor. Also like metaphorical uses, this use of the sentence remains perfectly
intelligible. This use of the sentence satisfies all the traditional criteria of metaphorical
use. Nevertheless, its use is a perfectly “ordinary” use of the sentence in which the
predicate is applied to an entity that satisfies the commonsense stereotypical proper-
ties of biologically normal boys. And the ordinary, literal meaning [YOUNG & MALE
& HUMAN] of the predicate is consistent with this use and interpretation of it. Young,
human hermaphrodites are biologically male (just not only male).
What does this mean? It means that the traditional criteria of metaphoricity do
not suffice to distinguish between so-called metaphorical and ordinary uses of lan-
guage. Faced with this consequence, one may assert that, for reasons unknown to
ourselves, our traditional characterization of metaphor is faulty and we must look
for subtler and more adequate criteria, or one may take a more radical position, which
was my choice in 1975. I asserted that so-called metaphors were semantically ordi-
nary—that metaphorical use of the poetic sort is of the same logical type as the third,
ordinary use of ‘My cousin isn’t a boy any longer’. My claim was that the semantics
of metaphorical language is no different from the semantics of ordinary language:
not looser (Sperber and Wilson 1986a), not a matter of different usage or inference
(Grice 1975a,b; Davidson 1981), not a matter of a special “metaphorical” sense. There
is no difference whatever in logical type, however different the perlocutionary ef-
fects of the “metaphorical” utterance upon the addressee may be (Austin 1962/1975).
There is a deep semantical reason for this similarity in logical type, and for the
failure of classical semantical theories to note it. Although it has different uses, the
negative sentence ‘My cousin is not a boy any longer’ is semantically unambiguous.
It has one sense. So does the term ‘boy’. Yet in its third use the term applies to an
individual, a hermaphrodite, that lies outside the extension of a predicate with the
intension [YOUNG & MALE & NONFEMALE & HUMAN] but lies inside the extension of a
predicate with the intension [YOUNG & MALE & HUMAN]—that is, of the predicate ‘boy’.
In the framework of Atlas (1974, 1975a,b, 1978b, 1979, 1989), one would say that
the English lexeme ‘boy’ is semantically nonspecific with respect to [NONFEMALE].
There are not two lexemes ‘boy1’ and ‘boy2’ with intensions [YOUNG & MALE & HUMAN]1
and [YOUNG & MALE & NONFEMALE & HUMAN]2. Rather, the use of ‘boy’ suggests—by
an inference formally akin to Peirce’s “abduction,” or to Aristotelian practical syllo-
gisms—the more specific interpretation, the sense attributed to the lexeme ‘boy2’.
The felt anomaly of the example sentence arises from the unexpected defeasibility
of this inference in the application of ‘boy’ to the young hermaphrodite. The word-
form boy is neither ambiguous nor used metaphorically in my example, and certainly
it is not ambiguous between a literal sense and a “metaphorical” sense, although the
standard accounts of metaphor would lead one to think so. The standard accounts of
metaphoricity cannot explain these linguistic observations, and the classical distinc-
tion between the literal and the figurative is not, I suggest, sustainable.
This special kind of nonambiguity of the word-form boy is pervasive in language
and of great theoretical interest. I (Atlas 1975a,b, 1977b, 1978a,b, 1979, 1989) have
argued elsewhere that a logical and linguistic notion as basic as negation, the mean-
ing of the free morpheme ‘not’, has a quite different, semantically nonspecific, mean-
ing in English and in all other natural languages from that of the extensional
truth-functional negations of two- and many-valued mathematical logics. The clas-
sical distinction between exclusion and choice negation senses of ‘not’ is not sus-
tainable.5 This difference between natural and formal language derives from the
conceptual economy of ordinary language. (For more on this theme, see Ziff 1972b
and Horn 1984b.) Likewise, the “metaphorical” use of ordinary language, which is
an ordinary use, provides us the economical alternative to vast elaborations of primi-
tive vocabulary and conceptual distinctions. Better to leave matters a little semanti-
cally underdeterminate (I do not mean ‘underdetermined’; see section 2.3 of this
chapter) than make the language impractical to use or impossible to learn. (For more
on this theme, see Atlas 1989: 7–24.)
In my view, human genius has shown its most original linguistic achievement
not in creating a purely “literary” use of language, which Gass rightly admires as an
aesthetic creation in Gertrude Stein’s writing, but in creating a purely “literal” use of
language, as in mathematics and physics. It is a greater linguistic leap, and it puts
one farther from one’s biological and evolutionary roots, to create the language of
pure mathematics than to borrow the language of the shopclerk to write a novel, as
Gass so poignantly noted.6
Modernist writers since Mallarmé have struggled to separate literary from ordi-
nary language, science from art, and writers from the populace.7 If only language
weren’t so public! What Gass and other literary modernists fail to perceive, for philo-
sophical and linguistic reasons of their own, is the theoretical banality of fictive, or
ordinary, language, and the theoretical abnormality of factive, or extraordinary, lan-
guage. The literary modernist aesthetes and the logical empiricists made the same
mistake in accepting the language of science as the norm. Aesthetes saw art only in
the abnormal language of the imaginative/literary minority. Positivists saw factual-
ity only in the normal language of the unimaginative/vulgar majority. The aesthete
had the right conception of language and the wrong conception of the imaginative.
The positivist had the right conception of the imaginative and the wrong conception
of language. Neither got what he wanted, and each got what the other wanted. In the
5For a language L, and the set V of admissible valuations val, val ∈ V, the choice negation –A of a
sentence A is such that val(–A) = 1 iff val(A) = 0, and the exclusion negation ¬A of a sentence A is such
that val(¬A) = 1 iff val(A) ≠ 1. Put sloppily, the difference is one between ‘false’ and ‘not true’. In a
bivalent language the negations are extensionally identical. In a non-bivalent language they diverge.
6Foran analogous argument differentiating ordinary seeing from artistic and scientific observation, see
N. R. Hanson (1969) and B. Russell (1913: 9).
7As noted earlier, James Wood (1998: 42) criticizes novelist Tom Wolfe for his use of “the most vulgar
word.” See Steiner (1975a).

weary, war-torn, irony-laden twentieth century I suppose that this result was to be
expected.8
2 Semantical nonspecificity, utterance

interpretation, and psychological modularity
In this section9 I review the distinction between semantical ambiguity and semantical
nonspecificity, with particular focus on the modularity versus nonmodularity of
mental processing and on the conceptual question of where to draw the boundary
line between the modular and the nonmodular. In doing so I examine the influen-
tial views of Jerry Fodor (1990a). I discuss the distinction between semantical
nonspecificity and ambiguity in both the visual and the verbal. I discuss the senses
in which perceptual mechanisms are inferential in character and introduce the dis-
tinction between encapsulated and unencapsulated processes. Then I turn explicitly
to the question whether speech perception is a modular process and criticize the ac-
count of Fodor (1990b). I draw on my discussion of visual and verbal nonspecificity
to develop an alternative account of speech perception and interpretation of utter-
ances. I pose the problem of speech perception in a novel—and what I take to be its
correct form—under the heading “A Language of Thought and the Nonmodularity
of Communicative Intentions: How Utterance-Interpretation can be both Hermeneutic
(Nonmodular) and Enormously Reliable (Functionally Modular).” And in the con-
cluding subsection I summarize my argument and point out the philosophical errors
underlying the ill-formedness of the old problem of speech perception.
2.1 Visual/verbal nonspecificity versus

visual/verbal ambiguity
Plane figures drawn without converging perspective offer no depth cues. Described
by the Swiss crystallographer L. S. Necker in 1832, the classic example is a figure
we take to display a cube (fig. 1.1.). The back face and front face are drawn the same
size; thus no size difference—the smaller as the back, the larger as the front—serves
to indicate which is the front and which the back. What is curious is that although we
recognize the figure as a cube-picture (and not as a truncated pyramid-picture, which
the figure would display if we always imposed converging perspective [Hochberg
1972: 56–57]—and I can see fig. 1.1 as a figure of a truncated pyramid, the back
face serving as the top, the pyramid being seen from above), what we see spontane-
8The argument of this section explains the linguistic foolishness to be commonly found in the views of
literary theorists such as Stanley Fish (1989: 4, 164). See also A. Sokal and J. Bricmont (1998) and
Thomas Nagel (1998).
9Parts of this section have appeared in my “On the Modularity of Sentence Processing: Semantical
Generality and the Language of Thought,” in Jan Nuyts and Eric Pederson (eds.), Language and
Conceptualization (Cambridge: Cambridge University Press, 1997), pp. 213–28, and in chapter 1 of
Atlas (1989) and appear here with the permission of Cambridge University Press and Oxford Univer-
sity Press.
FIGURE 1.1 Necker cube
ously reverses in depth. Alternately, the figure is seen as a left-directed, downward-

projecting cube seen from above or as a right-directed, upward-projecting cube seen
from below. (Occasionally, a viewer will see the figure as just a set of intersecting
line segments in the plane.)
Here is a story that the English psychologist Richard Gregory (1970, 1973, 1986)
tells: Given a (roughly) two-dimensional image on the retina, the brain tries to deter-
mine what three-dimensional object is being seen, or, what distal three-dimensional
object causes the proximal two-dimensional stimulus. Although the retinal image could
be the result of any number of different projections onto a plane of any number of dif-
ferent three-dimensional objects, the brain either rejects or never considers odd or com-
plex possibilities. The absence of standard cues for perspective in the figure means that
no unique solution to even the brain’s reduced problem can be computed, so, in turn,
it entertains a small, finite number of probable solutions. As the viewer continues to
look at the drawing, the seen cube spontaneously flips in and out of the page.
Psychologists have described multistable figures as depth-ambiguous. The am-
biguity of the noun phrase (NP) ‘the chicken that is ready to eat’ is syntactic: ‘the
chicken’ is either an underlying subject NP or an underlying object NP. The phrase
manifests distinct grammatical analyses, or underlying structural descriptions; it is
the superficial outcome of encoding distinct syntactical roles for the noun phrase ‘the
chicken’. A multistable Necker cube figure is not a figure that manifests distinct
perspectival analyses; it is not the outcome of encoding distinct depth cues. Rather,
it is constructed without conventional perspectival cues; the construction is depth-
perspective free. So the figure is neutral with respect to depth perspective. This
absence from the construction of the figure of perspectival cues for depth creates the
multistability, the spontaneous reversals of depth in what is seen.
The ambiguity of a description differs from the spontaneous multistability of a

figure. Unlike what is seen in the Necker figure as the mind tries to see what the fig-
ure “displays,” the different senses of an ambiguous phrase do not spontaneously
exchange places in the understanding as the mind tries to grasp what the phrase means.
An ambiguous sentence does not spontaneously perform semantic flips, overruling
the meaning being attended to by the mind. Nor does the grasp of one sense exclude
the simultaneous grasp of another sense: one can understand both meanings of a
presented two-ways ambiguous sentence. These observations emphasize important
differences between the mental processing of an ambiguous phrase and the process-
ing of figure 1.1.
It is sometimes forgotten that in 1967, in his Beckman lectures “Language and
Mind” at the University of California, Berkeley, Noam Chomsky pointed out an
observation of J. R. (Haj) Ross that a necessary condition of syntactical deletion was
sameness of sense of the deleted element with a formally identical element in the
sentence. Chomsky wrote:
(8) I don’t like John’s cooking any more than Bill’s cooking.
(9) I don’t like John’s cooking any more than Bill’s.
Sentence 9 is ambiguous. It can mean either that I don’t like the fact that John cooks
any more than I like the fact that Bill cooks, or that I don’t like the quality of John’s
cooking any more than I like the quality of Bill’s cooking.* However, it cannot
mean that I don’t like the quality of John’s cooking any more than I like the fact
that Bill cooks, or conversely, with “fact” and “quality” interchanged. That is, in
the underlying structure [associated with] 8 we must understand the ambiguous
phrases “John’s cooking” and “Bill’s cooking” in the same way if we are able to
delete “cooking.” (Chomsky 1972a: 33)
———
* There may also be other interpretations, based on other ambiguities in the structure “John’s
cooking,” specifically the cannibalistic interpretation and the interpretation of “cooking”
as “that which is cooked.”
It is characteristic of an ambiguity of this type that (9) is two-ways rather than per
impossibile four-ways ambiguous. Crossed interpretations are not possible as literal
meanings of the sentence. The reduced form allows only parallel interpretations as
possible meanings.
George Lakoff (1970) and A. Zwicky and J. Sadock (1975) adopt the following
two criteria:
1. The impossibility of a crossed, literal paraphrase for a conjunction-

reduced sentence S entails the ambiguity of S.
2. The possibility of a crossed, literal paraphrase for a conjunction-
reduced sentence S entails the nonambiguity of S. (The distinct,
parallel paraphrases do not express distinct senses.)
If my analysis of the depth-nonspecification, rather than depth-ambiguity, of the

Necker cube drawing is correct, there should be a visual analogue to the ambiguity
test just mentioned. One ought to be able to see a conjoined, reduced, double Necker
cube drawing so that simultaneously one cube is seen in one perspective and the other
cube in the other perspective. When I made this prediction in Jerry Sadock’s pres-
ence, he promptly drew figure 1.2. If you fill in the two missing edges of that figure,
you will discover that you will see two intersecting Necker-cubes in their respective,
different perspectives. The Necker cube drawing fails a spatial “reduced conjunc-
tion” test for depth-ambiguity.
One paradigm example of an ambiguous figure is W. E. Hill’s maiden/hag fig-
ure (fig. 1.3). Like the alternative readings of an ambiguous sentence that are often
unnoticed by listeners in conversation, the alternative displays of an ambiguous fig-
ure are often unnoticed by viewers. In both cases, unlike the Necker cube figure,
prompting is sometimes necessary before the alternative reading or display is per-
ceived or comprehended. And unlike the Necker cube figure, seeing the figure as a
young woman and seeing it as an old woman do not freely, attention-independently
alternate, though with attention to different parts of the figure, one can see it as a
young woman and then as an old woman. Once the several displays are recognized,
like the ambiguous sentence whose different senses can be grasped simultaneously,
the different “objects” displayed by an ambiguous figure, unlike the different pre-
sentations (the differently oriented displays) of the Necker cube figure, can be seen
simultaneously. For example, though many find it difficult to do, if I focus fixatedly
on the ‘+’ in figure 1.3 I can see Hill’s figure simultaneously as the maiden and as
the hag. (This is not a neurophysiological claim; this is a perceptual claim about what
it is possible to see.) This is not typical of spontaneously multistable figures like the
Necker cube drawing. It is the “split” aspects of multistable figures that make them
seem properly describable as ambiguous, but, ironically, it is precisely this character
that marks them psychologically as not ambiguous.
FIGURE 1.2 Double Necker cube

FIGURE 1.3 Old woman—young woman
The parallel I am drawing is this: in a non-display-ambiguous but display-gen-

eral figure like the Necker cube drawing, the figure is pictorially “nonspecific” with
respect to depth; in a non-sense-ambiguous but sense-nonspecific sentence, the sen-
tence is semantically nonspecific with respect to a semantic feature F—for example,
unspecified for either semantic gender [MALE] or [–MALE] in the case of ‘I’, ‘my’, and
‘neighbor’. I want to contrast both the Necker cube drawing and sense-nonspecific
sentences with the ambiguous maiden/hag drawing and sense-ambiguous sentences
as follows (see Atlas 1989):
Ambiguity: Visual (e.g., Maiden/Hag Drawing) and Verbal

a. Two (or more) underlying, specific structures of a figure/expression
b. Seeing-as can be overridden; it is possible to:
{ } { }
See displays
two (or more) simultaneously
Grasp senses
c. Fixation is possible: No spontaneous, attention-independent flips
between displays of a figure/expression.
d. Prompting is sometimes required to recognize alternate displays of a

figure/expression.
e. The figure/expression passes a Conjunction Reduction Test for
ambiguity.
Nonspecificity: Visual (e.g., Necker Cube Drawing) and Verbal

a. One underlying, nonspecific structure of a figure/expression
{ } { }
b. See presentations
two (or more) only alternately10
Grasp contents
c. No fixation is possible: Spontaneous, attention-independent flips
between presentations of a figure/expression.
d. Prompting is never required to recognize alternate presentations of a
figure/expression.
e. The figure/expression fails a Conjunction Reduction Test
for ambiguity.
2.2 Are perceptual mechanisms inferential and modular?

Now I want to introduce into my discussion a story from Jerry Fodor:
Changes in states of the retina, for example, register changes in the properties of
incident light, which are in turn caused by alterations in the arrangement of the distal
objects that radiate and reflect the light. To the extent that such proximal effects are
specific to their distal causes, cognitive processes with access to the one have grounds
for inference to the other.
So, the picture is that certain organic states register the proximal stimuli that
cause them, and that certain cognitive processes infer the arrangement of local dis-
tal objects from the organic effects of these proximal stimulations. In particular, I
assume that it’s the function of perceptual mechanisms to execute such inferences.
(1990b: 209)
Of course, in Fodor’s own view the organic states do not merely register the
proximal stimuli that cause them, they represent them. In addition, Fodor holds a
Modularity Thesis, according to which the mechanisms that execute these inferences
are both dedicated and encapsulated.
Why should we conceive of this perceptual process as inferential? One standard
argument, supported by the Necker cube phenomenon, is that sensation underdeter-
mines perception—that is, the proximal stimulation contains “less information” than
the perceptual beliefs that result. The extra information presumably must come from
a store of background knowledge. Thus perception is a “top-down” process. Hence,
it is inferential.
Fodor usefully points out that this argument from underdetermination actually
is independent of the inferentiality of perception: “Even if the information in the
10This claim can be elaborated; see my comment in Atlas (1989: 18, n.4).
proximal light uniquely determines the visible properties of the distal layout, the
inferentiality of the mental process that proceeds from representing the one to repre-
senting the other would not be impugned” (1990b: 210). The argument is, rather,
that perception fixes beliefs about distal objects. If the organic effects—for example,
on the retina—represent at all, they represent the proximal stimuli that cause them.
So we conceive of perception as a process in which “representations of proximal
stimuli causally determine beliefs about distal layouts,” and that is what Fodor means
by perception being inferential. Of course, by the same argument, the secretion of
hydrochloric acid in the stomach represents the ingesting of boeuf bourgignon, and
the representation of that ingestion causally determines the belief that one is no longer
hungry, from which, on Fodor’s views, it follows, absurdly, that digestion and satia-
tion are inferential.
Whatever the correct account of the relationship between inference and percep-
tion is, it cannot be Fodor’s. The problem with Fodor’s account, as I suggest later
(see section 4), may lie in its very first step, the notion that “stimulus meanings” (the
proximal stimuli) are “represented” at all. It is at least clear that in so far as the ques-
tion of an inference to a perceptual belief’s being justified is reduced to the question
of there being the “right” causal chain, Fodor has joined the ranks of classical repre-
sentationalists, theorists of knowledge who, knowingly in Fodor’s case, blur the dif-
ference between causality and justification (Rorty 1979: 139–48, 1998b; Putnam
1992b, Fogelin 1994: 175).
2.3 Is speech perception modular?

It does not follow from the fact that perception is, on Fodor’s view, inferential that it
is “conscious” or that it is “thinking” in the sense of problem solving. Take the case
of speech-perception. When an addressee hears a sound that is produced by another
human being whom the addressee believes to speak his (the addressee’s) language
and that is analyzed to be a token of an utterance-type in that language, an acoustic
representation has been unthinkingly transformed into a linguistic representation if
the addressee is in command of the language. Similarly, one points one’s head at the
Necker cube drawing, opens one’s eyes, and looks. The result in each case, nearly
instantaneously, is, respectively, an understood utterance-type and the perception of
a cube-picture. These processes are notable for the following features: speed; not being
consciously accessible; not being voluntary; and at least hypothetically, being algo-
rithmic, in the sense that a computable function succeeds in assigning to an acoustic
argument a linguistic value (Fodor 1975: 151–52; Fodor 1990c: 239, 1990b: 212, 214).
(One worry about these features is the existence of “garden-path sentences” like The
horse raced past the barn fell, which may well cause the parser to halt and the mind
to shift to a general-purpose problem solver [Fodor 1990b: 229, n.9].) Otherwise,
these features suggest the use of a dedicated, special-purpose processor rather than
the use of a generalized problem solver.
Furthermore, the linguistic description of the speaker’s utterance is one that the
addressee shares with other hearers of the speaker’s utterance-token who are co-
linguals with the addressee. Anyone who knows the language is in a position to give
a description of “what the speaker uttered,” despite differences among hearers in
background beliefs. By virtue of those differences in background beliefs, hearers may

disagree about what the speaker meant when, in, or by uttering what he uttered, but
they will not, ceteris paribus, disagree about the identity of the sentence the speaker
uttered (Fodor 1990b: 213). So far, I agree with Jerry Fodor, though I have been
careful to distinguish the product of my “sentence-understander” as a linguistic de-
scription of an utterance- and sentence-type, from the product in which Fodor is most
interested—namely, the propositional content P of the speaker’s communicative
intention that the addressee believe that the speaker believes P.
Furthermore, Fodor (1990b: 215) believes that what is most important for
his argument for a dedicated sentence-processor is the truth of a causal claim; for
example, the truth of the statement ‘Tokens of It’s raining are caused by intentions
to communicate the belief that it’s raining’, just as what is most important for his
argument for a dedicated visual processor is the truth of the statement ‘Tokens of a
Necker-cube-drawing-retinal-display-type are caused by a distal layout of a Necker-
cube-drawing’. The truth of such causal statements skips the part of the story in which
I am most interested: the intermediate step of the utterance-type between the acous-
tic form and the inferred belief of the speaker. Furthermore, unlike Fodor, I doubt
that an inference to the mental state of the speaker is algorithmic, in the sense of a
computation guaranteeing the assignment of a canonical description of a speaker’s
mental state to a canonical description of an acoustic representation of the speaker’s
utterance, even when the mental state is the speaker’s intention to utter a sentence of
a certain type, much less when the mental state is the speaker’s intention to commu-
nicate his belief P. Fodor (1990b: 212) inexplicably takes the view that of all the
inferences we make about others’ intentions from their behavior, there is just one
class of inference that is algorithmic: inference to communicative intention. On
Fodor’s view, one would have to conclude, absurdly, that when a bush burns there is
always oxidation of cellulose except in the one case in which Moses is watching.
Fodor (1990b: 214) more cautiously says, at just one point in his discussion, that
what the processor assigns to the acoustic representation is “the literal content of what
the speaker says,” a description from which talk of intention is absent. This descrip-
tion almost fits the description that I have given. Fodor ignores this description be-
cause he wishes to minimize the importance of the conventionality of language in
order to maximize the similarity of sentence perception with ordinary visual percep-
tion as he conceives it and, hence, to maximize the significance of causal relations
rather than conventional rules.
2.4 What do semantical and visual nonspecificity mean?

I shall turn Fodor’s argument upside down. My discussion of the Necker cube draw-
ing suggests that visual perception is more like sentence perception, as I conceive it,
than Fodor realizes. Like the product of a sentence-processor, the product of the vi-
sual processor is not at first a presentation of an external object, where the presenta-
tion is the “content” of the visual representation; it is a display of an object, but the
display, in the case of the Necker cube figure, is not determinately of one content
(the presented inward-projecting cube) or another content (the presented outward-
projecting cube). Nor is it ambiguous between them. It is a display that is nonspecific
between the two presentations (contents). An ambiguous figure would conventionally

display both contents; a semantically nonspecific figure conventionally displays
neither content—the figure is a conventional drawing of neither an inward-projecting
cube nor an outward-projecting cube.
The “natural” opposition between the linguistic and nonlinguistic is the opposi-
tion between “proposition” and “image.” What this description of the opposition
misses is the contrast for images that is exactly parallel to the contrast for sentences
between meanings, thoughts, and states of affairs: the contrast between display, pre-
sentation, and portrayal (Atlas 1989: 27–28). This is the contrast, for both sentences
and images, among what is uttered/displayed, what is meant, and what is referred to.
In Atlas (1989) I had suggested the existence of previously unnoticed similarities in
visual and verbal processing. The relationship between stimuli and interpretation in
each respective case shows some strikingly similar properties: semantical nonspeci-
ficity versus semantical ambiguity and the mental processing thereof.
If I am right about this parallel between visual processing and sentence process-
ing, there should be something in the perception of ordinary physical objects that
corresponds to the conventions of language in the perception of sentences. I think
there is. Why should an ordinary perceiver, “knowing the rules” of perspective com-
mon in our culture, see a three-dimensional cube picture when viewing the Necker
cube diagram rather than a two-dimensional pattern of lines in a plane? (Why should
an ordinary speaker, “knowing the grammar” of English, hear a statement that the
cat is on the mat rather than a pattern of sounds in a time interval?) Shimon Ullman’s
(1979) work long ago suggested that built into the computations that produce inter-
pretation of even quite minimal visual information about physical objects is the com-
putational equivalent of assumptions about the rigidity, three-dimensionality, and
surface continuity of the causal sources of the retinal stimuli. In the information pro-
cessing of which we are speaking, it plays the same role of constraining the interpre-
tation of retinal data that the grammar plays in constraining the interpretation of the
acoustical data.
What is the point of emphasizing what Fodor deemphasizes? The point is that
without what Fodor calls “literal content” rather than the content of the speaker’s
intention, one would not have the features of a dedicated sentence-processor that make
utterances immediately understandable without conscious awareness of the process.
Although a semantically nonspecific sentence-meaning will not provide all the in-
formation in the content of the speaker’s utterance-meaning, it will either constrain
the range of possible specific interpretations or provide the base for further infer-
ence. But upon receipt of the acoustical information, it will provide a semantical
information sketch fast. Then the sketch can be filled in from contextual information
or collateral beliefs of the addressee by processors that are not dedicated.
Information at the level of semantically nonspecific sentence-meaning or visu-
ally general pictorial display is the product, probably, of dedicated processors. These
processors are not cognitively penetrable; they are encapsulated. Their processes are
features of the “functional architecture” of the system, since they yield the interpre-
tations they do whatever one’s conscious beliefs in the context may be: the displayed
object is a three-dimensional rigid object, whatever its depth orientation is; or the
understood object is an English sentence with a “literal meaning,” whatever else the
speaker meant by it. These features are immune to bias from background informa-
tion (Fodor 1990c: 242): whatever depth orientation you think the cube should have,
it will still flip in and out of the page; whatever Nietzsche meant by what he uttered,
his sentence still means in English ‘God is dead’.11 If one could not tell relatively
quickly and accurately by bottom-up processing that a sound was an English word
or sentence, and what English word or sentence it was, when one heard it, one’s ability
to communicate would not be an efficient, hierarchically structured general capac-
ity—that is, a capacity that takes as input any word, bounded sequences of words, or
sentences of English, even if your knowledge of your own name makes recognition
of that one word by top-down processing relatively fast, and faster than your recog-
nition of some arbitrary word or sentence of English.
If what I have just said is plausible, then sentence-understanding at the level
of a computation of a semantically nonspecific semantic representation of a sen-
tence-type is a “modular” process in Fodor’s sense. In nonmodular processes, by
contrast, search mechanisms have access to any or all of my beliefs. Furthermore,
for each new bit of information that I access, I would have to recalculate the credi-
bility of a belief on the new evidence base. If memory searches are “costly” in time,
and if reconsideration of a belief in light of further evidence is also “costly” in time,
an encapsulated process would be a relatively faster process (Fodor 1990b: 219).
I am quick to understand what the sentence the speaker uttered meant, even if I am
not as quick to understand what the speaker meant by, in, or when uttering the
sentence.
On Fodor’s (1983: 91) own account in The Modularity of Mind, the questions
about modular systems that must be answered are, “What is the most that an encap-
sulated processor should be supposed to compute? Which aspects of the input can
plausibly be recognized without generalized appeal to background data?” On my view,
the most would be literal meanings of a sentence-type exemplified by an utterance-
token. Inferences to the communicative intention of a speaker of a sentence-token—
that is, to the content of a mental state of the speaker—fall outside the domain of a
modular system. Fodor’s account of the modularity of sentence-perception seems to
me mistaken.
2.5 A language of thought and the nonmodularity of

communicative intentions: How utterance-
interpretation can be both hermeneutic (nonmodular)
and enormously reliable (functionally modular)
I am of course committed to the mind’s having acoustic representations of utterances,
semantic representations of sentences, understandings of the contents of utterances—
that is, of what the speaker intended to mean—an addressee’s interpretation of what
11As an example of the prevalence of mistaken views about processing among philosophers, consider
Paul Churchland’s remark that “one learns very quickly to make the figure flip back and forth at will
. . . by changing one’s assumptions about the nature of the object or about the conditions of viewing”
(1988: 8). Churchland makes the usual mistake of treating the Necker cube drawing as an ambiguous
figure. It is obvious that the upward Necker cube image does not flip when one thinks to oneself, “Now
I believe this is a downward Necker cube.”
the speaker uttered, and functions from classes of one to classes of the other: map-
pings from sentence-meanings into speaker’s/hearer’s meanings, from mental se-
mantic representations into thoughts, which, of course, are identified by Fodor and
others with “sentences” in a “language of thought.” I appeal to a notion of seman-
tic representation in the mind, a mental “language” of sentence meaning. Fodor
and others have appealed to the notion of thinking as computation, so thoughts are
entities on which computations may be defined—thus “sentences” in a language
of thought.
Stephen Levinson reviews arguments that I and others have proffered in sup-
port of the conclusion, as Levinson puts it, that “semantic representations and con-
ceptual representations are not only distinct kinds of representations . . . they are not
isomorphic” (1997: 24). They are distinct kinds of representations, “contrary to as-
sumptions in many diverse quarters, including not only Fodoreans, Cognitive Lin-
guists and those of similar views, but also most branches of computational linguistics”
(1997: 24). But the contrast that I defended was not, without further assumptions,
one between types of representations. It was one between semantically specific
thoughts and semantically nonspecific sentences. In the history of modern philoso-
phy and psychology, from John Locke forward, philosophers have underemphasized
the autonomy of the language faculty (Chomsky). My arguments were intended to
redress the imbalance.
But my argument did not require that thoughts be “representations” in some
computational sense—for example, syntactical objects on which algorithms can be
defined. For those philosophers of mind and cognitive scientists who accept a ver-
sion of a “computational theory of cognition” for mental states like belief and desire
(Cummins 1989)—or a representational theory of the mind in general (Fodor 1975,
and others)—the moral of my argument was that many thoughts—that is, their “sen-
tences” in Mentalese—were not identical to, or synonymous with, any single sen-
tence in English, even after the parameters of tense, deixis, demonstratives, and
pronouns had been given values. (The content of a thought might be describable by
a complex relation among sentences of a natural language but not given by a single
syntactically well-formed sentence. [Note: a logical connective, such as the conjunc-
tion symbol ‘&’, is not a relation-symbol.]) The problem with the computational view
is that to the extent that Mentalese is “representational” like a human language, it
will, like all human languages, possess the semantical features, reviewed by Levinson,
that distinguish sentences from thoughts.
If Mentalese is actually a human, not a formal, language, as is the language of
thought, the inevitable semantical nonspecificity of its semantically interpreted syn-
tax could no more be identical with the contents of some of our thoughts than are the
literal meanings of sentences of English or Guugu Yimithirr. To the extent that
Mentalese does precisely express the contents of all our thoughts, to that extent it
will not be a linguistic representation; it will be a medium of thought, but it will not
be a language of thought. One English sentence can express the content of a thought,
but that does not entail that the meaning of the sentence is identical to the content of
the thought (Atlas 1989: 27–31). The mistake of conflating “a meaningful sentence
expressing an intentional content” with “the meaning of a sentence being an inten-
tional content” afflicts almost all philosophical discussion of the matter (e.g., Crane
1995: 194). Of course, it might have happened that thoughts are really thought in the
medium of an extensional formal language whose intended interpretations are ex-
pressed in classical model theory, but I’m betting that the biology of the brain is
unlikely to have evolved in such a Hilbertian and Tarskian manner.
Ultimately, the problem may not be the relationship between linguistic and con-
ceptual representations. The problem may be, with a rich enough notion of represen-
tation for there to be an interesting problem, the relationship between the
“representational” and the nonrepresentational.
Where I differ from Fodor is with his claim that
[what] is usually required in the intentional analysis of behavior is a kind of herme-

neutic sophistication that’s as far as can be from the execution of a rote procedure.
The notable exception is inferring intentional content from utterance form. Show
me an English speaker who utters “it’s about to rain” and I’ll show you an English
speaker who is, in all likelihood, thinking about the weather. This sort of inference
is enormously reliable.12 All you have to know about an English speaker is that he
made a certain sort of noise, and the intentional interpretation of his behavior is
immediately transparent. (Fodor 1990b: 213)
Unfortunately for Fodor’s view, science is not in the business of making notable
exceptions. Inferring communicative intentions from verbal behavior is not, pace
Fodor (1990b: 214), a solved problem. Although hermeneutic sophistication is a
symptom of a problem whose solution is never an “all likelihood,” “enormously re-
liable” one, unlike ordinary utterance interpretation, the problem, rather, is to ex-
plain how utterance-interpretation can be both hermeneutic and enormously reliable.
As I have argued elsewhere:
Linguists and philosophers find it natural to split their analyses of language into
three levels: the sentence (grammar), the statement (speech-act theory), the speaker
(pragmatics). There is a very strong tendency for logicians and syntacticians to focus
on the sentence, and for philosophers to want to reduce statements to speakers’ mean-
ings. I inveighed against this in Atlas (1975[a]), when I argued that Frege had three
notions of presupposition, one at each level. The sentence may be thought of as an
abstract object, e.g. as in Katz (1981), and studied mathematically. The speaker may
be thought of as a mental entity and studied psychologically. What, then, has hap-
pened to that convention-bound, truth-bearing, but intention-laden stuff that is pa-
role? (Atlas 1989: 3–4)
12A counterexample to Fodor’s claim: I meet you on Tottenham Court Road and say to you, “Rainy weather,
isn’t it?” and you reply, “Yes, it is” and pass on. In fact, only in the most etiolated sense was either of us
thinking about the weather. This commonsense philosophical observation is supported by the results of
brain scans using positron-emission tomography (PET). Regions of the left frontal and temporal lobes
(Broca’s and Wernicke’s areas) are not activated during mere speech production (e.g., repeating presented
words out loud); they become activated during conscious attention to a presented word-meaning and the
selection of a semantically relevant verbal response. But even then, with fifteen minutes practice in the
task, the task of choosing a semantically relevant word in response to a stimulus is taken over by the
motor areas of the brain, with no involvement of the temporal and frontal lobes. Thus it becomes possible
to utter words, and make sense of, for example, “Yes, it is,” without thought (Raichle et al. 1994).
I take all three levels of meaning seriously (see Levinson 1995: 110, 112; Lyons
1995b: 235, 237–39), but my topic in this book is the relation between literal sen-
tence-meaning and mid-level parole: convention-bound (“reliable”) and intention-
laden (“hermeneutic”).
A sketch of just such an explanation of reliable and hermeneutic utterance-
interpretation was given in Atlas (1979, 1984a, 1989), where semantically non-
specific, nonpropositional (hence non-truth-value-bearing) semantic representations
of sentences are arguments of inferential mental relations whose co-domains are
thoughts—for example, a speaker in asserting the sentence ‘I had a drink’ gener-
ally conversationally implicates (Grice 1975a) the sentence ‘I had an alcoholic
drink’, which is notated by: “I had a drink” » I had an alcoholic drink. These
relations produce reliable hermeneutic utterance-interpretations. Semantically non-
specific Semantic Representations of sentence-types and, thus, the inferential func-
tion called by Atlas and Levinson (1981) “inferences to stereotypes” constrain the
interpretation of sentence-tokens (utterances), making interpretation reliable even
though nonmodular (see also Atlas 1989; Horn 1984b; Levinson 1990; Récanati
1993: 260–68).
2.6 Conclusion
There is an issue about what processes it makes sense to model as modular processes,
and I have drawn the line between the modular and nonmodular at a different place
from Fodor (1990b). For me the modular stops at semantically nonspecific Semantic
Representations of sentences or semantically nonspecific pictorial representations of
objects. The mental contents of speakers’ intentions will have to be taken care of by
what I once called “inference to the best interpretation” (Atlas and Levinson 1981),
which is a formal analogue of, but not reducible to, “inference to the best explana-
tion” (Atlas and Levinson 1981: 50; Fodor 1983: 88), a concept central to American
philosophy of science since Charles Sanders Peirce’s discussion of “abductive infer-
ence.” And that, as the world knows, is not an inference that is encapsulated or
cognitively impenetrable.
So, to summarize: (a) My observations on the Necker cube drawing show that
figures have the same semantic properties of ambiguity and sense-nonspecificity
as do sentences. Visual perception is more like sentence-perception than Fodor sug-
gests. Further, (b) Fodor’s (1990b: 210) account of what it means for perception to
be inferential absurdly makes digestion and satiation inferential as well. (c) There is
no justification for the claim that a scientific understanding of intentions should di-
vide into two subclasses, communicative intentions and all the rest. A dedicated sen-
tence-processor that makes utterances immediately understandable will not provide
the information in the content of the speaker’s meaning. Rather, it will constrain the
range of possible specific interpretations of the speaker’s meaning or provide the
basis for further inference. I am quick to understand what the sentence meant, even
if I am not as quick to understand what the speaker meant. (d) The important issue
is where to draw the line between the modular and the nonmodular. Fodor has drawn
it the other side of communicative intentions, but such a division of intentions into
the modular communicative and the nonmodular noncommunicative has no concep-
tual or scientific defense. It is just another version of the mistaken Kantian division
of knowledge into the a priori and the a posteriori: linguistic knowledge of com-
municative intentions and empirical knowledge of other sorts of intentions. That
epistemological division, as Morton White (1950), W. V. O. Quine (1980c), Hilary
Putnam (1983b), and latterly Jerry Fodor (1998b) himself, have taught us, is an
untenable dualism.
This book is devoted to an examination of the relationship between literal sen-
tence-meanings and utterance-interpretations, where the interpretations derive in part
from a form of inference that is both reliable and hermeneutic—generalized conver-
sational inferenda, either defined on a domain of propositions—as Grice’s (1975a)
generalized coversational implicatures from “what is said” to what a speaker meant
or what a hearer thought the speaker meant (see chapter 2)—or defined on a domain
of semantic representations of sentences, including the semantically nonspecific,
nonpropositional representations of literal meanings (Atlas 1978a,b, 1979; Bach 1987,
1994a,b,c; Récanati 1993: 267 n.7).
As I have noted (1978a,b, 1979), the pragmatic inference mapping PRAGatlas:
{<CONTEXT, SEMANTIC REPRESENTATION>} → PROPOSITIONS, mapping a pair consisting
of a context and a semantic representation of, say, a negative sentence into an exclu-
sion or choice negation proposition, is unlike Grice’s conversational implicature
mapping PRAGgrice: {<CONTEXT, ASSERTION>} → PROPOSITIONS. My (Atlas 1979) class
of mappings takes contexts with Semantic Representations (SRs) into propositions
(thoughts)—the class [<K, SR> → P]. If the sentence with that SR was asserted, then
the content of the statement made in the context by the asserting of that sentence is
given by the proposition in the co-domain of the relation. Grice’s class of relations
takes contexts with asserted propositions into propositions (thoughts) conveyed by
what was asserted, namely the class [<K, “P” > → P]. In certain special cases, as in
certain sets of “eternal sentences” (Quine 1960), an SR is equivalent to the proposi-
tion standardly expressed in any context by asserting the sentence whose semantic
representation is SR.
If we can understand the logic of these sorts of inference, we will have taken an
important step in solving the logical aspect of the problem of utterance-interpretation
and in so doing begin to formulate a theory of the semantics-pragmatics interface.
3 Semantical underdeterminacy and pragmatic

interpretative inferences
In Seven Types of Ambiguity William Empson (1930) discusses a Samuel Johnson

poem, “The Vanity of Human Wishes,” used for illustrative purposes, as I do here,
by Jan Kooij:
What murdered Wentworth, and what exiled Hyde,

By kings protected, and to kings allied?
What but their wish indulged in courts to shine,
And power too great to keep, or to resign? (quoted in Kooij 1971: 122)
Line 3 of this quatrain has an “ambiguity” that we mark by the following syntactic
differences:
[their wish to shine] [indulged in courts]
[their wish [to shine in courts]] [indulged].
A further alleged “ambiguity” concerns the readings of the elliptical ‘their wish to
shine indulged’, where one may expand the phrases to ‘their wish to shine indulged
by themselves’ or to ‘their wish to shine indulged by others’. The elliptical phrase is
syntactically and sematically neutral between these expansions (Atlas 1989: 25; Bach
1994a,b).
Another type of ambiguity is lexical ambiguity. The verb ‘indulge’ can describe
a dispositional or occurrent state, and so be classed as having the semantical feature
[GENERAL] or [TEMPORAL] (Kooij 1971: 122). There is also a lexical ambiguity in ‘al-
lied’ as in ‘connected by marriage’ and ‘connected by treaty’. Both these cases are
instances of polysemy—having different but related senses for one word. This is to
be contrasted with homonymy—as in the linguistic word-form port, where the same
form realizes two words: ‘port1’ meaning [HARBOR], and ‘port2’ meaning [FORTIFIED
WINE] (Lyons 1977: 550). Ambiguities arising from word-sense or syntactic struc-
ture are “inherent” in the sentence. In understanding poetry, a reader selects one or
more “readings” from those his or her knowledge of the language can provide him
that seem to him appropriate in the poem.
Line 4 presents a rather different problem. ‘Power too great to keep’ can be
understood (in specific ways) as either (a) “power too great for Wentworth and Hyde
to keep (but not too great for others to keep)”; or (b) “power too great for anyone to
keep.” Likewise, ‘power too great to resign’ can be understood as (a') “power too
great for Wentworth and Hyde to give up (but not too great for others to give up)
willingly”; (b') “power too great for anyone to give up willingly”; (c) “power too
great for Wentworth and Hyde to give up without danger from having exercised it”;
or (d) “power too great for anyone to give up without danger from having exercised
it.” These are contextually relevant “specifications” of the generalized, nonspecific,
literal sense of the words, distinct and relevant fillings-in, so to speak, of the senses
of the words, but not themselves distinct senses of those very words. One is reading
meaning into the words, not reading meaning out of them. For example, in various
contexts, the word-tokens of the girl with the flowers can be literally and felicitously
be used to “present” the more specific contents “the girl wearing flowers,” “the girl
selling flowers,” “the girl carrying flowers in her hand,” “the girl strewing flowers,”
and so on. Each of the relations between the girl and the flowers is subsumable under
the [WITH] relation, but ‘with’ is not four or more ways ambiguous in sense (Kooij
1971: 110; Weydt 1972: 573; Brugman 1981). One should distinguish this phenome-
non of nonspecificity of sense from lexical and syntactic ambiguity (Zwicky and
Sadock 1975; Cruse 1986: 51–68).
The existence of various interpretations (contents) of ‘with’, appropriate to dif-
ferent contexts in which ‘the girl with the flowers’ is uttered, does not entail that ‘with’
is ambiguous in sense among ‘the girl wearing flowers,” “the girl selling flowers,”
and so on. These contextual specifications are not listed in my dictionary—the one
on my desk or the one in my head. I do not search an antecedently given list for a
sense that fits the context of utterance and select it as the interpretation that ‘the girl
with the flowers’ presents in that context. Rather, it seems that knowing the meaning
of ‘with’, knowing what ‘the girl with the flowers’ displays as its meaning, I under-
stand the fitting interpretation of ‘the girl with the flowers’ in the context to be, for
example, “the girl wearing the flowers.” Rather than select a specific sense, I con-
struct a specific interpretation.
Perhaps I may reinforce the point with a different example from Ruth Kempson
(1977: 132–35). Kempson’s example is a negation. The sentence ‘It wasn’t a woman
that came to the door’ may be used to state a proposition that would be true if a girl
came to the door or a proposition that would be true if a man came to the door, but as
Kempson (1975, 1977, 1988a) and I (Atlas 1974, 1975a,b, 1977b, 1978a,b, 1989)
argue, it is not ambiguous in sense merely because it may express the distinct propo-
sitions “It was a non-adult, human female that came to the door” and “It was an adult,
human non-female that came to the door.” Kempson and I demonstrate that the sen-
tence is sense-nonspecific, not ambiguous, between these interpretations; the literal
meaning of the sentence is neutral between them. The literal meaning of the sentence
is neither one nor the other. The literal meaning of the sentence, the product of word-
meanings and the syntactical rules of their combination, is neutral between the inter-
pretations. It is identical to neither of them. The sense of a sense-nonspecific sentence
is not a proposition, the bearer of determinate truth or falsity (Atlas 1974, 1977b,
1978a,b, 1979, 1989).
In our language, general terms (those with divided reference like the common
noun ‘apple’) contrasted with singular terms (like the definite description ‘the presi-
dent of the United States in 2004’) are general in sense; that is, their meanings are
nonspecific with respect to certain predicates, some of which are “basic” vocabulary
in constructing a lexicon of English. For example, to say that an object is an apple is
not by virtue of the meaning of ‘apple’ to attribute a size to the object, unlike ‘dwarf’,
or a color, unlike ‘palomino’. (Redness is only stereotypical for apples; ‘pumpkin-
sized green apple’ is not an oxymoron.)
These “basic” predicates are those from the lexicon of any successful grammar
that would be employed to state the meanings of English words, especially the con-
trasts among those lexemes that are members of “contrast sets,” as [MALE]/[–MALE]
and [MALE]/[FEMALE] for {‘father’, ‘mother’}, {‘stallion’, ‘mare’}, and so on (see
Grandy 1987: 261, 272; Lehrer 1974; Lyons 1977; Pulman 1983). Examples may be
found in decompositional theories of meaning proposed by J. J. Katz (1972), G. Lakoff
(1971b), and G. A. Miller and P. N. Johnson-Laird (1976), but I shall not assume the
correctness of any particular lexical decomposition theory for lexemes of English.
Some of the difficulties with them have been interestingly discussed by Pulman (1983)
and Grandy (1987). I would be content if ‘basic’ were a psycholinguistic meta-
predicate in a theory of language acquisition (see Lakoff 1987; Hornstein 1989: 40
n.6).
By contrast, images, especially mental ones, like Bishop Berkeley’s (1710) ideas,
have been supposed to be quite specific. The mental image that I conjure up when I
imagine a speckled chicken has been supposed to be an image of a chicken with eighty-
seven speckles, or one with eighty-eight speckles, for instance. Nevertheless, it seems
to me that I can have a perfectly good image of a speckled chicken without my image
being specified for the number of speckles. As Paul Ziff once wrote, “Staring at a
pictorial representation of a man in ordinary black opaque unbulgy riding boots we
need not ask: is that man a web-footed or a nonweb-footed fellow? A real live man
must be one or the other, but a man in a picture is not a real live one” (1972d: 136).
The same goes for feathered bipeds as well.
In Princeton, New Jersey, on 16 March 1972, Paul Grice put a dittographed copy
of “Logic and Conversation,” the second of his 1967 William James lectures in Harvard
University, into my hands. A year later the deeper significance of his distinction between
the literal meaning of an expression and what a speaker might have suggested, implied,
or conveyed by, when, or in asserting it—its “conversational implicata” as he termed
them, or more generally its “pragmatic interpretative inferenda,” as I term them—
became more apparent. Like Kempson (1975), I realized that negative, definite descrip-
tion sentences were not scope-ambiguous among external and internal negation
readings nor were they lexically ambiguous among exclusion and choice negation
readings but, as I demonstrated in June 1974 using Zwicky and Sadock’s (1975)
ambiguity tests, they were semantically underdeterminate: semantically nonspecific
for scope interpretations or nonspecific among the exclusion and choice negation
interpretations (see Kempson 1988a; Horn 1989; Iten 1998).13 The free morpheme
‘not’ in English does not generate an ambiguous sentence but rather a sentence
semantically nonspecific with respect to the exclusion and choice-negation interpre-
tations. It was then that I realized that a hearer’s pragmatic inferences would have to
“take up the semantic slack” (as Kent Bach 1994c would later put it) for there to be
specific and precise interpretations of a speaker’s utterance even in the case of an
“eternal sentence” in Quine’s (1960) sense. The exclusion/sentence-negation inter-
pretations and the choice/predicate-negation interpretations were pragmatic, inter-
pretative inferenda, the products of pragmatic interpretative inferences. Kempson
(1975) came to similar conclusions about ‘not’ but tried, wrongly I believe, to give
an extensional, logical disjunction account of the semantical nonspecificity of ‘not’
(Atlas 1984b, 1989; Kempson 1988a).
Much has since been made of the notion of semantical nonspecificity (with re-
spect to a semantical property) since I first wrote about it.14 But semantical
nonspecificity is not merely semantical underdetermination. Of the latter, Sperber
and Wilson wrote: “As is well known, linguistic structure grossly underdetermines
the interpretation of an utterance: the linguistic meaning is generally ambiguous, it
may be elliptical or vague, it contains referential expressions with undetermined ref-
erents, the intended illocutionary force is often not fully specified, and implicatures
13 For a language L, and the set V of admissible valuations val, val ∈ V, the choice negation –A of a
sentence A is such that val(–A) = 1 if and only if val(A) = 0, and the exclusion negation ¬A of a sen-
tence A is such that val(¬A) = 1 if and only if val(A) ≠ 1. Put sloppily, the difference is one between
‘false’ and ‘not true’. (I shall sometimes use the abbreviation ‘iff’ for ‘if and only if’.) In a bivalent
language, the negations are extensionally identical. In a non-bivalent language, they diverge.
14Atlas 1974, 1975a,b, 1977b, 1978b, 1979, 1989; see Zwicky and Sadock (1975), Bach (1982, 1987,
1994a,b,c), Gillon (1990), Récanati (1989, 1993), Sperber and Wilson (1986b), Taylor (1998), and
Wilson and Sperber (1981).
are not linguistically encoded” (1991: 544). None of this—except the infelicity of
saying that a meaning is ambiguous—would have surprised a student of sentence
structure.
But my observations in the mid-1970s were about the semantical nonspecificity
or underdeterminacy of negative eternal sentences (in Quine’s 1960 sense of ‘eter-
nal sentence’), not just about the phenomena of semantical underdetermination that
Sperber and Wilson (1986a) describe. These were not the usual problems of disam-
biguation, deixis, fixing references, tense, ellipsis, and vagueness (in the logician’s
sense of the term ‘vague’; see Williamson 1994, Keefe and Smith 1996). My obser-
vations indicated a radical semantical nonspecificity of the literal meaning of nega-
tive sentences in natural language. It is this radical semantic claim and its
consequences for the philosophy of logic and the philosophy of mind, not to men-
tion classical problems in the philosophy of language, that were my focus in Atlas
(1989, 1997a). In contemporary Chomskyan (1996c) terms, my investigation was,
in Atlas (1989), and continues to be in this book, one of Internalist Semantics and
the pragmatics that must complement it.
In Atlas (1978b) and in a lecture in the Department of Phonetics and Linguistics
in University College, London, in 1978, and later in a subsequently published paper
(Atlas 1979), I showed how semantical nonspecificity with respect to a semantical
property—for instance, the nonspecificity with respect to [MALE] and [–MALE] of
‘neighbor’ by contrast with the specification of the lexical feature [–MALE] in ‘poet-
ess’—requires inferential mechanisms to produce from the semantically nonspecific
literal meanings of sentence-types the interpretations of speakers’ utterances of
sentence-tokens. These inferences were similar to those that Grice (1975a) had de-
scribed for producing the pragmatic implications of what a speaker has asserted, but
the domain of my pragmatic inference relation, unlike Grice’s (1975a), as I (1978a,b,
1979) indicated, were not assertions but non-truth-evaluable semantic representations,
and its co-domain were propositions. As observed by Récanati fifteen years later:
The mechanism of implicature generation suggested by Grice is intended to account

for the step from what is said to what is communicated. But how are we to account
for the step from sentence meaning to what is said? What bridges the gap instituted
by there being a ‘free’ type of context-dependence pervasive in natural language?
Grice does not address this issue. However, as many people have suggested (e.g.,
Katz (1972: 449), Walker (1975: 157), Atlas (1979: 276–8), Wilson and Sperber
(1981: 156)), the pragmatic apparatus by means of which Grice accounts for con-
versational implicatures can also be used to account for the determination of what
is said on the basis of sentence meaning. (Récanati 1993: 236)
In the work cited by Récanati, Katz (1972: 449) notes the use of pragmatic inference
in determining speaker’s reference for a singular term, as does Walker (1975: 157),
who also mentions pragmatic inference in disambiguating the sense of a speaker’s
utterance.
Only Atlas (1974, 1975a,b, 1977b, 1978a,b, 1979), Atlas and Levinson (1981),
Bach (1982), Kempson (1975), Levinson (2000), and Wilson and Sperber (1981)
generalize to cases involving semantical underdetermination and semantical
underdeterminacy (i.e., semantical nonspecificity). Atlas (1974, 1975a, 1977b,
1978a,b, 1979), Kempson (1975), and Atlas and Levinson (1981) cite semantically
nonspecific negative sentences; Wilson and Sperber (1981) cite the example of the
underdetermined John plays well, and Bach (1982) cites the example of the seman-
tically underdeterminate (nonspecific) I love you too. Wilson and Sperber observe:
Nor is it just disambiguation and the assignment of reference in which the maxims
play role. Quite often, they lead the hearer to ascribe to an utterance some proposi-
tional content that is not strictly warranted by semantic rules alone. Suppose John
Smith is playing the violin in front of us, and I say to you:
(4) John plays well.
In these circumstances, I would naturally be taken as having expressed the propo-
sition in (5):
(5) John Smith plays the violin well.
The fact that John is taken as referring to John Smith, and that play is taken as
meaning ‘play a musical instrument’ rather than ‘play a game’, has already been
accounted for [in our earlier discussion]. The resulting proposition should be (6),
where the missing direct object in (4) is semantically interpreted in terms of a spe-
cific indefinite phrase:
(6) John [sic] plays some musical instrument well.
It is clear, though, that the hearer of (4) will normally interpret it as expressing not
(6), but the more specific (5). We shall not attempt a full account of this fact here.15
Note, however, that (5) entails (6), so that whenever (5) is true (6) will also be true,
but not vice-versa. [Statement] (5) therefore has more chances of being informa-
tive than (6), and one can conceive of circumstances in which (5), but not (6), would
satisfy the maxims of informativeness, and (5) would thus be preferred to (6).16 It
seems clear that any adequate account of how (4) is interpreted as expressing (5)
rather than (6) will have to appeal to the maxims of informativeness, and that these
may lead the hearer to choose a more specific interpretation than is warranted by
the semantic rules alone. (Wilson and Speber 1981: 158)
Bach, cited in Atlas (1989: 30), discusses another phenomenon—not semantical

underdetermination but semantical underdeterminacy (nonspecificity):
Jack tells Jill, ‘I love you too’. He could mean any one of several things, (a) that
just as Jill loves him, so he loves her, (b) that he loves Jill and someone else too,
(c) that like someone else, he too loves Jill, or (d) that he has love as well as some
other feeling for Jill. Whatever he may mean, he would be speaking literally and
15Atlas and Levinson (1981) had given an account of the interpretation of indefinite noun phrases. This
account was adopted by Horn (1984b) and discussed by Mey (1993). Bach (1994a) discusses the ex-
pansion of sentences like (4) into propositions like (5). For decades Morton White, following J. L. Austin
(1962/1975), has made philosophical use of the expansion of ellipitical sentences; see White (1965,
1979, 1993).
16Grice’s (1975a) Maxims of Quantity were: (a) Make your contribution as informative as is required
(for the present purposes of the exchange) and (b) Do not make your contribution more informative
than is required.
yet, I claim, the meaning of the sentence he is using, though univocal, does not fully
determine what he means in using it. The point is not merely that what he means is
a matter of his communicative intention, but that the sentence [type] is semanti-
cally non-specific. It does not have a definite truth-condition, even with Jack and
Jill fixed as the values of ‘I’ and ‘you’.17 A condition necessary for its [i.e., the
utterance’s] truth is that Jack loves Jill, but one of the four other conditions is nec-
essary as well, depending on which of the (a)–(d) is meant. However the sentence
is not thereby ambiguous, but merely semantically non-specific.18 (Bach 1982: 593)
Récanati (1993: 236) appeals to an example of reference specification and ei-

ther meaning disambiguation or the specification of a semantically nonspecific
expression John’s book, depending on whether one thinks that John’s book is am-
biguous, which is the classical view (including Chomsky’s); in this passage he is
notably noncommittal about the semantics of the expression, but in Récanati (1989)
he had (incorrectly) taken it to be ambiguous:
In the interpretation process, the referent of ‘he’ and the relation between John and
the book in ‘He has bought John’s book’ are selected so as to make what the speaker
says consistent with the presumption that he is observing the maxims of conversa-
tion. The speaker might have meant that Jim has bought the book written by John
or that Bob has bought the book sought by John. The hearer will select the interpre-
tation that makes the speaker’s utterance consistent with the presumption that he is
trying to say something true and relevant. (Davis 1991: 99)
Failing to cite the earlier research by Atlas, Katz, and Walker mentioned by Récanati
(1993: 236), M. Taylor adds:
Récanati (1989, 1993) suggests that we need to distinguish two kinds of pragmatic
processes. Primary pragmatic processes play a role in determining what is strictly
literally said by an utterance in context. Secondary pragmatic processes take the
outputs of primary pragmatic processes and generate further “implications” as out-
puts. Among the further implications generated by the application of secondary
pragmatic processes to the outputs of primary pragmatic processes are Gricean con-
versational implicatures. This nifty idea represents, I think, a great advance. (Tay-
lor 1998: 344)
In the 1970s as I lectured to various audiences on what in 1998 was flatteringly

called “a great advance,” I came to appreciate how difficult the logical and linguistic
tasks that Grice’s program of philosophical analysis posed for philosophers and lin-
guists—nothing less than a reconceptualization of the boundary between literal sen-
tence-meaning, speaker’s meaning, and hearer’s interpretation of speaker’s meaning,
17See Chomsky, who writes, “we cannot assume that statements (let alone sentences) have truth-condi-
tions. At most they have something more complex: ‘truth indications’ in some sense” (1996c: 52).
18Bach (1987: 200–2) and, in a slightly different way, Stich (1983: 111–18, 120–23) make a similar
point about the so-called ambiguities in the allegedly transparent, translucent, and opaque “readings”
of propositional attitude sentences. Proper appreciation of the point would radically alter the received
view on the semantics of propositional attitude sentences, but I shall not address that problem here.
what in linguistics is called “the semantics-pragmatics interface.” In Atlas (1977b,

1978a,b, 1979) I had tried to show that both sides of the boundary as previously
understood, the sentence-meaning-side and the speaker’s-meaning-side, would have
to be reconsidered, not because there was no longer a proper, or Fregeanly “sharp,”
boundary, but because the earlier boundary, between “propositions” as sentence-type
meanings and “propositions” as utterance-token meanings was no longer a theoreti-
cally satisfactory boundary. The boundary had moved, and what it bounded had
changed from “propositions” to pre-“propositional,” semantically nonspecific seman-
tic representations of literal sentence-meanings19—syntactically combined lexical
items—and from “propositions” to an addressee’s interpretation of a speaker’s ut-
terance, which might or might not turn out to be a proposition (Atlas 1978b, 1979,
1989, 1997a).
Kempson in her “Grammar and Conversational Principles” had noted:
One of the few people in the mid 1970s to recognize the gap between linguistic
content of a sentence and the articulation of the truth conditions of its associated
propositions was Jay Atlas, who argued in a series of papers (Atlas 1975, 1977[b],
1979) that the linguistic concept of sentence negation was weaker than any concept
sufficient to characterize the truth-theoretic properties of propositions that nega-
tive sentences express. (1988a: 141 n.2)
Independently Kempson (1975) had suggested an (unfortunately) extensional, dis-

junction version of semantical nonspecificity for ‘not’. Kempson claimed that the
negative sentences were to be semantically represented by a logical disjunction of
the two interpretations, the exclusion negation ¬φ and the choice negation –φ,
or ¬φ ∨ –φ, without noticing that this disjunction is logically equivalent to the
weaker exclusion negation ¬φ (see Atlas 1978b). But what we want is a semanti-
cally nonspecific representation of the negative sentence that is neutral in meaning,
as J. L. Austin (1962) would have put it, between the statements ¬φ and –φ. The
conceptual error is thinking that a representation of sentence-meaning that is neutral
between statements ¬φ and –φ is the same as one whose information is expressed
by the least weak statement whose information is common to both— the least upper
bound (sup) of both ¬φ and –φ in the lattice <P, > of the set P of statements
ordered by the logical consequence relation , viz. the join ¬φ ∨ –φ of the state-
ments. For φ and Ψ, the “weaker” common part entailed both by φ and by Ψ is obvi-
ously φ ∨ Ψ, since φ (φ ∨ Ψ), and Ψ (φ ∨ Ψ). The “weakest” statement entailed
by φ is a logical truth, e.g. (χ ∨ ¬χ) . Thus the least weak statement whose infor-
mation is common to both ¬φ and –φ is ¬φ ∨ –φ. Unfortunately this is not an
accurate model of the absence of semantical specification. It is extensional rather
than intensional; although the English expression ‘Tom is my neighbor’ is neutral in
sense between [TOM IS MY MALE NEIGHBOR], [TOM IS MY FEMALE NEIGHBOR], [TOM IS MY
HERMAPHRODITIC NEIGHBOR], and so on, it is not synonymous with the logical disjunc-
tion in logical English Tom is my male neighbor ∨ Tom is my female neighbor ∨
Tom is my hermaphroditic neighbor ∨ Tom is my non-human neighbor ∨ . . . Tom is
19What Noam Chomsky (1996c: 38) now calls “symbolic representations” of an “Internalist Semantics.”
my ADJ neighbor (Atlas 1978b, 1984b, 1989). It is even more clear that the nega-
tive English sentence-frame containing ‘not’—A(. . . not . . .)—will not be synony-
mous with the logical disjunction in logical English ¬A ∨ –A, which, as noted, is
logically equivalent to ¬A. A logical disjunction, though logically weak, is seman-
tically specific.
In Atlas (1979) I had argued that an account in which negative sentences The F
is not G were scope-ambiguous was redundant. If pragmatic inferences were a nec-
essary part of a theory of utterance-interpretation, and the sentences were ambigu-
ous, the inferences would be essential to disambiguation, to the selection of the
appropriate sense in the context of utterance. On the assumption that the addressee
analyzes the speaker’s sentence-token and discovers two or more senses, a choice of
the appropriate sense must be made in light of collateral information available in the
context. Thus some inferential mechanism must produce a decision on the best “fit”
between each sense and the context of utterance. By contrast, if the sentence is un-
ambiguous but semantically nonspecific for semantic features F1, F2, . . . Fn, the in-
ferential mechanism must give in the context an appropriate, more specific or precise,
interpretation of the semantically nonspecific sentence. I hypothesized that since the
same inferential mechanisms were at work in both the cases of ambiguity and of sense
nonspecificity—one to select a reading, the other to construct a more specific or pre-
cise interpretation—the difference between selection and construction was that the
construction of a specific interpretation operated on a pre-“propositional” semantic
representation that was too underdeterminate, not merely too underdetermined, to
carry a truth-value (Atlas 1979; Bach 1982). Contrary to Grice’s view that pragmatic
mechanisms operated only post-“propositionally” to give what the speaker implied
by asserting a sentence, what statement the speaker asserted was determined by prag-
matic inference in addition to the semantic interpretation of context-dependent
indexicals, demonstratives, reference-fixing of singular terms, tense, and others.
I (Atlas 1979) also observed that classical Gricean theory could not give an ade-
quate explanation of the inferential mechanism: sentence-negations Not (The F is G)
were transformed into predicate negations The F is non-G, and the references of singu-
lar terms were fixed, but no coherent account was given of the case in which Grice’s
sentence-negation meaning (the exclusion negation)—sometimes expressed in English
by It is not the case that P, though, as I pointed out in Atlas (1974), It is not the case
that P can also express the predicate/choice-negation interpretation; see Horn (1989)—
passed through the pragmatic mechanism unchanged, the case in which informative
enrichment was absent. According to the classical Gricean theorist, this case should
have been the semantically “unmarked” case, but no Gricean principle in the theory
could noncircularly explain it to be the unmarked case, the case of saying what one
means. In fact, it is linguistically the “marked” case, as Sir Peter Strawson (1950) noted.
By contrast, my claim (Atlas 1974, 1975a,b, 1977b, 1978b, 1979), that the nega-
tive sentence-meaning [THE F IS NOT G] is semantically nonspecific between the classi-
cal scope-interpretations meant that the inferential mechanism operated equally in both
cases: it produced the more specific choice-negation interpretation [– (THE F IS G)], and
it produced the more specific exclusion-negation interpretation [¬ (THE F IS G)]. In nei-
ther case has the speaker “said” something “propositional,” if by ‘saying’ one means
merely the act of uttering a meaningful sentence of a semantically nonspecific sort. Or,
in either case the speaker has “said” something interpretable, if by ‘saying’ one means
performing a locutionary act of producing a token of an utterance-type that is inter-
preted by one’s addressee to have a determinate sense and reference.20 Or, in one case
or the other the speaker has “said” what he meant, if by ‘saying’ one means perform-
ing an illocutionary act the content of which is correctly interpreted by the speaker’s
addressee to have the speaker’s intended sense and reference.
Having previously attempted in Atlas (1989) to put in canonical form my expla-
nation of the nature of semantically nonspecific sentence-meaning by developing and
applying to sentences the lexical notion of the semantical nonspecificity of
subsentential expressions, in this book I attempt to formulate, but not “reduce” or
“metaphysically explain” (whatever that means), principles underlying an addressee’s
interpretation of a speaker’s utterance. As in Atlas (1979), I radically revise and extend
Grice’s (1961, 1967, 1975a,b, 1989a) account of generalized conversational impli-
cata to an account of generalized conversational inferenda. My subject is not just
generalized conversational implicature; it is generalized interpretative inference. My
two semantic and pragmatic inquiries together provide philosophers and linguists with
the framework they need to reconceptualize traditional philosophical problems in
the spirit of Grice’s Linguistic Turn (see chapter 2, section 1) and in the spririt of
Chomsky’s (1996c, 2000) Internalist Semantics.21
My book Philosophy without Ambiguity (1989) examined negation, presuppo-
sition, definite descriptions, negative existence statements, and natural kind terms,
the construction of the correct theory for which requires the distinction between
ambiguity and semantical nonspecificity. Atlas (1977a,b, 1978a,b, 1979) and chap-
ter 4 of Atlas (1989) showed why the concept of semantical nonspecificity was also
essential for a correct account of speaker’s-utterance interpretation. My earlier book
having delimited the semantics-pragmatics boundary from the semantics side of
Chomsky’s (1996c, 2000) Internalist Semantics, this book attempts to construct a
post-Gricean theory of generalized interpretative inference of the sort first adumbrated
in Atlas (1978a,b, 1979) and Atlas and Levinson (1981), in order to delimit the
semantics-pragmatics boundary from the pragmatics side of Chomsky’s (1996c, 2000)
Performance Systems. Together the two books constitute a systematic study of the
semantics-pragmatics interface.
20See Ziff (1972a). Similar notions of interpreting utterances surfaced in others’ theories as well, nota-
bly the “explicatures” of the Relevance Theory of Sperber and Wilson (1986b); the “implicitures” of
Bach (1994a,b); the Pragmatic Intrusion of Katz (1972), Walker (1975), and Levinson (1988, 2000)
for the interpretation of singular terms; and the two stages of pragmatic inference in Récanati (1989),
who in Récanati (1993: 236) acknowledges the earlier work of Katz (1972) and Walker (1975) on ref-
erence-determination of definite descriptions and on disambiguation and of Atlas (1979) on interpre-
tative specification of nonspecific literal meaning (see Horn 1992a: 265).
21 Not, it may have disappointed Grice to realize, in exactly the letter of Grice. In June 1980 I asked
him whether, in light of my own and others’ suggested revisions of his Maxims of Conversation—e.g.,
Atlas and Levinson (1981) and Sperber and Wilson (1986b)—he thought his maxims as originally
formulated in 1967 were both accurate and complete. He replied succinctly, “Yes.” He was being slightly
disingenuous, as he himself, in Grice (1981, originally a lecture of 1970), proposed a new Maxim of
Manner in order to generate the existential “presuppositions” of definite descriptions from negative
definite description sentences. See chapter 4 in this volume.
As I have suggested elsewhere (Atlas 1975, 1977b, 1978a,b, 1979, 1981, 1989),
for theoretical purposes I distinguish among sentence-types or -tokens, utterance-
types or -tokens, literal meanings of sentence-types or -tokens, speaker’s interpreta-
tions of utterance-types or -tokens, and addressee’s interpretations of a speaker’s
utterance-types or -tokens. My Chomksyan Internalist Semantics claims that a sen-
tence-type or -token, which is a syntactical entity—a well-formed syntactical com-
bination of lexical items—has a literal meaning. I represent that literal meaning by a
semantic representation. An assertion (statement) or utterance of a sentence-token
will have its literal sentence-type meaning but also a semantic interpretation in the
context of utterance. A semantic interpretation of the utterance in a context will be
determined when it fixes the references of singular terms, indexicals, demonstratives,
and tense and will be determinate when it makes “precise” or makes “specific” the
interpretation of semantically nonspecific general terms in the utterance.
The semantic representation of a sentence will be semantically underdeterminate,
by virtue of its semantical nonspecificity, so that it might not “express a proposi-
tion” or carry a truth-value (depending on the relevance of the specific information
to the context of evaluation), as well as semantically underdetermined, by virtue of
its lacking values for its referential variables, so that it would not “express a propo-
sition” or carry a truth-value (depending on the relevance of determining the values
of the referential variables to the context of evaluation). The semantic interpretation
of a statement or assertion will be sufficiently determined and, in addition, determi-
nate so that it does identify the proposition expressed by the utterance and can be
assigned a truthvalue. Truth-conditional semantics is for the semantic interpretations
of utterance-types or -tokens, not for the semantic representations of sentence-types
or -tokens.
Classical Gricean pragmatics takes a semantic interpretation of an utterance and
its context and generates a pragmatic interpretation of the utterance in the context,
something intuitively described as “the addressee’s understanding of the speaker’s
intended meaning,” by combining the semantic interpretation of the assertion with the
asserter’s conversational implicata: something the addressee understands the speaker
to convey, suggest, or imply by, in, or when asserting the sentence-token in the con-
text. The Atlas (1979) version of post-Gricean pragmatics allows pragmatic inference
to map (a) semantic representations of sentences into semantic interpretations of utter-
ances of the sentences and (b) semantic interpretations of assertions (statements) of
the sentences into further semantic interpretations, which are further interpretations by
the addressee—what the addressee understands the speaker to have communicated in
making the asertion in that context. The latter interpretations are the classical Gricean
conversational implicata. Thus, the architecture of the theory is this:
1. Internalist sentence semantics. Sentence strings, well-formed

sequences of lexical items, are meaningful; they have semantic
representations.
2. Externalist statement semantics. Utterances of sentence-string-tokens
will, when nonspecific items have specific interpretations, when
vagueness is sharpened, and when variables have values, have well-
defined truth conditions in a context. An utterance with a well-
defined truth condition in a context will be said to “express a

proposition.” The truth condition of an utterance in the context is
described by the semantic interpretation. Some utterances are
assertions (statements).
3. Externalist speaker/addressee pragmatics. A speaker, in, by, or
when asserting a sentence-string-token, will convey, suggest, or
imply to an addressee further propositions [implicata]. The contents
of the assertion and of the implicata constitute the “total significa-
tion” of the speaker’s utterance for the addressee.
Psycholinguistically the picture is this: the mental realization of language L,

L = <Syntax, Vocabulary>, generates φ. Speaker S asserts utterance U, a phonetic
realization of φ. Addressee A interprets Uφ in the context K in order to know what
assertion S has made and what else S conveys by making the assertion. This two-
leveled semantics—internalist and truth-conditional—is an essential feature of the
Atlas (1979) version of post-Gricean pragmatics. (For discussion of these three lev-
els of representation and interpretation, see Atlas 1989, Levinson 1995, and Lyons
1995b). Chomsky (1996c, 2000) limits his semantics to Internalist Semantics of
the syntactically well-formed strings and relegates both truth-conditional seman-
tics and Gricean implicatures of utterances to the pragmatics of Performance Sys-
tems. This difference between me and Chomsky is merely terminological, but a
parallel difference between me and Jerry Fodor is not merely terminological (see
section 2).
In section 1, “Metaphor, Nonspecific Meaning, and Utterance Interpretation,” I
took a classical problem, that of metaphor, and approached it from the point of view
of my Internalist Semantics notion of semantical nonspecificity (with respect to a
semantical property). That section provided a gentle introduction to the notion of
semantical nonspecificity in an application to an old and familiar problem.
In section 2 “Semantical Nonspecificity, Utterance Interpretation, and Psycho-
logical Modularity,” I discussed the philosophical consequences of my semantical
and post-Gricean pragmatic theory for Fodor’s account of utterance-interpretation. I
posed the problem of utterance-interpretation in a new and, I suggested, conceptu-
ally more satisfactory way. Unlike me, Fodor thinks that language is an exception in
the realm of intentional action—that what “is usually required in the intentional analy-
sis of behavior is a kind of hermeneutic sophistication that’s as far as can be from the
execution of a rote procedure. The notable exception is inferring intentional content
from utterance form” (1990b: 213). On my view, though hermeneutic sophistication
is, as Fodor notes, normally a symptom of an analysis of intentional acts whose re-
sults are never “all likelihood,” “enormously reliable” ones, the challenge, rather, is
to explain how utterance-interpretation can be both hermeneutic and enormously
reliable.
That is the challenge that has directed the construction of the accounts discussed
in Atlas (1979, 1989) and in the present work. In my view, a satisfactory explanation
of utterance interpretation requires the recognition of two linguistic phenomena:
semantical nonspecificity of sentence and predicate expressions and generalized in-
terpretative inferences, either from semantic representations or from assertions.
Without semantical nonspecificity of the sentences, generalized conversational in-

terpretative inferences are “blind”; and without the contents of generalized interpre-
tative inferenda, the contents of semantically nonspecific sentences are “empty.”
Together they yield by hermeneutic means a reliable interpretation of a speaker’s
“meaning” in asserting a sentence. That is the post-Gricean thesis stated in Atlas
(1979), Atlas (1989: chap. 4) and developed further in this volume.
In chapter 2, “Grice’s Theory of Conversational Inference,” I expound on Grice’s
views, selectively and critically, from his first publication on the subject in 1961 to
his last remarks from his collected essays Studies in the Way of Words of 1989. My
treatment is not encyclopaedic or exhaustive. My focus is on a critical evaluation of
Grice’s views. It has been clear to me at least since Atlas (1975a,b, 1977a,b, 1978a,b,
1979) that assumptions of the classical, 1967 version of Grice’s account in the Wil-
liam James lectures in Harvard University would require revision in a radical way,
for two types of reasons:
1. Since the literal meanings, and so the semantic representations, of

many sentences being asserted are semantically nonspecific (for
some semantical feature) and thus in the philosophers’ sense
nonpropositions (not articulations of truth conditions and not bearers
of a determinate truth or falsity), independently even of deixis,
reference-fixing, tense, and so on, “what is said” is not a simple
function of context and literal meaning, as I demonstrated in Atlas
(1978a,b, 1979).
2. Grice’s Maxims of Conversation and his “idealized model” or
“theory” of conversational implicature entail inadequate taxonomies of
inference and are insufficiently constrained to avoid “overgenerating”
implicata that are logically contrary or contradictory. The post-Gricean
account expounded here is intended to remedy the descriptive inade-
quacy in Grice’s and the neo-Griceans’ account, without despairing of
Grice’s program entirely (cf. the enthusiastic attempt at “debunking”
Grice in Davis (1998), who throws out the baby with the tub and keeps
the dirty bathwater) and without begging further questions of the
account’s explanatory adequacy.
On the first problem of semantical nonspecificity, negative, definite description

sentences are a notorious case in point (see Horn 1989). As I argued in Atlas (1974,
1975b, 1977b, 1978a,b, 1979, 1989), pragmatic inference is antecedently required
even to construct a truth-value bearing interpretation (a “proposition,” Grice’s “what
is said”) from the literal, semantically nonspecific meaning of a negative sentence
uttered. Noam Chomsky recently suggested that “we cannot assume that statements
(let alone sentences) have truthconditions. At most they can have something more
complex: ‘truth indications’ in some sense” (1996c: 52). He added:
There is no question of how human languages represent the world, or the world as
it is thought to be. They don’t. . . . There is no reference-based semantics. . . . There
is a rich and intriguing internalist semantics, really part of syntax, on a par in this
respect with phonology. Both systems provide ‘instructions’ for performance sys-
tems, which use them . . . for articulation, interpretation, inquiry, expression of
thought, and various forms of human interaction. (Chomsky 1996c: 53)
I had written in Atlas (1979: 278 n.2), “The sentence The A is not B in one context
may be understood as L– [an external, exclusion negation] and in another as L+ [an
internal, choice negation]. These understandings [not senses] are related to the [lit-
eral] meaning . . . of the sentence as allophones are to the phoneme to which they
belong.” I had been engaged in a study in Chomskyan Internalist Semantics.
In chapter 2 I also discuss Grice’s (1989a) and A. P. Martinich’s (1980) views
on the role of Relevance in Grice’s account, arguing that Relevance cannot subsume
Grice’s Second Maxim of Quantity or Atlas and Levinson’s (1981) Maxims of Rela-
tivity and Informativeness. I show that the phenomena of hyperbole present obstacles
to Horn’s (1984b) and Sperber and Wilson’s (1986b) attempts to reduce Informa-
tiveness to Relevance. I conclude by discussing the central philosophical problem of
Grice’s pragmatics: his reduction of the semantics of utterances to the epistemology
of mental states.
In chapter 3 “The Rise of Neo-Gricean Pragmatics,” I discuss the project of Atlas
(1978b, 1979), Atlas and Levinson (1981), Horn (1984b, 1989, 1993, 1996b), Levin-
son (1987a,b, 1988a,b; 1991, 2000), and Huang (1994, 2000). We revise Grice’s
original account of the Maxims of Conversation so that both speaker-centered maxims
of production and addressee-centered maxims of comprehension (interpretation) are
formulated as part of a complete account (see Blutner 2000). Horn (1984b, 1989,
1993) has suggested a simplifying dualistic schema of Grice’s maxims—two anti-
thetical categories of informational adequacy, Q (Quantity), and economy, R (Rele-
vance)—the speaker saying enough for successful uptake by his addressee (Q) but
saying no more than is needed (R).22 Levinson (1987b, 1988a,b; 1991, 2000) and
Huang (1994, 2000) have developed and applied the theory to anaphora in English
and Chinese.
In chapter 4, “The Post-Gricean Theory of Presupposition,” I review the reduc-
tion of Strawsonian and Fregean presupposition to entailment and Gricean implicata,
comment on the inadequacies of Stalnaker’s (1974, 1999) notion of pragmatic pre-
supposition, and discuss the flaws in Grice’s (1981) attempted reduction.
In chapter 5, “Assertibility Conditions, Implicature, and the Question of Semantic
Holism,” I examine several theoretical problems in specifying the logical forms, truth-
conditions, and implicata of English sentences containing comparative adjectives,
equatives, adverbials of degree, and adverbial “approximatives” (e.g., almost). The
range of linguistic data explained and the depth of semantic analysis achieved is, I
believe, not to be found in competing theories.23 I also derive some surprising con-
sequences for the philosophical dispute concerning “meaning holism.”
22Horn (1984b: 17) notes: “The most detailed and careful discussion in the literature of Q vs. R clashes
in English is due to Atlas and Levinson (1981) (cf. Levinson 1983: Section 3.2 for related discussion).”
23Manfred Bierwisch (1989: 129) remarks of the analysis of comparatives and equatives of Atlas (1984a)
and in this book, “The most careful account of the relevant facts and relations concerning the com-
parative and the equative is given in Atlas (1984[a]).”
In the final chapter 6, “The Third Linguistic Turn and the Inscrutability of Lit-
eral Sense,” I examine a controversial question in the study of the semantics-
pragmatics interface: What does the English numerical adjective ‘three’ actually
mean? I reach the controversial conclusion that the meaning of numerical adjectives
is highly underdeterminate, and I introduce a Context Principle for Chomskyan
Internalist Semantics: only in the context of a noun phrase does a numerical adjec-
tive have a determinate meaning. I attempt to put post-Gricean pragmatics in histori-
cal perspective and consider its consequences for a philosophical theory of meaning
and for Chomskyan internalist semantics.
2
Grice’s Theory of
Conversational Inference
A Critical Exposition
1 The third linguistic turn
Thirty-five years ago Richard Rorty (1992 [1967]) published an anthology of essays
on the philosophical methods of the logical empiricists in the interwar years and of
the “ordinary language” philosophers in prewar and postwar Oxford, an anthology
entitled The Linguistic Turn. The phrase has caught on. But there has been a third
linguistic turn in philosophy—less dramatic than that of the early Wittgenstein’s
Tractatus Logico-Philosophicus and more systematic than the nuanced analysis of
language characteristic of J. L. Austin’s essay “A Plea for Excuses” (1956–57). It
has not been as ideological as either of the first two; it has been more tentative in its
claims, more sophisticated in its methodology, more sensitive to the demands of theory
construction. Its tutelary deities have been Noam Chomsky and W. V. O. Quine; its
progenitors Donald Davidson, Zeno Vendler, H. Paul Grice, and Robert Fogelin; and
its charter the texts of Fogelin (1967), Evidence and Meaning; Grice (1967), “Logic
and Conversation: The 1967 William James Lectures”; Vendler (1967), Linguistics
in Philosophy, Davidson (1967), “The Logical Form of Action Sentences”; Davidson
and Harman, eds. (1972), Semantics of Natural Language; and Davidson and Harman,
eds. (1975), The Logic of Grammar.
The Gricean strand of this development began with P. H. Nowell-Smith’s con-
cept of “contextual implication” in his 1954 book Ethics; it emerged with Paul Grice’s
notion of “conversational implication” from his 1961 essay on the causal theory of
perception, and it matured in Grice’s Maxims of Conversation of his 1967 William
James lectures “Logic and Conversation” and in Robert Fogelin’s “Rule of Strength”
45
from his 1967 Evidence and Meaning. Patrick Nowell-Smith and Paul Grice were
part of J. L. Austin’s Saturday morning meetings in Oxford in the 1950s. Fogelin
expresses his intellectual indebtedness to both Austin and Nowell-Smith. What dis-
tinguishes Grice’s approach is his emphasis on “the radical importance of distinguish-
ing (to speak loosely) what our words say or imply from what we in uttering them
imply; a distinction seemingly denied by Wittgenstein, and all too frequently ignored
by Austin” (Grice 1986: 59).
By his own testimony, Grice (1986: 59) was influenced by Quine’s work, which
“helped to throw light on the problem of deciding what kind of thing a suitable theory
[of linguistic phenomena of the kind with which in Oxford we had long been con-
cerned] would be, and also by his example exhibited the virtues of a strong method-
ology”; he added that “Quine’s influence on me was that of a model: I was never
drawn towards the acceptance either of his actual methodology or of his specific
philosophical positions” (1986: 60). Geoffrey Warnock (1973) reports that Austin’s
Saturday Morning Meeting group read Chomsky’s (1957) Syntactic Structures in the
Michaelmas term of 1959. Grice (1986: 60) describes Quine and Chomsky as his
“chief theoretical mentors.” He then describes his work of the 1960s as “principally
. . . an attempt to show, in a constructive way, that grammar (the grammar of ordi-
nary discourse) could be regarded as, in Russell’s words, a pretty good guide to logi-
cal form, or to a suitable representation of logical form” (1986: 60). For Grice this
was principally “the suggestion of notational devices together with sketchy indica-
tions of the laws or principles to be looked for in a system incorporating these de-
vices; the object of the exercise being to seek out hitherto unrecognized analogies
and to attain new levels of generality” (1986: 60).
Grice’s notational devices—for example, his square brackets—of the 1967
William James lectures and of the 1970 University of Illinois lecture “Presupposi-
tion and Conversational Implicature” (Grice 1981), have seemed to me unsuccessful
in reaching his objective, as I discussed in my essay “On Presupposing” (Atlas 1978b).
Ironically, given his goal, his informal discussion of the Maxims of Conversation
have been more adapatable to systematic theoretical linguistic explanations than were
his syntactical innovations. The syntactic way in which grammatical form could be
a guide to logical form was not successfully demonstrated by Grice, but the Grice-
Strawson Condition (Strawson 1954, 1964) that statements carry presuppositions of
the existence of a noun phrase designation only if the NPs are topic NPs has proven
fruitful in the linguistic and philosophical analysis of negative existence statements
(Atlas 1988, 1989) and justified Grice’s confidence in Russell’s Theory of Definite
Descriptions. A development of this idea in my Focal Noun Phrase Limitation Prin-
ciple (Atlas 1991a) determined the choice of logical subject NPs in logical forms for
cleft statements (Atlas and Levinson 1981; appendix 3 here) and for statements of
the surface form Only Tom VP, where ‘VP’ is a metavariable taking as values verb
phrases (Atlas 1991a, 1993, 1996b, 1997a, 2001).
Grice’s interest in and emphasis on the logical form of ordinary language sen-
tences parallel contemporaneous interests of Quine’s in chapter 4 (“Vagaries of Ref-
erence”) and chapter 5 (“Regimentation”) of Quine’s 1960 Word and Object; of
Anthony Kenny’s in his 1963 Action, Emotion, and Will; and of Davidson’s and
GRICE ’S THEORY OF CONVERSATIONAL INFERENCE 47
Vendler’s in their individual essays of 1967. Vendler self-consciously notes in his

preface to Linguistics in Philosophy that his essays represent “the gradual introduc-
tion of a new technique into analytic philosophy,” the point of which is “to show that
the more or less sophisticated data provided by structural linguistics can be used in
philosophical arguments” (1967: vii–viii).
Grice’s project—like other projects of the Third Linguistic Turn, and unlike that
of the Tractatus Logico-Philosophicus and logical empiricism or that of the later
Wittgenstein, Ryle, and Austin—is characterized by the conscious selection from,
or creation of, concepts in Chomskyan theoretical linguistics whose appropriation,
development, and application to problems in philosophy proper can reconceptualize,
and so overcome, the stalemates of competing, traditional philosophical “solutions.”
Cases in point, in addition to the Russell-Strawson debate on definite descriptions,
negation, and presupposition (Atlas 1977b, 1978a,b, 1989) and the question whether
‘ought’ logically implies ‘can’ (Sinnott-Armstrong 1984)1, are (a) the meaning of
causal statements (Vendler 1967), (b) the logic and linguistics of negative existence
statements (Atlas 1988), (c) the adequacy of propositional attitude sentences as analy-
ses of intentional locutions (Searle 1983; Marti 1993), (d) the semantics of natural
kind terms (Atlas 1980b, 1989; Chomsky 1995b, 2000: 148–53; Donnellan 1983,
1993), and (e) the reference of singular terms (Bach 1987). In Grice’s view, his in-
vestigation of “logic and conversation” was a prolegomenon to any future metaphysic
(Grice 1986: 53, 59). That is the larger philosophical significance of Grice’s account
of “logic and conversation.”
But it has been Grice’s concept of conversational implicature, and for philosophi-
cal purposes his notion of a generalized conversational implicature, that has been a
focus of both philosophical and linguistic discussion.2 As Grice emphasized, the
importance of the notion was its role in distinguishing what our words say or imply
from what we in uttering them imply. This distinction and its ramifications took Grice
beyond Wittgenstein and Austin, as he notes (Grice 1986: 59).3 It also significantly
influenced debates in linguistic theory on the semantical reductions of pragmatic facts,
1See Atlas (1974, 1975a,b, 1977b, 1978b), Bach (1987), Boër and Lycan (1976), Grice (1981), Russell
(1905, 1919, 1959), Sinnott-Armstrong (1984), Strawson (1950, 1952, 1964), and White (1993).
2Generalized conversational implicata are contents conveyed, suggested, or implied by a speaker in,
when, or by asserting a sentence, as a default, standard, or normal feature of the asserting of the sen-
tence in any and every speech context. Particularized conversational implicata, by contrast, are nonce
implications by a speaker that are implied by virtue of the sentence’s being asserted in a particular
speech context.
3Grice himself thought that his theory of implicature would cast light on problems of perception, knowl-
edge, and modality so as to correct what he saw as a mistaken and unreflective Wittgensteinian formu-
lation of these problems popular in the 1950s and 1960s among British philosophers (Grice 1961, 1986:
66). As P. M. S. Hacker remarks:
Grice’s distinction between the meaning of an expression and its conversational implicatures
can be brought to bear upon Wittgenstein’s accounts of the meaning of various expressions
in terms of use. For if Grice’s argument is correct, then Wittgenstein attributed features of
the use of expressions to their meaning, which are correctly ascribable not to their meaning
but to pragmatic principles of discourse. (Hacker 1996: 245)
illustrated by Gordon and Lakoff’s highly influential 1971 paper “Conversational

Postulates” and by the seminal Horn (1972).4 To quote R. A. Harris:
Ordinary language philosophy grew more important for generative semantics in the
seventies. Beginning with [J. R.] Ross’s fairly direct importation of Austin’s in-
sights about performatives [Ross 1970], which [Jerrold] Sadock and [Alice] Davison
took up at Chicago under [James] McCawley, it gained considerable momentum
under the influence that philosopher H. P. Grice’s [1967] conversational implicature
work had on both Lakoffs [Robin and George], on their students, and on [James]
McCawley. It is from this period that linguists began to develop a sense of some-
thing they called pragmatics, as distinct from what [is] called semantics. (Harris
1993: 185)
A decade later, in 1983, Geoffrey Leech published his Principles of Pragmatics

and Stephen Levinson published his survey Pragmatics. The latter book summarized
the extraordinary amount of creative work in philosophy of language and linguistics
that began in 1969 with Searle’s Speech Acts and in 1970 with Ross’s “On Declara-
tive Sentences” and that continued through the 1970s. Levinson chose to discuss
deixis, conversational implicature, presupposition, speech acts, and conversational
structure, establishing a curriculum that continues to influence textbooks years later—
for example, Green (1989), Grundy (1995), and Thomas (1995).
2 The language of perception
In 1961, in an essay in the Proceedings of the Aristotelian Society, H. Paul Grice,

then a tutorial fellow of St. John’s College, Oxford, addressed a problem in the theory
of knowledge that had exercised distinguished Oxford philosophers, in particular
H. H. Price in his 1932 book Perception. According to Price’s account, if one is per-
ceiving a material object M, the statement so describing one’s experience, namely
‘M is present to my senses’, is equivalent to ‘M causes a sense-datum with which I
am acquainted’, and “perceptual awareness” amounts to an inference from a psy-
chological effect to its material cause. Grice observed that since the phrase ‘present
to my senses’ was used by Price for one sense of the verb ‘perceive’, Price’s account
of the causal theory of perception gives it that (1a) and (1b) are equivalent:
(1) a. I am perceiving M.
b. I am having (or sensing) a sense-datum which is caused by M.
This talk of ‘sense-datum’ and ‘acquaintance’ is resonant with reminders of the 1910–
11 lectures of G. E. Moore (Some Main Problems of Philosophy, 1953) and Bertrand
Russell’s “shilling shocker,” Problems of Philosophy (1913). By the 1950s and 1960s,
4Gordon and Lakoff’s position was criticized in J. L. Morgan’s 1977 “Conversational Postulates
Revisited” and in Gazdar’s 1979a Pragmatics: Implicature, Presupposition, and Logical Form (see
Newmeyer 1996: 139–40).
Oxford philosophers were concerned to show how the philosophical term of art ‘sense-
datum’ should be understood, and thus to show what sense, if any, could be made of
the claims of a causal theory of perception. The philosophical statement in (2a) would
be explained by the ordinary statement in (2b) or by (2c):
(2) a. I am sensing a red sense-datum.

b. It looks red to me.
c. I seem to see something red.
In the Oxford of the late 1950s (at the time his essay appeared Grice was forty-eight
years old) a sense-datum theory of perception of this kind was under attack.5 What
interested Grice was the type of this attack, which he described as follows:
When someone makes such a remark as “It looks red to me” a certain implication is
carried. . . . It is implied either that the object referred to is known or believed by
the speaker not to be red, or that it has been denied by someone else to be red, or
that the speaker is doubtful whether it is red, or that someone else has expressed
doubt whether it is red, or that the situation is such that though no doubt has actu-
ally been expressed and no denial has actually been made, some person or other
might feel inclined towards denial or doubt if he were to address himself to the
question whether the object is actually red. . . . Let us refer to the condition which
is fulfilled when one or other of the limbs of this disjunction is true as the D-or-D
condition (‘doubt or denial’ condition). (Grice 1965: 441)
Grice then admits that “there would be something at least prima facie odd about my
saying ‘That looks red to me’ (not as a joke) when I am confronted by a British pillar
box in normal daylight at a range of a few feet.” But the critic’s thesis goes beyond
this, as Grice observes:
(a) that it is a feature of the use, perhaps of the meaning, of such locutions as “looks
to me” that they should carry the implication that the D-or-D condition is fulfilled,
and that if they were uttered by a speaker who did not suppose this condition was
fulfilled he would be guilty of a misuse of the locutions in question . . . , (b) that in
cases where the D-or-D condition is unfulfilled the utterance employing the “looks
to me” locution, so far from being uninterestingly true, is neither true nor false. (Grice
1965: 442)
But then, as Grice observes, the critic can close his attack upon the sense-datum
theorist:
The sense-datum theorist wants his sense-datum statements to be such that some
one or more of them is true whenever a perceptual statement is true; for he wants to
go on to give a general analysis of perceptual statements in terms of the notion of
5Grice (1986: 62) remarked, “I have never been very happy about Austin’s Sense and Sensibilia, partly
because the philosophy which it contains does not seem to me to be, for the most part, of the highest
quality, but more because its tone is frequently rather unpleasant.”
sense-data. But this goal must be unattainable if “looks to me” statements (and so
sense-datum statements) can be truly made only in the less straightforward percep-
tual situations [since in the straightforward situations, ones without doubt or con-
troversy, the “looks to me” idiom would be out of order and neither true nor false];
and if the goal is unattainable the [Causal Theory of Perception] collapses. (Grice
1965: 442)
Since Grice, in his essay, wishes to defend his own version of the Causal Theory
of Perception, he was forced to consider the strategy of the critic that he has
described.
He mentions a famous, parallel case, one that is both intellectually important for
Grice and of historical interest as well. Suppose that one asserted (anomalously):
(3) ??It is raining, but I don’t believe that it is raining.
What is it, G. E. Moore (1968: 535–44) had asked, that is peculiar about this state-
ment? Clearly there is something odd about the statement. The oddity might be ex-
plained if there were an “implication,” ‘I believe that it is raining’, that was “part of
the meaning” of ‘It is raining’, for then statement (3) would be logically contradic-
tory. But it seems clear that such an explanation is mistaken. ‘I believe that it is rain-
ing’ cannot be, literally, part of the meaning of ‘It is raining’; the oddity seems to
arise because of what “asserting” a sentence consists in. Sentence (3) is, literally,
logically consistent: it is possible for it to be raining and for the speaker not to be-
lieve that it is.
Grice’s diagnosis of his philosophical opponent’s thesis follows a similar pattern:
If I were to say ‘it looks red to me’ in a situation in which the D-or-D condition is
not fulfilled, what I say is . . . true, not “neuter”; while admitting that though true it
might be very misleading and that its truth might be very boring and its
misleadingness very important, one might still hold that its suggestio falsi is per-
fectly compatible with its literal truth.
Furthermore one might argue that though perhaps someone who, without in-
tent to deceive, employed the ‘it looks to me’ locution when he did not suppose the
D-or-D condition to be fulfilled would be guilty in some sense of a misuse of lan-
guage, he could be said not to be guilty of a misuse of the particular locution in
question; for, one might say, the implication of the fulfillment of the D-or-D con-
dition attaches to such locutions not as a special feature of the meaning or use of
these expressions, but in virtue of a general feature or principle of the use of lan-
guage. (Grice 1965: 442)
To show that the critic of the Causal Theory of Perception was mistaken, Grice
undertakes an examination of the concept of implication that figures centrally in the
dispute.
Grice (1965: 444–45, 448–49) distinguishes four cases:
(4) a. Smith has left off beating his wife.

b. She was poor but she was honest.
c. Jones has beautiful handwriting and his English is grammatical.

d. My wife is either in the kitchen or in the bedroom.
In (a) “what is implied is that Smith has been beating his wife.” In (b) “what is im-
plied is (very roughly) that there is some contrast between poverty and honesty, or
between her poverty and her honesty.” In (c) “I am reporting on a pupil at Collec-
tions. All I say is ‘Jones has beautiful handwriting and his English is grammatical’.
We might perhaps agree that there would here be a strong, even overwhelming, im-
plication that Jones is no good at philosophy.” In (d) “it would normally be implied
that he did not know in which of the two rooms she was.”
Of these examples Grice notes that (a) is a traditional case of “presupposition”
in Grice’s pupil’s P. F. Strawson’s (1950, 1952, 1964) sense: the truth of what is
“implied” is a necessary condition of the statement’s being either true or false, while
(b) could be false, rather than neither true nor false, whether or not there was a con-
trast between honesty and poverty, if she were poor and dishonest. Grice notes that
“what is said (or asserted)” in (a) is the source of the implication, but “what is said
(or asserted)” in (b) is not. He also observes that in both (a) and (b), the speaker implied
whatever is implied; that in (b) but not (a) the speaker’s words, in a precise sense,
imply whatever is implied; and that in neither (a) nor (b) “would it be evidently ap-
propriate to speak of his saying that, or of his saying that in that way, as implying
what is implied” (1965: 446).
Also in discussing examples (a) and (b), Grice (1965: 446) introduces the no-
tions of the detachability and cancelability of the “implication.” Since these notions
become important in the development of Grice’s views, it is of interest to consider
the first account that Grice gives. Of detachability in example (4a) Grice writes:
One cannot find a form of words which could be used to state or assert just what the
sentence ‘Smith has left off beating his wife’ might be used to assert such that when
it is used the implication that Smith has been beating his wife is just absent. Any
way of asserting what is asserted . . . involves the implication in question. I shall
express this fact by saying that in the case of (a) the implication is not detachable
from what is asserted (or simpliciter, is not detachable). (1965: 446–47)
And then of noncancelability Grice writes:
One cannot take a form of words for which both what is asserted and what is im-
plied is the same as for (a), and then add a further clause withholding commitment
from what would otherwise be implied, with the idea of annulling the implication
without annulling the assertion. One cannot intelligibly say ‘Smith has left off beat-
ing his wife but I do not mean to imply that he has been beating her’. I shall express
this fact by saying that in the case of (a) the implication is not cancellable (without
cancelling the assertion). (Grice 1965: 446–47)
By contrast, the implication of (b) is, in Grice’s view, detachable because “what
is asserted” is the same as “what is asserted” in She is poor and she is honest, but the
implication is not as resistant to cancelation as that of (a) is. Grice does not go so far
as to say that a denial of the implication of (b) is “unintelligible,” but he admits that
saying ‘She is poor but she is honest, though of course I do not mean to imply that
there is any contrast between poverty and honest’ is a “peculiar” way of conveying
that she was poor and honest. As I have pointed out in Atlas and Levinson (1981:
20), Grice (1961), unlike Karttunen and Peters (1979), distinguished between “pre-
supposition” and what he later called “conventional implicature” by saying, as in the
case of (a) and (b), that while the conventional implicatum was detachable and, to a
high degree, not cancelable, the presupposition was not detachable and not cancel-
able. Finally, Grice recognizes that the implication in (b) is attributable to the mean-
ing of a particular word: ‘but’.6
When Grice considers examples (d) and (c), he observes, interestingly, that in
asserting P or Q the implication that the speaker does not know whether P or
whether Q is a default implication: the implication is standardly, or normally, attrib-
utable to the speaker, and to the speaker’s asserting the sentence rather than some
other sentence, but there are contexts in which the implication would not be carried.
The default implication is ceteris paribus nondetachable. It is also cancelable: it is
intelligible and logically consistent to say ‘My wife is either in the kitchen or in the
bedroom; mind you, I’m not saying that I don’t know which’. Likewise the implica-
tion of (c) is cancelable: ‘Jones has beautiful handwriting and his English is gram-
matical; I do not of course mean to imply that he is no good at philosophy’. But the
implication is context-dependent: it is not standardly involved in the assertion of the
sentence, and it is not a default implication—a special context is required to attach
the implication to the utterance—for example, that it be uttered at Collections (stu-
dent evaluations) in an Oxford college.
It was important to Grice to distinguish the case of ‘or’ from the case of ‘but’.
The implication of ‘but’ is attributable solely to its lexical meaning. Grice (1965:
450) wished to argue that the implication of ‘or’ is not attributable solely to its lexical
meaning but to a “general principle governing the use of language.” His “first shot,”
as he put it, at formulating the principle was: One should not make a weaker state-
ment rather than a stronger one unless there is a good reason for so doing. On
the explicit assumptions that P is “stronger” than P or Q because P entails
P or Q and not conversely, and that an “obvious” reason for not making a state-
ment is lack of justification, warrant, or adequate evidence, Grice suggests that his
proposed principle of language use will explain the implication of the assertion
P or Q.
The form the Gricean argument takes must be something like this:
1. S asserted Ψ or χ.
2. S did not assert the logically stronger Ψ or assert the logically stronger χ.
6It is notable that in Grice’s (1965: 446–47) argument for the non-cancelability of the presupposition,
he only considers the cancelability of the presupposed sentence for the affirmative statement Smith
has left off beating his wife. Strawson (1950) had claimed that presuppositions were preserved under
negation, so Grice should have considered the implication of the negative Smith has not left off beat-
ing his wife as well. He does finally do so in Grice (1981), with the result that he concludes that
Strawsonian presuppositions are not detachable but are, in fact, cancelable. See chapter 4 in this vol-
ume for more on this subject.
3. A good reason for not asserting a statement φ is not having adequate

grounds for asserting φ.
4. So the speaker, it is presumed, does not have adequate grounds for
asserting Ψ and does not have adequate grounds for asserting χ.
5. If so, then the speaker does not know Ψ, and the speaker does not
know χ.
6. So the speaker means to imply in asserting Ψ or χ that the speaker
does not know that Ψ and does not know that χ.
Not only is it assumed in this form of argument that a speaker is capable of having
and of recognizing good reasons for his or her actions, it is assumed by Grice that
what counts as a reason for the action of making a statement—actually, in this case,
an inaction: not making the statement Ψ—is the same as what the speaker means to
imply in performing the alternative action by asserting the disjunctive sentence Ψ
or χ. But this assumption on Grice’s part is gratuitous, as I shall discuss further later
in this chapter.
It follows from Grice’s claim about (4d), that the speaker does not know in which
of the two rooms his wife was; that in asserting P or Q, the speaker does not know
that P and the speaker does not know that Q [¬KsP & ¬KsQ]. So ¬P is compatible
with all that speaker S knows, and ¬Q is compatible with all that speaker S knows.
But when one considers assertions of logical disjunctions in mathematical reason-
ing, a paradigmatic use of a disjunction is to prove a theorem θ by deducing it from
a disjunction, from both a statement φ and its contradictory ¬φ, since whatever is
a logical consequence of the necessary truth φ ∨ ¬φ is itself a necessary truth. The
advantage of such a proof is that one need not know whether φ is a mathematical
truth or whether ¬φ is a mathematical truth. One may, of course, use this proof
technique even when one does know which is true.
When I assert (R) My car keys are either in my pants pocket or on my desk, it
is true that an explanation for my act of asserting (R) may appeal to my not know-
ing which, but I do not think that I should be interpreted to imply, convey, suggest,
or mean by, in, or when asserting the disjunctive sentence (R) that I do not know
which, anymore than in asserting It’s raining Moore actually intended to convey,
suggest, or imply in, when, or by asserting that sentence that he believed that it
was raining, so that the total signification of the utterance would be [IT’S RAINING &
I BELIEVE IT], even if the best explanation for Moore’s act of asserting ‘It’s raining’
was that he believed that it was (see section 5 in this chapter). Thus I disagree fun-
damentally with Grice (1965, 1967, 1975a) and especially with linguists Gazdar
(1979a) and Levinson (1983, 2000), who construct a theory of “clausal” implicature
to explain this misguided intuition about or sentences (see chapter 3, section 2).
Grice’s view here seems just a lapse, as he later appeals to Moore’s Paradox sen-
tences—for example, It’s raining and I do not believe it, to distinguish between
what a speaker “implicates” and what a speaker “expresses.” In asserting U a speaker
S expresses his belief that U is true; he does not implicate that he believes U.
A speaker can assert You can/may have chocolate or vanilla and not implicate
that the speaker does not know which you can have. In fact, the speaker implicates
what is not entailed: you can/may have chocolate, and you can/may have vanilla.
You just cannot have them both.
Possible (P or Q) Possible P & Possible Q

“Possible (P or Q)” » Possible P & Possible Q
“Possible (P or Q)” » ¬Possible (P & Q)
Similarly for You can go to Grandma’s or you can go to the movies. In S’s offer of
a choice of alternatives, S is normally in a position to assert both limbs of the dis-
junct, which is why S is in a position to assert the disjunction. (Shades of Mathematical
Intuitionism!)
When I say pointedly to Jason Linder of a “choice” of dessert at my dinner table:
(T) You may have tartufo or you may have tartufo, I am not in doubt as to which he
may have, nor do I convey in or by asserting the disjunction that I do not know which
he may have. Such alleged implicata do not arise as generalized conversational im-
plicata from sentential ‘or’ as features of the meaning of P or Q, at least in the
case P or P. The classical Gricean theorist may riposte that my example merely
illustrates the intervening role of the Maxims of Manner: P or P is a circumlocu-
tion for P, but an arch or joke one, so one should not expect the usual conversa-
tional implications to be produced. But the mistake is already made: P or P is not
a circumlocution for P. It is certainly the case that one allegedly typical conversa-
tional implication but not both of asserting a disjunction P or Q, such as You may
not have both an eclair and pie in asserting You may have an eclair or pie, would not
be produced in asserting P or P, namely, You may not have both the tartufo and
the tartufo—that is, You may not have the tartufo. The modest point of the or state-
ment was to offer the tartufo or nothing, not to imply conversationally that Jason
could have no tartufo at all.
Since 1973 I have been pointing out the problem of conflating two claims: (a)
“P or Q” » ¬(P and Q) and (b) a speaker S’s asserting P or Q is explained by
¬KsP and ¬KsQ. The first claim is about the implicata of asserting ‘or’. The sec-
ond claim is purportedly explanatory of the behavior of asserters of ‘or’. Claim (b)
does not entail (a), and (a) is false as a general claim about ‘or’. Rather, English
‘or’ is semantically nonspecific (rather than ambiguous) between inclusive disjunc-
tion and exclusive disjunction. By Atlas and Levinson’s (1981) inference to the
best interpretation (discussed in chapter 3) the form P or Q is interpreted by de-
fault as exclusive disjunction, unless blocked—for example, by the use of ‘and/
or’, by the different form P or P, or by collateral contextual information (Fowler
1965: 422). This interpretation entails ¬(P & Q). Thus the alleged scalar
implicature from or to not and is actually another Atlas (1984a) hybrid: implicature-
composed-with-entailment inference. Thus the correct claim is (a') “P or Q” °
» ¬(P & Q). Both the ambiguity theorists and the classical Griceans were mis-
taken about ‘or’, but the ink continues to run (see chapter 5, section 1). I do not,
unlike Grice in 1961, think that (b) entails that asserting the or statement impli-
cates that the speaker does not know which alternative holds. But I shall pretend
otherwise in the rest of this section, for the sake of Grice’s argument. It seems to
me that to this day the phenomena of clausal implicatures have been misdescribed
and misunderstood. The theoretical situation is nearly as bad as the one for En-
glish conditional if . . . then sentences (see Sanford 2003).
The philosophical interest of the semantic distinction between ‘but’ and ‘or’,
let us recall, is an issue in the philosophical theory of perception. The critic of the
causal theorist’s explanation of sense-datum statements as statements using ‘looks
to me’ or ‘seems to me’ claimed that a ‘looks to me’ statement implied that there
was, or naturally could be, doubt or denial that an object is the way it looks. Since
ordinary statements describing the perception of observable properties of material
objects are not cases in which there is normally a doubt or denial, the critic rejects
the account of ‘looks to me’ statements, and so the account of sense-datum state-
ments, that the causal theorist requires. Furthermore, the objections are (a) that the
doubt-or-denial condition is a semantic “presupposition” of a ‘looks to me’ per-
ceptual report and (b) that the occurrence of the doubt-or-denial implication is at-
tributable, according to the critic, to the presence of the words ‘looks to me’ or
‘seems to me’.
These two features are philosophically important in the debate because they
allow the critic to claim that the causal theorist is committed to statements that, on
the theorist’s own view, are, in general, neither true nor false; since such statements
are obviously true or false, in adopting ‘looks to me’ language, the causal theorist
has shown himself or herself not to be using ordinary English and, presumably, not
understanding what he or she was saying.
It is also important that the implication be nondetachable and (to a high degree)
noncancelable, because it would not be enough for the critic of the causal theory to
show that just one way of characterizing sense-datum statements failed. It is essen-
tial that all ways of (equivalently; therefore nondetachably) characterizing sense-
datum statements should fail.
Grice then proceeds to argue that That pillar box looks red to me does not se-
mantically presuppose doubt-or-denial. His argument was this:
Suppose that I am confronted in normal daylight, by a perfectly normal pillar-box;

suppose further that I am in the presence of a normal, unsceptical companion; both
he and I know perfectly well that the pillar-box is red. However, unknown to him,
I suffer chronically from Smith’s Disease, attacks of which are not obvious to an-
other party; these attacks involve, among other things perhaps, the peculiarity that
at the time red things look some quite different colour to me. I know that I have this
disease, and I am having (and know that I am having) an attack at the moment. In
these circumstances I say, ‘That pillar-box looks red to me’. I would suggest that
here the doubt-or-denial condition is not fulfilled; my companion would receive
my remark with just that mixture of puzzlement and scorn which would please my
objector; and yet when he learnt about my attack of Smith’s Disease, he would cer-
tainly think that what I had said had been false [rather than neither true nor false].
(Grice 1965: 452–53)
Although this is an intuitively plausible counterexample to the critic’s claim that the
doubt-or-denial condition is semantically presupposed by statements of That pillar-
box looks red to me, the critic also wants to hold that the implications of these state-
ments are nondetachable and noncancelable.
In the course of discussing these claims, Grice (1961/1965:456) notes that if the
doubt-or-denial condition is nondetachable (from what is asserted), it is not nonde-
tachable by virtue of the implication’s being a logical implication. It is simply not a
logical implication that if this pillar-box looks red to me, then I, or someone else is,
or might be, inclined to deny that it is red or to doubt whether it is red. And even if
the implication is nonlogical but nondetachable, there is no reason to think that it is,
in consequence, not cancelable.7
What Grice now argues is that That pillar-box looks red to me nonlogically “im-
plies” the doubt-or-denial condition, but the implication is nondetachable and can-
celable. Like the implication of ‘or’, the implication is standardly implied by the
speaker, or by the speaker’s asserting what he did. Since such standard implications
are a matter of the use of language, not solely a matter of the meaning of words, the
implication is not solely a part of the meaning of the expression ‘looks to me’. So the
objection to explicating sense-datum statements in a causal theory of perception by
paraphrase into ‘looks to me’ statements fails.
It is particularly useful to note that the principle of language use in question is
“giving preference to the making of a stronger rather than a weaker statement in
the absence of a reason for not so doing” (Grice 1965: 459). One instance of such
a case, the stronger φ and the weaker φ or Ψ, was explained by the semantic
7Indeed, Grice’s example of ‘or’ would show why this is true. Interestingly Grice (1961, 1965: 456–
57) feels the need to dispose of a potential counterexample to his last thesis. He assumes that the impli-
cation of ‘but’ is nonlogical, and he has argued that it is (to a high degree) noncancelable. Now he
argues that if the implication of ‘but’ is noncancelable, the reason for its being noncancelable is that it
is detachable (by the use of ‘and’). Then, surprisingly, he wishes to conclude that if an implication is
nonlogical and not detachable, then it is cancelable: (D → ¬C) ∴ (¬D → C). Grice (1961, 1965: 457)
writes that “if you say that the implication of the fulfillment of the [doubt-or-denial] condition is
(a) not logical in character and (b) not detachable, then you must allow that it is cancellable.” If this
were correct, the objection to the Causal Theory would fail, as the objection claims that the D-or-D
implicatum is neither detachable nor cancelable.
The implicit argument here seems to be that since ‘but’ and ‘and’ are truth-functionally equiva-
lent yet distinct lexical items with distinct uses in the language, and since the occurrence of two nearly
synonymous items in the language would be superfluous unless some difference in use attached to them,
so the difference in form makes a difference in use by way of a difference in (noncancelable) “conven-
tional implicata” that the word-forms carry.
Plausible as this argument is, even if detachability of an implication φ from “what is asserted”
were to entail φ’s noncancelability, surely one should not conclude, as Grice seems to, that its
nondetachability entails its cancelability; that is just the Fallacy of Affirming the Consequent, as well
as an example of a conversational inference from ‘if’ to ‘if and only if’ (see chapter 3 in this volume).
What Grice has to show to defeat the critic is that it is possible for an “implication” to be cancel-
able while not detachable; on Grice’s own showing the default implication of ‘or’ is nondetachable yet
cancelable. He has shown what he needs to show. He does not need yet a further, bad argument.
It is noteworthy that Grice was surely assuming here, plausibly enough, that semantic presupposi-
tion is not a classical logical implication, since Strawson (1950, 1952) had explicitly tried to distinguish
presupposition from entailment, but since Grice had granted that presupposition is nondetachable and
(erroneously) that it is noncancelable, if it is also nonlogical, its properties would be a counterinstance to
the very thesis he has just tried to defend (by fallacious reasoning!).
entailment: φ φ ∨ Ψ, and the failure of the converse: φ ∨ Ψ φ (assuming it is

false that φ Ψ). Grice notes that It looks red to me and It is red are not instances
of an entailment ordering; they are logically independent.8 Nonetheless, Grice sug-
gests that, intuitively, one regards It looks red to me as the weaker. If that is cor-
rect, his principle of language use will justify the “standard implication”: if a speaker
asserts It looks red to me, he is in some doubt, or someone could have denied, that
it is red. Yet the implication is not part of the meaning of the sentence ‘It looks red
to me’. Thus Grice concludes that he has disposed “of what [he has] found to be a
frequently propounded objection to the idea of explaining the notion of a sense-
datum in terms of some member or members of the suggested family of locutions,”
as in ‘It looks red to me’ (1965: 460). And he suggests that a budget of philosophi-
cal objections of a like kind will be undercut by a like analysis:
1. You cannot see a knife as a knife, though you may see what is not a knife as a
knife.
2. When [G. E.] Moore said he knew that the objects before him were human hands,
he was guilty of misusing the word ‘know’.
3. For an occurrence to be properly said to have a cause, it must be something
abnormal or unusual.
4. For an action to be properly described as one for which the agent is responsible,
it must be the sort of action for which people are condemned.
5. What is actual is not also possible.
6. What is known by me to be the case is not also believed by me to be the case.
(Grice 1965: 459–60)9
It is, I hope, clear that a distinction between semantic features of statements, like
semantic entailment, semantic presupposition, and what Grice later called “conven-
tional implicature,” and pragmatic features, like what he later called “particularized
conversational implicature” and “generalized conversational implicature,” is crucial
to the success of Grice’s undercutting the criticism of the anti-sense-datum theorist.
In particular, the distinction between ‘but’ and ‘or’—that is, between what Grice
8I have referred to such orderings of lexical items as “Levinson scales,” but, as we see, Grice discusses
the sentential version in 1961. To my astonishment a referee for a prominent linguistics and philoso-
phy journal questioned my introduction of such an ordering on the grounds that no such notion had
been previously discussed in the published linguistics literature! Aside from the attitude to conceptual
innovation that such a remark betrayed, Paul Grice himself had published the sentential version of my
notion in the philosophical literature thirty years before I appropriated the lexical version of it for my
use in discussing implicature. For students of Grice’s work the matter was, I thought, common knowl-
edge, but apparently not. I withdrew the essay and published it elsewhere as Atlas (1991a); it has since
been the subject of discussion by de Mey (1991), Blok (1993), Horn (1992b, 1996b), and others. It is
now part of the linguistics literature.
9On (1), see “seeing as” in Wittgenstein’s (1967) Philosophical Investigations; on (2), see Wittgenstein’s
(1969) On Certainty; on (3), see J. Searle (1969b), “Assertions and Aberrations”; on (4), see Sinnott-
Armstrong (1984) and M. White (1993: 48–49); on (6), see Vendler (1975a), “On What We Know”;
Aune (1975), “Vendler on Knowledge and Belief,” and Vendler (1975b), “Reply to Professor Aune.”
(1967) later called a case of conventional implicature and a case of generalized con-
versational implicature—is the crux for Grice’s refutation of the criticism of the
causal, sense-datum theory.10
Having made good philosophical use of his division between the semantic and
pragmatic against the casual Wittgensteinianism of some 1950’s Oxford philoso-
phizing, six years later Grice presents a more systematic account of his principles
of the use of language. It is clear from his 1961 discussion that Grice wishes to
characterize, at least in the paradigm cases, the differences among semantic entail-
ment [E], semantic presupposition [SP] (4a), conventional (let us call it “lexical”
for convenience) implicature [LI] (4b), and particularized (4c) and generalized (4d)
conversational implicatures [PCI and GCI].11 If we look at the logical properties
of the classical semantic entailment relation—its reflexivity, transitivity, and mono-
tonicity—and compare semantic presupposition, conventional (lexical) implicature,
and conversational implicature with it, we have the mostly obvious results shown
in tables 2.1 and 2.2.
The pragmatic properties of entailment, conventional implicature, presupposi-
tion, and particularized and generalized conversational implicature—namely, detach-
ability, cancelability, and calculability (to be discussed in section 3)—are as shown
in table 2.2 (see Walker 1975 and Sadock 1978):
Grice was certainly not wrong to see differences here. In particular, his em-
phasis on the importance of the distinction between conventional implicata and gen-
eralized conversational implicata (e.g., between the implicata of ‘but’ and ‘or’),
however subtle their manifestations, seems well founded. Although Sperber and
Wilson (1986b, 1995) reject (incorrectly, I believe) the distinction, without it the
philosophical defense of Grice’s version of the Causal Theory of Perception would
have foundered.
10It is inexplicable or worse that Stephen Neale (1992: 524, n.18) should write, “The distinction be-
tween ‘generalized’ and ‘particularized’ conversational implicature is not represented in this diagram
. . . because it is theoretically inert (for Grice).” Grice himself writes:
I also distinguished . . . particular conversational implicatures that depended on particular
contextual features . . . and ones that I thought of as relatively general which I called GENER-
ALIZED IMPLICATURES. These are the ones that seem to me to be more controversial and at the
same time more valuable for philosophical purposes, because they will be implicatures that
would be carried (other things being equal) by any utterance of a certain form, though, as
with all implicatures, they are not to be represented as part of the conventional meaning of
the words or forms in question. (It is important that what is conversationally implicated is
not to be thought of as part of the meaning of the expressions that are used to get over the
implication.) And I thought that this notion of a GENERALIZED conversational implicature might
be used to deal with a variety of problems, particularly in philosophical logic, but also in
other areas. In these areas there seemed to me to be quite good grounds for suspecting that
some people have made the mistake of taking as part of the conventional meaning of some
form of expression what was really not part of its conventional meaning, but was rather a
non-conventional implication which would normally be carried, except in special circum-
stances, by the use of that form. (Grice 1981: 185)
11Grice notes that particularized conversational implicatures are “cases in which an implicature is car-
ried by saying that p on a particular occasion in virtue of special features of the context, cases in which
there is no room for the idea that an implicature of this sort is normally carried by saying that p” [a case
of a generalized conversational implicature]. (1989b: 37)
TABLE 2.1. Logical Properties of Semantic and Pragmatic Relations (1 = “yes”;

0 = “no”)a
Semantic Lexical Semantic Conversational
Entailment Implicature Presupposition Implicatures
Reflexive 1 0 0b 0
Monotonic 1 1 0c 0
Transitive 1 1 1 0
of a relation R among sentences: for every φ, φRφ. Transitivity: for every φ, Ψ, χ, (φRΨ &
aReflexivity
ΨRχ) → φRχ; Monotonicity: for any set A, ARφ → A∪{φ} R Ψ. See D. S. Scott (1971, 1973) and van
Fraassen (1971).
bThe reason that The king of France exists does not presuppose that the king of France exists is dis-
cussed in Atlas (1988, 1989: 91–118).
cShowing the failure of monotonicity of semantic presupposition is left as an exercise for the reader.
3 Rationality, cooperation, and the

imperatives of conversation
In the second of his William James lectures “Logic and Conversation” (Grice 1975a,b,
1989b, 1989c), at Harvard University, 1967, Grice briefly presents a description of
conversation as rational, cooperative, goal-directed behavior and formulates the
Cooperative Principle:
(5) Make your conversational contribution such as is required, at the stage at which it occurs,
by the accepted purpose or direction of the talk exchange in which you are engaged.
Then Grice describes the means by which this overarching goal of rational, coopera-
tive, conversational exchange is to be achieved: by following, or at least conforming
to, the conversational maxims, or what Grice (1989f: 370) later called “conversational
imperatives,” the observance of which promotes conversational rationality. Following
the example of one of his favorite philosophers, Grice divides linguistic rationality into
four categories: Quantity, Quality, Relation, and Manner (Grice 1986: 66, 103).
(6) Maxims of quantity

1. Make your contribution as informative as is required (for the current
purposes of the exchange).
2. Do not make your contribution more informative than is required.
(7) Maxims of quality

Try to make your contribution one that is true.
a. Do not say what you believe to be false.

b. Do not say that for which you lack adequate evidence.
TABLE 2.2. Pragmatic Properties of Semantic and Pragmatic Relations (1 = “yes”; 0 = “no”)
Generalized Particularized
Semantic Lexical Semantic Conversational Conversational
Entailment Implicature Presupposition Implicature Implicature
Cancelable 0 0 0a 1 1b
Detachable 0 1 0 0c 0d
Calculable 0 0 1e 1f 1
aCompare to treating presupposition as a GCI; see Grice (1978).
He didn’t {a. STOP} beating his wife —{he never beat her}
{b. stop} — {?he never beat her}
The focal stress negation in (a) interrupts the default GCI and permits the acceptable continuation (Atlas 1991a) (see
chapter 3, section 1). The (b) implicature is a GCI He used to beat her, which makes the continuation odd. See chap-
ter 4, section 6, and chapter 6, section 5.
bSadock (1978: 293) points out that cancelability will not be sufficient to show that a sentence is implicated, as the
failure of the utterance to be inconsistent with the denial of the sentence may be due to the ambiguity of the utterance:
the denial of the sentence forces an alternative reading on the ambiguous utterance. More plausibly, cancelability is
necessary; Grice (1989c: 44) certainly thought so.
cDefault value.
dNondetachability, as Grice (1989c: 43) himself points out, is neither necessary nor sufficient for the presence of a
conversational implicatum: it is not necessary since the implicatum may depend on manner of utterance, not on its
sense, so another utterance with the same sense but uttered in a different manner may not carry the same implicatum;
it is not sufficient since entailments and semantic presuppositions are also nondetachable. Its importance depends on
its distinguishing between conversational and conventional implicata. Sadock observes:
Whether there are absolutely equivalent paraphrases or not, the fact that it is difficult to tell if two expressions have the
same meaning makes the nondetachability test less useful in practice. Suppose the claim is made that a sentence such
as Can you open the door? does not conversationally implicate a request to open the door but rather conventionally
implicates it and that this claim is backed up by the observation that the implicature fails to go through if the synony-
mous periphrastic modal be able is substituted for can. Since the paraphrase detaches the implicature, the argument
goes, it cannot be conversational. This claim can all too easily be countered with the claim that can and be able are not
synonymous and that in fact this example PROVES that they are not. (Sadock 1978: 289)
For what it is worth, I can get the implicatum from Are you able to open the door?; in any case, Sadock’s point is well
taken. Happily for Grice, Sadock concludes:
Nevertheless, [detachability] is a pretty good [test], at least in extreme cases. Detachability in the absence of obvious
meaning differences of the right kind is a suspicious fact, and the more apparent paraphrases there are that succeed in
detaching an implicature, the more it looks as if the implicature must be conventional. On the other hand, the more
apparently synonymous expressions there are that fail to detach an implicature, the less the situation looks accidental
and the more it looks as if some principle, such as the Cooperative Principle, is in force. (Sadock 1978: 290)
eSee Grice (1978, 1989a: 269–282); nonmonotonicity suggests calculability.
fThough calculable, generalized conversational implicata are not typically calculated; see Morgan (1978).
(8) Maxim of relation

Be relevant.
(9) Maxims of manner12

Be perspicuous.
a. Avoid obscurity of expression.

b. Avoid ambiguity.
c. Be brief (avoid unnecessary prolixity).
d. Be orderly.
Of course Grice (1989b: 28) recognizes that conversation is pursued in conformity

with other sorts of maxims—for example, Be polite—and that one may convey mes-
sages by conforming or failing to conform to such norms of conversation (see Brown
and Levinson 1978, 1987). What Grice explicitly has in mind is a description of the
means by which the conversational goal of “a maximally effective exchange of in-
formation” can be achieved.
These maxims and the Cooperative Principle are offered in explanation of what
a speaker “implies,” “suggests,” or “means,” as contrasted with “what he said (as-
serted),” in uttering a sentence. Grice writes:
Suppose that A and B are talking about a mutual friend, C, who is now working in
a bank. A asks B how C is getting on in his job, and B replies, Oh quite well, I
think; he likes his colleagues, and he hasn’t been to prison yet. At this point, A
might well inquire what B was implying, what he was suggesting, or even what he
meant by saying that C had not yet been to prison. The answer might be any one of
such things as that C is the sort of person likely to yield to the temptation provided
by his occupation, that C’s colleagues are really very unpleasant and treacherous
people, and so forth.13 . . . It is clear that whatever B implied, suggested, meant in
this example, is distinct from what B said, which was simply that C had not been to
prison yet. (Grice 1989b: 24)
The explanation of this example that Grice gives is this:
In a suitable setting A might reason as follows: “(1) B has apparently violated the
maxim ‘Be relevant’ and so14 may be regarded as having FLOUTED one of the
12See also Grice (1981) and chapter 4, section 10 in this volume.

13 This‘so forth’ is important. The speaker’s utterance-interpretation of the sort illustrated in Grice’s
example is obviously “open-ended.” We shall encounter other cases in which the interpretation of what
the speaker meant by the statement is much more constrained.
14Grice evidently thought that a violation of relevance (Maxim of Relation) implied a failure to be
perspicuous (Maxim of Manner), or he thought (more likely) that one way for the speaker to violate
relevance would be to fail to be perspicuous, or he thought (even more likely) that if an addressee thought
that a speaker had failed to be perspicuous, the addressee would surely also be unable to see the rele-
vance (if any) of the speaker’s remark. I take it that Grice was not being perspicuous.
maxims conjoining [sic] perspicuity, yet I have no reason to suppose that he is opt-
ing out from the operation of the Cooperative Principle; (2) given the circumstances,
I can regard his irrelevance as only apparent if, and only if, I suppose him to think
that C is potentially dishonest; (3) B knows that I am capable of working out step
(2). So B implicates that C is potentially dishonest.” (Grice 1989b: 31)15
By “flouting a maxim” Grice means ‘blatantly failing to satisfy it’ and remarks:
On the assumption that the speaker is able to fulfill the maxim and to do so with-
out violating another maxim (because of a clash), is not opting out, and is not, in
view of the blatancy of his performance, trying to mislead. . . . How can his say-
ing what he did say be reconciled with the supposition that he is observing the
overall Cooperative Principle? This situation is one that characteristically gives
rise to a CONVERSATIONAL IMPLICATURE; and when a conversational im-
plicature is generated in this way, I shall say that a maxim is being exploited. (Grice
(1989b: 30)
In his later reflections, reported in “Retrospective Epilogue” to his Studies in

the Way of Words, Grice describes the arising of an implicatum from an assertion in
this way:
An implicatum . . . is the content [my emphasis] of that psychological state or

attitude which needs to be attributed to a speaker in order to secure one or an-
other of the following results; (a) that a violation on his part of a conversational
maxim is in the circumstances justifiable, at least in his eyes, or (b) that what
appears to be a violation by him of a conversational maxim is only a seeming, not
a real, violation; the spirit, though perhaps not the letter, of the maxim is respected.
(Grice 1989f: 370)
It is essential to conversational implicature that the content of the psychological

state of the speaker be one that the addressee is capable of reasoning out, the general
pattern of which Grice outlines as follows:
Calculability: Implicature argument schema

He has said that P; there is no reason to suppose that he is not observing the maxims,
or at least the Cooperative Principle; he could not be doing this unless he thought
that Q; he knows (and knows that I know that he knows) that I can see that the sup-
position that he thinks that Q is required; he has done nothing to stop me thinking
that Q; he intends me to think, or is at least willing to allow me to think, that Q; and
so he has implicated that Q. (Grice 1989b: 31)
Of course, if a speaker violates or flouts one of the conversational maxims at the

level of “what is said (asserted),” it is important to have some, even rough, charac-
15Obviously Grice meant “enjoin” instead of “conjoin” in this passage. It is also worth remarking on
the form of Grice’s argument. In step (2) Grice supposes that the speaker believes that C is potentially
dishonest. In step (3) the speaker implicates that C is potentially dishonest; he does not implicate that
he believes that C is potentially dishonest.
terization of what “what is said (asserted)” consists in. (After all, the reasoning just
described above begins with [The speaker] has said that P’.) So let us consider more
carefully “what is said” or “asserted.” Grice (1989b: 25) notes that in the case of an
ambiguous sentence, such as (10), one’s knowledge of the English language would
permit one to know that there are two possible senses.
(10) He is in the grip of a vice.
This sentence, independently of knowledge of context, could be expressed by either

(a) a male person or animal is unable to rid himself of a bad character trait or (b) a
male person is caught in a type of clamping tool.
Context permits semantic values to be given for parameters like tense and the
reference of ‘he’ and a choice of sense(s) to be made. Then one possesses an inter-
pretation of the utterance in the context, not merely the sense(s) of the sentence, that
Grice calls “what is said (asserted)” (Grice is indifferent whether or not what is as-
serted in uttering ‘John is handsome’ and ‘The first student whose last name begins
with ‘w’ in the Pomona College class of 1980 is handsome’ is the same, when ‘John’
and ‘the first student whose last name begins with ‘w’ in the Pomona College class
of 1980’ denote to the same individual.) In sum, fixing of reference, of tense, of deixis,
and disambiguation is assumed to yield, with literal sense(s), “what is said (asserted)”
in asserting a sentence-token.16
4 Relevance: Quantity [informativeness] subsumed

by relation [relevance]?
In “Retrospective Epilogue” Grice (1989f: 371–72) reflects on certain dependencies

among the maxims. The first concern, present even in the William James lectures of
1967, is whether the Second Maxim of Quantity (“Do not make your contribution more
informative than is required”) is superfluous, being supplanted by the Maxim of Rela-
tion (“Be relevant”). The second concern is whether violations of the Second Maxim
of Quantity amount to violations of cooperativeness at all. On this second concern Grice
is inclined to think that overinformativeness can be misleading to an addressee, and
thus uncooperative, in that the addressee will think that there is some point in the
speaker’s discursiveness. (Grice [1989b: 27] writes, “there may also be an indirect effect,
in that the hearers may be misled as a result of thinking that there is some particular
point in the provision of the excess of information.”) On the first concern, Grice (1989f:
372) reiterates his view that relevance and information are dependent notions, but nei-
ther early nor late does he explain just how the Second Maxim of Quantity could be
explained, or entailed, by the Maxim of Relation.
A. P. Martinich (1980: 218) provides a provocative discussion of this difficulty.
The reductive claim would argue: (a) if what a speaker says is relevant, it will not
16The notion of “what is said” has since come in for considerable discussion and notable disagreement
among philosophers and linguists: see Bach (1994a), Carston (1988), Levinson (1988b, 2000), Récanati
(1989), and Sperber and Wilson (1986b).
provide more information than is required, and (b) if what a speaker says is not
relevant, it is more informative than is required. If (a) were true, the Maxim of Rela-
tion would subsume the Second Maxim of Quantity; if (a) and (b) were both true,
the Maxim of Relation and the Second Maxim of Quantity would be equivalent.
Martinich’s counterexample to (a) is a line from François Mauriac’s Vipers’ Tangle.
Recalling his deceased grandson Luc, the miser says, “Everyone loved him, even I.”
In Gricean terms, the statement is, Martinich alleges, overinformative: ‘everyone loved
him’ entails ‘the miser loved him’; thus, alleges Martinich, the miser’s phrase ‘even
I’ is overinformative. Yet, observes Martinich, the statement is relevant: it indicates
that Luc was highly lovable and that, although the miser does not love widely, he
was at least able to love Luc. Martinich takes this example to show that some state-
ments are both more informative than is required and relevant, contrary to the reduc-
tionist claim (a).
If φ entails Ψ, the statement φ & Ψ is logically equivalent to φ and is truth-
conditionally neither more nor less informative than φ. If ‘even’ carried a conven-
tional implicatum, so that Even I loved him were logically equivalent to I loved
him (as Karttunen and Peters 1979 once suggested), the statement Everyone loved
him, even I would be, like φ & Ψ where φ entails Ψ, logically equivalent to Every-
one loved him. Since what is relevant in the context is the even I, what is overly
informative is not the phrase even I, contrary to Martinich’s claim, but rather its
conjunction with Everyone loved him, since that conjunction is, on the assump-
tions made here, logically equivalent to Everyone loved him. And the proposition
Everyone loved him is, in the context, overly informative if taken literally (and not,
as would be appropriate in this kind of case, hyperbole for Luc was lovable). With
this redescription of Martinich’s case, I think it is plausible to say that the speaker
has literally “said” (asserted) more than is required but was not thereby failing in
relevance. Thus the reductionist claim (a) is, indeed, false.
The observation that I have just made concerning hyperbole, which was not
considered by Martinich (1980), itself raises an interesting issue for Grice’s views.
Grice considers hyperbole under a class of conversational implicatures in which the
First Maxim of Quality (“Do not say what you believe to be false”) is flouted—that
is, blatantly violated.17 Grice’s (1989b: 34) example was Every nice girl loves a sailor.
(Apparently love and universal quantification just don’t mix.) Curiously, Grice gives
17The classic example of a flout (C-type) implicature is a flout of the First Maxim of Quantity, namely
of “exploitation, . . . a procedure by which a maxim is flouted for the purpose of getting in a conversa-
tional implicature by means of something of the nature of a figure of speech” (Grice 1989b: 33). This
is a variant of an example originally introduced in Grice (1965: 448–49). This is Grice’s account:
A is writing a testimonial about a pupil who is a candidate for a philosophy job, and his
letter reads as follows: “Dear Sir, Mr. X’s command of English is excellent, and his atten-
dance at tutorials has been regular. Yours, etc.” (Gloss: A cannot be opting out, since if he
wished to be uncooperative, why write at all? He cannot be unable, through ignorance, to
say more, since the man is his pupil; moreover, he knows that more information than this is
wanted. He must, therefore, be wishing to impart information that he is reluctant to write
down. This supposition is tenable only if he thinks Mr. X is no good at philosophy. This,
then, is what he is implicating.) (Grice 1965: 448–49)
The other cases Grice discusses, briefly, are the assertions of logical truths, such as War is war.
no gloss on his example at all. Hyperbole should have been of greater theoretical
interest to Grice, to Horn (1984b), and to Sperber and Wilson (1986b), for it demon-
strates the incorrectness of the subsuming of the Second Maxim of Quantity by the
Maxim of Relation [relevance], pace Horn (1984b) and Sperber and Wilson (1986b).
The statement John was loved by everyone could be relevant in a context, as
hyperbolic statements are, but, of course, it is “stronger” than would be required
for the simple exchange of information and, if taken literally of John in the ex-
ample, surely incorrect. At the level of what is implicated, by contrast, it would
communicate precisely what is required by the context: John was lovable. Thus it
is the Second Maxim of Quantity (“Don’t make your contribution more informa-
tive than is required [for the current purposes of the exchange]”) that is more effi-
cacious than the Maxim of Relation (relevance), just as the First Maxim of Quality
(“Don’t say what you believe to be false”) is more efficacious than the Maxim of
Relation: they both guide utterance-interpretation at the level of what is implicated.
The Maxim of Relation is incapable of guiding the interpretation of hyperbole, and
violations of the First Maxim of Quality alone will not distinguish hyperbole from
meiosis (understatement), from metaphor, or from irony. At the level of “what is
said (asserted),” the reduction thesis (a) is clearly false, as my reanalysis of
Martinich’s example shows. At the level of “what is implicated,” the Second Maxim
of Quantity plays an explanatory role that the Maxim of Relation cannot play, and
it cannot be subsumed by it.
As Levinson (1987a: 76) points out, both Wilson and Sperber (1981) and Horn
(1984b) incorrectly hold a reductionist view. Horn (1984b: 12) poses a rhetorical
question: “What would make a contribution more informative than required, except
the inclusion of material not strictly relevant to and needed for the matter at hand?”
The answer to this question commits Horn to reductionist theses: (a) (if what a speaker
says is relevant, it will not provide more information than is required) and (b) (if what
a speaker says is not relevant, it is more informative than is required). I have argued
against (a). Thesis (b) is strictly a nonstarter: its falsity is obvious. Consider the con-
text about John’s lovableness, and suppose that War is war is, gratuitously, and ir-
relevantly, uttered. Thesis (b)—if what a speaker says is not relevant, it is more
informative than is required—would predict that War is war is more informative than
is required, which is obviously absurd.
Martinich then offers a different argument against the subsumption of the Sec-
ond Maxim of Quantity (Quantity-2) by the Maxim of Relation (relevance). It is an
important instance of a familiar philosophical tactic:
If [Relation] and [Quantity-2] are jointly de trop, one could just as easily argue
that [Relation] is superfluous. For, if a speaker contributes more information than
is necessary, then the excess information of his contribution is irrelevant;
and, conversely . . . what is wrong with the objection against [Quantity-2] is that
it assumes that . . . faced with a choice . . . one should sacrifice [Quantity-2].
(Martinich 1980: 218)
If, as in Horn’s (1984b) rhetorical question, one were committed to a reductive equiva-
lence, so that on the grounds that φ Ψ, one reduced Ψ to φ, on the same grounds
one could just as well reduce φ to Ψ. Suppose further that one had a reductive defi-
nition of “more information” in terms of “relevance.” That, alone, would not show
that the notion of “more information” was irrelevant (pardon the expression); it would
merely show that whether or not it was logically irrelevant, it was no longer logi-
cally primitive in the vocabulary of the theory, and that if the notions used in the
supposed definition were justifiable in the theory, then the notion of “more informa-
tion” would be justifiable in the theory. The remaining case, one in which some of
the other maxims entailed the Maxim of Relation, subsuming it, would be the safest
claim to make if one thought that the Maxim of Relation were redundant, and, con-
versely, if one thought that the Maxim of Relation could be specialized to yield the
other maxims, the other maxims would be redundant.
What conversational relevance consists in is a large and disputed subject. I shall
briefly consider some of Grice’s own, late remarks in the “Retrospective Epilogue,”
which give us some clear indication of what Grice had in mind by relevance. In as-
sessing the Maxims of Quantity, Grice remarks:
To judge whether I have been undersupplied or oversupplied with information seems

to require that I should be aware of the identity of the topic to which the information
in question is supposed to relate; only after the identification of a focus of relevance
can such an assessment be made; the force of this consideration seems to be blunted
by writers like Wilson and Sperber who seem to be disposed to sever the notion of
relevance from the specification of some particular direction of relevance. (Grice
1989f: 371–72; see also Wilson and Sperber 1981 and Sperber and Wilson 1986b)
The role of topic in “what is said,” as in the logical form of cleft statements in
English, was emphasized in Atlas and Levinson (1981) (see appendix 3 of this vol-
ume), and its role in “what is implicated” was emphasized in Atlas (1984a) (see chapter
5), where it is needed to resolve contradictions in the implicata that standard appli-
cations of the First Maxim of Quantity can generate from statements like John is not
as tall as Brian. Statement-topic reappears again decisively in determining the cor-
rect logical form for Only John loves Sonia (see Atlas 1996b). Statement-topic is
crucial for a theory of implicature, as Grice (1989a) noted. Any theory that ignores
it, like Sperber and Wilson’s (1986b) Relevance Theory, cannot be a descriptively
adequate theory.
Grice’s notion of “a particular direction” in a conversation was anticipated, and
elaborated, by linguistic work in Conversational Analysis (see Levinson 1983: 284–
370) on rule-specified adjacency of two speech-acts (Sacks 1972; Levinson 1987a: 78;
Martinich 1980: 220–21), where what is “relevant” is what “should be done next” (as
in (11a)), and on sequential expectations, where what is “relevant” is what is “required
or wanted to achieve goals of the interaction” (as in (11b); Levinson 1987a: 77):
(11) a. A: Hello.
B: Hello.
b. A: Do you stock LT 188?
B: Big or small?
Finally, there is a notion of conversational topic, as contrasted with statement

topic, that describes the coherence of a discourse (see “formulation” in Sacks
(1972)).18
The philosopher A. P. Martinich independently observed the function of adja-
cency-pairs and of goal-directed sequential expectations in meeting the (vaguely
described) needs of “relevance.” Following Strawson (1964) he also emphasized
topical connectedness. In light of his observations, he suggests two Submaxims of
Relation:
a. Make your contribution one that moves the discussion towards its goal.
b. Express yourself in terms that will allow your hearer to tie your contribution
into the conversational context. (Martinich 1980: 220–21)
Martinich understands ‘context’ to include “the speaker and hearer, their common
perceptible environment, their previous utterances, and all of their relevant [my
emphasis] beliefs” (1980: 221). This notion of context is about as roomy as one
could ask for; unfortunately, it is not helpful to characterize the admittedly poly-
morphous notion of relevance by introducing submaxims intended to specify the
notion and then undermine the analysis by reintroducing the notion of “relevant
belief” in characterizing the concept of “conversational context” employed in the
submaxims themselves.
What is important and clear from Martinich’s (1980) discussion is the non-
reducibility of Grice’s Second Maxim of Quantity to the Maxim of Relation (rele-
vance). This is a significant result, and it is one to which I shall return, as a revised
and restructured, addressee-centered version of Grice’s speaker-centered Second
Quantity Maxim will play a significant role in the post-Gricean theory of inferenda
(e.g., Atlas and Levinson’s 1981 Maxims of Relativity and Principle of Informative-
ness to be discussed in chapter 3 of this volume).
Grice is looking at “information” from only one point of view: the speaker’s.
Grice’s point of view on quantity of explicit information was this: “Don’t say Ψ
& φ rather than φ if only φ is required.” A different post-Gricean point of view on
the Second Maxim of Quantity (“Don’t make your contribution more informative
than is required”) would be this: “Don’t say χ & φ rather than χ if it is evident to
18In his Principle of Relevance, Strawson (1964/1971a: 92) anticipates, the “topical connectedness” of
both conversational topic and statement topic, although he is not concerned to develop a theory of their
relationship:
We do not, except in social desperation, direct isolated and unconnected pieces of informa-
tion at each other, but on the contrary intend in general to give or add information about
what is a matter of standing or current interest or concern. There is a great variety of pos-
sible types of answer to the question what the topic of a statement is, what a statement is
“about”—about baldness, about what great men are bald, about which countries have bald
rulers, about France, about the king, etc.—and not every answer excludes every other in a
given case. This platitude we might dignify with the title, the Principle of Relevance.
(Strawson 1964, 1971a: 92)
both speaker and addressee that what is required—namely, χ & φ—is inferrable
from χ by the addressee (i.e., the speaker knows or is justified in presuming that
the addressee can infer it, and the addressee knows or is justified in presuming that
the speaker intends him to infer it).” In this case, too, if the speaker does provide
an excess of explicit information—something beyond χ—the addressee may be
misled, or, a possibility that Grice ignores, he may be correctly directed to the in-
tended interpretation by thinking that there is some particular point in the provi-
sion of the excess of explicit information. Grice had simply ignored the possibility
of implicit, “evidently” inferrable, information playing a role in the Maxims of
Quantity.19 A post-Gricean theory will not ignore that possibility.
5 Nontrivial applications, violations, and

flouts of the maxims
I have already indicated that Grice requires that a conversational implicature must
“be capable of being worked out” and that “to calculate a conversational implicature
is to calculate what has to be supposed in order to preserve the supposition that the
Cooperative Principle is being observed” (1989b: 31, 39–40). In the third William
James lecture “Further Notes on Logic and Conversation,” Grice puts an interesting
restriction on the character of implicata. He writes:
When I speak of the assumptions required in order to maintain the supposition

that the Cooperative Principle and maxims are being observed on a given occa-
sion, I am thinking of assumptions that are nontrivially required; I do not intend
to include, for example, an assumption to the effect that some particular maxim
is being observed, or is thought of by the speaker as being observed. (Grice 1989c:
41–42)
Let us refer to this condition as:
The nontriviality restriction on implicata

No implicata are of the types:
a. [Maxim] is fulfilled.
b. Speaker believes that [Maxim] is fulfilled.
Grice (1989c: 42) points out that this restriction has a consequence for understand-
ing G. E. Moore’s “paradox,” which, I observed, is a touchstone for Grice. The “para-
doxical statement” is an anomalous assertion (12b) instantiating the schema (12a):
(12) a. ?P but I do not believe P.

b. ?It’s raining but I do not believe it.
19This was a role for information that I had imputed to “presupposition” in my (1975a) discussion of
Frege and Dummett. See my discussion of the post-Gricean theory of presupposition in chapter 4 of
this volume.
The interpretation of this statement seems to require a speaker to have made a state-
ment the acceptance of which by any addressee (not excluding himself) the speaker
himself undermines by denying that he himself believes what he has said. Grice writes:
On my account, it will not be true that when I say that P, I conversationally impli-
cate that I believe that P; for to suppose that I believe that P (or rather think of myself
as believing that P) is just to suppose that I am observing the first maxim of Quality
[“Do not say what you believe to be false”] on this occasion.
I think that this consequence is intuitively acceptable; it is not a natural use of
language to describe one who has said that P as having, for example, “implied,”
“indicated,” or “suggested” that he believes that P; the natural thing to say is that
he has expressed (or at least purported to express) the belief that P. He has of course
committed himself, in a certain way, to its being the case that he believes that P,
and while this commitment is not a case of saying that he believes that P, it is bound
up, in a special way, with saying that P. (Grice 1989c: 42)
There are British linguists such as Stephen Levinson (1983: 105), and Ameri-
can philosophers such as Al Martinich (1980: 224–25), who have rejected Grice’s
view of Moore’s “paradox.” In fact, the conversational implicature analysis of Moore’s
paradox now seems to have become a “received view” (see comments in D. S. Clarke
1994: 145). Levinson’s explanation of the pragmatic anomaly is that the second con-
junct of statement (13) contradicts an alleged Quality Maxim (“Don’t say what you
believe to be false”) implicature of the first conjunct of the statement, which is that
one believes what one asserts.
(13) ??It’s raining and I don’t believe that it is.
Although he claims that “when one asserts something one implicates that one
believes it,” Levinson (1983: 105 n.7) nicely observes a major flaw in his own claim
and a fortiori in the current “received view” among philosophers. If S believes P were
implicated by S’s asserting that P, the implicatum would be consistently and intelli-
gibly cancelable—for example, like the cancelability of the implicatum not both of
or in The PM is patriotic or nationalistic—in fact she’s both. Note the absence of
linguistic anomaly in this utterance-type. By contrast, Moore’s “paradox” statements
are anomalous. Conversational implicata are deniable without inconsistency (with
“what is asserted”) and without anomaly. This evidence suggests that Moore’s “para-
dox” statement is not a case of implicature, despite the current popularity of the view
among philosophers.20
20Daniel Sperber (in conversation) questioned this argument, but his concern rests at best on an im-
plicature of what I said, not on what I said. My claim was that cancelability was a necessary condition
on the existence of an implicatum; I was not claiming that cancelability was a sufficient condition for
the existence of an implicatum. Once one shows that the anomaly of Moore’s Paradox cannot be ex-
plained by appeal to implicature, it remains open just what the explanation should be. If I understand
Sperber’s views (Sperber and Wilson 1986b), he is inclined to explain the anomaly by some kind of
“entailment,” and it is a thesis of this section that “entailment” is not a correct explanation.
2.1 Inadequacies of Grice’s implicature argument schema

Is there any reason for one to believe that an implicatum Bs[P] of S’s assertion
P exists? In fact, there is, and Grice himself unfortunately provides a rationale for
one.21 Recall Grice’s sketch of implicatural reasoning:
Implicature argument schema

He has said that P; there is no reason to suppose that he is not observing the maxims,
or at least the Cooperative Principle; he could not be doing this unless he thought
that Q; he knows (and knows that I know that he knows) that I can see that the sup-
position that he thinks that Q is required; he has done nothing to stop me thinking
that Q; he intends me to think, or is at least willing to allow me to think, that Q; and
so he has implicated that Q. (Grice 1989b: 31)
Martinich plausibly suggests the following instance of the Argument Schema:

“He has said that it is raining; there is no reason to suppose that he is not observing
the maxims, or at least the Cooperative Principle; he could not be doing this unless
he thought that he believed that it is raining; etc.” (Martinich 1980: 224–25). On
Grice’s grounds and Levinson’s grounds there is no implicatum Speaker believes
that P from the speaker’s assertion P, so the oddity of a lack of cancelation of the
(nonexistent) implicatum Speaker believes that P by the speaker’s denial I don’t
believe that P, and hence an inconsistency between what is implicated and what is
asserted, cannot be the correct explanation of the linguistic anomaly in Moore’s para-
dox statements. There must be something wrong with the formulation of Grice’s
Argument Schema.
The flaw in Grice’s formulation of his Argument Schema is in the clause he could
not be [observing the maxims or at least the Cooperative Principle] unless he thought
that Q. On Grice’s (1989f: 370) own view, an implicatum arises as the content of
that psychological state that a hearer needs to attribute to a speaker in order to ex-
plain, from the speaker’s point of view, why a maxim has been violated—for ex-
ample, because its fulfillment is less imperative than the fulfillment of another maxim,
a case of maxim clash (Grice’s class B implicata)—or arises as that part of the con-
tent of the total signification of the utterance that constitutes what the speaker “meant,”
in order to explain, from the speaker’s point of view, why, contrary to appearances,
a maxim or the Cooperative Principle has not really been violated or flouted—that
is, blatantly violated (Grice’s class C implicata).
21Interestingly, from the historical perspective, in 1942 Moore (1968:542), unlike Grice (1989b: 31),
apparently thinks that speaker S’s assertion of P does “imply” BS[P], in some sense of ‘imply’.
In the same year, in response to Moore, the American logician C. H. Langford (1968: 333) gives an
explanation of the Moore statement of the variant (Grice 1989b: 31) form B1st pers.[P] & ¬P, a form
later favored by Wittgenstein, that uses the Sincerity Condition on assertion and a Consistency Con-
dition of rational belief to derive the contradiction B[P] & ¬B[P]. Nevertheless, as I have dis-
cussed elsewhere (Atlas 1995), the linguistic anomaly of Moore’s “paradox” statements is not
explained by the rational speaker’s insincerity or by the sincere speaker’s irrationality. Likewise,
logical inconsistency is not an explanation of linguistic anomaly. See appendix 1 for a discussion of
Moore’s sense of ‘imply’.
I shall argue that Grice’s own formulation confuses the class B problem of ex-
planation of cooperative linguistic behavior with the class C problem of speaker’s
utterance-interpretation. The letter of Grice’s formulation in the Implicature Argu-
ment Schema is too imprecise to discriminate between these problems, and I will
argue that the conflation of these problems is the result of certain philosophical as-
sumptions about the explanatory roles of the semantic contents of assertions and the
mental contents of belief-states. In particular, Grice argued that semantic contents of
assertions could be subsumed, by analysis, under mental contents of psychological
states—of communicative intentions (Grice 1989a: 213–23, 117–37), for instance—
and that whatever was to be explained by semantic contents ultimately would be
explained by those mental contents. I shall show how Grice’s reductionist program
in the theory of meaning undermines his account of conversational implicature. (Ironi-
cally there is an immediate parallel to Russell’s 1903 error in his theory of denoting
concepts. Russell’s assumption that meanings were intentional objects undermined
his theory of reference.)
Grice suggests that conversational implicata can arise from violations of the
maxims or from exploitations (via blatant violations) of the maxims. (I discussed an
example of exploitation in section 3.) His example in which a maxim is violated but
its violation is to be “explained by the supposition of a clash with another Maxim” is
this:
A is planning with B an itinerary for a holiday in France. Both know that A wants
to see his friend C, if to do so would not involve too great a prolongation of his
journey:
(14) A: Where does C live?
B: Somewhere in the South of France.
(Gloss: There is no reason to suppose that B is opting out; his answer is, as he
well knows [my emphasis], less informative than is required to meet A’s needs.
This infringement of the first maxim of Quantity [”Make your contribution as in-
formative as is required . . .”] can be explained only [my emphasis] by the suppo-
sition that B is aware that to be more informative would be to say something that
infringed the second maxim of Quality, “Don’t say what you lack adequate evi-
dence for,” so B implicates that he does not know in which town C lives.) (Grice
1989b: 32–33)
The gloss that Grice provides is interesting because it indicates the distinct roles
that Grice, explicitly, thinks that implicature plays. He has written in the recent “Ret-
rospective Epilogue” to Studies in the Way of Words that “an implicatum . . . is the
content of that psychological state or attitude which needs to be attributed to a speaker
in order to secure one or another of the following results; [e.g.] . . . that a violation
on his part of a conversational maxim is in the circumstances justifiable, at least in
his eyes” (Grice 1989f: 370). But this content is not content that, intuitively and pre-
theoretically, I would have thought that a speaker “implied,” “suggested,” or “meant”
by, in, or when saying what he said. It is, rather, as Grice notes in his gloss, a hypo-
thetical explanation of the speaker’s infringement of the First Maxim of Quantity,
the hypothesis being that the speaker does not know in which town C lives, and in
conforming to the Second Maxim of Quality, the speaker says no more than Some-
where in the South of France.22
A’s hypothesis that B does not know where C lives is inferred as A’s best expla-
nation why, assuming that B is conforming to the Cooperative Principle, B has failed
to conform to the First Maxim of Quantity and was less informative than the context
required. Nevertheless, it is not clear in the example that when B “said” (asserted)
Somewhere in the South of France, B meant or suggested or implied in “saying”
Somewhere in the South of France that he did not know in which town C lived. It is
not clear that B intended by his choice of words to convey that, and, hence, it is not
clear that B meant it in saying what he said. (Despite animadversions by some lin-
guists, I claim, in agreement with Grice, that there are default implicata standardly
conveyed by the uttering of sentences, whether or not the speaker specifically in-
tended them to be conveyed—for example, generalized conversational implicata—
but in such cases, the speaker may have misled his addressee by his choice of words.)
What we have is a plausible explanation of B’s uninformative response. What
we do not have is a case of B’s intentionally conveying to his addressee a content
that he has not explicitly asserted. It certainly need not be the case that B utters ‘Some-
where in the South of France’ with the intention that by virtue of B’s utterance A
will come to believe that B does not know in which town C lived. For B to have this
communicative intention (M-intention in Grice’s 1969: 155; 1989a: 105 sense) it must
be true that B utters the sentence ‘Somewhere in the South of France’ intending A
(a) to believe that B does not know in which town C lives, intending A (b) to think B
intends A to have the belief in (a), and intending A (c) to think B intends the fulfill-
ment of the first intention, that of (a), to be effected by the fulfillment of the second
intention, that of (b). The example, as Grice describes it, gives us no reason to think
that these conditions for a communicative intention are fulfilled.
There is a further peculiarity in Grice’s account. He has stated that B’s infringe-
ment of the First Maxim of Quantity can be explained only by the hypothesis that B
is conforming to the Second Maxim of Quality. Why is it the only such explanation?
Couldn’t one also say that if B were aware that to be more informative would be to
say something (e.g., C lives in Marseilles) that infringed the First Maxim of Quality,
“Do not say what you believe to be false” (or even, “Do not say what you don’t be-
lieve to be true”), B should not say it? So, according to Grice’s reasoning, B could
implicate, again, that he does not know in which town C lives. There is more than
22Although Grice leaves the matter undiscussed, there is reason to think that he would accept the fol-
lowing generalization: if the Maxims of Quantity and Quality compete, the imperative to conform
to quality supersedes the imperative to conform to quantity. As he remarks in the “Retrospective
Epilogue”:
The [super] maxim of Quality, enjoining the provision of contributions which are genuine
rather than spurious (truthful rather than mendacious) does not seem to be just one among a
number of recipes for producing contributions; it seems rather to spell out the difference
between something’s being and (strictly speaking) failing to be, any kind of contribution at
all. (Grice 1989f: 371)
And he (Grice 1989b: 27) had said already that “other maxims come into operation only on the as-
sumption that this [super] maxim of Quality is satisfied.”
one plausible explanation of B’s infringement of the First Maxim of Quantity, more
than one reason to attribute the same psychological content to B’s mental states.
There are two subjects of concern to Grice. One is the rational explanation
of behavior, especially verbal behavior: the hypothesizing of reasons for which an
agent does what he or she does. The other is the nature of linguistic meaning:
utterance(type)-meaning and sentence(type)-meaning. The two concerns intersect at
speaker’s-meaning, what a speaker means by, in, or when saying what he or she has
said. So Grice wishes implicata to do double duty: first, as part of explaining why
agents, including speakers, do what they do; second, as part of interpreting the
sentence-tokens produced by speakers.
In philosophy this is a familiar demand to put on propositional contents. In the
case of belief, the attribution to an agent of an attitude toward the content of a belief-
report, the content of the that-clause, is part of commonsense explanations of the
agent’s actions, while the attribution of a particular content to an agent’s utterance is
part of commonsense interpretations of the agent’s speech. The propositional con-
tent is both a psychological item, the content of a belief, and a semantic item, the
meaning of a speaker’s utterance. The two roles converge when the agent intention-
ally states his beliefs in words: when he makes sincere assertions.
The difficulty is that the contents of implicata are not suited to play the double
role, in psychological explanation and in semantic interpretation, that Grice expects
them to play. It is plausible that whatever (in the way of reasons) explains why an
agent did (in a context) what he did is, at least in part, what the agent believed. By
contrast, whatever (in the way of reasons) explains why a speaker said (in the con-
text) what he said is not necessarily, even in part, what the speaker meant. B did not
mean, in saying Somewhere in the South of France, that he did not know in which
town C lived, even if his ignorance was his reason for saying it. Grice (1989f: 370)
was mistaken to view implicata as coherently playing both roles: disant-en and disant-
pour.23 The role for implicata is the one of conveyed speaker’s meaning.
Grice himself, not surprisingly, understood the difficulty. As he summarized his
analysis of ‘A meantNN something by x’ (1989a: 89), he claimed that the analysandum
of nonnatural (NN) meaning was
equivalent to ‘A intended the utterance of x to produce some effect in an audience

by means of the recognition of this intention’; and we may add to that to ask what
A meant is to ask for a specification of the intended effect (though, of course, it
may not always be possible to get a straight answer involving a ‘that’ clause, for
example, ‘a belief that . . .’).
And then he proceeded to raise the pertinent question:
Will any kind of intended effect do, or may there be cases where an effect is in-
tended (with the required qualifications) and yet we should not want to talk of
meaningNN? Suppose I discovered some person so constituted that, when I told him
23 For another unsatisfactory reductionist monism of meaning to action, see Donald Davidson (1980:
238–39, 1984a: 148–49).

that whenever I grunted in a special way I wanted him to blush or to incur some
physical malady, thereafter whenever he recognized the grunt (and with it my in-
tention), he did blush or incur the malady. Should we then want to say that the grunt
meantNN something? I do not think so. (Grice 1989e: 220–21)
And Grice makes the significant observation:
This points to the fact that for x to have meaningNN, the intended effect must be
something which in some sense is within the control of the audience, or that in some
sense of “reason” the recognition of the intention behind x is for the audience a rea-
son and not merely a cause. (Grice 1989e: 221)
But then Grice notices the problem:
It might look as if there is a sort of pun here (“reason for believing” and “reason for
doing”), but I do not think this is serious. For though no doubt from one point of
view questions about reasons for believing are questions about evidence and so quite
different from questions about reasons for doing, nevertheless to recognize an
utterer’s intention in uttering x (descriptive utterance), to have a reason for believ-
ing that so-and-so, is at least quite like “having a motive for” accepting so-and-so.
Decisions “that” seem to involve decisions “to” (and this is why we can “refuse to
believe” and also be “compelled to believe”). . . . It looks then as if the intended
effect must be something within the control of the audience, or at least the sort of
thing which is within its control. (Grice 1989e: 221)
Actually, Grice’s worry was appropriate. There was a pun here, and unlike Grice
I do think that it was serious. When Grice makes a remark of the form “a is at least
quite like b,” one wonders just what he had in mind. For on the face of it, having a
motive for accepting the divorce court judge’s statement that you owe your former
wife $2 million does not seem at all the same as having a reason to believe that you
owe your former wife $2 million; in such a case having a reason for believing so-
and-so does not at all seem quite like having a motive for accepting so-and-so.
How would a decision “that” involve a decision “to”? Brevet’s decision that
Throckmorton was offside does not logically or causally require that Brevet should
decide to do anything, not even if Brevet is the referee. (If Brevet is the referee, he
has an obligation, under the rules of the game, to make his decision known, but is
that what Grice meant by ‘involves’?) Decisions are a misleading case for Grice’s
purposes. And Grice’s observation that decisions “that” involve decisions “to” be-
cause one can “refuse to believe” and be “compelled to believe” is even more
peculiar.
I believe “that” there is a pair of spectacles perched on the bridge of my nose,
and, in fact, there is a pair of spectacles perched on the bridge of my nose, but am I
compelled to believe that there is a pair of spectacles perched on the bridge of my
nose? Some “evidence” is said to be “compelling”—in fact, the feelings of pressure
upon my nose and the visible rim of my spectacles might constitute such—but to
translate the “compellingness” of the evidence into my “being compelled” to acquire
a belief-state does seem just a bad pun.
Can I refuse to believe that there is a pair of spectacles perched on the bridge of
my nose? Suppose I am wearing my spectacles, and in the face of the supposed com-
pulsion to believe that I am wearing my spectacles, can I refuse to believe that I am
wearing my spectacles? I really don’t know what to say here.
Suppose there is no pair of spectacles perched on the bridge of my nose; now,
can I refuse to believe that there is a pair of spectacles perched on the bridge of my
nose? My face is a spectacle-free-zone, but Rogers Albritton says to me, “There’s a
pair of spectacles perched on the bridge of your nose!” Under what circumstances
would I say, “I refuse to believe it”?
In another case, I know what I would say. Rogers says, “Atlas, take out the
garbage!” I say, “I refuse.” Now, in this case of a response to a command, I know
what a refusal is. But when Rogers asserted, “There’s a pair of spectacles perched
on the bridge of your nose,” in what sense was his assertion a command that I should
believe it? Rogers may have, in a Gricean way, an intention that I should believe
it, and I might come to accept it by virtue of recognizing Rogers’s intention that I
should accept or believe it, but how does having that motive for accepting his state-
ment come even close to showing that my recognition of Rogers’s intention that I
should believe it itself constitute a good reason for my believing that the proposi-
tion is true? It would seem to be necessary, at least, that I not believe that Rogers
is a trickster, joker, or all-around con artist, believe that he is a reliable observer of
my face, that his reliable observations are reliably connected with his verbal asser-
tions, and so on.
Since an assertion is not a command that one believe what is asserted, the appli-
cation of notions like refusal and compulsion to assertion, and so to belief, seems an
unjustified extension of those notions if one is to take the notions literally.
Can I refuse to believe that there is no pair of spectacles perched on the bridge
of my nose? I certainly do not believe that there is no pair of spectacles perched on
the bridge of my nose, but am I by not believing it refusing to believe it? Not taking
out the garbage tout court is not refusing to take out the garbage (though one can
refuse to take out the garbage without saying that one is refusing to take out the gar-
bage, of course.)
If all that ‘I refuse to believe it’ means is ‘I do not believe it’, I have no diffi-
culty with Grice’s claims. But in that case, the claims make no philosophical point
about a connection between reasons for believing and reasons/motives for doing.
In sum, the observations from Grice’s essay “Meaning” do not succeed in se-
curing his claim that in the audience’s recognizing the intention of the speaker for
the audience to accept the utterance x the recognition is a reason for believing x and
not merely a cause of accepting it. He was acutely aware of the problem, wondered
whether it was serious, and found an interesting strategem to deny its seriousness.
The strategem, unfortunately, fails in its purpose. The audience’s recognition of a
reason or motive for or cause of a speaker’s saying x—namely, the intention to pro-
duce a belief in the audience by means of the recognition of the speaker’s intention—
is not so far a reason for rather than a cause of believing the content of what is said,
at least not for the reasons that Grice offers. In his original 1957 “Meaning” essay
(reprinted in Grice 1989e), Grice put his finger on a central difficulty in his own
account of “meaningNN.”
We can see Grice’s difficulty more explicitly if we consider an example of a

flouting of a Maxim, not merely a violation of one. If, in the same context Grice
described above, the discourse had been:

B: Somewhere on the planet Earth.
B’s remark would have been a flouting of the First Maxim of Quantity, not a violation
of it. Like the tautologies discussed by Grice (1989b: 33)—War is war—Somewhere
on the planet Earth is a blatant infringement of the injunction to be as informative
as required to meet A’s needs for information. If B is not opting out of the conversa-
tion, A can explain B’s response by hypothesizing that B does not know more pre-
cisely where C lives.
Explanation, though, is an inquiry-specific and question-specific activity: what
needs explanation, and of what type, depends on how problematic the “facts” are
taken to be (Bromberger 1992b). For us Earthlings, B’s response is highly uninfor-
mative. For Luke Skywalker, possessing an intergalactic spaceship and having this
discussion with Han Solo in a far-off galaxy, B’s (Solo’s) response is not quite so
uninformative. If B’s (Solo’s) reply is “informative enough”—for the purposes of
the questioner (Luke), merely specifying the galaxy might have been sufficient to
answer ‘Where does C live?’—that is because the point of the question could have
been the identification of C’s place of origin, broadly understood, or his biological
identification as a carbon-based organism of an Earth species, or . . . et cetera. Luke’s
explaining Solo’s choice of utterance Somewhere on the planet Earth requires Luke
to judge whether Solo has understood the point of Luke’s question and to hypothe-
size what point Solo has in fact taken Luke’s question to have if it is not the one
Luke intended. But explaining Solo’s choice to utter Somewhere on the planet Earth
is not eo ipso to interpret the content of Solo’s utterance.
The interest of this point about explanation is that in the case of War is war, the
blatantly uninformative logical truth of the statement makes Grice say that the state-
ment is “informative at the level of what is implicated, and the hearer’s identifica-
tion of [its] informative content at this level is dependent on his ability to explain the
speaker’s selection of this particular patent tautology” (1989b: 33). That is, Grice
asserts the dependency of interpretation on explanation that I have just denied. Grice
explicitly makes the semantic content of the utterance-object’s “conveyed meaning”
depend on the psychological content explaining the speaker’s intentional utterance-
act. Yet, though it is inferrable from Solo’s behavior that Solo does not know more
precisely where C lives, I claim that there remains an “open question”: What did Solo
“imply,” “suggest,” or “mean” in “saying” the hypothetical sentence-token Some-
where on the planet Earth? (Let me dub this point contrasting meaning with explaining
“the Open Question Argument.”)
Now, to return to our original example (14), why should A take a sentence to be
an implicatum of B’s utterance when the sentence is not a relevant response to A’s
WH-question, ‘Where does C live?’? Even if Grice’s statement “I don’t know” is
taken partially to explain why B answered as he did, why is it taken to be what B
meant when he answered as he did (the Open Question)? At the level of what is
implicated, the Maxim of Relation (“Be relevant”) is not being satisfied by that pu-
tative implicatum, nor is “I don’t know” satisfying the requirement of the First Maxim
of Quantity, insofar as what is needed is an informative reply to a WH-question. A
reason for saying P, or “I couldn’t justifiably say anything else,” is not necessar-
ily the same as what the saying of P is good for accomplishing in the way of com-
municative goals: conveying an “implication.”
If one didn’t know in which town C lived, and one thought the name of the town
was “relevant”—that is, what A needed (and in his gloss Grice does describe B as
knowing that he has said less than A needs), one could have said I don’t know where
C lives. That utterance, or the shortened I don’t know, would have been as brief a
reply as Somewhere in the South of France. One was not obeying a Maxim of Man-
ner (“Be brief”) in uttering Somewhere in the South of France that one would have
failed to obey by uttering I don’t know. Why would one bother to implicate I don’t
know which town C lives in when one could have asserted it?
The answer cannot be merely ‘to be truthful’—to obey the Maxim of Quality at
the level of implicature rather than at the level of assertion, since one was being truth-
ful, in the example, in asserting ‘Somewhere in the South of France’. Is one being
informative at the level of what is implicated? No, not if what is desired is informa-
tion about where C lives and not about what one doesn’t know. That one does not
know something is only in special circumstances useful information about one: for
example, what one’s inability to answer a question or solve a problem tells an exam-
iner, an interviewer, or a court of law. It tells A something informative, in that sense,
only if one should have known where C lives, but it still doesn’t answer A’s WH-
question. If, at the level of the implicatum, one is not being notably truthful, more
informative about where C lives, more perspicuous, or more relevant than at the level
of “what is said (asserted),” in what sense is one being more “cooperative” in impli-
cating than in asserting? Surely the answer for this example is “in no sense.” So why
does Grice think that I don’t know where C lives is implicated?
Of course, one could object that I am using an example from Group B in Grice’s
lecture, an example in which the point is only to illustrate the violation of a maxim
and to illustrate the explanation of the violation by the hypothesis of a clash with
another maxim. It is the Group C examples, not violations but flouts—blatant viola-
tions—where the Cooperative Principle is satisfied at the level of what is implicated.
The objector thinks that I should not judge example (14) by the standards set for
examples like (15), and so I should not conclude that there is something odd in Grice’s
claiming that in (14) B “implies,” “suggests,” or “means” that he does not know in
which town C lives by, in, or when asserting Somewhere in the South of France.
But it seems to me that every question that I can raise about the flout in (15) can
also be properly raised about the violation in (14) as well, for not only are the gram-
matical features of the examples the same, but the “flouting” character of (15) also
applies to (14) in the right context.

B: Somewhere in the South of France.

B: Somewhere on the planet Earth.
Just imagine a conversation (14) taking place when A and B are both in the South of
France and think that the South of France is the world. In that context, example (14)
would be both a Group B and a Group C example: there is a violation of Quantity to
be explained by a clash with Quality, and the violation is so blatant, as blatant as a
tautology (see Quine 1960: 66), as to count as a flout of the First Maxim of Quantity.
Example (14) in that context would be like (16):

B: Somewhere.
In cases (14), (15), and (16), one would infer from B’s reply that he does not
know more precisely where C lives, but it does not follow from that inference-to-an-
explanation of his “saying” what he did (and not something more precise) that he is
“implying,” “suggesting,” or “meaning” (as contrasted with “asserting”) in assert-
ing what he did that he does not know where C lives. A good explanation of, or rea-
son for, his saying what he did is not necessarily a good interpretation of what he
meant in saying what he did.
Of course I am not denying that sometimes the two might coincide, nor am I
denying a quite different claim: the converse claim that a good interpretation of what
he meant in saying what he did is at least part of, or necessary to, a good explanation
of, or reason for, his saying what he did.24 The tendency by philosophers, especially
Grice, to substitute one of these distinct claims for its converse is just a logical mis-
take. It is also a mistake, as Richard Rorty (1998b: 296) emphasizes, of a represen-
tationalist theory of knowledge (or perception) in which causality (causal explanation)
and justification are conflated, a conflating common to both British empiricism and
British idealism.
What discourses (14), (15), and (16) also show is that this putative distinction
between Class B violation-implicata and Class C flout-implicata is not a real distinc-
tion. Grice’s B-type “implicata” are not implicata at all. Grice uses B-type “impli-
cata” for purposes of psychological explanation, but they will not simultaneously
serve the purposes of C-type implicata as semantic interpretation. To distinguish
B-type explanations of maxim-clash from C-type flout implicata, Grice would have
to draw a principled distinction between discourses (14), (15), and (16), but there is
none to be drawn.
Since Grice conflates speaker’s-utterance interpretation with psychological ex-
planation, Grice’s Implicature Argument Schema is inadequate. That the speaker
could not be observing the maxims and the Cooperative Principle unless he believed
that Q does not show that Q is what the speaker “meant,” or even part of what the
speaker “meant,” in his utterance. (It does not even show, in Grice’s terms, that the
24This is a point occasionally emphasized by Hilary Putnam—for example, in “Reference and Under-
standing” (Putnam 1978b: 110–11). See also Atlas (1979: 276–77).
speaker S intends his addressee A to believe that the speaker S believes that P by
virtue of A’s recognizing that speaker S has the intention that A believe that speaker
S believes that P.25) The First Maxim of Quality enjoins the speaker not to say what
he believes to be false, and the Supermaxim of Quality enjoins him to try to make his
contribution one that is true. If the speaker aims to speak truly, we would expect the
speaker to say what he believes he has good reason to believe, but that he believes
that P is not part of what he “means” in uttering P (as if, for any P, S believes
that P could be logically entailed by the speaker- and context-dependent utterance-
meaning of P!). Thus, satisfying the implicature argument schema is not sufficient
for a sentence to be an implicatum of an utterance, because the schema conflates the
psychological explanans of an utterance-act with the linguistic implicatum of an
utterance-content.
By Grice’s Implicature Argument Schema, the First Maxim of Quality fails to
justify treating the Speech Act Sincerity Condition for assertions (Searle 1969a) as
an implicature. Grice’s Nontriviality Restriction on Implicatures (that the satisfac-
tion of a maxim is not implicated) is defensible. If a sentence is “calculable” from
the Implicature Argument Schema, it is not necessarily a conversational implicatum.
Some, such as Martinich (1980), treat the satisfaction of the Argument Schema as a
sufficient condition for a sentence being a conversational implicatum. Grice himself
merely claims that it is necessary: if a sentence is to be a conversational implicatum,
it must be possible to “work it out” according to the inferences in the Argument
Schema. Grice’s Argument Schema will not distinguish between sentences needed
for the psychological explanation of a speaker’s mental or verbal actions and sen-
tences needed for the interpretation of a speaker’s utterances. Semantic interpreta-
tion is not equivalent to psychological explanation, and the interpreted semantic
contents of statements are not to be identified with the explanatory mental contents
of belief-states, whatever Grice’s hopes in this matter for his reductionist program
of meanings to speaker’s intentions (Avramides 1989; Schiffer 1987). Grice’s intui-
tions that conversational implicature did not provide an explanation of Moore’s “para-
dox” were correct.
25See T. Baldwin (1992: 228) on Moore. M. Burnyeat (1967–68) makes Grice’s (1989d: 105) M-
intending essential to the concept of assertion. But even though speaker S asserts P BECAUSE S be-
lieves that P, it does not follow that a speaker S ASSERTS P because S believes that P (Dretske 1972;
Böer 1979).
The Rise of Neo-Gricean Pragmatics
1 The inconsistency of maxims and

the principle of informativeness
There seems to be a natural parallel between saying (1) and communicating the
more informative proposition (2) and saying (3) and communicating the more in-
formative (4).1
(1) John has three children.

(2) John has three children and no more than three children.
(3) The king of France is not bald.
(4) There is a king of France and he is non-bald.
The inference whereby (1) is used to communicate (2) by generalized conversa-

tional implicature has been much discussed under the rubric “scalar implicatures”
(Horn 1972; Gazdar 1976, 1977, 1979a).2 The parallel suggests that a similar ac-
1This section contains in part a revised version of sections 8–10 in Atlas and Levinson (1981: 32–43)
and appears with the permission of Academic Press.

2It is a First Quantity Maxim implicature from what the speaker did not say. He did not say four et al.
In the class of contexts in which it would be informative to say how many children John has, not say-
ing four, . . . is, ceteris paribus, behavior that conforms to the maxims. So the speaker would not be
conforming to the maxims, or being cooperative, if he had said four. . . . He would not be cooperative
unless he thought that John has no more than three children. He knows (and knows that the addressee
80
THE RISE OF NEO - GRICEAN PRAGMATICS 81
count might be given for the negative sentences. However, closer inspection indi-
cates that the parallelism between the implicatures induced by scalar items and those
induced by negation is illusory. The apparent parallelism exists merely because
the conjunction of any implicatum, however arrived at, with the logical conse-
quences of “what is said” will typically be more informative than those conse-
quences alone.3
In the case of many expressions, we may construct an ordering of items that meets
at least this condition: For an appropriately defined class of sentences, any sentence
containing the ith term of the ordering will entail a sentence like the original except for
“containing” the i+1st term of the ordering at one occurrence of the ith term in the
original sentence. Such an ordering we will call a “Horn Scale.” (We ignore several
complexities; see Gazdar 1979a: 55–58.) For example, consider the Horn Scales in (5):
(5) a. < . . . , . . . , four, three, two, one>

b. <necessarily, possibly>
c. <all, most, many, some, few>
d. <know, believe>
e. <must, should, may>
f. <and, or>
Sentences employing scalar words have generalized conversational implicatures of

these sorts:
1. If a speaker asserts a sentence A(. . .) containing a later, “weaker”

term in the scale—for example, A(three), A(possibly), A(some)—he
implicates the falsity of the “stronger” scalar variants: for example the
falsity of A(four), of A(necessarily), and of A(all).
2. If a speaker asserts the negative of a sentence containing an earlier,
“stronger” term in the scale—for example, not-A(four), not-
A(necessarily), not-A(all)—he implicates a “weaker” variant: for
example, A(three), A(possibly), A(some).
The explanation of the first sort of scalar implicature involves Grice’s First Maxim
of Quantity and the Maxims of Quality. If a speaker is in a position to assert that
John has five children, he should not say that John has three children; if he does as-
sert the latter, he may be taken to be in no position to assert a stronger statement—
for example, John has five children—and, in conformity with a consequence of the
knows that he knows) that the supposition that he thinks that John has no more than three children is
required to be cooperative. He has done nothing to stop the addressee’s thinking that John has no more
than three children and so intends the addressee to think that. So the speaker has implicated that John
has no more than three children.
3I assume that “what is implicated” is not a logical consequence of “what is said.” Here informative-
ness is narrowly understood so as to satisfy the condition that if φ is more informative than Ψ, Ψ does
not entail φ, and φ is neither logically true nor logically false. See Atlas (1975a,b), Harnish (1976: 362,
n.46), O’Hair (1969), Quine and Ullian (1978: 68), and Smokler (1966).
Maxim of Quality (namely, “Do not say what you do not know”) be taken not to
know whether John has five children. Thus from the fact that the speaker has not
asserted the stronger variant, it will be inferred that he does not know whether the
stronger variant is true. Gazdar (1979a) argues that in the case of Horn Scales, it will
be inferred that the speaker knows that the stronger variant is false.
If a similar explanation were to be given for negatives, we should posit a logical
scale (6) where choice negation –φ precedes exclusion negation ¬φ:4
(6) <not –φ, not ¬φ>
Then there should be two scalar implicatures: (a) if a speaker asserts an exclusion
negation, he implicates the falsity of the choice negation, and (b) if a speaker asserts
the negation of a choice negation, he implicates the exclusion negation. Thus Gazdar’s
(1979a) account predicts that the exclusion negation understanding of (3) will prag-
matically imply (7) and (8).
(3) The king of France is not bald.

(7) The speaker does not know that there is a king of France and that he is non-bald.
(8) The speaker knows that it is not the case that there is a king of France and that he is
non-bald.
Of course, what should be pragmatically implied is (9):
(9) The speaker knows that there is a king of France and that he is non-bald.
One response to the conflict between the apparent pragmatic implications would
be to abandon the Gricean claim that the literal meaning of ‘not’ in English is that of
exclusion negation. One account of negation introduces an updated version of the
traditional scope distinction and identifies ‘not’ with choice negation (Karttunen and
Peters 1979), just as Strawson (1950) did, but it has been argued that this suggestion
has serious defects (Atlas 1980a). An alternative nonclassical account argues that
the literal meaning of the free morpheme ‘not’ is neither a choice negation nor an
exclusion negation but is semantically nonspecific with respect to narrow versus wide
scope interpretations (i.e., between choice and exclusion interpretations; Atlas 1974,
1975a,b, 1977b, 1978a,b, 1979, 1989). But no matter whether classical or nonclassi-
cal semantics is preferable, it will still be necessary to find a pragmatic principle,
different from the one involved in scalar implicatures, that will offer an account of
the inference from (3) to (4)/(9).
4–φ, a choice negation of φ, is true (false) if and only if φ is false (true) for every admissible valuation
val∈V of the language. ¬φ, an exclusion negation of φ, is true if and only if φ is not true, for every
admissible valuation val∈V of the language. In a bivalent language exclusion-negation and choice-
negation functions are extensionally identical. In a non-bivalent language—one with truth-value gaps—
they are extensionally distinct.
An obvious problem to be solved is the inconsistency between such an infer-

ence and the typical Gricean arguments involving the First Maxim of Quantity, that
is, the epistemic inconsistency between the First Maxim of Quantity inference from
(3) to (8) and the intuitive inference from (3) to (9). The difference between the in-
ferences from the scalar expressions and from ‘not’ statements that we have described
is a general difference between kinds of pragmatic inference for two classes of ex-
pression. The Gricean inference from the First Maxim of Quantity accounts for one
class but not for the other.
I am concerned with the pragmatic principles that could be used to explain how
and why what is conveyed or communicated by an utterance is more definite or more
precise (or, what is not the same, more specific; see my discussion of these terms
later in this chapter) than the literal, or conventional, meaning of the sentence ut-
tered.5 For convenience in exposition, I follow Grice (1961, 1967, 1975a,b, 1978) in
identifying “what is said” with the sense of “the statement,” that is, with the truth
conditions that determine the truth-value of the statement.6 Where I intend to refer
to those inferences falling under maxims of conversation I shall speak of “conver-
sational implicatures.”7 The conjunction of “what is said” with “what is implicated”
will be “what is communicated,” the meaning a speaker conveys, what Grice (1989a:
118–20) called “the total signification of the utterance.”
I am interested in data in which “what is said” is augmented by generalized
conversational implicata so that “what is communicated” is standardly more infor-
mative than “what is said” (see Bach 1995). Here are familiar examples in which the
(b) sentences are implicata of asserting the (a) sentences (Gazdar 1979a; Horn, 1972,
1973; Grice, 1961, 1967, 1975a,b). Examples (10b1), (11b1), (14b1) are scalar; (12b2),
(13b1), and (14b2) are so-called clausal; (15b1) is negative scalar; and (16b1) is nega-
tive clausal.
(10) a. Some of the boys are at the party.

b1. Not all of the boys are at the party.
(11) a. Morton has three children.

b1. Morton has no more than three children.
5I do not identify the literal with the conventional. See Lakoff (1986).
6This is carefully put. What determines a truth-value does not have to be identical with a proposition,
or a Fregean Gedanke, although it is typically thought of as expressing truth conditions. Furthermore,
it presupposes that the spoken utterance, or written inscription, has a truth-value. Sense, in Frege’s or
in Grice’s usages, is certainly not the same as the linguistic meaning of the sentence-type of which the
statement is a token (see Burge 1990). For more on ‘what is said’, see Grice (1989a: 24–25, 87–88,
118–22). Had Grice not stuck me with it, I would have followed the advice of Paul Ziff (1972a), in his
“What Is Said,” and abjured in Atlas and Levinson (1981) the phrase ‘what is said’ as philosophically
obscurantist and conducive to a debate that confuses disagreement between theories with differences
in terminologies. The notion of “what is said” has since come in for considerable discussion and no-
table disagreement among philosophers and linguists: See Bach (1994a), Carston (1988), Levinson
(1988, 2000), Récanati (1989), and Sperber and Wilson (1986b).
7Grice uses “conversational implicature” in the narrow sense for inferences from floutings, or blatant
violations, of the maxims.

(12) a. Rick is a philosopher or a poet.

b1. Rick is not both a philosopher and a poet.
b2. Rick may not be a philosopher.
Rick may not be a poet.
(13) a. If John is at home, the brain machine will be on.

b1. John may not be at home.
The brain machine may be on.
(14) a. Marjorie believes that Babette is a Phi Beta Kappa.

b1. Marjorie does not know that Babette is a Phi Beta Kappa.
b2. Babette may be a Phi Beta Kappa.
Babette may not be a Phi Beta Kappa.
(15) a. Not all of the boys are at the party.

b1. Some of the boys are at the party.
(16) a. It’s not the case that Rick is both a philosopher and a poet.
b1. Rick is either a philosopher or a poet.
These implicata limit “what is said” by shrinking the range of possible states of affairs
whose descriptions are consistent with “what is said” to a smaller range of those states
of affairs whose descriptions are consistent with “what is communicated”—that is,
“what is said” plus “what is implicated.” “What is communicated” is more definite
than “what is said.” I shall argue that these more definite propositions are derivable
by the Gricean inference from the First Maxim of Quantity.
Other pragmatic inferenda enrich “what is said” by reshaping the range of the
possible states of affairs in the truth-set of “what is said” to a narrower range of pos-
sible states of affairs in the truth-set of “what is communicated.” “What is communi-
cated” is either (a) more precise or (b) more specific than “what is said.”
The Atlas and Levinson (1981: 35–36) contrast between the definite and, to coin
a term, the precific (a genus term), corresponds roughly to the Sperber and Wilson
(1986b) contrast between the implicatum (a proposition that the addressee understands
to have been conveyed by the speaker in stating the proposition that the addressee
understands him to have expressed in the sentence he uttered) and the explicatum
(the proposition understood by the addressee to have been expressed by the speaker
in asserting the sentence). With important qualifications to be noted, my (1981)
subcontrast in the “precific” between the precise (a species term) and the specific
(Levinson’s 1987a term) is roughly approximated by Récanati’s (1989) contrast
between the strengthened and the saturated and by Bach’s (1994a,b) contrast be-
tween the expanded and the completed among implicitures (not implicatures).
What is understood by the addressee to have been communicated to him via the
proposition [YOU ARE NOT GOING TO DIE], expressed by the speaker’s utterance You are
not going to die, would be the latter conceptually strengthened, or fleshed out, by
the addressee: [YOU ARE NOT GOING TO DIE FROM THIS CUT ON THE FINGER]. The speaker
could have been more explicit if he had inserted the words from this cut on the finger
into the utterance. Bach (1994a) calls the addressee’s making-what-is-meant-by-the-

speaker explicit “expansion”—a filling-out of the proposition understood by the
addressee to have been uttered by the speaker (see also White 1965: 59, 1993; Atlas
1989: 25). By contrast, “completion” is an addresse’s filling-in of a nonpropositional,
semantically nonspecific sentence-meaning of a sentence uttered by a speaker to yield
an understood proposition (see also Atlas 1974, 1975b, 1977b, 1979, 1989). The
source of the semantical nonspecificity in a sentence might be lexical—such as take
(see Ruhl 1989: 87; Cruse 1992)—or phrasal, as in the genitive John’s book. The
genitive phrase is traditionally taken to be ambiguous (e.g., by Chomsky 1972a);
Leech (1974) and, following Kay and Zimmer (1976: 29), Récanati (1989: 298, Davis,
1991: 99) take the genitive to be univocal but logically weak, merely expressing the
existence of a relation ρ between John and the book, as in [THE BOOK THAT BEARS SOME
ρ TO JOHN], rather than, as by Atlas, semantically nonspecific.
Views on univocality similar to Leech’s, Kay and Zimmer’s, and Récanati’s were
held by Grice, the classical Griceans, and Kempson and Cormack (1981). By con-
trast Bach (1994a: 129–30) and I hold that such words, phrases, and sentences are
semantically underdeterminate—that is, semantically nonspecific with respect to a
semantic feature, such as [–MALE], with respect to a constituent, as Gentlemen prefer
blondes [to brunettes], or with respect to structure, as The king of France is not bald,
McFee almost shot himself last night in the park in a plastic bag. (For discussion of
the differences among semantical nonspecificity, weak semantical univocality, el-
lipsis, and indexicality, see Atlas 1977b, 1978b, 1979, 1984b, 1989 and Bach 1982,
1994a: 130–33).)
My main concern has been with structural semantical underdeterminacy: se-
mantical nonspecificity of sentences containing quantified noun phrases, definite
descriptions, adverbial modifiers, numerical adjectival modifiers, and extensional
(e.g., ‘not’) and intensional (e.g., ‘believes’) operators. (For an attack on the ambi-
guity of ‘believe’ sentences, see Bach 1987: 210–14 and Stich 1983: 111–23.) Struc-
tural nonspecificity has been my concern as a philosophical logician because my
interest is in the logical and semantical structure of natural language (both classical
logical syntax and model theory, as well as what Chomsky 1996c calls “Internalist
Semantics”) and its relation to thought and to what Chomsky calls “Performance
Systems.” These structural questions, the most difficult and subtle kind of semantical
nonspecificity, are the ones that have the most significant bearing on logical form.
Some examples of generalized conversational inferenda in (19)–(27) and par-
ticularized conversational inferenda in (17)–(18) follow. Examples of precisification
are (17b1), (18b1), (19b1), (21b1), (22b1), and (23b1). Examples of specification are
(20b1), (24b1), (25b1), (26b1), and (27b1).
(17) Asymmetric conjunction (Schmerling 1975)

a. Kurt went to the store and bought some wine.
b1 Kurt went to the store in order to buy some wine and then bought some wine.
(18) Conjunction buttressing (Atlas and Levinson 1981)

a. Mart turned the switch and the motor started.
b1. First Mart turned the switch and then the motor started.
b1. Mart’s turning the switch indirectly caused the motor’s starting.
b3. Mart’s turning the switch directly caused the motor’s starting.
b4. Mart’s intentionally turning the switch caused indirectly/directly the motor’s
starting.
(19) Conditional perfection (Geis and Zwicky 1971)

a. If you mow the lawn, I’ll give you five dollars.
b. If you don’t mow the lawn, I won’t give you five dollars.
(20) Mirror maxim (Harnish 1976: 359)

a. Mart and David moved the cabinet.
b1. Mart and David moved the cabinet together.
(21) Continuity (Atlas and Levinson 1981)

a. Mikael ate the cake.
b1. Mikael ate the whole cake.
(22) Continuity (Atlas and Levinson 1981)

a. Eve ate the apples
b1. Eve ate all the apples.
(23) Membership categorization (Sacks 1972)

a. The baby cried and the mother picked it up.
b1. The baby cried and the mother of the baby picked it up.
(24) Bridging/definite reference (Clark and Haviland 1977;

Hawkins 1975, 1978)
a. It was a vase made of bronze and on the base of the vessel was the maker’s mark.
b1. It was a vase made of bronze and on the base of the vase was the maker’s mark.
(25) Inference to stereotype (Atlas and Levinson 1981)

a. Mikael said “Hello” to the secretary and then he smiled.
b1. Mikael said “Hello” to the (female) secretary and then he (Mikael) smiled.
(26) Negative specification (Atlas 1975a,b, 1977b; Horn

1978b, 1989)
a. The largest prime integer is not even.
b1. The largest prime integer is {not-even/odd}
b2. There is no largest prime integer that is even [or odd].8
8I did not paraphrase (26b ) as: It is not the case that the largest prime integer is even, since I showed
2
in Atlas (1974, 1977b, 1989) that this natural language version of the external sentence negation actu-
ally has just the same interpretations that (26a) does, contrary to decades of philosophical doctrine (see
Boër and Lycan 1976; Horn 1989). And one needs the “expansion” or odd to avoid an implicatum of
(26b2): There is a largest prime integer that is odd.
Negative incorporation (Horn 1989)

c. I do not like N.N.
c1. I dislike N.N.
Negative lowering (Horn 1989)

d. I do not believe Tom is tall.
d1. I believe Tom is not tall.
(27) Preferred local coreference (Levinson 1987a,b)

a. Tom came in and he sat down.
b1. Tom1 came in and he1 Tom1 sat down.
I shall argue that there is a general principle that licenses an inference from “what
is said” to the more precise or specific content of “what is communicated” even though
the particular ways in which the inferred proposition is constructed may differ from
case to case.9 I am interested in understanding the character of this inference from
informativeness. But first I shall discuss the Gricean inference from the First Maxim
of Quantity.
1.1 The inference from the first maxim of quantity and

its limitations
The implicata in (10)–(14) are alleged to be derivable by appeal to Grice’s First Maxim
of Quantity, namely, “Make your contribution as informative as is required for the
current purposes of the exchange.” A prototypical Gricean argument for this class of
implicatures goes as follows (Grice 1975a: 50):
(28) a. The speaker S has said φ.

b. There is a proposition Ψ, related to φ by virtue of entailing φ and/or by being more
9It is interesting to apply Kent Bach’s (1994a,b) taxonomy to examples (17)–(27). His notion of “ex-
pansion” implicitures covers the particularized conversational inferendum of Conjunction Buttressing

of (18b1), as well as the generalized conversational inferenda of the Mirror Maxim (20b1), Continuity
(21b1, 22b1), Membership Categorization (23b1), and Inference to Stereotype (25b1). His notion of
“completion” implicitures covers Negative Specification (26b1). Bridging/Definite Reference (24b1)
and Levinson’s (1987a,b) Preferred Local Coreference (27b1) are put by Bach into a separate category
of reference determination. Conditional Perfection (19), Conjunction Buttressings (18b2, 18b3, and 18b4),
and Negative Incorporation (26c1) are not implicitures but are particularized or generalized implicatures
on his view; he presumably does not accept my semantical nonspecificity of ‘not’ in (26c), although
he does elsewhere (see Bach 1987: 99n.). On my view a specification of (26c) is I do not-like N.N.,
which then, in a process of negative incorporation, becomes I dislike N.N. Asymmetric Conjunction
(17) does not fit Bach’s taxonomy, since he sees no generalized impliciture or implicature in that case;
I agree that the implicata are particularized. The Inference to Stereotype example (25b1) could also be
regarded as a lexical completion, as well as an expansion, since ‘secretary’ is nonspecific for semantical
gender. That last point raises the question of how well defined completion and expansion are: Are they
intended to be mutually exclusive? Does it matter for Bach’s account if they are not? And how much
does Bach’s notion of expansion depend merely on adding to surface structure, without consideration
of underlying semantical properties?
informative than φ, which it would be desirable to convey in view of the current

purposes of the exchange. (Here there is reference to the Maxim of Relation, “Be
relevant.”)
c. Proposition Ψ can be expressed as briefly as φ, so S did not say φ rather than Ψ sim-
ply in order to be brief—that is, to conform to a Maxim of Manner.
d. So S must intend the hearer to infer not Ψ or at least It’s not the case that S knows
that Ψ, for if S knew that Ψ, he would have infringed the First Maxim of Quantity
by saying φ.
e. Therefore, saying φ implicates not Ψ or at least It’s not the case that S knows that
Ψ.
Various versions of this argument have been rehearsed by Gazdar (1979a),

Harnish (1976), and Horn (1972). Schema (28) will suffice to represent these vari-
ous arguments. For purposes of our discussion, the salient feature of such an argu-
ment is its derivation of implicata from what is not said. Given that there is available
an expression of roughly equal length that is logically stronger or more informative,
the failure to employ the stronger expression conveys that the speaker is not in a
position to employ it. The inference will always result in a delimitation of what has
been said, in a more definite proposition being conveyed as “what is communicated.”
The argument relies crucially on the existence of equally brief expressions that
can be ordered in a Horn Scale of relative informativeness. When the items in the
scale are elements in a semantic field (see Grandy 1987), and where alternatives are
psychologically salient, the stronger inference to The speaker knows that the more
informative alternatives do not obtain is licensed. These are the well-known scalar
implicatures illustrated in (10b1), (12b1), and (14b1) and formalized by Gazdar (1979a:
58–59), relying partly on the work of Horn (1972).10
In other cases, the assertion of φ will implicate that the speaker is not in a position
to assert a stronger, more informative statement Ψ. Instead of an inference from φ to S
knows that not Ψ, as in the scalar cases, there is an inference from φ to S does not know
that Ψ and so to It’s compatible with what S knows that not Ψ. Some of these cases
have been formalized by Gazdar (1979a: 59) as “clausal implicatures.” These will
allegedly arise when a compound or complex sentence φ—for example, Ψ ∨ χ—
has a constituent sentence Ψ such that φ entails neither Ψ nor not Ψ and, on Gazdar’s
theory, presupposes neither as well. As there is usually a similar assertion that would
entail Ψ (e.g. Ψ itself), or entail its negation, the speaker is presumed not to know whether
Ψ is true. This theory allegedly accounts for (12b2), (13b1), and (14b2).
10It seems to have been assumed in the literature that only Horn Scales give rise to these strong
implicatures, but other types of cases exhibit the same behavior. The implicatum of Jane’s skirt is blue
(Harnish 1976) is not The speaker doesn’t know whether the skirt is any other color but rather The
speaker knows that the skirt is not any other color. Similarly, to say Jones is a doctor is to imply The
speaker knows that Jones is not (e.g.) an architect rather than The speaker does not know whether Jones
is an architect. It may be sufficient that a set of lexical items be “about” the same domain and provide
presumptively exclusive alternatives of equal saliency in order for the stronger implicata to obtain. The
theory of such matters remains highly underdeveloped.
For example, Levinson writes, “for the clausal alternates <(since φ, Ψ), (if φ,
Ψ)>, the use of the weaker conditional stands in opposition to the use of construc-
tions that would entail the embedded sentences (e.g., since φ, Ψ)” (2000: 36). The
scale <since, . . . , if . . .> is supposed to parallel <all, some>. Levinson (2000: 36 ex.
(15b)) claims that the following example exhibits clausal implicata:
“If there is life on Mars, the NASA budget will be spared.”
» {There may be life on Mars; There may not be life on Mars.}
Now, exactly how do we use the set of salient alternates as in the scalar cases to get the
clausal inference? How does the contrast between if and since give us {There may be
life . . . ; There may not be life. . . .}? From the assertion of the conditional, do we get
a scalar-like implicatum It is not the case that since there is life on Mars, the NASA
budget will be spared? And if so, how do we go from that implicatum to There may be
life on Mars? Levinson’s (2000: 36) suggestion is that the implicatum of asserting if φ
is φ is uncertain, since we have a contrast with since φ, which would entail φ. Is the
mechanism to deny since? That would not produce φ is not true.
If someone asserts If John comes, I’ll go, how does the speaker in general impli-
cate {Maybe John will come; Maybe John won’t come}?—or as Levinson (2000: 36)
puts it, It is uncertain whether John will come. Does the use of the if-clause by the
speaker generally implicate that it is epistemically contingent that John comes—that
as far as the speaker knows, it is possible that he will come and possible that he won’t?
Even if an explanation for the speaker’s asserting the conditional if φ, Ψ rather than
φ, so Ψ or Ψ since φ is the speaker’s uncertainty that φ, is Maybe φ, maybe not
φ what the speaker normally conveys, suggests, or implies by asserting if φ, Ψ?
I doubt it. There are plenty of cases in which the truth-value of the antecedent
of the conditional is certain for the speaker, as in If 7 + 5 = 13, then I’m a monkey’s
uncle. One does not have the speaker implicating that he is uncertain that 7 + 5 =
13; he is quite certain that it is not. So there is no clausal implicatum Maybe 7 + 5
= 13 and also no implicatum Maybe 7 + 5 ≠ 13. Consider some other examples:
(a) If you want dessert, you’ll have to eat your spinach; (b) If the president is a
genius, then I’m a monkey’s uncle; (c) If 2 + 2 = 5, then 0 = 1; (d) If 7 + 5 = 12,
then 7 + 6 = 13; (e) If Verdi were French, then he and Bizet would be compatriots.
In cases (b) to (e) it seems clear that a speaker, at least myself, is not uncertain
whether the antecedent clause is true; the clauses are not epistemically contingent.
The truth-values of the antecedent clauses of (b) to (e) are certain for the speaker,
at least for myself: false, false, true, false. In the case of example (a), in the normal
circumstances of utterance, the speaker is likewise in no doubt that the antecedent
clause is true—the parent expects the child to want dessert—but it could be that, in
a particular context of utterance, as far as the speaker actually knows, the anteced-
ent clause is false. The parent does not know that the child wants dessert. Yet that
is not the normal case of use of (a). That use is You’ll get dessert only if you eat
your spinach.
I certainly do not think that asserting those conditional sentences conveys, gen-
erally, that the antecedent clause is uncertain for the speaker or that it is possible as
far as the speaker knows that the clause is true or that it is possible as far as the speaker
knows that the clause is false. In short, there is no reason to think that there exist any
generalized clausal implicata of conditionals, pace Gazdar (1979a) and Levinson

(2000).
The data of (15) and (16) seem to require a more elaborate theory (q.v. Horn
1972). It is tempting to hypothesize, after the fashion of Generative Semantics,
Bertrand Russell, or Fred Sommers (1982), that not all derives from an underlying
some are not and none derives from an underlying all are not (and possibly further,
from an under-underlying not some). Then by the usual scalar implicature, saying
not all (i.e., some are not) implicates not all are not (i.e., not none), which by double
negation elimination is equivalent to some. Thus the pragmatic quantity scale, or-
dering “deeper” or otherwise “designated” readings, motivates a particular hypothe-
sis about semantic representations and about meaning postulates for lexical items
like ‘not’ in the mental lexicon. A more “surfacey” alternative would be the positing
of scales of negative items: <impossible, improbable/unlikely, . . .>, and <none/no,
. . . , not all, . . .>. But then one would have to give criteria for an item being a “nega-
tive” one and theorize about the difference between lexicalized and periphrastic nega-
tive items in a Horn Scale.11
The implicata of (17) to (27) are not derivable by the inference from the First
Maxim of Quantity. Indeed, the inference from the First Maxim of Quantity yields
results inconsistent with the data of (17) to (27). I have already discussed ‘not’ sen-
tences. Another example is Conditional Perfection. For example, saying (19a) intu-
itively implicates (19b1), and thus communicates the conjunction of (19a) and (19b1),
given in (19c).
(19) a. I’ll give you five dollars if you mow the lawn.
b1. If you don’t mow the lawn, I won’t give you five dollars.
c. I’ll give you five dollars if, and only if, you mow the lawn.
But by the First Maxim of Quantity, the speaker should have said the stronger sen-
tence (19c). As the speaker has not said (19c), the hearer must be intended to infer its
denial. Therefore, according to the inference from the First Maxim of Quantity, (19a)
implicates either its own falsehood or the falsehood of its intuitive implicatum (19b1).
Saying (19a) cannot implicate (19b1) through an inference from the First Maxim of
Quantity, but there is an implicature nonetheless.12 We must explain the data by ap-
11For more on the introduction of negative scales, see Atlas and Levinson (1981: 39, n.9), and on nega-
tive items, see (Atlas 1991a, 1993, 1996a,b, 2001). See also Sommers (1982: 61, 290, 360) and
Englebretsen (1996).
12Sentence (19b ) is only an implicatum; not all conditionals convey a biconditional, indicating the
1
defeasibility of the inference. Compare:
(i) I have a key in my pocket if the door is locked.
(ii) I have a key in my pocket if, and only if, the door is locked.
Sentence (ii) is not communicated by saying (i) because (ii) is incompatible with noncontroversial
presumptions and so is blocked, a mechanism formalized for background presumptions in Gazdar
(1979a). For alternative analyses of the “conditional perfection” inference, see J. van der Auwera (1995).
For discussion of the difference between “background” and “noncontroversial” presumptions, see the
following discussion and chapter 4.
peal to another form of argument, one that yields interpretations that supplement “what
is said” by positing that “what is meant” is a stronger proposition compatible both
with presumptions in the context and with “what is said.”
1.2 Informativeness
To account for the data, I formulated in Atlas and Levinson (1981: 40) the conversa-
tional maxims of relativity:
(29) Maxims of relativity
Speaker-centered
1. Do not say what you believe to be highly noncontroversial—that is, to be entailed
by the presumptions of the common ground in context K.
Hearer-centered
2. Take what you hear to be lowly noncontroversial—that is, consistent with the pre-
sumptions of the common ground in context K.
The essential notion here was “noncontroversiality,” which I contrast with what
is already presumed in the background of the context of utterance. So something
needed to be said about what noncontroversiality might consist of. I elaborated the
notion of noncontroversiality as follows:
(30) Conventions of noncontroversiality (among which are)

1. Convention of intension (common knowledge)
The obtaining of stereotypical relations among individuals is noncontroversial for
hearer H and speaker S in context K with respect to a statement A . (Atlas and
Levinson 1981: 40)
For example, if a predicate π(x) is semantically underdeterminate—semantically

nonspecific—with respect to semantical metapredicates Fi, 1 ≤ i ≤ n, that express
aspects of lexical meaning—for example, semantical gender, but for some k, 1 ≤ k
≤ n, a species fk of the semantical genus Fk expresses a stereotypical property of x,
for example as [FEMALE] expresses femaleness, then in saying π(τ) a speaker will
convey π(τ)fk to the addressee in accordance with the Second Maxim of Relativ-
ity and the Convention of Intension. This is illustrated by sentences (25a) and (31a)
to (32a), the assertion of which communicate (25b1), (31b), and (32b), generalized
implicata that are more informative than “what is said”:
(31) a. The secretary smiled.

b. The female secretary smiled.
(32) a. John had a drink.

b. John had an alcoholic drink.
Returning to the Conventions of Noncontroversiality, we have:
(30) 2. Convention of extension (Quine’s 1956/ 1976: 190

exportation)
If the statement A(t) is “about” t13, or if t is a topic noun phrase in the state-
ment A(t), then:
a. If t is a singular term, t exists is noncontroversial for speaker S in context K
with respect to A(t).14
13The semantic aspects of the intuitive notion of ‘aboutness’ that I am employing have been in part
explicated in Putnam (1958). Generalizing from a suggestion of Popper (1959: 122), I shall say that if
a statement φ is “about” the set σ and a statement Ψ is “about” the set β, φ is more informative than Ψ
if β is properly contained in σ [β ⊂ σ]. For example, All birds have wings and All crows have wings are
“about” birds and crows, respectively, but not “about” winged creatures, and the first statement about
birds is more informative than is the second about crows. On Putnam’s explication of “aboutness,” φ is
“about” σ if and only if Ψ is “about” σ, provided that φ and Ψ are logically equivalent. All birds have
wings and All crows have wings are also “about” the nonwinged, as they are equivalent to All nonwinged
things are nonbirds and All nonwinged things are noncrows. Thus the sentences may be taken to be
“about” the set-theoretic union of birds and the nonwinged and “about” the set theoretic union of crows
and the nonwinged, respectively. Because the former set properly contains the latter, by Popper’s cri-
terion All birds have wings is the more informative. It is also a feature of Putnam’s account that a sen-
tence and its negation are “about” the same thing and that φ(a) is “about” {a}.
A further feature of the commonsense notion of “aboutness” that is worthy of mention is its
intensionality. This is indicated by the nonreferential occurrence of t in φ is “about” t. The sen-
tence All winged horses are unridable is pre-theoretically “about” winged horses; All golden moun-
tains are unclimbable is “about” golden mountains. Of course, the extension of the predicates ‘winged
horse’ and ‘golden mountain’ is the same: the null set ∅. But intuitively, that is intensionally, All winged
horses are unridable is not “about” golden mountains, nor is All golden mountains are unclimbable
“about” winged horses.
The defeasible inference from φ is “about” t to ∃x(φ is “about” x) is an inference dubbed
“exportation” by W. V. O. Quine (1956/ 1976: 190). Quine, in a happy choice of terminology, called
his inference “implicative.” However logically dubious the existential “aboutness” inference is, it is
dubious in precisely the way Quine’s exportation is dubious. Quine’s classic example of exportation is
that inference from Ralph believes that Ortcutt is a spy to Ralph believes u(u is a spy) of Ortcutt: namely,
Ralph believes spyhood of Ortcutt, from which it follows, assuming one treats English proper names as
individual constants in classical first-order logic, (∃x)(Ralph believes u(u is a spy) of x): namely, Someone
is believed by Ralph to be a spy (1956/ 1976: 190). Our need for ∃x(φ is “about” x) is as pressing as
the need Quine recognizes for “relational” statements of belief. By the Convention of Extension, the
exported existential proposition is a matter in the context of utterance of (a) background presumption
or of (b) noncontroversiality. The proposition does not need to be true; it merely needs (a) to be pre-
sumed or (b) needs to be taken for granted by the parties to the discourse. It is their propositional atti-
tudes that affect how utterances in the context will be understood. (For other philosophical uses of the
topic/comment (aboutness/predication) distinctions, see Russell 1903 and Atlas 1988, 1989, 1991a,
1993, 1996b.)
14Gazdar dismisses a similar idea. He writes, “Naturally one can add to Grice’s maxims, perhaps along
the lines of: Assume referents exist unless you know they don’t, but then one can always invent no less
unreasonable sounding conversational maxims to deal with any example at all” (1977: 127). The re-
semblance between my suggestion and Gazdar’s strawman is only superficial. I do not conceive the
Convention of Extension as a “maxim” of conversation at all. It is part of a theory of context and
noncontroversiality. Its role is emphatically not that of a conversational maxim. Its acceptability will
rest on its value within such a theory of context and on its contribution to a neo-Gricean theory as a
whole.
b. If t denotes a set, ∃x(x ∈ t) is noncontroversial for speaker S in context K with
respect to A(t).
c. If t denotes a state of affairs or a proposition, t is actual and t is true are
noncontroversial for speaker S in context K with respect to A(t).
The two Conventions of Intension and Extension are theoretical principles of

noncontroversiality—principles of default interpretations—but what is noncontro-
versial in a context K for a speaker S is not necessarily common ground with respect
to a statement in that context, even though what is common ground in a context for
the speaker and addressee is noncontroversial in the context. Common ground in-
cludes already established mutual knowledge or belief, as well as what has already
been mutually taken for granted, whereas the takings-for-granted associated with an
utterance U in a context K can occur simultaneously with the utterance of U. In At-
las and Levinson (1981: 40, n.11), I wrote, “We do not conceive the Convention of
Extension as a ‘maxim’ of conversation at all. It is part of a theory of background
presumption, of noncontroversiality. Its role is emphatically not that of a conversa-
tional maxim.”
The concept of stereotypicality enters into just one of the Conventions of
Noncontroversiality—that is, the Convention of Intension. My maxims of conversa-
tion are the Maxims of Relativity, which are speaker-centered and addressee-cen-
tered revisions of Grice’s Second Maxim of Quantity (“Do not make your contribution
more informative than is required”). The notion of noncontroversiality plays an es-
sential role in the formulation of the Maxims of Relativity, but stereotypicality only
plays a role in one of the postulates of noncontroversiality.
Levinson (2000: 37) comments on Grice’s second maxim that the “underlying
idea is, of course, that one need not say what can be taken for granted.” Indeed,
Levinson’s formulation is one consequence of one construal of Grice’s Second Maxim
of Quantity, because if the speaker does offer more information than is required—
for example, information that is already mutually assumed in the context—then, as
Grice (1989b: 27) himself notes, “hearers may be misled as a result of thinking that
there is some particular point in the provision of the excess of information.” Another
consequence of excessive information is inferrable from Grice’s comments on the
maxims as examples of imperatives governing rational behavior. Grice noted, “If you
are assisting me to mend a car, I expect your contribution to be neither more nor less
than is required. If, for example, at a particular stage I need four screws, I expect you
to hand me four, rather than two or six” (1989b: 28). In Grice’s example the excess
is an obstacle to getting on with the job.
In Atlas and Levinson (1981) I made explicit the relationship between stereo-
typicality and the conversational Maxims of Relativity; the maxims appeal to a
notion of noncontroversiality, and the theory of noncontroversiality appeals to
stereotypicality. Atlas and Levinson wrote, “If a predicate Q is semantically nonspe-
cific with respect to predicates Pi (1 ≤ i ≤ n), but for some j (1 ≤ j ≤ n) Pj is stereotypi-
cal of Qs, then in saying Qt a speaker will convey Pjt in accordance with the
second maxim of Relativity and the Convention of Intension” (1981: 41). This was
then illustrated by the understanding of John had a drink as John had an alcoholic
drink. The noun ‘drink’ is semantically nonspecific with respect to [ALCOHOLIC], but
in saying John had a drink, speaker and addressee know that being alcoholic is ste-
reotypical of what is described by ‘a drink’ (unlike, for example, what is described
by the marked ‘a drink of water’). That common knowledge is part of their common
ground. The second Maxim of Relativity instructs the addressee to take what he hears
to be noncontroversial and consistent with the presumptions of the common ground,
which means, according to the first Convention of Noncontroversiality, to take what
he hears to be noncontroversial and consistent with the presumption that ‘. . . is a
drink’ is understood as ‘. . . is an alcoholic drink’.
As it happens, the stereotypical understandings of the utterances are more infor-
mative than the literal sense of the sentences uttered; they are more specific. Levinson
(2000: 37) takes the heuristic principle “what is expressed simply is stereotypically
exemplified” to be the essence of Atlas and Levinson’s (1981: 40–41) Principle of
Informativeness, saying that a “version of such a [heuristic] principle is given by Atlas
and Levinson (1981), who dubbed it the Informativeness Principle, and following this,
I [Levinson] call the heuristic the I-principle” (Levinson 2000: 38). In the application
of his heuristic, Levinson formulates a gloss as follows: “minimal specifications get
maximally informative or stereotypical interpretations” (2000: 37). This gloss may well
cause misunderstandings of Atlas and Levinson’s (1981) Principle of Informativeness.
In Levinson’s formulation it is unclear whether he thinks (a) that an interpreta-
tion is maximally informative if and only if it is stereotypical, or, more weakly,
whether he thinks (b) that an interpretation is stereotypical if informative, or (c) that
it is informative if stereotypical, or (d) that being maximally informative and being
stereotypical are inclusive or exclusive alternatives. I should have thought that none
of (a), (b), (c), (d) is correct. Atlas and Levinson (1981) is committed to none of them.
How should one characterize addressees’ understandings of utterances that are
more informative in a context than the literal senses of the sentences uttered by a
speaker? The context involves the background of presumptions common to speaker
and addressee, and it involves information that is understood by the addressee to be
noncontroversial for the speaker and so for the addressee’s interpretation of the
speaker’s utterance, about the topic of the utterance by, in, or when uttering the sen-
tence. I proposed in Atlas and Levinson (1981) and do so again here that among the
possible interpretations of the utterance in the context, the “best” interpretation is
the one—and my principle guarantees that there is a unique one if there are any—
that (a) is consistent with the presumptions of the common ground, (b) is consistent
with the propositions about the topic of the utterance that are understood by the ad-
dressee to be noncontroversial for the speaker at the time of utterance, and (c) is the
logically strongest among those satisfying the preceding conditions (a) and (b). Thus
the best interpretation for the addressee of a speaker’s utterance U in context K is as
informative a proposition as consistency with the common ground and consistency
with the noncontroversial propositions associated with U permit.
In the common ground and among the noncontroversial propositions are predi-
cations of stereotypical properties of objects being talked about in an utterance. Thus
stereotypicality is a constraint on interpretations of utterances, by making the most
informative interpretation consistent with common knowledge of stereotypical prop-
erties of objects. The analysis does not identify the most informative interpretation
with one that expresses stereotypical relations or properties, but it does constrain the
most informative interpretation to be the logically strongest interpretation that is

consistent with whatever stereotypes the speaker’s and addressee’s presumptions of
common ground and presumptions of noncontroversial information contain.
As I use it, ‘stereotype’ is a technical term, derived from Hilary Putnam’s (1975)
discussion in “The Meaning of ‘Meaning’” of natural kind terms, such as gold, water,
and tiger. Stereotype-predicates in the entry in the mental lexicon for the word tiger
include [TAWNY WITH BLACK STRIPES], [LARGE CAT], [CARNIVOROUS]. As is well known,
the related notion of a prototype has been developed by Rosch (1977) and Lakoff
(1987), among others.
Atlas and Levinson’s (1981) account is not captured by the brief description in
Levinson (2000: 37) that “brief and simple expressions thus encourage, by this heu-
ristic, a tendency to select the best interpretation to the most stereotypical, most ex-
planatory exemplification,” or in Levinson’s remark (2000: 40) that “I-inferences
[are] based primarily on stereotypical presumptions about the world.” The latter claim
is just false.
So what was Atlas and Levinson’s (1981) Principle of Informativeness? It was
a technical formulation of the notion of “best interpretation” for an addressee of a
speaker’s utterance in a context. I shall repeat it here in the form that I gave it in
Atlas and Levinson, with some clarifying emendations:
(33) Principle of informativeness

Suppose a speaker S addresses a sentence A to a hearer H in a context K. If H has n
competing interpretations U i (1 ≤ i ≤ n) of A in the context K with . . . information
contents INF(Ui), and GHA,K is the set of propositions that H takes to be noncontro-
versial for S in K with respect to A at the stage in the conversation at which A is ut-
tered, then the “best” interpretation U* of A for H in K is the most informative
proposition among the competing interpretations Ui that are consistent with the com-
mon ground CGK in the context and with the noncontroversial propositions GHA,K as-
sociated with the uttering of A in the context K. (Atlas and Levinson 1981: 40–41)
This “best” interpretation I called ‘the pragmatic content of A’—PRON(A) in the

notation of Atlas and Levinson (1981: 41). Informational contents INF had several
explications. Among the ones discussed in Atlas and Levinson (1981), I chose a
variant of the logically most straightforward, the set of a statement’s logical conse-
quences. So the “best” interpretation of A for the addressee is the logically strongest
proposition that is consistent with what is noncontroversial for the addressee H in K
with respect to A, namely with GHA,K, and consistent with the common ground CGK.
The notion of “competing interpretations” was left as a primitive notion in the
formulation of Atlas and Levinson’s (1981) theory. It was a complex function of the
literal meaning of the sentence uttered, stress, tone, and other aspects. The context
will enter to fix pronoun reference, and so on.
The principle permits the literal meaning of the speaker’s sentence φ to differ
from—and in particular to be logically weaker or semantically less informative than—
any of its hearer’s interpretations φui. This allows a speaker’s production strategy
correlative with the addressee’s comprehension strategy (Atlas 1975a): Literally say
less if you can count on the addressee to compensate for the less said by interpreting
more. Or, as Grice (1975a, 1989a) rather inexplicitly puts it in his Second Maxim of
Quantity, “Do not make your contribution more informative than is required.”
In Atlas and Levinson (1981: 40–41) I adopted a two-level explication of the
best interpretation of an utterance φ. The first level was INF(φU*); the second level
was PRON(φ). INF gave me the logically strongest interpretation consistent with the
common ground, and PRON gave me the strongest INF that was consistent with the
noncontroversial propositions about the topic of the utterance. Thus the “best” inter-
pretation of φ was PRON(φ).
The construction I used in Atlas and Levinson (1981) for the intuitive notion of
the “best interpretaiton” went like this: Let φu* be φuj for the least j, 1 ≤ j ≤ n, such
that INF({φuj} ∪ GK(φ)) = maxi INF({φui} ∪ GK(φ) ), 1 ≤ i ≤ n. The sentence φ will
tend to convey the pragmatic content PRON(φ) to the addressee H: PRON(φ) =
INF({φu*} ∪ Gφu*) where Gφu* is the set of propositions that are noncontroversial in
the context Kφ,that are “about” what φu* is “about,” and that are logically consistent
with φu*.
The constraint of consistency with the common ground was, in fact, a strong
and conservative constraint (Quine and Ullian 1978: 59, 66–68), as the obtaining of
“stereotypical” properties and relations would be elements of the common ground,
since both speaker and addressee would have the same information about stereotypes
mentally readily accessible to them, and each could be expected to expect the other
to access easily such information.
Two explications of the qualitative concept of a statement’s informational con-
tent have long been familiar to philosophers; they were proposed by Rudolf Carnap
(1942), Carnap and Y. Bar-Hillel (1952), John Kemeny (1953), Sir Karl Popper
(1959), and Carl Hempel (1960). The first explication identifies the informational
content of a statement with the set of its logical consequences—that is, IN(φ) = {Ψ:
φ Ψ}. The second identifies the content with the set of possible falsifiers of the
statement, descriptions of possible states of affairs incompatible with it: CON(φ) =
{Ψ: Ψ ¬φ}. The two views are subsumed under one notion of “semantic content”
in Carnap and Bar-Hillel (1953–54), and a quantitative concept is introduced. The
Carnapian definition given by H. Smokler restricts the classical logical consequence
relation (see Smokler 1966: 207, 210).
The standard application of Grice’s First Maxim of Quantity does not explain
the linguistic data. On Grice’s view, a speaker should tailor the form of his utter-
ances to what he thinks his hearer’s needs or interests in the conversation might be.
If a specification would enable the hearer to satisfy his needs or interests, there is a
presumption that the speaker should issue such a specification in his utterance. If the
speaker fails to be specific, it is assumed that he cannot be (Grice 1975a: 57). Thus
it would be predicted that in saying (25a) or (31a) or (32a) a speaker would not con-
vey that the secretary was female or that the drink was alcoholic.
Temporal, causal, and teleological relations between events are stereotypical
in our “commonsense” conceptual scheme. But (18) and (19) fall under the Max-
ims of Relativity and the Convention of Intension as particularized, not general-
ized, conversational implicata. The (a) sentences of (18) and (19) may be understood
in different ways in different contexts. In any particular context of utterance, the
chosen understanding results from an ‘inference to the best interpretation’, the

understanding that best “fits” both the shared background presumptions in the con-
text and the communicative intentions attributable to the speaker by the addressee
in light of “what he has said,” including what the addressee takes to be noncontro-
versial for the speaker and addressee with respect to the utterance in the context.
We have explicated this notion of best interpretation in our hearer-centered Prin-
ciple of Informativeness (33) (Atlas and Levinson 1981).15
2 The alleged inconsistency of scalar and

clausal implicata
Gazdar (1976, 1979a: 136) had suggested that asserting P or Q scalar-implicated not
both P and Q; for example:
(34) “John did it or Mary did it.” » Speaker knows that not(John did it and Mary did it.)
Asserting P or Q also produced the clausal-implicata of (35), which are inconsistent

with the scalar implicatum of (34):
(35) “John did it or Mary did it.” » {It is compatible with what the speaker knows that
Mary did it; It is compatible with what the speaker knows that John did it.}
That is, Gazdar claims that asserting P or Q implicates {Maybe P; Maybe Q}. The
clausal-implicata of (35) were alleged by Gazdar to be inconsistent with the scalar
implicatum of Speaker’s knowing not(John did it and Mary did it). So, to resolve the
alleged inconsistency, Gazdar (1979a: 132) adopted the hypothesis that the clausal-
implicata took precedence over the scalar-implicata, the former excluding the latter.
Gazdar’s argument committed a modal fallacy, since (It is compatible with what
the speaker knows that P) & (It is compatible with what the speaker knows that Q)
does not entail It is compatible with what the speaker knows that (P & Q). To see
this, let Q be not P. But that fallacious assumption is required to make the alleged
clausal-implicata of (35) inconsistent with the scalar implicatum of (34). So the Gazdar
mechanism of clausal implicata excluding scalar implicata in order to resolve an al-
leged inconsistency was not justified; the supposed implicata were not inconsistent.
When Grice (1989b: 26–27) first introduced the Second Maxim of Quantity (“Do
not make your contribution more informative than is required”), he immediately raised
questions about its status: whether it really is an imperative of cooperation, whether
it might be subsumed under “relevance”—and his doubts arise again in his “Retro-
spective Epilogue” when he remarks that “the impact [on implicature] of a real or
apparent oversupply [of information] is . . . problematic” (1989f: 372). (Notice that
Grice fallaciously infers that a speaker’s contribution that is more informative than
the addressee requires for purposes of their successful communication entails a
15Grice’s Second Maxim of Quantity, “Do not make your contribution more informative than is re-
quired,” is the speaker-centered maxim correlative with our hearer-centered maxim of relativity.
speaker’s contribution that is an oversupply for the addressee.) Yet in his original
remarks he indicates just where an important role for the Second Maxim can be played,
when he writes that “there may also be an indirect effect, in that the hearers may be
misled as a result of thinking that there is some particular point in the provision of
the excess of information” (Grice 1989b: 27).
It is just that Grice is looking at “information” from only one point of view.
Grice’s point of view on quantity of explicit information was this: “Don’t say Q & P
rather than P if only P is required.” Another point of view on the maxim would be
this: “Don’t say S & P rather than S if it is evident to speaker and addressee that what
is required, namely, S & P, is inferrable from S (where “inference” is not restricted
to logical deducibility).” In this case, too, if the speaker does provide an excess of
explicit information—perhaps something beyond S—the addressee may be misled,
or, a possibility Grice ignores, he may be correctly directed to the intended interpre-
tation by thinking that there is some particular point in the provision of the excess of
explicit information. Grice had simply ignored the possibility of implicit, evidently
inferrable, information playing a role in the Maxims of Quantity.16
Being more informative also appears in Gazdar (1976, 1979a). For example,
consider his First Maxim of Quantity argument to support his (mistaken) rule for
clausal quantity implicata:
IF one utters a compound or complex sentence [e.g., P or Q] having a constituent

which is not itself entailed or [potentially presupposed] by the matrix sentence and
whose negation is likewise neither entailed nor [potentially presupposed], THEN
one would be in breach of the maxim of quantity if one knew that sentence to be
true or false, but was not known to so know, since one could have been more infor-
mative by producing a complex sentence having the constituent concerned, or its
negation, as an entailment or a presupposition [e.g., P]. It follows that, ceteris pari-
bus, the utterance of such a complex sentence implicates that both the constituent
sentence and its negation are compatible with what the speaker knows. (Gazdar
1979a: 60–61)
Gazdar concludes that the speaker implicates Maybe P and Maybe ¬P.
How can we make Gazdar’s argument perspicuous? Here is one attempt. On the
assumption that the speaker is being cooperative in saying P or Q, that the Speaker
has not said the logically stronger/more informative P indicates that if P would be
informative in the context, the speaker is not in a position to assert P. So the speaker
does not know or believe P, and so ¬P is compatible with what the speaker knows or
believes. By symmetry, ¬Q is compatible with what the speaker knows or believes.
So on this reconstruction of Gazdar’s argument, asserting P or Q implicates Maybe
¬P and maybe ¬Q, but only on the assumption that P, Q would each be appropri-
ately informative in the context. On this reconstruction it does not follow from the
First Maxim of Quantity that asserting P or Q implicates Maybe P and maybe Q (see
Atlas 1990).
16Thiswas a role for informativeness that I had shown to be possible, and in fact necessary, in my (Atlas
1975a) discussions of “presupposition” in Frege and Dummett and in my (Atlas 1977a) discussion of
what David Lewis (1979) later called “accommodation.” See chapter 4 in this volume.
On a similar reconstruction of the argument for If P then Q, one could conclude

that asserting the conditional sentence clasually implicates Maybe P and Maybe not-
Q if one took the meaning of the English conditional to be the classical truth-func-
tional conditional. But there is no reason to hold such a view. In section 1 I argued
that there is no generalized conversational implicatum from assertions of the English
conditional either to Maybe P or to Maybe not-P.
The first problem, then, is that Gazdar’s Gricean argument does not support his
claim in (35). The second problem, already discussed, is that even if it did, there would
not be an inconsistency with the scalar implicature of (34).
But the quantifiers present a different case. For instance:
(36) Some, if not all, of the students were there.
It is alleged (Gazdar 1979a: 136) that asserting (36) scalar-implicates Speaker knows:
Not all of the students were there but clausal-implicates It is compatible with what
the speaker knows that: All of the students were there. And this is a genuine incon-
sistency. Of course, this analysis of clausal-implicata assumes, and Gazdar (1979a:
136) is explicit about it, that the semantic representation of (36) is the same as that of
(37):
(37) If not all of the students were there, some of them were.
Gazdar (1979a: 137) remarks that “this assumption is not critical,” but, in fact, it is
critical. The assumption is plausible but no more than plausible. For (37), on the
assumptions deployed in Gazdar’s (1979a: 136) reasoning, should be logically equiva-
lent to and, in this case, even a paraphrase (Sommers 1982) of (38):
(38) If some of the students were not there, some of them were.
This is because Gazdar takes not to be a sentential modifier of the if-clause of (37).
But (38) is not synonymous with (36). Sentences (36) and (37) do not have the same
semantic representation. Consequently, ‘if’ and ‘not’ in (36) do not generate the
clausal-implicata of a conditional statement (37) that Gazdar claims that they do, even
if Gazdar were correct, contrary to fact, that conditionals generate clausal implicata.
So there is no inconsistency with the scalar-implicatum Not all of the students were
there, and Gazdar has failed to explain the absence of the scalar-implicatum in (36)
by its inconsistency with an alleged but actually nonexistent clausal-implicatum of
an if-clause, which would take precedence over it and exclude it.
We need a better explanation of the absence of the scalar-implicatum (39) in
assertions of (36):
(39) Not all the students were there.
(36) Some, if not all, of the students were there.
The ‘if’ here is a “concessive” phrase (König 1988, 1991), like the ‘if’ in (40):
(40) He’s guilty of manslaughter if not murder.
It does not contradict the scalar-implicatum Not all of the students were there of some;
the “concessive” if not all does not contradict not all. Rather, the “concessive” if not
concedes the possibility expressed by all but also refuses to exclude not all, so that
the statement is used to implicate (41) or (42):
(41) It’s more than merely possible that all of the students were there.
(42) It’s more than likely that all of the students were there.
In effect, what the if not seems to do, as Horn (1972) noted, is to suspend the scalar
implicatum of (43):
(43) Some of the students were there.
The if not suspends (44) by indicating (implicating) (45), which is consistent with
(46) and is not as strong as to assert (47):
(44) Not all of the students were there,
(45) It’s not inferrable that not all of the students were there,
(46) It’s likely that all of the students were there,
(47) All of the students were there.
Thus Horn’s (1972) description of if not suspension anticipates the kind of analysis
given later by Horn (1985, 1989) for what he, somewhat misleadingly in my view,
calls “metalinguistic negation” but which I (Atlas 1983, 1990, 1996b) prefer to call
“focal negation.” Asserting the contrastively stressed (48) is consistent because not
in the first clause is used by the speaker to reject the corresponding affirmative as-
sertion of unstressed some on the grounds that its generalized conversational default
implicatum—not all—is false.
(48) SOME of the students were not there—ALL of them were.
Thus in asserting the stressed statement (48) the speaker implicates (49) and does
not mean the inconsistent (50) or the inconsistent (51):
(49) It is not felicitously assertible that some of the students were there—ALL of them were
(50) There were students who were not there—all of them were (there),
(51) Not any of the students were there—all of them were.

Notice that (52) is synonymous (on its natural reading) with (53) or with (54):
(52) He’s guilty of a grisly manslaughter if not murder.
(53) He’s guilty of a grisly manslaughter if not a grisly murder.
(54) He’s guilty of, if not a grisly murder, a grisly manslaughter.
This would be impossible if Gazdar’s analysis were correct, for Gazdar’s analy-
sans (55) of (52) is not synonymous with Gazdar’s analysans (56) of (53):
(52) He’s guilty of a grisly manslaughter if not murder.
(55) If he’s not guilty of a murder, then he’s guilty of a grisly manslaughter.
(53) He’s guilty of a grisly manslaughter if not a grisly murder.
(56) If he’s not guilty of a grisly murder, then he’s guilty of a grisly manslaughter.
So Gazdar’s conditional analysis of the concessive if not cannot be correct.
3 The resolution of inconsistent implicata

3.1 Conflicting implicata for negation and conditionals
I wish to resolve these apparent clashes between the Gricean generalized implicata
from the First Maxim of Quantity and Atlas and Levinson’s (1981) generalized
inferenda from informativeness. My strategy is to argue that <not—, not¬> and <if
and only if, if> are not scales properly so-called. If they are not, then there are no
generalized scalar implicatures of the sort just described, and so no clash between
informativeness and the First Maxim of Quantity. Only informativeness actually
applies to these cases. In fact, there are natural and independently motivated restric-
tions to put on Horn Scales.
First, to constitute a genuine scale for the production of generalized scalar im-
plicatures, the stronger item must be lexicalized to a degree equal to or greater than
that of the weaker item (see Levinson 2000: 80). Second, to constitute a genuine Horn
Scale, sentence-frames containing each item in a position on the scale entail sentence-
frames containing those in positions to its right, and all the items are “about” the same
property or relation (Gazdar 1977: 72, 181; 1979a: 57–58).17 The first restriction will
eliminate the generalized First Maxim of Quantity implicatures incompatible with
“negation specification” (26) and with “conditional perfection” (19). There is no
negation scale <–φ, ¬Ψ> because there is no free negative morpheme in English that
17Some caveats to this claim are offered in comments on Levinson scales in section 4 of this chapter.
See also Hirschberg (1985/ 1991) and Matsumoto (1995), and criticism of the latter in Levinson (2000).
yields sentences that mean just choice negation or just exclusion negation or that is
ambiguous between them.18 There is no argument of the form “Since the speaker did
not say The king of France is non-bald, he cannot mean it. And so he knows that it is
false.” Similarly, because there is no unitary lexeme in English like if that standardly
means the same as if and only if (the technical abbreviation iff does not count), there
is no Horn Scale 〈φ if and only if Ψ, if φ then Ψ〉 to support First Maxim of Quantity
generalized conversational implicata.
3.2 Implicatures of negated conjunctions

There are further cases of apparent clashes between First Maxim of Quantity gener-
alized conversational implicata and informativeness inferenda. Saying not (φ and
Ψ) seems to implicate φ or Ψ. Thus saying (16a) It’s not the case that Rick is
both a philosopher and a poet seems to implicate (16b1) Rick is either a philosopher
or a poet. The principle involved is the implication of the weakest item on the Horn
Scale by the denial of a strong one. The scale is <and, or>.19
By contrast, saying It’s not the case that Kurt went to the store and bought some
wine (colloquially, Kurt didn’t go to the store and buy some wine) does not generalizedly
implicate Kurt went to the store or he bought some wine. The inference from the Prin-
ciple of Informativeness results in “conjunction buttressing” (18) in particular contexts.
Sometimes φ and Ψ is informatively understood as φ in order to Ψ and then Ψ. The
two actions described separately in φ and in Ψ are teleologically related as means to end.
The sentence indicates one action under a complex description. The implicature from
the saying of not (φ and Ψ) to φ or Ψ that implicates the possibility of independent
alternatives cannot coherently arise. Indeed, saying Kurt didn’t go to the store and buy
some wine can in a particular context implicate Kurt neither went to the store nor bought
some wine. So, for some φ, Ψ, saying not (φ and Ψ) implicates not φ and not Ψ.
18In fact, the nonexistence of the Horn Scale negation “implicata” is evidence in support of my seman-
tic claim (Atlas, 1975b, 1977b) that not sentences are semantically nonspecific with respect to the choice
and exclusion interpretations and thus univocal rather than ambiguous between them. That is, not φ
literally means neither ¬φ nor –φ. Classical Gricean theorists also make a univocality claim but
identify the sense of not sentences with exclusion negation. Gazdar (1979a: 66) has held the latter
univocality view. Gazdar writes:
Allwood (1972), Atlas (1975[b]), and Kempson (1975: 95–100) have produced a series of
arguments which show that natural language negation is unambiguous, and consequently
that the Russellian and ambiguity invoking Strawsonian accounts of sentences like (3) [The
King of France is not bald.] and (11) [John doesn’t regret having failed.] are equally inade-
quate. . . . None of them have, to my knowledge, been challenged in the literature. (Gazdar
1979a: 66)
The former, semantical nonspecificity, univocality view was articulated in Atlas (1974, 1975a,b, 1977b,
1978a,b, 1979, 1989). Kempson (1975) offered an extensional disjunction of exclusion and choice
negations as an analysis of the univocality of natural language negation that does not capture the con-
cept of semantical nonspecificity (Atlas 1984b), and Allwood (1972) offered a purely wide-scope,
exclusion-negation account of negative sentences.
19Equivalently, the implicatum is predicted from the negative Horn Scale (neither . . . nor, . . . , not
( . . . and . . .) ) where A(W) » –(A(S)); for instance, “not both” » not(neither . . . nor)—that is, or.
However, for Gazdar (1979a: 59) not(P and Q) potentially clausal implicates,
by appeal to the First Maxim of Quantity, It’s possible for all the speaker knows that
P and It’s possible for all the speaker knows that Q. If the inference from informa-
tiveness yields The speaker knows that not-P and The speaker knows that not-Q in
cases like the one in question, the Principle of Informativeness is inconsistent with
Gazdar’s rule.
A simplified version of that rule is phrased informally by Gazdar (1979a: 60) as
follows: X potentially clausal implicates that for all the speaker knows Y, and, for all
the speaker knows, not Y, if and only if Y is a part of X but neither Y nor its negation
is entailed by X. In our example X = not (P and Q) and Y = P, where in the stereotypi-
cal course of things P is necessary for Q, P is a part of not (P and Q), and neither P
nor its negation is entailed by not (P and Q). Thus not (P and Q) potentially clausal
implicates It’s possible for all the speaker knows that P. Apart from Y (or not Y) being
entailed by (or being presupposed by) X, Gazdar’s rule takes no semantic relations
into account; in particular, no consideration is given to semantic relations between
parts of X. But that relation is crucial to the Kurt example earlier.
The same issues arise for potential scalar quantity implicatures. Saying P or Q
implicates not (P and Q). But if one says Socrates is mortal or everyone is mortal, which
is equivalent to Socrates is mortal, does one thereby implicate not (Socrates is mortal
and everyone is mortal), which is equivalent to not (everyone is mortal)? (It seems
most unlikely that a speaker would intend his assertion of the disjunction generally to
convey an obvious falsehood—except for some linguistically subtle doctors of theol-
ogy who do not believe the implicatum is a falsehood.) The contexts in which one could
appropriately employ Socrates is mortal or everyone is mortal may be a little odd; it is
not as if it were a premise for an argument that continues Socrates isn’t mortal; there-
fore, everyone is mortal. No such argument could possibly be sound, although it cer-
tainly has a valid form. The fact remains that, whatever an appropriate context might
be, assertoric use of a disjunctive sentence equivalent to Socrates is mortal is predicted
to implicate a sentence equivalent to Someone is not mortal in the same way that use of
P or Q is predicted to implicate but not both. If it is not obviously false, it also is not
obviously true that there is this generalized scalar implicature.
I take it that it is an open question whether Gazdar’s rules are adequate as they
stand. Properly reformulated in light of the semantic relations between P and Q,
Gazdar’s rules for the sentence not (P and Q) may not yield scalar or clausal impli-
cata that would contradict the implicata derived by the Principle of Informativeness.
3.3 Definite and indefinite noun phrase togetherness cases

As my third class of examples in which the First Maxim of Quantity and the Prin-
ciple of Informativeness apparently conflict, I discuss cases examined by Harnish
(1976). When a speaker asserts (57a) he implicates (57b), and when a speaker as-
serts (58a) he implicates (58b).
(57) a. Russell wrote “Principia Mathematica.”

b. Only Russell wrote “Principia Mathematica.”
(58) a. Russell and Whitehead wrote “Principia Mathematica.”

b. Russell and Whitehead jointly wrote “Principia Mathematica.”
By the First Maxim of Quantity, we may infer that as the speaker of (57a) failed
to be specific where it would be informative and generally useful to be so, the speaker
was in no position to assert (57b). Yet this conclusion conflicts with a stereotypical
relationship between books and authors, the norm of “one author per book” (cf.
Harnish’s 1976: 319 example, Leibniz and Newton invented the calculus). Given that
it is held—for example, by Gazdar (1979a)—that possible implicata inconsistent with
background presumptions are defeated, it is plausible to analyze the defeat of the
First Maxim of Quantity implicatum as one of this kind. However, incorrect infer-
ences from the First Maxim of Quantity will not always be neutralized through the
fortuitous intervention of contextual assumptions—the very ones that are employed
in my Principle of Informativeness. Perhaps we will finally find a real clash between
Gricean maxims and my Principle of Informativeness.
For example, there is a strong intuition that (59b), (60b), (61b), and (62b) are
the preferred interpretations of (59a), (60a), (61a), and (62a):
(59) a. Gilbert and Sullivan wrote “The Mikado.”

b. Gilbert and Sullivan jointly wrote “The Mikado.”
(60) a. Mart and David moved the cabinet.

b. They moved it together.
(61) a. Mart and David bought a piano.

b. They bought it together.
(62) a. Mart and David went to San Francisco.

b. They went there together.
Harnish (1976: 328) argues, correctly I believe, that the (a) sentences are not am-
biguous between “independent” and “cooperative” understandings. The preferred
interpretation is implicated. Harnish (1976: 358) points out that it is not at all clear
that the First Maxim of Quantity can explain this implicatum.
He does not explicitly say what the First Maxim of Quantity implicatum would
be. The hearer might argue that the speaker is in no position to make a relevant “co-
operative” claim, as he did not say Mart and David bought a piano together. Thus
the hearer would infer the “independent” understanding of the sentence. However,
he could also argue that because the speaker did not say Mart and David bought pianos
separately, the speaker was in no position to make that claim. So the hearer under-
stands him to mean the “cooperative” understanding of the sentence. The first Maxim
of Quantity implicatum is not well defined. (In chapter 5 we will again meet this fun-
damental logical problem with the formulation of Grice’s maxims.)
Harnish proposes a Gricean submaxim of manner: namely, insofar as possible,
if objects a, b, c, . . . F together, put their names together when reporting this F-ing.
This maxim is intended as one instance of a more general Grice-type maxim: “Make
your saying mirror the world” (Harnish 1976: 359). But such additional Gricean
maxims and submaxims of manner will not account for the data. For example, the
preferred interpretation of (63a)—in reply to Who took a shower? or in reply to What
did Mart and David do?—is (63b) rather than (63c):
(63) a. Mart and David took a shower.

b. Mart and David took showers separately.
c. Mart and David took a shower together.
In this example the “independent” interpretation is the preferred one. The asser-
tion (63a) conveys (63b), and, given our social norms, (63b) is predicted by our
Second Maxim of Relativity, the Conventions of Noncontroversiality, and the Prin-
ciple of Informativeness. The inference to the best interpretation of the assertion
(63a) yields the “independent” understanding (63b). Because the First Maxim of
Quantity implicata for (59) to (63) are not well defined, there is no clash between
the First Maxim of Quantity and our Principle of Informativeness. And the Maxim
of Manner is simply not a general explanation. We shall now sketch an account of
the implicata of (59) to (62).
The literal meaning of the sentence (61a) Mart and David bought a piano leaves
it open whether there was one piano-buying or two. The usual implicature restricts
the understanding to one. If the sentence is indeed a reduced form of Mart bought a
piano and David bought a piano, the literal meaning, under the assumption that this
reduction preserves meaning, is predictable. The conjunction also leaves it open
whether one or two pianos were bought. The same observation holds for Mart bought
a piano and so did David, which requires identity of sense of the deleted constituent.
I (Atlas 1977b: 329–30, 1989: 76–77) argued that it would follow from the last sen-
tence that David did what Mart did. The sentence requires sameness of action, but
that does not yet determine whether one or two piano-buyings are involved unless
the relevant criteria of identity of actions have been fixed. Given the meaning of the
sentence, at least it is clear that the criteria cannot require that the action-token (as
contrasted with action-type) be the same for David as for Mart. Thus the piano need
not be one and the same.
The “independent” implicatum entails that the event-tokens are different. Nor-
mally this would mean that more than one piano was involved, but it is imaginable
that in a short period of time Mart could buy and then sell a piano, which David then
bought, perhaps from Mart himself. It would not be semantically unacceptable, and
though unusual because incomplete, it certainly would not be false, to describe that
situation—one piano, two buyings—by (64a). We should represent such a situation,
following Davidson (1967), by (64b). The normal case would be (64c).
(64) a. Mart and David bought a piano.

b. ∃x∃e∃e' (Piano(x) & Buy(m,x,e) & Buy(d,x,e') & ¬(e=e'))
c. ∃x∃y∃e∃e' (Piano(x) & Piano(y) & ¬(x=y) & Buy(m,x,e) & Buy(d,y,e') & ¬(e = e'))
d. ∃x∃e(Piano(x) & Buy(m,x,e) & Buy(d,x,e))
By contrast the “cooperative” implicatum entails that the action-token (includ-

ing the piano) is the same for David as for Mart. Mart and David bought a piano
would then convey (64d). These implicata are not directly comparable: neither (64c)
nor (64d) entails the other. Nonetheless, we intuitively feel that the “cooperative”
implicatum (64d) is more specific or precise, perhaps because it is in Popper’s (1959)
sense a “riskier” proposition and more easily refuted. The fewer existential quantifi-
ers there are in an affirmative sentence, the more highly valued the sentence is. This
is a case in which the relevant notion of information is that determined by the class
of possible falsifiers of a proposition. Such a proposition is preferred as an interpre-
tation by my Principle of Informativeness unless it contradicts my background Con-
ventions of Noncontroversiality, as described in my Principle.
3.4 More indefinite noun phrase cases

As a further example of a class of cases in which the Principle of Informativeness
may clash with the First Maxim of Quantity, I consider sentences discussed by Grice
(1975a: 56), which I discussed first in Atlas and Levinson (1981: 49); for remarks
on this analysis, see Horn (1984b: 19) and Mey (1993: 78–79).20
(65) a. John is meeting a woman this evening.

b. The person to be met is someone other than John’s wife, mother, sister, or perhaps
even a close platonic friend.
(66) a. I broke a finger yesterday.

b. The finger is mine.
Grice argues plausibly that an inference from the First Maxim of Quantity will yield
(65b) from (65a). The failure to use a more informative expression than the indefi-
nite description a woman suggests that the speaker is in no position to provide a more
definite description of the kind normally relevant. The reverse implicatum in (66),
which Grice mentions in passing, presents an explanatory difficulty for the inference
from the First Maxim of Quantity. According to the First Maxim of Quantity, if the
speaker meant his own finger, he should have said so. Because he did not, he is as-
sumed not to be in the position to make that claim, that is, the finger was not his. But
the negation of this proposition is actually implicated.
Once again, the explanation of the inference lies in what speakers take as ste-
reotypical or conventional behavior. The use of the indefinite description a finger
leaves it indeterminate whose finger was broken, but the speaker’s breaking some-
one else’s finger would be regrettable if unintentional and contrary to our social norms
if intentional. As noted in the Second Maxim of Relativity, we are loathe to interpret
the utterance so as to impute an abnormal or unnatural act unless there are specific
indications to that effect. A similar explanation accounts for the implicatum in (67):
(67) a. I lost a book yesterday.

b. The book is mine.
20Jacob Mey (1993: 78) says of this analysis, in the version reported by Horn (1984b: 15), that “it de-
serves close attention for its painstaking analysis and elegant formulations of some original thoughts
on the subject of maxims.”
By contrast, First Maxim of Quantity implicatures seem in force in the follow-

ing cases (68) to (70):21
(68) a. I slept on a boat last night.

b. The boat is not mine.
(69) a. I slept in a car last night.

b. The car is not mine.
(70) a. I found a ring yesterday.

b. The ring is not mine.
The First Maxim of Quantity is part of an account of the speaker’s role in coopera-
tive communicative behavior. Informativeness is part of an account of the addressee’s
role in efficient communicative behavior. If one can communicate some specified
proposition P by asserting a less semantically specified sentence A, then in general it
will be more efficient to assert A and let the hearer make his inference to the best
interpretation (Atlas 1975b). In light of the distinct theoretical roles of Grice’s max-
ims and of the Principle of Informativeness, it is not a surprise that conflict might be
possible. For the class of indefinite descriptions just discussed, the empirical gener-
alization on the data seems to be that where there is an implicature at all (not all in-
definite descriptions yield them) the First Maxim of Quantity takes precedence over
the Principle of Informativeness unless the result contradicts our background Con-
ventions of Noncontroversiality. That is, ceteris paribus, constraints on production
take precedence over constraints on comprehension.
(71) Q > I.
If inconsistency with the network of background beliefs or incoherence with the

background abilities or capacities occurs (Searle 1983: 141–59, 1992: 175–96, 1995:
127–47, 1998: 31, 107–9), consistency requires that the informativeness inferendum
or implicatum be adopted instead. The case of indefinite noun phrases is the first
genuine case of clash between the First Maxim of Quantity, a speaker-centered prin-
ciple, and our Principle of Informativeness, an addressee-centered version of Grice’s
Second Maxim of Quantity.
If this is the correct generalization from the data, the question remains how to
explain it. The inconsistency in this first genuine clash, in indefinite noun phrase ex-
amples, between inferences from the First Maxim of Quantity and from my Principle
of Informativeness is resolved by a linguistic preference for the First Maxim of Quan-
tity implicatum unless background knowledge intervenes in favor of informativeness.
After all, where the First Maxim of Quantity implicature may be employed appro-
priately, it is reasonable to do so on the grounds that speakers are being cooperative in
Grice’s sense. Speaker and addressee have knowledge of each other’s reflexive com-
21Data in (67) to (70) are from Grice (1975a: 56) and Harnish (1976: 350). My intuitions differ from
Harnish’s on (68); he does not believe saying (68a) implicates (68b). Harnish provides no explanation
of his data. I provide a partial explanation in the following paragraphs.
municative intentions (Grice’s M-intentions). An addressee’s understanding the utter-

ance of a speaker relies on the addressee’s supposition that the speaker’s intentions
with which the utterance is performed are intended to be recognized by the addressee.
In particular, each expects the other to be truthful, relevant, perspicuous, and informa-
tive in what he or she says, and each expects the other to expect him or her to be so.
Unless knowledge by the addressee intervenes, and the more so if the knowledge can
be presumed to be mutual, as is knowledge of stereotypes, frames, scripts, and so on,
the normality of the speaker’s use of the language—its conformity with Grice’s im-
peratives for rational communication—is not questioned; thus an addressee’s inferences
from the First Maxim of Quantity are not standardly defeated.
Although addressees must share responsibility for successful uptake with the
speakers, speakers have the primary responsibility. When effective communication of
information is at stake, speakers’ utterances should conform to the First Maxim of
Quantity. The addressees, competent in the language, understand the limitations of the
resources of a speaker and of the grammar of their language in the speaker’s produc-
tion of sentencess. When appropriate, an addressee will pay the cost of comprehension
and conform to the Maxims of Relativity, the Conventions of Noncontroversiality, and
the Principle of Informativeness. The addressee-centered Maxims of Relativity work
on what the speaker said. The speaker-centered First Maxim or Quantity contrasts what
the speaker actually said with what he might have said.
The speaker’s choice among contrasting linguistic items is causally antecedent
to the interpretation of what he said, but the speaker’s choice is made strategically
(Atlas 1975a), anticipating how the addressee will interpret what will be said. The
form has to convey the content. As Levinson remarks, “inferences based on highly
constrained set of lexemes (Q-inferences) block those based on . . . stereotypes about
the world and the inference of maximal cohesion (I-inferences)” (2000: 161). As in
morphology and syntax, “the more specific rules block the application of more gen-
eral rules.” Unless there are special cognitive features in the situation to disrupt these
normal communicative and linguistic expectations, the ships in the conversational
fleet sail the seas of language undisturbed and undistressed.
This ordering Q > I, like those in Gazdar (1979a), is intended to save the lin-
guistic phenomena. The rationale offered in the preceding paragraphs is a sketch, a
gesture in the direction of a deeper explanation, in the form of a “plausibility argu-
ment.” But like Newton, I say, hypotheses non fingo.22
22There has been some unnecessary confusion about the goal of a rationale for the ordering principles.
A Gricean account of conversational inference is one sort of model by “reasons” and “rational behav-
ior” for the manifest linguistic facts: that addressees do in fact understand speakers to have conveyed,
suggested, or meant something more than what they literally asserted. By contrast, the preceding sketch
of a rationale for the ordering principle is an attempt at making sense of the particular ordering hy-
pothesis that generalizes over linguistic data. To confuse the latter, modest explanatory project in lin-
guistic and logical theory with the former philosophical attempt to characterize conversational inferences
by a model of communicative rationality is to fall into an error that one might have thought would
have been averted by careful reading of Atlas and Levinson (1981); for a cautionary example of such
an error, see W. Davis (1998: 58).
4 Reformulating the neo-Gricean maxims and

extending the ordering principle
To resolve the apparent conflict between scalar and clausal First Maxim of Quantity
generalized conversational implicata and to constrain the types of implicata, Horn
(1972) and Gazdar (1979a) had suggested that Horn Scales, composed of stronger
expressions S and weaker expressions W, would require that occurrences of S and W
in sentence frames A( ) should satisfy the entailment relation:
(72) a. A(S) A(W)
They would belong to a class of expressions in which they would be “expression

alternatives” (Gazdar 1979a):
(72) b. { . . . , S, . . . , W, . . . }.
Gazdar (1979a) noted that these constraints (a) and (b) permit the existence of a
scale <regret, know>, yet asserting “John knows that Columbus discovered America”
does not implicate John does not regret that Columbus discovered America. So I
propose here, as I did originally in Atlas and Levinson (1981), further constraints on
the formation of Horn Scales:
(72) c. Aboutness:
Expressions must be from a semantic field and be “conceptually homogeneous.”
(E.g., *<because, and> fails to satisfy this constraint.)
d. Lexicalization and Negative Scales (Levinson 2000):
Expressions in a scale <S, W> must be such that S is lexicalized to a degree equal to
or greater than W. (E.g., *<iff, if> and *<–,¬> are not scales). We permit the gen-
eration of negative scales, which I posited in Atlas and Levinson (1981: 39, n.9) to
explain:
“Not all of the boys are at the party” » Some of the boys are at the party.
“It’s not the case that Rick is both a philosopher and a poet” » Rick is either
a philosopher or a poet.
We permit the generation of Negative Scales <–W, –S> from <S, W>, e.g. <none,
not all> from <all, some>.
In Atlas (1991a) I made use of Levinson scales (see Grice 1965: 459), which record
the obvious fact that one can get generalized scalar implicata from items that are not
ordered by entailment, as in <succeed, try>, where “John tried to swim the Channel”
» John did not succeed in swimming the Channel, but where the scale is not
ordered by entailment, as the consistency of John succeeded without even trying
demonstrates.23
23Grice (1961, 1965: 459) had already noted this fact in his earliest discussion of implicata. Levinson
(1988a), in his Nijmegen lectures, has expanded the discussion of implicata arising from the structure
of the lexicon, and Hirschberg (1985) has done the same.
Finally, as a consistency requirement, Atlas and Levinson (1981: 50) suggested

an ordering, reminiscent of Gazdar’s (1979a) ordering of clausal over scalar impli-
cata, so that ceteris paribus First Maxim of Quantity (Q) generalized conversational
implicata would take precedence over, and exclude, Principle of Informativeness (I)
inferenda that were inconsistent with them:
(73) Consistency 1. Q > I.
We have supplemented Grice’s speaker-centered Second Maxim of Quantity with

the speaker- and addressee-centered Maxims of Relativity and the addressee-centered
Principle of Informativeness. A descriptively adequate account of pragmatic infer-
ences requires a communicative equilibrium between the two: the Division of Prag-
matic Labor (see Horn 1984b).
Since these ideas were first formulated in Atlas and Levinson (1981), Horn
(1984b, 1993) and then Levinson (1987a,b, 2000) have tried, in different ways, to
reorganize, revise, and extend them. Levinson (1987a, 2000) has suggested the fol-
lowing versions:
(74) Maxims of relativity (after Levinson 1987a: 66)

Speaker’s Maxim: “Don’t bother to say what is noncontroversial.”
Recipient’s Corollary: “Hear what is said as consistent with what is
noncontroversial.”
To give content to the maxims, we need to characterize the “noncontroversial,” which

I did in Conventions of Noncontroversiality and the Principle of Informativeness
(Atlas and Levinson 1981: 40, and section 1 here), reformulated by Levinson (1987a:
66) as follows:
(75) Convention of noncontroversiality

a. It is noncontroversial that referents and situations have stereotypical
properties.
b. The existence or actuality of what a statement is about is noncontroversial
(Atlas and Levinson 1981: 42).
(76) Principle of informativeness

The “best” interpretation of an utterance among its competing interpretations is the
most informative one that is also consistent with what is noncontroversial in the
speech-context.
Levinson (1987a: 66) then indicates the effect of these principles, remarking that
the “effect of this apparatus is to induce a more specific interpretation: what is com-
municated is a sub-case of what is said. Very often, the ‘best interpretation’ of what
is said will be a more specific interpretation in line with stereotypical expectations.”24
24Atlas and Levinson (1981) had distinguished between “precise” interpretations and “specific” inter-
pretations. What Levinson means here are the interpretations that I dub “precific.”
Horn (1984b: 13) suggested a super-categorization of Grice’s maxims into just

two opposing maxims (excluding the admittedly special case of the Maxim of Qual-
ity): a Q principle (”Make your contribution sufficient,” or Say as much as your hearer
needs, given the R principle) and an R principle (”Make your contribution neces-
sary,” or Say no more than your hearer needs, given the Q principle). Horn’s Q prin-
ciple subsumes the First Maxim of Quantity and its explanation of scalar and clausal
implicata; the R principle was intended to subsume Grice’s Maxims of Relation and
of Manner and Atlas and Levinson’s (1981) Principle of Informativeness inferenda
(the addressee-centered version of Grice’s Second Maxim of Quantity).
After Atlas and Levinson’s (1981) discussion of the inconsistencies between Q
and I implicata, Horn (1984b: 22–23) suggests, for his parallel Q and R principles, a
division of pragmatic labor:
While the Q Principle and the R Principle are diametrically opposed forces in in-
ference strategies and language change, it is perhaps in the resolution of the con-
flict between them that they play their major role in both ‘langue’ and ‘parole’. The
most general pattern for this resolution, the synthesis of the two antitheses, is sum-
marized in (16) and derived more explicitly in (17a–f).
(16) The use of a marked (relatively complex and/or prolix) expression [e.g. Larry
caused the car to stop] when a corresponding unmarked (simpler, less
‘effortful’) alternative expression is available [e.g. Larry stopped the car]
tends to be interpreted as conveying a marked message (one which the un-
marked alternative would not or could not have conveyed).
In particular, the reasoning goes as follows:
(17a) The speaker used marked expression E' containing ‘extra’ material . . . when
a corresponding unmarked expression E, essentially coextensive with it,
was available.
(17b) Either (i) the ‘extra’ material was irrelevant and unnecessary, or (ii) it was
necessary (i.e. E could not have been appropriately used).
(17c) 17b(i) is in conflict with the R Principle and is thus (ceteris paribus) to be
rejected.
(17d) Therefore, 17b(ii) . . .
(17e) The unmarked alternative E tends to become associated (by use or—through
conventionalization—by meaning) with an unmarked situation S, represent-
ing a stereotype or salient member of the extension of E (and E') [e.g. the
situation of stopping the car by the driver’s applying the sole of his shoe-
clad foot to the brake pedal]. (R-based inference; cf. Atlas and Levinson
(1981), (8b), (8c)).25
25Horn (1984b: 18) writes:

(8b) The Principle of Informativeness: “Read as much into an utterance as is consistent with what
you know about the world.” (Levinson 1983: 146–47)
Atlas and Levinson (1981: 42) formulate their informativeness-based inference as an inference to the
best interpretation:
(17f) The marked alternative E' tends to become associated with the complement
of [the situation] S with respect to the original extension of E (and E'). (Q-
based inference; cf. (6.12) [sic]). (Horn 1984b: 22–23)26
“The result,” Horn (1984b: 23) continues, “is an equilibrium which I shall call the
division of pragmatic labor.”
Horn then proceeds to discuss special cases of the division of pragmatic labor,
including: (a) Chomsky’s (1982: 65) Avoid Pronoun Principle, Kiparsky’s (1982)
Avoid Synonymy Principle, and McCawley’s (1978) account of the distribution of
lexical and periphrastic causatives, which is of particular interest to Horn. He remarks:
Lexical causatives (e.g. (28a)) tend to be restricted in their distribution to the ste-
reotypic causative situation: direct, unmediated causation through physical action.
This restriction can be viewed as a straightforward R-based
(28a) Black Bart killed the sheriff.
(28b) Black Bart caused the sheriff to die.
conversational implicatum—an inference ‘to the best interpretation’, in the language
of Atlas and Levinson (1981). The use of the relatively marked, morphologically
more complex periphrastic causative (e.g. (28b)) will then Q-implicate that the
unmarked situation does not obtain. (Horn 1984b: 27)
Horn’s example is of particular theoretical interest, in that asserting his (28a) I-

implicates the stereotypical, or usual, direct action, while the synonymous (28b) M-
implicates a nonstereotypical manner rather than the stereotypical action of the I
implicatum. In this case, the M implicatum excludes the I implicatum. And so, it seems
from the evidence of lexical causatives, M takes precedence over I:
(77) Consistency 2. M > I.
Finally, in light of the hypothesis (Consistency 1) in Atlas and Levinson (1981: 50)
and the hypothesis (Consistency 2), one would naturally hypothesize:
(8c) If a predicate Q is semantically nonspecific with respect to predicates Pi, 1 ≤ i ≤ n, but for some
j, 1 ≤ j ≤ n, Pj is stereotypical of Qs, then in saying Qt a speaker will convey Pjt.
[That is:
“The secretary smiled” » The female secretary smiled.
“John had a drink” » John had an alcoholic drink.]
26Horn (1984b: 21), section 7, writes:
(12) Q-based implicata:
(i) S entails W.
(ii) ‘W’ Q-implicates –S.
(iii) Normally, ‘not W’ = ‘less than W’, incompatible with S.
(iv) ‘not W’, asserted where S is given, reinterpreted as metalinguistic negation [see Horn (1985,
1989)].
(78) Consistency 3. Q > M > I.
The evidence in support of the hypothesis (Consistency 3) was provided not by

me but by Stephen Levinson (1987a,b, 1991), as we shall shortly see. It was Horn’s
(1984b) discussion of pronouns that led Levinson (1987a,b) to extend Atlas and
Levinson (1981), modify Horn (1984b), and, drawing inspiration from T. Reinhart
(1983), make a serious attempt at a pragmatic explanation of Chomsky’s (1982) bind-
ing conditions. Let us take an example and see the pragmatic reasoning in operation
(Levinson 1991: 112). Consider the interpretation of the pronouns in the indexed noun
phrases in the following sentences:
(79) a. John1 likes him2.

a'. John1 likes himself1.
b. John1 told her2 that he1 gave her2 a valentine.
b'. *John1 told her2 that himself1 gave her2 a valentine.
c. John1 told her2 that the man3 gave her2 a valentine.
Now, let us consider the following hypothetical explanation of the coreference

conditions in the preceding sentences. In (a), since a reflexive could have occurred
in the sentence in place of ‘him’, the non-use of a reflexive pronoun Q-implicates
that the reflexive interpretation is not intended, and the preferred interpretation is
that the NPs are not coreferential. In (b), since a reflexive ‘himself’ cannot occur
grammatically in the position that ‘he’ occupies (cf. (b')), the use of the pronoun ‘he’
cannot give rise to a Q implicature of non-coreference; so there is no higher-priority
implicature to exclude the I inference to coreference. In (c), since ‘the man’ is prolix
by contrast with a ‘he’ that could grammatically occur in the same position, the use
of ‘the man’ M-implicates non-coreference.
Levinson writes:
A pragmatic account of anaphora begins by noting that anaphoric expressions are

usually semantically general. Thus a locally anaphoric (coreferential) reading is
encouraged where a more semantically general term like ship follows a more se-
mantically specific term like ferry; and the reverse ordering discourages a locally
coreferential reading:
(1) a. The ferry hit a rock. The ship capsized.
b. The ship hit a rock. The ferry capsized.
The preference for a locally coreferential interpretation (as in (1)a.) can be attrib-
uted to the Gricean second maxim of Quantity, which I shall call the Informative-
ness Principle or I-principle for short (see Atlas and Levinson 1981; Levinson
1987a,b). This principle induces maximally informative and cohesive interpreta-
tions from minimal linguistic specification. It is balanced by a Manner maxim or M
principle, which induces from the use of a prolix or marked expression an inter-
pretation that is complementary to the one that would have been induced by the
I-principle from the use of a semantically general expression (this is what Horn 1984b
calls the ‘division of pragmatic labour’). Hence the association of anaphoric poten-
tial with pronouns and NP-gaps (both semantically general), and of independent
reference with full lexical NPs:
(2) The general anaphora pattern as M vs. I:

lexical NP > pronoun > NP-gap (Levinson 1991: 110)
In this ranking, Levinson argues, the higher-ranked, longer, more semantically

specific NP manner-implicates non-coreference, while the lower-ranked,
shorter, more semantically general, i.e. nonspecific, NP informativeness-conveys
coreference. Levinson continues:
A third inferential principle, Grice’s first maxim of Quantity (which we shall call
the Q-principle), provides a further essential ingredient. This induces a contrastive
interpretation between paired expressions of differential semantic strength or in-
formativeness. For example, the pair of quantifiers <all, some> form a Horn Scale
of this kind, such that the use of the semantically weaker some (as in some of the
boys came) Q-implicates the inapplicability of the stronger all to yield the interpre-
tation ‘not all of the boys came’. Only linguistic expressions that meet certain cri-
teria—notably salient opposition—form such Horn-scales capable of giving rise to
such Q-implicatures (see Atlas and Levinson 1981). One such salient contrast is
the opposition between ordinary pronouns and reflexives (see Levinson, 1987b, for
justification):
(3) The contrast <REFLEXIVE, PRONOUN>
forms a Horn Scale, so that the use of the Pronoun Q-implicates that the stronger
Reflexive could not truthfully have been used. A final ingredient in the pragmatic
account is a projection rule for generalized conversational implicatures, to the ef-
fect that, when inconsistent implicatures arise, priority is given according to the
following hierarchy:
Q > M > I.
Thus any tendency to read a pronoun as locally coreferential by the I-Principle will
be over-ruled by a Q-implicature to disjoint reference just where a stronger reflex-
ive could have been used to express coreference. (Levinson 1991: 111)27
Levinson formulates the following Conversational Norms for speakers and addressees:
(a) Where the syntax permits, a speaker intending coreference should use an
Anaphor [reflexive or reciprocal] by the Q-principle; the use of the weaker
Pronoun will Q-implicate disjoint reference by the Horn Scale <REFLEXIVE,
PRONOUN>.
27Levinson then importantly remarks:

A point that needs to be stressed (in the light of current fashions in pragmatics) is that it is
a special kind of implicature [implicata and inferenda] that is involved in these generaliza-
tions, namely GENERALIZED conversational implicature [or inference] (or GCIs or short). GCIs
are default interpretations, or preferred readings, in short, properties of utterance-types not
utterance-tokens.
They are generated by certain stable interpretive heuristics, not by nonce-inference of
the sort promoted in Relevance Theory or many kinds of computational pragmatics (see
Levinson 1989). Thus GCIs are stable, but defeasible, tendencies of interpretation; just the
sorts of things easily mistaken for grammatical stipulations. (Levinson 1991: 111–12)
(b) Otherwise, where a semantically general expression (e.g. a Pronoun) is used it

will I-implicate local coreference UNLESS: (i) an Anaphor (e.g. reflexive) could
have been used; (ii) a prolix form has been used which will M-implicate the
complement of the I-interpretation [of] a reduced form.
(Levinson 1991: 112)
Now I can summarize Levinson’s partial pragmatic explanation of Chomsky’s

Binding Conditions.
1. Binding Condition A: “Anaphors (reflexives and reciprocals) are

‘bound’—i.e., coindexed with a c-commanding NP, in their minimal
governing category.” This has no pragmatic account.
2. Binding Condition B: “Pronouns must be free—i.e., not coindexed
with a c-commanding NP, in their minimal governing category.”
Levinson’s (1991: 7) pragmatic explanation is that, given a Horn
scale like <himself, him>, a use of the weaker, less-specific him will
Q-implicate disjoint reference wherever the reflexive could have
been grammatically used; if one had meant the stronger, more
specific coreferential interpretation, one should have used the form
that encodes coreference.
3. Binding Condition C: “Lexical NPs must be free—i.e., not
coindexed with a c-commanding NP throughout the sentence.”
Levinson’s pragmatic explanation involves two points: (a) As in
Binding Condition B, there will be a Horn-like scale, e.g. <himself,
the man>, such that wherever a speaker could have used a reflexive
grammatically, he should have used it if coreference were intended.
(b) Unlike Binding Condition B, there will be an opposition
between the potential use of the pronoun and a use of a lexical NP,
an example of Horn’s (1984b) Division of Labor. For example, he,
a minimal form, will encourage an I-inference to local coreference;
the man will M-implicate disjoint reference, since the speaker
would seem to be avoiding the I-implicata induced by a use of the
pronoun.
Nothing here is said about the empty categories NP-trace and WH-trace, and there
are other data that do not have fully plausible, pragmatic explanations.
The interest, I take it, of Horn’s (1984b) and Levinson’s (1987a,b, 1991) efforts
is their success in giving genuinely plausible, pragmatic accounts of syntactic and
semantic phenomena by extending and revising the notions developed in Atlas and
Levinson (1981).28
I conclude this section with Levinson’s summary account of the two Maxims of
Quantity:
28For an application of Atlas and Levinson’s (1981) theory and its later developments to genuine prob-
lems of syntax, see Levinson (1987b, 1991, 2000) and Huang (1994), the latter providing the begin-
nings of an adequate account of anaphora in Chinese.
(80) Q Principle.
Speaker’s Maxim: Do not provide a statement that is informationally weaker than
your knowledge of the world allows, unless providing the strong statement would
contravene the I-principle.
Recipient’s Corollary: Take it that the speaker has made the strongest statement
consistent with what he knows, and therefore that: (a) if the speaker asserted A(W),
and <S,W> form a Horn scale, then one can infer not-A(S) [e.g. “some” » not
all]; (b) if the speaker asserted A(W), and A(W) fails to entail an embedded sen-
tence Q, which a stronger statement A(S) would entail, and {S,W} form a con-
trast set, then one can infer that the speaker does not know Q [e.g. “believes” »
does not know]. Levinson 1987b: 401–2, 2000: 76)
Levinson formulates the I Principle as follows:
(81) I Principle
Speaker’s Maxim (the Maxim of Minimization): Say as little as necessary—i.e.,
produce the minimal linguistic information sufficient to achieve your commu-
nication ends, unless providing the minimal statement would contravene the Q-
principle.
Recipient’s Corollary (the Enrichment Rule): Amplify the informational con-
tent of the speaker’s utterance, by finding the most SPECIFIC [or PRECISE]
interpretation, up to what you judge to be the speaker’s M-intended point (see
Grice 1989a: 105). [The ‘M’ here does not abbreviate ‘Manner’ but denotes
Grice’s reflexive communicative intentions.] Specifically: (a) Assume that ste-
reotypical relations obtain between referents or events, UNLESS (i) this is in-
consistent with what is taken for granted [what is “noncontroversial”]; (ii) the
speaker has broken the maxim of Minimization by choosing a prolix expres-
sion. (b) Assume the existence or actuality of what a sentence is “about” if that
is consistent with what is taken for granted. (c) Avoid interpretations that mul-
tiply entities referred to (assume referential parsimony); in particular, prefer
coreferential readings of reduced NPs (pronouns or zeros). (Levinson 1987b:
402, 2000: 114–15)
Finally, as we have seen earlier, we have the following:
(82) Projection rule for generalized

conversational implicata
When inconsistent implicata arise, priority is given by the hierarchy Q > M > I.
Quantity implicata take precedence over incompatible Manner implicata, which, in

turn, take precedence over Informativeness inferenda or implicata. Thus any tendency
to read a pronoun as locally coreferential by the I principle will be overruled by a Q
implicatum to distinct reference just where a stronger reflexive could have been used
to express coreference.
5 A concluding remark
In this chapter I have expounded the neo-Gricean theory for generalized conversa-
tional inferences that was developed by Horn (1972, 1976), Atlas and Levinson
(1981), Horn (1984b, 1993), Levinson (1987a,b, 1991, 2000), and Huang (1994). I
have tried to show that these principles not only improve on Grice’s formulations
(especially as they indicate the equilibrium required by both speaker-centered and
addressee-centered principles, and hence illustrate the proper relationship between
production and comprehension principles) but also have descriptive adequacy for a
wide range of linguistic phenomena. No competing linguistic theory or philosophi-
cal account has these predictive virtues with such modest theoretical commitments
and with such empirically verifiable constructs.
In the next chapter I employ my post-Gricean theory in the philosophical analy-
sis of Fregean and Strawsonian presupposition, whose properties have raised impor-
tant questions in twentieth-century philosophy of language and philosophical logic:
questions of assertion, entailment, negation, existence, and, of course, implicature.
We shall see that there has not always been a careful division in the analyses be-
tween what is properly semantic or logical and what is properly pragmatic, just as
Grice had suspected. In chapter 5 I examine the post-Gricean theory in the philo-
sophically interesting case of comparative adjectives and adverbials of degree. These
sentences provide a superb case study for the philosophical ideas of Donald Davidson
and Paul Grice, Peter Geach, and Sir Peter Strawson, and their analysis has conse-
quences for current philosophical debates over Meaning Holism. In chapter 6, I con-
sider the theoretically vexing case of numerical adjectives, and I draw some morals
for the role of post-Gricean pragmatics.
The Post-Gricean Theory

of Presupposition
T his chapter undertakes to illustrate how one type of post-Gricean pragmatic theory
of presupposition could be constructed and defended. In the course of the discussion,
most of the substantive problems about presupposition will be encountered. I press the
case for one type of post-Gricean pragmatic theory, reducing the concept of presuppo-
sition to its component concepts of entailment and generalized conversational inferenda
along the lines of Atlas (1975a,b; 1977a; 1979), Atlas and Levinson (1981), and Grice
(1981), and I pay particular attention to Grice’s (1981) attempt to make the same reduc-
tion. Grice uses the concept of ambiguity rather than the concept of semantical non-
specificity that I have for so long insisted is crucial in any successful theory. I also discuss
Grice’s and Stalnaker’s (1974) use of the notion of common ground, which I believe is
misapplied by them. I argue for a paradigm shift in treating presupposition, arguing
that the notion introduced by Strawson (1950), Ballmer (1975), Karttunen (1973), and
Atlas (1975a,b), later popularized by David Lewis (1979) under the name “accom-
modation,” offers the basic notion to reconstruct our theories of “presupposition.”
1 The linguistic phenomena of “presupposition”
The data of presupposition are represented by the relationships a competent speaker

of English takes to hold among his understandings of the affirmative, negated, and
modal statements in (1), (2), and (3), of the questions in (4), and the inferred state-
ments in (5). The inferred statements are preserved under negation, modals, and
question-formation.
118
THE POST - GRICEAN THEORY OF PRESUPPOSITION 119
(1) a. The king of France is bald. (Russell 1905, Strawson 1950)

b. After the separation of Schleswig-Holstein from Denmark, Prussia and Austria quar-
reled. (Frege 1892)
c. All of John’s children are asleep. (Strawson 1952)
d. John has stopped smoking.
e. Brian regrets that he drank too much.
f. Only Josh threw up at Carnivale in Venice.
g. It’s the queen of England who raises the best racehorses.
(2) a. The king of France is not bald.

b. After the separation of Schleswig-Holstein from Denmark, Prussia and Austria did
not quarrel.
c. Not all of John’s children are asleep.
d. John has not stopped smoking.
e. Brian does not regret that he drank too much.
f. Not only Josh threw up at Carnivale in Venice.
g. It’s not the queen of England who raises the best racehorses.
(3) a. The king of France might be bald.

b. After the separation of Schleswig-Holstein from Denmark, Prussia and Austria might
have quarreled.
c. All of John’s children might be asleep.
d. John might have stopped smoking.
e. Brian might regret that he drank too much.
f. Maybe only Josh threw up at Carnivale in Venice.
g. Maybe it’s the queen of England who raises the best racehorses.
(4) a. Is the king of France bald?

b. After the separation of Schleswig-Holstein from Denmark, did Prussia and Austria
quarrel?
c. Are all of John’s children asleep?
d. Has John stopped smoking?
e. Does Brian regret that he drank too much?
f. Did only Josh throw up at Carnivale in Venice?
g. Is it the queen of England who raises the best racehorses?
(5) a. There is a king of France.

b. Schleswig-Holstein did separate from Denmark.
c. John has children.
d. John smoked.
e. Brian drank too much.
f. Josh threw up at Carnivale in Venice.
g. Someone raises the best racehorses.
Each of the statements and questions (contrast to sentences) in (1) to (4) is, on
utterance, understood as “implying,” in some sense to be explained, the correspond-
ing sentences in (5). Competing theories have explained the “implication” (a) as a
semantic or logical one (using a nonclassical logical implication—e.g., Bergmann
1981; Martin 1979; Seuren 1985, 2000; van Fraassen 1971); (b) as a condition on
the use of statements to perform felicitous speech-acts (Gordon and Lakoff 1971);
(c) as a Gricean inference (Grice 1989b), from the literal understandings of the state-
ments in context to what speakers in, by, or when making the statements communi-
cate (“inference,” by contrast with an entailment relation between sentences, is
understood as a change in cognitive attitude by a speaker and addressee), or (d) as a
feature of speakers’ and addressees’ background information already available, or
created in real time, when the statements are made (Stalnaker 1974; Thomason 1990).
All these approaches manifest varying degrees of “semanticity” or “pragmaticity.”
Recent efforts by linguists—for example, the Discourse Semantics of Seuren (1985)
and van der Sandt (1988) or the Discourse Representation Theory of Kamp (1981) and
Kamp and Reyle (1993) and a similar approach of Heim (1982), or Dynamic Seman-
tics, illustrated in Chierchia (1995) and Chierchia and McConnell-Ginet (2000)—re-
veal the hybrid, semantico-pragmatic nature of the theoretical frameworks adopted.
Rather than survey all of these approaches to presupposition, and there are excellent
surveys to be found elsewhere, as in Beaver (1997), Horn (1992a), and Soames (1989),
I describe here a post-Gricean, pragmatic theory, distinct from those just mentioned,
which I and others have developed from H. P. Grice’s (1989a) account of conversa-
tional inference (see Atlas 1975a,b, 1977a, 1979; Kempson 1975, 1988a; Wilson 1975;
Atlas and Levinson 1981; Horn 1984b, Kempson 1986; Bach 1994a; Levinson 2000).
2 Preserving presupposition under negation
One of Frege’s employments of the notion of presupposition occurs in a footnote to

a discussion of adverbial clauses in “On Sense and Reference” (1892/1970a). Frege
wrote:
The sense of the sentence ‘After Schleswig-Holstein was separated from Denmark,
Prussia and Austria quarreled’ can also be rendered in the form ‘After the separation
of Schleswig-Holstein from Denmark, Prussia and Austria quarreled’. In this version
it is surely sufficiently clear that the sense is not to be taken as having as a part the
thought that Schleswig-Holstein was once separated from Denmark, but that this is
the necessary presupposition in order for the expression ‘after the separation of
Schleswig-Holstein from Denmark’ to have a reference at all. (Frege 1970a: 71n.)
I shall call this notion of Frege’s “referential presupposition.”

Since Frege took places, instants of time, and time intervals to be logical ob-
jects, to be designated by singular terms, if there were no true thought that Schleswig-
Holstein was once separated from Denmark, there would have been no event of
Schleswig-Holstein’s separating from Denmark, and there could be no specification
of an event at which Prussia and Austria quarreled as after the event of the alleged
separation of Schleswig-Holstein from Denmark. The existence of the event—an
event in which and hence a time of occurrence at which Schleswig-Holstein sepa-
rated from Denmark—is required to give a reference to a singular term designating

a time interval that would include a time or temporal subinterval at which Prussia
and Austria quarreled. Thus, suppose the logical form of the main clause Prussia
and Austria quarreled is
(6) ∃tQ(p,a,t)
There is some time or time-interval at which Prussia and Austria quarreled.
If so, one way to understand the semantical effect of a subordinate clause is for it to
determine the relevant domain of temporal instants or intervals that fixes the truth
conditions of the main clause. If the domain is restricted to times later than that of
the separation of Schleswig-Holstein from Denmark—namely, later than time to—
then the logical form of Prussia and Austria quarreled after the separation of
Schleswig-Holstein from Denmark would be (7a):
(7) a. ∃tQ(p,a,t)
t ∈T
where T = {t: t > to}. The truth of the clause Schleswig-Holstein separated from
Denmark at to is then a semantical determinant of the truth conditions of the original
sentence, since it specifies the domain of quantification T (Thomason 1973). In such
circumstances it is understandable that Frege should write of someone who believes
that it is false that Schleswig-Holstein was separated from Denmark, so that the do-
main of quantification of (7a) is ill defined:
He will take our sentence . . . to be neither true nor false but will deny it to have any
reference [on Frege’s view, a truth-value], on the ground of absence of reference for
its subordinate clause. (Frege 1970a: 71)
It is clear that on Frege’s semantical account of presupposition, on which the

falsity of a presupposition entails the lack of truth-value of the sentence with that
presupposition, the negative sentence Prussia and Austria did not quarrel after the
separation of Schleswig-Holstein from Denmark would have either the logical form
(7b) or the logical form (7c):
(7) b. ¬∃tQ(p,a,t)
t∈T
At no time after the separation of Schleswig-Holstein from Denmark did Prussia and
Austria quarrel.
c. ∃t ¬Q(p,a,t)
t∈T
At some time after the separation of Schleswig-Holstein from Denmark, Prussia and
Austria did not quarrel.
On either the wide-scope (7b) or the narrow-scope (7c) negation understanding,

the negative English sentence preserves the presupposition that there was a time at
which Schleswig-Holstein was separated from Denmark, since the specification of

the domain of quantification is antecedent to the assignment of truth conditions to
the negative sentence. Thus the notion of a semantical presupposition as a semantical
determinant of the truth-value of the sentence possessing the presupposition offers
one explanation of the preservation of the presupposition under ordinary, main-verb
negation (see the data in (1), (2), and (5)).
The invariance of presupposition under negation is also noted by Frege in an
example sentence that contains a proper name and a one-place predicate, but his
discussion of this example has features notably distinct from the example just dis-
cussed. It raises a number of questions about negation that recur in the discussion
of presupposition.
3 Alleged ambiguity of negation and

contrasts among presupposition,
assertion, and direct entailment
For Frege, a logically perfect language would be one in which each well-formed sin-
gular term designates an object. In ordinary, logically imperfect languages, singular
terms do not satisfy this requirement: for example, Vulcan and the cold-fusion reac-
tion are not guaranteed a reference merely by virtue of being singular terms in the
language. Frege claims:
If anything is asserted [my emphasis] there is always an obvious presupposition

that the simple or compound proper names used have reference. If one therefore
asserts [my emphasis] ‘Kepler died in misery’, there is a presupposition that the
name ‘Kepler’ designates something; but it does not follow that the sense of the
sentence ‘Kepler died in misery’ contains the thought that the name ‘Kepler’ desig-
nates something. If this were the case the negation would have to run not:
Kepler did not die in misery
but:
Kepler did not die in misery, or the name ‘Kepler’ has no reference.
That the name ‘Kepler’ designates something is just as much a presupposition for
the assertion [my emphasis]:
Kepler died in misery
as for the contrary assertion. (Frege 1970a: 69)
In this passage Frege claims that the presupposition of an assertion and of its main-
verb negation are the same, and he offers an argument to support it.
It is evident from his argument that the notion of P contains a thought Q was
assumed by Frege to be representable by P has the form Q & . . . or by P di-
rectly entails Q. (The quotation marks around ‘φ’ in ‘φ’ are Quine’s 1981 quasi-
quotation marks. The notion of “direct entailment” is, roughly, the entailment of a
subformula; see Atlas 1991a). In the case of the negative assertion Not P, it was
obvious to Frege that the Fregean senses (the truth conditions) of Not P and Not
P ∨ Not Q were not the same, as long as Not Q did not entail Not P. That con-
dition would be guaranteed if ‘‘Kepler’ has no reference’ did not entail ‘Kepler did
not die in misery’. But what ensures that this non-entailment obtains?
If the negative sentence ‘Kepler did not die in misery’ is interpreted as an exclu-
sion negation, paraphrased in English by ‘It is not true that Kepler died in misery’, a
vacuous singular term in the complement clause might be thought to yield for the
statement the value “true” (rather than as Frege might have thought, because ‘Kepler’
would lack a reference, no truth-value at all). Since the exclusion negation ¬φ of
a statement φ is true in a valuation if and only if φ is not true, even if φ is not
true because neither true nor false, ¬φ will be true. But then there is an entailment
of ‘Kepler did not die in misery’ by ‘‘Kepler’ has no reference’, and Frege’s argu-
ment, which supports the claim that statements and their main-verb negations that
contain singular terms share the presupposition that the terms are referentially non-
vacuous, fails!
Thus Frege’s argument requires that the main-verb negation not be an exclusion
negation, but a choice negation. The choice negation –φ of a statement φ is true
(false) in a valuation if and only if φ is false (true). If the choice negation is para-
phrased in English by ‘Kepler didn’t die in misery’, but ‘Kepler’ has no reference,
one’s intuition is that the choice negation is not true. So Frege’s argument survives.
But it survives on the assumption that the negative, ordinary language sentence
expresses a choice negation, typically a narrow-scope negation, not an exclusion ne-
gation, typically a wide-scope negation. The choice negation permits a failure of truth-
value for a sentence with a false (not-true) presupposition, but an exclusion negation
will be true even though a presupposition is not true, as we have seen. For these rea-
sons, van Fraassen (1971) formalizes the semantical notion of presupposition using
choice negation, but it commits a theorist of semantical presupposition to a false claim
about English: the ambiguity of simple ‘not’ sentences.
Frege’s argument also illustrates another aspect of presupposition. Since Not
P is not equivalent to Not P ∨ Not Q, P is not equivalent to P & Q. So Q
is not contained in P. The thought that ‘Kepler’ has a reference is not contained in
the affirmative assertion. If the thought is not contained in the assertion, not asserted
as part of it or directly entailed by it, it must be presupposed.
Here we have the contrast of presupposition with both assertion and direct entail-
ment. If one asserts Kepler died in misery or asserts Kepler did not die in misery, one
does not therein assert ‘Kepler’ has a reference. Similarly, if one asserts these state-
ments, the proposition ‘Kepler’ has a reference is not a subformula (or clausal con-
stituent) of the asserted content. If ‘not’ is understood as a choice negation, ‘‘Kepler’
has a reference’ will be semantically entailed by the negative sentence ‘Kepler did not
die in misery’ but not directly entailed by it, just as ‘‘Kepler’ has a reference’ will be
semantically entailed by the affirmative sentence but not directly entailed by it.
The difference between the affirmative statement semantically entailing that
‘Kepler’ has a reference and the negative not semantically entailing ‘‘Kepler’ has a
reference’ depends on construing the negative statement as expressing an exclusion
negation. Note that although the exclusion negation does not entail ‘‘Kepler’ has a
reference’, even the exclusion negation could be understood to have the referential
presupposition that ‘Kepler’ has a reference. It is just that if the name does have a
reference, the exclusion negation will be equivalent to the choice negation. The point
is that the exclusion negation can be true whether or not the name has a reference. Its
truth-value is unaffected by the obtaining of the referential presupposition. Hence it
is not the case that in a use of the negative sentence understood as an exclusion ne-
gation a speaker cannot presuppose that the name has reference. But if the presuppo-
sition fails, the exclusion negation will be true, while the choice negation will be
neither true nor false.1
In reconstructing Frege’s argument I, speaking for Frege, have implicated that
not only is the English sentence ‘It’s not true that Kepler died in misery’ capable of
expressing the exclusion negation of ‘Kepler died in misery’ but that it is only ca-
pable of expressing the exclusion negation—that the sentence just means the exclu-
sion negation. This is a traditional linguistic assumption in twentieth-century logic
and philosophy—for example in Whitehead and Russell (1970: 6), Frege (1970b:
123), and Strawson (1952: 78).2
Frege’s argument for the preservation of referential presupposition under main-
verb negation will not succeed without the assumption that main-verb negation ‘not’
(or ‘nicht’ in his case) is semantically a choice negation, which allows for sentences
that are neither true nor false. For further, independent reasons, Frege also accepts a
negation that is a wide-scope, exclusion negation. Both Frege and Russell (1905, 1919)
assume that ‘not’ (or ‘nicht’) sentences are ambiguous.3
4 Frege, the first pragmatic presuppositionalist
The roles of context and background knowledge were seen very early to be impor-
tant for one conception of presupposition. Frege himself, in addressing the anti-realist
prejudices of his late-nineteenth-century philosophical audience of German ideal-
ists, made use of the notion. He wrote:
Idealists or sceptics will perhaps long since have objected: “You talk, without fur-
ther ado, of the Moon as an object: but how do you know that the name ‘the Moon’
1So exclusion negation is not a presupposition “plug” in Karttunen’s (1973) sense.

2Theassumption is, I have argued, false (see Atlas 1974, 1975a, 1977b, 1978b, 1989; Bach 1987;
Kempson 1988a; Horn 1989).
3This assumption too, I have argued, is false. The free morpheme ‘not’ does not create ambiguities in
English. It is semantically nonspecific with respect to scope or with respect to exclusion and choice
negation just as ‘neighbor’ is semantically unspecified for gender, unlike ‘stallion’ or ‘poetess’ (see
Zwicky and Sadock 1975; Atlas 1974, 1975a, 1977b, 1978b, 1979, 1989; Horn 1989).
Happily the semantical nonspecificity (generality) of ‘not’ will not undermine Frege’s argument,
since he couched the argument in terms of assertions rather than sentences. A semantically negative
sentence can be used to make a specific, choice-negation assertion. Just as long as one treats asser-
tions—tokens of utterance-types in which a possible specification of a semantically nonspecific sen-
tence is achieved, for example, by the use of collateral information available to speakers and addressees
(mutual “knowledge”) in a context of utterance—as the carriers of presuppositions, Frege’s argument
for the preservation of referential presuppositions under main-verb negation in sentences of the form
Proper Name + Verb Phrase will be sustained.
has any reference? I reply that when we say ‘the Moon’, we do not intend to speak
of our idea of the Moon, nor are we satisfied with the sense alone, but we presup-
pose a reference. . . . Now we can of course be mistaken in the presupposition, and
such mistakes have indeed occurred . . . ; in order to justify mention of the refer-
ence of a sign it is enough, at first, to point out our intention in speaking or think-
ing.” (Frege 1970a: 61)
This passage is actually the first in “On Sense and Reference” in which Frege
uses the notion of presupposition. He was making a commonsense, linguistic and
philosophically realist objection to the views of philosophical skeptics and idealists.
Ordinary speakers take for granted the existence of the Moon, the satellite of the
Earth—not an idea in the mind or a concept—and intend to be referring to it, the
object, when they utter the singular term ‘the Moon’.
In summary, then, two of Frege’s three notions of presupposition are as follows
(Atlas 1975a):
(8) a. Semantical presupposition

The thought P semantically presupposes the thought Q if and only if the truth of Q
is a necessary condition on P’s having a truth-value. (The paradigm case for Frege
is where the presupposition is a semantical determinant of the interpretation of P.
For example, the truth of Schleswig-Holstein separated from Denmark is a semantical
determinant of the truth-valuedness of After the separation of Schleswig-Holstein
from Denmark, Prussia and Austria quarreled.)
b. Pragmatic presupposition
A speaker S pragmatically presupposes a thought P in speech context C if and only
if in C the speaker S assumes, believes, or takes for granted that P is true.4
5 Pragmatic presupposition continued
In “Pragmatics” Stalnaker wrote:
To presuppose a proposition in the pragmatic sense is to take its truth for granted,
and to presume that others involved in the context do the same. This does not imply
that the person need have any particular mental attitude toward the proposition, or
that he need assume anything about the mental attitudes of others in the context.
Presuppositions are probably best viewed as complex dispositions which are mani-
4Frege’s third notion of presupposition in “On Sense and Reference” is what I call “assertoric presup-
position”; this is roughly that the speaker literally means what he says and that it has a truth-value. I
glossed this in Atlas (1975a) by:
(iii) Assertoric presupposition
An assertion by a speaker S of a sentence A presupposes (a) that the speaker S intends to mean
what his words literally mean, and (b) that the semantical presuppositions of the thought expressed
by A are true.
fested in linguistic behavior. One has presuppositions in virtue of the statements he

makes, the questions he asks, the commands he issues. Presuppositions are propo-
sitions implicitly supposed before the relevant linguistic business is transacted.
(Stalnaker 1972: 387–88)
Karttunen (1973, 1974) and others took Stalnaker’s notion to be a sincerity condi-
tion on the utterance by a speaker of a sentence in a context. Stalnaker’s notion, in
contrast with Frege’s notion of pragmatic presupposition, requires that the supposi-
tions of the speaker be assumed by him to be those of his audience as well. Stalnaker’s
presuppositions are what the speaker takes to be common background for the par-
ticipants in the context. Grice (1967, 1981), Schiffer (1972), and Lewis (1969) had
employed similar notions. Stalnaker uses a Gricean formulation:
A proposition P is a pragmatic presupposition of a speaker in a given context just

in case the speaker assumes or believes that P, assumes or believes that his addressee
assumes or believes that P, and assumes or believes that his addressee recognizes
that he is making these assumptions, or has these beliefs. (Stalnaker1974: 200)
6 Accommodation in conditionals and

factive-verb statements
Karttunen (1973), seconded by Atlas (1975a, 1977a), had noted a weakness in

Stalnaker’s account. Karttunen pointed out that a counterfactual conditional like If
Bill had a dime, he would buy you a Coke is sincerely uttered in some contexts in
which the speaker does not assume that his audience assumes that Bill does not have
a dime. One point of uttering the sentence is to inform the audience that Bill does not
have a dime. On Stalnaker’s (1972) account, the proposition that Bill does not have
a dime is not a pragmatic presupposition in that context, and, on Stalnaker’s general
principle that “any semantic presupposition of a proposition expressed in a given
context will be a pragmatic presupposition of the people in that context,” not a se-
mantic presupposition of the counterfactual conditional. That was a conclusion that
Karttunen rejected, so he rejected Stalnaker’s general principle.
Likewise Atlas (1975a: 37) emphasized that “the assumption of common back-
ground knowledge is too strong to be applicable to speech-situations as universally
as Stalnaker and others would like.” Atlas also noted that “there are two strategies of
The Communication Game that are especially relevant to the problem of presup-
position, the strategy of Telling the Truth and the complementary strategy of Being
Informative” (1975a: 40). As noted in (1d,e) and (5d,e) factive-verb statements,
so-called because, for example, Geoffrey knows that P is analyzed as Geoffrey
knows the fact that P, are said to presuppose P. It is clear that there is an entail-
ment of the complement from the affirmative statement: Geoffrey knows that P P,
and of the object-language version of the referential presupposition: Geoffrey knows
that P Geoffrey exists. From an understanding of the negative statement, we may
also infer these propositions. For the neo-Gricean the question was how to explain
the inferences from the negative statement by appeal to Grice’s (1967) model of con-
versation as a rational, cooperative communication of information.
If we take the Kiparskys’ (1971) analysis of factive sentences seriously, we have,

in effect, two referential presuppositions: ‘Geoffrey’ has a reference; ‘the fact that
P’ has a reference; and in the object-language: ‘Geoffrey exists’, P. According to
Grice, when a speaker means more than he literally says and expects the hearer to
recognize that he does, the speaker’s expectations and the hearer’s interpretation are
governed by Grice’s (1967) Maxims of Conversation. One particularly important pair,
for our purposes, are the Maxims of Quantity: (a) Make your contribution as infor-
mative as is required by the current purposes of the exchange and (b) Do not make
your contribution more informative than is required. The Post-Gricean account of a
factive presupposition of a ‘know’ statement can then be sketched as follows.
Negative sentences of the form Geoffrey does not know that P are not scope-
ambiguous but rather semantically nonspecific between presuppositional and
nonpresuppositional understandings (Atlas 1975a: 42, n.23). On the semantical
nonspecificity view, the negative sentence is not ambiguous; it is univocal, and the
exclusion-negation and choice-negation “senses” are instead contextual specifica-
tions of the indeterminate literal meaning of the negative sentence. The literal mean-
ing is neither the exclusion-negation nor the choice-negation interpretation, but it is
something to which contextual information is added to produce in the hearer a choice-
negation understanding of the speaker’s utterance or an exclusion-negation under-
standing of the speaker’s utterance in the context. On Strawson’s (1950) view, the
choice negation will be true or false if P is true, and neither true nor false if P is not
true. On the post-Gricean view, the truth of P is inferred by the hearer in order to
construct a more informative understanding of the negative sentence than the
sentence-meaning that is semantically nonspecific between exclusion and choice ne-
gation. The syntax and meaning (the syntactical combination of its meaningful parts)
of the sentence constrain, but do not alone specify, what a hearer understands a speaker
to mean literally by an utterance of the sentence.
The specification of ‘not’ as a choice or an exclusion negation is also made by
the hearer in interpreting the speaker’s utterance. The semantical nonspecificity of
‘not’ in the sentence leaves it open to the hearer to make an “inference to the best
interpretation” of the utterance (Atlas and Levinson 1981; see also chapter 3 in this
volume).
The hearer’s inference that ‘Geoffrey’ has a reference is an interpretative one,
in order to explain most plausibly the speaker’s asserting Geoffrey does not0 know1
that P2, instead of the differently stressed utterance Geoffrey1 does not know that P0—
Geoffrey1 doesn’t exist, and may be a real-time accommodation, taking the speaker
at his word: thus, ‘Geoffrey’, as referring to an actual individual.
I observed, like Karttunen in the case of conditionals, that speakers can make
use of presuppositional sentences to be informative. The analysis was as follows (Atlas
1975a: 42–43): If a speaker intends to be informative, in this case about Geoffrey’s
ignorance, the speaker must intend, and the hearer recognize, another understanding
of the negative sentence (i.e., one other than the nonpresuppositional, exclusion-
negation understanding). This understanding is one in which the speaker presumes
that the proposition expressed by the complement of ‘know’ is true and hence a pos-
sible object of Geoffrey’s knowledge. This presumption by the speaker is necessary
whenever he intends his utterance to be informative [to the hearer about Geoffrey’s
ignorance]. Likewise, the hearer presumes that the speaker intends to be informa-
tive, and so assumes that the speaker presumes that the complement is true. If the
hearer does not know or believe, prior to the speaker’s utterance, that the comple-
ment is true, his or her presumption, ceteris paribus, that the speaker’s utterance is
meant to be informative provides him or her with good reason to accept that the
speaker believes that the complement as true and, hence, assuming that the speaker
has good reason for his belief, to accept that the complement is true. In this way, the
speaker, by reporting Geoffrey’s ignorance, can remedy the hearer’s ignorance.
Thus was recognized, for conditionals (Karttunen 1973), and for factive-verb
statements (Atlas 1975a: 42–43), the possibility of unpresupposed “presuppositions,”
which were given a theoretical explanation as part of the strategy of being informa-
tive in the communication game.
Later the notion was given the name “accommodation” by Lewis (1979) in his
“Scorekeeping in a Language Game,” and an earlier variant of the concept than
Lewis’s appeared in Ballmer (1975, 1978) and in the last few sentences of Strawson
(1950).
Having explicitly introduced accommodation into my discussion of presuppo-
sition in 1975a, I (1975a: 43) went on to claim that the best explanation of the “pre-
suppositions,” either presupposed or accommodated, of factive-verb statements could
be sketched as follows. The more informative use of the negative sentence to report
ignorance depends on a semantic property of ‘know’: S knows that P entails P.
For a person S to have his ignorance reported by S does not know that P, P must
be true. The informative use of the negative sentence is one in which the sentence is
understood in such a way that its truth conditions are “factive”; the complement of
‘know’ is taken to be true. This understanding of the negative sentence allows one to
infer the truth of P, just as the affirmative sentence S knows that P entails the truth
of P.
The description just given was, I believed, the best way to account for the lin-
guistic claim that S knows that P “presupposes” P. Stalnaker (1974: 206) had given
an account of factive-verb statements that missed the significance of accommoda-
tion, and one that merely demonstrated the alleged infelicity of asserting x knows
that P in a context in which speaker and hearer had not mutually acknowledged the
truth of P. Thomason (1984) gave a less detailed explanation of the factive-verb
presuppositions but had recognized the importance of accommodation (see Thomason
1990).
Later discussions differed from my 1975a,b account in their various empha-
ses on mechanisms of inference—practical reasoning about plans in Thomason
(1990) or abductive inference (inference to the best explanation) in Hobbs et al.
(1993), although practical reasoning in implicature had been discussed by Atlas
and Levinson (1973) and inference to the best explanation by Atlas and Levinson
(1981)—and in their appeal to a Stalnakerian (1974) assumption that strict felicity
conditions on the assertion of sentences containing singular terms or factive verbs
must be presumed to be satisfied in the speaker’s and hearer’s common ground.
For Stalnaker and Thomason, accommodation is the repair of the alleged infelic-
ity. I, by contrast, believe that there is no infelicity in asserting negative factive
statements in these contexts.
Since presuppositions were, pre-theoretically, preserved under main-verb nega-

tions, the explanatory task was to explain inference to the “presupposition” from both
the affirmative and the negative assertions. Part of the explanation was semantic, as
Frege noted: (a) The discoverer of the elliptical orbits of the planets died in misery
Some unique individual discovered the elliptical orbits of the planets; (b) Geoffrey
knew that Strayer was a first-class medievalist Strayer was a first-class medieval-
ist. The explanation for the presuppositional inference from the negative assertions
was both pragmatic (in Grice’s sense) and semantic. Some notion of an “informative
interpretation” of the negative utterance, and some “strategies of communication”
by virtue of which an informative interpretation could be arrived at, were obviously
necessary.5
What distinguishes presuppositional phenomena from pragmatic phenomena like
conversational implicature is that presupposition is a heterogeneous relation. A pre-
supposed proposition is a semantical entailment from the affirmative statement and
the content of the conversationally implicated, specific interpretation of the nega-
tive statement in a context, an interpretation that is beyond the literal meaning of the
semantically nonspecific negative sentence; the same content is “contained in” the
interpretation that is implicated by the speaker in the assertion of the negative sen-
tence as is entailed by the asserted content of the affirmative sentence.
Since Grice (1967) did not recognize the existence of semantically nonspecific
negative sentences, nor did Boër and Lycan (1976) in their attack on the semantic
definition of presupposition, and Stalnaker (1974) and Thomason (1990) did not
recognize the existence of felicitous accommodation, the Atlas (1975a, 1979) and
Atlas and Levinson (1981) account of presupposition was unavailable to them.
7 Grice on presupposition, implicature, and

the ambiguity of negation
In this section I compare the post-Gricean treatment (Atlas 1975a, 1978b, 1979; Atlas
and Levinson 1981) of presupposition as a heterogeneous relation combining entail-
ment and an extended concept of Grice’s generalized conversational implicature with
that given independently by Grice (1981) himself in his essay “Presupposition and
Conversational Implicature.” Atlas, Levinson, and Grice agree that There is a king
of France is entailed by The king of France is bald, although Grice takes the view,
unlike myself, that the negative sentence ‘The king of France is not bald’ is structur-
ally ambiguous between a wide-scope and a narrow-scope ‘not’. When a speaker
asserts the (allegedly) ambiguous negative sentence, then, according to Grice, “with-
out waiting for disambiguation, people understand an utterance of The king of France
is not bald as implying (in some fashion) the unique existence of a king of France.
5My own discussion (Atlas 1975a) was indebted to Grice’s (1967) model of conversation in which there
was a rational constraint on the communication of information, to Aristotle’s model of informative-

ness (Categories 2b7–2b22), and, under the influence of Schelling (1960), to my introduction of the
notion of a strategy, and not just maxims, of communication. It is the notion of a strategy that led me
to recognize and emphasize how unpresupposed “presuppositions” could be communicated—that is,
to recognizing the importance of what Lewis (1979) later labeled “accommodation.”
This is intelligible,” Grice continues, “if on one reading (the strong one), the unique
existence of a king of France is entailed, on the other (the weak one), though not
entailed, it is conversationally implicated. What needs to be shown, then, is a route
by which the weaker reading would come to implicate what it does not entail” (Grice
1981: 189).
What is notably common to Atlas (1975a, 1978b, 1979), to Atlas and Levinson
(1981), and to Grice (1981) is the claim that Strawson’s classic presupposition of the
existence of a reference of a singular term in a statement is a heterogeneous relation
between the statement and its presupposition, an entailment of the affirmative, and
somehow a conversational implicature of the asserter of the negative statement.
Nevertheless, the differences between the views are also striking.
The first striking difference between Grice (1981) and Atlas (1975a etc.) is
Grice’s claim that the negative sentence is ambiguous and Atlas’s contrasting claim
that the unambiguous negative sentence is semantically nonspecific between the
“weak” and “strong” understandings, which are not readings or senses. This subtle
semantic difference has often been misunderstood, and its consequences have been
underappreciated. Grice’s (1981) account allows us to see, in a more dramatic way
than usual, the consequences of this apparently small difference in theory.
What is incoherent in Grice’s (1981: 189) account—incoherent because it com-
bines assumptions of Grice’s ambiguity account with implications of Atlas’s (1975a)
semantical nonspecificity account—is the remark that hearers somehow “without
waiting for disambiguation . . . understand an utterance of The king of France is not
bald as implying (in some fashion) the unique existence of a king of France.” On
Grice’s own (1967, 1989b) original discussion of conversational implicature, sen-
tences were taken as disambiguated, so that statements had well-defined truth condi-
tions, before the reasoning resulting in a conversational implicatum was applied to
the statement in its context. But the intent of Grice’s remark as quoted here is that on
either sense of the negative sentence there will be an implication “in some fashion,”
an entailment from the strong sense and a conversational implicatum from the weak
sense. The implication of the existence of a unique king of France is overdetermined;
it is a double implication, the same from each sense but on semantic grounds for the
first sense and on pragmatic grounds for the second sense.
But if hearers really do not wait for disambiguation (and Grice’s justification
for that claim is an utter mystery), then there is no need to generate a conversational
implicatum from the (alleged) weak sense. Before disambiguation, all that can be
required for there to be any implication at all from the sentence is that there be an
implication from some sense of the negative sentence—not from all senses of the
negative sentence. Since the strong sense will entail the existence of a unique king
of France, the implication, of some fashion, can be explained without appeal to any
conversational implicature at all.
Grice (1981: 189) claims that the affirmative ‘The king of France is bald’ logi-
cally implies the existence of a unique French king, and its ambiguous negative
‘The king of France is not bald’ has a double implication: an entailment from
its strong negative sense and a conversational implicatum from its weak negative
sense.
By contrast, the post-Gricean claims that the affirmative statement The king of
France is bald logically implies the existence of a unique French king, and the se-
mantically nonspecific negative sentence ‘The king of France is not bald’ may, when
asserted, be understood by an addressee in a context to “implicate”—in the Atlas
(1979, 1989), post-Gricean technical sense (see Levinson 2000: 256–59)—distinct
propositions constructed (or inferred) from the nonpropositional, semantically non-
specific literal meaning of the negative sentence. The strong implicatum in a context
entails the existence of a unique French king; the weak implicatum in a context does
not. (I do not rule out a speaker misleading an addressee, and an addressee misun-
derstanding a speaker, by an addressee constructing both propositions in a context,
analogous to what on the ambiguity account would be the addressee recognizing both
[alleged] senses of the sentence. It is an empirical fact that for asserted sentences in
most contexts addressees do not consciously recognize more than one of the alterna-
tive senses or consciously recognize more than one of the inferred alternative under-
standings. I discuss the mechanism of the interpretative inference in the following
section 10.)
It is actually interesting how the philosopher who once proposed a Modified
Occam’s Razor, Senses are not to be multiplied beyond necessity (Grice: 1989c: 47),
lands in this commitment to ambiguity. Sluga had pointed out to him (Grice 1981:
188, n.2—an acknowledgment omitted from Grice 1989a: 271)—that one could treat
the king of France either as a quantifier or as a primitive singular term. If the former,
then the sentence would be open to scope-ambiguities, like typical quantified sen-
tences. But, Grice notably observes:
If there were a clear distinction in sense (in English) between, say, The king of France
is not bald and It is not the case that the king of France is bald (if the former de-
manded the strong reading and the latter the weak one), then it would be reasonable
to correlate The king of France is bald with the formal structure that treats the iota
operator [Russell’s definite description operator] like a quantifier. But this does not
seem to be the case; I see no such clear semantic distinction.” (Grice 1981: 188)
I want to emphasize Grice’s (1981: 188) last remark, since it was also made in-
dependently by me (Atlas 1974) and S.-Y. Kuroda (1977: 105), who also saw no
such clear semantic distinction. Unlike Grice, I noted that the observation suggested
that there was no scope-ambiguity for these negative English sentences—they were
semantically nonspecific with respect to scope, which further suggested that the for-
malization of these English sentences in a formal language whose structure imposed
a scope-ambiguity would miss a semantically important feature of negative English
sentences.
Rather than draw that conclusion, Grice (1981: 188) used his observation to mo-
tivate his choice of the formalization of the definite description as a logically primitive
singular term and concluded, “We are then committed to the structural ambiguity of
the sentence The king of France is not bald.” As a result, Grice’s explanatory task was
to show how, from an ambiguous negative sentence, Strawson’s presupposition that
there exists a unique French king (at the time of utterance) can be explained.
Grice had already observed that
in the original version of Strawson’s truth-gap theory, he did not recognize any
particular asymmetry as regards the presupposition that there is a king of France,
between the two sentences, ‘The king of France is bald’ and ‘The king of France is
not bald’; but it does seem to be plausible to suppose that there is such an asymme-
try. (Grice 1981: 187)
8 The evidence of cancelability
Grice then continues:
I would have thought that the implication that there is a king of France is clearly
part of the conventional force of The king of France is bald; but that this is not clearly
so in the case of The king of France is not bald. . . . An implication that there is a
king of France is often carried by saying [The king of France is not bald], but it is
tempting to suggest that this implication is not, inescapably, part of the conven-
tional force of the utterance of [‘The king of France is not bald’], but is rather a
matter of conversational implicature. (Grice 1981: 187)
Then, as had I (Atlas 1975a, 1979), Boër and Lycan (1976), and Atlas and Levinson
(1981), among others, Grice argues that the so-called presupposition of the negative
statement is (a) cancelable, (b) nondetachable, and (c) justifiable by argument from
Grice’s Maxims of Conversation as a conversational implicatum. (Roughly, this
means that (a) one can assert the statement and deny the “presupposition” without
inconsistency; that (b) other statements synonymous with this statement, but not dif-
fering wildly in manner, will carry the same “presupposition”; and (c) the inference
to the existence of a unique king of France is justifiable by principles governing ra-
tional information exchange in conversation.) Grice argued that the proposition that
there is a king of France is both explicitly cancelable (by outright denial) and con-
textually cancelable (by inconsistency with background information).
Grice offered the following support for those pragmatic features of the “presup-
posed” proposition. He wrote, “if I come on a group of people about whether the king
of France is bald, it is not linguistically improper for me to say that the king of France
is not bald, since there is no king of France” (Grice 1981: 187). As I have mentioned
earlier, speakers rarely notice the ambiguity of their utterances, and Grice is a case in
point. He did not notice the ambiguity of his wording and placement of the comma in
his indirect discourse sentence; he should have written in direct discourse: “for me to
say The king of France is not bald, since there is no king of France.” He also argued
that the proposition was contextually cancelable. He described an example as follows:
It is a matter of dispute whether the government has a very undercover person who
interrogates those whose loyalty is suspect and who, if he existed, could be legiti-
mately referred to as the loyalty examiner; and if, further, I am known to be very scep-
tical about the existence of such a person, I could perfectly well say to a plainly loyal
person, Well, the loyalty examiner will not be summoning you at any rate, without, I
would think, being taken to imply that such a person exists. (Grice 1981: 187)
But the more compelling example is the one he then goes on to give:
Further, if I am well known to disbelieve in the existence of such a person, though

others are inclined to believe in him, when I find a man who is apprised of my po-
sition, but who is worried in case he is summoned, I could try to reassure him by
saying, The loyalty examiner won’t summon you, don’t worry. Then it would be
clear that I said this because I was sure there is no such person. (Grice 1981: 187)
Notice, as I (Atlas 1974) observed, felicitousness does not require Grice to have said
It’s not the case that the loyalty examiner will summon you, don’t worry.
9 Pragmatic intrusion
The issue of nondetachability is more subtle. Here one looks for roughly synonymous
ways of making an assertion in which differences of manner are not so pronounced as
to swamp the similarities of meaning. Levinson writes, “Most analysts hold that pre-
supposition cannot be reduced to matters of implicature and that presuppositions are
attached to their lexical or syntactic triggers (and are thus not detachable in Grice’s
sense . . .)” (2000: 111), as if nondetachability were a problem for the post-Gricean
reduction. To the contrary, one expects generalized conversational implicata to be non-
detachable. Syntactic triggers, like the syntactic structure of clefts, are accounted for
on the post-Gricean view if, as Atlas and Levinson (1981) show, the logical form of
clefts, from which the implicata are generated, is distinct from that of the related simple
declaratives, whose syntax does not trigger a “presupposition.” As for lexical triggers,
‘knows that P’ and ‘believes justifiably and non-accidentally the fact that P’ will trig-
ger the same “presupposition” that P is true. Grice’s (1981) and my arguments are
designed to show that the post-Gricean account “saves the phenomena” of presuppo-
sition. Of course, one must understand the phenomena first.
For example, as suggested in Atlas (1974, 1975a, 1977b), The king of France is
not bald and It’s not the case that the king of France is bald can have the same presup-
position that there is a king of France, contrary to the tradition in philosophical logic
that claimed the latter statement to express only the wide-scope or exclusion negation
interpretation. Grice (1981: 188) agrees with this linguistic judgment. But Grice holds
that these sentences are semantically ambiguous and that the narrow-scope, choice
negation sense entails the existence of a king of France, while the wide-scope, exclu-
sion negation sense, when asserted, conversationally implicates the existence of a king
of France. Atlas (1975a,b, 1977b, 1978a,b, 1979) holds that these sentences are not
ambiguous but univocal, semantically nonspecific between the choice negation and
exclusion negation understandings. Thus the inferential mechanism of conversational
implicature will map semantically nonspecific, nonpropositional semantic representa-
tions into choice negation or into exclusion negation propositions, depending on the
context (Atlas 1978a, 1979). As discussed in Atlas and Levinson (1981), those singu-
lar terms in the statement that are Topic NPs are “noncontroversially,” by default, given
status in the interpretations as referring terms.
This is not a processing model, but it does raise the further question of the pro-
cessing of these interpretations. Do we resolve the nonspecificity of ‘not’ before
identifying singular terms as topic NPs? Or do we interpret the utterance The king of
France is not bald in the left-to-right surface string order, determining the topic NP
status of the singular term and then the interpretation of ‘not’? How does informa-
tion in different contexts affect the processing?
The post-Gricean account is non-Gricean, since the classical Gricean view took
the semantic representation of sentence-types to be literal meanings incomplete
only in contextual specification of the reference of singular terms, demonstratives,
indexicals, tense, and so on and took the semantic interpretations of sentence-
tokens (utterances) to be completed propositions, the content of “what is said.” Thus
I was committed to what later was labeled by Levinson (1988b, 2000) pragmatic
intrusion: the intrusion of pragmatically inferred content into the truth conditions of
what Grice (1989b) called “what is said.”6
10 Reduction of presuppositions
to conversational implicata
Grice (1981: 185) briefly characterizes the speaker’s implicatum as the content that
“would be what he might expect the hearer to suppose him to think in order to pre-
serve the idea that the [conversational] maxims are after all, not being violated.” The
neo-Gricean explanation of referential, factive, and cleft presuppositions (Atlas 1975a,
Atlas and Levinson 1981) depended on the hearer supposing the speaker not to be
violating Atlas and Levinson’s (1981: 40) neo-Gricean Maxims of Relativity, which
were refinements of Grice’s First and Second Maxims of Quantity: “Make your con-
tribution as informative as is required (for the purposes of the exchange)” and “Do
not make your contribution more informative than is required” (Grice 1989b: 26).
(9) Maxims of relativity
1. Do not say what you believe to be highly noncontroversial—that is, to be entailed by

the presumptions of the common ground.
2. Take what you hear to be lowly noncontroversial—that is, consistent with the pre-
sumptions of the common ground.
The first maxim is a speaker-orientated production maxim; the second maxim is a

hearer-orientated comprehension maxim. It is important that the production maxim
6The view of Atlas (1978a, 1979) bears a strong family resemblance, noted by Horn (1989: 433; 1992a),
to the views developed by the “London School,” in their notions of “explicature” in Relevance Theory,
and Kent Bach’s notion of “impliciture”; see Kempson (1986, 1988a), Sperber and Wilson (1995),
Carston (1988), Blakemore (1992), Récanati (1989, 1993), Bach (1994a).
is a prohibition, a “do not” maxim, and that the comprehension maxim is an obliga-
tion, a “must do” maxim. It is also important to note the difference between a sen-
tence being entailed by a set of sentences in the common ground and a sentence being
merely logically consistent with a set of sentences in the common ground. The con-
sistency requirement was designed to permit the kind of informative statement ac-
commodation for referential and factive presuppositions that I described in Atlas
(1975a, 1977a). If a singular term were introduced by its use in a statement that would
be more informative under an interpretation requiring the singular term to be a refer-
ring term in that statement, and its having a reference was consistent with the previ-
ously established common ground, nothing in my maxim would stand in the way of
such an informative interpretation—it would be “noncontroversial”—whether or not
the existence of the reference of the singular term had already been established as
part of the common ground. Atlas and Levinson’s (1981) Maxims of Relativity were
designed to accommodate accommodation.
It also should be noted that my Maxims of Relativity were couched in terms of
noncontroversiality and of common ground, constrained by a mini-theory of noncon-
troversiality. Among the axioms of that mini-theory were:
(10) Axioms of noncontroversialty

a. If A(t) is “about” t, or if t is a topic NP in the statement A(t), then if t is
a singular term, the proposition t exists is noncontroversial for speaker S in con-
text K with respect to A(t).
For example, if one asserts The king of France is not a good philosopher,
the proposition that the king of France exists is, in the context, noncontroversial.
b. The obtaining of stereotypical relations among individuals is noncontroversial for
speaker S and addressee H in context K with respect to A(t).
For example, if one asserts The king of France had a drink, that the king of
France had an alcoholic drink would be noncontroversial. (Atlas and Levinson
1981: 40)
What I showed in Atlas (1975a,b) and Atlas and Levinson (1981) and have reviewed
here is the explanation for the inference to There is a king of France from a speaker’s
assertion of The king of France is not bald (see (10a)). The construction of reason-
ing to a default, noncontroversial, informative interpretation of a semantically non-
specific negative sentence is required if Grice’s third criterion for the existence of
a conversational inference is to be met, and the reasoning I constructed depended
on an elaboration and a revision of Grice’s Maxims of Quantity (being as informa-
tive as is required). This in-my-technical-sense “implicated” interpretation of the
semantically nonspecific sentence-token makes use of what is noncontroversial with
respect to the utterance, as long as it is consistent with the common ground of the
context in which the sentence-token is uttered. The existential proposition that there
is a king of France is entailed by that interpretation, but the interpretation is origi-
nally constructed from the existential proposition’s being noncontroversial with
respect to the utterance in the context.
Grice himself adopts a Russellian analysis of The king of France is bald, a con-
junction of three independent clauses (cf. Strawson 1950, 1971b: 5):
(11) The king of France is bald.

(A) There is at least one king of France.
(B) There is not more than one king of France.
(C) There is nothing which is a king of France and is not bald. (Grice 1981: 189)
The account of presupposition that Grice gave of the presupposition of the negative
statement The king of France is not bald depends on a distinction between denied
and undenied conjuncts:
It would be reasonable to suppose that the speaker thinks, and expects his hearer to
think, that some subconjunction of A and B and C has what I might call common-
ground status and, therefore, is not something that is likely to be challenged. One
way in which this might happen would be if the speaker were to think or assume
that it is common knowledge, and that people would regard it as common knowl-
edge, that there is one and only one [king of France]. (Grice 1981: 190)
Thus the speaker who asserts The king of France is not bald would be under-
stood to deny only the third conjunct (C) {Nothing that is a king of France is not
bald, Whatever is a king of France is bald}, since the argument just quoted was sup-
posed to “show that, in some way, one particular conjunct is singled out” (Grice 1981:
190). Of course, if one takes it as common knowledge that there is a unique king of
France, and then denies conjunct (C), as Grice proposes, one gets the conjunction
(a) There is a unique king of France & there is at least one king of France that is not
bald, which is supposed to be an interpretation of (b) The king of France is not bald.
The (weak) denial, namely the interpretation of It’s not the case that the king of France
is bald as ¬((A & B) & C), conjoined with the common ground (A & B), is supposed
to give ((A & B) & ¬C) as the interpretation of The king of France is not bald. But
the question is, why should the common ground intervene in utterance-interpretation
in this way? Grice believes that its commongroundedness is a sufficient and obvious
explanation; I do not. A theory of how and why common ground enters into utter-
ance-interpretation is needed. Grice does not offer one; Atlas and Levinson (1981)
do, in their notions of noncontroversial propositions that are consistent with com-
mon ground propositions in a context of utterance and that enrich the sentence-
meanings of the utterances with respect to which they are noncontroversial.
Sentence (a) entails There is a unique king of France, and an utterance of (b),
on Grice’s (1981: 189) own showing, “without waiting for disambiguation,” implies
“(in some fashion) the unique existence of a king of France.” Grice (1981: 189) has
already remarked that “what needs to be shown is a route by which the weaker read-
ing could come to implicate what it does not entail.” But what Grice (1981: 190) has
just shown is how the (weak) denial of The king of France is bald, in conjunction
with the common ground, entails There is a unique king of France, which is what he
has explicitly claimed a speaker implicates by it, but that it does not entail alone.
Does this mean that Grice reduces implicature to a context-relative entailment,
an entailment from the assertion and the propositions of the common ground of the
context of utterance? If he were to do so, he would start to sound like a Sperber and
Wilson (1986b) Relevance Theorist. But a common-ground-dependent entailment
from the (weak) denial is merely a fact about context and the assertoric content of a
weak negation. It is not a theory of an inference to the best interpretation of the nega-
tive utterance in the fashion of Atlas and Levinson (1981). Unlike Atlas, Grice thinks
the negative sentence already has an interpretation; it is the weak negation. The ques-
tion for Grice is, why should that interpretation generate the “implication” that there
is a unique king of France?
Pace Grice (1981: 190), the answer cannot be that the existential proposition is
already assumed in the context. The existence of a context-relative entailment is no
explanation of “the route by which the weaker reading could come to implicate what
it does not entail,” since context-relative entailment does not possess the logical prop-
erties of implicature or presupposition. Even though the context provides premisses,
the relation is an entailment; it is monotonic, unlike implicature (which is defeasible),
and it is unlike presupposition (which is preserved under main-verb negation). Un-
like an entailment of a statement, an implicatum of a speaker can also be canceled
(i.e., negated without a resulting contradiction). Although the existential proposition
is entailed by the hearer’s choice-negation interpretation of the speaker’s utterance,
the proposition can be negated without an inconsistency with what the speaker liter-
ally uttered, since the literal meaning of the speaker’s semantically nonspecific, nega-
tive sentence-token does not entail the existential proposition.
What makes it plausible to Grice that a speaker could assert the weak denial ¬((A
& B) & C) and in the context mean [implicate] ((A & B) & ¬C) by his assertion
is that the truth of the latter is a way of guaranteeing the truth of the former—and so
Grice is fallaciously thinking that the latter is a way of expressing what a speaker
could mean by the former, not merely a way of entailing its truth. G. E. M. Anscombe
(1981b) long ago warned us against this fallacy of taking a truth-guaranteeing con-
dition—a fact that verifies a statement—to be a meaning (see Atlas 1989: 62–64),
what I shall call ‘The Anscombe Point’.
Further, Grice has given an analysis of the presupposition of The king of France is
not bald as a pragmatic presupposition (in Stalnaker’s 1974 sense) of speakers and
addressees even when they assert the so-called weak sense “It’s not the case that the
king of France is bald.” But as Stalnaker (1974) explicitly notes, speaker’s presuppo-
sition is not an account of presupposition as a generalized conversational implicature!
The sources of these difficulties in Grice’s (1981) analysis of The king of France
is not bald are clear. The first source is his self-consciously accepting the conjunc-
tion of three independent propositions as an analysis of The king of France is bald
(for reasons that I discuss next) and his concomitant commitment to the scope ambi-
guity of The king of France is not bald. The ambiguity assumption, combined with
his ignoring the Anscombe Point, leads Grice into an incoherent account in his at-
tempt to derive the referential “presupposition” from his (weak) external negation
reading of the negative sentence. It is here that the subtle difference between ambi-
guity and semantically nonspecific univocality has blatant and devastating conse-
quences for the success of Grice’s (1981) attempt to reduce referential presupposition
to implicature. And it is here that the Atlas (1975a, 1979) and Atlas and Levinson
(1981) analysis succeeds where Grice’s (1981) fails.
Most extraordinarily, Grice (1981: 189) adds to his 1967 Maxims of Manner the
maxim “Frame whatever you say in the form most suitable for any reply that would be
regarded as appropriate” and adopts as the canonical form of what is to be denied in

the negative sentence ‘The king of France is not bald’ the Russellian expansion (R)
There is at least one king of France & at most one king of France & whoever is a king
of France is bald of the abbreviated, affirmative form (D) The king of France is bald.
Then Grice (1981: 192) argues that because a speaker has uttered the abbreviated af-
firmative form (D) rather than (R), the hearer, noting that the speaker did not say (R)
when he could have done so, and assuming that the speaker is conforming to the new
Maxim of Manner—namely, that the speaker is to tailor his assertions to possible de-
nials by the addressee—intends his (the hearer’s) possible denials of (D) to deny the
third Russellian conjunct, which Grice assumes to be equivalent to denying baldness
rather than existence or uniqueness. Then, Grice (1981: 192) adds, a speaker who now
asserts the syntactical negative of (D) is speaking as one who himself would have in-
terpreted the denial of (D) as denying baldness of the king of France. To generalize,
perhaps unfairly, Grice’s analysis: a Speaker S typically implicates Ψ in asserting φ if
there exists a φ' such that (a) φ' paraphrases φ, (b) in asserting φ' one asserts or other-
wise explicitly raises the question of the truth of Ψ, (c) in asserting φ one does not as-
sert or otherwise explicitly raise the question of the truth of Ψ. Such an analysis is prima
facie an absurd account of the implicatures of φ.
One problem, among several, with Grice’s analysis is that the denial of the
third Russellian conjunct (that someone is a non-bald French king) is not equiva-
lent to denying baldness of the king of France. Once this is admitted, Grice’s analysis
loses all plausibility as an explanation of the inference from the exclusion-negation
interpretation of The king of France is not bald to its choice-negation interpretation.
Of course, in addition, it is antecedently questionable that Grice should assume that
the underlying structure for The king of France is bald is just the three-conjunct
Russellian analysis, which he does, circularly, by asking what structure would be
most suitable for distinct denials expressing scope-distinct choice negations. And,
as I have already mentioned, the lack of evidence of disambiguation has an even
simpler explanation than the one Grice considered: the negative, definite descrip-
tion sentence is not ambiguous. But no one, except Atlas (1974, 1975a,b, 1977b)
and Kempson (1975, 1988), had argued for the theoretical possibility that the sen-
tence was semantically nonspecific with respect to the scope of negation, although
Sir Peter Strawson (personal communication 1978) informed me that he had enter-
tained a similar hypothesis.
What Grice (1981: 187) has argued successfully is that the inference to the exis-
tence condition for the use of a definite description is not a Strawsonian presupposi-
tion; it is a cancelable and highly nondetachable inference, on the now familiar ground
that it is linguistically acceptable and logically satisfiable to assert Don’t worry, the
bogeyman won’t get you; there is no bogeyman, which should have been a linguis-
tically anomalous and, if the second conjunct is true, a truth-valueless conjunction
on a Strawsonian theory of the presuppositions of these main-verb negation sentences
(Atlas 1988, 1989). Let me emphasize this: Grice (1981: 187) shows that existence
of the denotation is not a semantical presupposition of the use of a definite descrip-
tion. It is an implicatum or inferendum.
Grice (1981: 188) has also partially anticipated the point made in Kuroda (1977)
and in Atlas (1974), in which I observed that the English sentences The king of France
is not bald and It is not the case that the king of France is bald have precisely the
same range of interpretations, both presuppositional and nonpresuppositional alike
(for discussion, see Boër and Lycan 1976; Horn 1978b, 1989). The point of my ob-
servation was to support my claim of the absence of scope-ambiguity for ‘not’ in the
semantic representation of the English sentences. Grice, instead, noted that it was
not plausible to think that The king of France is not bald expressed just the predicate
negation sense in which the sentence entailed the existence of a unique French king—
the reading that Strawson (1950) had given it—or to think that It is not the case that
the king of France is bald expressed just a sentence negation sense in which the
existence of a unique French king was not entailed. Grice inferred from these obser-
vations that the English sentence The king of France is not bald is structurally am-
biguous. He made two correct observations, so far as they went, and then drew just
the wrong conclusion.
The theoretical situation, as I described it in “Reservations about the Standard
Grice View,” was this:
Taking the negative sentence [The king of France is not bald] in isolation, compe-
tent speakers know that it has (at least) two uses or understandings. Independently
of context, the understandings are phenomenologically of equal status, neither judged
less a function of the meaning of the sentence than the other. But the account in the
[classical Gricean, but not Grice’s (1981)] theory is “unequal,” in that it is split be-
tween the semantics and the pragmatics, and the understandings are of different theo-
retical status [the exclusion negation understanding is semantic; the choice negation
understanding is pragmatic]. (Atlas 1979: 275–76)
In the neo-Gricean view the literal sense of the sentence ‘The king of France is not
bald’ is the wide-scope, exclusion negation. It is not held to be ambiguous, unlike
what Grice (1981) himself thought. The neo-Gricean pragmatic view, motivated in
part by theoretical considerations from the linguistic theory of Generative Seman-
tics, was that the correct syntactical analysis of the negative sentence gave it a clause-
external negation, and the semantic representation of a sentence with that syntactical
structural description was the wide-scope exclusion negation. I continued:
The classical pragmatic view, without the semantical generality of negation, per-
mits the negative sentence interpreted as a sentence negation to implicate the exis-
tential “presupposition” [as Grice 1981 suggests but fails to justify]. The negative
sentence interpreted as a predicate negation staightforwardly entails it. And of course
there are contexts in which no implicature of the sentential negation is intended.
Letting L– stand for the sentential negation, L+ for the predicate negation, the func-
tion PRAG for the Gricean inference, and K for kinds of context, we may abbrevi-
ate the claims by the formulas:
PRAG(K*,L–) = L+,
PRAG(K**,L–) = L–.
In the second case [in which the sentential negation meaning of the negative sen-
tence is passed through the pragmatic inference machine unchanged] the pragmat-
ics adds nothing to the semantical interpretation; in the first case [in which the
sentential negation meaning of the negative sentence is transformed into the predi-
cate negation implicatum] it obviously does. The standard view permits this kind
of asymmetry in the theory. The first case is paradigmatic; the second case is de-
generate. Why should there be this difference? (Atlas 1979: 275)
Grice (1981: 189) had remarked, recall, that “without waiting for disambigua-
tion, people understand an utterance of The king of France is not bald as implying
(in some fashion) the unique existence of a king of France.” For the radical pragma-
tist in linguistics, and for Grice (1981), too, the first case was paradigmatic. What
neither asked was my question, “Why should there be this difference?” The post-
Gricean answer I gave in Atlas (1979) was, “There should not.” I continued:
A Gricean view combined with the representation of the semantical generality of

negation, which was among the positions I advanced as viable options in the early
1970’s, remedies this theoretical “asymmetry” in the standard Gricean view. Let L#
stand for the nonspecific semantic representation of The A is not B. Then the
pragmatic theory does theoretical work in BOTH cases:
PRAG(K*,L#) = L+
PRAG(K**,L#) = L–.
Understandings that phenomenologically are of equal status are theoretically of equal
status—produced in the same way by the same mechanism. The problem of explain-
ing the degenerate case, that is, the case where PRAG(K**,L–) = L–, simply van-
ishes. (Atlas 1979: 278)7
It is important to understand that the semantic representation L# of ‘The king of

France is not bald’ does not express a proposition, is not a description of truth con-
ditions, or of a set of possible worlds, and so on. It is a semantically nonspecific,
prepropositional, nonpropositional representation of the literal meaning of the En-
glish sentence string.
Ruth Kempson in her “Grammar and Conversational Principles” had noted:
One of the few people in the mid 1970s to recognize the gap between linguistic
content of a sentence and the articulation of the truth conditions of its associated
propositions was Jay Atlas, who argued in a series of papers (Atlas 1975[a], 1977[b],
1979) that the linguistic concept of sentence negation was weaker than any concept
sufficient to characterize the truth-theoretic properties of propositions that nega-
tive sentences express. (1988a: 141 n.2)
Noam Chomsky has recently suggested that “we cannot assume that statements (let
alone sentences) have truth conditions. At most they can have something more com-
plex: ‘truth indications’ in some sense” (1996c: 52). And he added:
7[Note in the original text] “It may be helpful to linguists to consider a parallel with phonology. The
sentence The A is not B in one context may be understood as L– and in another as L+. These under-
standings are related to the . . . meaning L# of the sentence as ALLOPHONES are to the PHONEME to
which they belong.”
thought, and various forms of human interaction. (Chomsky 1996c: 53)
I had written, “The sentence The A is not B in one context may be understood
as L– [an external, exclusion negation] and in another as L+ [an internal, choice nega-
tion]. These understandings [not senses] are related to the literal meaning . . . of the
sentence as allophones are to the phoneme to which they belong” (Atlas 1979: 278
n.). It seems that I have been engaged in a study in Chomskyan Internalist Seman-
tics. I continued:
On my view understanding the speaker’s utterance is knowing a proposition that

the context [permits one to] construct from “the meaning” of a univocal, semanti-
cally nonspecific sentence. Not only is the nonspecific meaning of the sentence made
specific in the understanding of the utterance, but the propositional content of the
utterance that is constructed by inference is one that contributes to [but is not ex-
hausted by (see Chapter 2)] an explanation for the uttering of the sentence in the
context. (Atlas 1979: 278)
Again it is important to note that the propositional content of the utterance-

interpretation is constructed by pragmatic inference PRAG from a semantically non-
specific, nonpropositional representation of the literal meaning of the sentence. I
finished the discussion with a rhetorical flourish:
In seeing concretely for the first time how linguistics contributes to the solution of
philosophical problems, we discover that Fregean semantics is inadequate to our
explanatory tasks. The age of classical semantics is over, and we are finally expelled
from the Cantorian paradise. (Atlas 1979: 279)
In a similar vein Noam Chomsky recently wrote:
Turning to principle III of the [Fregean] model [i.e., “Language is a set of well-
formed expressions, and its semantics is based on a relation between parts of these
expressions and things in the world”], human languages differ radically from Fregean
symbolic systems in just about every crucial respect. (Chomsky 1996c: 48)
11 Common ground and context as the source

of presuppositions
The second source of Grice’s difficulties is his appeal to common ground in the simple
way he appeals to it and the way that Stalnaker (1974) appeals to it in “Pragmatic
Presuppositions.” An explanation using speaker’s presupposition is not equivalent
to an account using conversational implicatures.
Grice’s error was the tacit assumption that by putting There is a unique king of
France into the common ground, its common-groundedness would explain its pre-
suppositional—as contrasted with “assertoric” and with “entailed”—status. But
common-groundedness of a proposition can provide no such contrast with the
assertoric status or entailments of a proposition. It matters not whether There is a
unique king of France belongs to the speaker’s and addressee’s “common knowl-
edge” if what needs to be explained is why and how Grice’s weak reading of The
king of France is not bald, if asserted, yields as an implicatum There is a unique king
of France. Grice’s attempted explanation by appeal to common knowledge of ‘There
is a unique king of France’ still results on his account in an entailment of ‘There is a
unique king of France’ from the common ground and the weak reading of The king
of France is not bald, because it is entailed by the common ground alone. What is
missing is an account of why using the common-ground status of ‘There is a unique
king of France’, when combined with the weak negation interpretation of ‘The king
of France is not bald’, explains an inference to the presuppositional interpretation of
the utterance The king of France is not bald, an inference to the token of the type The
king of France having a reference, where the token’s having a reference is neither
entailed by the literal meaning of the utterance (on either my view or Grice’s view of
the literal meaning of the negative sentence) nor asserted in it.
If one needed more proof that “common knowledge” is not essential to explain-
ing why people understand an utterance of The king of France is not bald to “imply
(in some fashion)” the existence of a unique king of France, it is the existence of
accommodation. And, despite his own (misguided) attempt to explain presupposi-
tion as an implicatum somehow arising from common ground, Grice was aware of
accommodation:
It is quite natural to say to somebody, when we are discussing some concert, My

aunt’s cousin went to that concert, when one knows perfectly well that the person
one is talking to is very likely not even to know that one had an aunt, let alone know
that one’s aunt had a cousin. So the supposition must be not that it is common knowl-
edge but rather that it is noncontroversial, in the sense that it is something that you
would expect the hearer to take from you (if he does not already know). . . . That is,
to take my word for. (Grice 1981: 190)
That is why the account in Atlas and Levinson (1981: 40–41) appealed to noncon-
troversiality in stating the Maxims of Relativity, and why the referential “presuppo-
sitions” are expressed as Conventions of Extension, a subclass of Conventions of
Noncontroversiality:
(12) If a speaker’s statement A(t) in context K is “about” t, or if t is a topic NP in a

statement A(t) , then:
a. if t is a singular term, t exists is noncontroversial for the speaker S in context
K with respect to A(t) ;
b. if t denotes a state of affairs or a proposition, t is actual and t is true are non-
controversial for the speaker in context K with respect to A(t) .
It is also why among the Conventions of Noncontroversiality is the following Con-

vention of Intension:
(13) The obtaining of stereotypical relations among individuals is noncontroversial for the
speaker and addressee in context K with respect to A(t) .
If others were not as clear about the notion of noncontroversiality as Grice was, others
were clear about accommodation. Lewis (1979) acknowledges Stalnaker’s discus-
sion, and Stalnaker is quite explicit:
A speaker may act as if certain propositions are part of the common background
when he knows that they are not. He may want to communicate a proposition indi-
rectly, and do this by presupposing it in such a way that the auditor will be able to
infer that it is presupposed. In such a case, a speaker tells his auditor something in
part by pretending that his auditor already knows it.
When a conversation involves this kind of pretense, the speaker’s presupposi-
tions, in the sense of the term I shall use, will not fit the definition sketched above.
That is why the definition is only an approximation. I shall say that one actually
does make the presuppositions that one seems to make even when one is only pre-
tending to have the beliefs that one normally has when one makes presuppositions.
(Stalnaker 1974: 202)
I shall return to the bizarreness of Stalnaker’s (1974) account in the second para-
graph of the preceding quotation. I merely want to note here the evident utility of the
phenomena of accommodation in understanding presupposition and to distinguish
the linguistic phenomena from Stalnaker’s analysis of it as a pretended “speaker’s
presupposition.” (Stalnaker’s 1974: 200 definition of a speaker’s pragmatic presup-
position in a context is “A proposition P is a pragmatic presupposition of a speaker
in a given context just in case the speaker assumes or believes that P, assumes or
believes that his addressee assumes or believes that P, and assumes or believes that
his addressee recognizes that he is making these assumptions, or has these beliefs.”)
The notion that is required to explain presupposition is not the common knowl-
edge that is appealed to in Stalnaker’s (1974: 200) definition of a speaker’s prag-
matic presupposition in a context. Common-ground status of a proposition for a
speaker and addressee will in a speech-context be sufficient for a proposition to be
noncontroversial in that context, but common-ground status is not necessary for it to
be noncontroversial. The trouble with Stalnaker’s analysis of accommodation is that
accommodation supposedly occurs against a background of common knowledge in
which, for example, the existence of Grice’s aunt’s cousin is established and in which
an individual is identifiable as the reference of ‘my aunt’s cousin’. Among users of
language who draw on a shared family or work experience, or a fund of media-instilled
popular culture, or twelve years of primary and secondary education for purposes of
a particular conversation, the amount and significance of common knowledge may
on occasion be important and on other occasions unimportant.
The point is that for purposes of a theoretical explanation of presupposition,
common knowledge does not matter. The theoretically important notion, as Grice
(1981) and Atlas and Levinson (1981) recognized, is noncontroversiality, not common-
groundedness. What is common ground in a context for a speaker and addressee is
also noncontroversial with respect to a speaker’s utterance and an addressee’s inter-
pretation of the utterance, but what is noncontroversial with respect to a speaker’s
utterance in a context at the time of utterance need not antecedently have been in the
common ground of speaker and addressee but achieves its status simultaneously with
the utterance (Atlas 1975a, 1977a). Lewis’s (1979) notion of accommodation does
not distinguish between the addition of a proposition to the common ground simul-
taneously with the making of a statement and the specification of a proposition as
noncontroversial simultaneously with the making of a statement.
The speaker’s implicata that constitute the “presuppositions” of assertions can
reinforce propositions already in the common ground of a conversation, or they can
introduce propositions into the common ground, or they can be recognized and then
dismissed, never even entering the common ground of a conversation, because they
belong to a separate store of information that we characterize as noncontroversial.
This store of noncontroversial information is accessible for use in a conversation; it
need not be explicitly a part of the common ground, or part of mutual knowledge,
for purposes of a particular conversation. But what is noncontroversial on the occa-
sion of an utterance need not have been stored at all. A speaker’s expectation that an
addressee will charitably take the speaker’s word that a singular term t is non-
vacuous is not the same as a speaker and addressee’s expectations that they have in
common the thought t exists. What they linguistically have in common is not a
background belief; it is a language-based practice or convention of interpretation that
allows certain bits of language, such as singular terms, charitably to have a taken-
for-granted semantic evaluation in the course of making and understanding asser-
tions, but only if the singular terms are topic noun phrases in the assertions or if the
singular terms purport to designate what the statement is “about.” (See Davidson 1967;
Grandy 1973; Atlas 1988, 1989: 112.)
Reflecting on accommodation, Stalnaker (1974: 202) remarks that “presuppos-
ing is thus not a mental attitude like believing, but is rather a linguistic disposition—
a disposition to behave in one’s use of language as if one had certain beliefs, or were
making certain assumptions.” This is certainly closer to the truth than his notion of a
speaker’s pragmatic presupposition; it is his notion of a speaker’s pretended prag-
matic presupposition.
But is pretense the theoretically useful notion in explaining presupposition and
accommodation? I shall now explain to you my use of the predicate ‘x leaps tall
buildings at a single bound and flies faster than a speeding bullet’. The predicate is
true of x if and only if x leaps buildings at a single bound and flies faster than a speeding
bullet or x pretends to. Now I assert Rogers Albritton leapt buildings at a single bound
and flew faster than a speeding bullet and delight in its truth: in Rogers’s wrapping
himself in a blue cape, extending his arms into the air above his head, and taking
little leaps off the ground.
Is Grice (1981: 190) pretending to believe that his aunt has a cousin? Is he pre-
tending to believe that his addressee believes that Grice’s aunt has a cousin? Is he
pretending to believe that his addressee recognizes that he is pretending to believe
that his aunt has a cousin? And what does any of this have to do with Grice’s taking
it for granted that his addressee will take it for granted that Grice’s aunt has a cousin
when Grice asserts My cousin’s aunt went to that concert?
But, then, I am not saying anything that Stalnaker and Sadock have not already
recognized. Stalnaker (1974: 202–3, n.3) reports a counterexample of Sadock’s (per-
sonal communication):
(14) I am asked by someone whom I just met, “Are you going to lunch?” I reply, “No, I’ve
got to pick up my sister.”
Stalnaker admits that the I of the example “seems to presuppose that [he] has a sister
even though [he] does not assume that the [first] speaker knows this. Yet the state-
ment is clearly acceptable,” which it would not be if Stalnaker’s view were correct
that absent the mutual beliefs or assumptions, an assertion relying on them would be
infelicitous.8 Stalnaker continues, “and it does not seem right to explain this in terms
of pretense.”
Indeed. The situation for Stalnaker’s (1974) account of pragmatic presupposi-
tion is now this: first, “mutual knowledge” was seen to fail to account for accommo-
dation; so, second, mutual knowledge was changed to pretended mutual knowledge;
finally, pretended mutual knowledge was seen to fail to account for accommodation.
What now?
Stalnaker (1974: 203, n.3) has two replies, in the first of which he appeals to a
notion of Gricean implicature, the very notion that in his essay “Pragmatic Presup-
positions” Stalnaker proposes for explanatory purposes to replace by his notion of a
speaker’s presupposition in a context! Unfortunately this cannot rescue Stalnaker,
since he glosses Grice’s notion of implicature in terms of the notion of common
background knowledge: “the addressee infers that the speaker accepts that Q from
the fact that he says that P because normally one says that P only when it is common
background knowledge that Q,” thus missing his own point about accommodation.
8In (14) ‘my sister’ is not a topic NP, so on the Grice-Strawson condition (Atlas 1989: 112) that the
existence of a referent of a simplex singular term is presupposed in making a statement only if the sin-
gular term is a topic-designating expression in the statement, or only if the statement is “about” my
sister, the existence of a referent of ‘my sister’ is not presupposed in the statement. Although the state-
ment carries no referential presupposition with respect to ‘my sister’, if it is obligatory for me to pick
up my sister, and I did not have a sister, I could not fulfill the stated obligation. So a “preparatory con-
dition” (Searle 1969a) on statements of obligation, that it be possible to meet the condition expressed,
would not be satisfied. On the informative understanding of the denial of the obligation, the same pre-
paratory condition for the obligation denied still holds, had I replied, “Yes, I’ve not got to pick up my
sister.” On the uninformative understanding of the negative sentence, in which I have no sister, it is
nonetheless true (vacuously) that I’ve not got to pick her up. The sentence in question is a modal sen-
tence, and Atlas (1988, 1989) explicitly restricts the discussion of presupposition to extensional state-
ments; the application of the theory to intensional contexts was given in Atlas (1991b). The example
in (14) is not a case of referential presupposition, but it is case in which, for Stalnaker and others, pre-
suppositions have been identified with Felicity Conditions on the performance of speech-acts (Austin
1962/1975; Searle 1969a). ‘My sister’ could have been a topic NP in a statement similar to the one in
example (14): As for my sister, I’ve got to pick her up. But that would not have been an answer to the
question Are you going to lunch?, where the topic NP is you in the normally stressed utterance of the
question.
Stalnaker’s second reply to Sadock’s example is to consider the option of deny-

ing that there is a presupposition at all in the example, to claim that the example is an
exception to the usual cases of presupposition. Stalnaker refuses to undertake this
strategy of dealing with Sadock’s example, on the grounds of the complexity of ac-
counting for both cases in which a speaker does presuppose the existence of a unique
reference of a singular term and cases in which he does not and the consequent loss
of the simplicity of the generalization that “a speaker always presupposes the exis-
tence of a unique referent.” (A falsehood is no loss. The falsity of Stalnaker’s simple
generalization follows from the Grice-Strawson Condition (Atlas 1988, 1989) that a
statement A(t) presupposes t exists only if t is a topic NP in A(t) or only
if A(t) is about t.)
Stalnaker describes his program in “Pragmatic Presuppositions” as follows:
The contrast between semantic and pragmatic claims can be either of two things,
depending on which notion of semantics one has in mind. First, it can be a contrast
between claims about the particular conventional meaning of some word or phrase
on the one hand, and claims about the general structure or strategy of conversation
on the other. Grice’s distinction between conventional implicatures and conversa-
tional implicatures is an instance of this contrast. Second, it can be a contrast be-
tween claims about the truth-conditions or content of what is said—the proposition
expressed—on the one hand, and claims about the context in which a statement is
made—the attitudes and interests of speaker and audience—on the other. It is the
second contrast that I am using when I argue for a pragmatic rather than a semantic
account of presupposition. (Stalnaker 1974: 212)
Atlas and Levinson’s (1981) and Grice’s (1981) claim is that no adequate theory
of presupposition can maintain the contrast that Stalnaker proposes. Pragmatic intru-
sion (Levinson 1988b, 2000) and semantical nonspecificity (Atlas 1975a, 1989) show
that the content/context distinction is just one more philosophical myth—another
untenable dualism like the figurative/literal distinction (chapter 1) and the a priori/a
posteriori distinction (Putnam 1976).
As Sadock’s example shows, Stalnaker’s theory of context manifestly fails to
save the presuppositional phenomena, and his only hope of responding to the
counterexamples is the Gricean theory of conversational implicature that he wants
his theory of context to supplant.
But there is an even simpler objection to Stalnaker’s account of presupposition
as a speaker’s pragmatic presupposition. Consider the case of simple conditionals:
If P then Q. Stalnaker’s account of conditionals is given briefly:
We need first the assumption that what is explicitly supposed becomes (temporarily)
a part of the background of common assumptions in subsequent conversation, and
second that an if clause is an explicit supposition. (Stalnaker 1974: 211)
So, If the king of France is a serial killer, his mother will be ashamed of him now
requires for its felicitous assertion that the speaker assume (temporarily) that there is
a king of France, assume that his addressee assumes that there is a king of France,
and assume that his addressee recognizes that the speaker is making these assump-
tions and that the speaker assume (temporarily) that the king of France is a serial
killer, assume that his addressee assumes that the king of France is a serial killer and
assume that his addressee recognizes that the speaker is making these assumptions.
Note that the condition imposed by Stalnaker is not that one “entertains the thought,”
or “considers the consequences of,” but rather that one assumes . . . But do you?
Assume, I mean. I don’t assume, even temporarily, that there is a king of France when
I reflect on the conditional If the king of France is a serial killer, his mother will be
ashamed of him. (I would so reflect if I interpreted the conditional in the following
way: As for the king of France, if he is a serial killer, then his mother will be ashamed
of him. But that is not the interpretation in question.) And I don’t assume, even tem-
porarily, that the king of France is a serial killer. I know that there is no king of France;
why should I assume that he’s a serial killer? Do I pretend to assume these things?
No, I don’t do that, either. How about “taking it for granted”? I might take it for granted
that there was a king of France if a speaker asserted the conditional, not myself know-
ing whether there was, but do I take it for granted that the king of France is a serial
killer if a speaker asserts the conditional If the king of France is a serial killer . . . ?
No, I don’t, precisely because the king of France is a serial killer occurs in an if clause.
I don’t take the contents of if clauses for granted. So, on Stalnaker’s theory of con-
texts, I cannot felicitously or appropriately assert If the king of France is a serial
killer, his mother will be ashamed of him. Yet, surely, I can felicitously assert this
conditional.
What is worse, Stalnaker’s position is inconsistent with his own view of condi-
tionals. Consider for the moment that asserting If P then Q has the feature of re-
quiring the common background assumptions that Stalnaker claims: the speaker
temporarily assumes P, etc. Hence, according to Stalnaker, the speaker pragmati-
cally presupposes P. But Stalnaker (1974: 208) writes, “if a speaker explicitly sup-
poses something, he thereby indicates that he is not presupposing it, or taking it for
granted. So when the speaker says ‘if I realize later that P,’ he indicates that he is not
presupposing that he will realize later that P.” So Stalnaker’s (1974) account of the
context of conditionals commits him to a speaker both presupposing and not presup-
posing the content of the if clause.
None of the attitudes of the speaker and addressee that Stalnaker has consid-
ered—believing, assuming, taking for granted—correctly characterizes the linguis-
tic behavior or associated psychological states of speakers of a language, their attitudes
toward what they say or hear. Stalnaker believed that to explain the presuppositions
of assertions one should use the concept of “speaker’s pretended presuppositions”
to give a theory of linguistic contexts in which assertions are made. I believe, with
Grice, that in order to explain the presuppositions of assertions, one should use the
concept of “speaker’s conversational implicata of an assertion.” Stalnaker (1974: 202–
3, n.3) never answered Sadock’s counterexample to his theory of contexts.
It is peculiar that some of those who first noted the phenomena of accommoda-
tion have largely misunderstood its implications. The phenomena of accommoda-
tion show that the word ‘presupposition’ misnames the linguistic facts. Referential
presuppositions are a special case of accommodations; accommodations are not a
special case of presuppositions. What we want a logical and linguistic theory of is
accommodation, not presupposition. Accommodations are linguistically primary;
presuppositions are secondary—they are special cases in which t exists is not merely
accommodated as noncontroversial but also in which t exists already belongs to
the common ground. (The Newtonian motions that are primary and paradigmatic are
those at constant speed in a straight line, at constant velocity, not those that require
the application of a net force. Sadock’s objection is to Stalnaker’s account of pre-
supposition as Galileo’s objections are to Aristotle’s account of motion.) A post-
Gricean theory of conversational implicature is a theory of one type of utterer’s
meaning; it is a theory of accommodation, or, more broadly, as Levinson (2000) has
recently put it, a theory of presumptive meanings. Noncontroversiality, taking the
speaker at his word, not common-groundedness, is the notion for describing the basic
paradigm.
12 A final remark
According to the post-Gricean account in Atlas (1975a, 1979, 1989) and Atlas and
Levinson (1981), the resources for an explanatory pragmatic theory of “pre-
suppositional” inference consist in these elements: (a) the semantical nonspecificity
of ‘not’, nonspecific between choice-negation and exclusion-negation understandings;
(b) a post-Gricean mechanism for utterance-interpretation of semantically nonspecific
negative sentences; (c) principles of noncontroversial, default interpretations of state-
ments containing singular terms that are topic NPs; (d) a defensible topic/comment
distinction for statements; (e) the Grice-Strawson Condition that permits a
presuppositional inference to the referentiality of a simplex singular term in a state-
ment only if the term is a topic NP (see Atlas 1988, 1989) or only if the statement is
about the purported reference of the term; and ( f ) a distinction between propositions
that are noncontroversial and propositions that are in the common ground.
In this chapter I argued for a shift in paradigm. Accommodation is not a periph-
eral notion of presupposition: it is the central notion of presupposition. An adequate
theory of presuppositional phenomena will place the data within a Gricean theory of
presuppositional inference as an entailment of a proposition from an affirmative state-
ment and its “implicated” interpretation from an assertion of the related main-verb
semantically non-specific negative sentence, as suggested by Atlas (1975a, 1979),
and Atlas and Levinson (1981), and, in part and ignoring Grice’s mistaken appeal to
ambiguity and his mistaken application of the notion of common ground, by Grice
(1981).9
9A shorter version of this chapter has appeared in L. Horn and G. Ward (eds.) Handbook of Pragmatics
(Oxford: Blackwell, 2004), and appears here with the permission of B. H. Blackwell, Ltd.
ASSERTIBILITY CONDITIONS , IMPLICATURE , AND SEMANTIC HOLISM 149
Assertibility Conditions, Implicature,

and the Question of Semantic Holism
Almost but Not Quite
1 Comparative adjectives, adverbials of degree, and

adverbial approximatives
1.1 Disposing of the entailment view
In The Reader over Your Shoulder: A Handbook for Writers of English Prose by
Robert Graves and Alan Hodge, Principle Six of the Principles of Clear Statement is
“There should never be any doubt left as to how much, or how long” (1979: 83).1
Graves and Hodge offer the following information about English usage:
There is a popular scale of emotional approximation (not to be found in any dictio-

nary or table of measures) for estimating the comparative degrees of success in,
say, catching a train. It may be legitimately used in prose and goes something like
this: “Not nearly, nearly, almost, not quite, all but, just not, within an ace, within a
hair’s breadth—oh! by the skin of my teeth, just, only just, with a bit of a rush,
comfortably, easily, with plenty to spare.” (Graves and Hodge 1979: 83)
Just how do Graves and Hodge know this about almost and not quite?
A simple beginning in trying to explain a logical ordering of almost and not quite
would be to suggest that the statement Moore almost understood the notion “mate-
rial object” analytically entails Moore did not understand the notion “material ob-
1The first section is a revised version of Atlas (1984b) and appears with the permission of the editors of
Linguistics and Philosophy and Reidel Publishing Co.
149
ject.” If it were also true that Moore did not understand the notion “material object”
entails Moore did not quite understand the notion “material object,” it would fol-
low that Moore almost understood the notion “material object” entails Moore did
not quite understand the notion “material object.” So almost and not quite would be
ordered because any almost-state-of-affairs would also be a not-quite-state-of-affairs,
but not conversely.
The use of almost with understand indicates that there are comparative degrees
of success in understanding and that Moore “approximately” understood the notion
“material object”; after all, Graves and Hodge refer to their scale as one of approxi-
mation. Even so, there is something odd in claiming that the statement Moore did
not understand the notion “material object” is necessitated by Moore almost under-
stood the notion “material object.” If the former is part of the meaning of the latter,
it is to my ear a wee, small part of it. So I offer an intuitive datum: it is not self-
evident that x almost F’d analytically entails x did not F.
The normal counterargument to demonstrate the entailment would be to exam-
ine the conjunction of x almost F’d with the denial of x did not F for the presence of
a contradiction, thus: Moore almost understood the notion “material object” and he
understood it. But, surely, that is no obvious contradiction. So I offer another intui-
tive datum: it is not self-evident that x almost F’d and x F’d is contradictory. What,
then, is the relationship between the statements, if it is not entailment? For it is equally
clear that asserting Moore almost understood the notion “material object”—or, to
change the example, Tom almost swam the English Channel—permits an inference
to something like Tom didn’t swim the English Channel or Moore didn’t understand
the notion “material object.”
These intuitions are fine, but in addition I offer two arguments reductio ad ab-
surdum to show that x almost F’d does not entail x did not F. Suppose for adjectives,
determiners, or verbs F, the statement schema A(almost F) entails the schema A(not
F). (a) Then Almost all swans are almost white entails Almost all swans are not white,
which entails Not all swans are not white, which entails Some swans are white. But
the proposition Some swans are white is just what a speaker who asserts Almost all
swans are almost white chooses the word almost to avoid conveying. (b) It is intu-
itively evident that if there were no white swans, it could still be true that almost all
swans were almost white. However, if A(almost F) entailed A(not F), and there were
no white swans, it would be false that almost all swans were almost white, as we
have just seen in (a). Conclusion: A(almost F) does not entail A(not F). The simple
explanation of the ordering of almost and not quite is the wrong explanation.2 What
is the relationship between almost and not quite?
1.2 Post-Gricean pragmatics

The answer to the question requires an enriched pragmatic theory, which for conve-
nience I will dub “post-Gricean pragmatics,” to distinguish it from early views pro-
2Hitzeman (1992: 236–37) challenges this argument. See appendix 2 for my reply to her instructive
argument.
pounded by Grice (1967, 1975a), Horn (1972, 1976), Sadock (1981), the early Levin-
son (1983), and others, whose views I shall refer to by the Sadockian term “radical
pragmatics.” In this chapter I show that the standard, radical pragmatics application
of Grice’s Maxims of Conversation are inaccurate, inadequate, and logically inco-
herent, thus reinforcing the discussion of chapters 2 and 3. The correct account of
the logical, semantic, and pragmatic relationships among statements containing de-
gree and approximative adverbials with gradable predicates or comparative similar-
ity (equative) predicates is given by a post-Gricean, neo-Gricean pragmatics. Such a
pragmatic account contains Conventions of Noncontroversiality (Atlas and Levinson
1981; see also chapter 3 in this volume), which involve common knowledge of ste-
reotypes, cognitive preferences for some assertibility conditions (not truth conditions)
to others, the use of first-person-based (egocentric) concepts of spatial orientation,
and “metaphorical” extensions of those concepts.
Post-Gricean pragmatics has greater explanatory value in linguistics, more philo-
sophical plausibility, logical consistency, and stronger connections with recent work
in cognitive psychology than the canonical views of Grice’s “Logic and Conversa-
tion” (Grice 1967, 1975a,b; see Lakoff 1987; Horn 1989).
In expounding these views I shall use as a foil an account of almost but not quite
that would be typical of one kind of classical Gricean account, an account not neces-
sarily held by any linguist or philosopher, one that in this chapter I shall attribute to
the radical pragmatist. By criticizing this account I wish to show the need for, and
advantages of, post-Gricean pragmatics. In the course of this chapter, I shall answer
the following questions:
1. a. For gradable predicates F, does x does not F mean, or entail, x does

not quite F?
b. Conversely, for which predicates F, if any, does x does not quite F
mean, or entail, x does not F?
2. a. Does x does not quite F presuppose, or does its assertion conversa-
tionally implicate, x almost F’s?
b. If either, how is this to be explained?
3. For which predicates F, if any, does x almost F’s entail, or does its
assertion conversationally implicate, x does not F?
4. a. Does asserting x almost F’s conversationally implicate x does not
quite F?
b. If so, how is this to be explained?
5. a. Does John is as tall as Brian mean John is exactly as tall as Brian
(i.e., John’s height is the same as Brian’s), as some radical pragma-
tists hold? Or does John is as tall as Brian mean John is at least as
tall as Brian (i.e., John’s height is equal to or greater than Brian’s),
as some logical conservatives hold? Or does it mean neither, as
post-Gricean pragmatists hold?
b. Which view can explain the semi-redundancy of John is as tall as
Brian and Brian is as tall as John?
6. a. Does asserting John is not as tall as Brian conversationally
implicate John is shorter than Brian as some radical pragmatists
hold, or does the former entail the latter? Does asserting John is
not exactly as tall as Brian conversationally implicate John is
shorter than Brian, as some radical pragmatists and post-Gricean
pragmatists hold, or does the former entail the latter?
b. How are the implicatures, or entailments, to be explained?
7. a. Can John is as tall as Brian be used to mean that John and Brian
are the same height (so, John is exactly as tall as Brian)?
b. If so, how?
8. What are the entailments and implicata among John is exactly as
tall as Brian; John is as tall as Brian and Brian is as tall as John;
John is as tall as Brian; Brian is not taller than John; John is at
least as tall as Brian and their negations?
9. a. What is the logical form of John is as tall as Brian?
b. What are the logical forms of the other sentences in (8)?
My answers to these questions rely on the following semantic and pragmatic claims,
which I defend in the course of this chapter:
A. It is essential to distinguish simple gradable predicates from gradable

accomplishment and achievement predicates (Vendler 1967: 103–7).
(a) For gradable F, x does not F entails x does not quite F.
(b) Conversely, for gradable accomplishment and achievement
predicates G, x does not quite G analytically entails x does not G.
B. It is essential to construct a Horn Scale S of degree adverbials,
analogous to <all, some>, from which implicata can be predicted
(see Gazdar 1979a: 55–59; Atlas and Levinson 1981: 33; Levinson
1983: 132–36; see also chapter 3 in this volume):
(S) <quite/ wholly/ entirely, almost, nearly, mostly/ mainly, rather, partly>
Here, mostly means for the most part/ mainly / more than half; the
approximative almost means for the greatest (proper) part; and quite
is polysemous between wholly and rather. Scale S shows that (a) for
gradable F asserting x almost F’s implicates x does not quite F;
(b) for gradable accomplishment and achievement predicates G,
asserting x almost G’s implicates x does not quite G, which analyti-
cally and “directly” entails x does not G. Thus “x almost G’s” D °
» x does not G. That is, x almost G’s does not entail x does not G; x
almost G’s “quasi-entails” x does not G. Notice that a quasi-entail-
ment is not an entailment; it is the composition of the implicature
relation » with the entailment relation . (c) For gradable F
asserting x does not quite F implicates x almost F’s.
C. For unmarked, gradable F in a comparative similarity relation (an
“equative”) expressed by x is as F as y—for example, x is as tall as
y, x is as F as y means Whatever (measures of) F-ness y has, x has
also.
My notion of measures of F-ness is not a notion of the “extent” to which an in-

dividual x exemplifies the gradable predicate F. In that case, it is natural to identify
the extent of x’s tallness with x’s height. If the extent of John’s tallness is ej, and the
extent of Brian’s tallness is eb, John’s height is the same as Brian’s if and only if ej =
eb. John is taller than Brian if and only if ej > eb. This is not my conception of mea-
sures of tallness. On the conception of extents, an individual exemplifies tall to a
unique extent; on my conception of measures, an individual exhibits many measures
of tallness. On the conception of extents, the extent to which an individual exempli-
fies tall is the individual’s height; on my conception of measures, the individual’s
height is the least upper bound of the measure of tallness that the individual has
(i.e., x’s height = sup {m: x has m of tallness}). Of course on my view, if an indi-
vidual has measure m of tallness, he has m' if 0 < m' ≤ = m.
It is essential to note that x has some measure m of tallness does NOT entail x is
tall; “having tallness” and “being tall” are distinct concepts.
One difference between scientists and pure mathematicians is that scientists are
interested in the results, whatever the proofs, while mathematicians are interested in
the proofs, whatever the results. The impatient reader who wants the neo-Gricean
answers to my nine questions immediately will find them summarized in the foot-
note.3 The patient reader who enjoys argument may proceed to subsection 1.3. The
use of almost but not quite is part of our daily linguistic practice. My post-Gricean
3 1a. For gradable predicates F, x does not F entails x does not quite F.
1b. x does not quite F does not mean x does not F; for gradable, accomplishment and
achievement predicates G, x does not quite G analytically and directly entails x does
not G.
2a. x does not quite F does not presuppose x almost F’s; asserting x does not quite F
conversationally implicates x almost F’s.
2b. Quite and almost form a Gricean Horn (1972, 1976) Scale of degree adverbials
analogous to <all, some>, where all of the N VP entails some of the N VP, where
asserting some implicates not all, and asserting not all implicates some. The Horn
Scale is <quite/wholly/entirely, almost, nearly, mostly/mainly, rather, partly>, where
mostly means {for the most part/mainly/more than half}; almost means for the
greatest proper part; and quite is polysemous: (i) wholly, (ii) rather.
3. x almost F’s does not entail, does not presuppose, and its assertion does not
conversationally implicate x does not F; for accomplishment and achievement
predicates G, asserting x almost G’s implicates x does not quite G, which analyti-
cally and directly entails x does not G.
4a. Asserting x almost F’s conversationally implicates x does not quite F.
4b. By the Horn Scale described in 2b.
5a. John is as tall as Brian means neither John is exactly as tall as Brian nor John is at
least as tall as Brian.
5b. Only the neo-Gricean view can explain the semi-redundancy of John is as tall as
Brian and Brian is as tall as John.
6a. John is shorter than Brian entails John is not as tall as Brian, but not conversely.
Asserting John is not exactly as tall as Brian conversationally implicates John is
shorter than Brian.
6b. The implicature cannot be explained by Grice’s (1975a) or Sadock’s (1981) radical
pragmatics. Post-Gricean pragmatics will explain it.
7a. Asserting John is as tall as Brian implicates, in a post-Gricean fashion, John is
exactly as tall as Brian.
pragmatics makes explicit what the tacit linguistic knowledge underlying this prac-
tice consists in, and that brings us a certain kind of philosophical enlightenment, in
the spirit of Wittgenstein, Austin, Dummett, Strawson, and Grice.
1.3 Post-Gricean criticism of the radical

pragmatics account
The radical pragmatics analysis that I criticize here, as discussed in Sadock (1981),
is the following:
x almost F’d
(D) i. asserts x almost F’d
(A) ii. implicates x did not F
x did not quite F
(C) i. asserts x did not F
(B) ii. presupposes x almost F’d
For purposes of argument, I shall take this analysis as a paradigm of a kind of radical
pragmatics account and attribute it to a hypothetical figure, the radical pragmatist.
The pattern of analysis is clever and familiar. In 1971 Charles Fillmore suggested
that criticize and blame had assertoric and presuppositional parts; blame presupposed
that the object of blame is at fault while asserting that he is responsible; criticize
presupposed that he is responsible while asserting that he is at fault. The radical
pragmatist makes the pattern more pragmatic by a substitution of implicates for
presupposes.
Does saying almost implicate not?

The radical pragmatist believes that in answer to Did Tom swim the English Chan-
nel? Almost but not quite is acceptable, but ?Almost and not quite is not. Accord-
ingly, a substitution of and for but that similarly fails to preserve acceptability occurs
in the pair: (a) John ate some, but not all, of the cake, (b) ?John ate some, and not
all, of the cake. (The latter sentence may be used acceptably to deny that John ate all
of the cake, as Sadock (1981: 264) notes, so there is some oddity in the radical
pragmatist’s questioning the acceptability of the sentence.) If saying John ate some
of the cake implicates John did not eat all of the cake, a speaker can make the
implicatum explicit, reinforcing it, by an assertion. If the implicatum had been an
7b. The explanation is provided in Atlas and Levinson (1981: 44).

8. See table 5.1 and 5.2 in chapter 5.
9a. John is as tall as Brian means Whatever (measures of) tallness Brian has, John has
also. Quantifying over the measures of tallness that each has, or j* and b*, the
logical form is (∀b*)(∃j*)( j* = b*).
9b. j is exactly as tall as b (∀b*)(∃j*)( j* = b*) & (∀j*)(∃b*)(b* = j*); b is taller than j
( j is shorter than b) (∃b*)(∀j*)( j*<b*); j is at least as tall as b (∀b*)(∃j*)( j* ≥ b*).
(easily recognized) entailment, reinforcement would have been redundant. But no

whiff of redundancy attaches to John ate some, but not all, of the cake. Likewise, the
sentences Tom almost swam, but didn’t swim, the English Channel and Tom almost,
but not quite, swam the English Channel are not redundant. From the latter state-
ment the radical pragmatist concludes that saying almost implicates not quite. (The
radical pragmatist believes that not quite means not, as we shall see.)4
Besides reinforceability, another characteristic of implicature is cancelability.
Even though saying John ate some of the cake implicates John did not eat all of the
cake, it is acceptable to deny the implicatum by saying John ate some, and in fact
all, of the cake. The negation of a conversational implicatum may be asserted without
contradiction. Thus it is acceptable to say Not only did John eat some of the cake, in
fact he ate all of it. The analogous sentence Not only did Tom almost swim the English
Channel, in fact he did swim it is noncontradictory, but the radical pragmatist claims
that it is a “bit odd,” even “very odd” (Sadock 1981: 263–64). This suggests to him
that, for some reason, it is harder to cancel the (alleged) implicatum not of almost
than the implicatum not all of some. He calls the (alleged) implicatum of almost a
“strong” implicatum. In short, the radical pragmatist concludes that saying almost
conversationally implicates, even “strongly” implicates, but does not entail, not.5
4Larry Horn (personal communication) has made the following noteworthy observation: “The radical
pragmatist’s basic position seems to be that P and Q is always redundant when Q is (entirely) entailed,
presupposed (cf. the Kiparskys’ ‘Fact’ on this claim), or asserted by P, but not when Q is merely impli-
cated by P. Thus pragmatic inferences may be reinforced nonredundantly but logical inferences may
not be. (The same claim is also made in Horn (1972, 1976): chap. 2). But while this claim may be largely
correct for and conjunctions, it is clearly false when the conjunction is but, vitiating the radical
pragmatist’s test, and arguments derived therefrom. Thus consider:
It’s odd that she married him, but she did marry him.
I don’t know why I love you, but I do (love you).
I regret that I must say this, but say it I must.
Nobody but John can do it, but he can do it.
I’ve sinned only once, but I have sinned.
In each case the material in the but clause is semantically or logically inferrable from the material in
the first clause, yet (redundant as they might be) the Q parts of each P but Q conjunction may be as-
serted without deviance. . . . Notice that the well-formed cases of reinforcement involving but all con-
tain one negative clause (either with overt negation or with inherent negation, as in the case of regret,
only, odd, where the presence of the negative is confirmed by the fact that these are all polarity trig-
gers: It’s odd that he ate anything, etc.) and one affirmative clause. It is this contrast in polarity that
requires but.” (For more on only and negative polarity, see Atlas (1996b, 1997b, 2001).
5The radical pragmatist’s appeal to “strong” implicature is an ad hoc explanation for the alleged diffi-
culty of canceling the implicature in Not only did Tom almost swim the English Channel, in fact he did
swim it. On Sadock’s (1981: 264) assessment, this radical pragmatics account’s most pressing diffi-
culty arises from this: if saying almost implicates not, the implicatum should be cancelable, and can-
celation is allegedly difficult. But I believe that the sentence can be asserted acceptably, just like the
equivalent: Not only did Tom almost swim the English Channel, in fact he quite swam it. There is no
difficulty in canceling the implicatum. On this point Sadock was too pessimistic about the chances for
a successful pragmatic account.
Cancelation proceeds in the usual way. Asserting A or B implicates not(A and B); so A or B, and
in fact A and B is consistent, even if “a bit odd.” Likewise, saying Not only did Tom almost swim . . .
Does not quite presuppose almost?

If the radical pragmatist had concluded that saying almost implicates not quite rather
than concluding that saying it implicates not, then an analogue with not all/some would
have suggested that saying not quite also implicates almost. Just as saying some impli-
cates not all, so saying not all implicates some: thus, saying John did not eat all of the
cake implicates but does not entail John ate some of the cake. The implicatum is rein-
forceable without redundancy: John didn’t eat all, in fact he ate some, of the cake. The
implicatum is also cancelable: John didn’t eat all of the cake, in fact he ate none of it.
The radical pragmatist concluded that saying almost implicates not because the
data showed that saying almost implicates not quite, and he or she believed, for rea-
sons I shall examine, that not quite means not. Then, instead of arguing for the con-
verse of saying almost’s implicating not quite—that is, that saying not quite would
implicate almost—he or she argued instead that not quite presupposes almost. This
argument, I believe, is faulty, so let us examine it.
In answer to the question Did Tom swim the English Channel?, the radical prag-
matist believes that Almost but not quite is acceptable but that [?]Not quite but al-
most is odd. He or she notices a parallel with the acceptable It rained and Tom realized
that it had and the odd ?Tom realized that it had rained and it had. The explanation
of the oddity—the redundancy—is that the second conjunct asserts what the first
presupposes. (Actually, such an oddity would be just as well explained if the first
conjunct merely entailed the second.) Well, how odd is the sentence?
Such sentences have natural contexts: Tom realized that it had rained and—he
had reason to since—it had. What would be distinctly odd is ?Tom realized that it
had rained but it had. It is the contrast conventionally implicated (in Grice’s origi-
nal sense) by but rather than the presupposition (if there is such) of the second con-
junct by the first that is essential to explaining the oddity of ?Tom realized that it had
rained but it had.
Just as there is a point in the use of but to contrast some with all in John ate
some, but not all, of the cake, so there is in Tom almost, but not quite, swam the English
Channel: it reinforces the contrast of almost and quite. This is roughly the contrast
between nearly and wholly. For example: Are you almost/nearly finished with your
meal? and Are you quite/wholly finished with your meal? Quite and almost are ad-
verbs of degree and approximation. Just as asserting some implicates not all, assert-
ing almost implicates not quite; but just as asserting not all implicates some, so denying
quite, or asserting not quite, implicates almost.
implicates Not only did Tom not quite swim . . . For achievement predicates like swam the English
Channel, Not only did Tom not quite swim . . . entails Not only did Tom not swim . . . , which entails
Tom did not swim . . . So “part” of what is implicated, not entailed, is then denied without contradic-
tion in the continuation . . . in fact he did swim it. (My position on this point was clarified for me by a
comment of Scott Soames’s.)
Hitzeman (1992: 228) claims that the following is contradictory: ?Bill ate almost all of the cake
and he ate all of the cake. I do not see why the alleged “oddity” should imply logical inconsistency,
even if the statement is odd. But cancelation can be expressed by the acceptable Bill ate almost all,
and in fact all, of the cake or by the acceptable Bill ate almost all of the cake, in fact he ate all of it.
Hitzeman’s evidence does not support the noncancelability claim, much less the inconsistency claim.
If these neo-Gricean suggestions are correct, there is not the parallel between
Almost but not quite / [?]Not quite but almost and It rained and Tom realized that it
had / ?Tom realized that it had rained and it had that the radical pragmatist relies on.
In fact, I think there is no parallel. Not quite but almost is simply not odd: Tom didn’t
quite swim the English Channel, but he almost swam it is acceptable, and certainly it
is not parallel to ?Tom realized that it had rained and it had or ?The Queen of En-
gland is gracious and she exists. If the alleged oddity of Not quite but almost were
parallel to ?Tom realized that it had rained and it had, and so explained by not quite
presupposing almost, the parallel with the acceptable It rained and Tom realized that
it had should also be acceptable, but it isn’t: ?Tom almost swam the English Channel
and he didn’t quite swim it. The latter requires a substitution of but for and to ensure
acceptability: Tom almost swam the English Channel, but he didn’t quite swim it.
By contrast, replacement of and by but in the acceptable It rained and Tom re-
alized that it had yields the acceptable It rained but Tom realized that it had. So if
the radical pragmatist’s argument that the alleged parallel between Almost but not
quite / ?Almost and not quite and Some, but not all / ?Some, and not all supports the
claim that saying almost implicates not quite, the lack of parallelism between Almost
but not quite / ?Almost and not quite and It rained but Tom realized that it had / It
rained and Tom realized that it had would undermine, on his own grounds, his claim
that not quite presupposes almost.
There is one further intuition that pushes him to the view that not quite presup-
poses (or has as a Karttunen-Peters 1979 style conventional implicature) almost. He
believes that Tom didn’t quite swim the English Channel is true if Tom did not even
come close, but that it is inappropriate in just the same way as is any utterance whose
presuppositions are not satisfied (Sadock 1981: 263).
Quite and almost are not the only items in the relevant degree adverbial scale.
Asserting Tom did not quite swim the English Channel implicates Tom partly swam
the English Channel. Since Tom quite swam the English Channel entails Tom partly
swam the English Channel, we see exhibited the formal properties associated with Tom
did not quite swim the English Channel presupposing Tom partly swam the English
Channel. If a speaker asserted not quite, then one would infer at least that the speaker
did not believe that Tom did not partly swim the English Channel and ceteris paribus
the stronger claim that the speaker did believe that Tom did partly swim the English
Channel (as in the usual First Maxim of Quantity implicata; Grice 1975a).
The content of the speaker’s belief in the case of this stronger inference is just
what one has been tempted to call the presupposition of the assertion. For if this is
false—if Tom scarcely swam the English Channel at all—and the speaker knew or
believed this, he would be asserting the sentence in circumstances in which an audi-
ence would, in general, be mistaken in its inference from what the speaker said to
what the speaker believed. When the communally shared understanding of the use
of the language systematically fails to produce knowledge of the speaker’s beliefs
from what the speaker says, one takes the speaker’s utterance to be inappropriate to
its circumstances. He is misusing the community’s language. This neo-Gricean ac-
count explains the radical pragmatist’s intuition that if Tom paddled about Dover,
the statement Tom didn’t quite swim the English Channel would be a case of true
English understatement but inappropriate in the circumstances.
As far as the claim that not quite presupposes almost goes, I’ve questioned the
acceptability judgment that initially suggested it. I have just shown that the kind of
parallelism argument employed in support of the presupposition analysis can be re-
versed. The demonstrated nonparallelism undermines the presupposition analysis. I
have argued that the intuitions about conditions of appropriate utterance can be given
a neo-Gricean, and in this case classically Gricean, pragmatic explanation. The radi-
cal pragmatist has tried to show, I believe unsuccessfully, that saying almost impli-
cates not and that not quite presupposes almost. I now turn to the third element in
this kind of account, the view that not quite means not.
Does not quite mean not?

As I have already explained, essential to the view that saying almost implicates not
and not just not quite is the view that not quite means not. Alleged evidence for this
is the alleged contradiction in ?Not only did Tom not quite swim the English Chan-
nel, in fact he did swim it. But at most a contradiction would show that Tom did not
quite swim the English Channel entails Tom did not swim the English Channel or,
equivalently, if Tom swam the English Channel, he quite (i.e., wholly) swam it.
However, the inference x G’s, therefore x quite G’s is correct only for accomplish-
ment and achievement predicates. For example, if Alexander cut the Gordian knot,
he quite cut it; if Albert knew the answer, he quite knew it. (The accomplishment
predicate indicates the completion of a bounded process or sequence.) This evidence
cannot show that not quite means not. At best, for some predicates G, not quite G
entails G (see Vendler 1967: 103–7).
What does almost mean?

If x almost F’s is asserted, and thereby implicates anything, it does so by virtue of
the language user’s understanding of the sense [X ALMOST F’S] of x almost F’s. A clas-
sical Gricean can offer a provocative discussion of the semantics of ‘almost’, to which
I now turn.
If Tom almost swam the English Channel, it is tempting to take this as support
for the counterfactual: Had things been otherwise than they were, but not markedly
so, Tom would have swum the English Channel. It could also be taken as support for
the modal claim: It was possible for Tom to have swum the English Channel; or, Tom
could have swum the English Channel. To support, however, is not to entail, as the
consistency of Tom almost swam the English Channel, but he couldn’t really have
swum it indicates. Tom could have swum the English Channel does not express a
necessary condition of Tom almost swam the English Channel. Nonetheless, the clas-
sical Gricean adopts the former condition as his characteristically minimal truth con-
ditions for the latter. Explicitly, Tom almost swam the English Channel is true just in
case there is a possible world, not very different from the actual world, such that Tom
did swim the English Channel (Sadock 1981: 259).
One defense of this minimal, classical Gricean proposal is its claim to capture the
vagueness of almost. For example, the classical Gricean notes the existence of vague-
ness in sentences employing quantitative predicates: My filing cabinet is almost six feet
tall; Stephen is almost six feet tall. Cabinets that are five and one-half feet tall make the
statement true; cabinets that three feet tall make the statement false; and cabinets with
intermediate heights make determinate answers difficult to give. The same holds for
persons, except the relevant boundaries shift. If Stephen is five and one-half feet tall,
the statement is false. The classical Gricean theorist justifies his minimal formulation
of the truth conditions for almost by claiming to have preserved in his translation not
very different from the actual world in the metalanguage the vagueness of almost in
the object language. The preservation of vagueness is an appropriate constraint on trans-
lation, one that has not been properly understood by Donald Davidson or his followers
in applying the ideas of Tarski’s theory of truth to natural language, or by Richard
Montague or his followers (see Katz 1978; Keenan 1978; Atlas 1980b: 139). But if
one is to justify a minimal formulation of truth conditions on such grounds, one might
wonder whether vagueness is being preserved in the right way (Atlas 1980b: 139).
Consider the sentence The Newtonian theory of the precession of Mercury’s orbit
almost produces the observed value (to within experimental error). We are to under-
stand this to be true just in case there is a possible world not very different from the
actual one in which the Newtonian theory of the precession of Mercury’s orbit does
produce the observed value (in that world). One way to imagine such a possible world
would be as a state of affairs with planets and a sun with masses, relative velocities,
and other physico-chemical properties indistinguishable from those of our planets
and sun, but as a state of affairs in which Newton’s theory of gravitation is true. After
all, it might have been true. In that state of affairs the theory would, of course, pro-
duce the observed value (in that world).6
But there is no more difficulty in the judgment that a Newtonian universe is radi-
cally different from our actual Einsteinian one than there is in the judgment that a
foot-high filing cabinet is not almost six feet tall. For example, in a Newtonian uni-
verse it is a brute, albeit almost magical fact about physical bodies that the inertial
mass of a body is quantitatively the same as its gravitational mass. This is just a con-
sequence of Galileo’s Law, that objects released from rest from the same position
near the surface of the earth will drop (in a vacuum) at the same acceleration, no
matter how massive they are. In an Einsteinian universe this is not a brute fact about
physical bodies; it is explained by the indistinguishability of accelerating reference
frames from frames at rest in inverse-square force fields. It is a feature of physical
motion itself. On this classical Gricean account, The Newtonian theory of the pre-
cession of Mercury’s orbit almost produces the observed value is true just in case
there is a possible world, not very different from the actual world, in which the
Newtonian theory produces the observed value. As far as I have any intuitions about
the similarity of possible worlds, these truth conditions are not satisfied. In contrast,
there is a strong intuition that the statement in question in fact is true. More graphic
examples of the same sort would be Light almost doesn’t bend in gravitational fields
and Planets travel in almost circular orbits. Intuitively these statements are true, but
6Those readers with Kripke’s intuitions about proper names may find the following variant less trou-
bling: The Newtonian theory of the precession of the orbit of the planet nearest the sun almost pro-
duces the observed value. For a refutation of Putnam’s (1975, 1978a) related analysis of natural kind
terms, see Atlas (1980b: 134–36, 1989: 136–39) and Chomsky (1995b: 44–45; 2000: 148–51).
any possible world in which light does not bend in gravitational fields is radically
different in nature from the actual world, as is an anti-Keplerian world in which planets
travel in circular orbits.7
There is a more serious difficulty. In the metalinguistic translation of the object-
language sentence-schema ((NP)(almost VP)), the almost is semantically misplaced.
On the classical Gricean theorist’s analysis, the sentence schema is equivalent to That
((NP)(VP)) is almost true. Yet A rose by any other name would almost smell as sweet
is hardly equivalent to That a rose by any other name would smell as sweet is almost
true or to That a rose by any other name would smell as sweet is possibly true. The
classical Gricean theorist is treating the adverb almost as if it were exclusively a sen-
tence adverb.
The classical Gricean theorist’s first fallback position, an alternative that Sadock
(1981) mentions, is that the truth conditions of Tom almost swam the English Channel
are the even more minimal condition that there is a possible world in which Tom swam
the English Channel, plus a “conventional implicature” (of the Karttunen-Peters type)
that the possible world in question is not very different from the actual one. This ac-
count of the truth conditions of the sentences I have been considering would give them
correct rather than incorrect truth-values, but it would also predict that the sentences
would be semantically anomalous in our actual world. That prediction is just false.
If one now opts for an even more purely pragmatic account (where Tom almost
swam the English Channel is true just in case there is a possible world in which Tom
swam the English Channel, but it would be infelicitous to assert the sentence were
the possible world in question very different from our own), the sentences I have
been considering would be inappropriate to assert. That account just seems false. In
addition, the statement Tom is almost bald would be predicted to be true, even though
actually Tom has a full head of hair.
Sadock (1981) was sensitive to deficiencies in the classical Gricean theorist’s
sentence-adverbial truth conditions for almost. He observes that on the analysis there
is no evident truth-conditional difference between John is almost six feet one inch
tall and John is about six feet one inch tall. But from almost we know that John is not
six feet one inch tall and that John is less than six feet one inch tall, both of these
being, on Sadock’s view (discussed later in this chapter), generalized conversational
implicata of asserting the sentence. Since implicata depend on the sense of the utter-
ance, if one statement engenders implicata that another doesn’t, the statements would
be expected to differ in sense. In that regard the apparently intelligible sense of It
was almost true that Tom swam the English Channel is very misleading. The sentence-
adverbial paraphrase is, in general, either unintelligible or, when intelligible, inac-
curate. It is clear that It is almost true that John is as tall as Brian, even if logically
coherent, which I doubt, since I doubt the conceptual coherence of degrees of truth,
cannot have the sense of John is almost as tall as Brian. Asserting the latter seems to
implicate John is shorter than Brian; asserting the former clearly does not.
7Cf. Sadock (1981: 258–59, n.2) where Sadock relativizes the truth conditions to doxastically possible
worlds—belief worlds—including logically and physically impossible but imaginary worlds. In these
circumstances, it seems to me, the similarity relation among worlds becomes opaque, and the judg-
ments of similarity become unreliable.
1.4 The problem of inconsistent classical Gricean inferences

Neo-Gricean pragmatics explains this apparent implicature as follows: (a) Saying
John is almost as tall as Brian implicates John is not quite/wholly as tall as Brian;
(b) John is not quite/ wholly as tall as Brian entails Brian is (uniformly) taller than
John;8 (c) Brian is taller than John analytically entails John is shorter than Brian.
So John is shorter than Brian is “part of,” or analytically entailed by, what is impli-
cated by asserting John is almost as tall as Brian. If we compose the relations of
implicature » and entailment , then the argument shows that John is almost as
tall as Brian ° » John is shorter than Brian. We can give the composition of the
8In the formalization offered here, John is not quite/wholly as tall as Brian is represented in part by:
(∃b*)(∀j*)( j* < b*); that is, some measure of tallness that Brian has is greater than any measure of
tallness that John has. In mathematical vocabulary, Brian is “uniformly” taller than John, which for
reasons to be discussed I take to represent Brian is taller than John. (‘In part’ because I treat a2 is not
quite/wholly as F as a1 as analytically and logically equivalent to a1 is as F as a2 & a1 is uniformly
F-er than a2. Note that this formally entails that Brian is “measurewise” taller than John: (∀j*)(∃b*)(b*
> j*); thus for each measure of John’s tallness there is some measure of Brian’s tallness that exceeds
it.) Hence John is not quite/wholly as tall as Brian entails Brian is taller than John.
Interestingly, from the point of view of our formalization, Brian is taller than John does not in
our formalization entail Brian is as tall as John. That is because though there is a measure of tallness
that Brian may have, a measure of “highness” as it were, that is greater than any measures that John
may have (and hence for any measure that John has, there is some measure that Brian has that is greater),
it does not thereby follow that for any measure that John has, there is some measure that Brian has that
is identical to it. Thus, a mockingbird sitting at the top of a California live oak tree could be “taller”
than I am, in the sense that it is higher off the ground than I. ‘High’, according to some, is polysemous
between ‘great extension, as distance from bottom to top’ and ‘great elevation’. ‘Tall’ according to the
lexicographers of Merriam-Webster’s Collegiate Dictionary (2003) at the G. & C. Merriam Co., Spring-
field, Massachusetts, is restricted to the former sense; according to the lexicographers of The Ameri-
can Heritage Dictionary of the English Language, 3rd ed. (1996) at Houghton Mifflin Co., Boston,
Massachusetts, ‘tall’ is not so restricted. According to the latter, the mockingbird is taller than I am;
according to the former, the mockingbird is merely higher than I am. Having been educated in western
Massachusetts, I always knew that Bostonians couldn’t speak (American) English. The Springfield
lexicographers are right, of course; the mockingbird in question is not taller than I am. Now, whether
‘tall’ is actually polysemous, or whether it is really monosemous and semantically nonspecific between
[EXTENT] and [ELEVATION], I shall not be tempted to discuss here (and, as far as I can tell, neither was
Ruhl 1989). If the latter is the case, then I shall stipulate that in the cases in question in my arguments,
I mean ‘tall’ in the [EXTENT] interpretation (not sense). But in one of its senses or in that interpretation,
one needs to guarantee that if Brian is measurewise taller than John, then Brian is as tall as John, if one
wishes to guarantee that there is an “analytical entailment” from Brian is [uniformly] taller than John
to Brian is as tall as John. One would need a Carnapian Meaning Postulate for English ‘taller’ and ‘as
tall as’: English (∀x)(∀y)(x is [measurewise] taller than y → x is as tall as y). (In the formalization
that I have adopted: E (∀x)(∀y)((∀y*)(∃x*)(x* > y*) → (∀y*)(∃x*)(x* = y*)).) If one adopts that Mean-
ing Postulate for English, then John is not quite/wholly as tall as Brian becomes analytically equiva-
lent to Brian is [uniformly] taller than John; the former does not merely entail the latter. But entailment
suffices for my argument.
I have already defended proposition (a): saying John is almost as tall as Brian implicates John is
not quite/wholly as tall as Brian. The further entailment proposition (d) that John is not quite/wholly
as tall as Brian entails John is not as tall as Brian is intuitively evident, and my formal analysis pre-
dicts it. My explanation actually makes a confirmed linguistic prediction. (The entailment in (d) is
equivalent to: Brian is as tall as John & Brian is [uniformly] taller than John entails John is not as tall
as Brian. That obviously (and my analysis agrees) reduces to the question whether Brian is [uniformly]
two relations a technical name—‘quasi-entailment’—whose relata are an assertion

and a sentence (proposition). Thus the assertion John is almost as tall as Brian quasi-
entails the sentence (proposition) ‘John is shorter than Brian’. (Do not forget that a
quasi-entailment is not an entailment.)9
By contrast, the classical Gricean theorist explains this implicature as follows:
(a) Saying John is almost as tall as Brian implicates John is not as tall as Brian;
(b) saying John is not as tall as Brian implicates John is shorter than Brian; (c) there-
fore, by an assumption of transitivity, saying John is almost as tall as Brian impli-
cates John is shorter than Brian.10
My interest particularly lies in the classical Gricean theorist’s provocative argu-
ment for (b): saying John is not as tall as Brian implicates John is shorter than Brian:
taller than John entails John is not as tall as Brian. My formal analysis confirms that the entailment
holds.) The implicature from x is almost as F as y to x is not quite/wholly as F as y and the analytical
entailment from x is not quite/wholly as F as y to x is not as F as y make the equative as F as similar
to an achievement predicate G—for example, swam the English Channel. Thus we should expect to
get sentences with modifiers like easily, and we do: John is easily as tall as Brian, John is within a
hair’s breadth of being as tall as Brian.
9There is another possible explanation of the inference on the theory that I am offering, but it is not
adequate, and the reasons for its inadequacy are of some theoretical interest, so I shall take the oppor-
tunity to mention it. One could argue: (a) Saying John is almost as tall as Brian implicates John is not
quite as tall as Brian; (b) John is not quite as tall as Brian entails John is not as tall as Brian; (c) John
is not as tall as Brian entails John is not as tall as Brian or Brian is not as tall as John; (d) John is not
as tall as Brian or Brian is not as tall as John entails John is not exactly as tall as Brian; (e) saying
John is not exactly as tall as Brian implicates John is shorter than Brian. So John is shorter than Brian
is, loosely speaking, inferrable from saying John is almost as tall as Brian.
There is nothing incorrect in the premises of this argument, and I have hedged the conclusion by
the phrase “loosely speaking.” Nonetheless there are problems. The first problem is that premise (c)
requires the classical “irrelevant” entailment from P to P ∨ Q. The entailment is classically valid but in
this argument linguistically unmotivated. Premises (c) and (d) show that John is not as tall as Brian
entails John is not exactly as tall as Brian, which is the intuitively acceptable, logically equivalent
contrapositive of John is exactly as tall as Brian entailing John is as tall as Brian. The problem is that
the linguistic, as contrasted with the purely logical, justification, for (c) lies nowhere to hand. Nothing
independently motivates it; thus it would be better to reduce the argument to one in which (c) and (d)
are replaced by one premise (cd) John is not as tall as Brian entails John is not exactly as tall as Brian.
But there is a remaining problem. Premise (e) introduces the “saying” of John is not exactly as
tall as Brian into an argument that proposes to explain why “saying” John is almost as tall as Brian
implicates John is shorter than Brian. But in saying John is almost as tall as Brian one does not say
(assert) John is not exactly as tall as Brian, even though the latter is entailed by an implicatum of the
former, unless one goes in for an act of mentally asserting one of the entailments of an implicatum of
a verbal assertion when one needs to generate a further implicatum—namely, that John is shorter than
Brian. So if such an argument is to have psycholinguistic plausibility, one would have to explain why
this particular entailment gets “mentally” asserted in order to yield the appropriate implicatum. That is
a problem for which I have no, and expect no, good solution, at least in the immediate future.
For further discussion of this sort of argument, see Horn (1996b), and Atlas (1996b, 1997b).
10Sadock writes: “I assume that conversational implicatures are ordinarily transitive; that if A conver-
sationally B, and B conversationally implicates C, then A conversationally implicates C as well.[*]”

He continues in his footnote, “[*] William Lycan (personal communication) has pointed out that there
is no particular reason to expect conversational implicatures to be transitive. A conversational impli-
cature, according to Grice, is generated on the basis of the fact that a particular linguistic form,
Without the implicature [John] is not as tall as [Brian] would say only that [John]
and [Brian] are not exactly the same height. This is not an especially interesting
fact since two people chosen at random rarely are. Therefore implicatures are strongly
favored with negative statements like [John] is not as tall as [Brian]. Furthermore,
it is clear why the implicature of [John] is not as tall as [Brian] is that [John] is less
tall than [Brian], rather than vice versa. The sentence is about [John]. It would clearly
be saying more about [John] to say that his height was greater than [Brian’s] than
to say that it was not greater. So by failing to say that [John] is taller than [Brian],
the speaker of [John] is not as tall as [Brian] . . . conveys that [John] is not taller,
via [Grice’s First] maxim of Quantity [”Make your contribution as informative as
is required for the current purposes of the exchange”]. (Sadock 1981: 269)
Quite important in this argument, and unfortunately ignored in most classical Gricean
arguments, is the question of topic—what a statement is about.11 However, the claim
with a particular meaning, is uttered. Since the first-order implicature is not uttered, we should not expect
it to produce further conversational implicatures. Yet Grice himself talks at times as if it can. In dis-
cussing the effect of an abrupt change of topic in a conversation, following the remark ‘Mrs X is an old
bag’, Grice writes: ‘B has blatantly refused to make what he says relevant to A’s preceding remark. He
thereby implicates that A’s remark should not be discussed and, perhaps more specifically, that A has
committed a social gaffe’ (Grice 1975a: 54). Here it is quite reasonable to suppose that the more spe-
cific implicature is a result of the conveyance of the more general one. After all, if B had changed the
topic by saying That remark should not be discussed, the more specific implicature would also go
through” (Sadock 1981: 269).
From sympathy with Sadock and Grice I once tried to make out the case for the following kind of
transitivity: P entails Q; Q presupposes R; R pragmatically implies T; so P pragmatically implies T. (I
failed miserably, as William Harper, Rob van der Sandt, and Scott Soames pointed out to me.) But
were this transitivity correct, the classical Gricean theorist’s analysis—saying x almost F’s asserts x
almost F’s and implicates x does not F; saying x does not quite F asserts x does not F and presup-
poses x almost F’s—would be structurally absurd, no matter what the truth conditions x almost F’s
of x almost F’s are. My argument was this: If x does not quite F means [X DOES NOT F], the former en-
tails the latter. But by contraposition, satisfied by entailment, x F’s entails x quite F’s. If x does not
quite F presupposes x almost F’s, by the invariance of presupposition under negation, x quite F’s
presupposes x almost F’s. If asserting x almost F’s implicates x does not F, one says that the asser-
tion-content x almost F’s “pragmatically implies” x does not F. It follows, by the transitivity in
question, that x F’s pragmatically implies x does not F, which is obviously absurd.
Still, the structure of the analysis does not seem all that absurd, so one suspects that the transitiv-
ity principle is incorrect, which indeed it is. For example, let P = Jonathan knew that Mart regretted
that Laura seduced Jim or Rick, Q = Mart regretted that Laura seduced Jim or Rick, R = Laura se-
duced Jim or Rick, and T = Laura did not seduce both Jim and Rick. It is absurd that Jonathan knew
that Mart regretted that Laura seduced Jim or Rick pragmatically implies Laura did not seduce both
Jim and Rick, or so it seems to many.
A simpler and clearer case is a counterexample to the transitivity schema P entails Q, Q prag-
matically implies R; so P pragmatically implies R. Rob van der Sandt’s simple counterexample is: Let
P = φ, Q = φ ∨ Ψ; R = ¬(φ & Ψ).
My own objection to the principle of transitivity for implicatures is the following reductio ad ab-
surdum argument: according to the principle, asserting some implicates not all; asserting not all impli-
cates some; so, asserting some implicates some. And that is absurd.
11As I have emphasized in a study of cleft sentences undertaken with Stephen Levinson, topic and
comment enter in an essential way into Gricean reasoning (Atlas and Levinson 1981; Atlas 1981, 1988,
1989, 1991a,b, 1993a, 1996b; Horn 1981a; chapter 4 and appendix 3 in this volume).
that it would be saying more about John to say that he is taller than Brian rather than
not taller is, taken literally, rather too unsurprising. Typically negatives are less in-
formative than the corresponding affirmatives. The classical Gricean theorist begins
his discussion of the implicatures of negative statements like John is not as tall as
Brian from precisely this basic intuition.
As far as this reasoning goes, one could equally well argue that it would be say-
ing more about John to say that he is shorter than Brian rather than to say that he is
not shorter. So, following this reasoning, by failing to say that John is shorter than
Brian, the speaker of John is not as tall as Brian conveys that John is not shorter, via
Grice’s First Maxim of Quantity. Thus saying John is not as tall as Brian, according
to this reasoning, communicates John is taller than Brian, the intuitively incorrect
inferendum.
The earlier application of this reasoning yielded the implicatum John is shorter
than Brian, the intuitively correct implicatum. As far as this reasoning goes, John
is not as tall as Brian, when asserted, communicates both John is taller than Brian
and John is shorter than Brian. This absurdity should be enough to make the good
classical Gricean theorist question the use of Grice’s reasoning from the First Maxim
of Quantity in the case of negative sentences. (This absurdity was first pointed out
by Stephen Levinson; see Atlas and Levinson 1981 and chapter 3, section 1 in this
volume).
One might think that the question should not have been whether it would be saying
more about John to say that he is more F than Brian rather than to say that he is not
more F than Brian, since as we have just seen, tall and short are equally good substi-
tutions for ‘F’ in the argument. The question should have been whether it would be
saying more about John to say that he is taller than Brian rather than to say that he is
shorter than Brian. This does strike me as the more provocative question; perhaps it
is the question that the radical pragmatist intended to raise.
Now, in what sense is it to say more about John to say that he is taller than
Brian? Surely in no simple sense is there a greater quantity of information about
John in John is taller than Brian than in John is shorter than Brian. As I have re-
marked, either would be more informative than the radical pragmatist’s interpre-
tation of John is not as tall as Brian; Grice’s type of argument from his First Maxim
of Quantity will absurdly justify both. However, the issue is not just information,
it is orientation.
Linguistic evidence, largely owed to J. R. Ross, suggests asymmetries in our ways
of talking and thinking. Contrast, for example, bigger and better with ?better and
bigger, fore and aft with ?aft and fore, more or less with ?less or more, down and out
with ?out and down, That’s the long and short of it with ?That’s the short and long
of it, up and down with ?down and up. Of the same family as up and down are peak
and valley, rise and fall, over and under, high and low, above and below, ascending
and descending, raise or lower, top and bottom.
There is evidence that speakers have a general preference for stating a spatial
relation between x and y as x is above y rather y is below x, or x is taller than y rather
than y is shorter than x.
Up also occurs more frequently as an affix, as do high and top: mountaintop/
?mountainbottom; upstart/?downstart; highlight/?lowlight. Spatial terms also have
wide metaphorical uses: He was in high/low spirits, His spirits rose/fell. Happi-
ness is “up”; sadness is “down”: His income rose/fell last year, GDP is up/down
this quarter, The number of papers Harman writes keeps going up. More is
“up”; less is “down”: Things are looking up, He does high-quality work, His
work went down last year. Good is “up”; bad is “down.” (See Cooper and Ross
1975, and Lakoff and Johnson 1980, to whom I owe the data of the preceding three
paragraphs.)
The important difference between John is taller than Brian and John is shorter
than Brian is that tall is the unmarked, more frequently occurring item. When we
ask a person’s height, we ask How tall is he?, not How short is he? The sentence
John is as tall as Brian can be used to say what the classical Gricean, radical prag-
matist takes it to mean literally: that John’s height is the same as Brian’s (Sadock
1981: 268). We could not use short with that understanding; John is as short as
Brian “presupposes” Brian is short, but John is as tall as Brian does not “presup-
pose” Brian is tall.12
1.5 Post-Gricean semantics of the equative as F as

There is some temptation by logical conservatives to take John is as tall as Brian
to mean literally that John’s height is equal to or greater than Brian’s. There are a
number of difficulties with this view. First, it would seem peculiar to say that John
is as tall as John, where John is used coreferentially, means that John is identical
in height to himself or taller than himself. Similarly, this analysis seems peculiar
for A man is as good as his word, which does not, I suspect, mean that a man’s
goodness is the same or greater than his word’s; or for a bump on his head as big
as an egg, which does not mean a bump on his head whose size is the same or larger
than the size of an egg. Second, if as tall as were paraphrased by at least as tall as,
then John is as tall as, or at least as tall as, Brian would be redundant; but it isn’t.
Third, the following pairs will distinguish between as tall as and at least as tall as:
(a) x is as tall as y, if not taller/and perhaps taller and x is at least as tall as y, ?if
not taller/??and perhaps taller.
The first example x is as tall as y, if not taller/and perhaps taller offers a criticism
not only of the account of meaning of x is as tall as y (which both the radical pragma-
tist and I reject) but also of the radical pragmatist’s own view. The force of this ex-
ample was made manifest to me by the acceptability of the variant I know that x is as
tall as y, and I suspect that he is taller.13 If x is as tall as y means x’s height is the same
as y’s height, the assertion x is as tall as y, and perhaps taller would mean ?x’s height
12There are other important differences between as short as and as tall as. Contrast the paired items: x
is at most as tall as y; ??x is at most as short as y; x is at least as tall as y, {?if not / ?and perhaps}
taller; x is at least as short as y, {if not shorter / and perhaps shorter}. Consistent with my analysis of
presupposition, I think that John is as short as Brian entails Brian is short but John is as tall as Brian
does not entail Brian is tall. This is almost the only remark I shall make in this section that bears on the
vexed question, Which is the correct semantical primitive, the relational predicate x is taller than y or
the apparent one-place predicate x is tall? See subsection 1.5 and note 13 in this chapter.
13This example was suggested to me by Scott Soames as a criticism of the radical pragmatist’s view and
as evidence for the claim that x is at least as tall as y and x is as tall as y express the same proposition. I
is the same as y’s height and perhaps greater, which would imply ?x’s height is the
same as y’s height, and perhaps not the same, a much odder remark than the accept-
able x is as tall as y, and perhaps taller. Neither the logical conservative nor the radical
pragmatist can account for the semantics of the expression.
However, since as F as is a relation, we want a characterization of its logical
structure. I have already mentioned the acceptable reflexivity of John is as tall as
John. It is also obvious that transitivity holds: If John is as tall as Brian, and Brian is
as tall as Mart, John is as tall as Mart. The relation is not symmetric. (a) The
nonsymmetry is most obvious in the marked cases—in x is as short as y—which
obviously does not entail y is as short as x.14 (b) The symmetry of x is as F as y would
mean that x is as F as y entails y is as F as x; thus y is not as F as x entails x is not as
F as y. Consider the sentence Atlas is not as handsome as Richard Chamberlain. If
symmetry obtained, this would entail Richard Chamberlain is not as handsome as
Atlas. Since the assumption of symmetry logically implies that a truth entails a false-
hood, which is impossible, x is as F as y is not symmetrical.
Another feature of x is as F as y worth remarking is the way in which y is a ref-
erence point for the comparison. Compare the items in the sequence <Jay is as strong
as a mouse, Jay is as strong as a Welsh terrier, Jay is as strong as John, Jay is as
strong as an ox, Jay is as strong as Hercules>. Negative cases can be relatively un-
informative or relatively informative, depending on the reference point. On the clas-
sical Gricean, radical pragmatist’s view asserting Jay is not as tall as the tallest person
in the world would implicate Jay is shorter than the tallest person in the world, which
does not tell one very much about Jay’s height, except that he may be rather tall for
all the speaker knows or believes. Similarly, an example like Jay is not as bald as a
cue ball leaves it open that Jay may be rather bald.
To accommodate (a) the acceptability of x is as F as y and perhaps F-er, (b) the
transitivity, reflexivity, and nonsymmetry of x is as F as y, and (c) the role of y as a
reference point for the comparison, I adopt an analysis of x is as F as y different from
the radical pragmatist’s x’s F-ness is the same as y’s F-ness and from the logical
conservative’s x is at least as F as y. My post-Gricean view is the following:
Analysis. For unmarked, gradable F, x is as F as y means Whatever (measures of )

F-ness y has, x has also.
Corollary. For unmarked, gradable F, x is not as F as y means Some (measure of )

F-ness that y has is not a (measure of ) F-ness that x has.
It may be of heuristic value to give a simple model of this matter. Let the measures
in question be identified with real numbers in the closed interval [a,b]. These measures
agree with the critical point and disagree with the logical conservative’s claim that [X IS AT LEAST AS
TALL AS Y]is a reading of x is as tall as y. The schema x is as tall as y entails x is at least as tall as y, but
not conversely. The importance of the example x is as tall as y if not taller was also emphasized by
Morton White in conversation.
14On second thought, the invalidity of the contrapositive is more obvious.
are thought of as measures m of F-ness that x has (abbreviated: HAS(x,u(F(u)),m)).

Then we have the following formalizations:
(1) a. x is as F as y
b. (∀m)(HAS(y,u(F(u)),m) → (∃m')(HAS(x,u(F(u)),m') & m' = m))15
(2) a. x is not as F as y
b. ¬(∀m)(HAS(y,u(F(u)),m) → (∃m')(HAS(x,u(F(u)),m') & m' = m))
where the variable m ranges over [a,b].

As I have already said, my notion of measures of F-ness is not a notion of the extent
to which an individual x exemplifies the gradable predicate F(x). In the latter case it
is natural to identify the extent of x’s tallness with x’s height. If the extent of John’s
tallness is ej, and the extent of Brian’s tallness is eb, John’s height is the same as Brian’s
if and only if ej = eb, and John is taller than Brian if and only if ej > eb. This is not my
conception of measures of tallness. On the conception of extents, an individual exem-
plifies tall to a unique extent; on my conception of measures, an individual exhibits
many measures of tallness. On the conception of extents, the extent to which an indi-
vidual exemplifies tall is the individual’s height; on my conception of measures, the
individual’s height is the least upper bound of the measures of tallness that the indi-
vidual has: that is, x’s height = sup {m: HAS(x,u(T(u)),m)}. Of course, on my view, if
an individual has measure m of tallness, he has measure m' if 0 < m' ≤ m. It is important
to note that x has some measure m of tallness does not entail x is tall; ‘having tallness’
and ‘being tall’ are not the same concept. Confusion of the two has led to regrettable
muddle about the semantic relationship between comparative and positive adjectives.16
On this model we have the following formalizations:
(3) a. x is exactly as F as y
b. (∀m)(HAS(y,u(F(u)),m) ↔ (∃m')(HAS(x,u(F(u)),m') & m = m'))17
(4) a. x is at least as F as y18

b. (∀m)(HAS(y,u(F(u)),m) ↔ (∃m')(HAS(x,u(F(u)),m') & m≤m'))
15Note that this logical form is equivalent to the simpler (∀m)(HAS(y,u(F(u)),m) → HAS(x,u(F(u)),m)).
16For a case in point, notice the juxtaposition of sentences (1a) Sean is tall and (1b) How tall is Sean?
and the discussion in Klein (1980: 1–4).
The idea of extents is Pieter Seuren’s (1973). A related notion, degrees, is Max Cresswell’s (1976)
and Ewan Klein’s (1980: 30). I fully agree with Klein’s remark that his notion of degrees does not “play
any essential role in the interpretation of comparatives” (Klein 1980: 33). I shall criticize his proposed
analysis in terms of comparison classes only briefly here (see note 19), since (a) whatever semantic
analysis is proposed must account for the semantic relationships that I describe here, and (b) I am pri-
marily concerned with the pragmatics rather than the best logical form of comparatives. All that said,
I admit that I think that Klein’s construction of comparison classes is an awkward device.
17Note that this is equivalent to the simpler logical form (∀m)(HAS(y,u(F(u)),m) ↔ HAS(x,u(F(u)),m))
18The adverbial at least in the relation at-least-as F as should be distinguished from the sentence modifier
At least P, or P, at least.
(5) a. y is F-er than x (x is ‘F -er than y, where ‘F is the relation converse to F)

For example, y is taller than x is to be paraphrased: y is as tall as x, and y is uni-
formly taller than x, where ‘uniformly taller’ is a technical term, to be contrasted
with ‘measure-wise taller’. The converse relation predicate ‘F is, in this example,
‘shorter’. Thus the representation of y is F-er than x is that of y is as F as x, and y is
uniformly F-er than x. The logical form is:
b. (∀m)[HAS(x,u(F(u)),m) → (∃m')(HAS(y,u(F(u)),m') & m' = m)] &
(∃m)[HAS(y,u(F(u)),m) & (∀m')(HAS(x,u(F(u)),m') → m > m')].
The reason for the first clause y is as F as x in the analysis of y is F-er than x is
to ensure that even when the second technical clause y is uniformly F-er than x is
true, according to which some measure of F-ness possessed by y exceeds any mea-
sure of F-ness possessed by x, y possesses lesser measures of F-ness that are pos-
sessed by x as well—F-ness lower down, so to speak. There is an interesting question
of the range of substitutions for the relational symbol ‘F’ for which this constraint is
true. If Brian is taller than John, it seems theoretically acceptable to say that Brian
also has a measure of “tallness” that John has as well. If a chocolate cake is bigger
than a chocolate cupcake, it seems intelligible to say that the former has a “bigness”
(a size, lurking within it, so to speak) that the latter has as well. And if one mountain
is higher than another, it seems plausible to say that the former also has a “highness”
that the latter has as well. But if one airplane is higher (has greater altitude) than
another, should we say that it also has a “highness” of the lower plane as well? “High-
ness,” which is not the same as altitude, is not an ordinary, commonsense property,
so the question may be intuitively undecidable. But one’s intuitions seem to go against
the last case. These intuitions suggest why Seuren’s (1973) analysis in terms of “ex-
tents” (distances or sizes) is so plausible, even though I am not adopting it here. I am
adopting in my analysis a logically more basic notion, measures, the least upper bound
of which is an extent. (See note 8 and note 19 in this chapter.)
I have also adopted the technical notion “uniformly F-er” by contrast with “mea-
sure-wise F-er” to explicate the linguistic meaning of y is F-er than x: for example,
y is taller than x. The rationale can best be seen by considering the converse relation
predicate x is shorter than y. For example, when we judge that a rigid bar x is shorter
in length than a rigid bar y, we lay one against another with an end of one coincident
with the end of another, and we determine whether x’s other end is exceeded by y or
not—whether some part of y fails to overlap every part of x. The same procedure
holds for judging y is longer than x. As long as we wish a necessary equivalence
between y is longer than x and x is shorter than y, I see no escape from this account.
The concept logically weaker than “uniformly F-er” would be “measure-wise F-er,”
where for each measure of F-ness that x has there is some measure of F-ness that y
has, but not necessarily the same measure of y’s F-ness for all of the measures of x’s
F-ness.19
19My analysis differs from both Seuren’s (1973) and Klein’s (1980), which are, in my view, inaccu-
rate. Seuren’s formalization would be, where e ranges over extents: (∃e)(y is tall to e & – x is tall to e).
An alternate formalization can make these relationships more perspicuous. Let

‘x*F’ range over {m: HAS(x,u(F(u)),m)} and ‘y*F’ range over {m: HAS(y,u(F(u)),m)}.
Suppressing the subscripts for the sake of cleanliness and quantifying over x’s and
y’s measures of F-ness, we have the following:
(6) a. x is as F as y
b. (∀y*)(∃x*)(x* = y*)
(7) a. x is exactly as F as y
b. (∀y*)(∃x*)(x* = y*) & (∀x*)(∃y*)(y* = x*)
(8) a. x is at least as F as y
b. (∀y*)(∃x*)(y* ≤ x*)
(9) a. y is F-er than x, x is ‘F-er than y (y is as F as x & y is uniformly F-er than x)

b. (∀x*)(∃y*)(y* = x*) & (∃y*)(∀x*)(y* > x*)
In the beginning of this section, I argued that x is as tall as y was synonymous

with neither x’s height is the same as y’s height nor x is at least as tall as y, and so
not ambiguous between them. I then claimed that an analysis of x is as F as y should
reflect the role of y as point of comparison; the transitivity, reflexivity, and
nonsymmetry of x is as F as y; and the acceptability of x is as F as y and possibly
F-er/if not F-er, so x is as F as y should not entail x is not F-er than y.
It is easy to see that the formalization (∀y*)(∃x*)(x* = y*) of x is as F as y shows
it to be transitive, reflexive, and nonsymmetric. It is also obvious that the logical form
of x is as F as y does not entail the logical form of x is not F-er than y. Finally, the order
of the quantifiers in the logical form of x is as F as y indicates the role of y as a point of
comparison. The analysis that I have proposed seems to account for the linguistic data
as well or better than the alternatives, without making untenable claims of synonymy
or ambiguity. It remains to deal with the questions of logical equivalence, entailment,
and truth conditions and to examine the radical pragmatic analysis of x is almost as F
Klein’s formalization, where N ranges over Montaguvian functions taking, for example, the inter-
pretation of tall into the interpretation of very tall, is similar: VN[N{^tall}(b) & – N{^tall}(a)]. Quite
apart from the empty characterization of the range of values of N, typical of Montague grammar,
Klein’s formula merely means that among the family of comparison classes of individuals with respect
to which judgments of tallness are made, there is some comparison class relative to which b is tall,
but a is not. I cannot decide whether this claim is merely a matter of metaphysical faith, but it is
clear that in b is taller than a we are comparing b with a, and, unfortunately, that element is missing
from Klein’s formalization. The work of comparing will have to be done by comparing the heights
of individuals in Klein’s comparison classes. The formalization merely tells us that in some com-
parison class in which b is tall it is false that a is tall, not that some comparison class contains indi-
viduals taller than individuals in some other comparison class. We must add that information to Klein’s
analysis for it to be adequate. But that relation among individuals, of course, is just what we were
trying to analyze! Ironically, Klein (1980: 4) makes a related point against Cresswell (1976). For a
recent discussion see Kennedy (2001).
as y, x is not as F as y, and their implicatures in light of what we have learned about the
logical form of x is as F as y. To those questions I now turn.20
20After I had developed my analysis of x is as F as y presented here and in an earlier version in Atlas
(1984b), James McCawley brought to my attention the papers by Seuren (1973) and Klein (1980). I have
disagreed with their accounts of x is taller than y (see note 19); there is a syntactical likeness, but a fun-
damental semantic difference, in Klein’s and my formalizations of x is as tall as y. (Contrast Klein’s (1980:
38) vN[N{^tall}(y) → N{^tall}(x)] with my (∀m)(HAS(y,u(Tall(u)),m) → (∃m')(HAS(x,u(F(u)),m')
& m' = m)) and with my (∀y*)(∃x*)(x* = y*). Klein (1980: 38) incorrectly takes x is as tall as y to be
equivalent to x is at least as tall as y and, incorrectly, either takes them to be synonymous (Klein 1980:
28) or takes the latter to express one sense of the former (Klein 1980: 29). Here I wish to consider data
from Klein (1980) that bear on my analysis of x is as F as y.
It has been observed that the following intuitively valid argument is linguistically in order (Klein
1980: 28):
(A) Hussars must be at least six feet tall.
John is a Hussar.
So: John is six feet tall.
As expected, one may assert John is six feet tall, in fact he is six feet 2 inches without inconsistency.
One would say the same of:
(B) Hussars must be at least six feet tall.

John is a Hussar.
Brian is as tall as John.
So: Brian is six feet tall.
By contrast, in the following, McZ seems to be contradicting McY (Klein 1980: 29):
(C) McX: How tall is John?
McY: Six feet tall.
McZ: No, he’s six feet two inches tall.
The data in (A) and (B) suggest to some that John is six feet tall and Brian is as tall as John mean,
respectively, John is at least six feet tall and Brian is at least as tall as John, where John is not more
than six feet tall, or Brian is not taller than John, is only implicated, not entailed (e.g., Klein 1980: 38;
see note 22 in this chapter). The data in (A), (B), and (C) together suggest to some that John is six feet
tall and Brian is as tall as John are ambiguous between John is at least six feet tall and John is exactly
six feet tall, or between Brian is at least as tall as John and Brian is exactly as tall as John (e.g., Klein
1980: 29).
Since I have already argued that the expressions mean neither at least nor exactly, which are the
views of the logical conservative and the radical pragmatist, respectively, a fortiori they are not ambigu-
ous between them. On my analysis, (C) is understood differently from (A) and (B) because the question
How tall is John? asks what John’s height is. McY’s answer, Six feet tall, means that six feet is the least
upper bound of the measures of tallness that John has on our standard scale of inches and feet. Thus McZ’s
statement that the least upper bound is six feet two inches is inconsistent with McY’s statement.
In (A) Hussars must be at least six feet tall means that Hussars must be no shorter than six feet
(tall). It follows that John is at least as tall as six feet (tall), which entails that John has a six feet tall-
ness. So we properly conclude John is six feet tall, which means that John has that measure of tallness,
which of course is consistent with his having a greater measure of tallness. The point is even clearer in
(B). Brian is six feet tall is derivable from {Brian is as tall as John, John is six feet tall}—that is, from
{Whatever tallness John has, so does Brian; John has a six feet tallness}.
The analysis I have given seems to provide perfectly natural explanations of the linguistic ac-
ceptability of these arguments without appeal to dubious claims of synonymy and ambiguity (e.g., Klein
1980: 29, 38).
1.6 Comparing the neo-Gricean and my post-Gricean

views of equatives
On the radical pragmatist’s view:
a. John is as tall as Brian means John’s height is the same as Brian’s.

b. So, John is not as tall as Brian means John’s height is not the same
as Brian’s.
c. Asserting John is not as tall as Brian implicates John is shorter than
Brian.
d. Asserting John’s height is not the same as Brian’s implicates John is
shorter than Brian.
One should observe that {a,b,c} entail a theoretical absurdity. The radical pragma-
tist takes John is as tall as Brian to be synonymous with John’s height is the same as
Brian’s or John is exactly as tall as Brian. It then follows that (α) John is not as tall
as Brian is synonymous with (β) Brian is not as tall as John.21 But on the view (c),
asserting α implicates John is shorter than Brian; mutatis mutandis, asserting β im-
plicates Brian is shorter than John. Thus the allegedly synonymous sentences α and
β, when asserted, implicate incompatible statements. Since implicata are nondetach-
able from what is asserted, a difference between the generalized conversational im-
plicata implies a difference in sense in what is asserted. One cannot maintain without
absurdity that the quantity implicata of α and β are different—indeed, incompatible—
while holding that α and β are synonymous. Thus {a,b,c} lead to absurd theoretical
consequences.
It is intuitively clear that John’s height is the same as Brian’s entails John is as
tall as Brian. It also seems to me intuitively clear that John’s height is not the same
as Brian’s does not entail John is not as tall as Brian; thus it does not seem contra-
dictory to say John’s height is not the same as Brian’s, but John is as tall as Brian.
So John is as tall as Brian does not entail John’s height is the same as Brian’s. In
short, John’s height is the same as Brian’s (or John is exactly as tall as Brian) en-
tails, but is not entailed by, John is as tall as Brian. Clearly, then, John is as tall as
Brian cannot be synonymous with John’s height is the same as Brian’s; (a) and (b)
are false. On the view that I outlined in subsection 1.5, for an unmarked F in an
equative relation, x is as F as y means Whatever F-ness y has, x has also. The intui-
tive facts just noted are:
x’s height is the same as y’s
x is exactly as tall as y
⇓ ⇑
x is as tall as y
These are an immediate consequence of my analysis.
21Sentencesα and β are synonymous since α means John’s height is not the same as Brian’s and ß
means Brian’s height is not the same as John’s; those are synonymous.
It also seems intuitively clear that John’s height is not the same as Brian’s en-
tails John is not as tall as Brian or Brian is not as tall as John; this intuitive fact is
also predicted by my analysis. Given this fact, we can easily prove that John is ex-
actly as tall as Brian is logically equivalent to John is as tall as Brian and Brian is
as tall as John. This is an intuitively satisfying result.
The proof is this: since John’s height is not the same as Brian’s entails John is
not as tall as Brian or Brian is not as tall as John, it follows by contraposition (i)
that (γ) John is as tall as Brian and Brian is as tall as John entails (δ) John’s height
is the same as Brian’s. Since identity is symmetric, and since x’s height is the same
as y’s entails x is as tall as y, it follows (ii) that (δ) John’s height is the same as Brian’s
entails (γ) John is as tall as Brian and Brian is as tall as John. Thus by (i) and (ii), γ
⇒ δ ⇒ γ. So γ ⇔δ. And:
⇓ ⇑
x is as tall as y and y is as tall as x
⇓ ⇑
x is as tall as y
As far as (c) is concerned, the anomaly of ?John is not as tall as Brian and/but John
is not shorter than Brian suggests that John is not as tall as Brian entails, rather than
its assertion implicating, John is shorter than Brian. Conversely, it is intuitively clear
that John is shorter than Brian entails John is not as tall as Brian. These data are
predicted by my analysis. Thus, the radical pragmatist’s (c) is false.
I have argued that John’s height is the same as Brian’s (or John is exactly as tall
as Brian) entails, but is not entailed by, John is as tall as Brian. What moves some of
us to think that there is logical equivalence, or, stronger, synonymy, between these
sentences? The explanation is that asserting John is as tall as Brian implicates John’s
height is the same as Brian’s—that is, John is exactly as tall as Brian. This implicature
is not an inference from Grice’s First Maxim of Quantity but is an underlying in-
stance of conditional perfection (Geis and Zwicky 1971), the implicature from John
has a tallness m if Brian does to John has a tallness m if, and only if, Brian does.22 If
John and Brian have the same measures of tallness, then John’s height is the same as
Brian’s, and conversely. This implicature explains Sadock’s observation (Sadock
1981: 268) that John is as tall as Brian “could be used to mean” that John and Brian
are the same height. So far, then, we have:
⇓ ⇑
⇓ ⇑
x is as tall as y
22See example (101) in Atlas and Levinson (1981: 33) and discussion in Atlas and Levinson (1981:
44). The explanation of this implicature rests on a post-Gricean Maxim of Relativity and Principle of
Informativeness, which includes the language user’s knowledge of stereotypes (Atlas and Levinson
1981: 40–41). See chapter 3 in this volume.
and:
y is taller than x (x is shorter than y)
⇓ ⇑
x is not as tall as y
As I shall argue in section 5.7 that (d) is true, my analysis implies the relationships
exhibited in tables 5.1 and 5.2.23
My post-Gricean analysis explains the errors of the radical pragmatist and the
logical conservative. Those who take x is as tall as y to be synonymous with x is
exactly as tall as y and those who take x is as tall as y to be synonymous with x is at
least as tall as y make two errors. X is as tall as y and x is at least as tall as y are not
even logically equivalent. X is exactly as tall as y and x is as tall as y also fail to be
logically equivalent; they are interpolated by the nonequivalent x is as tall as y and
y is as tall as x.
My analysis also explains the air of semi-redundancy in asserting John is as tall
as Brian and Brian is as tall as John. The at-least-as-tall-as view of the meaning of
as tall as will not explain this semi-redundancy, since the sentence would mean John’s
height is equal to or greater than Brian’s and Brian’s height is equal to or greater
than John’s. That account of meaning imputes to the original sentence too little re-
dundancy; one clause does not even entail the other, although the two together entail
the correct truth conditions. The radical pragmatist’s view of the meaning of as tall
as will not explain this semi-redundancy either, since the sentence would then mean
TABLE 5.1. The Semantics and Pragmatics of Equatives

—» x’s height is the same as y’s
⇓ ⇑
⇓ ⇑
x is as tall as y
⇓ ⇑
y is not taller than x (x is not shorter than y)
⇓ ⇑
x is at least as tall as y
Entailment from lower to higher: ⇑, Entailment from higher to lower: ⇓,
Implicature:—»
23Sadock (1981: 269–70) claims that asserting x is not exactly as F as y does not implicate x is ‘F-er
than y; thus asserting John is not exactly as tall as Brian does not implicate John is shorter than Brian.
What Sadock seems to deny is just what I seem to hold in (d). The resolution of the apparent conflict is
that Sadock interprets not exactly as inexactly/approximately. Obviously that is not the interpretation that
I intend. For pointing out errors in the versions of tables 1 and 2 in Atlas (1984a), I am grateful
to Elizabeth Staegemann at the Second European Summer School on Language, Logic, and Information,
University of Leuven, Belgium, August 1990. I have corrected the views of Atlas (1984a) in this section.
TABLE 5.2. The Semantics and Pragmatics of

Negative Equatives
—» x is not at least as tall as y
⇓ ⇑
y is taller than x (x is shorter than y)
⇓ ⇑
x is not wholly/quite as tall as y
⇓ ⇑
x is not as tall as y
⇓ ⇑
x is not as tall as y or y is not as tall as x
⇓ ⇑
x is not exactly as tall as y
x’s height is not the same as y’s
John’s height is the same as Brian’s and Brian’s height is the same as John’s. That
account of the meaning imputes to the original sentence too much redundancy or
repetition. My view explains just the right amount of redundancy. Asserting John is
as tall as Brian implicates John’s height is the same as Brian’s. That implicatum
followed by . . . and Brian is as tall as John is redundant, since the implicatum John’s
height is the same as Brian’s means Brian’s height is the same as John’s, which entails
Brian is as tall as John. Thus the second clause “partially reinforces” the implicatum
of the first clause.
Among these views my post-Gricean pragmatic analysis is the only one com-
patible with a simple linguistic fact: x is as tall as y, x is at least as tall as y, and x
is exactly as tall as y are three distinct expressions of English whose different words
make different contributions to their meanings. It is prima facie incredible that
the meaning of one should be identical to that of another. I have tried to show the
inadequacies of any such view—those of the logical conservative and the radical
pragmatist.24
24It is possible to be a hybrid, the conservative pragmatist. Such a theorist—for instance, Horn, E. Klein,
and perhaps Levinson and Gazdar—would hold that x is as tall as y is not only logically equivalent to
but also synonymous with x is at least as tall as y. The conservative pragmatist makes the further claims
that asserting x is as tall as y implicates x is not taller than y, via Grice’s First Maxim of Quantity, and
that the occurrence of at least in x is at least as tall as y cancels that implicatum. It is realized that this
claim entails the curious consequence that nondetachability cannot be a necessary condition of con-
versational implicature, since the occurrence of at least, which cancels the implicatum, queerly de-
taches it as well (see Sadock 1978). (On my view, asserting x is as tall as y implicates x is exactly as
tall as y, which self-evidently entails x is not taller than y—a straightforward process.) His view does
not account for the implicature of John is shorter than Brian by asserting John is not exactly as tall as
Brian (see section 1.7). It does not explain the role of y as a reference point for the comparison in x is
as tall as y. Lastly, it does not satisfactorily explain the semi-redundancy of John is as tall as Brian
and Brian is as tall as John. For these reasons I am not yet convinced that the conservative pragmatist
has an adequate theory, but I shan’t expound my criticism further here. (See Horn 1981b.)
1.7 Post-Gricean pragmatic inference

The radical pragmatist takes John is not as tall as Brian to mean John’s height is not
the same as Brian’s, the assertion of the former allegedly implicating John is shorter
than Brian. But if the former only entails the latter, as I hold, the sentence John is not
as tall as Brian in question in the radical pragmatist’s argument, in section 1.4, should
be John is not exactly as tall as Brian, and the implicature (b), section 1.4, “saying
John is not as tall as Brian implicates John is shorter than Brian,” allegedly explained
(Sadock 1981: 269) by the radical pragmatist’s argument, should be instead: saying
John is not exactly as tall as Brian implicates John is shorter than Brian. John’s being
taller than Brian, rather than John’s being shorter than Brian, would be the unmarked,
preferred, stereotypical ground for asserting John is not exactly as tall as Brian.25
If John’s being taller were the speaker’s ground, it would have been uncoopera-
tive of him to have asserted the less informative, negative sentence John is not ex-
actly as tall as Brian than the sentence John is taller than Brian. From the fact that
the speaker did not assert the briefer, more informative, psychologically preferred
John is taller than Brian, the hearer infers that, because he or she is assumed to be
cooperative in the conversation, the speaker was in no position to make that asser-
tion, or at least that as far as the speaker knows or believes, John is not taller than
Brian. The speaker’s best reason for not saying John is taller than Brian would be
the speaker’s knowledge or belief that it is false.26 In light of what the speaker says,
the hearer infers that John is not taller than Brian, or at least that the speaker does not
know or believe that John is taller than Brian. The former, combined with what the
speaker says, allows the hearer to infer John is shorter than Brian, which the radical
pragmatist, unjustifiably on the principles of his own theory, takes to be implicated
by asserting John is not exactly as tall as Brian—that is, by asserting John’s height
is not the same as Brian’s.
Grice’s reasoning, from what it would have been informative to say but the
speaker did not say, does not distinguish between the two grounds for saying John is
not exactly as tall as Brian: (a) John is taller than Brian; (b) John is shorter than Brian.
However, it is crucial in the reasoning just sketched that users of the language “know,”
as part of their background presumption in the conversation, that the first ground
would be the psychologically preferred ground for the assertion. And if it were true,
there would be a clear preference for asserting it instead of the negative sentence
John is not exactly as tall as Brian.
My analysis also avoids the following absurdity. Suppose the speaker asserted John
is not exactly as short as Brian. If that meant literally that John’s height is not the same
as Brian’s, then by my previous pragmatic reasoning, and in general from the argu-
ment that shows that conversational implicata are nondetachable from what is asserted,
asserting John is not exactly as short as Brian would implicate John is shorter than
25It is now relevant to quote again the first sentence of the radical pragmatist’s argument, discussed in
section 1.4: “Without the implicature [John] is not as tall as [Brian] would say only that [John] and
[Brian] are not exactly the same height” (my emphasis).
26It is a “best” reason insofar as we are concerned with an information-exchange language game, as we
are in discussing Grice (1975a,b).

Brian. But that would be absurd. This absurdity does not arise on my post-Gricean ac-
count because John is not exactly as short as Brian does not literally mean John’s height
is not the same as Brian’s (see section 1.4 and note 12 in this chapter).
Stephen Levinson and I (Atlas and Levinson 1981) have characterized the inter-
pretation of utterances in light of cognitive biases by what we call Conventions of
Noncontroversiality—the language user’s common knowledge that “up” is better than
“down” and that tallness is “up,” shortness is “down.” These are stereotypes that we
use uncritically in the interpretation of utterances. It is essential to the explanation of
the hearer’s reasoning that it considers both the truth conditions of the statement made
and its various assertibility conditions, concerning which cognitive biases will oper-
ate. Some assertibility conditions are distinguishable from others by cultural and
psychological preferences. This should not surprise us, as it is the nature of assertibility
conditions to have such connections, unlike realistically interpreted truth conditions,
or possible states of affairs. Assertibility conditions are an essential part of what a
language-user knows when he or she knows the use of words and sentences in his or
her language.27 When we revise the Gricean account of pragmatic inference by re-
fining its maxims of conversation, supplementing it by Conventions of Noncon-
troversiality (e.g., truisms about stereotypes), and inferences from utterances to their
“best” interpretations, we avoid the absurdities that the classical radical pragmatist’s
Gricean inference sketch engendered (see Atlas and Levinson 1981).
1.8 Summary of the analysis

Conversational implicata are predictable from a Gricean Horn Scale of degree (1972/
1976) and approximative adverbials, <quite/wholly/entirely, almost, nearly, mostly/
mainly, rather, partly>, where mostly means for the most part/mainly/more than half;
almost means for the greatest ( proper) part; and quite is polysemous: (i) wholly,
(ii) rather. For gradable F, x does not F entails x does not quite F; for accomplish-
ment and achievement predicates G, x did not quite G entails x did not G. Given the
scale of adverbials, saying x almost F’s implicates x does not quite F. Saying x does
not quite F implicates x almost F’s. For accomplishment and achievement predicates
G, an analytical entailment of the implicatum of x almost G’s (namely, of x does not
quite G) is x does not G. Thus “x almost G’s” ° » x does not G, for accomplish-
ment and achievement predicates ‘G’.
For unmarked F in a comparative similarity (equative) relation expressed by
x is as F as y—for example, x is as tall as y, x is as F as y means Whatever (measures
of ) F-ness y has, x has also. The proper account of the pragmatics of degree and
approximative adverbials with comparative similarity (equative) expressions requires
a coherent inference sketch, and that in turn requires post-Gricean Conventions of
Noncontroversiality: common knowledge of stereotypes, cognitive preference for some
assertibility (not truth) conditions to others, the use of first-person-based (egocentric)
concepts of spatial orientation, and metaphorical extensions of those concepts. Those
are the contributions of the post-Gricean pragmatics that I present in this book.
27Fora discussion of the relation between knowledge of use and the lexical meaning of expressions,
see Atlas (1975a, 1978b, 1979, 1989). See also Dummett (1979).
(10) a. Horn Scale = <quite/wholly/entirely, almost, nearly, mostly/mainly, rather, partly>.

b. For gradable F, x does not F x does not quite F.
c. For gradable accomplishment and achievement predicates G, x did not quite G x
did not G.
d. For gradable F, (i) “x almost F’s” » x does not quite F and (ii) “x does not quite
F ” » x almost F’s
e. Hence, for gradable accomplishment and achievement predicates G, “x almost G’s”
° » x does not G.
f. For equatives, “x is almost as F as y” » x is not quite as F as y; “x is not quite as
F as y” » x is almost as F as y; “x is almost as F as y” ° » x is not as F as y.
2 Comparative adjectives and the semantics-

pragmatics interface
Levinson (1988b, 2000: 199–205) discusses an example of Deirdre Wilson’s (1975:

151):
(11) Driving home and drinking three beers is better than drinking three beers and driving
home.
If, in Gricean fashion, and is taken to be the truth-functional connective &, (11) would
have the form:
(12) α & β is better than β & α,
where & is a symmetric connective and α and β are nominalizations of the sentences
x drives home and x drinks three beers. If A & B is logically equivalent to B & A,
which it is, then it seems that (12) must, trivially, be false. But, intuitively, one wants
to say that (11) is true.
The Atlas and Levinson (1981) explanation of its truth would be that the use of
and Informativeness-implicates and then in statement (11) (see chapter 3 in this vol-
ume). Then the “total signification” of an utterance of (11) would be:
(13) Driving home and then drinking three beers is better than drinking three beers and then
driving home.
This proposition could be true. This phenomenon suggests that the assignment of
truth conditions that make (11) contingently true or false rather than necessarily false
depends on the implicata of the statement’s parts. This Levinson (1988b) dubs “prag-
matic intrusion” (Gazdar 1980; Carston 1988).
Of course, this example is contentious, since and might be polysemous among
[&], [& THEN], and so on, rather than univocal, as in Grice’s view. We need to find
examples of “pragmatic intrusion” that are not so contentious.
Parenthetically, it might be worth mentioning that there is a tendency to think
that the reinterpretation of and as [& THEN] is sufficient to repair the trivial falsity of
(11). This is evidently false for the statement in (14):
(14) Driving home and drinking no beers is better than drinking no beers and driving home.
This trivial falsity will not be repaired by the paraphrase in (15):
(15) Driving home and then drinking no beers is better than drinking no beers and then
driving home,
This still amounts to:
(16) Driving home is better than driving home.
Levinson (1988: 26) suggests as an example of intrusion:
(17) Eating some of the cake is better for my health than eating all of it.
Levinson alleges that this statement is, on its literal reading, necessarily false, since
that I eat all of the cake entails that I eat some of the cake, and so the allegedly less
preferable state of affairs guarantees the more preferable state of affairs. So how could
it be less preferable?
If Levinson were right, one way out of the trivial falsity of (17) would be to use
the First Maxim of Quantity implicatum not all of the cake, as in:
(18) Eating some but not all of the cake is better for my health than eating all of it.
This is an interesting example, and it certainly avoids the problems with and of ex-
ample (11). The trouble is that Levinson’s argument in favor of the “nonsensicality”
of (17) is simply not plausible. The phrase ‘better than’ creates an intensional, not an
extensional, context: that is one complication. Thus it makes perfectly good sense to
say, and it could be true so far as I know even though, necessarily, a Euclidean tri-
angle is equiangular if and only if it is equilateral:
(19) Being equiangular is better than being equilateral.
Levinson (1988b: 26, 2000: 201–2) thinks that the comparative relations are neces-
sarily irreflexive, but the cases he has in mind are the simple, extensional relations
between individuals, such as:
(20) ??John is taller than himself.
Levinson’s examples have used nominalized sentences or nominalized predicate

phrases, not names or singular terms.
The complexity of intensional contexts aside, Levinson’s argument that there
could not be a state of affairs that necessitated another state of affairs more prefer-
able than it is simply not convincing. It is perfectly obvious that:
(21) Eating some of the cake is better for my health than eating some of the cake and drink-
ing arsenic.
This statement is not false and is not “nonsensical,” yet the latter state of affairs guar-
antees the former and, on Levinson’s account, (21) should be false or “nonsensical.”
The same goes for both (22) and (23):
(22) Being short or fat is better than being short and fat.
(23) Being married to someone is better than being married to everyone.
On similar grounds, Levinson marks the sentences in (24) as anomalous, but it is

precisely the semantical nonspecificity of the predicates in the subject phrase that
permits these sentences to be acceptable, contrary to Levinson’s acceptability judg-
ments, but just as Horn’s (1984b) Division of Labor predicts. In the interpretation of
sentence (24a), vehicle tends to take on the reference of the complement of car within
the domain of vehicles by a First Maxim of Quantity implicatum [the speaker did
not say car, so he did not mean car; he meant a non-car vehicle] , and it is precisely
the generality of a word like vehicle that permits its literal understanding to be so
easily narrowed by the action of implicatural inferences:
(24) a. Having a vehicle is better than having a car.

b. Having a child is better than having a boy.
Levinson’s efforts to demonstrate the necessity of “pragmatic intrusion” from First

Maxim of Quantity implicata in order to preserve the felicity of the comparative better
than seem to me to fail, while the acceptable (24), which on his incorrect analysis of
comparatives with better than he takes to be semantically anomalous, does show the
effect of implicatural narrowing. What is not shown is that only the effect of implicata
saves the literal interpretation from trivial falsehood or a logical contradiction. This is
a major problem for Levinson’s argument, since he wishes to claim that only an impli-
cated interpretation can save such statements from an alleged semantical anomaly or
necessary falsehood of a kind that I have shown does not, in fact, exist.
A better attempt might be his cases of Atlas and Levinson (1981) informative-
ness (I)-inferenda, such as (25) (not one of Levinson’s own examples):
(25) John and Inma being married is less probable than John being married and Inma being
married.
The I-inferendum of the first phrase is John and Inma being married to each other
and statistically, for randomly chosen John and Inma, (25) is presumably true. Harnish
(1976: 328) argued that expressions like x and y are married were not ambiguous
between x and y are married to each other and x is married to someone and y is
married to someone. I am sympathetic to Harnish’s claim, but I can easily imagine
that anyone who took and to be polysemous could just as well take these expressions
to be polysemous. So I find Levinson’s confidence that these sorts of examples are
more compelling than the and cases misplaced. If the univocality claim can be sus-
tained, then, literally, (25) would be paraphrasable by (26):
(26) John being married and Inma being married is less probable than John being married
and Inma being married.
This would be a trivially false statement. So it does seems that the natural interpreta-
tion of (25) and the natural assignment of truth-value to it rely on the contribution of
the I-inferendum of a “part” to the interpretation of the “whole.”
Perhaps an example of another I-inferendum with less chance of an ambiguity
analysis would add more plausibility to Levinson’s (2000: 201) thesis; for example:
(27) Hammering the nail into the wood is safer than hammering the nail some way into the
wood.
The I-inferendum hammering the nail all the way into the wood, like “eating the apple”
conveying eating the whole apple (Atlas and Levinson 1981: 36), suggests a non-
trivial, true, inferred interpretation rather than an underdeterminate, not false, “lit-
eral” reading of (27). The semantical underdeterminacy of into the wood makes
underdeterminate what proposition a token of (27) is expressing and therefore what
truth-value the token possesses in various states of affairs. There is simply no an-
swer to the question, “How far into the wood does the nail have to be to be into the
wood?” (except the answer: a non-zero distance).
Theorists like Leech (1974) and Récanati (1989, 1993), unlike myself, tend to
treat hammering the nail into the wood as meaning hammering the nail some way
into the wood. This makes the literal meaning of (27) anomalous, having the form α
is safer than α. My analysis, that (27) is literally underdeterminate but has as a de-
fault interpretation Hammering the nail all the way into the wood is safer than ham-
mering the nail some way into the wood, which, in turn, has a Gricean implicatum,
Hammering the nail all the way into the wood is safer than hammering the nail some
but not all the way into the wood, is consistent with the linguistic judgments we make.
Levinson had hoped to show that there were cases of the form α is F-er than β
in which α is entailed by, but not necessarily logically equivalent to or synonymous
with, β, where the trivial falsity or semantical anomaly of the literal interpretation of
the statement is repaired by an implicated interpretation. Unfortunately, this hope is
unfulfilled in Levinson’s examples, since the entailment of α by β does not produce
the semantic anomaly that Levinson expects, and there is no dramatic repair for an
implicated interpretation to make. The implicated interpretation is not required to
rescue the literal interpretation from infelicity.
The upshot is that the only example free from contentious dispute about ambi-
guity or about synonymy is example (27) in Levinson (2000: 201). That example is
complex, because for the literal, underdeterminate sense of the first phrase its pos-
sible entailment by the second phrase is not well defined until a specific interpreta-
tion is given to it. But that complexity suffices to make an important point: that
although Gricean implicata are not necessary to save a statement from semantical
anomaly or logical falsehood, informativeness inferenda are sometimes required to
transform a semantically underdeterminate sense of a sentence-type into a semanti-

cally specific sentence-token to which it becomes possible to assign truth or falsity
in light of the facts.
Levinson thought that “pragmatic intrusion” meant that the truth condition of
the whole is a function of the truth-conditions of the implicated parts. That is not
only what it means. It more importantly means that whether the whole has truth con-
ditions at all depends on whether the informativeness inferenda add their content to
the meaning of the whole: the thesis is that the truth-evaluable meaning of the whole
is a function of the meanings of the inferred or implicated parts. Ironically, the fail-
ure of Levinson’s examples to make a satisfactory case for the modest neo-Gricean
thesis about the contribution of Gricean implicata to truth conditions (Katz 1972;
Walker 1975) shows the importance of the latter post-Gricean thesis about the con-
tributions of the contents of Informativeness inferenda to the literal meaning of the
sentence (Atlas 1979).28
3 Philosophical consequences: Semantic atomism,

holism, and corporatism
The semanticists who think that almost VP analytically entails not VP typically do
so because they view the meaning of ‘almost’ as “containing” the meaning of ‘not’,
in the familiar metaphor in which the meaning of ‘bachelor’ “contains” the meaning
of ‘male’. This commits such theorists to an analytic/synthetic distinction for expres-
sions, but I agree with the view of Putnam (1983b), and later Fodor and LePore (1992),
that what Quine (1980a) showed in “Two Dogmas of Empiricism” and White (1950)
showed in “The Analytic/Synthetic: An Untenable Dualism” was that a semantical
distinction between sentences cannot be reconstructed from a (nonexistent) episte-
mological distinction between a priori and a posteriori knowledge. Quine’s arguments
do not show that there is no usable distinction between the analytic and synthetic to
be drawn in linguistic theory—a point that Jerry Katz (1990: 184–94) has insisted
on for many years.
If there were an analytic entailment from almost to not, which there actually is
not, then the explanation of speaker’s intuitions about the inferences they make from
almost sentences to not sentences would appeal to linguistic knowledge of that ana-
lytical entailment from almost to not without appealing to facts about all inferences
that speakers of English could make; the explanation is semantically non-holistic.
Thus an analytical entailment from almost to not would make the linguistic world
safe for semantical non-holism: atomism (Fodor and LePore 1992: 32–35) or molecu-
larism (Dummett 1981: 599–600; Fodor and LePore 1992: 22–23).
It is an interesting sociological fact that at a time during which almost all phi-
losophers of language and mind were meaning holists of one stripe or another, lexi-
28The thesis of “pragmatic intrusion,” in the defensible “sense” version just described, was first articu-
lated in Atlas (1979). Variants of it now exist in the views of the London School influenced by Deirdre
Wilson’s Relevance Theory (see Sperber and Wilson 1986b; Kempson 1986, 1988a; Carston 1988;
Blakemore 1992; and Récanati 1989, 1993). See also Atlas (1979, 1989: 62–64).
cal and formal semanticists of natural language remained committed to an analytic/

synthetic distinction and so to meaning non-holism. As Fodor and LePore (1992: 32)
asked in 1992, “Why is almost everyone a meaning holist?” The analysis of almost
inferences that I have given in this chapter, and gave in a slightly different form in
Atlas (1984a), permit a reformulation of the question to “Why is anyone a meaning
holist?” An explanatorily adequate linguistic theory must reject meaning holism as
an empirically unsatisfactory position.
What are the commitments of the kind of explanation that I have offered here
for inferences among adverbials? First, I have relied on analytical entailments
among adverbial predicate and sentence expressions in a language; for example,
John is not wholly as tall as Brian analytically entails John is not as tall as Brian,
and upon lexical relations underlying the analytical entailment from John is taller
than Brian to Brian is shorter than John.29 Second, I have relied on implicatures
among assertions of sentence expressions in a language. The mechanism of im-
plicature requires that there be lexical Horn and Levinson Scales of lexical items
from the language, like the adverbial scale discussed in this chapter, in order for
there to be adequate explanations of utterance-interpretation. That is, a speaker of
English must know that almost belongs to such a scale, and hence that almost bears
a special semantic relation, not to every word in the language but to an identifiable
finite set of adverbial expressions whose semantical relations to each other in the
scale are known to competent speakers. And every language will possess such sets
of kinds of expressions.
The actual linguistic relationship between almost VP and not VP, for a certain
class of VPs, shows that it is not required that every inferential relationship among
sentences containing occurrences of almost VP and not VP be representable by a
componential analysis or a meaning postulate of the kind found in idealized dic-
tionaries or in the philosophical writings of semantic atomists, whether mentalis-
tic or behavioristic. It shows that an explanatorily efficacious metalinguistic concept
of a lexical scale, if it is to be psychologically realistic, imputes to a speaker the
linguistic knowledge that an expression has the property of belonging to a natural
linguistic kind—of being an element of a Horn or Levinson Scale. But it shows
that that property may be enough to explain some inferences that on the face of
them seem to be analytical entailments, such as almost VP to not VP. Thus we have
replaced a purely atomistic account of the meaning of utterance-types, one in which
[NOT] is a component of [ALMOST] and so justifies an analytic conditional If . . . al-
most . . . then . . . not . . . by a more “corporatist” (White 1981: 17–23, 1986) ac-
count that is not yet holistic.
Besides the non-holism implied by the explanatory use of analytical entailments,
it is clear that some semantical meta-properties, like belonging to a particular scale,
are an-atomic in Fodor and LePore’s (1992: 1) sense: a property is an-atomic just in
case if anything has it, at least one other thing does. (Non-an-atomic properties are
atomic properties, by definition.) But the implied competence in the language is “cor-
29Chomsky (1996c: 52) has recently remarked that “Among the intrinsic semantic relations that seem
well-established on empirical grounds are analytic connections among expressions.”
poratist” not holistic, since the speaker must grasp the semantic relations—the en-
tailment relations—among almost, quite, not, and nearly within a finite set of lexemes
that belong to a scale. But in order to do this, the speaker does not need to grasp
entailment relations holding between all expressions in the language. Nor does the
competent speaker need to grasp every linguistic kind or every “projectible”
(Goodman 1983) semantical predicate that relates meaningful words, phrases, or sen-
tences in the language to understand one lingustic kind or one “projectible” semantical
predicate; his or her understanding of the meaning of the semantical metalanguage
is not holistic. And if the speaker’s understanding of the metalanguage is not holis-
tic, and the metalanguage contains a translation of the object language, it would be
bizarre to think that an understanding of the predicates of the metalanguage would
be non-holistic but an understanding of the relata of those predicates (the other ex-
pressions of English) must be holistic.
Pragmatic inference is not directly represented in the meanings of the terms,
neither as a result of derivational rules operating on the formal representations of the
meanings of the words, as in formal logical implications, nor as a result of derivations
from meaning postulates (even though generative semanticists and discourse repre-
sentation theorists and discourse semanticists have tried to put it there). Implicatural
and inferendal inference does require that a speaker understand the more global prop-
erty that a term possesses by virtue of belonging to a lexical scale, a natural linguis-
tic kind, and thus its relations to the other lexical items of the scale. Hence, it is clearly
possible—since it is actual—that inferences that might have been solely explained
by analytical entailments, and lexically based, might be better explained by lexical
scales and inferenda and implicata combined with semantic and analytical entailments.
Conversely, a language in which an inference is a composition of inferendal or
implicatural relations with an entailment relation could possibly become a language
in which an inference is solely explained by an analytical entailment, through con-
ventionalization, lexicalization, and grammaticalization (Horn 1984b, 1989: 260–
62; Hopper and Traugott 1993). If the data to be explained are intuitive inferences
from one utterance-type to another utterance-type, there are systematic reasons for
preferring an entailment analysis or preferring a generalized inferendal or implicatural
analysis, depending on which best explains the range of linguistic data (e.g., the lat-
ter in the case of almost).
If one takes the phenomena of generalized conversational inferenda and impli-
cata seriously, one cannot escape an explanation that makes essential use of lexical
meaning, and hence a semantical distinction between the analytic and synthetic, and
essential use of an-atomic semantic properties, like belonging to a particular Horn or
Levinson Scale of lexical items. Meaning holism is inadequate if, as has been argued
(Fodor and LePore 1992), it is inconsistent with a principled distinction between the
analytic and synthetic, which is required by the explanatory use of analytical entail-
ment in linguistic explanation; semantical atomism is inadequate if it is inconsistent
with the essential use in linguistic explanation of an-atomic semantic properties, which
are required by linguistic explanation of generalized conversational inferenda and
implicata. Linguistic explanation requires the philosophically unexciting middle
ground—semantic molecularism or semantic corporatism. Thus ordinary science
deflates the views of the exciting philosophical extremes, as Quine (1960: 15–16)
and Rorty (1979: 319), where philosophical battles are typically fought: on the mar-
gins of science. If philosophy of science is philosophy enough, as Quine once re-
marked, philosophy of linguistics is philosophy of language enough. Neither atomism
alone nor holism alone has any explanatory value for empirical linguistic theory. They
are both mere philosophical conceits.
6
The Third Linguistic Turn and

the Inscrutability of Literal Sense
1 The problem of numerical adjectives
Just at the time that the classical 1967 Gricean program received its canonical form
and application in linguistics—in the radical pragmatics of Jerrold Sadock’s (1978)
“On Testing for Conversational Implicature” and his (1981) elegant essay “Almost”—
the neo- and post-Gricean programs began. One began in London with a lecture by
Atlas (1978a), as well as in three essays: Atlas (1978b, 1979) and Atlas and Levinson
(1981). The latter—“It-Clefts, Informativeness, and Logical Form: An Introduction
to Radically Radical Pragmatics (Revised Standard Version)”—appeared in the same
volume in which Sadock’s “Almost” essay appeared. That was followed by Atlas
(1984a), “Comparative Adjectives and Adverbials of Degree: An Introduction to
Radically Radical Pragmatics”; Horn (1984b), “Toward a New Taxonomy for Prag-
matic Inference: Q-Based and R-Based Implicature”; Levinson (1987a), “Minimi-
zation and Conversational Inference”; Levinson (1987b), “Pragmatics and the
Grammar of Anaphora”; Horn (1989), A Natural History of Negation; Levinson
(1991), “Pragmatic Reduction of the Binding Conditions Revisited”; Horn (1993),
“Economy and Redundancy in a Dualistic Model of Natural Language”; Huang
(1994), The Syntax and Pragmatics of Anaphora; and Levinson (2000) Presumptive
Meanings: The Theory of Generalized Conversational Implicatures. (Another depar-
ture from the classical Gricean program also began in London with the 1981 essay
of Deirdre Wilson and Dan Sperber, “On Grice’s Theory of Conversation,” leading
eventually to the Relevance Theory of Sperber and Wilson 1986b.)
185
There is another irony in that just as Sadock formulated the problematic for the
classical Gricean approach in his lecture “Whither Radical Pragmatics?” at the
Georgetown University Round Table on Languages and Linguistics in 1984, Horn
offered at the Round Table his “Toward a New Taxonomy for Pragmatic Inference:
Q-Based and R-Based Implicature,” which reformulated Atlas and Levinson (1981)
and gave one compelling answer to the question in Sadock’s title.
In his lecture Sadock discussed two types of pragmatic views, the Gricean and
the discourse-functional. In what follows I discuss the Gricean example of what
he called ‘excessive pragmatism’, other examples that he listed being my (1977b),
Kempson and Cormack (1981), and his own Sadock (1975). Although I do not
think that my (1977b) essay was excessive, and I defended the view more fully in
Atlas (1989), I do think that some Gricean analyses have been excessive in their
reduction of semantic properties to Gricean properties—Sadock’s (1981) own analy-
sis of almost F and Horn’s (1992b, 1996b) analysis of Only Muriel voted for Hubert,
for example. But it is instructive to consider Sadock’s example of “excessive
pragmatism.”
Saddock (1984: 142) presents a case “in the tradition of Grice where it is clear,”
he claims, “that a pragmatic explanation can be taken so far as to undermine its own
structural foundations.” According to Horn (1972, 1976) and Gazdar (1979a) nu-
merical expressions like ‘two’, though alternately understood to mean [AT LEAST 2]
or [EXACTLY 2], were not ambiguous. On their view a statement like Tom has two lovers
literally means Tom has at least two lovers, but its assertion implicates by a general-
ized conversational implicature Tom has at most two lovers. Thus what is literally
asserted is a lower-bound on the number of Tom’s lovers, and what is implicated is
an upper-bound on the number of Tom’s lovers. Hence what is communicated is that
Tom has exactly two lovers. Thus, Sadock observes, the statement would be false if
Tom has 1 lover but only misleading in some contexts if Tom has 3 lovers. (I distin-
guish the numerical adjective ‘three’ from the arithmetic numeral ‘3’.) But Sadock
imagines a more radically pragmatic analysis of the meaning of numerical expres-
sions derived from the phenomenon of Horn Scale reversal in which a statement entails
an upper bound, for example, Tom ran the 100 in ten seconds, meaning [TOM RAN THE
100 IN AT MOST 10 SECONDS], not a lower bound as in the previous example. Then
Sadock suggests that scale reversal might show that ‘ten’ expresses neither a lower
nor an upper bound. Then ‘ten’ would apparently mean merely [SOME NATURAL NUM-
BER], or [SOME POSITIVE RATIONAL NUMBER], etc.
There is a simpler argument against the radical pragmatics analysis other than
the one from scale-reversal phenomena, which are clearly matters of common
encyclopaedic knowledge and which raise White’s and Quine’s questions of the dis-
tinction between lexical knowledge and factual knowledge (White 1950; Quine 1980c;
Putnam 1983b). On the classical Gricean view to assert (a) Luckily after the car crash
Tom had two hands is (implausibly in my view) to say literally that luckily after the
car crash Tom had at least two hands and to implicate that luckily after the car crash
he had no more than two hands (as if car crashes were in the causal business of add-
ing hands to persons unfortunate enough to experience them, except in certain lucky
cases), and to assert (b) After the car crash Tom had two fingers is (plausibly) to say
that after the car crash Tom had at least two fingers and to implicate that he had no
THE THIRD TURN AND LITERAL SENSE 187
more than two fingers. Sadock then asks, in effect, what does ‘two’ mean, if it does
not mean [AT LEAST 2] in (a) but does mean [AT LEAST 2] in (b)?
An extreme, pragmatic answer, Sadock suggests, would be merely that ‘two’
means nothing except whatever justifies the entailment of substituends with lower-
scale numerical expressions—for example, A(two) A(one), and whatever fails to
justify the same relation of A(two) to A(three). On this extreme view ‘two’ itself
possesses no meaning that expresses a lower-bound [AT LEAST 2] or an upper-bound
[NO MORE THAN 2]. Sadock remarks:
On this theory, the natural number words we have in English would mean hardly
anything at all.1 Such a result might offend our sense of propriety, but what is re-
ally wrong with it is that it cannot work. There is simply not enough conventional
content left in the number words for a pragmatic theory to use as input.2 On a theory
that gives the natural number words a semantic lower bound ([AT LEAST n]), the
precise use of these words ([EXACTLY n], e.g. Two is the positive square root of four)
. . . can be modeled as a contextual setting of an indeterminate upper bound equal
to a semantic lower bound (as in Horn’s account). But if there is no precise upper
or lower bound, it is not at all clear how we could ever account for the precise use.”
(Sadock 1984: 143)
I shall argue, as I did in Atlas (1983, 1990), that English sentences containing
three are semantically nonspecific among [AT LEAST 3], [EXACTLY 3], and [AT MOST 3]
interpretations. This account demystifies the classical Gricean claim that three meant
[AT LEAST THREE]. (One would have wondered why at least three did not mean [AT
LEAST AT LEAST . . . THREE].) My claim is that three means none of [AT LEAST 3], [EX-
ACTLY 3], [AT MOST 3]. (I distinguish the natural numerals ‘0’, ‘1’, ‘2’, ‘3’, . . . from
the English numericals ‘zero’, ‘one’, ‘two’, ‘three’, . . . .) The puzzle for the classi-
cal Gricean theorist was how ‘1 + 2 = 3’ could mean anything mathematically pre-
cise, since he interpreted its sense as that of the English sentence ‘One plus two is
three’, and then analyzed the latter as [ONE OR MORE PLUS TWO OR MORE IS THREE OR
MORE]. He recognized, of course, that this was no way to do arithmetic.
My elementary answer was to point out a fallacy of equivocation. The cardinal
numeral ‘3’ is not synonymous with the English numerical ‘three’. Since ‘3’ is a
1Atlas (1983, 1990) agrees with Sadock’s (1984) premise that ‘two’ means neither [AT LEAST 2] nor [NO
MORE THAN 2] but regards his conclusion that ‘two’ has hardly any meaning at all as a non sequitur and
suggests how we can account for the precise use. The premises of the pure pragmatic position do have
the consequence that ‘two’ cannot mean [EXACTLY 2], and I agree with that conclusion; it is clear that
if two Ns] does not entail that the cardinality of the set of Ns is equal to or greater than 2 (two Ns
|{x: Nx}| ≥ 2) and does not entail that the cardinality of the set of Ns is equal to or less than 2 (two Ns
|{x: Nx} | ≤ 2), it does not follow that ‘two’ has no meaning. It merely follows that the meaning of
two Ns does not entail that the cardinality of the set of Ns is 2 (two Ns | {x: Nx} | = 2), i.e., that
two Ns does not mean [EXACTLY 2 Ns]. But it does not follow from the claim that ‘two’ means none
of [2 OR MORE], [EXACTLY 2], and [2 OR FEWER] that ‘two’ is meaningless; or that ‘two’ means [SOME NATURAL
NUMBER ].
2Sadock thinks this because, as suggested earlier for ‘ten’ and ‘two’, the conclusion appears to be that
‘ten’, or ‘two’, merely means [SOME NATURAL NUMBER], a fallacious conclusion, as I have already ex-
plained (see the previous note).
canonical name for the natural number 3, the integer that in Dedekind-Peano arith-
metic is the successor of the successor of the successor of 0, or s(s(s(0))), and ‘three’
is a numerical English determiner (or, perhaps it is not a determiner but a numerical
adjective), there should have been less temptation than there was to conflate the two
(not, notice, ‘the 2’).
On the classical Gricean account of “what is said” in numerical statements, a pe-
culiarity arises that conversational inference must resolve. If one asserts Tom has one
lover, one has not claimed something literally inconsistent with what one asserts in the
statement Tom has two lovers. Similarly, if one asserts Tom has three lovers, one has
not asserted something literally inconsistent with what is asserted in the statement Tom
has two lovers; in fact, on this view, the former entails the latter. On the classical Gricean
account, for sentence frames A( ) like the above, A(num) is consistent with A(num'),
the strong-to-weak scale of numerical expressions being <num', num, one>, just as φ
∨ (ψ ∨ χ) is consistent with (ψ ∨ χ). Obviously, on this view A(num') entails A(num),
since each numerical expression takes its [AT LEAST] interpretation rather than its [EX-
ACTLY] interpretation. On the classical Gricean view, the [EXACTLY] interpretations arise
from the implicatures of asserting a statement that literally means Tom has at least . . .
lovers, the assertion of which implicates [TOM HAS NO MORE THAN . . . LOVERS] by a scalar
implicature appealing to Grice’s First Maxim of Quantity (“Say as much as is required
in the context”). (On scalar implicatures, see Horn 1972, 1976, 1989; Gazdar 1979a;
Hirschberg 1985.)
Until 1984 Sadock and others were willing to live with this peculiarity, but when
Sadock (1984: 143) concluded that the classical Gricean theorist cannot even justify
the [AT LEAST] interpretation as the literal meaning of the numerical expressions, he—
like the pre-1905 Bertrand Russell when confronted with Meinong’s acceptance of
the being of the round square—thought things had gone too far. Sadock retreated to
the view that ‘three’ in ‘Three is the square root of nine’ (actually, three is the posi-
tive square root of nine) has its [EXACTLY] meaning. Sadock observed that now one
requires a new pragmatic principle of “loose talk” to derive [3 OR MORE] from three,
“conveying less than his words imply, rather than more,” although how this prin-
ciple is to work, and why it does not convey [3 OR FEWER] is left unexplained.
But is Sadock’s argument a genuinely principled objection to the classical Gricean
view? The classical theory could hold, and this was Horn’s (1972) position, that when
context changes the direction of a numerical expression scale by reversing the nu-
merical value of what counts as literally or metaphorically “high” and what counts
as “low” on the scale, the literal sense [AT LEAST . . . ] of a numerical phrase still en-
tails a lower bound [NO LESS THAN . . . ] on the reverse scale. To use Sadock’s (1984:
143) example of a change in direction of a scale: to break 100 in golf is to have a
score less than 100, while to break 100 in bowling is to have a score greater than
100. Sadock (1981: 260) also illustrated the change of direction of a scale by the two
examples It’s almost freezing, It’s almost melting, putting 0°C as the low point in the
former case and as the high point in the latter case. Sadock (1984: 143) admits that
a classical Gricean like Horn (1972, 1976) is not required to abandon the lower bound
analysis of the meanings of the numerical expressions, but Sadock claims, as I have
indicated, that “an utterly pragmatic” interpretation of scalar implicata would do so.
On such an extreme pragmatic view even the lower bound [AT LEAST] is not seman-
tic. So if asserting A(three) only implicates A(at least 3) and only implicates A(no
more than 3), what is its literal meaning? Sadock (1984) opts for identifying the
meaning of three with the [EXACTLY] interpretation.
Consider the following parody of Sadock’s reductio argument of the purely prag-
matic analysis of ‘three’, this version of Sadock’s argument purporting to show that
the determiner ‘the’ has hardly any meaning. The determiner ‘the’ occurs in the col-
lective term ‘the boys’, which is disguised syntactically as a definite plural count
noun phrase in ‘The boys lifted the piano’, meaning [THE BOYS TOGETHER LIFTED THE
PIANO], and not to be mistaken for the plural distributive term ‘the boys’ in ‘The boys
played the piano’, meaning [EACH OF THE BOYS PLAYED THE PIANO]. On Sadock’s ac-
count, the extreme pragmatist should have confronted these facts: that for some (col-
lective) instances of the term the Ns, The Ns VP1 Ns VP1 and The Ns VP1 Any
of Ns VP1: For example, The boys lifted the piano Boys lifted the piano and The
boys lifted the piano Any of the boys lifted the piano, and for some (distributive)
instances of the term the Ns, The Ns VP2 Any of the Ns VP2 and The Ns VP2 Ns
VP2 . He would conclude that the definite NP the N means either [Ns BUT NOT ANY OF
THEM] or [Ns AND ANY OF THEM], and so ‘the’ allegedly means hardly anything at all,
just as on the extreme pragmatist’s view two Ns means either [2 OR MORE Ns] or
[2 OR FEWER Ns], and so ‘two’ means hardly anything at all. This is surely a reductio
ad absurdum of Sadock’s form of argument. How could ‘the’ mean nothing at all,
and how could that form of argument be correct?3 Even an “excessive pragmatist”
could not represent his position by adopting the characterization of his argumenta-
tion that Sadock suggests.
In “Almost” Sadock (1981: 259) had suggested in a classical Gricean fashion
that almost P meant [P IS POSSIBLE IN A WORLD NOT VERY DIFFERENT FROM THE ACTUAL
WORLD], but Atlas (1984a) criticized this account; its posited meaning is too weak to
explain the entailments and implicatures of almost P and it gets the truth-conditions
wrong (see chapter 5, section 1.1). The neo-Gricean theorist (Atlas 1984a) raises for
Sadock’s (1981) classical Gricean analysis of almost precisely the question that
Sadock (1984) had raised for the classical Gricean analysis of the English numerical
expressions. In “Whither Radical Pragmatics?” Sadock (1984) hypothesizes that ‘two’
should literally mean [EXACTLY 2] (a similar view is taken by Carston 1985); the in-
terpretation [2 OR MORE] would supposedly arise from a pragmatic principle of “loose
speaking,” as in J. L. Austin’s example France is hexagonal, meaning [FRANCE’S SHAPE
IS HEXAGONALISH]. In his essay Sadock does not propose a mechanism for imple-
menting the principle of loose talk, however. Thus one may raise for Sadock’s hy-
pothesis just the question that he raised for the classical Gricean theorist: How does
the principle of loose talk and the allegedly literal, exact meaning [EXACTLY 2] of ‘two’
explain the inexact interpretation [AT LEAST 2]? If Tom has 2 or more fingers is
inferrable by mysterious means via Tom has 2-ish fingers from Tom has two fingers,
i.e. from Tom has exactly 2 fingers, asserting the latter would also implicate via the
3Note that the conclusion is not that ‘the’ means nothing at all independently of a sentence-frame in
which it occurs; the argument purports to hold as well for syncategorematic ‘the’ that in the frame in
which it occurs it still means nothing. Thus the argument purports to hold for ‘the’ as an incomplete
symbol in Bertrand Russell’s sense.
same principle of loose talk that Tom has 2 or fewer fingers equally well. So Sadock’s
analysis of numerical expressions by appeal to a principle of loose talk does not,
without further theoretical refinement, explain the data.
Sadock’s objection was to a possible, extreme pragmatics hypothesis that nu-
merical expressions have no meaning of their own except the minimum required to
justify the following: A(two) A(three), but A(two) A(one); and the contrasting
account that he offers is just as open to his own objections as is the account of the
classical Gricean theorist. So I do not see the justification for Frederick Newmeyer’s
(1986: 179) conclusion in his Linguistic Theory in America that “this idea of ‘radical
pragmatics’ . . . has been successfully challenged (see Sadock 1984).” Sadock was
being too pessimistic in the implicit answer he gave to the question in his title “Whither
Radical Pragmatics?” Nowhere. That is the answer Frederick Newmeyer (1986) takes
him to have not only suggested but demonstrated. The argument that allegedly dem-
onstrates it is fallacious. So, as Mark Twain would have remarked, Newmeyer’s (1986)
report of the demise of classical Gricean pragmatics was an exaggeration, although
the classical Gricean version of linguistic pragmatics was being supplanted by a neo-
Gricean pragmatic theory at the very time 1984–86 that the classical view’s obituary
was too hastily written by Sadock (1984) and Newmeyer (1986). Even though Sadock
(1984) did not successfully challenge radical pragmatics, he was correct to conclude
that if the goal of radical pragmatics was “a detailed and correct account of the intri-
cate symbiosis that characterizes the association between structure and function in
natural language, it is an important and exciting linguistic enterprise that deserves to
flourish” (1984: 147) This was precisely the end in view in Atlas (1975b, 1977b),
which Sadock (1984: 141, 147 n.1) claimed, mistakenly I believe, was “excessive
pragmatism,” as well as the goal explicitly of the neo-Gricean Atlas and Levinson
(1981) and Atlas (1979, 1984a, 1989).
2 The meanings of numerical adjective

noun phrases
In sentences like (1) and (2), from Carston (1985, 1988), the normal interpretations
would be ‘at least 17’, ‘at least 18’, ‘at least 3’, and so for (3), it would seem, ‘at least
4’ as in (4).4
(1) In Britain you have to be seventeen to drive a motorbike and eighteen to drive a car.
(2) Mary needs three A’s to get into Oxford.
(3) Mrs. Smith has four children if not five.
4This section contains revisions of part of the third Lecture in the series “Implicature and Logical Form:
The Semantics-Pragmatics Interface” that I delivered in the Second European Summer School in Lan-
guage, Logic, and Information, in the Katholieke Universiteit Leuven, Belgium, 6–10 August 1990. It
also contains part of my lecture “Atlas Meets Seuren on Numerical Adjectives” delivered in the Work-
shop in Honor of Pieter Seuren at the Max Planck Institute for Psycho-linguistics in Nijmegen, The
Netherlands, 2 July 1999, as well as part of Atlas (1989).
(4) Mrs. Smith has at least 4 children if not at least 5.
Suppose we follow Gazdar (1979a) and Grice (1975a, 1989a) in the analysis of the
conditional as a material, truth-functional conditional. Then (3) is logically equiva-
lent to:
(5) Mrs. Smith has five children or Mrs. Smith has four children.
On the Gricean (Horn 1972; Levinson 1983, 2000) ‘at least’ interpretation of the
numerical adjectives, this would be:
(6) Mrs. Smith has at least 5 children or Mrs. Smith has at least 4 children.
This, in turn, on the Gricean account of ‘or’ as the inclusive disjunction, would be
logically equivalent to:
(7) Mrs. Smith has at least 4 children.
In turn, (7) on the Gricean account would be synonymous with (8):
(8) Mrs. Smith has four children.
There are reasons for being unhappy with the claim that Mrs. Smith has four chil-
dren if not five should be logically equivalent to Mrs. Smith has at least 4 children,
or to Mrs. Smith has four children, much less the claim that the semantic representa-
tions, or meanings, of those sentences are the same. But my interest here is the ques-
tion of interpretation.
The alternative ‘exactly’ interpretation of (3) would be (9), which on the Gricean
view of the conditional would be logically equivalent to (10):
(9) Mrs. Smith has exactly 4 children if not exactly 5.
(10) Mrs. Smith has exactly 5 or Mrs. Smith has exactly 4 children.
At first sight this interpretation does seem pretty strange for (3), and one sees why,
by contrast, the ‘at least’ interpretation is plausible. But let’s consider the following
context: there is a special government assistance plan for families with exactly 4 or
5 children. Two clerks, A and B, are discussing Mrs. Smith’s case:
A: Does Mrs. Smith qualify?
B: Mrs. Smith has four children, if not five.
In this context I think that it would be uncooperative for B to have said, “Mrs. Smith
has four children if not more.”5 From what B does say, A takes B to have implicated
5Stephen Levinson disagrees with this intuition, but what is relevant to the assistance plan is 4 or 5
children, and B has been less than relevant.
that Mrs. Smith does not have more than 5 children. So A interprets B to mean “Mrs.
Smith has exactly 4 or Mrs. Smith has exactly 5 children” and not to mean “Mrs.
Smith has at least 4 children.” As far as interpretations supporting the claim that ‘four’
literally means ‘at least 4’ go, why doesn’t this evidence, and similar evidence, pro-
vide just as good support for the claim that ‘four’ literally means ‘exactly 4’? What
privileges the data in support of ‘at least 4’?
In his discussion of Modified Occam’s Razor, Senses are not to be multiplied
beyond necessity, Grice writes:
One should not suppose what a speaker would mean when he used a word in a cer-
tain range of cases to count as a special sense of the word, if it should be predict-
able . . . that he would use the word (or the sentence containing it) with just that
meaning. If one makes the further assumption that it is more generally feasible to
strengthen one’s meaning by achieving a superimposed implicature than to make a
relaxed use of an expression (and I don’t know how this assumption would be jus-
tified), the Modified Occam’s Razor would bring in its train the principle that one
should suppose a word to have a less restrictive rather than a more restrictive mean-
ing, where choice is possible. (Grice 1989c: 47)
As I (Atlas 1979: 269) have discussed elsewhere, the “strengthening” assumption

can be justified by discovering that there is an intelligible inference that brings about
the strengthening of a speaker’s meaning—intelligible in the sense that such infer-
ences can be formulated and rationalized—but no intelligible inference that brings
about the relaxation of a speaker’s meaning. Grice’s observation pushes theory in
the direction of assigning ‘at least 4’ as the literal meaning of ‘four’ and explaining
the ‘exactly’ interpretation by appeal to Grice’s Conversational Maxims.
Larry Horn (1972: 38) claimed that statement (11) was logically consistent:
(11) I have 3 children: in fact I have more.
Thus in (11) the noun phrase ‘3 children’ cannot mean ‘exactly 3 children’. He then
claimed that statement (12a) is “normally understood as negating the at least” inter-
pretation except when the numeral is contrastively stressed, as in (12b):
(12) a. John does not have $175.

b. John won not £100 but £1,000 in the lottery,
In (12b) the ‘exactly’ interpretation is possible (Horn 1972: 39). Given such evidence,
Horn offered a classical Gricean hypothesis: that ‘3 children’ literally means ‘at least
3 children’ in sentences like (11), but that in other contexts there may be a Gricean,
First Maxim of Quantity scalar conversational implicature to ‘not 4 children’, ‘not 5
children’, . . . , i.e. to no more than 3 children. In that case the utterance-meaning of
the statement would be ‘exactly 3 children’. For example:
(13) A: How many children does Jack have?

B: Jack has two children.
In this question-answer sequence, the answer (13B) allegedly literally means Jack
has at least 2 children, but the context of the sequence’s question (13A) How many
children does Jack have? encourages the First Maxim of Quantity Scalar Implicature.
‘Two children’ in (13B) conveys ‘no more than 2 children’, an inference that Horn
explains by an entailment ordering, by virtue of the scale of numerical words < . . . ,
three, two, one >, and Grice’s First Maxim of Quantity : “Say as much as is required
for purposes of the exchange.” Speaker B did not say ‘three children’, ‘four chil-
dren’, . . . , so A may infer that B’s saying ‘three children’, . . . , would be saying
more than is needed for purposes of the exchange. The best explanation of B’s not
saying ‘three children’, not saying ‘four children’, . . . , is that B believes that Jack
has no more than 2 children, the propositional content of that posited belief being
the conversational implicatum of asserting a sentence containing ‘two children’.
By contrast Seuren (1993: 228) claims that in the question-answer sequence B’s
answer in the possible situation in which Jack has more than 2 children would be a
false answer. If, as the Griceans hold, B’s answer literally meant Jack has at least
2 children, the answer in the same situation would be true, though, since B’s state-
ment would conversationally implicate Jack has no more than 2 children, B’s an-
swer, though literally true, would be misleading and uncooperative to assert in the
situation in which Jack has more than 2 children.
Seuren also claims that there is no logical inconsistency in statement (14):
(14) Jack does not have two children: he has three.
The consistency implies that there is a possible situation in which the statement is
true. The truth of (14) entails that ‘two children’ in the negative clause Jack does not
have two children of (14) could mean ‘exactly two children’ or could mean ‘at most
two children’.6
In his theory of semantic syntax, Seuren would analyze the answer (13B) Jack has
two children as The number of Jack’s children is 2, and the question (13A) How many
children does Jack have? as The number of Jack’s children is what? For the topic of the
question How many children does Jack have? is the number of Jack’s children, and the
answer Jack has two children is assertorically equivalent to the comment of a statement
with contrastive stress (15a), with topicalization (15b), or of the English cleft (15c):
(15) a. Jack has TWO children.

b. As for the number of Jack’s children, it’s 2.
c. It’s two children that Jack has.
Seuren distinguishes the truth conditions of statement (15a) Jack has TWO children
from the truth conditions of a statement of the canonical sentence in (16).
(16) Jack has two children.

6The negation expressed by ‘not’ in the statement is claimed to be descriptive, not “metalinguistic” in
Horn’s (1985) sense. Kempson (1986: 80) agrees with this second observation of Seuren’s and argues
that such statements as (14) can be descriptive, not “metalinguistic,” negations. I discuss “metalinguistic”
negation later in this chapter.
Briefly, for Seuren (1993) the truth conditions of the canonical statement (16) are
Jack has at least 2 children; the truth conditions of the contrastively stressed state-
ment (15a) Jack has TWO children and of the answer (13B) Jack has two children
are Jack has exactly 2 children.
Happily, here is where I may express agreement with Seuren. The topic/com-
ment analysis of the statement in the question-answer sequence in (13) seems to me
accurate. “Contrary to commonly held views,” as Seuren notes, “cleft constructions
or constructions with contrastive (focus or comment) accent may differ truth-condi-
tionally from non-cleft, non-contrastive (i.e. canonical) constructions” (1993: 229).
My agreement with Seuren’s uncommon view is found in Atlas and Levinson (1981:
16–18, 50–55) and Atlas (1991a), where I argued that the canonical (17a), the stressed
(17b), the cleft statement (17c), and the only Proper Name statement (17d) should
have the logical forms (and truth conditions) in (18 a,b,c,d):
(17) a. Sam wants Fido.

b. SAM wants Fido.
c. It’s Sam who wants Fido.
d. Only Sam wants Fido.
(18) a. λx(W(x, f ))(s)

Sam wants Fido.
b. λy(W(s,y))( f )
Fido is wanted by Sam.
c. λy(y = ΓzSz)(ΓxW(x, f ))
The group of individuals who want Fido is identical to the group that Sam-izes (i.e.,
to Sam; Quine 1960).
c′. ∃xW(x, f ) & ∀y(W(y, f ) → y = s)
Someone wants Fido and Sam does if anyone does.
d. ∃x∀y[(x = y ↔ W(y, f )) & (W(y, f ) → y = s)]
Exactly one individual, and Sam if anyone, wants Fido.
Here (18c), on the distributive reading, is logically equivalent to (18c'). I shall not
defend again the logical analysis that I have given for cleft statements (Atlas and
Levinson 1981) and for only Sam statements (Atlas 1996b), but I shall use these re-
sults to evaluate Seuren’s claims for his linguistic intuitions. (For a discussion of the
semantics and pragmatics of cleft statements, see appendix 3.)
Applying the Atlas and Levinson (1981) logical analysis to the contrastively
stressed statement (15a) Jack has TWO children, or to the cleft statement (15c) It’s
two children that Jack has, the logical paraphrase is The group of Jack’s children is
a group of size 2, given in (19b):
(19) a. It’s two children that Jack has.

b. λu(Size-2(u))(Γx(C(x,j))
The group of Jack’s children is a group of size 2.
c. Jack has exactly 2 children,
Here a group is of size 2 if and only if the cardinality of the corresponding set whose
elements are the members of the group is 2. Or, more simply put, the paraphrase of
(19a) It’s two children that Jack has is (19c) Jack has exactly 2 children.
It follows from Atlas and Levinson’s (1981) logical analysis of clefts that the
truth conditions of It’s two children that Jack has are not the same as the truth con-
ditions of Jack has two children. The cleft entails the canonical statement, but the
converse is false; the canonical statement does not entail the cleft. In fact, from my
(1991a) analysis of only Proper Name statements, there are the entailment rela-
tions in (20):
(20) Only John hit Brian It was John who hit Brian John hit Brian
Here Only John hit Brian is logically equivalent to It was John who hit Brian, which
entails but is not entailed by John hit Brian. (Horn 1981a, 1996b denies that It was
John who hit Brian entails Only John hit Brian, and he denies that Only John hit
Brian entails John hit Brian; in Horn (2002) he finally accepts the latter entailment.)
Thus Atlas and Levinson’s (1981) logical analysis of clefts provides a defense of
Seuren’s (1993) intuitive, linguistic claims: the truth conditions of Jack has TWO
children, the answer to How many children does Jack have? are distinct from the
truth conditions of the canonical statement Jack has two children. For Seuren the
truth conditions of the canonical statement Jack has two children are just as the Clas-
sical Griceans supposed: Jack has at least 2 children. Thus Seuren (1993: 232) finds
the following statement logically consistent:
(21) Jack has two children, and they live in Kentucky, and he has three children, and they
live in Texas.
By contrast, Seuren (1993: 233–34) believes that in statement (22a) ‘they’ is a “spe-
cifically referring term” only if Jack has exactly 2 children, and ‘they’ makes refer-
ence to them:
(22) a. Jack has two children: they live with their mother.
b. Jack has two children: each of them lives with his mother.
So Seuren (1993: 234) believes that one does not need to appeal to Grice’s (1975a,b,
1989a) Maxims of Conversation to produce the interpretation exactly 2 children of
two children in (22a). The Theory of Discourse Anaphoric Reference takes over the
explanatory task. But on Seuren’s view it remains unclear why the ‘exactly 2’ inter-
pretation of (22a) does not also occur in the second clause of (21), producing an in-
consistency in (21) if the referring character of ‘they’ requires the ‘exactly’
interpretation. I should have thought Seuren’s requirement is too strong: Jack has
two children is not synonymous with, nor does it entail, Jack has only two children,
as Seuren himself had noted in (15) and (16). If we give up Seuren’s requirement,
(21) has a consistent interpretation in which the first ‘they’ refers to a subgroup of
two, and the second ‘they’ refers to a subgroup of three children. Notice that sen-
tence (21) does not entail that John has at least 5 children: the two children living in
Kentucky could also live in Texas; the sentence does not say Jack has two children,
and they only live in Kentucky. Thus the ‘exactly 2’ interpretation can be at best a
pragmatic inference from ‘two children’ in an assertion of (21) and a fortiori in an
assertion of (22a).
Whether the Theory of Discourse Semantics will provide an account of anaphoric
reference that makes a Gricean account unnecessary is a large issue that is tangential
to my concerns in this book. I will merely mention that in light of Dowty’s (1980)
and Reinhart’s (1983) suggestions, Levinson (1987b, 1991, 2000: 287) has provided
the beginnings of a Gricean explanation of two of Chomsky’s (1982, 1986) Binding
Conditions that unlike the syntactical accounts predicts the preferred interpretation
of John told her that he gave her a valentine, and Huang (1994) argues that a Gricean
account is necessary for the explanation of anaphora in Chinese (see chapter 3).
An example of Robyn Carston’s (1985, 1988: 174–75) presents a prima facie
difficulty for the classical Gricean view, as well as for Seuren’s claims for the truth
conditions of the canonical statement forms. She alleges that statement (23) has an
‘exactly’ interpretation under which it is true and an ‘at least’ interpretation under
which it is false in the same circumstances. (I shall note later in this chapter that the
sentence cannot have the latter interpretation.)
(23) If there are three books by Chomsky in the shop, I’ll buy them all.
Carston (1988: 171–73) criticizes the classical Gricean view by arguing that an al-
leged implicatum that might alter the truth conditions of what she calls “the propo-
sition expressed” cannot be a genuine implicatum but instead must be part of the
“explicit content”—an “explicature”; classical Gricean implicata, she notes, do not
affect truth conditions of “the proposition expressed.” So ‘three books’ cannot liter-
ally mean ‘at least 3 books’ but implicate ‘at most 3 books’.
Carston (1988: 155) identifies “the proposition explicitly expressed”—Sperber and
Wilson’s (1986b: 72–73) “propositional form of the utterance”—with Grice’s “what
is said.” Unfortunately, this is a misreading of Grice. Grice’s “what is said” is, roughly,
the disambiguated literal sense of the words taken with reference-, tense-, and deixis-
fixed values, the content of an Austinian (1962/1975) illocutionary act—and what would
one expect from a regular participant in J. L. Austin’s Saturday mornings?
Grice gave no attention to the specification of semantically nonspecific ex-
pressions, what Bach (1994a) terms a “completion” of a semantically nonspecific,
nonpropositional expression. Pace the sanguine remarks of Bach (1994a: 141) that
Grice (1978: 116, 119) also anticipated Bach’s notion of the conceptual “expansion”
of a proposition expressed by a sentence used non-literally to communicate a propo-
sition that would be expressed by an expansion of the contents of the thought liter-
ally expressed by the utterance, Bach’s notion of a conceptual expansion is also not
anticipated in Grice (1978). (For expansion, see White 1965: 59 and Atlas 1989: 25.)
For example, the thought expressed by You won’t die expands to the proposition YOU
WON’T DIE FROM THE CUT ON YOUR FINGER. (Bach 1994a differs from Récanati 1989 in
his account of “what is said”—Récanati identifying “what is said” with the expanded
proposition, Bach identifying “what is said” with the proposition literally expressed
by the utterance.)
“What is said” is distinguished by Grice from what he calls “the total significa-
tion of the utterance.” The addition of the content of Grice’s conventional implicata,
his generalized and particularized conversational implicata, and, at an early stage of
his thinking, presuppositions to “what is said” transforms it into Grice’s “total signi-
fication” of the utterance. What is needed by Carston, Récanati, and others here is a
distinction between what is indirectly suggested, implied, or conveyed—(a) what is
“implicitated” (in Bach’s 1994a sense) and (b) what is implicated (in Grice’s sense)—
and what is directly asserted. Unfortunately, Grice himself did not restrict his de-
scriptions of assertions to ‘suggest’, ‘imply’, or ‘convey’ when describing what a
speaker does by, in, or when uttering a sentence. He used his favorite word ‘means’
as well: what the speaker meant by, in, or when uttering a sentence. Grice simply
did not have a precise notion of the content of an “assertion” and explicitly avoided
constructing one. (Bach 1994a creates one for him, a Carnapian notion of a “struc-
tured proposition.” For more on the problem of “what is said,” see Levinson 2000:
170–98.)
Of course, Grice realized that his minimal sense of “what is said,” as in the case
of statements containing logical connectives, did not match an intuitive notion of
“what the speaker meant” in uttering a sentence. His comments on or and if . . . then
make that clear, but his comments on his conventional implicata of words like but
make the point even more clearly. In the case of conventional implicata, “what is
said” is not even the whole of the explicit lexical content of the sentence uttered.
Grice merely claimed that implicata, other than manner implicata, should be partly
determined by “what is said,” and he was clear that synonymous statements “say the
same thing”; these claims were sufficient to justify nondetachability of conversational
implicata. Carston’s (1988: 155) notion of a proposition explicitly expressed approxi-
mates Grice’s notion of what a speaker meant, not his notion of “what is said.”
In Atlas (1983, 1990) I observed that in sentences like (23) with numerical
modifiers of plural count nouns, the plural N books was actually a collective term, as
evidenced by the phrase them all. I distinguished (23) from (24):
(23) If there are three books by Chomsky in the shop, I’ll buy them all.
(24) If there are three books by Chomsky in the shop, I’ll buy each of them.
My thesis was that asserting three N scalar implicates no more than 3 only if N is a
distributive term, not a collective term. In the case of the collective term, the numerical
adjective expresses the size of a group designated by the collective term, which en-
tails the ‘exactly’ interpretation. Seuren’s (1993: 233–34) observation that they makes
“specific reference” (Seuren 1985: 459–63) to a group of Jack’s children in (22a)
Jack has two children: they live with their mother is in my view a consequence of
they being an anaphor for a collective term. One may contrast with the interpretation
of (22a) the distributive interpretation of two children in (22b) Jack has two chil-
dren: each of them lives with his mother; the default interpretation of (22b) is not
Seuren’s exactly 2 children. (Carston’s alleged ‘at least 3’ interpretation of Carston’s
example (23) is actually semantically anomalous. The sentence cannot have that in-
terpretation, though my sentence (24) may.)
To make matters even more complex, Horn (1978c) pointed out that an asser-
tion of (25), where ‘sides’ is a plural, distributive term, possesses the no more than
3 implicatum that is absent from assertions of (26), where ‘three’ is lexically incor-
porated and does not mean at least 3:
(25) x is a figure with three sides.
(26) x is a three-sided figure.
These two observations suggested that the specification of the interpretation of the
noun as collective or distributive was a crucial feature in determining the interpreta-
tion of ‘three’. Or, to paraphrase Frege, only in the context of a noun phrase does a
numerical modifier have a meaning.
This then suggested (Atlas 1983, 1990), compatibly with the account of the se-
mantically nonspecific meaning of ‘not’ (Atlas 1974, etc.), that the English word
‘three’ should be distinguished from the Arabic numeral ‘3’ (the name of a natural
number) and that English ‘three’, like ‘not’, is semantically nonspecific: semanti-
cally unspecified for the ‘at least’ and ‘exactly’ interpretations just as ‘not’ is seman-
tically unspecified for the exclusion and choice negation interpretations. For example,
the open sentence (27) containing the numeral ‘3’ is not synonymous with (25) x is
a figure with three sides, nor is (25) synonymous with (28):
(27) x is a figure with 3 sides
(28) x is a figure with at least three sides.
Furthermore, the arithmetic statement (29) is not redundant in the way (28) would
be on the classical Gricean reading of ‘three’ as ‘at least 3’ given in (30):
(29) x is a figure with at least 3 sides
(30) x is a figure with at least [at least 3] sides.
These semantic observations indicate a clear difference between ‘three’ and ‘3’.
Unlike Carston’s Relevance Theoretic objections to Grice’s notion of “what is
said,” my objection was to the internal incoherence in the classical Gricean’s view.
The Gricean must argue that in assertions of (29) x is a figure with at least 3 sides,
the asserted at least 3 cancels, because it is in minimal opposition to, the default,
generalized conversational implicatum at most 3 sides of asserting three sides that is
typically inferable from the allegedly synonymous (25) x is a figure with three sides.
Since the Griceans (e.g., Levinson 1983, 2000; Horn 1972) suppose three sides to be
synonymous with at least 3 sides, the synonymy claim implies that the implicatum
no more than 3 sides is, bizarrely, detachable from assertions of (25) x is a figure
with three sides. Since generalized conversational implicata are not detachable from
synonymous assertions (except by features of the manner of the assertion), these
synonymy claims suggested an internal incoherence in the Gricean theory. It is for
these reasons—recall the analogous argument against the Gricean view of equatives
in Atlas (1984a) and in chapter 5 in this volume—that I rejected the classical Gricean
analysis of the meaning of three N et al.
As discussed in chapter 5, a problem analogous to that of the semantic relation
between three and at least 3 arose for the classical Gricean analysis of the equative
x is as tall as y. That analysis would claim that x is as tall as y is not only logically
equivalent to but also synonymous with x is at least the same height as y. It makes
the further claims that asserting x is as tall as y implicates x is not taller than y, via
Grice’s First Maxim of Quantity, and that the occurrence of at least in x is at least
the same height as y cancels that implicatum. It is realized that the latter claim en-
tails the curious consequence that nondetachability would then not be a necessary
condition of conversational implicature, since the occurrence of at least, which can-
cels the implicatum, would, oddly, on the classsical Gricean view that x is as tall as
y is synonymous with x is at least the same height as y, also detach the conversa-
tional implicatum x is not taller than y from “what is said” in x is as tall as y. My
(1984a) view of as tall as, discussed in the previous chapter, will escape this anomaly
of unexplained detachment, since I reject the synonymy assumption. On my view,
the assertion x is as tall as y Informativeness-implicates x is exactly as tall as y, which
entails x is not taller than y and y is not taller than x, and logically entails but is not
entailed by or synonymous with x is at least as tall as y.
Synonymy is also central to a problem that Sadock raised about detachability:
Why should it not be possible to find two expressions that differ just in that one
contains the denial of what the other conversationally implicates? It has been
claimed, for example in Horn 1973, that my saying I ate some of the cake conver-
sationally implicates that I did not eat all of the cake. Now the doctrine of non-
detachability claims that if this is a bona fide case of conversational implicature,
then there should not be a lexical item that means just what some does but which,
when substituted for some in the sentence above, detaches the implicature [from
the utterance]. This happens to be true for English, but the cancelability doctrine
would seem to indicate that it does not need to be true. Why could there not be a
word that happened to mean ‘some and perhaps all’? . . . What principle . . . pre-
vents a lexical item in a natural language from having a partially redundant lit-
eral sense? (Sadock 1978: 290)
As a case in point, Sadock considers or and and/or:
In English we have a form and/or. Logically speaking, this has just the same mean-
ing as or, but, because of a redundancy it contains, it has the property of canceling
the implicature from P ∨ Q to ¬ (P & Q). . . . But adopting [these Gricean analy-
ses] involves giving up the claim that non-detachability is a necessary characteris-
tic of conversational implicature. (Sadock 1978: 291)
The weakness in Sadock’s argument is the assumption that Grice’s doctrine of

nondetachability should apply to statements resulting from the substitution of logi-
cally equivalent, rather than synonymous, expressions. The expressions or and and/
or are logically equivalent if or designates inclusive disjunction ∨ and and designates
conjunction &: (P ∨ Q) ((P & Q) ∨ (P ∨ Q), but their logical equivalence does
not entail their synonymy or “sameness of meaning.”
Grice remarks:
The implicature is non-detachable insofar as it is not possible to find another way

of saying the same thing (or approximately the same thing) which simply lacks the
implicature. The implicature which attaches to the word try exhibits this feature.
One would normally implicate that there was a failure, or some chance of failure, if
one said A tried to do x; this implicature would also be carried if one said A at-
tempted to do x, A endeavoured to do x, or A set himself to do x. (Grice 1978: 115,
1989c: 43)
Then Grice himself observes a complication, due to the Maxim of Manner, of

nondetachability being a necessary condition on conversational implicature, as
follows:
This feature is not a necessary condition of the presence of a conversational

implicature, partly because, as stated, it does not appear if the implicature depends
on the manner in which what is said has been said, and it is also subject to the limi-
tation that there may be no alternative way of saying what is said, or no way other
than one which will introduce peculiarities of manner, such as by being artificial or
long-winded. (Grice 1978: 115, 1989c: 43)
Let us grant that nondetachability must take into account the matter of manner. This
hardly seems a theoretical, though possibly a practical, difficulty in applying the
maxims. The alleged limitation, that there may be no alternative way of saying what
is said, does not strike me as a major difficulty, either, since I am prepared to allow
that when there is in the language no alternative way of saying what is said, then the
condition of nondetachability is vacuously satisfied. Sadock (1978: 289) mentions
vacuous satisfaction of nondetachability as a problem for nondetachability being a
sufficient condition for the existence of a conversational implicature, since every-
thing conveyed by an expression is, when no paraphrase is available, trivially non-
detachable. But Grice is worrying about nondetachability being a necessary condition
for the existence of conversational implicature, and this seems to me to be an empty
worry. In the case of Sadock’s example, the necessity of nondetachability is trivial
and theoretically harmless. Since from Grice’s (1989b : 25) characterization of “what
is said,” “what is said” is not invariant under substitution of logically equivalent
expressions, Sadock’s putative counterexample—substituting and/or for or—does
not preserve the sameness of “what is said,” and thus is no counterexample to
nondetachability being a necessary condition of conversational implicature. Further-
more, the meaning of or is not the same as the meaning of and/or even if, “logically
speaking” (Sadock 1978: 291), or is logically equivalent to and/or. Linguistically
speaking, or is not synonymous with and/or.
Since x is as tall as y is not synonymous with x is at least the same height as y,
the apparent paradox of simultaneous cancelation and detachability of x is not taller
than y from x is as tall as y by the supposedly synonymous x is at least as tall as y
does not show that nondetachability is not a necessary condition on conversational
implicata arising from maxims other than manner. My analysis (Atlas 1984a, chap. 5)
of comparative adjectives and equatives applied now to numerical noun phrases shows
that a sentence A( . . . three . . . ), if the frame A( ) contains a collective term
(e.g., Three children moved the piano) or a lexically incorporating term (e.g. x is a
three-sided figure), is not synonymous with A( . . . at least 3 . . . ), and a speaker
asserting it does not implicate A(. . . at most 3 . . . ) , and so fails to generate an
interpretation A(. . . exactly 3 . . . ) by Grice’s First Maxim of Quantity. Indepen-
dently Carston (1985, 1988) came to the same conclusions.
Like Kempson (1986: 82–83) I also rejected the view that ‘three’ meant literally
‘exactly 3’, on the grounds that John hit three balls is not synonymous with John hit
only three balls any more than John hit Brian is synonymous with John hit only Brian
(Atlas 1991a, 1993, 1996b) and on the grounds that no pragmatic principle weaken-
ing exactly 3 to at least 3 was coherent (see Kempson 1986: 82–83).
If three N does not literally mean at least 3 N and does not literally mean
exactly 3 N, I concluded that it must be a semantically nonspecific noun phrase.
Unlike Carston (1985, 1988: 174), who also concluded that three N was unam-
biguous and meant none of at least three N, at most three N, and exactly three
N, I (Atlas 1990) offered arguments for the semantical nonspecificity of three N,
employing the ambiguity tests of Zwicky and Sadock (1975) that I had discussed in
Atlas (1974, 1977b, 1989).
3 Ambiguity Tests
I had employed these tests to refute the received view that negative definite descrip-
tion sentences, as (31), are ambiguous:
(31) The king of France is not wise.
Such sentences were alleged to have two senses and two Russellian logical forms,
one the narrow-scope predicate negation (32a), usually represented in English by
(32b), and the other the wide-scope sentence negation (33a) usually represented in
English by (33b):
(32) a. ∃x[(Kx &∀y(Ky → y = x)) & ¬Wx ]

b. The king of France is non-wise.
(33) a. ¬∃x[(Kx &∀y(Ky → y = x)) & Wx ]

b. It is not true that the king of France is wise.
{ }
the case
Since Russell’s formal language is bivalent, it does not matter whether the negation
in (32a) and (33a) is choice negation – or exclusion negation ¬ . (Recall that
the exclusion negation ¬A of a sentence A is true if and only if A is not true, for
every admissible valuation of the language; the choice negation – A is true [false]
if and only if A is false [true], for every admissible valuation of the language. In a
bivalent language, exclusion and choice negation are extensionally identical func-
tions. In a non-bivalent language—one with truth-value gaps—they are distinct.)
Russell (1919: 179) and Whitehead and Russell (1927: 69) thought of the ambi-
guity as a scope ambiguity, described by Russell in terms of the singular term ‘the
king of France’ having “primary occurrence” in (32b) and “secondary occurrence”
in (33b). Russell would have said that, if true, the two readings are true for different
reasons. When (32b) is true, it is because a unique, extant king of France fails to be
wise. When (33b) is true, it is either because there is no king of France or because
there are more than one or because a unique, extant king fails to be wise; Russell
(1905) omits to consider the case, only relevant thirty-five years later, where there
might be no France. Typically it is observed that the narrow-scope predicate nega-
tion represented by (32b) entails the wide-scope sentence negation represented by
(33b), but not conversely, that the logical form (32a) represented by (32b) entails
that there exists a king and that he is unique, but the logical form (33a) represented
by (33b) does not entail either. Since the affirmative statement The king of France is
wise entails the existence and uniqueness of a French king, it is usually said that both
the affirmative statement and the negative sentence (31) on its narrow-scope reading
represented by (32b) presuppose the existence and uniqueness of a French king. It is
said that the wide-scope reading of the negative sentence (31), allegedly represented
by (33b), does not have this presupposition.
There are two questions that seem never to have been raised explicitly prior to
Allwood (1972, 1977), Atlas (1974, 1975a,b), and Kempson (1975). First, do the
English sentences (33b), which supposedly represent univocally one reading of (31),
each have two understandings instead, and, further, understandings that allegedly
constitute two readings, the wide-scope one (33b) allegedly represents and the narrow-
scope one that (32b) allegedly represents? In short, as Allwood and Atlas asked, are
(31) and (33b) paraphrases? And then, second, as Atlas and Kempson asked, do the
intuitive differences in understandings in (31), and possibly (33b), constitute a genuine
ambiguity, or is the difference a matter of sense nonspecificity with respect to choice-
and exclusion-negation interpretations in both (31) and (33b)?
I shall first simply report a fact about my own, and I believe, ordinary American
English speakers’ idiolects. Just as my linguistic intuitions detect presuppositional
and nonpresuppositional understandings of (31), they detect the very same under-
standings in (33b)! In these respects (33b) and (31) behave like paraphrases in my
idiolect. The standard philosophical view that (33b) expresses only one understand-
ing is more a logician’s prejudice than an empirical linguistic judgment. (For agree-
ment with this paraphrase claim of Atlas 1974, see Kuroda 1977 and Grice 1981.
See also Boër and Lycan 1976: 48–52 and Horn 1978b, 1985, 1989.)
The question that is now before us is whether the sentences of (31) and of (33b)
are each ambiguous between a nonpresuppositional understanding, which on the stan-
dard view would be “the” reading of (33b), and a “factive” or presuppositional under-
standing, which on the standard view would be “the” reading of (32b). The answer is
that they are not ambiguous. The sentences are semantically nonspecific between the
presuppositional and nonpresuppositional understandings.
In defense of my answer, I shall appeal to tests for ambiguity and sense non-
specificity discussed by Zwicky and Sadock (1975). One of the difficulties of the
example is that the presuppositional understanding entails the nonpresupppositional

understanding of (31) and of (33b). This is a semantical property of what Zwicky
and Sadock call “privative opposites.” For example, ‘dog’ meaning MALE CANINE
and ‘dog2’ meaning CANINE are privative opposites with respect to the semantic
feature [MALE]. The same is true for ‘mother’ and ‘parent’ with respect to [FEMALE].
One of Zwicky and Sadock’s tests is for ambiguity of privative opposites. If the
expression is truly ambiguous, it ought to be possible to assert the general case
and deny the specific case without contradiction. ‘Dog’ is ambiguous, so it is pos-
sible to say without inconsistency That’s a dog, but it isn’t a dog or That dog isn’t
a dog, meaning THAT CANINE IS NOT A MALE CANINE. If such cases were inconsis-
tent, it would indicate that the expression was not ambiguous. In our cases of ne-
gation, we have
(34)
{ ?The king of France is not wise
?It’s not { true / the case } that the king of France is wise,}
but {he / the king of France } is not non-wise.
The second conjunct is the denial of the presuppositional understanding, which on

the standard view would be the denial of the reading represented by (32). Since I
find the sentences semantically anomalous, we have our first indication that neither
(31) nor (33b) is ambiguous.
Sadock (personal communication) has suggested that these sentences might not
sound so bad to some ears because lexical negations are usually “poseurs”; ‘unhappy’
differs from ‘not happy’. There is a tendency to read ‘non-wise’ as ‘definitely stupid’.
Sadock suggests, as a clearer example, that the following sentence is anomalous:
(35) ?The king of France is not wise and it’s not even the case that the king of France is wise.
But it ought not be anomalous if the negative sentence were ambiguous and took its
narrow-scope-of-negation reading in the first clause.
Another ambiguity test is by “semantic differentia.” When a sentence has rela-
tively similar understandings, as I suggest (31) does, or (33b) does, but these differ
only by one’s being sense-specified and the other sense nonspecific for some par-
ticular semantic feature, such as [FACT], the feature must be such that the lexicons of
natural languages can plausibily fail to use it. A lexical feature for the age of the
referent of a third-person personal pronoun might be, unlike semantic gender, an
example of a possible lexical feature that English need not adopt. Certainly in En-
glish there is no formal marking in a negative sentence for a presuppositional, or
nonpresuppositional, understanding of the sentence. (Of course, this observation is
also supported by my earlier claim that the form of (33b) does not specify the
nonpresuppositional understanding of the negative sentence.)
If the difference in understanding were very great, it would point to ambiguity
rather than sense nonspecificity. But, in contrast with the two understandings of ‘They
saw her duck’, the difference in the understandings of (31), or of (33b), is not very
great. To see explicity that this is so, recall Russell’s (1905) argument from “On
Denoting.” Suppose we could enumerate all the individuals who are wise. We look
down the list, and we do not find the king of France. Of course the reason we do not
may be either that he is extant but fails to be wise (the standard account of the truth
of (32)) or that he does not exist, part of the standard account of the truth of (33b).
But the similarity in the two cases is significant; we do not find the king of France on
the list. This similarity is more compelling than the difference. It is at least suffi-
ciently compelling to shift the burden of linguistic proof of the ambiguity of (31), or,
on my view, of (33b), on to those parties who claim it.
This argument also confronts the claim of polysemy: ambiguity among closely
related meanings. A defender of the ambiguity of (31) and possibly of (33b) might
well want to argue that exclusion negation and choice negation are closely related,
distinct meanings. I would argue that paradigm cases of polysemy are constituted
by meanings in ways quite different from the way choice negation and exclu-
sion negation are logically and semantically related. Consider, for example, the
polysemous ‘hit’ in ‘John hit Brian and the highway divider’ but the nonpolysemous
‘knocked’ in ‘Enraged, Brian knocked over his wife and also an innocent bystander’.
Choice and exclusion negations are, in Zwicky and Sadock’s (1975) sense, priva-
tive opposites. The meanings of polysemous expressions are typically not priva-
tive opposites, ‘dog’ and the like being unusual cases. Nevertheless, this does not
mean that ‘not’ might not be a case of polysemy like ‘dog’, so I shall consider further
ambiguity tests.
Of the two syntactic tests that I shall mention from Zwicky and Sadock (1975),
I shall not employ here the now anachronistic but still revealing test of transforma-
tional potential. (It will be easy for readers of Zwicky and Sadock to see that the results
confirm that (31), or (33b), is not ambiguous.) But I shall consider Ross’s, Chomsky’s,
and Lakoff’s conjunction test.
The test is an identity test, about which I wrote in chapter 1, section 2.1, in my
discussion of the depth nonspecificity of the Necker cube drawing. Following Chomsky
(1957: 35–36) and Grinder and Postal (1971: 269), Zwicky and Sadock illustrate the
way conjunction reduction requires identity of sense (not identity of reference):
If:
(59) Morton tossed down his lunch
were unspecified (rather than ambiguous) as to whether Morton bolted his lunch or
threw it to the ground, then the parallel example:
(60) Oliver tossed down his lunch
would also be unspecified, and the reduced sentence:
(61) Morton and Oliver tossed down their lunches
would have four understandings, not two [as in the case of ambiguity, which per-
mits only the two parallel readings: (i) Morton and Oliver bolted their lunches
(ii) Morton and Oliver threw their lunches to the ground], because the identity con-
dition on conjunction reduction cannot require identity of elements that are not part
of syntactic structure. But (61) lacks the crossed understandings (except as a joke),
and we conclude that (59) is ambiguous. (Zwicky and Sadock 1975: 18)
Thus, Zwicky and Sadock (1975) adopt criterion (A):

(A) The impossibility of a crossed, literal paraphrase for a conjunction-reduced

sentence S entails the ambiguity of S. (The distinct, parallel paraphrases express dis-
tinct senses.)
In another example Zwicky and Sadock appeal to criterion (B) to demonstrate

nonambiguity (semantical nonspecificity):
(B) The possibility of a crossed, literal paraphrase for a conjunction-reduced sen-

tence S entails the non-ambiguity of S. (The distinct, parallel paraphrases do not
express distinct senses.)
Zwicky and Sadock write:
We can now return to examples

(6) Melvin became as tall as any of his cousins
(7) Melvin became taller than the average Ohioan
(8) Melvin became the tallest linguist in America
and show that they exhibit no ambiguity with respect to whether Melvin or his cir-
cumstances change. The reduced sentences:
(87) Melvin became as tall as any of his cousins, and then the same thing hap-
pened to Martin.
(88) Melvin became taller than the average Ohioan, and then the same thing hap-
pened to Mervyn.
(89) Melvin became the tallest linguist in America, and the next year the same
thing happened to Merton.
all permit the crossed understandings. (Zwicky and Sadock 1975: 22–23)
(There are a number of subtleties in Zwicky and Sadock’s account of ambiguity tests
that I discuss at length in Atlas 1989: 25–65 but shall not discuss here.)
Now consider the pro-form reduced sentences:
(36)
{The
It is not { true / the case } that the king of France is wise}
king of France is not wise
and the same thing goes for the queen of England.
These are structurally related to:
(37)
{It’s not { true / the case } that the king of France is wise}
The king of France is not wise
and
{it’sthe notqueen{ trueof England

/ the case } that the queen of England is wise.}
is not wise
If the negations in the clauses were ambiguous between different “scope” readings,
we would have four possible senses of the conjunction: factive/factive, factive/
nonfactive, nonfactive/factive, and nonfactive/nonfactive. On the ambiguity hypoth-

esis the pro-form, conjunction-reduced sentence in (36) would eliminate the second
and third, crossed understandings as possible literal interpretations of the sentence.
Ambiguity permits only the parallel readings.
My idiolect accepts the crossed understanding nonfactive/factive in (38) as a
literal interpretation of (36):
(38)
{ItTheis notking{true/ the case} that the king of France is wise}
of France is not wise
(since France is not a monarchy), and the same thing goes for the queen of England
(who overindulges her children).
If crossed understandings are acceptable as literal interpretations of the conjunction-

reduced sentence, and if the meanings of the reduced and unreduced sentences are
related by identity-of-sense constraints—as they are, then the constituents have the
same, univocal, nonspecific sense, even if utterances of the constituents do not have
the same understanding. This means that the sentence is not ambiguous.
The same conclusions hold for adverbials other than ‘not’; in these cases the
presupposition is on the verb phrase:
(39) a. Jane didn’t walk slowly (she walked rapidly) and the same goes for Max (he stood
still).
b. Sharon didn’t butter the toast with a knife (she used a spoon) and the same goes for
Leo (he stabbed it).
The presuppositional aspects of the sentences in (39) were ignored in the Russellian
treatment Davidson (1967) offered of the logical form of action sentences. They were
also ignored in the lengthy debates that ensued. From the point of view of an ad-
equate semantic theory of natural language, ignoring them was a mistake. (Again,
there are subtleties to attend to in the case of privative opposition, in which one ex-
pression is more specific and entails the other more general expression but not
conversely, as ‘dog1’ and ‘dog2’ and, on the standard view, ‘not1’ and ‘not2’. For dis-
cussion see Atlas 1977b: 328–30, 1989: 74–77.)
According to four Zwicky and Sadock (1975) tests for ambiguity—transforma-
tional potential, conjunction reduction, semantic differentia, and anomaly of priva-
tive opposites—the difference between presuppositional and nonpresuppositional
understandings of the negative sentence ‘The king of France is not wise’, or of ‘It’s
not {true/ the case} that the king of France is wise’, is not a difference in sense. The
sentence is not ambiguous; it does not have two or more logical forms. It does not
contain syntactical “scope ambiguities” with respect to ‘not’. Nor does it contain a
lexically ambiguous ‘not’, one that is an exclusion negation and another that is a choice
negation. These four tests suggest that each negative sentence is sense nonspecific
with respect to the exclusion- and choice-negation understandings. Quine (1974: 484)
actually once remarked that “the English ‘not’ obviously is not the sign of negation
in the logical sense.” But I (Atlas 1974) was never satisfied with knowing that ‘not’
is not a one-place, extensional, sentence operator; what I showed (Atlas 1974, 1975a,b,
1977b, 1989) was that ‘not’ is a semantically nonspecific adverbial. This conclusion
is not what two millennia of logical and philosophical doctrine would have led one
to expect.
Now let us apply the conjunction-reduction ambiguity test to sentences with
numerical noun phrases. For example, the sentence in (40) is an acceptable reduced
form, one grammatically possible, literal interpretation of which is (41):
(40) The council houses are big enough for families with three kids, which is enough to
move a living room rug.
(41) The council houses are big enough for families with [at most 3 kids], [exactly 3 kids]
is enough to move a living room rug.
If three were semantically ambiguous between exactly 3 and at most 3, the reduced
form (40) could not have the meaning of (41). Yet in fact it can. Likewise the sen-
tence in (42) is also an acceptable reduced form one grammatically possible, literal
interpretation of which is (43):
(42) The council houses are big enough for families with three kids, which, by the way,
qualifies Mrs. Smith for child-care assistance and will allow her to lease a bigger house.
(43) The council houses are big enough for families with [at most 3 kids], [having at least
3 kids], by the way, qualifies Mrs Smith for child-care . . .
If three were semantically ambiguous between at least 3 and at most 3, the reduced
form (42) could not have the meaning of (43). Yet, in fact, it can. The possibility of
these interpretations as literal interpretations for the reduced sentences demonstrates
that three N is not ambiguous but is semantically nonspecific among at least 3, at
most 3, and exactly 3. The connection between ‘3’ and ‘three’ is roughly this: in
extensional contexts ‘exactly 3’ is substitutable salva veritate for ‘exactly three’. But
‘3’ is an Arabic numeral, a canonical name for a positive integer; ‘three’ is a numeri-
cal adjective of English.
The inference from the semantically nonspecific three N to the specification at
least 3 N or to the specification exactly 3 N is justified by Atlas and Levinson’s (1981)
Maxims of Relativity and the Principle of Informativeness or by Horn’s (1984b)
R-implicatures. The lexical meaning of the noun phrase three N is nonspecific be-
tween these specific interpretations. Thus exactly 3 and at least 3 are products of a
pragmatic, context-dependent, Gricean-like process of utterance-interpretation but
not of a classical First Maxim of Quantity conversational implicature from at least 3
to no more than 3. Rather, like the inference that I have discussed from the semanti-
cally nonspecific, scope-unspecified semantic representation of The king of France
is not bald to a choice-negation proposition (or, in certain contexts, to an exclusion-
negation proposition), the addressee’s pragmatic inference takes a semantically non-
specific semantic representation, which is what the language-knower knows when
he knows the literal meaning of the sentence-type or sentence-token, and yields a
numerically specified utterance-interpretation. We now have a genuine pragmatic
contribution to truth conditions of utterances, since antecedent to the determination

of a truth-value is the construction of an interpretation that is able to carry a truth-
value (Atlas 1979). Sperber and Wilson (1986b) later made a similar point under the
rubric ‘explicature’, as have Récanati (1989, 1993) and Bach (1994a), the latter under
the rubric ‘completion’. The semantically specified interpretation of the utterance is
not identical to the semantically nonspecific reading of the sentence; it is inferred
from contextually available information (Atlas 1979). The English word ‘three’ is
semantically nonspecific, semantically neutral between the exactly 3 and the at least
3 interpretations. So, ‘three’ has a meaning in combination with distributive, collec-
tive, or lexically incorporating terms, as in Three children are in the room, Three
children moved the piano, x is a three-sided figure; the meaning of the whole deter-
mines the meaning of the part, or, to paraphrase Frege, only in the context of a noun
phrase does a numerical modifier have a meaning.
A collective term is understood to require semantically a size-term modifier (ex-
actly 3), while asserting a sentence in which three N occurs scalar-implicates no
more than 3 ONLY IF the NP is understood distributively. In neither case does the
English word ‘three’, by itself, mean EXACTLY 3or mean AT LEAST 3. In combina-
tion with a collective term, or with lexical incorporation, three takes on a meaning;
(e.g., the meaning of the collective term three children and of three-sided is that of
exactly 3 children and exactly-3-sided, respectively), but this is not because ‘three’
means the same as ‘exactly 3’. Nor does ‘three’ mean the same as ‘3’.
By partisans of Relevance Theory, the admitted weakness of the First Maxim of
Quantity implicature analyses is mistakenly taken to support the program of Relevance
Theory. By partisans of Seuren’s (1985) Discourse Semantics, and similar programs
like Kamp’s (1981; Kamp and Reyle 1993) Discourse Representation Theory, anaphoric
reference is mistakenly taken to eliminate totally the need for neo-Gricean pragmatic
inference. Both theories ignore the implications of Atlas (1974, 1975b, 1978a,b, 1979,
1983), Atlas and Levinson (1981), and Horn (1984b) for the classical Gricean analy-
sis: the development of Grice’s Second Maxim of Quantity analyses as Atlas and
Levinson’s (1981) Maxims of Relativity implicatures or Horn’s (1984b) R-implicatures.
That post-Gricean revision, like the analysis that I have offered here, makes essential
use of semantical nonspecificity. Seuren wishes to avoid the complications that seman-
tically nonspecific expressions imply for a formal logic and to eliminate the need for
Gricean explanations entirely. My discussion, although critical of certain details of
Carston’s (1988) and Seuren’s (1985, 1993) views, is in large agreement with their
motivating linguistic intuitions and with their justified skepticism that Grice’s First
Maxim of Quantity implicata are sufficient to explain the “strengthening” (Grice 1989c:
47) of “what is said.” Their analyses focus welcome attention on two questions that
have been my concern to address in Atlas (1978a,b, 1979, 1989) and in Atlas and
Levinson (1981):
1. What are the literal senses that must be attributable to sentence-types

to best explain what natural, conversational inferenda and implicata
in fact arise from assertions of their tokens?
2. What types of inference best explain how those inferenda and
implicata arise from the literal senses of the sentences asserted?
Horn (1992b: 172–75), reflecting on the arguments (Atlas 1990) that I have just re-
hearsed here and on arguments of Sadock (1984), Carston (1985, 1988), Horn (1972),
and others, admits that cardinal numerals pose special problems for a classical Gricean
scalar implicature analysis, though, unlike the relevance theorists, he does not wish
to give up on scalar implicatures entirely. He points out a difficulty for the classical
Gricean, noted by Sadock (1984: 142–43), that the arithmetic statement in (44) would
be true since it would have, on the Classical Gricean analysis, the true interpretation
of (45):
(44) 2+2=3
(45) 2 or more plus 2 or more is 3 or more.
Horn (1992b: 173) then comments, “It is plausible, as Atlas (1990) has suggested,
that mathematical values are simply lexically distinct from the corresponding nu-
meral words of natural language, which themselves are unspecified among their ‘ex-
actly n’, ‘at least n’ and ‘at most n’ values.”
Another special feature of the numerals, noted by Horn (1972) and again by
Carston (1988), Hirschberg (1985), and Sadock (1984), is the possibility of context-
dependent scale reversal. For example, note the difference between (46a) and (46b)
(Horn 1992b: 173):
(46) a. That bowler is capable of a round of at least 100 (and maybe even 110).
b. That golfer is capable of a round of at least 100 (and maybe even 90).
A third feature, noted by Horn (1972: 37–38), Hirschberg (1985), and Atlas
(1983, 1990), and as discussed here, is the behavior of numerical terms when they
are lexically incorporated. As Horn remarks:
A triple (three-base hit) is not (at least) a double (two-base hit), although the list of
players with two base hits in a game may include those with three. . . . Atlas (1990)
argues persuasively that the ‘exactly n’ interpretation of incorporated cardinal [nu-
merals] is to be linked to the collective or group readings which themselves sys-
tematically exclude minimalist treatment. This extends to the reading of Carston’s
[(a)], as Atlas points out, citing the contrast between that sentence and its distribu-
tive (and scalar-implicating) counterpart [(b)]:
a. If there are three books by Chomsky in the shop, I’ll buy them all.
b. If there are three book by Chomsky in the shop, I’ll buy each of them.
Koenig independently notes the ‘exactly n’ intepretation of sentences like Three

boys carried a sofa up the stairs (*in fact four) and comes to the same conclusion:
‘only distributed readings of count phrases give rise to scalar implicatures’ (Koenig
1991: 4). (Horn 1992b: 173–74)
Horn concludes:
In sum, while we can accept Atlas’s argument (1990: 15) that “only in the context
of an NP does a numeral modifier have a meaning,” no analogous conclusion fol-
lows for the full range of scalar values. The signs point to a mixed theory in which
sentences with cardinals [i.e., cardinal numerals] may well submit naturally to a
post-Gricean pragmatic enrichment analysis of what is said, while other scalar predi-
cations continue to submit happily to a neo-Gricean minimalist implicature-based
treatment. (Horn 1992b: 175)
4 Classical Gricean pragmatics revived?
Recently Levinson (2000: 87–90) defended the original observations of Horn (1972).
The assertion (47a) is claimed to have the literal meaning of (47b) and the general-
ized conversational implicatum (47c). His reason for holding to the classical Gricean
view is that it is possible consistently to suspend the implicatum with an ‘if not four’,
a ‘and perhaps more’ and a ‘or possibly five’ as in (47d). This shows, he believes,
that the implicatum (47c) is not entailed by (47a).
(47) a. John has three children.

b. John has at least three children.
c. John has at most three children.
d. John has three children, if not four / and perhaps more / or possibly five .
Unfortunately, nothing of the sort is shown. Even if John has three children meant
what John has exactly three children does, (47d) would still be consistent under that
interpretation, as illustrated in (48):
(48) John has exactly three children, if not exactly four / ?and perhaps more / or possibly
exactly five.
Here there is only an infelicity, not an inconsistency, with the continuation . . . and
perhaps more, and not even an infelicity with . . . if not exactly four or with . . . or
possibly exactly five.
The one further piece of evidence that Levinson (1983: 116; 2000: 88) adduces
is an alleged contextual cancellation of the alleged implicatum at most three in a
context like (49):
(49) A: Does John qualify for the Large-Family Benefit?

B: Sure, he has three children all right.
Here an exact specification is irrelevant in the context. But I have already argued
that the neo-Gricean hypothesis that the literal meaning of (49B) is Sure, he has at
least three children all right is inconsistent with the Zwicky and Sadock (1975) tests
for the semantical nonspecificity of three children. Levinson’s classical Gricean
hypothesis is also prey to an infinite regress argument that the literal meaning is Sure,
he has (at least)n three children all right, 1 ≤ n , for positive integers n. Nothing in
Levinson’s remarks explains how the context’s failing to need an exact specification
blocks the strengthening of the alleged literal meaning at least three by what is sup-
posedly a default implicatum at most three. Does the default strengthening subcon-
sciously occur, and then the interpreter of the utterance, understanding that the in-
formation is “de trop,” then reinterpret the utterance in a weaker form? How can this
context block the exactly three interpretation, since if having at least three children
qualifies John for the Large-Family Benefit, obviously having exactly three children
qualifies him as well. The reinterpretation of the assertion cannot depend on the in-
correctness of the answer Sure, he has exactly three children all right, since it is not
incorrect, but on, if anything, the interpreter’s judgment that it is inappropriate be-
cause it gives too much information. But if cancelation depends on that judgment,
the strengthening by the default implicature has already occurred, the interpretation
exactly three children is already available, and though it provides more information
than needed, it correctly answers A’s question in (49). So why should A take it upon
himself, once his question has been answered, to reinterpret B’s utterance by sub-
tracting the information at most three children from the original default interpreta-
tion? Just because A realizes that the weaker, allegedly literal meaning at least three
children would also answer his question?
Linguistic cancelation phenomena indicate, as Levinson assumes, that a propo-
sition is not an entailment of an assertion. Contextual cancelation phenomena is an-
other matter. Levinson has no theory at all, except the model of Gazdar’s bucket
(Gazdar 1979a), which is inappropriate to Levinson’s example, that would require
B’s background knowledge to include the information that John has three or more
children prior to B’s assertion John has three children in order to be inconsistent
with, and so cancel, B’s potential implicatum John has at most three children. But
B’s knowledge does nothing for A’s interpretation of B’s assertion. And if A already
knows that John has three or more children, A did not need to ask B the question in
(49). Levinson has no explanation of the contextual cancelation of the alleged de-
fault implicatum at most three children of B’s asserting John has three children.
In an earlier discussion of an example of the same sort, Levinson (1983: 115–
16) remarks that “implicatures can just disappear when it is clear from the context
of utterance that such an inference could not have been intended as part of the
utterance’s full communicative import,” because “it is clear from the context that
all the information that is required is whether” in order to meet the requirements of
the Benefits Scheme John has at least three children. But again, that is just a non
sequitur. Even if it is clear that all that is required to answer A’s question is whether
John has at least three children, that epistemic fact about the minimum that has to
be known to answer A’s question does not determine the result of A’s inference to
the best interpretation of B’s utterance in the context. One wonders by virtue of
what miracle the default, generalized conversational implicata of assertions “just
disappear” because the speaker could have meant less that what his assertion would
standardly convey and he still succeed in communicating appropriate information
to the addressee.
On Levinson’s (2000: 37) own view of his Informativeness Heuristic, “minimal
specifications get maximally informative or stereotypical interpretations,” Levinson’s
heuristic would yield the irrelevant but maximally informative John has exactly three
children, contrary to Levinson’s own claim about (49). Levinson (2000: 37) uses his
heuristic to motivate Grice’s Second Maxim of Quantity, “Do not make your contri-
bution more informative than is required,” but they make incompatible predictions
about (49). If B is conforming to that maxim, his utterance of three children possibly
should have the interpretation at least 3 children. On my view of the semantical
nonspecificity of ‘three children’, it can have that interpretation, and so uttering it
conforms to Grice’s Second Maxim of Quantity. I have already indicated in chapter 3,
section 1, of this book that Levinson’s formulation of the Informativeness Heuristic
was too simple. The explanation of (49) is a case in point.
Of course Levinson (2000: 88–89) is aware of the arguments of Sadock (1984)
and Horn (1992b), Atlas (1983, 1990), Kempson (1986), and Carston (1985, 1988),
as discussed here, as well as work by T. Fretheim (1992), J. van Kuppevelt (1996),
and R. Scharten (1997), which throw doubt on the classical Gricean analysis of car-
dinal adjectives. But he is unwilling to give it up:
Still, when due allowance is made for the special role of number words in math-
literate cultures, and consequent possible conventionalization of the ‘exactly’ read-
ings, there are a number of reasons to hang on to a scalar interpretation of ordinary
language numeral expressions in general. (Levinson 2000: 90)
And what might those reasons be? He continues:
One central piece of evidence is provided by those languages that have a finite se-
ries of numerals. Many Australian languages, for example, have just three number
words, which are glossed as ‘one’, ‘two’, and often ‘three’. The scalar prediction is
clear in those cases: we have a finite scale <’three’, ‘two’, ‘one’>, where ‘one’ or
‘two’ will implicate ceteris paribus an upper bound; but because there is no stron-
ger item ‘four’, the cardinal [numeral] ‘three’ should lack this clear upper bound-
ing by [generalized conversational implicature]. And this is clearly the case in, for
example, Guugu Yimithirr: nubuun can be glossed ‘one’, gudhirra ‘two’, but
guunduu must be glossed ‘three or more, a few’. . . . None of the other theories makes
this correct prediction. (Levinson 2000: 90)
As an alternative to Levinson’s classical Gricean view, my (Atlas 1983, 1990) and

Carston’s (1985, 1988) semantical nonspecificity view of the meaning of numerical
noun phrases is consistent with the facts that Levinson presents; the alternative theory
also makes the correct prediction. The finite scale <“n”, “n-1”, . . . , one>, where “n”
denotes the natural language translation of the cardinal numeral : of the cardinal num-
ber n , implies that the upper-bounding scalar implicature not “n + 1” N of asserting
“n” N is not a default implicature, so on my account one would expect specifications
of exactly “n” N or of at least “n” N but not at most “n” N , at least as long as the
quantitative concept AT LEAST and logical concept EXACTLY, involving identity
= , were expressible in the language. Nothing in Levinson’s data commits one to
Levinson’s classical Gricean account of the semantics of numerical noun phrases.
Levinson’s claim that only the classical Gricean analysis of numerical NPs explains
the data of languages with finite scales is mistaken.
As it happens, guunduu in Guugu Yimithirr, pace Levinson (2000: 90), cannot
be unambiguously glossed as both ‘three or more’ and ‘a few’. ‘A few boys’ is a
consistent NP in Zwarts’s (1998) sense—their predicate negations entail their sen-
tence negations—but ‘three or more boys’ is not a consistent NP. In fact, I would
treat the upward monotonic, consistent NP ‘a few boys’ in the same way I treat the
downward monotonic consistent NP ‘none of the boys’ (See Atlas 2001: 16–17)—
namely, as a functor (singular or collective term) instead of a classical quantifier NP.
Levinson’s (2000: 178) account of “indexical resolution” is also affected by his
classical Gricean view. He wishes to show that there are implicatural-like contribu-
tions to “what is said”—to the proposition literally expressed. (Many of these are the
“explicatures” of Relevance Theory [Sperber and Wilson 1986b].) A typical example
is (50a), whose literal meaning is allegedly (50b):
(50) a. Take THESE three drinks to the three people over THERE; take THESE four to the
four people over THERE.
b. Take THESE 3 or more drinks to the 3 or more people over THERE; . . .
c. Take THESE 3 or fewer drinks to the 3 or fewer people over THERE; . . .
d. Take THESE (exactly) 3 drinks to the (exactly) 3 people over THERE; . . .
Here the implicatum is allegedly (50c), to give the communicated content (50d).
Levinson (2000: 178) wishes to challenge the assumption “that the assignment of
indexical values has nothing to do with pragmatic inference,” a challenge that I sup-
port—but not because the literal meaning of (50a) is (50b). What Levinson (2000:
183) thinks is a Horn scalar implicature is actually an Atlas and Levinson (1981)
Maxim of Relativity/Informativeness implicature instead—a view with which Horn
(1992b) agrees. My I-implicature analysis likewise explains Levinson’s (2000: 183)
“ellipsis unpacking” data in (51):
(51) A: Which side got three goals?

B: Tottenham Hotspurs.
This also explains his intuition that the scalar implicata of the antecedent of a condi-
tional can affect the truth-evaluation of the conditional:
(52) If each side in the soccer game got three goals, then the game was a draw.
Statement (52) is an example of my Fregean context-principle that only in the con-

text of a sentence (token) does a numerical NP have a (specific) meaning.
Levinson offers a further and most interesting argument, using Horn’s (1985)
notion of a “metalinguistic negation.” Levinson’s (2000: 213) datum (53a) is given
the classical Gricean literal meaning (53b) and the implicatum (53c):
(53) a. He has four children rather than three.

b. He has 4 or more children rather than 3 or more children.
c. He has no more than 4 children rather than no more than 3 children.
d. #He has at least 4 children but not at least 3 children.
The classical Gricean semantic analysis results in the hypothesis that what is negated
in the understood and not three children of (53a) is the First Maxim of Quantity sca-
lar implicatum of three children—that is,what is negated in (53a) is the implicatum

at most 3 children. Hence the negation in (53a) must be metalinguistic in Horn’s
(1985) sense; thus, “Do not infer ‘at most 3 children’.” For if one adopts the classi-
cal Gricean ‘at least 3 children’ analysis, simply negating ‘at least 3 children’ by the
use of ‘rather than’ in (53a) would yield a literal meaning for (53a) that entails the
contradictory (53d). Since our linguistic judgment is that the literal meaning of (53a)
is not logically inconsistent, we need an alternative analysis of the role of negation
in (53a). Levinson’s “metalinguistic negation” account avoids the mistaken predic-
tion that the utterance-meaning of assertion (53a) is inconsistent, even though it ac-
cepts that the literal meaning of the sentence is inconsistent.
But there is another explanation than Levinson’s. If one abandons the classical
Gricean semantic analysis of ‘three children’ as ‘at least 3 children’ and abandons
the First Maxim of Quantity scalar implicature to ‘at most 3 children’, one does not
need to hypothesize a “metalinguistic negation” to avoid the mistaken prediction that
the literal meaning of the consistent sentence (53a) entails the inconsistent (53d). The
Principle of Informativeness and the Maxims of Relativity (Atlas and Levinson 1981:
40–41) generate implicatures from the semantically nonspecific numerical noun
phrases in sentence (53a) ‘He has four children rather than three’ that give a finite
range of consistent, literal utterance-interpretations (54 a–f):
(54) a. He has at least 4 children but not exactly 3.

b. He has at least 4 children but not at most 3.
c. He has exactly 4 children but not exactly 3.
d. He has exactly 4 children but not at most 3.
e. He has at most 4 children but not at least 3.
f. He has at most 4 children but not exactly 3.
None of these consistent, literal interpretations of the sentence is produced by

Levinson’s (2000: 213) classical Gricean theory, which necessarily appeals to Horn’s
(1985) “metalinguistic negation” of an utterance to avoid predicting an inconsistent
literal meaning for an obviously consistent sentence. Abandoning the classical Gricean
semantic analysis means that there is no motive to posit a metalinguistic utterance-
negation; it is theoretically otiose. Since Horn’s pragmatic “metalinguistic negation”
analysis is prey to various difficulties—as, for example, those discussed in Kempson
(1986)—the semantical nonspecificity analysis allows one to drop the “geocentric”
Gricean hypothesis that ‘three children’ has the same meaning as ‘at least 3 children’,
the epicycle that negation is metalinguistic in utterances of (53a), and the false pre-
diction that the literal meaning of the sentence in (53a) is logically inconsistent. Since
I have already been accused of “the radical move of throwing out the model-theoretic
baby with the ambiguist bathwater” (Horn 1989: 423) in my treatment of ‘not’, why
stop there? What Levinson’s ‘rather than’ datum shows is that I was not radical
enough. Levinson’s (2000: 213) justification for hypothesizing the “metalinguistic
negation” of rather than utterances is to save the pragmatic phenomena of the con-
sistency of utterance-meanings, but he fails to save the semantic phenomenon, the
consistent ‘rather than’ sentence-meaning. Once one realizes that the sentence-
meaning of (53a) is logically consistent, there is no justification for the “metalinguistic
negation” analysis of utterances of (53a)! Language, like the solar system, is better
understood without a static model. Having thrown out the model-theoretic baby and
the ambiguist bathwater, it is now time to get rid of at least a few of the “metalinguistic
negation” tubs.
5 Horn’s metalinguistic negation
To that end I shall consider another case in which “metalinguistic negation” is ap-
pealed to in order to resolve an alleged contradiction. Horn (1989: 374) and Levinson
(2000: 212) give examples of “metalinguistic negation”:
(55) a. SOME men aren’t chauvinists—ALL men are chauvinists.

b. It’s not a [vα:z], it’s a [veiz]. (Levinson 2000: 212)
c. I didn’t trap two monGEESE—I trapped two monGOOSES.
d. It’s not stewed bunny, honey, it’s civet de lapin.
e. I’m not his daughter—he’s my father. (Wilson 1975: 152)
Horn notes that these examples involve
contrastive intonation with a final rise within the negative clause (the ‘contradic-
tion contour’ of Liberman and Sag [1974] or the fall-rise of Ladd [1980], as the
case may be), followed by a continuation in which the offending item is replaced
by the correct item in the appropriate lexical, morphological, and phonetic garb—
RECTIFICATION, to borrow the label of Anscrombre and Ducrot (1977). (Horn
1989: 374)
Horn believes the same holds of the presupposition-canceling example in (56). It is

clearly linguistically acceptable to assert:
(56) The king of France is not bald—(because) there is no king of France.
Horn (1985, 1989) has argued that the interpretation of ‘not’ required to explain the
acceptability of this utterance does not suggest the claim that ‘not’ is semantically
ambiguous, between exclusion (external) and choice (internal) negation readings.
Rather, Horn believes that the statement is pragmatically ambiguous (Donnellan
1966), and in the statement above ‘not’ has the force of a metalinguistic operator,
indexed to a speaker, I object to U, where U is an utterance-token or utterance-
type. He further observes:
The principal resemblance between the instances of marked negation [in (55)] and
the classical examples of presupposition-canceling negation [in (56)] . . . is that both
types occur naturally only as responses to utterances by other speakers earlier in
the same discourse-contexts, or as mid-course corrections after earlier utterances
by the same speakers. It is for this reason that I seek to encompass all these examples
under the general rubric of meta-linguistic negation: they all involve the same ex-
tended use of negation as a way for speakers to announce their unwillingness to
assert something in a given way, or to accept another’s assertion of it in that way.

Given the behavioral resemblances just cited . . . , as well as Occamist consider-
ations, there is no obvious reason not to collapse the presupposition-canceling ne-
gation of [(56)] with the negation attaching to conversational implicature in [(55a)],
to pronunciation [(55b)], to morphology or syntax [(55c)], to register, speech level,
or social attitudes [(55d)], and to perspective or point of view in [(55e)]. (Horn 1989:
374–75)
As Horn (1989: 371) also notes, the examples in (55) all have a focus of negation:
some in (a), ‘[vα:z]’ in (b), -geese in (c), stewed bunny in (d), and his daughter in
(e). For those reasons, I (Atlas 1983) preferred calling this use of ‘not’ a “focal nega-
tion” instead of a meta-linguistic negation, reserving the latter for metalinguistic
sentences as classically understood, for example:
(57) The plural of ‘mongoose’ is not ‘mongeese’; it is ‘mongooses’.
If example (56) is really like (55), it should have the same properties of intonation,
stress, and focus. Before we consider whether it does, let’s consider an example in
which English syntax provides the focus, a cleft sentence:
(58) It isn’t the king of France who is bald—(because) there is no king of France.
As McCawley (1981: 241) observed, the statement made in asserting (58) does not
presuppose in its first clause that there is a king of France. In the cleft statement
‘the king of France’ is a focal not a topic NP. By the Grice-Strawson Condition
(Atlas 1988, 1989, 1991a, 1996b) that there is a referential presupposition of a state-
ment in which a simplex singular term occurs only if the singular term is a topic
noun phrase in the statement, there would be predicted to be no referential presup-
position of there being a king of France in a statement of the first clause of (58).
Thus there is no reason to expect a statement of (58) to be linguistically anoma-
lous or logically inconsistent; there is no contradiction between a presupposition
of the first clause and the assertion of the second, since there is no presupposition
There is a king of France of the asserted first clause at all. So the explanation of
the consistency is not that there must be a “metalinguistic negation” in (58), para-
phrased by (59):
(59) I (the speaker) object to ‘It’s the king of France who is bald’—(because) there is no
king of France.
As Karttunen and Peters (1979: 46–47) pointed out, uses of focal negation in-
hibit the usual negative polarity items, such as any in (60c):
(60) a. Chris managed to solve some problems.

b. Chris didn’t manage to solve any problems.
c. Chris didn’t MANAGE to solve {some/*any} problems—he solved them easily.
But the negative cleft statement allows negative polarity items, such as any in (61):
(61) It wasn’t managing to solve {some /any} problems that Chris accomplished.
By Horn’s criterion that negative polarity items are inhibited by “metalinguistic”

negation, the acceptability of example (61) suggests that ‘not’ in the cleft statement
is not a “metalinguistic” negation. The same goes for the negative cleft in the first
clause of (58), repeated as (62):
(62) It isn’t the king of France who is bald—(because) there is no king of France.
Here an ordinary negation combined with the king of France being in focus gives us
a linguistically acceptable, logically consistent statement. And the same holds for
(63), understood to have its typical stress and intonation (marked by a stress accent
and horizontal bars for intonation levels) and for (64):
—————————
——
—–
(63) The king of France/ isn’t bald—there is no king of France.
—–— ––
———––– – —–
(64) The king of France1 is not 2 bald 0—there is no king of France.
In both (63) and (64) the NP the king of France typically has greater than normal
stress, making clear that it is a focus and not a topic NP. Why should this be?
A speaker who asserts (63) or (64) will put greater than normal stress on the king
of France because the speaker’s whole utterance includes a clause there is no king of
France. A speaker who believes that there is no king of France and who wishes to
communicate un-misleadingly with his addressee would not utter the clause ‘the king
of France is not bald’ with stress and intonation that would mislead the addressee or
garden-path the addressee’s comprehension of the speaker’s utterance. Cooperative-
ness in the conversation would require that the speech-production mechanism pro-
duce the clause with phonetic properties that would not cause the addressee to infer
that ‘the king of France’ was a referring singular term, because generating such an
inference would create an inconsistency with the remainder of the speaker’s utter-
ance. So, taking advantage of the addressee’s competence in recognizing stress and
intonation of an utterance, the speaker indicates that ‘the king of France’ is not a topic
NP. By the Grice-Strawson Condition, this implies that the assertion of the first sen-
tence of (63) or of (64) will not “presuppose” that ‘the king of France’ has a refer-
ence. The presupposition is not canceled by a contradictory clause in the rest of the
sentence being asserted; it was never there to be contradicted.
What is doing the work here is not a contradiction intonation contour. What is
doing the work is stress—a focus NP stress, or, more weakly, a non-topic-NP
stress—that prevents the occurrence of the presuppositional interpretation that there
is a king of France. The theoretical description ‘presupposition-canceling’ for these
statements presupposes that there is a presupposition to be canceled. That is false.
Without a presupposition, “presupposition cancelation” is a myth. In (63) or (64)
there is no more a presupposition that there is a king of France than there is one in
the cleft (65):
(65) It isn’t the king of France who is bald.
And as Atlas and Levinson (1981) and McCawley (1981: 241) observed, there is no
such presupposition for a focal NP ‘the king of France’ in a cleft statement.
Why then, one might well ask, has it been so common to describe such cases as
presupposition-canceling? The answer, I believe, is an artifact: the projection prob-
lem. Linguists have assumed that compositionality applies to the presuppositions of
constituent clauses of compound and complex sentences: the presuppositions of the
whole are a function of the presuppositions of its parts, where it is assumed that the
presuppositions of a part are independent of the presuppositions, or assertions, of
another part. But that is a too-simple account of what the presuppositions of the parts
are. Sometimes whether a part has a presupposition is a look-ahead function of other
parts of the sentence that is being asserted.
Since there is no such referential presupposition in the utterance of the first clause
of (63) or (64), there is no need for a “metalinguistic” interpretation of ‘not’ in that
uttered clause to resolve an inconsistency in the whole statement, since there is no
logical inconsistency between the parts of the utterance. These observations under-
mine the explanation of (63) as a “metalinguistic” negation given by Horn (1985,
1989) and Burton-Roberts (1989a,b) (see Atlas 1991b). For what is essential here is
not semantic or pragmatic properties of ‘not’ in the assertion; what is essential is the
focused, non-topic, prosodic character of the occurrence of the singular term in the
negative utterance.
It also follows from this analysis that the attempt to assimilate classic “presup-
position-canceling” examples like (63) to focal negation examples like (55) is a mis-
take. It is therefore no surprise to find data, from Seuren (1988: 195), Horn (1990:
498), and Carston (1998: 328), that distinguish the inconsistent focal negation ex-
amples (66a) and (66b) from the consistent, so-called presupposition-canceling (66c):
(66) a. #It’s not true that we saw some mongeese; we saw some mongooses.
b. #It’s not true that he’s my father; I’m his daughter.
c. It’s not true that the king of France is bald; there is no king of France.
Contrary to popular opinion, what needs to be explained, and has been explained in
this chapter, is the fact that there is no contradiction—and so no cancelation—
between the first clause as uttered and the second clause of the utterance. What is
also explained is the fact that by conforming to Grice’s Cooperative Maxim (“Make
your contribution such as is required for current purposes of the exchange.”), the
speaker, by his use of stress on ‘the king of France’, avoids an interpretation of the
utterance by the addressee that is logically inconsistent or linguistically anomalous,
or an interpretation that imposes gratuitous processing difficulties on the addressee
in comprehending such a denial of ‘The king of France is bald’.
I would modify Grice’s Modified Occam’s Razor to read: Do not multiply “prag-
matic ambiguities” beyond necessity. The Grice-Strawson Topichood Condition on
the existence of presuppositions of NPs removes the necessity of positing “pragmatic

ambiguity” in (56), (63), and (64), since there is no inconsistency for the ambiguity
to resolve.7
6 Neoclassical Semantics
In Atlas (1978a,b, 1979), reflecting on the semantics of the free morpheme ‘not’ in
definite description sentences, on presupposition, and on the misuse of the concept
of ambiguity by philosophers and linguists, I suggested that conversational inferences
had two roles. The first was the enrichment of the semantic representations of sen-
tences to construct well-defined interpretations of utterances of the sentences that
were capable of bearing a truth-value. The second was the classical Gricean infer-
ence from contents “said” to contents “implicated.” In Atlas (1978a,b, 1979), the first
of these a colloquium in the Department of Linguistics and Phonetics, University
College, London, I explicitly claimed that pragmatic inferences must play both these
roles, because the semantic representations of many types of sentences, the products
of the syntactical rules and the lexicon, were semantically underdeterminate—that
is, semantically nonspecific with respect to a semantical feature [F], such as seman-
tic gender, as in the gender-nonspecific ‘neighbor’ by contrast with the specific ‘he’
and ‘she’. I had long argued that negative definite description sentences The F is
not G were not semantically scope-ambiguous with respect to negation and were
not ambiguous between a choice negation lexeme ‘not1’ and an exclusion negation
lexeme ‘not2’ (Gazdar 1979a: 64–66).8 The semantical underdeterminacy, rather
than sense ambiguity, of these sentences meant that the (meaningful) sentences
did not represent, on any reading, the speaker’s meanings, which would be an exclu-
sion negation or a choice negation (see Atlas 1975b, 1977b, 1978a,b, 1979). The se-
mantically nonspecific semantic representation (SR) was not identical to any specific
truth condition, either the exclusion negation or the choice negation. A negative sen-
tence with a semantically nonspecific SR would have to be enriched, made specific
or precise (see chapter 3), by an inferential mechanism exactly like the inferences
made in conformity with the Maxims of Relativity and the Principle of Informa-
tiveness (Atlas and Levinson 1981), which are the hearer-centered analogues of
Grice’s speaker-centered Second Maxim of Quantity (“Do not make your contribution
more informative than is required”). (See chapter 3 in this volume for discussion.)
In Atlas (1979) I had argued that an account in which negative sentences The F
is not G were scope-ambiguous was redundant. If pragmatic inferences were a nec-
essary part of a theory of utterance-interpretation, and the sentences were ambiguous,
the inferences would be essential to disambiguation, to the selection of the appropri-
7There is more to say about the argument that I have sketched here, but the essentials of the argument
are present. See Burton-Roberts (1989a,b, 1997, 1999), Carston (1996, 1998, 1999), Chapman (1996),
Foolen (1991), Geurts (1998), Horn (1985, 1989), Kempson (1975, 1986), Levinson (2000), McCawley
(1991), Seuren (1990, 2000), van der Sandt (1991), Wilson (1975), and Yoshimura (1998).
8Note that this point is independent of the features of deixis, reference of singular terms, ellipsis, tense,
or other parameters that must be fixed in order to produce an interpretation of the utterance.
ate sense in the context of utterance. On the assumption that the addressee analyzes
the speaker’s sentence-token and discovers two or more senses, a choice of the ap-
propriate sense must be made in light of collateral information available in the con-
text. Thus some inferential mechanism must produce a decision on the best “fit”
between each sense and the context of utterance. If the sentence is unambiguous but
semantically nonspecific for semantic features F1, F2, and so on, in the context the
inferential mechanism must give an appropriate, more specific or precise, interpre-
tation of the semantically underdeterminate sentence. I hypothesized that since the
same inferential mechanisms were at work in both the cases of ambiguity and of sense
underdeterminacy (nonspecificity)—one to select a reading, the other to construct a
more specific or precise interpretation—the difference between selection and con-
struction was that the construction of a specific interpretation operated on a pre-
“propositional” semantic representation that was too underdeterminate to carry a
truth-value. Contrary to Grice’s view that Gricean mechanisms operated only post-
“propositionally” to give what the speaker conveyed or suggested by, in, or when
asserting a sentence, what statement the speaker asserted was determined by prag-
matic inference in addition to the semantical interpretation of context-dependent
indexicals, demonstratives, singular terms, tense, and so on.
I (Atlas 1979) also observed that classical Gricean theory could not give an ad-
equate explanation of the inferential mechanism: sentence negations Not (The F is
G) were transformed into predicate negations The F is non-G, and the references of
singular terms were fixed, but no coherent account was given of the case in which
Grice’s sentence-negation meaning, namely the exclusion negation (sometimes ex-
pressed in English by It is not the case that P, though, as I pointed out in Atlas 1974,
and as Grice 1981 noted—see chapter 4—It is not the case that P can also express
the predicate or choice-negation interpretation [Horn 1989]), passed through the
pragmatic mechanism unchanged, the case in which informative enrichment was
absent. According to the classical Gricean theorist, this case should have been the
semantically “unmarked” case, but no Gricean principle in the theory could non-
circularly explain it to be the unmarked case, the case of saying what one means. In
fact, it is linguistically the “marked” case, as Strawson (1950) first noted.
By contrast, as I mentioned in chapter 1, my claim (Atlas 1974, 1975a,b, 1977b,
1978b, 1979) that the negative sentence-meaning [THE F IS NOT G] was semantically
unspecified for scope meant that the inferential mechanism operated equally in both
cases: it produced the more specific choice-negation interpretation [–(THE F IS G)],
and it produced the more specific exclusion-negation interpretation [¬(THE F IS G)].
In neither case has the speaker “said” something propositional, if by ‘saying’ one
means merely the act of uttering a meaningful sentence of a semantically nonspe-
cific sort. In neither case has the speaker “said” something interpretable, if by ‘say-
ing’ one means performing a locutionary act of producing a token of an utterance-type
that is interpreted by one’s addressee to have a determinate sense and reference (see
Ziff 1972a). Nor, in one case or the other has the speaker “said” what he meant, if by
‘saying’ one means performing an illocutionary act, the content of which is correctly
interpreted by the speaker’s addressee to have the speaker’s intended sense and ref-
erence. (Similar notions of interpreting utterances later surfaced in others’ theories
as well, notably the “explicatures” of the Relevance Theory of Sperber and Wilson
[1986b] and Carston [1988]; the Pragmatic Intrusion of Katz [1972], Walker [1975],
and Levinson [1988, 2000] for the interpretations of singular terms; the two stages
of pragmatic inference in Récanati [1989]—who later in Récanati [1993: 267] ac-
knowledges the earlier views of Atlas [1979], as does Horn [1992a: 265]—and the
“implicitures” of Bach [1994a,b].)
My paradigm for Sadock’s “symbiosis” between the structure and function of
sentences is the relationship sketched in chapter 1 between the semantical representa-
tion of the semantically nonspecific sentence-type and the pragmatic interpretation of
the speaker’s utterance of the semantically nonspecific sentence-token (the token has
the same semantic properties as the type; see Searle 1979b). Just as Sadock (1984) was
asking “Whither Radical Pragmatics?”, post-Gricean pragmatics was going there—see
Atlas (1984a), Horn (1984b), and Levinson (1987a,b)—while parallel developments
were occurring in London and Paris (see Sperber and Wilson 1986b).
The remedy for the difficulties of classical Gricean pragmatics that Sadock (1984)
highlighted is a neoclassical semantics: not φ is semantically nonspecific among
exclusion ¬φ and choice –φ negation interpretations (Atlas 1975a,b, 1977b,
1978a,b, 1979, 1989), and a sentence containing three Ns is semantically nonspe-
cific among [3 OR MORE Ns], [EXACTLY 3 Ns], and [3 OR FEWER Ns] interpretations. As I
(Atlas 1983, 1990) have argued, when N is a collective term, a sentence contain-
ing three Ns takes the [EXACTLY 3 Ns] interpretation. When N is a count noun, a
sentence containing the noncollective term three Ns is semantically nonspecific
among the interpretations, while collateral information in the context permits the
addressee to construct the “best” interpretation of the statement in the context.
As for what ‘not’ or three N literally means, as contrasted with the contribution
that ‘not’ or the noun phrase three Ns makes to the addressee’s interpretation of a
speaker’s statement in which a token of ‘not’ or three Ns occurs, the lexical mean-
ing of ‘not’ or of three N is not a psycholinguistic observable (Atlas 1989: 145–49).
An interpretation, even a self-interpretation of one’s own utterance of ‘not’ or of three
Ns, is already the result of the interaction of semantic with pragmatic processing.
Users of the language can no more say, on the evidence of their own interpreta-
tion of their own utterance of ‘not’ sentences or of three N sentences, what the
lexical content of ‘not’ or of three N in their mental lexicon is than a quantum
physicist can say what the nonsuperposed biological state, alive or dead, of
Schrödinger’s cat in the unopened box is. The perception of the deadness (or alive-
ness) of Schrödinger’s cat in the opened box shows the vitality (or morbidity) of the
cat to be a physical observable. The interpretation of statements made by asserted
sentences containing ‘not’ or the noun phrase three Ns may show that the contents
of statements are psycholinguistic observables, but not all subsentential meaning com-
ponents of a sentence are psycholinguistic observables. My point is not Quine’s (1960)
indeterminacy of radical translation or Davidson’s (1984b) indeterminacy of radical
interpretation, and not Quine’s (1969) inscrutability of reference. It is the inscruta-
bility of literal sense.
Sadock (1984) was partly right. Utterances of the noun phrase three N will
generally Informativeness-implicate (by default) the exact interpretation [3 N], but
that intuitive linguistic observation does not entail that the literal sense of three N
is [3 N]. There is a physical reality described by a Schrödinger wave function Ψt(x),
though only |Ψt(x)|2 is measurable.9 Similarly, there is a semantical reality to ‘not’

and to ‘three’ that contributes content to the literal meaning of the constituent in which
it occurs, although only statements of the sentence have context-specific truth con-
ditions that are interpretable. The assumption that one has introspective access to the
meaning of each word of a language or each lexeme of the mental lexicon that oc-
curs in a sentence, even if one is not a behaviorist like Quine (1960) in the psychol-
ogy of language, is gratutitous.
The novelty of my argument was that the free negative morpheme ‘not’ is such
a sense-inscrutable expression; the further novelty is that ‘three’ and similar numeri-
cal words are also such sense-inscrutable expressions. To think that the meaning of
each word in every syntactical constituent must be a semantical atom whose mean-
ing is “visible” to the mental eye of introspection is to make a naive and theoreti-
cally gratuitous assumption about the transparency of linguistic meaning to the mind,
as gratuitous as the Newtonian assumption that both the position and momentum of
each physical atom in each physical state must each be pointwise definable and “mea-
surable” by Newton’s God. A successful semantico-pragmatic theory of utterance-
interpretation no more requires such an assumption than a successful atomic theory
requires the Newtonian assumption.
On the first page of the first chapter of Atlas (1989: 7) I quote lines from Paul
Ziff’s (1972d) essay, “Something about Conceptual Schemes.” Reflecting on the same
essay Ruhl (1989: 87–91) remarks (p. 86) that “a considerable part of alleged lexical
meaning is actually supplied by other means: words are highly abstract in inherent
meaning, often too much so for conscious understanding.” The novelty of my and Ruhl’s
conclusions lay in their claims that (a) a considerable part of utterer’s-meaning is
nonlexical, and, I added, nonsentential, and (b) literal meaning is abstract (Atlas 1989:
10–24) and accessible to consciousness only with difficulty if at all (Atlas 1989: 8).
The assumption of the Scrutability of Literal Sense is encouraged by the assump-
tion that the semantic representation of a sentence’s literal meaning is restricted to a
representation of truth conditions in an extensional metalanguage. The well-formed
formulae of First-Order Quantification Theory with identity have parsable, scrutable
constituents: individual variables, sentential connectives, predicate expressions, quan-
tifiers. The same is true of intensional languages like Dana Scott’s and Richard
Montague’s (Scott 1970; Montague 1974; Dowty et al. 1981). But, to quote Chomsky
once again:
We cannot assume that statements (let alone sentences) have truth-conditions. At

most they can have something more complex: ‘truth indications’, in some sense. . . .
thought, and various forms of human interaction. (Chomsky 1996c: 52–53)
9For a good layperson’s introduction to basic quantum mechanical notions, I recommend Nick Herbert’s
(1985) Quantum Reality (Garden City, New York: Anchor Press/Doubleday).

If, like Quine and Davidson, one rejects the existence of the analytic/synthetic
distinction and accepts the indeterminacy of radical translation/interpretation, one
accepts truth or reference but rejects its scrutabilty. If, like Grice, Chomsky, Katz,
and myself, one accepts the existence of an analytic/synthetic distinction and rejects
the indeterminacy of radical translation/interpretation, one accepts meaning but re-
jects its scrutability. Davidson and Quine’s position is like that of a classical physi-
cist who, when confronted with the de Broglie wavelike nature of the electron, would
conclude that electrons are “creatures of darkness”—that there cannot be any elec-
trons. This would make the observed electrical properties of matter inexplicable. My,
Chomsky’s, and Grice’s position is like that of a quantum physicist who concludes
that though there are electrons, they are quite different in nature from what classical
physicists had believed. The observed electrical properties of matter are explicable,
although quite differently explained. It no longer makes sense to think of the elec-
tron as a tiny, Newtonian, billiard ball, the electron having precisely determinate
position and momentum, and it no longer makes sense to think of the meaning of
every linguistic expression as a Russellian concept, the expression having precisely
determinate literal meaning and denotation. But lightning still strikes, and sentences
still mean.
Appendix 1
On G. E. Moore’s Term ‘Imply’
T he form of words that generates G. E. Moore’s “paradox” arose, among other

places, in a controversy between Moore and Charles L. Stevenson in the early 1940s.
In a 1942 reply to a discussion of Moore’s ethical views by Stevenson, Moore wrote:
I think Mr. Stevenson’s actual view is that sometimes, when a man asserts that it
was right of Brutus to stab Caesar, the sense of his words is (roughly) much the
same as if he had said “I approve of Brutus’ action: do approve of it too!” the former
clause giving the cognitive meaning, the latter the emotive. But why should he not
say instead, that the sense of the man’s words is merely “Do approve of Brutus’
stabbing of Caesar!”—an imperative, which has absolutely no cognitive meaning,
in the sense I have tried to explain? If this were so, the man might perfectly well be
implying that he approved of Brutus’ action, though he would not be saying so, and
would be asserting nothing whatever, that might be true or false, except, perhaps,
that Brutus did stab Caesar. (1968: 542)
On the matter of this notion of “implying,” Moore immediately comments on

the crucial distinction between implying and asserting:
There seems to me nothing mysterious about this sense of “imply,” in which if

you assert that you went to the pictures last Tuesday, you imply, though you don’t
assert, that you believe or know that you did; and in which, if you assert that
Brutus’ action was right, you imply, but don’t assert, that you approve of Brutus’
action. (542)
225
226 APPENDIX 1
But he goes further to give a commonsense explanation of the implication in question:
That you do imply this proposition about your present attitude, although it . . .
does not follow from what you assert, simply arises from the fact, which we all
learn by experience, that in the immense majority of cases a man who makes
such an assertion as this does believe or know what he is asserting: lying, though
common enough, is vastly exceptional. And this is why to say such a thing as “I
went to the pictures last Tuesday, but I don’t believe that I did” is a perfectly
absurd thing to say, although what is asserted is something which is perfectly
possible logically: it is perfectly possible that you did go to the pictures and yet
you do not believe that you did. . . . Your saying that you did, does imply [in a
sense other than logical implication] . . . that you believe you did; and this is
why “I went, but I don’t believe that I did” is an absurd thing to say. 1
1Although this was Moore’s claim in 1942, it is clear from an only recently published manuscript in
the Moore archive in the University Library, Cambridge University, U.K., tentatively dated by Tho-
mas Baldwin (Moore 1993: 207–12) as 1944, that Moore was admirably explicit about both his data
and the strength of his purported explanation of them.
(a) “I start from this: that it’s perfectly absurd or nonsensical to say such things as ‘I don’t believe
it’s raining, but as a matter of fact it is’ . . . I’m just assuming that it is absurd or nonsensical to say such
things. But I want it noted that there is nothing nonsensical about merely saying these words. I’ve just
said them; but I’ve not said anything nonsensical. And W. [Wittgenstein] pointed out another proof that
there isn’t. He pointed out (I think) that there’s nothing nonsensical in saying ‘It’s quite possible that
though I don’t believe it’s raining, yet as a matter of fact it really is’ or ‘If I don’t believe it’s raining, but
as a matter of fact it really is, then I am mistaken in my belief’. In all these cases the very same identical
words are said, but they are said in a context with other words, so that there is nothing nonsensical about
them.” Wittgenstein’s point was later called “the Frege Point” by Peter Geach (1972b).
(b) Moore (1993: 207) concludes, “It’s absurd to say them in the sort of way in which people
utter sentences, when they are using these sentences to assert the proposition which these sentences
express. I will call this ‘saying them assertively’. I don’t want to say that to utter sentences assertively
is the same thing as making an assertion.”
This last, interesting distinction is not elaborated upon by Moore. One application of it would be
to Geach’s (1972b) Frege Point. In asserting the conjunction P & Q one utters Q assertively, but one
does not assert Q (pro Geach 1972b and pace Stalnaker 1974). I also take it that utterances that purport
to be assertions, in that they are uttered in circumstances typical of assertions with intonation and stress
appropriate to assertion, including pretended assertions of the sort to be heard onstage in a play, ex-
hibit the same phonetic, syntactic, and semantic properties as “real” assertions except in the latter case
actual truth or falsity. Any explanation of the oddity of the Moore utterance-type should explain its
oddity regardless of its being a “real” assertion.
(c) After some discussion of the difference that tense makes and choice of pronoun makes, Moore
reformulates his “paradox.” He takes it that when he asserts I don’t believe it’s raining, but as a matter
of fact it is and another asserts of him Moore doesn’t believe that it’s raining, but as a matter of fact it
is, the same proposition is expressed. But why should one expression of the proposition be absurd and
another expression of it not be absurd? That seems paradoxical. Moreover, it is possible for the propo-
sition in question to be true. And that, Moore (1993: 209) remarks, is paradoxical: “It is a paradox that
it should be perfectly absurd to utter assertively words of which the meaning is something which may
quite well be true.”
These characterizations of what is paradoxical about Moore’s statements are much more interest-
ing than the blank observation that it is absurd to utter a Moore sentence assertively.
(d) Finally, Moore (1993: 211) makes the following remarkable admission: “I think the things
. . . wouldn’t be absurd to say, unless it was true that by saying a thing assertively we imply that we
believe it. But I don’t know that this fully explains why it is absurd [my emphasis].” How right he was!
ON MOORE ’S TERM ‘ IMPLY ’ 227
Historically it is of interest to see that Moore himself was never in doubt that
this notion of “implication” was not logical implication and that it is a “default” in-
ference justified by the psychological facts of speech in an “immense majority of
cases.” But given Moore’s parallel between belief and approval in the factual and
ethical cases, and his willingness to use the word ‘imply’ in both cases, he might
have asked why it is not linguistically absurd to assert It was right of Brutus to stab
Caesar, but I don’t approve of it and it is linguistically absurd to assert ?I went to the
pictures but I don’t believe I did. For surely, Stevenson and Moore would admit, it is
a fact, which we all learn from experience, that in the immense majority of cases a
man who makes such a claim as this (Brutus’s action was right) does approve of what
he is claiming.
Moore’s explanation requires that the factual and ethical cases should be on
all linguistic fours, but they clearly are not. Moore’s explanation is a complete non-
starter. In 1944, on Moore’s return to England from the United States after the Battle
of the Atlantic had been won by the Allies, Ludwig Wittgenstein first encounters
Moore’s “paradox” in Moore’s lecture to the Moral Sciences Club. Wittgenstein
writes to Moore, with some detectable impatience, in October 1944, that the para-
dox is an “‘absurdity’ which is in fact something similar to a contradiction, though
it isn’t one. . . . You have said something about the logic of assertion . . . [that] it
makes no sense to assert ‘p is the case, and I don’t believe that p is the case’. This
assertion has to be ruled out . . . just as a contradiction is” (Wittgenstein 1974: 177).
In his 1942 essay “The Notion of Analysis in Moore’s Philosophy,” the Ameri-
can logician C. H. Langford mentions the “paradox” in the form I believe P but not
P (B1st per [P] & ¬P):
I want to cite an example which is due to A.M. MacIver and which is worth repeat-
ing on its own account.[a] Suppose someone to remark: “He thinks that he has been
to Grantchester but he has not.” The person referred to may entertain this proposi-
tion as an hypothesis. But suppose he actually asserts the proposition: “I think that
I have been to Grantchester but I have not.” This sounds self-contradictory, and the
reason is that he will actually be saying that he thinks that he has been to Grantchester,
whereas the but-clause in the indicative mood will signify or mean pragmatically
[b] that he does not think so.[c] (Langford 1968: 333)
aSee Analysis, Vol. 5 (1937–38), 43–50, and Journal of Symbolic Logic, Vol. 3 (1938), 158.
b[Langford (1968: 332) wrote, “We may call what a man does not state, but intends his lin-
guistic behavior to signify, his pragmatical meaning, and we may distinguish this from the
sense of his words, which is the proposition expressed by them. (This term has been used by
Charles Morris in the same or a similar sense.)”
cThe course of Moore’s argument in “A Defence of Common Sense” will be clearer if this
distinction between formal and pragmatical contradiction is carefully observed. For in say-
ing that certain philosophers contradict themselves when they assert, in effect, “There have
been many other human beings beside myself, and none of them, including myself, has ever
known of the existence of other human beings,” Moore is not holding that a formal contra-
diction can be derived from the sense of those words, but only that the pragmatical meaning
of such an assertion is incompatible with its literal meaning.
228 APPENDIX 1
The peculiarity of Langford’s analysis is that he is diagnosing the “apparent

self-contradiction” of a statement of the form B1st per [P] & ¬ P by the hypothesis
that in saying ¬ P the speaker “pragmatically means” ¬B1st per [P]. There is
no recognition from Langford of the need to discuss the obvious scope distinction
between the wide-scope negative belief proposition ¬B1st per [P] that is allegedly
the speaker’s pragmatical meaning and the narrow-scope negative proposition
B1st per [¬P]. Without defense, Langford chooses the wide-scope negative for-
mula in order to explain the apparent self-contradiction by an actual logical con-
tradiction between the sense of ‘I believe I have been to Grantchester’ and the
speaker’s pragmatical meaning of asserting ‘I have not’, or “I don’t believe I have.”2
Unfortunately, he thinks it is self-evident that such a logical inconsistency between
these different sorts of “meaning” suffices to explain the apparent “oddness” of
asserting the Moore sentence. No theory providing the explanatory connection is
offered at all.
George Lakoff (1975: 264–65) first discusses Moore’s “paradox” in the form
P & B[¬P]. Lakoff’s solution is a version of the Speech Act Sincerity Condition and
Rationality Assumption solution: In addition to the Sincerity Condition If x (sincerely)
asserts P, x believes P, Lakoff makes three rationality assumptions: (a) belief
conjunction-elimination: B[P & Q] → B[P] & B[Q], (b) belief-reduction: B[B[P]] →
B[P], and the consistency assumption (c): B[P] → ¬B[¬P]. He uses them to deduce
the contradiction B[¬P] & ¬B[¬P] from the assumption of a sincere, rational assertion
of a Moore sentence. By explaining P & B[¬P] rather than Moore’s form P &
¬B[P], Lakoff avoids the dubious S4-like axiom: B[P] → B[B[P]], which would other-
wise be required by his type of explanation, but Lakoff’s form does not pose the same
problem as Moore’s (1968: 543) original 1942 form.3
2A plausible reconstruction of Langford’s reasoning is possible: à la Moore, in asserting ¬P the speaker
S pragmatically means B1st per [¬P]. Then consistency of belief requires the axiom: B[¬P] → ¬B[¬¬P],
while “modest”—that is, classically and intuitionistically acceptable—Belief Double Negation yields
the axiom: B[P] → B[¬¬P], from which there follows the theorem: B[¬P] → ¬B[P]. If pragmatic
meaning is preserved under logical consequence modulo the axioms of rational belief, then if the speaker
pragmatically means B[¬P], he means ¬B[P]. Such consequences are plausible if a speaker in-
tends his utterings to signify the logical consequences (or perhaps, more restrictedly, the direct logical
consequences [Atlas 1991a: 137]) modulo the axioms of rational belief of what he intends his utterings
to signify.
3The derivation is:
1. x sincerely asserts P & B[¬P] (assumption)
2. B[P & B[¬P] (sincerity 1.)
3. B[P] & B[B[¬P]] (belief & elimination 2.)
4. B[P] (simplification 3.)
5. B[B[¬P] (simplification 3.)
6. B[¬P] (belief reduction 5.)
7. ¬B[¬P] (belief consistency 4.)
8. B[¬P] & ¬B[¬P] [& introduction 6.7.]
By reductio ad absurdum, if the principles of rational belief hold, x cannot sincerely assert: P & B[¬P].
ON MOORE ’S TERM ‘ IMPLY ’ 229
Lakoff (1975: 265) goes on to claim that the same assumptions will allow a
deduction of a contradiction from the assumption of a sincere assertion of the origi-
nal Moore sentence form P & ¬B[P] to explain the latter’s “oddness,” but this
claim is incorrect. To deduce a contradiction in the same manner, he needs
to assume in addition the S4-like iteration axiom that he (1975: 264) agrees is
dubious.4
Langford and Lakoff explain the alleged peculiarity of the assertion by deduc-
tions of a logical contradiction of two sorts. In Langford’s case, precisely because
the original assertion sounds self-contradictory to him, he claims that a logical in-
consistency can be derived from the conjunction of the speaker’s pragmatical mean-
ing and the literal sense. In Lakoff’s case, because the assertion sounds odd to him,
he claims that a sincere rational assertion of the Moore sentence is logically incon-
sistent with the principles of rational belief and of speech-act theory. It seems to me
an open question whether the Moore statement does sound self-contradictory, and a
closed question that even if it seemed “odd,” the alleged reductio of the rational, sin-
cere assertion of the Moore’s paradox sentence could not explain the “oddness” of
“asserting” a Moore’s paradox sentence. Of course, traditionally, that the statement
sounds “odd” is the linguistic datum to be explained.
Another difficulty with Langford’s explanation is that there are a denumerable
number of self-contradictory statements that are perfectly felicitous, so that the logi-
cal inconsistency of the total signification of a statement cannot be sufficient for its
assertoric “oddness.” In the case of Lakoff’s (1975) explanation, since neither a
speaker’s insincerity nor the failure of the speaker’s beliefs to conform to Lakoff’s
three axioms of rational belief could be explanatory of the Moore statement’s sounding
“odd,” it is bizarre to think that the “impossibility” of the “rational,” sincere asser-
tion of a Moore sentence could explain the “odd” sound of uttering a Moore sen-
tence assertively (as Moore 1993: 207 would put it), even if the impossibility of a
sincere, “rational” assertion explains why either the utterance is insincere, “irratio-
nal,” or not an assertion—oddities all, but not an oddity of uttering a sentence asser-
tively. None of those consequences would explain the particular “oddity” of uttering
4The derivation is:

1. x sincerely asserts P & ¬B[P] (assumption)
2. B[P & ¬B[P]] (sincerity 1.)
3. B[P] & B[¬B[P]] (belief and elimination 2.)
4. B[P] (simplification 3.)
5. B[¬B[P]] (simplification 3.)
6. B[B[P]] (belief iteration: B[P] → B[B[P]] 4.)
7. ¬B[¬¬B[P]] (belief consistency 5.)
8. ¬B[B[P]] (modest Belief Double negation: B[B[P]] → B[¬¬B[P]] 7.)
9. B[B[P]] & ¬B[B[P]] (& introduction 6.8.)
So by reductio ad absurdum, if the principles of rational belief hold, x cannot sincerely assert: P &
¬B[P]. But this derivation requires belief iteration, contrary to Lakoff’s claim. Without going into an
extensive discussion, to see that belief iteration is at least controversial, note that David Rosenthal (1993)
adopts belief iteration as a criterion for conscious belief states.
230 APPENDIX 1
the Moore sentence assertively.5 The oddity remains even if the Moore utterance
merely purports to be an assertion; furthermore, irrationality of belief and insincer-
ity of belief are psychological properties of the asserter, not linguistic properties of
the utterance, so it is a category error to think that they can explain the linguistic
oddity of the sentence uttered assertively.6
5In response to this claim, Paul Benacerraf (personal communication) has remarked, correctly I believe,
that an adequate account must appeal to principles that every speaker/hearer can be expected to “know,”
in the sense in which we know our language and its uses.
6See note 1(b).
Appendix 2
On Hitzeman (1992) on ‘Almost’
I offer two arguments reductio ad absurdum to show that x almost F’d does not en-
tail x did not F. Suppose for adjectives, determiners, or verbs F, the statement schema
A(almost F) entails the schema A(not F). (a) Then Almost all swans are almost white
entails Almost all swans are not white, which entails Not all swans are not white,
which entails Some swans are white. But the proposition Some swans are white is
just what a speaker who asserts Almost all swans are almost white chooses the word
almost to avoid conveying. (b) It is intuitively evident that if there were no white
swans, it could still be true that almost all swans were almost white. However, if
A(almost F) entailed A(not F), and there were no white swans, it would be false that
almost all swans were almost white, as we have just seen in (a). Conclusion: A(almost
F) does not entail A(not F). The simple entailment explanation of the ordering of
almost and not quite is the wrong explanation. What is the relationship between al-
most and not quite?
Hitzeman (1992) believes that the transition in (a) from Almost all swans are
almost white to the entailed Almost all swans are not white requires that I assume
(incorrectly in her view) that almost all swans is an upward-entailing (i.e., upward
monotonic in the sense of J. Barwise and R. Cooper 1981) generalized quantifier noun
phrase. Actually, what I imputed to the entailment theorist was the claim that x is
almost white analytically entails x is not white, for free x, or, equivalently (in Classi-
cal but not Intuitionistic logic), that ∀x(x is almost white) analytically entails ∀x(x is
not white). On what I take to be Hitzeman’s assumption—viz., that {x: x is almost
white} ⊆ {x: x is not white}—her claim, that the hypothesis that almost all swans is
upward-entailing is a necessary condition for the correctness of my (a), is false (even
231
232 APPENDIX 2
though, on her, admittedly plausible, assumption about the subset relation between
the extensions of the predicates ‘x is almost white’ and ‘x is not white’, the hypoth-
esis is a sufficient condition for my claim).
For example, for the non-monotonic generalized quantifier Only Tom (Atlas
1996b), the entailment theorist about almost would certainly be committed to the claim
that Only Tom was almost drunk entails Only Tom was not drunk. The status of this
claim as an entailment claim is unaffected by the fact that only Tom is not an upward-
entailing generalized quantifier. In fact, on the traditional, and incorrect, view that
only Tom is a downward monotonic generalized quantifier, the traditional entailment
theorist would be caught in a logical cleft stick: he would want to claim that ‘x is
almost drunk’ analytically entails ‘x is not drunk’, and so necessarily {x: x is almost
drunk} ⊆ {x: x is not drunk}, and he would want to claim that the substitution of the
superset Not Drunk for its set Almost Drunk in an allegedly downward-entailing
quantifier only Tom necessarily preserves truth—which is an absurd conjunction of
claims. So the theorist cannot both be traditional about the downward monotonicity
of only Tom and simultaneously hold that almost F entails not F. (This result, alone,
is worth the price of admission, since both views are standardly held, and they are
logically inconsistent. No semantic theorist can consistently hold that only Tom is
downward monotonic and that almost F entails not F.) The solution to the traditional
difficulty is that neither part of the traditional position is correct: only Tom is non-
monotonic, and almost F does not entail not F. Hitzeman’s argument, which con-
fuses a sufficient with a necessary condition, does not undermine my (a).
Since Hitzeman (1992) thinks that my reductio argument requires the assump-
tion that almost all N is upward monotonic, she attempts to defang my argument by
arguing that almost all N is not upward monotonic. She offers an alleged
counterexample to the plausible, upward monotonicity of almost all N, plausible even
to Hitzeman because the following argument (A) seems valid to her: (A) Almost all
dogs run, Every individual that runs moves ∴ Almost all dogs move. Her alleged
counterexample is the alleged invalidity of the argument: (B) Almost all men are
fathers, Every individual that is a (biological) father is male ∴ Almost all men are
male. But how one could think that the premises of this argument (B) might be true
while simultaneously its conclusion not true is quite beyond me. If it is true that al-
most all men are fathers, and that every individual that is a (biological) father is male,
then it is surely true that almost all men are male, even though it is also true that all
(biological, non-transsexual) men are male. So what is Hitzeman’s objection to ar-
gument (B)—that is, to the upward monontonicity of almost all N? She asserts that
the concluding sentence Almost all men are male is “strange,” but not “strange” in
the way she thinks that the analytical truth All men are male is “strange.” She thinks
the former is “strange” because it (allegedly) entails Some man is not male!
But for her to argue in this way against the validity of argument (B) fails for two
different reasons: first, because the issue is not strangeness of the concluding sen-
tence of (B) but its truth in any model in which the premises are true and, second,
because even if “strange” meant ‘false’, her argument overtly begs the question against
my view that almost does not entail not. She argues: almost entails not; so Almost all
men are male is “strange,” thus necessarily false; so the (alleged) invalidity of argu-
ment (B) shows that Atlas cannot assume that Almost all N is an upward monotonic
ON HITZEMAN ON ‘ ALMOST ’ 233
quantifier. This argument just blatantly begs the question against my position that
almost does not entail not. And, as I argued above, she (falsely) thinks that upward
monotonicity is necessary (by contrast with sufficient) for the first claim of my re-
ductio argument against the entailment of not by almost to be sound.
Hitzeman (1992: 236) has been kind enough to say that she “believe[s] that the
most interesting argument against the entailment hypothesis is . . . due to Atlas
(1984a)”—that is, the argument in question here. Since the semantically most so-
phisticated attack on the pragmatic position of Sadock (1981) and me (Atlas 1984a)
that I am defending here is Hitzeman’s (1992), involving as it does an attempt to
defeat my argument by arguing (incorrectly) for the non-monotonicity of almost all
N, I find it a highly instructive argument.
According to criteria for upward monotonic generalized quantifier noun phrases
analogous to the ones discussed in Zwarts (1996, 1998) and Atlas (1996b: 282) for
downward monotonic generalized noun phrases, we have the following obvious cri-
teria for upward monotonicity of the NP:
a. NP(VP1 and VP2) NPVP1 & NPVP2

b. NP(VP1 and VP2) NPVP1 v NPVP2
c. NPVP1 v NPVP2 NP(VP1 or VP2)
d. NP(VP1 and VP2) NPVP1
e. NP(VP1) NP(VP1 or VP2)
where and entails the Boolean intersection interpretation of the extensions of the Verb
Phrases, and or means nothing more restrictive than the Boolean union of the exten-
sions of the verb phrases. These criteria (a) to (e) have obviously correct instantiations
for almost all N, for example, (a) Almost all college boys smoke and drink Almost
all college boys smoke & Almost all college boys drink. Thus the evidence is power-
ful that almost all N is upward monotonic, contrary to Hitzeman’s (1992) claim and
consistent with my argument against almost F entailing not F.
234 APPENDIX 3
Appendix 3
The Semantics and Pragmatics of

Cleft Statements
A unified theory?
Linguists who recognize that statements may have the same truth conditions but dif-
ferent semantic representations have suggested that the semantic representations of (1a),
(1b), and (1c) are (2a), (2b), and (2c), respectively (see Gazdar 1979a: 124–25):1
(1) a. Sam wants Fido.

b. What Sam wants is Fido.
c. It is Sam who wants Fido.
(2) a. Wants(Sam, Fido)

b. λx(Wants(Sam,x))(Fido)
c. λx(Wants(x,Fido))(Sam)
But with normal statement stress on the last word of (1a), Fido has “unmarked
information focus” (Halliday 1967), which is consistent with the general convention
that old information precede new information. Similarly, the pseudo-cleft (1b) con-
This appendix is in part a revised version of Atlas and Levinson (1981: 16–18, 50–57) and appears
with the permission of Academic Press.
1For a discussion of lamba-abstraction, see Rudolf Carnap (1958), pp. 129–31. See also L. T. F. Gamut
(1991), pp. 102–16 and Peter B. Andrews (1986).
234
CLEFT STATEMENTS 235
forms to this convention. The “focus” (Chomsky 1972b) of (1b) is Fido; its “presup-
position” is Sam wants something. The same analysis would give for (1c) the
“focus” Sam and the “presupposition” Something wants Fido. In other words, the
cleft rhetorically parallel to the pseudo-cleft and the simple sentence is not (1c) as
Gazdar and the order of items in surface structure seem to suggest, but (3), which
contravenes the convention that old information precede new information. Gazdar’s
logical form for (3) would be (4):
(3) It is Fido that Sam wants.
(4) λx(Wants(Sam,x))(Fido)
The “focus” of (1b) and of (3) is Fido; the “presupposition” of (1b) and (3) is
Sam wants something. In Gazdar’s logical forms for clefts and pseudo-clefts, the
“focus” corresponds to the logical subject; the “presupposition” corresponds to the
logical predicate. And the logical form of the pseudo-cleft What Sam wants is Fido
is identical to that of the cleft It is Fido that Sam wants; that is, (2b) = (4).
I take it that a statement with contrastive stress SAM wants Fido has Sam as
“focus” and Something wants Fido as “presupposition.” So the contrastively stressed
(5) rhetorically parallels (1c) and (6). The Gazdar logical form for the latter two
is (2c):
(5) SAM wants Fido.
(2a) Wants(Sam,Fido)
(1c) It is Sam who wants Fido.
(6) Who wants Fido is Sam.
(2c) λx(Wants(x,Fido))(Sam)
Similarly, the normally stressed statement (1a) and the contrastively stressed (7)
parallel (3) and (1b). The Gazdar logical form for the latter two is (2b):
(1a) Sam wants Fido.
(7) Sam wants FIDO.
(3) It is Fido that Sam wants.
(1b) What Sam wants is Fido.
(2b) λx(Wants(Sam,x))(Fido)
236 APPENDIX 3
If, like Gazdar, one were to assume that (8) parallels (9), one might also assume
that (1c) parallels (1a):
(8) It was John who went.
(9) John went.
(1c) It is Sam who wants Fido.
(2c) λx(Wants(x,Fido))(Sam)
(1a) Sam wants Fido.
But, as we have just seen, statements (1a) and (1c) are not parallel at all. The order of
items in the surface structures of (1c) and (1a) is misleading. Misled by surface struc-
ture one might assume that the “logical subject” of both sentences is Sam. On this
assumption, in (1c) the “focus” corresponds to the logical subject. But in the nor-
mally stressed simple sentence (1a), the normal “focus” is Fido; the “focus” there
does not correspond to the logical subject.
Thus we shall be guided by “focus” rather than by surface order. I reject the
suggestion that statement (8) pairs with the normal stressed (9); it pairs with the con-
trastively stressed (10).2
(10) JOHN went.
I believe that the identification of logical subject and “focus” in (1c) is just as
mistaken as the same identification in the normally stressed (1a) would be. It seems
to me that the semantic representations offered by Gazdar and others fail to make the
semantic and pragmatic data cohere. (For a discussion of the pragmatics of clefts,
see Prince 1978.) One of the aims in this section is to sketch an account that will
unify the semantics and pragmatics.
The logical form of clefts and its explanatory value
One attraction of Grice’s views has always been its semantical conservatism. The
Fregean notion that “sense” is truth conditions, the identification of a set of English
expressions, which are frustratingly resistant to systematization, with the logical
constants, which are our paradigm of semantic systematization, and the scrupulous
2IndependentlyHalvorsen (1978: 6) has argued against A. Akmajian (1970) that It was himself that
John wanted Mary to describe pairs with John wanted Mary to describe HIMSELF, not with *John
wanted Mary to describe himself.
adherence to a policy of austerity in positing senses have contributed to theoretical

simplicity in our theory of language. Simplicity is indeed a virtue of theories:
simplemindedness is not. There has been a regrettable temptation to adopt a logical
primitivism when theorizing about conversational inference. The canonical languages
of our logical theories are constructed to achieve pellucidity, but a certain measure
of complexity is compatible with, indeed on my view required by, a satisfactory
use of truth-conditional semantics within a pragmatic theory.
Logical primitivism would take the familiar (but incorrect) claim that (11) and
(12) have the same truth conditions to imply (fallaciously) that (11) and (12) have
the same logical form (13). Gazdar’s less primitive (but still incorrect) suggestion
would give (11) the logical form (14).
(11) It was Max that Jane kissed.
(12) Jane kissed Max.
(13) Kiss(Jane,Max)
(14) λx(Kiss(Jane,x))(Max)
In adopting a logical form, I am locating the sentence in a network of entailment

relations that is described by the particular logical theory I am employing. But I am
also interested in hypothesizing logical forms that are explanatory—that account for
entailment relations by exhibiting semantically significant structure in the sentence.
Such an account will begin to explain how the relations between the parts of the
sentence contribute to the meaning of the whole. It will illuminate the similarities
and differences between related sentences. It will (on the standard view) provide the
extensional sentence “meaning” upon which inferential mechanisms must operate
to yield the understanding of an utterance. The assignment of logical form to a sen-
tence is not only relative to the logical theory employed, it is relative to the compre-
hensive theory in which logical forms have an explanatory place. Indeed, even the
pragmatic features of the sentence, its use in the language, can in principle bear on
the assignment of logical form, especially if the resulting form increases the overall
coherence and explanatory power of the theory.3
The logical forms in (13) and (14) are logically equivalent, but they are distinct:
whereas (13) has a primitive two-place predicate symbol true of Jane and Max, (14)
has a complex one-place predicate symbol true of Max. Whereas (13) expresses a
relation between Jane and Max—it is “about” the pair <Jane, Max>—(14) expresses
a property of Max: it is “about” him. And it is precisely here that the flaws of (14)
become obvious.
If one recalls the semantical similarities between clefts and pseudo-clefts, the
pseudo-cleft (15) will highlight the properties of (11):
3My indebtedness to the writing and teaching of Donald Davidson shows itself here, as does my diver-
gence from his views (cf. Davidson 1967, 1970).

238 APPENDIX 3
(15) What Jane kissed was Max.
This sentence is “about” what/whom Jane kissed, which is specified or identified as

Max. Likewise (11) is actually “about” whom Jane kissed, which is then specified or
identified as Max.
It was argued in Atlas and Levinson (1981) that clefts exhibit the following
behavior:
(I) It was Max that Jane kissed

a. entails Jane kissed Max; the latter does not entail the former
b. entails Jane kissed someone
c. entails but does not “presuppose” Jane kissed (exactly) one person
d. is “about” what/whom Jane kissed.
(II) It wasn’t Max that Jane kissed

a. entails Jane didn’t kiss Max; the latter does not entail the former
b. “presupposes” or its use implicates Jane kissed someone
c. does not “presuppose” Jane kissed (exactly) one person;
d. is “about” what/whom Jane kissed.
The logical forms (13) and (14), and their negations, obviously cannot satisfy these
conditions. Is there any logical form that will meet all these conditions and in the
process yield the correct pragmatic inferences? The answer is yes; it is just a more
complex logical form than is typically suggested.
The correct logical form for (16a) It was Max that Jane kissed involves λ ab-
straction (Carnap 1958: 129–31) to formulate a complex one-place predicate-sym-
bol and my “collection term operator” γ to formulate a singular term γxA(x) for a
collective. The logical form (16b) of (16a) has precisely the properties described in
(I). It may be paraphrased in English by (16c).
(16) a. It was Max that Jane kissed.

b. λx(x = Max)(γxKiss(Jane,x))
c. A group of individuals kissed by Jane is identical to Max.
If ûA(u) = {a}, that is, the extension of A(u) is just one object a, Hilbert’s
(1927) term ∈uA(u) designates the descriptum of |uA(u). If the extension of
A(u) is larger, ∈is a choice function; ∈uA(u) designates some one of the indi-
viduals in the extension (but we do not know which). The expression ∈uA(u) may
be paraphrased by a u such that if anything has A, u has A. The basic axioms govern-
ing the use of the term are ∃uA(u) ↔ A(∈uA(u)) and ∀uA(u) ↔ A(∈u¬A(u)).
Thus the selection operator allows one to make a statement the force of which is purely
existential while employing a designating singular term.
Paul Ziff, Jaakko Hintikka (1973), and I (Atlas 1972) independently have re-
marked on the need for ∈-terms in giving the logical forms for sentences of a natural
language. Ziff and Hintikka noted it for coreference phenomena, as in ‘John wants
to catch a fish and eat it for supper’ (Hintikka 1973). I (Atlas 1972) made a system-
atic use of the ∈-term in giving a theory of truth for English: the problem was to
characterize the circumstances in which I met a man who wrote “Lolita”; therefore,
the man I met wrote “Lolita” would be an acceptable inference. Once again, the heart
of the matter is coreference—the coreference of event-terms.
The indeterminateness of Hilbert’s ∈-term makes it attractive to some (for ex-
ample, R. M. Martin 1979: 214), as a paraphrase of indefinite plural noun phrases.
Although I am in agreement with Martin’s suggestions in some respects, his claim
that Hilbert’s ∈-term (that is, the selection description) correctly formalizes indefi-
nite plural noun phrases seems to me mistaken. It also seems incorrect to define con-
textually the ∈-term as Martin (1958: 55; 1979: 214) does. Attributing the definition
to Frederic B. Fitch, Martin (1958) contextually defines the ∈-term as follows:
B(∈xA(x)) = df ∃xA(x) & ∀x[A(x) → B(x)]. The second conjunct of the definiens
seems too inclusive to be an accurate analysis of the definiendum. (For discussion of
the ∈-term, see A. C. Leisenring 1969.) But the condition in the definiens does cap-
ture an important concept in mathematics and in linguistics, which we can use in the
analysis of collective terms and so of clefts. Just as the | operator attaches to a for-
mula A to produce an individual term |xA(x), so my γ operator attaches to a
formula A to produce a collective term γxA(x). By a collective term I mean one
that denotes a group. For example, the plural noun phrase the boys may be used as a
collective term in The boys (collectively; that is, a group of boys) are at the party;
the sentence is true if and only if there are boys and every boy (in the group) is at the
party. I contextually define the formula B(γxA(x)) by ∃xA(x) & ∀x[A(x) → B(x)].
The γ operator is indifferent to the distinction between singular and plural:
γxA(x) is consistent with both the singular the A and plural the As and so
captures a linguistic feature of collective nouns. Collective nouns in English are some-
times grammatically plural (for example, cattle, clergy), sometimes grammatically
singular (for example, furniture), and sometimes either (for example, family).
A collective noun can designate a group collectively, and so behave as a denot-
ing term, or designate a group distributively, for example, in the United States, The
administration, who have . . . are . . . or in the U.K., The government, who have . . .
are . . . ; plural count nouns, like the boys in our example above, can mimic the be-
havior of collective nouns. Martin’s definiens, which I accept as roughly correct for
γxA(x), though not for ∈xA(x), captures the distributive use of the collective
term in the truth conditions for sentences containing it.
It is also linguistically possible for a collective noun, and so even for plural count
nouns, to designate a group of one as well as a more normal group of more than one.
It is a virtue of my γ-operator that it allows this possibility. The cleft sentence It was
Max that Jane kissed is “about” the collectivity, not excluding a group of one, that
Jane kissed. It can easily be demonstrated that the logical form for clefts that em-
ploys our γ-operator explains precisely those characteristics of clefts that I have ar-
gued in Atlas and Levinson (1981) are properly attributable to them.
There is one final observation supporting formalization of collectivity by our
γ-operator. Collective nouns like flock, herd, forest, and group can act as “sortal clas-
sifiers” when they are attached to count nouns—for example, flock of sheep. And it
has been claimed that “sortal classifiers” have properties in common with determin-
ers (Lyons 1977: 464). If this is correct, there is a suggestive analogy between a
240 APPENDIX 3
determiner like the and a “sortal classifier” like group. Likewise, there is an analogy
between the |-operator and our γ-operator, which we have exploited.
The advantages of this semantic representation are manifold. First, it explains
the data in (I). It is easy to see that the entailment relation is as claimed in (Ia), as
λx(x = Max)(γxA(x)) A(Max); but A(Max) λx(x = Max)(γxA(x)). Condition (Ib)
then follows immediately from (Ia). It is easy to prove that λx(x = Max)(γxA(x))
E!|xA(x), so part of (Ic) is explained. It may be worth remarking on the reasons for
this entailment.
The formula λx(x = Max)(γxA(x)) is definitionally equivalent to ∃xA(x) &
∀y(A(y) → y = Max), from which it follows immediately that Jane kissed (exactly)
one person. The sentence Jane kissed someone follows from the contribution of that
Jane kissed to It was Max that Jane kissed. But the proposition Jane kissed (exactly)
one person follows because of the contingent fact that the specification in the (sur-
face) main-clause focus constituent of It was Max that Jane kissed lists but one item:
namely, Max. That “asserted” fact adds Jane kissed (at most) one person to the “pre-
supposed” Jane kissed someone to give Jane kissed (exactly) one person.
The felicitousness of the discourse (17) was inexplicable to Halvorsen (1978).
(17) McX: Was it Mart and Rick that Laura kissed?

McY: She kissed only John.
It is explained without the machinery of “ordered implications” (Wilson and Sperber

1979). What is being contradicted is not a “presupposition” but an assertion, and there
is no problem of felicitousness.
Furthermore, the fact that lists can be of any finite length receives a natural ac-
commodation. Lists are sequences (or vectors), and can be the values of individual
variables. The expression Kiss can be treated as a “multigrade” predicate Kiss, so a
sequence of any length (including infinite length) can be one of its arguments. The
sentence schema It was N1, N2, . . . Ni–1, and Ni that Jane kissed specifies a sequence
[sa]ia = 1 of i terms where sa = Na. The logical form is λx(x = [sa]ia = 1)(γxKiss( j,x)) with
x ranging over i-term sequences of individuals. If we wish to accommodate any num-
ber of terms, we may let sequences be infinite and identify the subsequence [sa]ia=1,
sa = Na, with the sequence [ta]∞a = 1 such that ta = sa, a = 1,2, . . . , i, and ta = D, i + 1
≤ a, where D is the domain of individuals.
The logical form preserves the intuition that the cleft is a property rather than a
relation statement, thereby showing that it is “about” its logical subject and remov-
ing the incoherence between the semantics and pragmatics noted in subsection 1. Thus
it explains datum (Id). It is “about” what/whom Jane kissed, that is, “about”
γxKiss(Jane,x).4
4If you believe in a uniqueness implicature, so that you believe that (Ic) should read ‘“presupposes”
Jane kissed (exactly) one person’, you can be satisfied by the logical form λx(x = Max) (|xKiss(Jane,x)).
The logically equivalent form |xKiss(Jane,x) = Max resembles an underlying syntactic structure
adopted for clefts by Harries-Delisle (1978). Harries-Delisle produces syntactical arguments to show
that the underlying structure of the “equational sentence” The one whom Jane kissed is Max also underlies
It was Max that Jane kissed, where the one is a neutral head noun marked for person (third) and in
English at least, for number and for humanness. Logically the sentence is recognized to involve
Classical Gricean pragmatics posits the external negation (18b) for the negative
sentence (18a).5 The Atlas-Levinson neo-Gricean revision of Gricean pragmatics,
by the Principle of Informativeness, yields as a generalized conversational inferendum
of (18a) the internal negation (18c). On either understanding of (18a), (IId) is explained.
(18) a. It wasn’t Max that Jane kissed.

b. ¬λx(x = Max)(γxKiss(Jane,x))
c. λx(¬(x = Max))(γxKiss(Jane,x))
Condition (IIc) is explained, as the implicatum (18c) does not entail Jane kissed (ex-
actly) one person. This completes the explanation of (Ic). The implicatum (18c)
“directly entails” (Atlas 1991), and so (18a) “quasi-entails” (see chapter 5) by the
composition of the generalized conversational inferendum relation with the direct
entailment relation, Jane kissed someone.6 Thus (IIb) is explained.
And finally, the implicatum (18c) entails Jane didn’t kiss Max, but the converse
is not the case. So (IIa) is explained.
My logical form for clefts, which incorporates “topic” into its singular term as
logical subject, confirms Strawson’s (1954, 1964), and also H. P. Grice’s and G. J.
Warnock’s, intuitions about the filling-in of truth-value gaps, explains the anomaly
of It’s NO one that Jane kissed, as Ivan Sag observed (personal communication),
and explains the falsity rather than truth-value-lessness of It’s the king of France that
is bald, a datum noted but not explained by J. McCawley (1981: 241).
Further remarks on the analysis
L. R. Horn (1981a) has discussed Halvorsen (1978) and Atlas and Levinson (1981)
on the semantics of clefts. In that essay, after agreeing, mostly, with the criticisms
identity, but grammatically the underlying structure is subject/predicate, and there are various compli-
cations about the occurrence of the copula in the underlying structure. We begin from roughly the same
semantic intuitions. It is interesting to discover syntactical arguments in support of an underlying struc-
ture whose basic features at least approximate those of the logical form that we have posited.
The choice of the collective term as a logical subject is consistent with the Focal Noun Phrase
Limitation Principle discussed in Atlas (1991a, 1993, 1996b).
5I (Atlas 1975a,b, 1977b, 1978b, 1979, 1989) argued that negation was univocal—that it was semanti-
cally nonspecific rather than ambiguous. I also suggested how a Classical Gricean theory could accept an
identification of negation in English with the external negation in ordinary logic by giving up the iden-
tification of external negation with the “literal meaning” of a sentence-type. Instead, the pragmatic theory
would rest content with describing the understandings of utterances, and for the sake of theoretical sim-
plicity, though contrary to the linguistic facts of markedness, make the external negation the “unmarked”
understanding of an utterance-type (Atlas 1979). My own theoretical solution was not that; it was to rec-
ognize as essential to a full theory a semantic representation of the sentence-type, the product of the gram-
matical rules and the lexicon, but that account entailed that for the negative sentences the semantic
representation was of a univocal, scope-nonspecific rather than scope-ambiguous negative sentence.
6For the notion of direct entailment , see Atlas (1991a) and chapter 5 here: “A” » B; B
D D C; so,
“A” ° » C.
242 APPENDIX 3
that Levinson and I (Atlas and Levinson 1981) make of Halvorsen (1978), Horn
presents an interpretation of those (Atlas and Levinson 1981) views and on that foun-
dation presents some astute criticisms of those views. In what follows I shall assess
(a) the interpretation of Atlas and Levinson (1981) offered by Horn (1981a) and (b)
Horn’s criticism of Atlas and Levinson (1981) offered on the basis of that interpre-
tation. I hope to show that our (Atlas and Levinson 1981) actual views, and the views
of this section, survive the critical onslaught. Since there is also a parallel between
my (1981) treatment of clefts and my (1991a, 1996b) treatment of only Proper
Name, defusing Horn’s objections to the former will indirectly make more plau-
sible my treatment of the latter.
Halvorsen (1978) had claimed that statement (19a) had the same truth condi-
tions as statement (19b), and so “make the same assertion,” but (19a), unlike (19b),
in the sense of L. Karttunen and S. Peters (1979) “conventionally implicated,” or
presupposed, (19c) and something approximating (19d):

b. Jane kissed Max.
c. Jane kissed someone.
d. Jane kissed only one person.
e. Jane kissed at most one person.
“Conventional implicata,” in this sense, are preserved under negation and question-
ing. First, I (Atlas and Levinson 1981: 21) argued that Halvorsen was incorrect to
claim that (19a) “conventionally implicates” (19c). To show that it does, using tests
from Karttunen and Peters (1979), Halvorsen claimed that (20b) does not follow from
(20a), by contrast with (20c), which does, he believes, follow from (20a):
(20) a. I just discovered that it was Max that Jane kissed.

b. I just discovered that Jane kissed someone.
c. I just discovered that Jane kissed Max.
Example (20) is supposed to parallel example (21), a paradigm case of “conventional

implicature”:
(21) a. I just discovered that Max managed to write a brief.

b. I just discovered that it is difficult for Max to write a brief.
c. I just discovered that Max wrote a brief.
I claimed, pace Halvorsen, that (20b) no less followed from (20a) than (20c) did. So
I rejected the claim that (20a) “conventionally implicated” (20b). I rejected it on the
grounds that Halvorsen’s intuition was less compelling than its opposite, and also on
the grounds (Atlas and Levinson 1981: 24) that Halvorsen had supposed that in the
negative case It wasn’t Max that Jane kissed there should be the same “conventional
implicature,” which, in turn, meant that It wasn’t Max that Jane kissed—she didn’t
kiss anybody should be unacceptable. Since I found It certainly wasn’t Max that Jane
kissed—in fact, Jane didn’t kiss anyone perfectly acceptable, I concluded that
Halvorsen had no support for his claim.
The claim that (19a) “conventionally implicated” the Uniqueness Implicatum
(19d) Jane kissed only one person, or the Exhaustiveness Implicatum (19e) Jane
kissed at most one person, was open to my objection (Atlas and Levinson 1981: 25)
that on Halvorsen’s view (22) should be anomalous, which in fact, it isn’t:
(22) It wasn’t John that Mary kissed—it was Mart and Rick.
In light of data like (22), Halvorsen (1978: 16) had inferred that the correct “exhaus-
tiveness implicatum” would have the form (23):
(23) Mary kissed n0 persons.
where n0 had “no particular value.” Horn (1981a) charitably reformulated this ex-
haustiveness implicatum as (24):
(24) Mary kissed at most one thing [within some contextually defined set].
But then he agreed with my rejection of Halvorsen’s claim that (19a) “convention-
ally implicates” (24).
On my analysis of the logical form of clefts, the truth conditions of It was Max
that Jane kissed are equivalent to Jane kissed someone & Jane kissed at most Max.
It then follows, from a cleft sentence with a single, non-vacuous, Proper Name in
focus position (and does not follow in other cases), Jane kissed (exactly) one person.
The theory was formulated specifically for referring singular terms in focus posi-
tion; I did not discuss indefinite noun phrases in focus position. After agreeing that
my theory had the virtue of showing that the negative sentence It wasn’t Max that
Jane kissed, on its exclusion negation interpretation, did not entail Jane kissed (ex-
actly) one person, and so, unlike Halvorsen’s (1978) incorrect account, did not com-
mit the speaker of the negative sentence to that entailment, Horn (1981a) offered three
objections to my theory. I now shall discuss those objections.
Using an example with an indefinite noun phrase in focus position, Horn (1981a)
described my view as claiming that It was a pizza that Lauren ate entails Lauren ate
a pizza but not conversely, that it entails, and its negation presupposes, Lauren ate
something, and that it entails, but does not presuppose, Lauren ate (exactly) one thing.
Then he further glossed my view as implying that the truth conditions of It was a
pizza that Lauren ate are Lauren ate a pizza and only a pizza (alternatively, Lauren
ate a pizza and Lauren ate (exactly) one thing). Unfortunately, this interpretation of
my account was not accurate. On my account the truth conditions of It was a pizza
that Lauren ate would be Lauren ate something and Lauren ate at most a pizza.
Furthermore, on the theory of ‘only’ in Atlas (1991a, 1996b), the truth conditions of
Lauren ate only a pizza would be Lauren ate (exactly) one thing and Lauren ate at
most a pizza. Finally, it does not follow from my account that It was a pizza that
Lauren ate entails Lauren ate (exactly) one thing; so Horn’s characterization of that
244 APPENDIX 3
entailment as following from my account was incorrect, as were the truth conditions
that he imputed to my account.
Horn (1981a) then properly objected to his version of my theory that, on the face
of it, Lauren ate a pizza and Lauren ate (exactly) one thing does not express the truth
conditions of It was a pizza that Lauren ate. And I certainly agree with him, but my
account entailed no such claim. It claimed that the truth conditions were Lauren ate
something and Lauren ate at most a pizza. So Horn’s first objection missed the mark.
Horn’s second objection was that my theory failed to predict that the assertion
of Lauren ate a pizza with the denial of the cleft sentence is semantically anomalous,
as in (25a) or (25b):
(25) a. #Lauren ate a pizza, but it wasn’t a pizza that she ate.
b. #I know Lauren ate a pizza, but it wasn’t a pizza that she ate.
c. #Lauren ate a pizza but Lauren ate something and whatever Lauren ate was a non-
pizza.
d. #∃x[A(l,x) & Px] & ∃x A(l,x) & ∀x[A(l,x) → ¬Px].
Horn wrote, “There is no obvious way to rule out the infelicitous sequences . . . if we
are to insist, with [Atlas and Levinson], that clefts entail exhaustiveness” (1981a:
130). Let us see if this is so.
Since my theory states that the truth conditions of the default, choice negation
understanding of the utterance-type It wasn’t a pizza that Lauren ate is analogous to
that of It wasn’t Max that Jane kissed, the truth conditions of Horn’s sentence (25a)
would be (25c ,d). Statement (25c) is semantically odd by virtue of expressing a bla-
tant contradiction. Thus my account explains the anomaly of (25a) by the anomaly
of (25c). The oddity is more obvious when focus position contains a singular term
rather than an indefinite description, as in (26):
(26) a. #Jane kissed Max, but it wasn’t Max that she kissed.
b. #Jane kissed Max, but Jane kissed someone & Jane did not kiss Max.
My theory has no difficulty explaining by appeal to my account of negative clefts

the anomaly of (26a), as it derives from the anomaly in (26b).
Horn also complained that my theory could not explain the redundancy (the
“pointlessness”) of asserting the simple declarative and the affirmative cleft as well:
(27) a. Jane kissed Max, but it was Max that she kissed,
b. Lauren ate a pizza, but it was a pizza that she ate.
My theory gives the following truth conditions for (27a,b):
(28) a. Jane kissed Max, but Jane kissed someone & Jane kissed at most Max.
b. Lauren ate a pizza, but Lauren ate something & Lauren ate at most a pizza.
It seems to me that if one’s linguistic intuitions would cause one to regard (27) as
redundant, so, too, mutatis mutandis, would one regard (28) as redundant. So my
theory would explain the redundancy of (27) by the redundancy of (28). Horn’s sec-
ond objection misses the mark.
Finally, Horn objected that my theory could not explain the alleged oddity
of (29a):
(29) a. It wasn’t John that Jane kissed, it was John and Brian.
b. Jane kissed some group & Jane did not kiss John & Jane kissed some group & Jane
kissed at most the group John and Brian.
on Horn’s assumption that It was John that Jane kissed entails Jane kissed (exactly)
one person and Jane kissed no one else.
On the choice negation understanding of the negative sentence, the truth-
conditions of sentence (29a) would be (29b). For myself, I do not find (29a) infe-
licitous, but if I were to find it infelicitous, I would find (29b) infelicitous in the
same way. In either case, my theory seems able to explain the varying linguistic
intuitions. Of course, the natural, felicitous understanding of (29a) would appeal
to an exclusion negation, or to Horn’s (1985, 1989) “metalinguistic” negation, in
which case (29a) would definitely not be anomalous and would not have the truth-
conditions of (29b). Thus, contrary to Horn’s claim, my theory predicts just the
anomaly it should.
As far as I can see, none of Horn’s three objections succeeds. The reason for
Horn’s objections to fail to hit the mark is Horn’s inadvertent misconstrual of my
theory. He thought that on my theory It was a pizza that Lauren ate meant Lauren
ate a pizza and only a pizza or Lauren ate a pizza and Lauren ate (exactly) one thing.
It does not. On my theory the truth conditions are Lauren ate something and Lauren
ate at most a pizza. Furthermore, although I claimed that It was Max that Jane kissed
entailed Jane kissed (exactly) one person, my theory does not imply that It was a
pizza that Lauren ate entails Lauren ate (exactly) one thing. On my theory there is
an important difference between indefinite noun phrases and definite noun phrases
in focus position.
Furthermore, on my (1991a) analysis of ‘only’ statements, It was Max that Jane
kissed is logically equivalent to Jane kissed only Max, but It was a pizza that Lauren
ate is not logically equivalent to Lauren ate only a pizza; contrast Lauren ate some-
thing and Lauren ate at most a pizza with Lauren ate (exactly) one thing and Lauren
ate at most a pizza. On my theory it is true that Lauren ate only a pizza entails It was
a pizza that Lauren ate; the converse entailment fails. My theory implies that Horn’s
alleged analysans Lauren ate a pizza and only a pizza of It was a pizza that Lauren
ate is logically stronger than, on my theory, it should be; my analysans is the weaker
Lauren ate something and Lauren ate at most a pizza. In short:
It was Max that Jane kissed Jane kissed only Max.
Lauren ate only a pizza It was a pizza that Lauren ate.
It was a pizza that Lauren ate Lauren ate only a pizza.
Finally, since Horn suggests that the Exhaustiveness Condition is conversation-
ally implicated in the assertion of a cleft sentence, he is worried by the apparent dif-
ficulty in canceling the Exhaustiveness Condition in sentences like:
246 APPENDIX 3
(30) a. (#)It was a pizza that Lauren ate; indeed, it was a pizza and a calzone.
The consequence of this, as Horn notes, is that:
if we tentatively assume, on the basis of the (purported) deviance of [(30a)] and the
difficulty of establishing context-cancelation, that the [alleged] exhaustiveness
implicature is not cancelable, we arrive at a curious conclusion about the relation
of cancelability and implicature. . . . If it turns out to be correct that exhaustiveness
is a non-cancelable generalized conversational implicature of cleft sentences, we
must conclude that non-cancelablity is . . . not a sufficient condition for concluding
that an inference is conventional in nature.
This, as Horn notes, is indeed a “provocative possibility.”

I would explain (30a) rather differently. On my theory, the truth-conditions of
(30a) are (30b):
(30) b. Lauren ate something & Lauren ate at most a pizza; indeed, Lauren ate something
& Lauren ate at most a pizza and a calzone.
In my analysis cancelability of the exhaustiveness condition properly fails, as the

condition is not an implicatum, and Horn’s provocative possibility is, happily, a mere
possibility and produces no anomaly in the theory of implicature.
The semantics-pragmatics interface
Within the philosophy of language and linguistic theory, there have been attempts to
investigate the relationship between semantics and pragmatics, to map a boundary
between the two domains, and to understand the mechanics of their interaction. In
the original 1981 work, our aim was to exemplify one approach through which our
understanding might be improved and to make evident the explanatory power of such
an approach. A benefit for linguistics is the retrieval of the hope that the phenomena
known as “presupposition” can be reduced to matters of entailment on the one hand
and nonconventional, generalized conversational inferenda on the other. The ingre-
dients making this hope viable are (a) a refinement of the role of logical form,
(b) the formulation of general principles of conversational inference, and (c) an ap-
peal to a Chomsky Internalist Semantics notion of semantical underdeterminacy. (See
chapter 4.)
The original intuition that Levinson and I tried to explicate is that there is sig-
nificant semantic structure, explicable by logical forms and by even more abstract
Internalist Semantics semantic representations, over and beyond propositions, sets
of possible worlds, or truth conditions. This structure meshes closely with pragmatic
principles to produce via the Performance System informative, defeasible inferenda.
There were two problems. First, we needed to find some independent condition on
logical forms that express the same truth conditions. This condition would distin-
guish a semantic representation of an English sentence from a sentence logically
equivalent to it. Second, we needed to make explicit that, in fact, there are two
countervailing pragmatic principles governing informativeness, not simply a hodge-

podge of conflicting inferences.
The successful development of this approach would have several benefits. The
one illustrated in the preceding section is the reduction of some well-known pre-
suppositional phenomena to the case of an abstract semantic representation interact-
ing intimately in the Performance System with pragmatic principles of the sort due
to H. P. Grice to give utterance-interpretations. Alternative accounts treat presuppo-
sition as irreducible, a special species of conventional, non-truth-conditional infer-
ence that requires specific lexical items and syntactic structures to be associated with
the inferences. This is accomplished not by rule but item by item (Gazdar 1979a;
Karttunen and Peters 1979). On our theory a few general principles will explain a
wide range of data. Apart from the strength and simplicity of theory thereby achieved,
our account attempts to answer to the intuition that presuppositions arise in part be-
cause of the semantic properties of the statements yielding them, but it avoids the
incoherencies of accounts of “semantic presupposition” (see Boër and Lycan 1976).
One example of a simplification attributable to a more delicate use of logical
form is the unification of the presuppositional behavior of clefts, factives, and defi-
nite descriptions, as illustrated in (31):

b. λx(Gx)(γxFx)
c. Mikael knows that California is exciting.
d. K(m,|P(P = ^C & Tr(P))
e. The Prince of Wales is clever.
f. G(|xFx)
But whatever the success of this reduction, the issues raised here bear on how the
relation between semantics and pragmatics, between Chomsky’s Internalist Semantics
and the Performance System, should be construed: What the relationship between
semantic representations, truth conditions, inferenda, implicata, and logical forms
is; what conditions of adequacy (e.g., predicting “aboutness” and reading off inferenda
and implicata) semantic representations and logical forms should satisfy. These
problems are central to a complete theory of sentence meaning and utterance-
interpretation.
The analysis given for cleft sentences is the product of a combination of ele-
ments: (a) semantic representations, (b) inferenda and Gricean implicata, (c) logical
forms, and (d) topic noun phrase constraints on the structure of semantic representa-
tions and logical forms. Only all the elements, working coherently together, manage
to give an accurate and adequate explanation of how the semantic representation and
logical form of the English cleft structure produces its inferenda, truth conditions,
entailments, and conversational implicata.
248 APPENDIX 4
Appendix 4
A Note on Notation
I use notation and typography from logic, mathematics, and linguistics in this book.
There are also the usual literary conventions in the use of quotation marks to deal
with. As a guide to readers, I briefly sketch my usage here.
1. Logical symbols of a Formal Object Language

x, y, . . . Individual variables
a, b, . . . Individual constants
F11, . . . , G11, . . . Predicate letters
P, P1, P2 . . . Q, Q1, Q2, . . . Sentence letters
& Truth-functional conjunction (and)
∨ Truth-functional disjunction (or)
→ Truth-functional conditional (only if )
↔ Truth-functional equivalence (if and only if )
¬ Exclusion negation (not)
– Choice negation (not)
∀ Universal quantifier ( for every individual)
∃ Existential quantifier ( for at least one individual)
= Identity
(|x)F(x) Inverted sans-serif capital ‘I’ for definite descriptions
(γx)F(x) Lower-case Greek gamma for group descriptions
Logical necessity
248
A NOTE ON NOTATION 249
2. Symbols of meta-languages
a. Italics
Tom is clever The name of an utterance-type, a meaningful
form of words in their orthographic representation, is given in
italics. (This convention differs from that of Sir John Lyons
[1995a: 23–24] in which italicized expressions denote form
without meaning. I shall sometimes follow Sir John’s convention,
but I shall explicitly say so when I do, as, for example, in (b)
below.)
b. Single quotation marks
‘Tom’ Single quotation marks are used to indicate the name of a
name or other expression by putting the named expression
between single quotation marks.
‘Tom is clever’ The name of a sentence (i.e., a system-sentence,
which is generated by the grammatical rules in a generative
grammar, to be distinguished from the product of an act of
utterance) is expressed in single quotation marks. Single-quoted
English expressions will be items with both form (written and
spoken) and meaning: for example, the word ‘man’, represented
by its stem-form, has the word-forms man and men.
c. Double quotation marks
A direct quotation.
An assertoric force indicator. ‘“Tom is clever”’ names an assertoric
utterance-type in English.
d. Dot quotation
•__• Wilfrid Sellars’s (1963: 204–5) dot quotation to denote a

sortal term for inter-linguistically synonymous expressions. Thus
the French sentence ‘La neige est blanche’ and the English
sentence ‘Snow is white’ are both a •snow is white•.
e. Square bracketed small capitals
[MALE] Square-bracketed expressions denote sense components
or other items of meaning.
f. Greek letters
µ, µ', . . . µ1, µ2 . . . Metalinguistic variables for expressions of
an object language.
φ, Ψ, χ, . . . Metalinguistic variables for statements of an object
language, formal or informal (natural).
σ, τ Metalinguistic variables for singular terms of an Object
Language, formal or informal (natural).
250 APPENDIX 4
π, π1, π2, . . . ρ, ρ1, ρ2, Metalinguistic variables for predicate and

relation symbols
∆, Γ Metalinguistic variables for sets of object language
statements.
g. Quasi quotation
__ W. V. O. Quine’s (1981) quasi-quotation marks (corner
quotes) to speak of specific contexts of unspecified expressions.
For example: φ ↔ Ψ is the result of writing φ and then ‘↔’ and
then Ψ, or equivalently, is the result of putting φ for ‘φ’ and Ψ for
‘Ψ’ in ‘φ ↔ Ψ’. Hence, µ is µ, since the result of putting the
expression µ for the symbol ‘µ’ in the symbol ‘µ’ is just the
expression µ. For example, ¬φ is true iff φ is not true.
h. Schematic letters
P, Q, . . . are schematic letters for statements of an object lan-
guage. They are not metavariables for which one would substitute
a name of a statement; they are placeholders for which one
substitutes the sequence of expressions that constitutes the
statement, not the name of the statement. Thus the expression
‘(P → Q)’ is a statement-schema or statement-form, an instance of
which would be ‘(Tom is clever → Tom sings)’.
‘F’, ‘G’, . . . are schematic letters for predicate and relation
symbols of an object language.
i. Concatenation
‘^’ is the concatenation symbol for concatenating expressions.
j. Double vertical bars and brackets
φA The intension or propositional content of the statement φ in
an interpretation A. Given an interpretation A and a set of possible
worlds I, for the world i, φ iA ∈ {T,F}: for example, if the
statement under that interpretation is true in world i, φ iA = 1
(letting ‘1’ stand for ‘True’). See D. S. Scott (1970a, 1971,
1973).
φ[f] A specific [f] interpretation of an F-nonspecific literal
sentence-meaning [φ] of an English sentence φ.
µ[f] A specific [f] interpretation of an F-nonspecific literal
expression-meaning [µ] of an English expression µ.
For example, for the gender-nonspecific lexeme ‘he’ in English,
he[FEMALE] = [she]. Also, horse[MALE] = [stallion], and
horse[FEMALE] = [mare].
[[φ]] Truth-conditions for φ under the intended interpretation A*
of φ. Thus, usually φA* = [[φ]].
A NOTE ON NOTATION 251
k. Double vertical turnstiles

V Logical consequence relation between sentences or utter-
ance-types with respect to a class V of admissible valuations val(φ)
of sentences φ of a language L: for example, A V B. See Scott
(1970, 1971, 1973); B. van Fraassen (1971). A B is to be read
‘A does not entail B’.
» Implicatural relation between asserted (actual or possible)
tokens of utterance-types and sentences (expressing thoughts or
semantic contents): for example, “Snow is almost white” » Snow
is not quite white.
k. Horizontal turnstiles
Truth in an interpretation (e.g., A φ iff φA =1).
S Derivable in a formal system S.
l. Linguistic judgment notations
* Ungrammatical
# Semantically anomalous
? Unacceptable
?? Marginally acceptable
3. Set theoretic notation
∈ Set membership
∩ Set intersection
∪ Set union
⊆ Subset relation
ø Empty set
û(φ(u)) The set of individuals that satisfy φ(x), viz. {x : φ(x)}.
u(φ(u)) Sometimes it will be convenient to use the intensional
version of the extension: the attribute of Fness. I shall denote the
attribute by: u(φ(u)), where the variable preceding the open
sentence is without the circumflex.
A, B, . . . Sets
X, Y, . . . Variables ranging over sets
BIBLIOGRAPHY
Akmajian, A. (1970). “On Deriving Cleft Sentences from Pseudo-cleft Sentences,” Linguis-
tic Inquiry, 1: 149–168.
Allwood, J. (1972). “Negation and the Strength of Presupposition.” Logical Grammar Re-
ports No. 2. Department of Linguistics, University of Göteborg, Sweden.
———. (1977). “Negation and the Strength of Presupposition,” rev. ed., in O. Dahl (ed.),
Logic, Pragmatics, and Grammar (pp. 11–52). Göteborg: Department of Linguistics.
Andrews, P. B. (1986). An Introduction to Mathematical Logic and Type Theory: To Truth
through Proof. Orlando: Academic Press.
Anscombe, G. E. M. (1981). Metaphysics and the Philosophy of Mind. Minneapolis: Univer-
sity of Minnesota Press.
———. (1981b). “Before and After,” in G. E. M. Anscombe, Metaphysics and the Philoso-
phy of Mind (pp. 180–95). Minneapolis: University of Minnesota Press.
Anscombe, J.-C., and Ducrot, O. (1977). “Deux mais en français?” Lingua: 23–40.
Atlas, J. D. (1972). “A Davidsonian Approach to Demonstrative Inference.” Presented at a
Rutgers University colloquium with Dana Scott and Philip Johnson-Laird on logic and
language, 26 April 1972.
———. (1974). “Presupposition, Ambiguity, and Generality: A Coda to the Russell-Strawson
Debate on Referring.” Typescript. Department of Philosophy, Pomona College, Clare-
mont, California.
———. (1975a). “Frege’s Polymorphous Concept of Presupposition and Its Role in a Theory
of Meaning,” Semantikos, 1: 29–44.
———. (1975b). “Presupposition: A Semantico-Pragmatic Account,” in G. Gazdar and
S. C. Levinson (eds.), Pragmatics Microfiche (1.4: pp. D13–G9). Cambridge: Cambridge
University Press.
———. (1977a). “Presupposition Revisited,” in G. Gazdar and S. C. Levinson (eds.), Prag-
matics Microfiche, (2.5: pp. D5–D11). Cambridge: Cambridge University Press.
———. (1977b). “Negation, Ambiguity, and Presupposition,” Linguistics and Philosophy,
1: 321–36.
253
254 BIBLIOGRAPHY
———. (1978a). “Presupposition and Grice’s Pragmatics.” Presented at a colloquium, De-

partment of Phonetics and Linguistics, University College, London, May 1978.
———. (1978b). “On Presupposing,” Mind, 87: 396–411.
———. (1979). “How Linguistics Matters to Philosophy: Presupposition, Truth, and Mean-
ing,” in D. Dinneen and C. K. Oh (eds.), Syntax and Semantics 11: Presupposition
(pp. 265–81). New York: Academic Press.
———. (1980a). “A Note on a Confusion of Pragmatic and Semantic Aspects of Negation,”
Linguistics and Philosophy, 1: 321–36.
———. (1980b). “Reference, Meaning, and Translation,” Philosophical Books, 21: 129–40.
———. (1981). “Is ‘Not’ Logical?” Proceedings of the 11th International Symposium on
Multiple-Valued Logic (pp. 124–28). New York: Institute of Electrical and Electronics
Engineers.
———. (1983). “Comments on ‘Metalinguistic Negation and Pragmatic Ambiguity’ by Larry
Horn, Yale University, June 1983.” Manuscript, the Institute for Advanced Study,
Princeton, New Jersey, December 1983.
———. (1984a). “Comparative Adjectives and Adverbials of Degree: An Introduction to
Radically Radical Pragmatics,” Linguistics and Philosophy, 7: 347–77.
———. (1984b). “Grammatical Non-Specification: The Mistaken Disjunction Theory,” Lin-
guistics and Philosophy, 7: 433–43.
———. (1988). “What Are Negative Existence Statements About?” Linguistics and Philoso-
phy, 11: 371–93.
———. (1989). Philosophy without Ambiguity. Oxford: Clarendon Press.
———. (1990). “Implicature and Logical Form: The Semantics-Pragmatics Interface.” Five
Lectures, 6 August 1990–10 August 1990, Second European Summer School in Lan-
guage, Logic, and Information, Katholieke Universiteit, Leuven, Belgium.
———. (1991a). “Topic/Comment, Presupposition, Logical Form, and Focus Stress Implicatures:
The Case of Focal Particles only and also,” Journal of Semantics, 8: 127–47.
———. (1991b). “Negative Existence Statements, Fictional Objects, and Intensional Con-
texts: The Relation between Bedeutung and Topic/Comment.” Presented at the Sixteenth
International Linguistic Agency Symposium, Multidisciplinary Research on Reference:
History and Present State of the Art, Duisberg, Germany, 18–22 March 1991.
———. (1993). “The Importance of Being ‘Only’: Testing the Neo-Gricean versus Neo-
Entailment Paradigms,” Journal of Semantics, 10: 301–18.
———. (1994). “Do It in DOS,” PC Laptop Computer Magazine, Vol. 6, No. 11 (Novem-
ber), pp. 40–45.
———. (1995). “G. E. Moore’s Paradox, Wittgenstein’s Philosophy of Mind, and the Gram-
mar of First-Person Belief.” Unpublished ms. Department of Philosophy, Pomona Col-
lege, Claremont, California.
———. (1996a). “Negative Quantifier Noun Phrases: A Typology and an Acquisition Hy-
pothesis.” Presented at Perspectives on Negation, Pionier Conference on Negation,
sponsored by the NWO Pionier Project “Reflections of Logical Patterns in Language
Structure and Language Use,” University of Groningen, The Netherlands, 24–26 Au-
gust 1996.
———. (1996b). “‘Only’ Noun Phrases, Pseudo-Negative Generalized Quantifiers, Nega-
tive Polarity Items, and Monotonicity,” Journal of Semantics, 13: 265–328.
———. (1997a). “On the Modularity of Sentence Processing: Semantical Generality and the
Language of Thought,” in J. Nuyts and E. Pederson (eds.), Language and Conceptuali-
zation (pp. 213–28). Cambridge: Cambridge University Press.
———. (1997b). “Negative Adverbials, Prototypical Negation and the De Morgan Tax-
onomy,” Journal of Semantics 14: 349–67.
BIBLIOGRAPHY 255
———. (1998). “Adverbial Verb Phrases and Downward Monotonicity: Negativity and the
De Morgan Taxonomist’s Dilemma.” Presented at First Annual International Confer-
ence on Negation: Syntax, Semantics, and Pragmatics, European Studies Research In-
stitute, North West Centre for Linguistics, University of Salford, Manchester, United
Kingdom, 30 October 1998–1 November 1998.
———. (2001). “Negative Quantifier Noun Phrases: A Typology and an Acquisition Hypoth-
esis,” in J. Hoeksema, H. Rullman, V. Sánchez-Valencia, and T. van der Wouden (eds.),
Perspectives on Negation and Polarity Items (pp. 1–23). Amsterdam: John Benjamins.
Atlas, J. D., and Levinson, S. C. (1973). “What Is an Implicature? Part 1: Kenny Logic.”
Unpublished ms., Mathematical and Social Sciences Board Workshop on the Pragmat-
ics of Natural Language, University of Michigan, Ann Arbor.
———. (1981). “It-Clefts, Informativeness, and Logical Form: An Introduction to Radically
Radical Pragmatics (Revised Standard Version),” in P. Cole (ed.), Radical Pragmatics
Auden, W. H. (1968). The Dyer’s Hand. New York: Vintage.
Aune, B. (1975). “Vendler on Knowledge and Belief,” in K. Gunderson (ed.), Minnesota
Studies in the Philosophy of Science: 7. Language, Mind, and Knowledge (pp. 391–99).
Minneapolis: University of Minnesota Press.
Austin, J. L. (1956–57). “A Plea for Excuses,” in Proceedings of the Aristotelian Society 57
(pp. 1–30).
———. (1962/1975). How to Do Things with Words. Cambridge: Harvard University Press.
Avramides, A. (1989). Meaning and Mind: An Examination of a Gricean Account of Lan-
guage. Cambridge: MIT Press.
Bach, K. (1982). “Semantic Nonspecificity and Mixed Quantifiers,” Linguistics and Philoso-
phy, 4: 593–605.
———. (1987). Thought and Reference. Oxford: Clarendon Press.
———. (1994a). “Conversational Impliciture,” Mind and Language, 9: 124–62.
———. (1994b). “Meaning, Speech Acts, and Communication: Introduction,” in R. M.
Harnish (ed.), Basic Topics in the Philosophy of Language (pp. 3–20). Englewood Cliffs,
N.J.: Prentice Hall.
———. (1994c). “Semantic Slack,” in S. Tzohatzidis (ed.), Foundations of Speech Act Theory
(pp. 267–91). London: Routledge.
———. (1995). “Standardization vs. Conventionalization,” Linguistics and Philosophy, 18:
677–86.
Baldwin, T. (1992). G. E. Moore. London: Routledge.
Ballmer, T. (1975). “Einführung and Kontrolle von Diskurswelten,” in D. Wunderlich (ed.),
Linguistische Pragmatik (pp. 183–206). Berlin: Athenaion-Verlag.
———. (1978). Logical Grammar: With Special Consideration of Topics in Context Change.
Amsterdam: North-Holland.
Barwise, J., and Cooper, R. (1981). “Generalized Quantifiers and Natural Language,” Lin-
guistics and Philosophy 4: 159–219.
Beaver, D. (1997). “Presupposition,” in J. van Benthem and A. ter Meulen (eds.), Handbook
of Logic and Language (pp. 939–1008). Amsterdam: Elsevier Science.
Bergmann, M. (1981). “Presupposition and Two-Dimensional Logic.” Journal of Philosophi-
cal Logic 10: 27–53.
———. (1982). “Metaphorical Assertions,” Philosophical Review, 91: 229–45.
Berkeley, G. (1948–57). “A Treatise concerning the Principles of Human Knowledge.” In
A. Luce and T. E. Jessop, (eds.), The Works of George Berkeley. 9 vols. London: Nelson.
Originally published 1710.
Bierwisch, M. (1989). “The Semantics of Gradation,” in M. Bierwisch and E. Lang (eds.),
256 BIBLIOGRAPHY
Dimensional Adjectives: Grammatical Structure and Conceptual Interpretation (pp. 71–

261).Berlin: Springer-Verlag.
Black, M. (1954). “Metaphor,” in M. Black, Models and Metaphor (pp. 41–60). Ithaca: Cornell
University Press.
Blakemore, D. (1992). Understanding Utterances: An Introduction to Pragmatics. Oxford:
Blackwell.
Blok, P. (1993). “Interpretation of Focus.” Ph.D. diss., University of Groningen, The
Netherlands.
Blutner, R. (2000). “Some Aspects of Optimality in Natural Language Interpretation,” Jour-
nal of Semantics, 17: 189–216.
Böer, S. (1979). “Meaning and Contrastive Stress,” Philosophical Review, 88: 263–98.
Boër, S., and Lycan, W. (1976). The Myth of Semantic Presupposition. Ohio State Work-
ing Papers in Linguistics No. 21. Columbus: Department of Linguistics, Ohio State
University.
Bromberger, S. (1992a). On What We Know We Don’t Know. Chicago: University of Chi-
cago Press.
———. (1992b). “An Approach to Explanation,” in S. Bromberger On What We Know We
Don’t Know (pp. 18–51). Chicago: University of Chicago Press.
Brown, P., and Levinson, S. C. (1978). “Universals in Language Usage: Politeness Phenom-
ena,” in E. Goody (ed.), Questions and Politeness: Strategies in Social Interaction,
(pp. 56–310). Cambridge: Cambridge University Press.
———. (1987). Politeness: Some Universals in Language Usage. Studies in Interactional
Sociolinguistics 4. Cambridge: Cambridge University Press.
Brugman, C. (1981). “The Story of ‘over’.” Bloomington: Indiana University Linguistics Club.
Burge, T. (1990). “Frege on Sense and Linguistic Meaning,” in D. Bell and N. Cooper (eds.),
The Analytic Tradition (pp. 30–60). Oxford: Blackwell.
Burnyeat, M. (1967–68). “Belief in Speech.” Proceedings of the Aristotelian Society 68: 227–
248.
Burton-Roberts, N. (1989a). “On Horn’s Dilemma: Presupposition and Negation,” Journal
of Linguistics, 25: 95–125.
———. (1989b). The Limits to Debate. Cambridge: Cambridge University Press.
———. (1991). Review of Atlas (1989). Mind and Language, 6: 161–76.
———. (1997). “On Preservation under Negation,” Lingua 101: 65–88.
———. (1999). “Presupposition-Cancelation and Meta-linguistic Negation: A Reply to
Carston,” Journal of Linguistics, 35: 347–64.
Carnap, R. (1942). Introduction to Semantics. Cambridge: Harvard University Press.
———. (1958). Introduction to Symbolic Logic and Its Applications. Trans. W. H. Meyer
and J. Williamson. New York: Dover.
Carnap, R., and Bar-Hillel, Y. (1952). An Outline of a Theory of Semantic Information. Tech-
nical Report No. 247. Cambridge: Research Laboratory of Electronics, MIT.
———. (1953–54). “Semantic Information,” British Journal for the Philosophy of Science,
4: 147–57.
Carroll, L. (1963). The Annotated Alice. Cleveland: World.
Carston, R. (1985). “A Reanalysis of Some ‘Quantity Implicatures.’” Unpublished ms., Uni-
versity of London.
———. (1988). “Implicature, Explicature, and Truth-Theoretic Semantics,” in R. M.
Kempson, (ed.), Mental Representations: The Interface between Language and Reality
———. (1996). “Metalinguistic Negation and Echoic Use,” Journal of Pragmatics, 25: 309–
30.
BIBLIOGRAPHY 257
———. (1998). “Negation, ‘Presupposition’, and the Semantics/Pragmatics Distinction,”

Journal of Linguistics, 34: 309–50.
———. (1999). “Negation, ‘Presupposition’, and Metarepresentation: A Response to Noel
Burton-Roberts,” Journal of Linguistics, 35: 365–90.
Chapman, S. (1996). “Metalinguistic Negation: Some Theoretical Implications,” Journal of
Linguistics, 32: 387–402.
Chierchia, G. (1995). Dynamics of Meaning: Anaphora, Presupposition, and the Theory of
Grammar. Chicago: University of Chicago Press.
Chierchia, G., and McConnell-Ginet, S. (1990). Meaning and Grammar. Cambridge: MIT
Press.
Chomsky, N. (1957). Syntactic Structures. The Hague: Mouton.
———. (1965). Aspects of the Theory of Syntax. Cambridge: MIT Press.
———. (1972a). Language and Mind. New York: Harcourt Brace Jovanovich.
———. (1972b). “Deep Structure, Surface Structure, and Semantic Interpretation,” in N.
Chomsky, Studies on Semantics in Generative Grammar (pp. 62–119). The Hague:
Mouton.
———. (1982). Lectures on Government and Binding: The Pisa Lectures. 2nd ed. Dordrecht:
Foris.
———. (1986). Knowledge of Language. New York: Praeger.
———. (1995a). The Minimalist Program. Cambridge: MIT Press.
———. (1995b). “Language and Nature,” Mind, 104: 1–61.
———. (1996a). Powers and Prospects: Reflections on Human Nature and the Social Order.
Boston: South End Press.
———. (1996b). “Language and Thought: Some Reflections on Venerable Themes,” in
N. Chomksy, Powers and Prospects: Reflections on Human Nature and the Social
Order (pp. 1–30). Boston: South End Press.
———. (1996c). “Language and Nature,” in N. Chomsky, Powers and Prospects: Reflec-
tions on Human Nature and the Social Order (pp. 31–54). Boston: South End Press.
———. (2000). New Horizons in the Study of Language and Mind. Cambridge: Cambridge
University Press.
Churchland, P. (1988). “Perceptual Plasticity and Theoretical Neutrality: A Reply to Jerry
Fodor,” Philosophy of Science, 55: 167–87.
Clark, H., and Haviland, S. E. (1977). “Comprehension and the Given-New Context,” in
R. Freedle (ed.), Discourse Production and Comprehension (pp. 1–40). Hillside, N.J.:
Erlbaum.
Clarke, D. S. (1994). “Does Acceptance Entail Belief?” American Philosophical Quarterly,
31: 145–55.
Coffa, J. A. (1991). The Semantic Tradition from Kant to Carnap. Cambridge: Cambridge
University Press.
Cooper, W. E., and Ross, J. R. (1975). “World Order,” in R. E. Grossman, J. San, and
T. Vance (eds.), Papers from the Parasession on Functionalism (pp. 63–111). Chicago:
Chicago Linguistics Society.
Crane, T. (1995). The Mechanical Mind. London: Penguin.
Cresswell, M. (1976). “The Semantics of Degree,” in B. Partee (ed.), Montague Grammar
Cruse, D. A. (1986). Lexical Semantics. Cambridge: Cambridge University Press.
———. (1992). “Monosemy vs. Polysemy” (review of Ruhl, 1989), Linguistics, 30: 577–99.
Cummins, R. (1989). Meaning and Mental Representation. Cambridge: MIT Press.
Davidson, D. (1967). “The Logical Form of Action Sentences,” in N. Rescher (ed.), The Logic
of Decision and Action (pp. 81–95). Pittsburgh: University of Pittsburgh Press.
258 BIBLIOGRAPHY
———. (1970). “Action and Reaction,” Inquiry, 13: 140–48.

———. (1980). Essays on Actions and Events. Oxford: Clarendon Press.
———. (1981). “What Metaphors Mean,” in S. Sacks (ed.), On Metaphor (pp. 29–46). Chi-
cago: University of Chicago Press.
———. (1984a). Inquiries into Truth and Interpretation. Oxford: Clarendon Press.
———. (1984b). “Radical Interpretation,” in D. Davidson, Inquiries into Truth and Inter-
pretation. (pp. 125–39). Oxford: Clarendon Press.
Davidson, D., and Harman, G. (eds.) (1972). Semantics of Natural Language. Dordrecht: D.
Reidel.
———. (eds.) (1975). The Logic of Grammar. Belmont, Calif: Wadsworth.
Davis, S. (ed.) (1991). Pragmatics. New York: Oxford University Press.
Davis, W. A. (1998). Implicature: Intention, Convention, and Principle in the Failure of
Gricean Theory. Cambridge: Cambridge University Press.
de Mey, S. (1991). “‘Only’ as a Determiner and Generalized Quantifier,” Journal of Seman-
tics, 8: 91–106.
Donnellan, K. (1966). “Reference and Definite Descriptions,” Philosophical Review, 75: 281–
304.
———. (1983). “Kripke and Putnam on Natural Kind Terms,” in C. Ginet and S. Shoemaker
(eds.), Knowledge and Mind (pp. 84–104). New York: Oxford University Press.
———. (1993). “There Are Words for That Kind of Thing: An Investigation of Two Thought
Experiments,” in J. E. Tomberlin (ed.), Philosophical Perspectives 7: Language and Logic
(pp. 155–71). Atascadero, Calif.: Ridgeview.
Dowty, D. (1980). “Comments on the Paper by Bach and Partee,” in J. Kreiman and A. Ojeda
(eds.), CLS 16: Papers from the Sixteenth Regional Meeting of the Chicago Linguistics
Society: Parasession on Pronouns and Anaphora (pp. 29–40) Chicago: Chicago Lin-
guistics Society.
Dowty, D., Wall, R. E. and Peters, S. (1981). Introduction to Montague Semantics. Dordrecht:
Reidel.
Dretske, F. (1972). “Contrastive Statements,” Philosophical Review, 81: 411–37.
Dummett, M. (1979). “What Does the Appeal to Use Do for the Theory of Meaning?” in
A. Margalit (ed.), Meaning and Use (pp. 123–35). Dordrecht: Reidel.
———. (1981). Frege: Philosophy of Language. 2nd ed. London: Duckworth.
———. (1993). Origins of Analytical Philosophy. Cambridge: Harvard University Press.
Empson, W. (1930). Seven Types of Ambiguity. London: Chatto and Windus.
Englebretsen, G. (1996). Something to Reckon With: The Logic of Terms. Ottawa: University
of Ottawa Press.
Fillmore, C. J. (1971). “Verbs of Judging: An Exercise in Semantic Description,” in C. J.
Fillmore and D. T. Langendoen (eds.), Studies in Linguistic Semantics (pp. 273–90).
New York: Holt, Rinehart and Winston.
Fish, S. (1989). Doing What Comes Naturally: Change, Rhetoric, and the Practice of Theory
in Literary and Legal Studies. Durham: Duke University Press.
Fodor, J. A. (1975). The Language of Thought. New York: Thomas Y. Crowell.
———. (1983). The Modularity of Mind. Cambridge: MIT Press.
———. (1990a) A Theory of Content and Other Essays. Cambridge: MIT Press.
———. (1990b). “Why Should the Mind be Modular?” in A Theory of Content and Other
Essays (pp. 207–230). Cambridge: MIT Press.
———. (1990c). “Observation Reconsidered,” in A Theory of Content and Other Essays
(pp. 231–51). Cambridge: MIT Press.
———. (1998a). In Critical Condition: Polemical Essays on Cognitive Science and the Phi-
losophy of Mind. Cambridge: MIT Press.
BIBLIOGRAPHY 259
———. (1998b). “Review of Christopher Peacocke’s A Study of Concepts” in In Critical

Condition: Polemical Essays on Cognitive Science and the Philosophy of Mind (pp. 27–
34). Cambridge: MIT Press.
Fodor, J. A., and LePore, E. (1992). Holism: A Shopper’s Guide. Oxford: Blackwell.
Fogelin, R. (1967). Evidence and Meaning. New York: Humanities Press.
———. (1988). Figuratively Speaking. New Haven: Yale University Press.
———. (1994). Pyrrhonian Reflections on Knowledge and Justification. New York: Oxford
University Press.
Foolen, A. (1991). “Metalinguistic Negation and Pragmatic Ambiguity: Some Comments on
a Proposal by Laurence Horn,” Pragmatics, 1: 137–57.
Fowler, H. W. (1965). A Dictionary of Modern English Usage. 2nd ed. New York: Oxford
University Press.
Frege, G. (1970a). “On Sense and Reference,” in P. Geach and M. Black (eds.), Translations
from the Philosophical Writings of Gottlob Frege (pp. 56–78). Oxford: Blackwell. Origi-
nally published 1892.
———. (1970b). “Negation,” in P. Geach and M. Black (eds.), Translation from the Philo-
sophical Writings of Gottlob Frege (pp. 117–35). Oxford: Blackwell. Originally pub-
lished 1919.
Fretheim, T. (1992). “The Effect of Intonation on a Type of Scalar Implicature,” Journal of
Pragmatics, 18: 1–30.
Frost, R. (1979). The Poetry of Robert Frost. New York: Henry Holt.
Gamut, L. T. F. (1991). Logic, Language, and Meaning: Vol. 2. Intensional Logic and Logi-
cal Grammar. Chicago: University of Chicago Press.
Gass, W. (1970). Fiction and the Figures of Life. New York: Knopf.
Gazdar, G. (1976). “Formal Pragmatics for Natural Language Implicature, Presupposition,
and Logical Form.” Ph.D. diss., University of Reading, U.K.
———. (1977). Implicature, Presupposition, and Logical Form. Bloomington: Indiana Uni-
versity Linguistics Club.
———. (1979a). Pragmatics: Implicature, Presupposition, and Logical Form. New York:
Academic Press.
———. (1979b). “A Solution to the Projection Problem,” in C-K Oh and D. Dinneen (eds.),
Syntax and Semantics 11: Presupposition (pp. 57–89). New York: Academic Press.
———. (1980). “Pragmatics and Logical Form,” Journal of Pragmatics, 4: 1–13.
Geach, D. T. (1972a). Logic Matters. Berkeley: University of California Press.
———. (1972b). “Assertion,” in Logic Matters (pp. 250–69). Berkeley: University of Cali-
fornia Press.
Geis, M., and Zwicky, A. M. (1971). “On Invited Inferences,” Linguistic Inquiry, 2: 561–65.
Geurts, B. (1998). “The Mechanisms of Denial,” Language, 74: 274–307.
Gillon, B. S. (1990). “Ambiguity, Generality, and Indeterminacy: Tests and Definitions,”
Synthese, 85: 391–416.
Glucksberg, S. (2001). Understanding Figurative Language. New York: Oxford University Press.
Goodman, N. (1976). Languages of Art. Indianapolis: Hackett.
———. (1983). Fact, Fiction, and Forecast. Cambridge: Harvard University Press.
Gordon, G. and Lakoff, G. (1971). “Conversational Postulates,” in CLS 7: Papers from the
Seventh Regional Meeting of the Chicago Linguistics Society (pp. 63–84). Chicago: Chi-
cago Linguistics Society. Repr. in P. Cole and J. L. Morgan (eds.), Syntax and Seman-
tics 3: Speech Acts (pp. 83–106). New York: Academic Press, 1975.
Grandy, R. (1973). “Reference, Meaning, and Belief,” Journal of Philosophy 70: 439–52.
———. (1987). “In Defense of Semantic Fields,” in E. Lepore (ed.), New Directions in Se-
mantics (pp. 259–80). New York: Academic Press.
260 BIBLIOGRAPHY
Grandy, R., and Warner, R. (eds.) (1986). Philosophical Grounds of Rationality. Oxford:
Clarendon Press.
Graves, R., and Hodge, A. (1979). The Reader over Your Shoulder: A Handbook for Writers
of English Prose, 2nd ed. New York: Random House.
Gregory, R. L. (1970). The Intelligent Eye. London: Weidenfeld and Nicolson.
———. (1973). “The Confounded Eye,” in R. L. Gregory and E. H. Gombrich (eds.), Illu-
sion in Nature and Art (pp. 49–95). London: Duckworth.
———. (1986). Odd Perceptions. London: Methuen.
Green, G. M. (1989). Pragmatics and Natural Language Understanding. Hillsdale, N.J.:
Erlbaum.
Grice, H. P. (1961). “The Causal Theory of Perception,” Proceedings of the Aristotelian
Society, Vol. 25(Suppl.): 121–52.
———. (1965). “The Causal Theory of Perception,” in R. J. Swartz (ed.), Perceiving, Sens-
ing, and Knowing: A Book of Readings from Twentieth-Century Sources in the Philoso-
phy of Perception (pp. 438–72). Garden City, N.Y.: Anchor Books, Doubleday.
———. (1967). “Logic and Conversation: The 1967 William James Lectures.” Harvard Uni-
versity, Cambridge, Massachusetts.
———. (1969). “Utterer’s Meaning and Intentions,” Philosophical Review, 78: 147–77. Repr.
in Grice (1989a), pp. 86–116.
———. (1975a). “Logic and Conversation,” in P. Cole and J. L. Morgan (eds.), Syntax and
Semantics 3: Speech Acts (pp. 41–58). New York: Academic Press.
———. (1975b). “Logic and Conversation,” in D. Davidson and G. Harman (eds.), The Logic
of Grammar (pp. 64–75). Encino, Calif.: Dickenson.
———. (1978). “Further Notes on Logic and Conversation,” in P. Cole (ed.), Syntax and
Semantics 9: Pragmatics (pp. 113–27). New York: Academic Press.
———. (1981). “Presupposition and Conversational Implicature,” in P. Cole (ed.), Radical
Pragmatics (pp. 183–97). New York: Academic Press.
———. (1986). “Reply to Richards,” in R. Grandy and R. Warner (eds.), Philosophical
Grounds of Rationality (pp. 45–106). Oxford: Clarendon.
———. (1989a). Studies in the Way of Words. Cambridge: Harvard University Press.
———. (1989b). “Logic and Conversation,” in H. P. Grice, Studies in the Way of Words
(pp. 22–40). Cambridge: Harvard University Press.
———. (1989c). “Further Notes on Logic and Conversation,” in H. P. Grice, Studies in the
Way of Words (pp. 41–57). Cambridge: Harvard University Press.
———. (1989d). “Some Models for Implicature,” in H. P. Grice, Studies in the Way of Words
———. (1989e). “Meaning,” in H. P. Grice, Studies in the Way of Words (pp. 213–23). Cam-
bridge: Harvard University Press.
———. (1989f). “Retrospective Epilogue,” in H. P. Grice, Studies in the Way of Words
Grinder, J., and Postal, P. M. (1971). “Missing Antecedents,” Linguistic Inquiry, 2: 269–312.
Grundy, P. (1995). Doing Pragmatics. London: Edward Arnold.
Guenthner, R., and Guenthner-Reutter, M. (eds.) (1978). Meaning and Translation. London:
Duckworth.
Hacker, P. M. S. (1996). Wittgenstein’s Place in Twentieth-Century Analytic Philosophy.
Oxford: Blackwell.
Hahn, L. E. and Schilpp, P. A. (eds.) (1986). The Philosophy of W. V. Quine. LaSalle, Ill.:
Open Court.
Halliday, M. A. K. (1967). “Notes on Transitivity and Theme in English, Part 2,” Journal of
Linguistics, 3: 199–244.
BIBLIOGRAPHY 261
Halvorsen, P.-K. (1978). The Syntax and Semantics of Cleft Constructions. Texas Linguistic
Forum, 11. Austin: Department of Linguistics, University of Texas.
Hanson, N. R. (1969). Perception and Discovery: An Introduction to Scientific Inquiry. San
Francisco: Freeman, Cooper.
Harman, G. H. (1965). “The Inference to the Best Explanation,” Philosophical Review, 74:
88–95.
Harnish, R. M. (1976). “Logical Form and Implicature,” in T. G. Bever, J. J. Katz, and D. T.
Langendoen (eds.), An Integrated Theory of Linguistic Ability (pp. 313–91). New York:
Crowell.
Harries-Delisle, H. (1978). Contrastive Emphasis and Cleft Sentences,” in J. Greenberg,
C. Ferguson, and E. Moravcsik (eds.), Universals of Human Language, Vol. 4: Syntax
(pp. 419–86). Stanford, Calif.: Stanford University Press.
Harris, R. A. (1993). The Linguistic Wars. New York: Oxford University Press.
Hawkins, J. A. (1975). “The Pragmatics of Definiteness,” in G. Gazdar and S. C. Levinson (eds.),
Pragmatics Microfiche (1.3: pp. C2–G10). Cambridge: Cambridge University Press.
———. (1978). Definiteness and Indefiniteness. London: Croom Helm.
Heim, I. (1982). “The Semantics of Definite and Indefinite Descriptions.” Ph.D. diss., De-
partment of Linguistics, University of Massachusetts, Amherst.
Hempel, C. G. (1960). “Inductive Inconsistencies,” Synthese, 11: 439–69.
Herbert, N. (1985). Quantum Reality. Garden City, N.Y.: Anchor/Doubleday.
Hilbert, D. (1927/1967). “The Foundations of Mathematics,” in J. van Heijenoort (ed.), From
Frege to Gödel: A Sourcebook in Mathematical Logic, 1879–1931 (pp. 464–79). Cam-
bridge: Harvard University Press.
Hintikka, K. J. J. (1973). “Grammar and Logic: Some Borderline Problems,” in K. J. J.
Hintikka, P. Suppes, J. M. E. Moravcsik (eds.) Approaches to Natural Language: Pro-
ceedings of the 1970 Stanford Workshop on Grammar and Semantics (pp. 197–214).
Dordrecht: Reidel.
Hirschberg, J. (1985). A Theory of Scalar Implicature. Technical Report MS-CIS-85–86.
Moore School of Electrical Engineering, Philadelphia: University of Pennsylvania.
———. (1991). A Theory of Scalar Implicature. New York: Garland.
Hitzeman, J. (1992). “The Selectional Properties and Entailments of ‘Almost’,” in C. P.
Canakis, G. P. Chan, and J. M. Denton, (eds.), CLS 28: Papers from the Twenty-eighth
Regional Meeting of the Chicago Linguistics Society, 1992: Vol. 1. The Main Session
(pp. 225–38). Chicago: Chicago Linguistics Society.
Hobbs, J., Stickel, M., Appelt, D., and Martin, P. (1993). “Interpretation as Abduction,”
Artificial Intelligence, 63: 69–142.
Hochberg, J. (1972). “The Representation of Things and People,” in E. H. Gombrich and
J. Hochberg, Art, Perception, and Reality (pp. 47–94). Baltimore: Johns Hopkins Uni-
versity Press.
Hopper, P., and Traugott, E. (1993). Grammaticalization. Cambridge: Cambridge Univer-
sity Press.
Horgan, P. (1974). Approaches to Writing. New York: Farrar Straus Giroux.
Horn, L. R. (1972). “On the Semantic Properties of Logical Operators in English,” Ph.D. diss.,
University of California, Los Angeles.
———. (1973). “Greek Grice: A Brief Survey of Proto-Conversational Rules in the History
of Logic,” in C. Corum, T. Smith-Stack, and A. Weiser (eds.), CLS 9: Papers from the
Ninth Regional Meeting of the Chicago Linguistics Society (pp. 205–14). Chicago: De-
partment of Linguistics, University of Chicago.
———. (1976). “On the Semantic Properties of Logical Operators in English.” Indiana Uni-
versity Linguistics Club. Bloomington: University of Indiana.
262 BIBLIOGRAPHY
———. (1978a). “Remarks on Neg-Raising,” in P. Cole (ed.), Syntax and Semantics 9: Prag-
matics (pp. 129–220). New York: Academic Press.
———. (1978b). “Some Aspects of Negation,” in J. Greenberg, C. Ferguson, and E. Moravcsik
(eds.), Universals of Human Language Vol. 4, (pp. 127–210). Stanford: Stanford Uni-
versity Press.
———. (1978c). “Lexical Incorporation, Implicature, and the Least Effort Hypothesis,” in
D. Farkas, W. Jacobsen, and K. Todrys (eds.), Parasession on the Lexicon (pp. 196–
209). Chicago: Chicago Linguistics Society.
———. (1981a). “Exhaustiveness and the Semantics of Clefts,” in V. Burke and J. Pustejovksy
(eds.), Proceedings of the Eleventh Annual New England Linguistics Society Confer-
ence, November 7–9, 1980 (pp. 125–42). Department of Linguistics, Amherst: Univer-
sity of Massachusetts.
———. (1981b). “A Pragmatic Approach to Certain Ambiguities,” Linguistics and Philoso-
phy, 4: 321–58.
———. (1984a). “Ambiguity, Negation, and the London School of Parsimony,” Proceed-
ings of the New England Linguistics Society, 14: 108–31.
———. (1984b). “Toward a New Taxonomy for Pragmatic Inference: Q-Based and R-Based
Implicature,” in D. Schiffrin (ed.), Georgetown University Round Table on Languages
and Linguistics 1984. Meaning, Form, and Use in Context: Linguistic Applications
(pp. 11–42). Washington, D.C. Georgetown University Press.
———. (1985). “Metalinguistic Negation and Pragmatic Ambiguity,” Language, 61: 121–74.
———. (1989). A Natural History of Negation. Chicago: University of Chicago Press.
———. (1990). “Showdown at Truth-Value Gap: Burton-Roberts on Presupposition” [Re-
view], Journal of Lingusitics, 26: 483–503.
———. (1992a). “Pragmatics, Implicature, and Presupposition,” in W. Bright (ed.), Interna-
tional Encyclopedia of Linguistics, (Vol. 3, pp. 260–66). New York: Oxford University
Press.
———. (1992b). “The Said and the Unsaid,” in C. Barker and D. Dowty (eds.), SALT II
Proceedings of the Second Conference on Semantics and Linguistic Theory, May 1–3,
1992 (pp. 163–92). Ohio State University Working Papers in Linguistics No. 40. Co-
lumbus: Ohio State University, Department of Linguistics.
———. (1993). “Economy and Redundancy in a Dualistic Model of Natural Language,”
Suomen kielitieteellinen yhdistys (SKY) 1993 (pp. 33–72). Helsinki: Linguistic Associa-
tion of Finland.
———. (1996a). “Comments.” Unpublished ms., Department of Linguistics, Yale Univer-
sity, New Haven.
———. (1996b). “Exclusive Company: Only and the Dynamics of Vertical Inference,” Journal
of Semantics, 13(1):1–40.
———. (2002). “Assertoric Inertia and NPI Licensing.” Lecture, Annual Meeting of the
Chicago Linguistics Society, April 2002, Chicago, Ill.
Hornstein, N. (1989). “Meaning and the Mental: The Problem of Semantics after Chomsky,”
in A. George (ed.), Reflections on Chomsky (pp. 23–40). Oxford: Blackwell.
Huang, Y. (1994). The Syntax and Pragmatics of Anaphora: A Study with Special Reference
to Chinese. Cambridge Studies in Linguistics 70. Cambridge: Cambridge University
Press.
———. (2000). Anaphora: A Cross-Linguistic Study. Oxford: Oxford University Press.
Iten, C. (1998). “Because and although: A Case of Duality?” in V. Rouchota and A. H. Jucker
(eds.), Current Issues in Relevance Theory (pp. 59–80). Amsterdam: John Benjamins.
Jacobs, R. A., and Rosenbaum, P. S. (eds.) (1970). Readings in Transformational Grammar.
Waltham, Mass.: Ginn.
BIBLIOGRAPHY 263
Johnson, M. (1981). Philosophical Perspectives on Metaphor. Minneapolis: University of

Minnesota Press.
Kamp, H. (1981). “A Theory of Truth and Semantic Representation,” in J. Groendijk,
T. Janssen, and M. Stokhof (eds.), Formal Methods in the Study of Language (Vol. 1,
pp. 277–321). Amsterdam: Mathematical Centrum.
Kamp, H., and Reyle, U. (1993). From Discourse to Logic. Dordrecht: Kluwer.
Karttunen, L. (1973). “Presuppositions of Compound Sentences,” Linguistic Inquiry, 4: 169–
93.
———. (1974). “Presupposition and Linguistic Context,” Theoretical Linguistics, 1: 182–94.
Karttunen, L., and Peters, S. (1979). “Conventional Implicature,” in C-K Oh, and D. Dinneen
(eds.), Syntax and Semantics 11: Presupposition (pp. 1–56). New York: Academic Press.
Katz, J. J. (1972). Semantic Theory. New York: Harper and Row.
———. (1978). “Effability and Translation,” in F. Guenthner and M. Guenthner-Reutter (eds.),
Meaning and Translation (pp. 191–234). London: Duckworth.
———. (1981). Language and Other Abstract Objects. Oxford: Blackwell.
———. (1990). The Metaphysics of Meaning. Cambridge: MIT Press.
Kay, P., and Zimmer, P. (1976). “On the Semantics of Compounds and Genitives in English,”
in R. Underhill (ed.), Sixth California Linguistics Association Proceedings (pp. 29–35).
San Diego: Campanile.
Keefe, R., and Smith, P. (eds.) (1996). Vagueness. Cambridge: MIT Press.
Keenan, E. (1978). “Some Logical Problems in Translation,” in F. Guenthner and
M. Guenthner-Reutter (eds.), Meaning and Translation (pp. 157–89). London: Duckworth.
Kemeny, J. (1953). “A Logical Measure Function,” Journal of Symbolic Logic, 18: 289–308.
Kempson, R. (1975). Presupposition and the Delimitation of Semantics. Cambridge: Cam-
bridge University Press.
———. (1977). Semantic Theory. Cambridge: Cambridge University Press.
———. (1986). “Ambiguity and the Semantics-Pragmatics Distinction,” in C. Travis (ed.),
Meaning and Interpretation (pp. 77–103). Oxford: Blackwell.
———. (1988a). “Grammar and Conversational Principles,” in F. J. Newmeyer (ed.), Lin-
guistics: The Cambridge Survey. Vol. 2. Linguistic Theory: Extensions and Implications
———. (ed.) (1988b). Mental Representations: The Interface between Language and Real-
ity. Cambridge: Cambridge University Press.
Kempson, R., and Cormack, A. (1981). “Ambiguity and Quantification,” Linguistics and
Philosophy, 4: 259–310.
Kennedy, C. (2001). “On the Monotonicity of Polar Adjectives,” in J. Hoeksema, H. Rullman,
V. Sánchez-Valencia, and T. van der Wouden (eds.), Perspectives on Negation and Po-
larity Items (pp. 201–21). Amsterdam: John Benjamins.
Kenny, A. (1963). Action, Emotion, and Will. London: Routledge and Kegan Paul.
———. (1966). “Practical Inference,” Analysis 26: 65–75.
Kiparsky, P. (1982). “Word Formation and the Lexicon,” in F. Ingemann (ed.), Proceedings
of the 1982 Mid-America Linguistics Conference. Lawrence: University Press of
Kansas.
Kiparsky, P., and Kiparsky, C. (1971). “Fact,” in D. D. Steinberg and L. A. Jakobovits (eds.),
Semantics, (pp. 345–69). Cambridge: Cambridge University Press.
Kittay, E. (1987). Metaphor: Its Cognitive Force and Linguistic Structure. Oxford: Clarendon.
Klein, E. (1980). “A Semantics for Positive and Comparative Adjectives,” Linguistics and
Klein, H. (1997). “Adverbs of Degree in Dutch.” Ph.D. diss. University of Groningen, The
Netherlands.
264 BIBLIOGRAPHY
———. (1998). Adverbs of Degree in Dutch and Related Languages. Amsterdam: John
Benjamins.
Koenig, J-P. (1992). “Scalar Predicates and Negation,” in L. M. Dobrin, L. Nichols, and
R. M. Rodriquez (eds.), CLS 27: Papers from the Twenty-seventh Regional Meeting of
the Chicago Linguistics Society 1991. Part 2: The Parasession on Negation (pp. 140–
König, E. (1988). “Concessive Connectives and Concessive Sentences: Cross Linguistic
Regularities and Pragmatic Principles,” in J. Hawkins (ed.), Explaining Language Uni-
versals (pp. 145–66). Oxford: Blackwell.
———. (1991). The Meaning of Focus Particles: A Comparative Perspective. London: Croom
Helm.
Kooij, J. (1971). Ambiguity in Natural Language. Amsterdam: North-Holland.
Kuroda, S.-Y. (1977). “Description of Presuppositional Phenomena from a Non-Presuppositional
Point of View,” Lingvisticae Investigationes, 1: 63–162.
Ladd, D. (1980). The Structure of Intonational Meaning. Bloomington: Indiana University
Press.
Lakoff, G. (1970). “A Note on Ambiguity and Vagueness,” Linguistic Inquiry, 1: 357–59.
———. (1971). “On Generative Semantics,” in D. D. Steinberg and L. A. Jakobovits (eds.),
Semantics: An Interdisciplinary Reader in Philosophy, Linguistics, and Psychology
———. (1975). “Pragmatics and Natural Logic,” in E. L. Keenan (ed.), Formal Semantics of
Natural Language (pp. 253–86). Cambridge: Cambridge University Press.
———. (1977). “Linguistic Gestalts,” in W. Beach, S. Fox, and S. Philosoph (eds.), CLS 13:
Papers from the Thirteenth Regional Meeting of the Chicago Linguistics Society (pp. 236–
———. (1986). Two Metametaphorical Issues. Berkeley Cognitive Science Report No. 38.
Berkeley: University of California, Institute of Cognitive Studies.
———. (1987). Women, Fire, and Dangerous Things. Chicago: University of Chicago Press.
Lakoff, G., and Johnson, M. (1980). Metaphors We Live By. Chicago: University of Chicago
Press.
Langendoen, D. T. (1971). “Presupposition and Assertion in the Semantic Analysis of Nouns
and Verbs in English,” in D. Steinberg and L. Jakobovits (eds.), Semantics: An Interdis-
ciplinary Reader in Philosophy, Linguistics, and Psychology (pp. 341–44). Cambridge:
Cambridge University Press.
Langford, C. H. (1968). “The Notion of Analysis in Moore’s Philosophy,” in P. A. Schilpp,
(ed.), The Philosophy of G. E. Moore (Vol. 1, pp. 321–42). LaSalle, Ill.: Open Court.
Originally published in 1942.
Leech, G. (1974). Semantics. London: Penguin.
———. (1983). Principles of Pragmatics. London: Longman.
———. (1990). Semantics. 2nd ed. London: Penguin.
Leezenberg, M. (1995). “Contexts of Metaphor.” Ph.D. diss., Institute for Language, Logic,
and Computation, University of Amsterdam.
———. (2001). Contexts of Metaphor. Oxford: Elsevier.
Lehrer, A. (1974). Semantic Fields and Lexical Structure. Amsterdam: North Holland.
Leisenring, A. C. (1969). Mathematical Logic and Hilbert’s ε-symbol. London: Macdonald.
Levinson, S. C. (1978). “Pragmatics and Social Deixis.” Unpublished ms., University of
Cambridge.
———. (1983). Pragmatics. Cambridge: Cambridge University Press.
———. (1987a). “Minimization and Conversational Inference,” in J. Verschueren and
M. Bertucelli-Papi (eds.), The Pragmatic Perspective, (pp. 61–129). Amsterdam: John
BIBLIOGRAPHY 265
Benjamins. Repr. in A. Kasher (ed.), Pragmatics: Critical Concepts (Vol. 4. pp. 545–
612). (London: Routledge, 1988).
———. (1987b). “Pragmatics and the Grammar of Anaphora: A Partial Pragmatic Reduc-
tion of Binding and Control Phenomena,” Journal of Linguistics, 23: 379–434.
———. (1988a). “The Nijmegen Lectures.” University of Nijmegen, The Netherlands.
———. (1988b). “Generalized Conversational Implicature and the Semantics/Pragmatics
Interface.” Typescript. Department of Linguistics, Stanford University.
———. (1989). “Relevance,” Journal of Linguistics, 21: 455–72.
———. (1990). “Interactional Biases in Human Thinking.” Presented at the Workshop on
the Social Origins of Human Intelligence, Wissenschaftcolleg zu Berlin, May 1990.
———. (1991). “Pragmatic Reduction of the Binding Conditions Revisited,” Journal of Lin-
guistics, 27: 107–61.
———. (1995). “Three Levels of Meaning,” in F. R. Palmer (ed.), Grammar and Meaning:
Essays in Honour of Sir John Lyons (pp. 90–115). Cambridge: Cambridge University
Press.
———. (1997). “From Outer to Inner Space: Linguistic Categories and Non-Linguistic Think-
ing,” in J. Nuyts and E. Pederson (eds.), Language and Conceptualization (pp. 13–45).
Cambridge: Cambridge University Press.
———. (2000). Presumptive Meanings: The Theory of Generalized Conversational Implica-
tures. Cambridge: MIT Press.
Lewis, D. K. (1969). Convention. Cambridge: Harvard University Press.
———. (1979). “Scorekeeping in a Language Game,” Journal of Philosophical Logic, 8:
339–59.
Liberman, M., and Sag, I. (1974). “Prosodic Form and Discourse Function,” in M. LaGaly,
R. Fox, A. Bruck (eds.), CLS 10: Papers from the Tenth Regional Meeting of the Chi-
cago Linguistics Society (pp. 402–15). Chicago: Chicago Linguistics Society.
Linsky, L. (ed.) (1952). Semantics and the Philosophy of Language. Urbana: University of
Illinois Press.
Locke, J. (1975). An Essay Concerning Human Understanding. Edited by P. H. Nidditch.
Oxford: Oxford University Press. Originally published 1690.
Lodge, D. (1996). The Practice of Writing. New York: Penguin.
Lyons, J. (1977). Semantics. Cambridge: Cambridge University Press.
———. (1995a). Linguistic Semantics. Cambridge: Cambridge University Press.
———. (1995b). “Grammar and Meaning,” in F. R. Palmer (ed.), Grammar and Meaning:
Essays in Honour of Sir John Lyons (pp. 221–249). Cambridge: Cambridge University
Press.
Marti, G. (1993). “The Source of Intensionality,” in J. E. Tomberlin (ed.), Philosophical
Perspectives 7: Language and Logic. (pp. 197–206). Atascadero, Calif.: Ridgeview.
Martin, J. N. (1979). “Some Misconceptions in the Critique of Semantic Presupposition,”
Theoretical Linguistics, 6: 235–82.
Martin, R. M. (1958). Truth and Denotation. Chicago: University of Chicago Press.
———. (1979). Pragmatics, Truth, and Language. Dordrecht: Reidel.
Martinich, A. P. (1980). “Conversational Maxims and Some Philosophical Problems,” Philo-
sophical Quarterly, 30: 215–28.
Matsumoto, Y. (1995). “The Conversational Condition on Horn Scales,” Linguistics and
McCawley, J. (1978). “Conversational Implicature and the Lexicon,” in P. Cole (ed.), Syntax
and Semantics 9: Pragmatics (245–59). New York: Academic Press.
———. (1981). Everything That Linguists Have Always Wanted to Know about Logic: But
Were Ashamed to Ask. Chicago: University of Chicago Press.
266 BIBLIOGRAPHY
———. (1991). “Contrastive Negation and Metalinguistic Negation,” in L. Dobrin, L. Nichols,

and R. Rodriguez (eds.), CLS 27: Papers from the Twenty-seventh Regional Meeting of
the Chicago Linguistics Society. Part 2: The Parasession on Negation (pp. 331–44). Chi-
cago: Chicago Linguistics Society.
———. (1993). Everything That Linguists Have Always Wanted to Know about Logic: But
Were Ashamed to Ask. 2nd ed. Chicago: University of Chicago Press.
Mey, J. (1993). Pragmatics: An Introduction. Oxford: Blackwell.
Miller, G. A. and Johnson-Laird, P. N. (1976). Language and Perception. Cambridge: Harvard
University Press.
Montague, R. (1974). “English as a Formal Language,” in R. Thomason (ed.), Formal Phi-
losophy: Selected Papers of Richard Montague (pp. 188–221). New Haven: Yale Uni-
versity Press.
Moore, G. E. (1953). Some Main Problems of Philosophy. London: George Allen and Unwin.
Lectures from 1910–11.
———. (1959). Philosophical Papers. London: George Allen and Unwin.
———. (1968). “A Reply to My Critics,” in P. A. Schilpp (ed.), The Philosophy of G. E.
Moore. Vol. 2, (pp. 535–677). LaSalle, Ill.: Open Court. Originally published in 1942.
———. (1993). G. E. Moore: Selected Writings. Ed. by Thomas Baldwin. London: Routledge.
Morgan, J. L. (1977). “Conversational Postulates Revisited,” Language, 53: 277–84.
———. (1978). “Two Types of Convention in Indirect Speech Acts,” in P. Cole (ed.), Syn-
tax and Semantics 9: Pragmatics (pp. 261–80). New York: Academic Press.
Mortimer, J. (2000). The Summer of a Dormouse. New York: Viking.
Muka6ovský, J. (1970). “Standard Language and Poetic Language,” in D. C. Freeman (ed.),
Linguistics and Literary Style (pp. 40–56). New York: Holt, Rinehart and Winston.
Munitz, M. K. and Unger, P. K. (eds.) (1974). Semantics and Philosophy. New York: New
York University Press.
Nagel, T. (1998). “The Sleep of Reason.” Review of Sokal and Bricmont (1998), New Re-
public, 12 October: 32–38.
Neale, S. (1992). “Grice and the Philosophy of Language,” Linguistics and Philosophy, 15:
509–59.
Newmeyer, F. J. (1986). Linguistics Theory in America. 2nd ed. New York: Academic Press.
———. (1996). Generative Linguistics: A Historical Perspective. London: Routledge.
Nowell-Smith, P. H. (1954). Ethics. Harmondsworth: Pelican.
Nuyts, J., and Pederson, E. (eds.) (1997). Language and Conceptualization. Cambridge:
Cambrdige University Press.
Oh, C-K, and Dinneen, D. (eds.) (1979). Syntax and Semantics 11: Presupposition. New York:
Academic Press.
O’Hair, S. G. (1969). “Implication and Meaning,” Theoria, 35: 38–54.
Ortony, A. (ed.) (1993). Metaphor and Thought. 2nd ed. Cambridge: Cambridge University
Press.
Peacocke, C. (1992). A Study of Concepts. Cambridge: MIT Press.
Popper, K. (1959). The Logic of Scientific Discovery. New York: Basic Books.
———. (1963). Conjectures and Refutations. New York: Basic Books.
Price, H. H. (1932). Perception. London: Methuen.
Prince, E. (1978). “A Comparison of WH-Clefts and it-Clefts in Discourse,” Language, 54:
893–906.
Pulman, S. G. (1983). Word Meaning and Belief. London: Croom Helm.
Putnam, H. (1958). “Formalization of the Concept ‘About’,” Philosophy of Science, 25: 125–
30.
———. (1975). “The Meaning of ‘Meaning’,” in K. Gunderson (ed.), Minnesota Studies in
BIBLIOGRAPHY 267
the Philosophy of Science: 7. Language, Mind, and Knowledge (pp. 131–93). Minne-
apolis: University of Minnesota Press.
———. (1976). “‘Two Dogmas’ Revisited,” in G. Ryle (ed.), Contemporary Aspects of Phi-
losophy (pp. 202–13). Stocksfield: Oriel Press.
———. (1978a). “Meaning, Reference, and Stereotypes,” in F. Guenthner and M. Guenthner-
Reutter (eds.), Meaning and Translation: Philosophical and Linguistic Approaches
(pp. 61–81). London: Duckworth.
———. (1978b). Meaning and the Moral Sciences. London: Routledge and Kegan Paul.
———. (1978c). “Reference and Understanding,” in H. Putnam, Meaning and the Moral
Sciences (pp. 97–119). London: Routledge and Kegan Paul.
———. (1983a). Philosophical Papers: Vol. 3. Realism and Reason. Cambridge: Cambridge
University Press.
———. (1983b). “‘Two Dogmas’ Revisited,” in H. Putnam, Philosophical Papers: Vol. 3.
Realism and Reason (pp. 87–97). Cambridge: Cambridge University Press.
———. (1992a). Renewing Philosophy. Cambridge: Harvard University Press.
———. (1992b). “A Theory of Reference,” in H. Putnam, Renewing Philosophy (pp. 35–
59). Cambridge: Harvard University Press.
Quine, W. V. O. (1956). “Quantifiers and Propositional Attitudes,” Journal of Philosophy,
53: 177–87. Repr. in Quine (1976), pp. 185–96.
———. (1960). Word and Object. Cambridge: MIT Press.
———. (1969). “Ontological Relativity,” in W. V. O. Quine, Ontological Relativity and Other
Essays (pp. 26–68). New York: Columbia University Press.
———. (1974). “First General Discussion,” Synthese, 27: 471–508.
———. (1976). The Ways of Paradox and Other Essays. Rev. and enl. ed. Cambridge: Harvard
University Press.
———. (1980a). From a Logical Point of View. 2nd ed. rev. Cambridge: Harvard University
Press.
———. (1980b). “On What There Is,” in W. V. O. Quine, From a Logical Point of View
(pp. 1–19). Cambridge, Massachusetts: Harvard University Press.
———. (1980c). “Two Dogmas of Empiricism,” in W. V. O. Quine, From a Logical Point
of View (pp. 20–46). Cambridge: Harvard University Press.
———. (1981). Mathematical Logic. Rev. ed. Cambridge: Harvard University Press.
Quine, W. V. O., and Ullian, J. (1978). Web of Belief. New York: Random House.
Raichle, M. E. (1994). “Visualizing the Mind,” Scientific American, 270(4 April): 58–64.
Raichle, M. E., Fiez, J. A., Videen, T. O., MacLeod, A.-M. K., Pardo, J. V., Fox, P. T., and
Petersen, S. E. (1994). “Practice-Related Changes in Human Functional Anatomy dur-
ing Non-Motor Learning,” Cerebral Cortex, 4(January/February): 8–26.
Récanati, F. (1989). “The Pragmatics of What Is Said,” Mind and Language, 4: 295–329.
Repr. in S. Davis, Pragmatics (New York: Oxford University Press, 1991), pp. 97–120.
———. (1993). Direct Reference: From Language to Thought. Oxford: Blackwell.
Reinhart, T. (1983). Anaphora and Semantic Interpretation. London: Croom Helm.
Rorty, R. (1979). Philosophy and the Mirror of Nature. Princeton: Princeton University Press.
———. (1992). The Linguistic Turn. Chicago: University of Chicago Press. Originally pub-
lished 1967.
———. (1998a). Truth and Progress. Philosophical Papers Vol. 3. Cambridge: Cambridge
University Press.
———. (1998b). “Dewey between Hegel and Darwin,” in R. Rorty, Truth and Progress
Rosch, E. (1977). “Human Categorization,” in N. Warren (ed.), Advances in Cross-cultural
Psychology (Vol. 1, pp. 1–49). New York: Academic Press.
268 BIBLIOGRAPHY
Rosenthal, D. (1993). “Thinking That One Thinks,” in M. Davies and Glyn W. Humphreys
(eds.), Consciousness (pp. 197–223). Oxford: Blackwell.
Ross, J. R. (1970). “On Declarative Sentences,” in R. A. Jacobs and P. S. Rosenbaum (eds.),
Readings in Transformational Grammar (pp. 222–72). Waltham, Mass.: Giun.
Ruhl, C. (1989). On Monosemy: A Study in Linguistic Semantics. Albany: State University
of New York Press.
Russell, B. (1903). Principles of Mathematics. Cambridge: Cambridge University Press.
———. (1905). “On Denoting,” Mind, 14: 479–93. Repr. in Russell (1956), pp. 41–56.
———. (1913). Problems of Philosophy. Oxford: Oxford University Press.
———. (1919). Introduction to Mathematical Philosophy. London: George Allen and Unwin.
———. (1956a). Logic and Knowledge: Essays 1901—1950. Ed. R. Marsh. London: George
Allen and Unwin.
———. (1956b). “Mathematical Logic as Based on the Theory of Types,” in B. Russell Logic
and Knowledge: Essays 1901–1950 (pp. 59–102). Ed. R. Marsh. London: George Allen
and Unwin. Originally published 1908.
———. (1959). My Philosophical Development. New York: Simon and Schuster.
Ryle, G. (1949). The Concept of Mind. London: Hutchinson.
Sacks, H. (1972). “On the Analyzability of Stories by Children,” in J. Gumperz and D. Hymes
(eds.), Directions in Sociolinguistics: The Ethnography of Communication (pp. 325–45).
New York: Holt, Rinehart and Winston.
Sacks, S. (ed.) (1981). On Metaphor. Chicago: University of Chicago Press.
Sadock, J. M. (1975). “Larry Scores a Point,” Pragmatics Microfiche (1.4: pp. G10–G13).
Cambridge: Cambridge University Press.
———. (1978). “On Testing for Conversational Implicature,” in P. Cole (ed.), Syntax and
Semantics 9: Pragmatics (pp. 281–97). New York: Academic Press.
———. (1981). “Almost,” in P. Cole (ed.), Radical Pragmatics (pp. 281–97). New York:
Academic Press.
———. (1984). “Whither Radical Pragmatics?” in D. Schiffrin (ed.), Georgetown Univer-
sity Round Table on Languages and Linguistics (pp. 139–49). Washington, D.C.:
Georgetown University Press.
Sanford, D. (2003). If P, then Q: Conditionals and the Foundations of Reasoning. 2nd ed.
London: Routledge.
Scharten, R. (1997). “Exhaustive Interpretation: A Discourse-Semantic Account,” Ph.D. diss.,
Catholic University Nijmegen, The Netherlands.
Schelling, T. (1960). The Strategy of Conflict. Cambridge: MIT Press.
Schiffer, S. (1972). Meaning. Oxford: Clarendon Press.
———. (1987). Remnants of Meaning. Cambridge: MIT Press.
Schmerling, S. F. (1975). “Asymmetric Conjunction and Rules of Conversation,” in P. Cole
and J. L. Morgan (eds.), Syntax and Semantics 3: Speech Acts (pp. 211–32). New York:
Academic Press.
Scott, D. S. (1970). “Advice on Modal Logic,” in K. Lambert (ed.), Philosophical Problems
in Logic: Recent Developments (pp. 143–73). Dordrecht: The Netherlands: Reidel.
———. (1971). “On Engendering an Illusion of Understanding,” Journal of Philosophy, 68:
787–807.
———. (1973). “Background to Formalization,” in H. Leblanc (ed.), Truth, Syntax, and
Modality (pp. 244–73). Amsterdam: North-Holland.
Searle, J. (1969a). Speech Acts. Cambridge: Cambridge University Press.
———. (1969b). “Assertions and Aberrations,” in K. T. Fann (ed.), Symposium on J. L. Aus-
tin (pp. 205–18). London: Routledge & Kegan Paul.
BIBLIOGRAPHY 269
———. (1979a). Expression and Meaning: Studies in the Theory of Speech Acts. Cambridge:
Cambridge University Press.
———. (1979b). “Literal Meaning,” in J. Searle, Expression and Meaning: Studies in the
Theory of Speech Acts (pp. 117–36). Cambridge: Cambridge University Press.
———. (1983). Intentionality. Cambridge: Cambridge University Press.
———. (1992). The Rediscovery of the Mind. Cambridge: MIT Press.
———. (1995). The Construction of Social Reality. London: Allen Lane, Penguin Press.
———. (1998). Mind, Language, and Society. New York: Basic Books.
Sellars, W. (1963). Science, Perception, and Reality. New York: Humanitics Press.
Seuren, P. (1973). “The Comparative,” in F. Kiefer and N. Ruwet (eds.), Generative Gram-
mar in Europe (pp. 528–64). Dordrecht: Reidel.
———. (1985). Discourse Semantics. Oxford: Blackwell.
———. (1988). “Presupposition and Negation,” Journal of Sematnics, 6: 175–226.
———. (1990). “Burton-Roberts on Presupposition and Negation,” Journal of Linguistics
26: 425–53.
———. (1993). “Why Does 2 mean ‘2’? Grist to the anti-Grice Mill,” in E. Hajicová (ed.),
Functional Description of Language (pp. 225–35). Proceedings of the Conference on
Functional Description of Language in Prague, November 24–27, 1992. Prague: Fac-
ulty of Mathematics and Physics, Charles University.
———. (2000). “Presupposition, Negation, and Trivalence,” Journal of Linguistics, 36: 261–
97.
Sinnott-Armstrong, W. (1984) “‘Ought’ Conversationally Implies ‘Can’,” Philosophical
Review, 93: 249–61.
Smokler, H. (1966). “Informational Content: A Problem of Definition,” Journal of Philoso-
phy, 63: 201–11.
Soames, S. (1989). “Presupposition,” in D. Gabbay and G. Guenthner (eds.), Handbook of
Philosophical Logic, Vol. 4 (pp. 553–616). Dordrecht: Reidel.
Sokal, A., and Bricmont, J. (1998). Fashionable Nonsense: Postmodern Intellectuals’ Abuse
of Science. New York: Picador.
Sommers, F. (1982). The Logic of Natural Language. Oxford: Clarendon.
Sperber, D., and Wilson, D. (1986a). “Loose Talk,” Proceedings of the Aristotelian Society,
86: 153–71. Repr. in Davis (1991), pp. 540–49.
———. (1986b). Relevance: Communication and Cognition. Oxford: Blackwell.
———. (1995). Relevance: Communication and Cognition. 2nd ed. Oxford: Blackwell.
Stalnaker, R. C. (1968). “A Theory of Conditionals,” in N. Rescher (ed.), Studies in Logical
Theory (pp. 98–112). American Philosophical Quarterly Monograph Series, No. 2.
Oxford: Blackwell.
———. (1972). “Pragmatics,” in D. Davidson and G. Harman (eds.), Semantics of Natural
Language (pp. 380–97). Dordrecht: Reidel.
———. (1974). “Pragmatic Presuppositions,” in M. K. Munitz and P. K. Unger (eds.), Se-
mantics and Philosophy (pp. 197–213). New York: New York University Press.
———. (1999). Context and Content. New York: Oxford University Press.
Steinberg, D. D. and Jokobovits, L. A. (eds.) (1971). Semantics: An Interdisciplinary Reader
in Philosophy, Linguistics, and Psychology. Cambridge: Cambridge University Press.
Steiner, G. (1975a). After Babel. New York: Oxford University Press.
———. (1975b). “Word against Object,” in G. Steiner, After Babel (pp. 110–235). New York:
Oxford University Press.
Stern, J. (1983). “Metaphor and Grammatical Deviance,” Noûs, 17: 577–99.
———. (1985). “Metaphor as Demonstrative.” Journal of Philosophy, 82: 677–710.
270 BIBLIOGRAPHY
———. (1991). “What Metaphors Do Not Mean,” in P. A. French, T. E. Uehling Jr., and
H. K. Wettstein (eds.), Midwest Studies in Philosophy 16: (13–52). Minneapolis: Uni-
versity of Minnesota Press.
———. (2000). Metaphor in Context. Cambridge: MIT Press.
Stich, S. P. (1983). From Folk Psychology to Cognitive Science: The Case against Belief.
Cambridge: MIT Press.
Strawson, P. F. (1950). “On Referring,” Mind, 59: 320–44. Repr. in Strawson (1971a),
pp. 1–27.
———. (1952). Introduction to Logical Theory. London: Methuen.
———. (1954). “A Reply to Mr Sellars,” Philosophical Review, 63: 216–31.
———. (1964). “Identifying Reference and Truth-values,” Theoria, 30: 96–118. Repr. in
Strawson (1971a), pp. 75–95.
———. (1971a). Logico-Linguistic Papers. London: Methuen.
———. (1971b). “On Referring,” in P. F. Strawson, Logico-Linguistic Papers (pp. 1–27).
London: Methuen.
———. (1971c). “Identifying Reference and Truth-values,” in P. F. Strawson, Logico-
Linguistic Papers (pp. 75–95). London: Methuen.
Taylor, M. (1998). Truth and Meaning. Oxford: Blackwell.
Thomas, J. (1995). Meaning in Interaction: An Introduction to Pragmatics. London: Longman.
Thomason, R. H. (1973). “Philosophy and Formal Semantics,” in H. Leblanc (ed.), Truth,
Syntax, and Modality (pp. 294–307). Amsterdam: North Holland.
———. (1984). “Accomodation, Meaning, and Implicature: Interdisciplinary Foundations
for Pragmatics.” Unpublished ms., University of Pittsburgh.
———. (1990). “Accomodation, Meaning, and Implicature: Interdisciplinary Foundations
for Pragmatics,” in P. R. Cohen, J. Morgan, and M. E. Pollack (eds.), Intentions in Com-
munication (pp. 325–64). Cambridge: MIT Press.
Tuchman, B. (1978). A Distant Mirror: The Calamitous Fourteenth Century. New York:
Knopf.
Ullman, S. (1979). The Interpretation of Visual Motion. Cambridge: MIT Press.
Valéry, P. (1961). The Art of Poetry. New York: Vintage.
van der Auwera, J. (1997). “Conditional Perfection,” in A. Athanasiadou and R. Dirven (eds.)
On Conditionals Again (pp. 169–90). Amsterdam: John Benjamins.
van der Sandt, R. (1988). Context and Presupposition. London: Croom Helm.
———. (1991). “Denial,” in L. Dobrin, L. Nichols, and R. Rodriguez (eds.), CLS 27: Papers
from the Twenty-Seventh Regional Meeting of the Chicago Linguistics Society. Part Two:
the Parasession on Negation (pp. 331–44). Chicago: Chicago Linguistics Society.
van Fraassen, B. (1971). Formal Semantics and Logic. New York: Macmillan.
van Kuppevelt, J. (1996). “Inferring from Topics: Scalar Implicatures as Topic-dependent
Inferences,” Linguistics and Philosophy, 19: 393–43.
Vendler, Z. (1967). Linguistics in Philosophy. Ithaca: Cornell University Press.
———. (1975a). “On What We Know,” in K. Gunderson (ed.), Minnesota Studies in the Phi-
losophy of Science: 7. Language, Mind, and Knowledge (pp. 370–90). Minneapolis: Uni-
versity of Minnesota Press.
———. (1975b). “Reply to Professor Aune,” in K. Gunderson (ed.), Minnesota Studies in
the Philosophy of Science: 7. Language, Mind, and Knowledge (pp. 400–402). Minne-
apolis: University of Minnesota Press.
von Klopp, A. (1993). “Negation: Implications for Theories of Natural Language.” Ph.D. diss.,
University of Edinburgh, Scotland.
Walker, R. C. (1975). “Conversational Implicatures,” in S. Blackburn (ed.) Meaning, Refer-
ence, and Necessity (pp. 133–81). Cambridge: Cambridge University Press.
BIBLIOGRAPHY 271
Warnock, G. (1973). “Saturday Mornings,” in I. Berlin, L. Forguson, D. Pears, G. Pitcher,

J. Searle, P. Strawson, and G. Warnack. Essays on J. L. Austin (pp. 31–45). Oxford:
Clarendon.
Weydt, H. (1972). “Le Concept d’ambiguité en grammaire transformationelle-générative et
en linguistique fonctionelle,” La Linguistique, 8: 41–72.
White, M. G. (1950). “The Analytic and the Synthetic: An Untenable Dualism,” in S. Hook
(ed.), John Dewey: Philosopher of Science and Freedom (pp. 316–30). New York: Dial.
Repr. in L. Linsky (1952), 272–86, and in White (1973), 121–37.
———. (1952). “The Analytic and the Synthetic: An Untenable Dualism,” in L. Linsky (ed.),
Semantics and the Philosophy of Language (pp. 272–86). Urbana: University of Illinois
Press.
———. (1956). Toward Reunion in Philosophy. Cambridge: Harvard University Press.
———. (1963a). Toward Reunion in Philosophy. New York: Atheneum.
———. (1963b). “The Analytic and the Synthetic,” in M. G. White, Toward Reunion in
Philosophy (pp. 133–47). New York: Atheneum.
———. (1965). Foundations of Historical Knowledge. New York: Harper and Row.
———. (1973a). Pragmatism and the American Mind. New York: Oxford University Press.
———. (1973b). “The Analytic and the Synthetic: An Untenable Dualism,” in M. G. White,
Pragmatism and the American Mind: Essays and Reviews in Philosophy and Intellec-
tual History (pp. 121–37). New York: Oxford University Press.
———. (1979). “Oughts and Cans,” in A. Ryan (ed.), The Idea of Freedom: Essays in Honour
of Isaiah Berlin (pp. 211–19). New York: Oxford University Press.
———. (1981). What Is and What Ought to Be Done. New York: Oxford University Press.
———. (1986). “Normative Ethics, Normative Epistemology, and Quine’s Holism,” in L. E.
Hahn and P. A. Schillp (eds.), The Philosophy of W. V. Quine. (pp. 649–62). La Salle,
Ill.: Open Court.
———. (1993). The Question of Free Will: A Holistic View. Princeton: Princeton University
Press.
Whitehead, A., and Russell, B. (1970). Principia Mathematica to * 56. Cambridge: Cam-
bridge University Press. Originally published 1910.
Williamson, T. (1994). Vagueness. London: Routledge.
Wilson, D. (1975). Presuppositions and Non-Truth-Conditional Semantics. New York: Aca-
demic Press.
Wilson, D., and Sperber, D. (1979). “Ordered Entailments: An Alternative to Presuppositional
Theories,” in C-K Oh and D. Dinneen (eds.), Syntax and Semantics 11: Presupposition
———. (1981). “On Grice’s Theory of Conversation,” in P. Werth (ed.), Conversation and
Discourse (pp. 155–78). London: Croom Helm.
Wittgenstein, L. (1967). Philosophical Investigations. Oxford: Blackwell.
———. (1969). On Certainty. Oxford: Blackwell.
———. (1974). Letters to Russell, Keynes, and Moore. Oxford: Blackwell.
Wood, J. (1998). “The Flying Trapezius,” New Republic, 219 (14 December 1998): 37–42.
Yoshimura, A. (1998). “Procedural Semantics and Metalinguistic Negation,” in R. Carston
and S. Uchida (eds.), Relevance Theory: Applications and Implications (pp. 105–22).
Amsterdam: John Benjamins.
Ziff, P. (1972a). “What Is Said,” in G. Harman and D. Davidson (eds.), Semantics of Natural
Language (pp. 709–21). Dordrecht: Reidel.
———. (1972b). Understanding Understanding. Ithaca: Cornell University Press.
———. (1972c). “What Is Said,” in P. Ziff, Understanding Understanding (pp. 20–38). Ithaca:
Cornell University Press.
272 BIBLIOGRAPHY
———. (1972d). “Something about Conceptual Schemes,” in P. Ziff, Understanding Under-

standing (pp. 127–41). Ithaca: Cornell University Press.
Zwarts, F. (1996). “Negation: A Notion in Focus,” in H. Wansing (ed.), Perspektiven der
analytischen Philosophie (Perspectives in Analytical Philosophy) (Vol. 7, pp. 169–194).
Berlin, New York: Walter de Gruyter.
———. (1998). “Three Types of Polarity,” in F. Hamm and E. Hinrichs (eds.), Plurality and
Quantification (pp. 177–238). Dordrecht: Kluwer.
Zwicky, A., and Sadock, J. (1975). “Ambiguity Tests and How to Fail Them,” in J. P. Kimball
(ed.), Syntax and Semantics 4 (pp. 1–35). New York: Academic Press.
INDEX
||-, 37, 251 Akmajian, A., 236n.

||-», 29, 251 Albritton, R., 75, 144
2, 188 allophone, 43, 140n., 141
‘2’, 187, 187n.,188 Allwood, J., 102n., 202
and two, 188 almost all N, 231–3
3, 188 almost F 149, 152, 153n., 154, 154n., 156–7,
‘3’, 187–8, 198, 201, 207–9 160, 182, 189, 231
relation between and three, 207, 209. See also does not entail not F, 150, 231–3
three N Sadock on, 154–60
almost true, 160
abduction, 13, 15, 29, 128 ambiguity, 3n., 12, 14, 24, 124, 129, 132, 137–
aboutness, 88n., 92n., 101, 109, 134–5, 142, 8, 161n., 169–70, 170n., 179–80
144, 148, 163, 163n., 237–9, 241, 247. the chicken that is ready to eat, 18
See also focal noun phrase limitation conjunction-reduction test for, 19–20, 22, 204
principle; Grice-Strawson condition; criteria of, 19, 21–2, 205
topic depth, 18
abstraction, 3n. Grice on, 63
acceptability judgments, 154–8, 166, 169, 214– of John’s book, 36
17 lexical, 33
accessibility of literal meaning, 222 and non-specificity, 17–22, 36, 38, 54, 118,
accommodation, 98n., 118, 126–9, 129n., 135, 123–4, 124n., 127, 129–31, 133, 137
142, 144–5 pragmatic, 215, 219
versus common ground, 143–4, 148 processing of, 20–22
Grice on, 142 of scope, 33, 38, 123, 127, 202
and presupposition, 147–8 tests, 19, 33, 201–6, 206–8
accomplishment predicates, 152, 158, 162n., types of, 31
176 of W. E. Hill’s maiden/hag drawing, 21
acquisition of language, 32 See also depth-nonspecificity of the Necker
achievement predicates, 152, 156n., 158 cube
adjacency pairs, 66–7 analytic, the, 150, 152, 153n., 161, 161n.,
adverbials, 206 162n., 176, 181–2, 186, 223
affirming the consequent, 56n. See also fallacy anaphor, 114, 197
273
274 INDEX
anaphora, 43, 113–14, 195–6, 197, 208. See background presumption, 90n., 91, 92n., 93,
also division of pragmatic labor 143
an-atomic property, 182 capacities (Searle), 107
and, 51, 85–6, 154–55, 155n., 157, 177, 179, and noncontroversiality, 90n., 91, 92n.
233. See also Boolean intersection; Baldwin, T., 226
but Ballmer, T., 118, 128
and/or, 54, 199–200 bank, 12
anomaly, 12–13, 69, 70n. Bar-Hillel, Y., 96
Anscombe, G. E. M., 137 Barwise, J., 231
Anscombe Point, 137 basic predicate, 32
Anscombre, J.-C., 215 Beaver, D., 120
any, 216 begging the question, 232–3. See also circular
a priori/a posteriori distinction, 30, 146, 181, reasoning
186 being F 153, 167–8
Argument Schema (Grice), 52–3, 62, 70–1, 78– and having F-ness, 153, 167
9, 175–6. See also calculability being tall, 153
Aristotle, 12, 15, 129n. belief, 50, 53, 62, 68–70, 73–5, 78, 85, 157,
as F as, 66, 151–2, 153n., 154n., 161, 165, 160n.
165n., 169, 173, 199–200 and conjunction-elimination, 228
and achievement predicate, 162n. and consistency, 228
almost, 161 knowledge of, 157
at least, 151, 165, 165n., 173 and reduction, 228
exactly, 151, 153n., 154n., 167, 169, 171–3 and S4 axiom, 228
non-symmetry of, 166 and saying, 157
not quite, 161 Benacerraf, P., 5, 230n.
reference point of, 166 Bergmann, M., 4n., 120
reflexivity of, 166 Berkeley, G., 32
transitivity of, 166 best interpretation, 54, 94–6, 105, 107, 112,
assertibility conditions, 151, 175–6 127, 137, 176, 211, 221. See also
cognitive biases in, 175–6 inference to the best interpretation
and truth-conditions, 157, 176 better than, 177–9
and use-conditions, 157 Bierwisch, M., 43n.
See also truth-conditions binding condition (Chomsky), 113, 196
assertion, 63, 79, 79n., 122 pragmatic reduction of, 115
and believing, 157, 175 bites shrewdly, 11, 13
and direct entailment, 123 Black, M., 14
and entailment, 123, 142 blame, 154
and implicature, 154 Boër, S., 132, 139, 202, 247
mental, 162n. Boolean intersection, 233
and Moore’s “implication,” 225–6 Boolean union, 233
and presupposition, 123, 142, 154 boy, 14–15
asymmetry, 132, 139–40, 164 bridging reference, 86
Atlas, J. D., 30, 34, 37–8, 39n., 41, 89–90, Brown, P., 13
93, 118, 122, 127–9, 131, 138–41, Brunot, F., 6
159n., 187, 197–8, 201–2, 209–10, Burge, T., 83n.
218, 220, 233, 238 burning bush, 24
Leuven Lectures (1990), 190n. Burnyeat, M., 79n.
at least “n,” 187, 191, 196–7, 210 Burton-Roberts, N., 4n, 218
at most “n,” 187, 201 but, 50, 52, 55–6, 56n., 57, 154–5, 155n., 156–
atomism (semantic), 181, 183–4 8, 197
Auden, W. H., 4, 6, 10 philosophical interest of, 55
Austin, J. L., 15, 37, 45, 189, 196
Avoid Pronoun. See division of pragmatic labor calculability, 52–3, 58, 60, 62, 79
Avoid Synonymy. See division of pragmatic and argument schema, 62
labor of conventional implicature, 60
of conversational implicature, 60
Bach, K., 33, 35–6, 36n., 38, 84–5, 87n., 120, of entailment, 60
196–7, 208, 221 of presupposition, 60
INDEX 275
can, 47, 53–4 completion. See completed interpretation

cancelability, 51, 52n., 55–6, 56n., 132–3, 137, compositionality, 218, 237
155, 155n., 156, 156n., 174n., 198–9, composition of relations, implicature and
211 entailment, 54, 152. See also quasi-
of conventional implicature, 60 entailment
of conversational implicature, 60 comprehension strategy, 95–6, 107–8, 134, 211,
of entailment, 60 217
and presupposition, 127, 133, 138, 215–18 computational theory of mind, 27
of presupposition, 60 concept, 223
Carnap, R., 96, 238 conceptual economy of ordinary language, 16
Carroll, L., 10 conditional perfection, 56n., 86, 90, 90n., 101,
Carston, R., 189–90, 196–8, 208–9, 212, 218, 172
221 conditionals, 89–90, 99–101, 126, 146–7, 191,
category error, 12 213
causality and concessive if not, 99–101
and convention, 24 Gazdar’s analysis of, 99–101
and dedicated process, 24 with quantifiers, 99
and justification, 22–3, 74–5, 78 Stalnaker on, 146–7
and perception, 22 and suspension of implicatures by if not,
and reason, 74–5 100
and sense-data, 48–50, 57–8 conjunction
causal theory of perception, 48–50, 57–8 asymmetric, 85
causative verb buttressing, 85, 102
lexical, 112 negated, 102–3
for Manner takes precedence over conjunction-reduction test for ambiguity. See
Informativeness, 112 ambiguity
paraphrastic, 112 consistent generalized quantifier noun phrase,
See also division of pragmatic labor 212–13
Chierchia, G., 120 constraints on interpretation. See interpretation
child, 14–15 constraints on production/ comprehension,
choice function e, 238 107
choice negation construction of interpretation. See
see negation interpretation; selection of sense
Chomsky, N., 12, 19, 34, 39, 42–3, 45–6, 85, content
112–13, 115, 140–1, 159n., 182n., 196, information, 96
204, 222, 235, 246–7 mental, 27, 71, 73
Churchland, P. M., 26n. pragmatic, 95, 141
circular reasoning. See petitio principii and presentation, 24–5
Clarke, D. S., 69 propositional, 30, 32, 37, 42, 73, 141, 196
clash (maxims), 62, 70–1 and image, 25
clausal implicature. See implicature and psychological state, 71
cleft sentence, 66, 163n., 193–5, 216–18, 234– semantic, 71, 73
5, 237–8, 242 content/context distinction, 146
exhaustiveness condition for, 243, 245 context dependence, 52, 67
only Indefinite NP in focus of, 243 context principle, 198, 208, 210, 213
uniqueness condition for, 243 conventional implicature
collective term, 189, 197, 201, 208 see implicature
g operator for, 238–9 conventionalization, 111, 183, 212
See also group conversational topic, 67, 67n. See also topic
common ground, 93–4, 118, 128, 135–6, 141–3 Cooper, R., 231
consistency with, 96, 135 Cooperative Principle (Grice), 59
Grice on, 136 coreference, 113–16, 238–9
and noncontroversiality, 93–4, 142–4, 148 local, 87, 113
communicative intention. See intention See also pronoun
comparative similarity. See as F as corporatism (semantic), 181–3. See also White,
completed interpretation, 84, 196, 208 M. G.
and expanded interpretation, 84, 196 Crane, T., 27
See also interpretation creatures of darkness, 223
276 INDEX
Cresswell, M., 167n., 169n. dog, 203–4, 206

criticize, 154 Donnellan, K., 215
crossed interpretation, 19, 206. See also double negation. See negation
ambiguity (tests); nonspecificity doubt-or-denial condition (Grice), 49–50
Cummins, R., 27 Dowty, D., 196
drink, 93–4
Davidson, D., 3n., 15, 45, 73n., 105, 144, 159, Ducrot, O., 215
206, 221, 223, 237n. Dummett, Sir M., 176n., 181
Davis, W., 108n.
Davison, A., 48 each other, 179
deBroglie, L., 223 Empson, W., 30
decision encapsulation, 17, 26, 29
that, 74 Englebretsen, G., 90n.
to, 74 entailment, 37, 56–8, 69, 81, 81n., 96, 109,
Dedekind, R., 188 122–3, 137, 142, 150, 155, 161, 161n.,
dedicated processor, 23, 25 237
default implication, 52, 72, 83, 227 analytic 150, 152, 153n., 161, 161n., 162n.,
default interpretation. See interpretation 176, 181, 231
defeasibility, 90n., 137. See also monotonicity context-relative, 136–7, 142
(non-) and contraposition, 163n.
definite description, 18, 38–9, 46, 159n., 201–2, direct, 122–3, 136, 152, 153n., 241
215–19 and grounds, 158
definite interpretation, 83 and implicature/presupposition, 137
and “precific” interpretation, 84 and meaning, 237
and Quantity One, 84, 88 properties of, 58–60
See also interpretation and reinforcement, 154–5
definite Noun Phrase, 245 See also logical consequence relation
degrees, 167n. equative, the. See as F as
degrees of truth, 160 equivocation, 187, 209. See also fallacy
depth-nonspecificity of the Necker cube. See eternal sentence, 33–4
nonspecificity even, 64
detachability, 51, 52n., 55–6, 56n., 132–3, 171, events, 105, 120–1
174n., 175–6, 197–200 exactly as F as. See as F as
of conventional implicature, 60 exactly “n” 187, 191, 195–6, 201, 209–10
of conversational implicature, 60 exclusion negation. See negation
of entailment, 60 exhaustiveness condition for clefts. See cleft
of presupposition, 60 sentence
and synonymy, 171, 174n., 175–6, 198–9 existence statement (singular)
determiner, 239–40 negative, 8
discourse representation theory, 120, 208 expanded interpretation, 35, 35n., 84–5, 86n.,
discourse semantics, 120, 208 196
disjunction (logical), 33, 37–8, 53–4, 97–8, and completed interpretation, 84–5
103, 162n. See also interpretation
exclusive, 54 expansion. See expanded interpretation
and Gazdar, 97–8 explanation, 76
inclusive, 54 psychological, 13, 53–4, 70–3, 78–9
See also Boolean union; or and implicature, 53–4, 70
display, 17, 19, 24–5 and utterance-interpretation, 70–1, 73, 76,
and portrayal, 25 78, 89, 141
and presentation of content, 24–5 explicature, 39n., 84, 134n., 196, 208, 213, 220.
distributive term, 189, 197–8, 208 See also relevance (Theory)
division of pragmatic labor (Horn), 110–13, exploitation, 62, 64n., 71
179 and violation, 61, 61n., 64, 64n., 71
and Chomsky, 112 See also maxims of conversation
and Kiparsky, 112 exportation, 92n.
and Levinson, 113 express, 27, 41, 53, 69, 157
and McCawley, 112 expressive alternatives, 109
INDEX 277
extension, 222, 237 Gregory, R., 18

convention of, 92–3, 142 Grice, H. P., 12–13, 15, 30, 33, 39, 39n., 40,
extents (of F-ness), 153, 161n., 167–8 42, 46, 49, 49n., 50–51, 55, 58, 58n.,
and measures (of F-ness), 161n. 97–8, 129–33, 136, 138, 142,192, 200,
plausibility of, 168 202–3
on presupposition, 132–3
factive verbs, 119, 126–8 on Wilson and Sperber, 66
neo-Gricean account of, 127 Grice-Strawson condition, 46, 145–6, 216, 218–
fallacy, 187n. 19
of affirming the consequent, 56n. Grinder, J., 204
of equivocation, 187, 209 grounds. See assertibility conditions
modal, 97 group, 194
figurative/literal distinction, 11, 15, 146. See size of, 195, 239
also metaphor See also collective term
Fillmore, C., 154 Guenthner, R., 10
first dogma of modernism, 4, 11 Guugu Yimithirr, 212
Fitch, F., 239
flouting a maxim, 12, 17, 61–2, 64, 64n., 76–8 Hacker, P. M. S., 47n.
philosophical importance of, 76–9 Halliday, M., 234
See also maxims of conversation; violating a Halvorsen, P.-K., 236n., 240–2
maxim Hamlet, 11
F-ness, 167–8 Harman, G. H., 13, 45
having and being F 167 Harnish, R., 86, 103, 107n., 179
focal negation. See negation Harper, W., 163n.
focal noun phrase limitation principle, 46, Harries-Delisle, H., 240n.
240n., 241 Harris, R. A., 48
focus, 217, 235–6, 242, 245 having tallness, 153
Fodor, J. A., 17, 22, 28, 28n., 181–2 Heim, I., 120
Fogelin,R., 4n., 23, 45 he is in the grip of a vice, 63
France is hexagonal, 189 Hempel, C. G., 96
Frege, G., 37, 120–2, 141, 208 high, 161n.
Frege Point, 226n. Hilbert, D., 238–9
Fretheim, T., 212 Hill, W. E., 20
Frost, R., 9–10 Hintikka, K. J. J., 238
Hirschberg, J., 101n., 109n., 209
gamma [g] operator. See collective term hit, 204
garden-path sentence, 23, 217 Hitzeman, J., 150, 156n., 231–3
Gass, W., 3n., 4, 11 Hobbs, J., 13
Gazdar, G., 53, 97–9, 101, 102n., 103, 234–6, Hochberg, H., 17
247 Hodge, A., 149
Gazdar’s bucket, 211 holism (meaning), 181– 4
Geach, P., 226n. homonymy, 3n.
Geis, M., 86, 172 Hopper, P., 183
generality (semantic, aka nonspecificity), 32, Horgan, P., 5
113–15, 179. See also nonspecificity Horn, L., 16, 43n., 65, 100, 106, 110, 120,
generalized conversational inferenda, 30, 38 124n., 139, 155, 155n., 179, 183, 185,
generative semantics, 48, 90, 139 192, 195, 198, 202, 207, 209–10, 212–
genitive Noun Phrase, 19, 36, 85 13, 215–18, 221, 241–2, 244–6
Glucksberg, S., 4n. on only, 195
Goodman, N., 12, 183 on Q and R principles, 111
Gordon, G., 120 Horn scale, 81, 88n., 90, 101, 110, 114, 152,
gradable predicates, 151 176, 182, 186
and accomplishment, 152 negative, 102n.
and achievement, 152 restrictions on, 101, 109
grammaticalization, 183 Hornstein, N., 32
Grandy, R., 32, 88, 144 how many, 192–3, 195
Graves, R., 149 Huang, Y., 115n., 185, 196
278 INDEX
hybrid implicature, 54. See also composition of “imply” (Moore, G. E.), 69n., 70n., 225–6. See
relations also assertion; Moore’s paradox
hyperbole, 43, 64–5 incoherence of classical Gricean semantics and
pragmatics, 80–7, 162–4, 171, 174n.,
I love you too, 35 175–6, 198–201, 210–15
imagination, 4–5, 11 indefinite Noun Phrases, 35n., 106–8, 243, 245
implicature, 13, 30, 40, 47, 47n., 58, 58n., 83, for Quantity takes precedence over
83n. Informativeness, 107, 110
argument schema for, 70, 78–9 indeterminacy
and Bach, K., 87n. of radical interpretation, 221, 223
and belief, 62 of radical translation, 221, 223
class B and class C, 64n., 70–1, 76–8 indexical resolution (Levinson), 213
clausal, 53–5, 84, 88–90, 97–9, 103 inference
fallacious arguments for 97–9 to the best explanation, 13, 29, 128, 193
inconsistency of with scalar implicata, 97 to the best interpretation, 13, 29, 54, 94, 97,
continuity, 86 107, 112, 127, 137, 176, 211
conventional, 52, 57–8, 156, 197 conversational, 30
properties of, 59–60 to stereotype, 86, 93
conversational, 47, 57, 58n., 83, 83n. See also best interpretation
generalized, 47, 47n., 57, 58n.,114n. inferendum, 33, 39
particularized, 47, 47n., 57, 58n. infinite regress, 187, 210
properties of, 59–60 information
default, 52 content, 96
and entailment, 69n., 137 and existential quantification, 106
epistemic form of, 54 Grice on, 61, 67–8, 93, 97–8
and explanation, 54, 70 and negative statements, 163–4
in explanation versus interpretation, 73, 79 old/new, 234–5
inconsistent, 83 Popper on, 106
and indefinite Noun Phrases, 106 informativeness, 43, 59, 63, 65, 67, 80, 80n.,
and maxim clash, 70 83, 91, 94, 97–8, 105, 127–8, 135, 164,
narrow sense of, 83n. 172, 172n., 175, 177, 179–81, 199, 207,
and negative statements, 38, 82–4, 101–3, 213, 241
109, 119–24, 127, 129–42, 163–4 and efficient communication, 107
nontriviality restriction on, 68 and entailment, 81n.
philosophical importance of, 57–8 and Levinson’s heuristic, 94–5, 211–12
problem of the degenerate case for, 140 and pragmatic intrusion, 180–1
properties of, 59–60 principle of, 95, 110, 113
Quantity One not well defined and reduction by relevance, 63–8, 97–8
reinforcement of 154–5, 155n., 156, 174 and stereotypicality, 94
scalar, 54, 80–1, 83, 87–88, 103, 152, 153n., inscrutability
155n., 156, 163n., 164, 171, 175–8, 188, of literal meaning, 221–3
192–3, 208, 213–14 of reference, 221
with collective Noun Phrase, 197 intension, 222
with distributive Noun Phrase, 197 convention of, 91, 93, 143
with lexically incorporated numerical intention
adjective, 198 communicative, 24, 28–30, 36, 41, 71–2
inconsistency of with clausal implicata, Grice’s M-intention, 72, 108
97 interface (semantics/pragmatics), 30, 37, 39–41,
negative, 90 246
and speaker’s meaning, 73 internalist semantics, 34, 37n., 39–40, 42–3,
suspension of, 100 141, 246–7
and transitivity, 59–60, 163n. interpretation (utterance), 13, 33–4, 38, 40–1,
See also belief; calculability; cancelability; 63, 65, 70–3, 78–9, 94–5, 107–8, 127–8,
detachability; incoherence of classical 136–7, 144, 161n., 207, 211, 220–1
Gricean semantics and pragmatics; and Class C implicata, 71
informativeness; maxims of conversation cognitive biases in, 175–6
impliciture (Bach, K.), 39n., 84, 87n., 134n., and common ground, 136
197, 221 competing, 95
INDEX 279
completed, 84–5, 196, 208 Kooij, J., 30

constraining of, 25, 29, 94–5, 107, 127 Kraus, K., 6
construction of, 32–9, 127–8, 131, 133–5, Kripke, S., 159n.
180–1, 219–21 Kuroda, S-Y., 131, 138, 202
and context, 211
convention or practice of, 144 Ladd, D., 215
default, 54, 93, 114n., 134, 148, 221 Lakoff, G., 3n., 14, 19, 32, 48, 83n., 95, 120,
definite, 83–4, 88 204, 228–9
expanded, 35, 35n., 84–5, 86n., 196 Lakoff, R., 48
and explanation, 70–3, 76, 78, 78n., 79, 89, lamda [l] abstraction, 234n., 238
141 Langendoen, T., 14n.
hermeneutic (nonmodular) and reliable Langford, C. H., 227–9
(modular), 26–30 language of thought, 26–9. See also mentalese;
and meaning, 161n. thought
“precific”, 84, 110n., 220 lattice, 37
precise, 33, 83, 85 least upper bound (sup), 37, 153
saturated, 84 Leech, G., 48, 85, 180
specific, 33, 83–5, 94, 110, 110n., 220 Leezenberg, M., 4n.
strengthened, 52–3, 84–5, 192, 210–11 Lehrer, A., 32
See also best interpretation; explanation; Leibniz and Newton invented the calculus, 104
representation (semantic); selection of Leisenring, A., 239
sense LePore, E., 181–2
intonation, 215–17 Levinson, S. C., 13, 27, 53, 65, 93–4, 108–10,
into the wood, 180 113–16, 164, 177–9, 185, 191n., 210,
introspective access, 222 212–15
intuitionism, 54 on Moore’s paradox, 69, 129, 133
I-principle (Levinson), 116 Levinson scale, 57n., 109, 182
Iten, C., 33 Lewis, D. K., 98n., 118, 128, 129n., 143
it’s not the case that p/ it’s not true that p, 38, lexical decomposition, 32, 203
86n., 124, 124n., 131, 133, 137, 139 lexical incorporation, 198, 201, 208–9
lexicalization, 101, 109, 183, 203
Jabberwocky, 10–11 Liberman, M., 215
John plays well, 35 Linder, J., 54
John’s book, 36, 85 linguistic turn (third), 39
Johnson, Dr. S., 30 lists, 240
Johnson-Laird, P. N., 32 literal sentence-meaning, 24–5, 32, 37, 40, 42–
Johnson, M., 14n. 3, 207
and utterance-interpretation, 30
Kamp, H., 120, 208 See also interpretation; meaning
Kaplan, D., 5 literariness, 9, 11
Karttunen, L., 64, 82, 118, 124n., 126, 157, local coreference, 87. See also coreference
160, 216, 242, 247 Locke, J., 7
Katz, J., 32, 34, 159, 181, 221, 223 logical consequence relation, 37, 58–60. See
Kay, P., 85 also entailment
Keenan, E., 159 logical form, 237
Kemeny, J., 96 logically perfect language, 122
Kempson, R., 32–3, 37, 102n., 120, 138, 140, looks red (Grice), 49–57
193n., 201–2, 212 loose talk, 14, 188–90, 201
Kennedy, C., 169n. luckily, 186
Kenny, Sir A., 13, 46 Lycan, W., 132, 139, 162n., 202, 247
Kiparsky, P. & C., 127, 155n. Lyons, J., 32, 239, 249
Kittay, E., 4n.
Klein, E., 167n., 168n., 169n. maiden/hag figure (Hill), 21
knocked, 204 manner (maxim of), 54, 60–1, 61n., 200
know, 126–8 Grice’s new, 39n., 137–8
knowledge of language, 181–4 See also maxims of conversation
Koenig, J-P., 209 marked/unmarked [expression], 38, 111–13,
König, E., 99 164, 171, 175, 220, 241n.
280 INDEX
Mart and David bought a piano, 105–6 mental image, 32–3

Marti, G., 47 metalanguage, 160, 183, 216, 222
Martin, J. N., 120 metalinguistic negation. See negation
Martin, R. M., 239 metaphor, 3–17, 165
Martinich, A. P., 63–7 Grice on, 12
Matsumoto, Y., 101 Mey, J., 106n.
Mauriac, F., 64 Miller, G., 32
maxims of conversation, 12, 59, 61, 61n., 83, mirror maxim (Harnish), 86, 104
83n., 106–16, 137–8 modularity, 17, 22, 26
clash, 62, 70–1 modal fallacy, 97
explanatory inadequacy of Grice’s, 84, 96, modernism, 3–17. See also first dogma of;
151 second dogma of
explanation by, 76–9 modified Occam’s razor, 131, 192
exploitation of, 62, 64n., 71 molecularism (semantic), 181. See also
flout of, 12, 17, 61–2, 64, 64n., 76–8 corporatism
Manner, 39n., 61, 61n. monotonicity (non-monotonicity), 58–9, 90n.,
Manner (new), 137–8 137
Manner takes precedence over of conventional implicature, 58–9
Informativeness, 112 of conversational implicature, 58–9
Quality, 59, 72, 72n., 81 of entailment (logical consequence), 58–9
Quantity, 59 of presupposition, 58–9
Quantity and Quality, 71–2, 72n. See also defeasibility
Quantity takes precedence over Montague, R., 159, 169n., 222
Informativeness, 107, 110 Moore, G. E., 48, 50, 68
reduction of Quantity to Relation, 63–8 Moore’s paradox, 50, 53, 68–69, 69n.,70,
Relation (relevance), 61, 61n. 70n.,157, 225–30
Relativity, 91, 94, 134 and implicature, 69
violation of, 61, 61n., 64, 71 Levinson on, 69
See also causative verb; indefinite Noun Mortimer, Sir J., 13n.
Phrases; manner; meaning; mirror Moses, 24
maxim; Quality; Quantity mostly, 152
may, 54 Muka6ovský, J., 6–8, 11
McCawley, J., 14, 48, 170n., 216, 218, 241 multigrade predicate, 240
McConnell-Ginet, S., 120 multi-stable figure, 18
meaning my aunt’s cousin, 142–3
conventional, 83 my cousin isn’t a boy any longer, 14
and entailment (logical consequence)
relations, 237 natural kind term, 3n., 39, 47, 95, 159n., 182
Grice’s account of, 73–5 natural number, 188
Grice’s confusion of and explanation, 76–9 Neale, S., 58n.
literal, 24–5, 30, 32, 37, 40, 42–3, 73, 83, 94, necessary/ sufficient conditions, 56, 60, 69,
131, 137–41, 149–53, 165–76, 185–7, 69n., 78, 79, 143, 199–201, 208, 229,
207–10, 219–23 231–3
sentence-type, 30, 37 Necker, L. S., 17
speaker’s, 42, 73 Necker cube, 17–18
presumptive, 148 double, 20
and truth-conditions, 181, 237 See also nonspecificity
and use, 157–8, 160, 172, 176 negation, 16, 16n., 32–3, 33n., 37–8, 80, 82
utterance-token, 37 ambiguity tests for, 201–6
See also explanation; interpretation choice, 16, 16n., 33, 33n., 38, 82, 82n.,123–
meaning postulate (Carnap), 161n. 4, 133, 198, 201, 204, 206–7, 219–20,
measures of F-ness, 153, 161, 167 245
and extents, 153, 167 and clefts, 245
measure-wise F-er than, 161n., 168 descriptive, 193n.
meiosis (understatement), 65 double, 90
membership categorization, 86 exclusion, 16, 16n., 33, 33n., 38, 82, 82n.,
mentalese, 27, 90, 95–6. See also language of 102n., 123–4, 133, 201, 204, 206–7,
thought; thought 219–20, 245
INDEX 281
focal, 60n., 100, 215–16, 218 not F not entailed by almost F 150
lexicalized versus paraphrastic, 90 not quite F 149, 152, 153n., 154n., 156–8,
metalinguistic, 100, 112n., 193n., 213–19, 245 231–3
and presupposition, 56n., 60n., 122–4 Sadock on not quite F 154
presupposition-canceling, 127, 133, 138, See also nonspecificity
215–18 Nowell-Smith, P. H., 45
See also nonspecificity; not numeral, 187, 209
negative incorporation, 87 numerical adjective, 44, 80, 187, 190, 209
negative lowering, 87 equivocation on, 209
negative polarity item, 216 numerals versus numerical adjectives, 209
negative scale, 82, 90, 90n., 109
negative specification, 86 obligation, 145n.
neighbor, 34 observable property, 221–2
Newmeyer, F., 190 observation, artistic versus scientific, 16n.
Newton, Sir I., 222 only Indefinite noun phrase (focus noun phrase
Nietzsche, F., 26 in clefts), 243
noncontroversiality, 90n., 91–2, 92n., 93, 134– only Proper Name, 66, 194–5, 232, 242
5, 142, 144, 148, 176 open question, 76–7
axioms of 135 and meaning versus explaining, 76
and cognitive biases, 176 or, 51–6, 56n., 57, 69, 97–8, 196–7, 199–200,
and common ground, 90n., 91, 93–4, 136, 233
142–4, 148 clausal implicata of, 97–8
conventions of, 91–2 and disjunction, 38, 54
and stereotypicality, 93 exclusive, 54
See also take one’s word for Gazdar on clausal implicatures of, 98
non-monotonic quantifier, 232 implicatures of, 52–4, 97–8
nonspecificity of sense (aka semantic inclusive, 54
generality), 15–16, 17–22, 24–6, 29–39, nonspecificity of, 54
41–2, 54, 93,113, 137, 146, 161n., 179– philosophical importance of 56, 56n.
80, 187, 196, 198, 202, 205, 207, 212, quasi-entails not both, 54
214, 219, 241n., 246 scalar implicata of, 97–8
and ambiguity, 15–16, 17–22, 32, 38, 85, See also disjunction; nonspecificity
118, 124n., 127, 130–1, 133–4, 137 ordered implication, 240
Bach, K. on 35–6 ordering principles for application of maxims,
and constituent structure, 85 107–8, 108n., 109–16. See also
criteria of, 22 projection rules for application of
and disjunction, 38 maxims
evidence from absence of Horn Scale orientation (spatial), 164
negation implicata, 102n. Ortony, A., 4n.
of independent/cooperative, 104 ought, 47
of John’s book, 36
of the Necker cube, 17–22 parole, 28–9
of not, 33–4, 37–9, 82, 127–8, 131, 138, Peano, G., 188
201–6 Peirce, C. S., 13, 29
of or, 54 perception
and the problem of Gricean semantics, 198 causal theory of, 48, 55–6
processing of, 20–22, 26 and inference, 22–3
of pronouns, 113–14 and language, 48
and semantic features, 85 and modularity, 22
of three N 198, 207–12, 214 and representationalists, 23
and underdetermination, 33 and underdetermination argument, 22–3
non-triviality restriction (Grice), 68, 79 performance system, 39, 141, 246–7
not, 16, 32–3, 37–8, 80, 82, 86, 86n., 99, 122–4, perspicuous, 61, 61n.
127, 129, 139, 151–2, 193, 193n., 198, Peters, S., 64, 82, 157, 160, 216, 242, 247
204, 206–7, 219, 221 petitio principii. See begging the question
ambiguity tests for, 201–6 Phillips, M. A., 3n.
not F 150–2, 153n., 154, 154n., 158, 182, phoneme, 43, 140n., 141
231 politeness, 61
282 INDEX
polysemy, 161n., 177, 179, 204 proposition. See content (propositional)

Pomona College, 3n. propositional attitude, 36n.
Popper, K., 92n., 96, 106 propositional form of an utterance, 196
P or P, 54 prototype, 95
possible world, 158–160, 160n., 189 pseudo-cleft sentence, 234–5, 237
Postal, P., 204 Pulman, S. G., 32
practical reasoning, 13, 15, 128 Putnam, H., 3n., 15, 23, 30, 78n, 92n., 95, 146,
pragmatic intrusion, 38, 39n., 133–4, 146, 177– 159n., 186
81, 207–8, 221
Atlas on 181 Q Principle
“precific” interpretation, 84, 110n., 220 and Horn, 43, 111
and definite interpretation, 84 and Levinson, 116
See also interpretation Quality (maxim), 59, 72, 72n., 81
precise interpretation, 83–5, 110n., 219–20. See and Quantity, 71–2, 72n.
also interpretation See also maxims of conversation
presentation, 24 Quantity (maxims), 59, 114, 127, 157, 163–4,
presumption, 90n., 91. See also background 179
presumptions; noncontroversiality and Quality, 71–2, 72n.
presupposition, 46, 50–2, 52n., 55–6, 56n., 58– Quantity One, 81, 87, 96, 104, 107, 114, 164,
60, 68, 118, 128, 137–8, 156–7, 202, 172, 174n.,179, 188, 192–3, 199, 201,
206, 216–17, 242, 247 207–8, 213–14
assertoric (Frege), 125n. Quantity One inadequacy, 96
cancelation of, 56n., 127, 133, 138, 215–218 Quantity One inconsistent implicata, 164
data of, 119 Quantity One not well defined, 104
and entailment, 137 Quantity Two, 65, 93, 97, 208, 212
and focus, 235 Quantity Two reduction by Relation
“plug” for, 124n. (relevance), 63–8
pragmatic (Frege), 125 See also maxims of conversation
pragmatic (Stalnaker), 125, 128, 137, 143 quasi-entailment, 54, 152, 161–2, 241. See also
projection problem for, 218 composition of relations
reduction (Grice) of to implicature, 137–8 quasi-quotation, 122
referential, 120, 127 Quine, W. V. O., 30, 33–4, 46, 78, 92n., 96,
semantical, 121–2, 125 122, 184, 186, 194, 206, 221–3
See also focal noun phrase limitation
principle; Grice-Strawson condition radical pragmatics, 151, 190. See also
pretense, 144–5 Sadock, J.
Prince, E., 236 rather than, 213–14
Princeton University, 3n. reality (semantic), 222
principle of informativeness, 95. See also reason
informativeness for believing, 74–5
privative opposition, 203 and cause, 74–5
processing model, 134, 211 for doing, 74–5
production strategy, 95–6, 107–8, 134, 217 and motive, 74–5
projectible predicate, 183 Récanati, F., 34, 36, 84–5, 180, 196–7, 208, 221
projection problem for presupposition, 218. See rectification, 215
also presupposition reductio ad absurdum argument, 65, 150, 164,
projection rules for application of maxims 166, 171, 175–6, 187–9, 228, 228n.,
for anaphora, 114 229, 231
for Manner versus Informativeness, 112 redundancy, 155–6, 165, 244
for Quantity, Manner, and Informativeness, reflexive pronoun. See pronoun
113 reflexivity, 58–9
for Quantity versus Informativeness, 107 of conventional implicature, 58–9
See also ordering principles of conversational implicature, 58–9
pronouns, 113–15, 236n. of entailment, 58–9
Horn scale <reflexive, pronoun> 114 of presupposition, 58–9
See also coreference; nonspecificity regret/know problem, 109
proper name, 122–3, 159, 243 reinforcement of implicata
and presupposition, 122–3 see implicature
INDEX 283
Reinhart, T., 113, 196 semantic slack, 33

Relation (maxim), 61, 61n. semi-redundancy, 151, 153n., 173, 174n.
Quantity Two reduced by, 63–8 sense-datum, 48–50, 57–8
See also informativeness; maxims of sentence semantics (internalist), 40, 220
conversation sequential expectations, 66–7
Relativity (maxims), 43, 67, 91, 93, 105, 172n., Seuren, P., 120, 167n., 168, 168n., 193–7, 208,
207 218
relevance, 61, 61n. signification (total), 41, 53, 70, 83, 197
Grice on, 66, 97–8 Sinnott-Armstrong, W., 47
Martinich on, 67 Sluga, H., 131
reduction of Quantity Two by, 63–8, 97–8 Smokler, H., 96
Strawson on 67n. Soames, S., 120, 156n., 163n., 165n.
Theory, 114n., 134n., 136, 208, 213, 220 Sommers, F., 90, 99
representation, 22–3, 26–8, 141 sortal classifiers, 239–40
and interpretation, 40, 219–22 specific interpretation, 83, 85, 94, 110n., 219–
semantic, 27, 29, 34, 40, 133–4, 140, 219–22 20. See also interpretation
Reyle, U., 120, 208 Sperber, D., 4n., 14–15, 33, 33n., 34–5, 65–6,
Rorty, R., 23, 45, 78, 184 69n., 84, 120, 185, 208, 213, 220, 240
Rosch, E., 95 square brackets (Grice), 46
Rosenthal, D., 229n. Staegemann, E., 173n.
Ross, J. R., 19, 48, 164, 204 Stalnaker, R., 14n., 118, 120, 125–6, 128, 137,
R Principle (Horn), 111, 207–8. See also 141, 143–7, 226n.
informativeness statement semantics (externalist), 40
Ruhl, C., 161n., 222 stereotype, 14–15, 32, 93–6, 135, 175–6
Russell, B. A.W., 12, 48, 71, 90, 189n., 201–3, as a constraint on interpretation, 94–5, 106,
223 176
Ryle, G., 5, 12 conversational inference to, 29, 86, 93
and informativeness, 94
S4 characteristic axiom, 228–9 and noncontroversiality, 93, 176
Sacks, S., 4n. See also assertibility conditions;
Sadock, J., 14, 19–20, 33, 33n., 48, 60n., 145– informativeness; interpretation
7, 151, 154–5, 157–8, 160n., 162n., 163, Stern, J., 4n.
163n., 172,173n., 185–7, 199, 202, 204– Stevenson, C. L., 225
5, 209, 212, 221 Stich, S., 36n.
on almost, 154 Strawson, P. F., 38, 51, 52, 52n., 56n., 67n.,
on not quite, 154 118–19, 124, 127–8, 130–2, 138–9, 202,
Sag, I., 215, 241 220, 241
salient alternatives, 88 on the negative marked case, 38
Sanford, D., 54 on accommodation, 128
saturated interpretation, 84 strength (rule of), 45
and strengthened, 84 strengthened interpretation, 52–3, 84, 192, 210–
See also interpretation 11
saying, 38–9 and saturated, 84
scalar implicature. See implicature See also interpretation
scale reversal, 188, 209 stress, 60n., 79n., 100, 127, 193–5, 213, 215,
Scharten, R., 212 217, 235
Schelling, T., 129n. succeed/try, 109
Schrödinger, E., 221 sufficient conditions. See necessary/sufficient
Scott, D. S., 222, 250–1 conditions
Searle, J., 107, 145n. suspension of implicata. See implicature
second dogma of modernism, 12–13 synonymy, 170–1, 171n., 172, 174n., 197, 199–
selection description, 238 200. See also detachability
selection of sense, 31–2, 38, 220. See also
construction of interpretation take, 85
Sellars, W., 249 take for granted, 92n., 93, 144, 147
semantical determinant, 121 Stalnaker on 125
semantic differentia, 203 See also background presumption;
semantic representation. See representation presumption
284 INDEX
take one’s word for, 127, 142, 144, 148. See truth-indications, 36n., 42, 140, 222
also noncontroversiality Twain, M., 190
tall, 161n., 164 two, 186, 188–9. See also ‘2’; 2
being tall versus having tallness, 167
and short, 164 Ullian, J., 96
taller, 161, 161n., 182 Ullman, S., 25
measure-wise taller, 161n. underdeterminate, 33–5, 180–1, 246
uniformly taller, 161 versus underdetermined, 16, 33–5, 38
Tarski, A., 159 understatement, 65
tautology, 64–5, 76 uniformly F-er than, 161, 161n., 168. See also
Taylor, M., 36 taller
the, 189, 189n. uniqueness condition for clefts. See cleft
the chicken that is ready to eat, 18. See also sentence; exhaustiveness condition for
ambiguity; definite description clefts
third-use argument, 14–17 upward-monotonic generalized quantifier noun
Thomason, R., 120–1, 128 phrase, 231, 233
thought, 27–8 use-conditions, 157–8, 160, 172, 176
content of, 27 utterance-interpretation, 63, 107–8, 207, 211.
expressing a versus being a 27 See also interpretation
medium of versus language of 27
See also content; language of thought; vagueness, 34, 158–9
mentalese Valéry, P., 8–9, 11
three N 80, 187, 187n., 198, 201, 207–10, 212– van der Auwera, J., 90n.
13, 221 van der Sandt, R., 120, 163n.
ambiguity tests for, 206–8 van Fraassen, B., 120, 123, 251
equivocation on three and ‘3’ 209 van Kuppevelt, J., 212
See also nonspecificity; not; 3; ‘3’ Vendler, Z., 45, 47, 152, 158
three or more N, 213 violation of a maxim. See maxims of
together, 103–6 conversation
too, 35
topic, 46, 66–7, 92, 92n., 94, 134–5, 144, 145n., Walker, R., 34, 36, 221
148, 163, 192–4, 216–17, 240–1, 247 Warnock, G., 46, 241
conversational, 67, 67n., 163n. what is said, 42, 62–3, 63n., 81n., 83, 83n., 134,
See also aboutness; focal noun phrase 188, 196–7
limitation principle; Grice-Strawson White, M. G., 30, 35n., 166n., 181–2, 186, 196
condition Whitehead, A. N., 124, 202
topicalization, 193 Wilson, D., 4n., 14–15, 33–5, 65, 84, 120, 185,
total signification. See signification 208, 213, 215, 220, 240
transitivity, 58–9, 162, 163n. with, 31
of conventional implicature, 58–9 Wittgenstein, L., 46–7, 47n., 226n., 227
of conversational implicature, 58–9, 162, Wolfe, T., 4, 16n.
162n. Wood, J., 4, 16n.
of entailment, 58–9 word, 249
of presupposition, 58–9 word-form, 249
translation, 9–10, 159–60, 183
constraints on, 159 you are the cream in my coffee, 12–13
transparency of meaning, 222
Traugott, E., 183 Ziff, P., 16, 33, 83n., 238
truth-conditions, 36, 40n., 83, 83n., 157–60, Zimmer, P., 85
173, 176, 194, 208, 222 Zwarts, F., 212–13, 233
and meaning, 40, 181, 221–2 Zwicky, A., 19, 33, 33n., 86, 124n., 172, 202,
See also assertibility conditions 204–5

Jay David Atlas - Logic, Meaning, and Conversation Semantical Underdeterminacy, Implicature, and Their Interface

Загружено:

Сведения о документе

Оригинальное название

Авторское право

Доступные форматы

Поделиться этим документом

Поделиться или встроить документ

Параметры публикации

Этот документ был вам полезен?

Это неприемлемый материал?

Авторское право:

Доступные форматы

Jay David Atlas - Logic, Meaning, and Conversation Semantical Underdeterminacy, Implicature, and Their Interface

Загружено:

Авторское право:

Доступные форматы

Logic, Meaning,

OXFORD UNIVERSITY PRESS

JAY DAVID ATLAS

Copyright © 2005 by Jay David Atlas

B831.5 .A75 2000

T he flight to Frankfurt on the Lufthansa 747-400 passed slowly, as August flights

2 Grice’s Theory of Conversational Inference: A Critical Exposition 45

3 The Rise of Neo-Gricean Pragmatics 80

4 The Post-Gricean Theory of Presupposition 118

5 Assertibility Conditions, Implicature, and

Appendix 1 On G. E. Moore’s Term ‘Imply’ 225

Appendix 2 On Hitzeman (1992) on ‘Almost’ 231

Appendix 3 The Semantics and Pragmatics of Cleft Sentences 234

Appendix 4 A Note on Notation 248

We cannot assume that statements (let alone sentences)

The linguistic turn was, I think, an uncompleted revolu-

1 Metaphor, nonspecific meaning, and utterance

Twentieth-century studies of literary style and early, influential studies in philoso-

some fundamental cleavage between language that is figurative or literary, contain-

1.1 The first dogma

An imaginary enemy is a nonexistent enemy, which, as Gilbert Ryle (1949) would

follows. The aesthetic function of a term is incompatible with communicative func-

It is blue-butterfly day here in spring,

Even in as uncomplicated a structure as this, it is evident that paraphrase into prose

Frost’s definition of poetry as the untranslatable element in language looks plau-

’Twas brillig and the slithy toves

Il brilgue: les tôves lubricilleux

Es brillig war. Die schlichte Toven

The first systematic discussion of metaphor in Western philosophy occurs in

1.2 The second dogma

The second dogma of literary modernism is the reduction of metaphoricity to

1.3 The third use argument

word.” See Steiner (1975a).

2 Semantical nonspecificity, utterance

2.1 Visual/verbal nonspecificity versus

FIGURE 1.1 Necker cube

ously reverses in depth. Alternately, the figure is seen as a left-directed, downward-

The ambiguity of a description differs from the spontaneous multistability of a

1. The impossibility of a crossed, literal paraphrase for a conjunction-

If my analysis of the depth-nonspecification, rather than depth-ambiguity, of the

FIGURE 1.2 Double Necker cube

FIGURE 1.3 Old woman—young woman

The parallel I am drawing is this: in a non-display-ambiguous but display-gen-

Ambiguity: Visual (e.g., Maiden/Hag Drawing) and Verbal

d. Prompting is sometimes required to recognize alternate displays of a

Nonspecificity: Visual (e.g., Necker Cube Drawing) and Verbal

2.2 Are perceptual mechanisms inferential and modular?

2.3 Is speech perception modular?

background beliefs. By virtue of those differences in background beliefs, hearers may

2.4 What do semantical and visual nonspecificity mean?

between the two presentations (contents). An ambiguous figure would conventionally

2.5 A language of thought and the nonmodularity of

[what] is usually required in the intentional analysis of behavior is a kind of herme-

3 Semantical underdeterminacy and pragmatic

In Seven Types of Ambiguity William Empson (1930) discusses a Samuel Johnson

What murdered Wentworth, and what exiled Hyde,

The mechanism of implicature generation suggested by Grice is intended to account

Bach, cited in Atlas (1989: 30), discusses another phenomenon—not semantical

Récanati (1993: 236) appeals to an example of reference specification and ei-

In the 1970s as I lectured to various audiences on what in 1998 was flatteringly

what in linguistics is called “the semantics-pragmatics interface.” In Atlas (1977b,

Independently Kempson (1975) had suggested an (unfortunately) extensional, dis-