Вы находитесь на странице: 1из 13

Linguistic Society of America

The Non-Cyclic Nature of Japanese Accentuation Author(s): Masayoshi Shibatani Source: Language, Vol. 48, No. 3 (Sep., 1972), pp. 584-595 Published by: Linguistic Society of America Stable URL: http://www.jstor.org/stable/412036 Accessed: 28/10/2010 18:58
Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in the JSTOR archive only for your personal, non-commercial use. Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at http://www.jstor.org/action/showPublisher?publisherCode=lsa. Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed page of such transmission. JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.

Linguistic Society of America is collaborating with JSTOR to digitize, preserve and extend access to Language.

http://www.jstor.org

THE NON-CYCLIC NATURE OF JAPANESE ACCENTUATION


MASAYOSHI SHIBATANI

University of California, Berkeley


While the cyclic application of a phonological rule hinges upon the erasure of inner labeled brackets or phrase boundaries, the late (or post-cyclic) pitch assignment rules of Japanese crucially depend on the information provided by the inner phrase boundaries. Thus we find a basic incompatibility between the principle of the transformational cycle and the nature of Japanese accentuation. This paper first points out a serious flaw in McCawley's cyclic analysis of the accentual system of Japanese, and then offers a non-cyclic re-analysis. Since English is being decycled by a number of people, what follows helps to further weaken the theory of the transformational cycle in phonology.

1. The principle of the transformational cycle was first formulated by Chomsky, Halle, & Lukoff 1956 in their study of English accentuation. Since then the principle has been applied to languages such as Russian (Halle 1963, Lightner 1965), German (Kiparsky 1966), French (Schane 1968), and Japanese (McCawley 1968). Having seen these applications of the principle to a variety of languages, Chomsky & Halle (1968:24) assume that the principle is 'one of the conditions, intrinsic to the language-acquisition system, that determines the form of the languages acquired', and is therefore a language universal. The applicability of the principle is further taken by McCawley 1968 as one of the typological criteria which groups languages like Chinese and Vietnamese on the one hand vs. languages like Russian, English, and Japanese on the other; in the latter group, 'the reduction rules affecting accent MUST APPLY A CYCLE IN which deduces the accentual shapes of any constituent from the accentual shapes of the constituents of that constituent' (183, emphasis supplied). The principle of the transformational cycle, however, has been challenged by a number of people, e.g. Ross 1969 and Lee 1969, who are undertaking non-cyclic re-analyses of the English accentual system. Japanese, as noted above, is one of the languages contributing to Chomsky & Halle's assumption of the universality of their principle. In this paper, I shall challenge that assumption by showing that Japanese can no longer be counted as a language requiring the cyclic application of phonological rules.' 2. Before I examine McCawley's cyclic analysis and point out an unfortunate
preparation of this paper was supported by a Grace W. Drake Scholarship of the University of California and by a National Science Foundation grant (GS-2386) made to the Phonology Laboratory of the Department of Linguistics, University of California at Berkeley. The content of this paper was presented in spring 1971to the Seminar on Suprasegmental Phonology at Berkeley, conducted by William S-Y. Wang, to whom I would like to express my deep gratitude for his stimulating discussion and comments. I must also thank John Ohala, Richard Stanley, and John Crothers, who read an earlier version of this paper and provided me with valuable comments on both theoretical and stylistic points. James McCawley, as an associate editor of LANGUAGE, suggested certain improvements in the final version. Needless to say, I alone am responsible for the content of the paper. 584
1 The

THE NON-CYCLIC NATURE OF JAPANESE ACCENTUATION

585

consequence of it, I must acknowledge the insightfulness of his investigation of the Japanese accentual system.2 My understanding of Japanese accentuation owes a great deal to McC's study, and the non-cyclic re-analysis proposed in ?4 is essentially based on his investigation and formal interpretations. Let us first briefly sketch the accentual system of Japanese along the lines of McC's work. Japanese accentuation involves two types of rule: the first has to do with placement, shifting, and elimination of accents, while the second assigns phonetic pitches to the output of the first type. The assignment of pitches depends on the location of accents, specifically on the first (leftmost) accent. The pitch assignment rule can be tentatively stated as follows: (1) Within a minor phrase, the initial mora is low pitched, unless it is accented, and all the moras following an accented mora are also low pitched. All the remaining moras are high pitched.3 Since this pitch assignment rule interprets the underlying forms and the output of the first type of rule, there is no need to assign pitches directly in the lexicon. Lexical items, if accented at all, thus need be specified only as to where the accent falls. Thus the underlying forms in the left column below are assigned their pitches as shown in the right column.4 'life' indti (2) i'noti koko'ro kokoro 'heart' atama' 'head' atdmd miyako miydko 'city' In isolation atamna and miyako, respectively final-accented and unaccented, display the same pitch contour, but atama has an accent on the last mora when an unaccented enclitic is added: 'head (nominative)' dtdmdgd (3) atama' ga miydko gd miyako ga 'city (nominative)' The accent on the last mora of atama has the effect of lowering the pitch of the following enclitic. The first type of rule mainly manipulates accents when formatives are combined, so that the pitch rules assign correct pitches to a stretch of formatives. E.g., when accented nouns in 2 are combined with accented enclitics, an accent elimination rule is called for, which McC (139-40) gives as follows: (4) When a phrase contains two or more accents, all but the first accent is eliminated. The rules so far given produce the correct phonetic pitch shapes for a phrase
2 This does not meanthat McC'sbookis easy to follow. In fact, it is partly due to the of of organization the bookthat someof the unfortunate consequences his treatmenthave never been broughtup before, even escapingone reviewer'sscrutiny; Okuda1970has missedthe problems discussedbelow.-Referencesbelowto McCareto his 1968 completely workunless otherwisespecified. Thereare fourtypes of morain Japanese: CV, V, the so-calledsyllabicnasal, and the consonant followedby a homorganic 'mint'. immediately e.g. consonant, the firstk in hakka 4The mark ' is used as an accentmarker,and we speak of the i' of i'noti 'life' as an accentedmora.The acuteindicateshighpitch; graveindicateslow pitch. Thus, in6ti is to be interpreted [HiLoLo]. as

586

VOLUME NUMBER3 (1972) LANGUAGE, 48,

containing two or more accented formatives, as in the following (when the first element is unaccented, the accent of the enclitic becomes the first accent of the phrase, as in 5d): itntimdd&'even a life' (5) a. i'noti ma'de -(4)-* i'noti made -(1)b. koko'ro ma'de -(4)-* koko'ro made -(1)-k6k6r3mdi 'even a heart' c. atama' ma'de -(4)-- atama' made -(1)-- dtdmdmWd 'even a head' d. miyako ma'de -(4)-* (not applicable) -(1)miydkomdd& 'even y a city'. There is a type of enclitic that does not behave like the above. When enclitics of this second type are combined with accented nouns, the accent of the enclitics rather than that of the nouns is manifested. Such enclitics are lexically marked with a morpheme feature as [+predominating]; and another rule, which precedes 4, is posited by McC (140) to delete the accent of the preceding noun: (6) All accents preceding a [+predominating] morpheme are removed. Gu'rai 'as much as', being marked as [+predominating], has this effect: (7) a. i'noti gu'rai -(6)-* inoti gu'rai -(1)-- in6tigzirdi 'as much as a life' b. koko'ro gu'rai -(6)-. kokoro gu'rai -(1)-* kok6r6guirdi'as much as a heart' c. atama' gu'rai -(6)-* atama gu'rai -(1)-* dtdmdg^rdi'as much as a head' d. miyako gu'rai -(6)-(not applicable) -(1)-* miydk6ogurdi'as much as a city'. The distribution of the accent in verbs and adjectives is very restricted, so much that the lexicon requires the marking of each verb and adjective with a morpheme feature [iaccented].5 A rule then assigns an accent to the [+accented] verbs and adjectives. McCawley (145) posits the following rule:6 S Verb 8
_

(8)

Adj

-}

[+acc] /

(C)(V) &

[+accented] But since an accent does not always appear on the penultimate syllable in inflected forms of verbs and adjectives, a rule that shifts the originally assigned accent is required. On the assumption that the penultimate accent is basic,7 a rule that specifically shifts an accent to the final syllable is needed. This ACCENT
ATTRACTION

the verbs and adjectives are supplied with the present tense, provisional, im6 It has been arguedby Martin1967 that adjectivesare alwaysaccentedeither on the stem-finalsyllable or the stem-penultimate syllable;but we will followMcC'sanalysisin this paper. 6 S, M, and & stand for syllable, mora,and morpheme boundary, respectively.Wewill follow McC'sboundarymarkersand their conventionsthroughout this paper:[@]major phraseboundary,[%]minorphraseboundary,[#]wordboundary,[#i]internalwordboundfor ary, [:] compound boundary,[a]syllableboundary accentplacement. 7 The assumption justifiedby McC (143-5). is

RULEshifts the stem-penultimate accentone morato the rightwhen

THE NON-CYCLICNATURE OF JAPANESEACCENTUATION perative, or honorific morpheme, as in the following examples:


PRESENT PAST

587

tabe & ta tabe & ru [+acc] [+acc] ta'be ta ta'be ru Rule 8 tabe' ru (not applicable) (attraction) tdbUtd tdbri Rule 1 taka & ta taka & i b. 'high' [+acc] [+acc] ta'ka i Rule 8 ta'ka ta taka' i (not applicable) (attraction) ta'ka 'kar ta (not applicable) ('kar insertion) Rule 4 ta'ka kar ta (not applicable) ta'ka kat ta (not applicable) (assimilation) tkdkakttd Rule 1 tdkdi Verbs and adjectives with a marking of [-accented] undergo neither 8 nor the accent attraction rule. (9) a. 'eat' Several other rules place, shift, or eliminate accents; but since they are specifically needed only for compound nouns, we will defer their discussion for the moment. By all the rules described above, most of the minor phrases, e.g. noun or verb phrases with one noun or verb respectively, can be assigned their correct pitches. The pitches relevant within minor phrases are [Hi] and [Lo], and no significant gradations of pitch are observed. But if one looks at major phrases, e.g. noun or verb phrases with more than one noun or verb, syllables with mid pitches appear. Thus, in major phrases like those of Figure 1, the first minor phrases have high and low pitches, while the second minor phrases have mid and low pitches.8 McC interprets this phenomenon as follows (172-3; bracketed insertions are my own):
Each of these phrases consists of two [or more] pieces, the first with a PRIMARY ACCENT, the second [and the rest if they exist] with a SECONDARYACCENT. Moreover, the two accents

are on the same syllable as would be accented if the corresponding pieces were pronounced
independently. To account for this phenomenon, it is thus NECESSARY TO ASSUME [emphasis

supplied] that each piece goes through regular accentuation rules (at each end of which each piece contains exactly one accent) and that the accent on the second [and the rest if they exist] is reduced from primary to secondary.

The rule needed reduces not only the second accent, but all accents after the first in a minor phrase. McC finds this accent reduction rule very similar to the accent elimination rule (4)-which, we may recall, eliminates all accents after the first (leftmost) one within a minor phrase. Thus McC proposes (173):
there are not two rules here but really one, an accent reduction rule which applies
IN A
8 McC (177) admits that there is a problem in introducing major phrase boundaries. Even though the exact definition of major phrase and minor phrase may not be entirely clear, our discussion holds. For a general discussion on boundary phenomena, see Stanley 1970.-In examples below, the macron indicates mid pitch.

588

LANGUAGE, VOLUME 48, NUMBER 3 (1972)

@% tabMtU% YImast%@ 'is eating'

@% yikkdri % hanasJmass7t %@ 'spoke slowly'

@% kb'utto % mtr;

%@ 'if (1) were to try putting on (a hat)'

FIGURE 1

CYCLE [emphasis supplied], first operating on smaller constituents

stituents,

and then on larger conreducing all accents [by one degree] except the first each time it applies.

He formulates the accent reduction rule (182) as follows (by convention, all the existing accents in a particular constituent are reduced by one degree when accent is placed somewhere else in that constituent): (10) Accent reduction rule: S -, (lacc) in env.... This rule 'accents the first primary accent in the constituent', thereby reducing all other accents in the constituent to which the rule is applying. McC also reformulates the accent placement and attraction rules, as given above, so that they read as follows (181): (11) Accent placement rule: S AdVjb -+ (lacc) / [+accented] (12) Accent attraction rule:9 (C)(V) &

(Present J ProvisionalI a -) (lacc) / (lacc) M (lacc) M & & Provison al -, r Imperative (lacc)/) [Honorific J Having reformulated the rules for cyclic use, McC (173) derives phrases like kabuttemitara as in Figure 2. For the tertiary accent, which is not pronounced, McC (173-4) offers the
' Facing a number of problems in noun compounds, McC (171) concludes that accents are placed on syllable boundaries rather than moras. This conclusion enables him to view what I have been calling unaccented morphemes as pre-accented morphemes, e.g. 'miyako 'city'. In my re-analysis I have followed this practice. Accordingly, ko' and 'ro in koko'ro 'heart' are spoken of as post-accented and pre-accented moras, respectively.

THE NON-CYCLIC NATURE OF JAPANESE ACCENTUATION

589

kabu'r & te % mi' & ta'ra

originalaccents
FIRSTCYCLE: accent reduction SECOND CYCLE: accent reduction

1
1 1

1
1 2

1
2 3

FIGURE2

following account:
One might give a rule to eliminate these accents; however, there is no need to do so: it will suffice to revise the rules for converting accent into pitch [our Rule 1] in such a way as to insure that the correct pronunciation results.

Now McC states the pitch assignment rules (174) as follows: (13) a. Everything in a minor phrase becomes high or mid pitched, depending on whether the strongest accent in the phrase is primary or non-primary. b. Everything after the first mora of the strongest accented syllable becomes low pitched. c. The first mora of the phrase becomes low pitched if the second is not low pitched. These rules are post-cyclic and interpret the output of the cyclic rules. Thus they interpret and assign pitches in the following manner to the output of Fig. 2: kabutte mi tara (14) 1 23 HiHiHiHi MidMidMid (by 13a) HiHiLoLo MidLoLo (by 13b) LoHiLoLo (not applicable) (by 13c) kdbuttU mitdrd 3. Having examined the process that has led McC to treat Japanese accentuation cyclically, we may now note a serious flaw in his analysis. McC's basic motivation for the cyclic analysis is, as he states, the similarity between Rule 4, which eliminates the accents following the leftmost one, and the phenomenon shown in Fig. 1, where high pitched syllables in a series of minor phrases are reduced to mid, except for the leftmost phrase. Let us, first, ask ourselves whether Rule 4 is required at all. McC's motivation for positing it is to account for the phenomenon shown in 5, where only the syllables up to the first occurrence of the accent are assigned high pitches. However, this does not mean that the second or the third occurrences of an accent must be eliminated. In fact, the pitch assignment of Rule 1 does not depend on the erasure of the non-first occurrences of accent. In full generality, Rule 1 assigns low pitches to all moras, regardless of whether or not they are accented, following the leftmost accented mora. The leftmost accented mora, since it is not preceded by another accented mora, is not low pitched; but all the following accented moras

590

LANGUAGE, VOLUME 48, NUMBER 3 (1972)

do follow an accented mora, namely, the leftmost one, and thus must be low pitched as Rule 1 says. Thus the underlying forms in 5 can be assigned correct pitches by Rule 1, even without going through accent elimination Rule 4. All this indicates that we do not really need Rule 4. Thus McC's basic motivation for a cyclic analysis is not very strongly founded, since it depends on a rule that is not necessary. Second, McC, looking at the phenomena in Fig. 1, thinks that the phrase which has high pitches and the one with mid pitches are due to a primary accent and a secondary accent, respectively. But there is no reason to assume that high, mid, and low pitches are derived from different underlying accent values. One can equally well posit that what is going on in Fig. 1 is just a SURFACE PITCH ADJUSTMENT, which leaves only the high pitched syllables in the leftmost minor phrase high, and that there are no grades of abstract accent-it is only a question of whether or not an accent is there. Thus I should like to claim that it is NOT 'necessary to assume... that ... the accent on the second piece is reduced from primary to secondary' (173). Unless McC finds some more compelling reason, there is no need to think that Japanese accentuation involves graded accents: all surface pitch gradations can be accounted for as surface pitch adjustment. In fact, McC has such an adjustment rule in his system-our Rule 13c. The above two arguments should weaken the case for a cyclic analysis; furthermore, there is a most unfortunate consequence of the cyclic analysis proposed by McC. Let us look at his example of derivation reproduced in Fig. 2 and in 14, examining carefully how cyclic application of a phonological rule operates. In the first cycle, the accent reduction rule applies to the maximum string that contains no boundaries. This rule does not apply at the word level, since each word presumably has one accent; hence the first maximum strings will be minor phrases like %kabu'rte % and %mi' ta'ra %. After the application of the accent reduction rule in the first cycle, the minor phrase boundaries are ERASED, so that the same reduction rule can apply to the entire phrase @kaburtemitara@ in the second cycle. (This is why McC does not have a vertical line separating the two minor phrases in the second cycle of Fig. 2.) Thus the principle of the transformational cycle crucially hinges upon the erasure of inner phrase boundaries. Post-cyclic pitch assignment rules (13) assign pitches to the output of the second cycle in our example, which is a major phrase containing no minor phrase boundaries. Now let us re-examine how pitch assignment rules are stated by McC. Note that all three rules in 13 depend on the information provided by MINOR
PHRASE BOUNDARIES.10

But when cyclic applicationsof the accent reduction

rules have erased minor phrase boundaries, the pitch assignment rules, which depend on them, cannot assign correct pitches. Thus, in addition to lacking real motivation, McC's cyclic analysis simply does not work. The principle of the transformational cycle works only when a language has no post-cyclic rules that depend on the information provided by inner labeled brackets or phrase boundaries."1The late pitch assignment rules in Japanese
10McCawley's second rule, given here as 13b, is not quite accurate; it must be so worded that 'everything' refers to the moras WITHIN A MINOR PHRASE. 11I am speaking about the original principle formulated by Chomsky, Halle, & Lukoff, which has been adopted by many linguists, including McC. It has been pointed out to me by

THE NON-CYCLICNATURE OF JAPANESEACCENTUATION

591

crucially depend on the information provided by minor phrase boundaries. Accordingly, the principle of the transformational cycle that requires the erasure of inner labeled brackets or phrase boundaries involves a requirement which cannot be met by Japanese accentuation. In other words, Japanese accentuation is non-cyclic in nature. 4. The remaining task of this paper is to show that a non-cyclic analysis accounts for the accentual phenomena of Japanese without involving any more complication than McC's cyclic analysis. The following analysis is essentially a re-analysis of McC; therefore it does not pretend to account for more than he attempts to account for. In particular, the rules for noun compounds are subject to further revision. As outlined in the foregoing sections, Japanese accentuation is rather straightforward, except as regards the rules for noun compounds. The accentual phenomena in noun compounds are the most difficult aspect of the entire system. I shall posit the following three rule schemata for them (recall that we assume, with McC, that accents are placed on syllable boundaries rather than moras; cf. fn. 9):
RULES: (15) COMPOUND Cl. < [-acc] / # M?
M
_

MO o

] #i Mo
] i

fMo [+acc] Mo 1 #i -[+de-accenting]J

a b c

(M)
(M) b

C2. a -- [+acc] / # [

M_

C3. r -, [-acc] / # [ ] #i . Mo[+acc] Mo# Rule Cla accounts for a compound like e'#imonoga'tari--e-monoga'tari 'picture story' by deleting the accent of the first noun when it is followed by another accented noun. Rule Clb accounts for compounds such as no'oryoku#ibetu--* nooryoku-betu'difference in ability' and 'huhei#ikao -- huhei-gao 'expression of displeasure', as pointed out by Okuda (750). Okuda's account marks these second compound elements, e.g. betu, kao, as [+de-accenting], which removes the accent of the first noun. Rule Clc is proposed for compounds like 'soroban#i tama' -(Cla)--+ soroban#itama' -(Clc)-soroban-dama 'abacus bead', and sya'kai#ito'o -(Cla)--* syakai#ito'o -(Clc)-syakai-too 'socialist party'. Rule C2a is required for compounds like ni'#i'kuruma -(Cla)-nifi'kuruma -(C2a)--+ ni#i'ku'ruma -(C3)--- ni-gu'ruma 'cart'. Rule C2b accounts for the case where the second element is not longer than two moras but contains the compound boundary [:], as in ka' :zoku#ihu:ro' -(Cla)-kazoku#ihu:ro' -(Clc)-kazoku#ihu:ro-(C2b)-- kazoku-bu'ro'family bath'. RichardStanleythat one can modifythe principle as not to eraseinnerbrackets,but to so just checkthem when a cyclic rule applies,so that a post-cyclicrule can still referto the brackets.Furthermore, since McCdoes not delete surface-structure nodesup to the tree whatis a minorphrase. phoneticlevel, he may be able to referto the nodesin determining
But this means there is no need to introduce boundary markers at all, for all P-rules can be assumed capable of referring back to surface-structure tree nodes. It is clear, in any case, that in Japanese there is no real motivation for a cyclic analysis; therefore my arguments against McC's cyclic analysis hold even if he should adopt Stanley's suggestion or defend

his analysisalongthe line of argument givenabove.

592

VOLUME NUMBER3 (1972) LANGUAGE, 48,

Rule C3 deletes the pre-accent of the second element when it is internally accented (cf. ni-gu'ruma). The earlier predominating Rule 6 can be formulated as an accent deletion rule, and the accent attraction rule (12) can be reformulated in terms of a transformational rule as follows:
(16) a. PREDOMINATING RULE:

a -- [-acc] /

. Mo & [+predominating]

b. ACCENT ATTRACTION RULE:

(Present Provisional M M[+ac] Imperative LHonorific 1 2 3 4 5 - 313245 We also need to posit an accent shift rule in order to account for a shift of the accent following a moraic consonant or a vowel mora which in turn follows another vowel. This rule will account for the shift of an accent observed in cases like 'yuu:bin#i'hako -(Cla)-> yuubin#i'hako -(accent shift)-> yuubi'n-bako 'mailbox':
(17) ACCENT SHIFT RULE:
o o

where M = moraic consonant or V/V.. M[+acc] 2 --21 1 Since we admit, following McC, pre-accented morphemes such as 'nu'si 'owner' and 'uta' 'song', we need to posit a rule that deletes the pre-accent when these morphemes are phrase-initial (otherwise, the pitch assignment rules give wrong pitches):
(18) ACCENT DELETION RULE: o- -> [-acc] / %

This rule deletes the pre-accent of 'nu'si, giving nu'si, which finally comes out as nzsi. All the rules given above are what I earlier called the 'first type' of rule; they manipulate underlying abstract accents. Now, we reformulate McC's pitch assignment rules-though these are best understood and stated as a convention which interprets the output of the accent-manipulating rules and assigns a phonetic pitch shape permitted in the language. Thus it is appropriate to look first at the pitch shapes permitted in standard Japanese. If we take a three-mora word, there are eight logically possible shapes: d. fiMMI g. ~MI~I[1 (19) a. fIS7l/L[ b. MM MMM e. MMI h. c. 'L[]1M f. But actually only three shapes occur phonetically: 19c, e, and f. Now if we look at what can and what cannot occur phonetically, we realizethat severe constraints govern pitch shapes in Japanese. These constraints are analogous to segmental surface-phonetic constraints (SPC's) such as the ones prohibiting [tu] and [til phonetically in Japanese.12We shall, therefore, state the constraints governing
12

Cf. Shibatani 1971for the notion of SPC's in segmental phonology.

THE NON-CYCLIC

NATURE

OF JAPANESE

ACCENTUATION

593

pitch shapes in terms of a positive SPC:


(20) SPC
FOR PITCH SHAPES OF MINOR PHRASES:

cML [-Ml [.Lo This states that the pitch shape of a minor phrase must satisfy the specifications given.l3 More precisely, it requires that the first two moras must disagree in the specifications of the feature [Lo], if a minor phrase contains two or more moras. This constraint rules out the occurrence of 19a, b, g, and h, all of which have feature specifications identical for the first two moras. Next, the SPC states that, once pitch drops from non-low to low, the following moras within a minor phrase can no longer have a non-low pitch. This constraint rules out the occurrence of 19d, which has a high pitch following a pitch drop. As observed, Japanese accentuation involves a high degree of redundancy in its phonetic pitch shapes, and the SPC captures this redundancy. This SPC furthermore describes what is and what is not possible as a surface-phonetic pitch shape in Japanese. Without having such a constraint, it is necessary to apply all the rules to all possible underlying forms in order to find out what is a possible phonetic pitch shape. Thus a description with SPC's has more descriptive power than one without them. Having studied the surface phonetic constraint that governs the pitch shapes of minor phrases, we can now state a convention for pitch assignment:
FOR PITCHASSIGNMENT: Assign pitches to minor phrases (21) CONVENTION

according to SPC 20 on the condition that a pitch drop occurs only at the leftmost accent of a minor phrase. The condition of the convention makes it necessary to identify the leftmost post-accented mora as the last non-low mora of a minor phrase, for it says that a [-Lo Lo] sequence occurs only on the two moras that have the leftmost accent between them. The condition further restricts the pitch shape of unaccented (or pre-accented) words to [Lo --Lo], for it rules out the other possibility of [-Lo Lol]. Unaccented words cannot have a pitch drop-they can only raise the pitch in conforming to SPC 20. Then, given forms like %M M' M M % and %M M M M %, for example, the convention assigns the pitch shapes [Lo -Lo Lo Lo] and [Lo - Lo - Lo - Lo],respectively. The convention, in otherwords, superimposes the phonetic suprasegmental features onto sequences of moras according to a phonetic-structure well-formedness condition captured by SPC 20. Thus the convention uses two types of information-where the accent is and what wellformed phonetic pitch shapes are; whereas the rules considered earlier use only the information provided in the input string. The pitch assignment convention given in 21 yields only two levels of pitch, [- Lo] and [Lo]. But since there are three levels of pitch in a major phrase, we need to further specify the [- Lo] pitch in such a way that the [- Lo] pitched moras are realized as [Hi] pitched moras in the leftmost minor phrase, and as [Mid] pitched moras in the other minor phrases of a major phrase. This can be achieved by a redundancy rule posited as SPC 22, which states (a) that a major phrase may
13 I am grateful to William S-Y. Wang and John Crothers for their suggestion concerning the formalization of this SPC.

594

LANGUAGE, VOLUME 48, NUMBER 3 (1972)

Convention21 SPC 22

@%kabu'r & te % mi' & ta'ra %@ -Lo Lo Lo Lo-Lo Lo Lo Lo Hi Lo Lo Mid Lo Lo


FIGURE3 a

Rule C1a Accent attraction Convention21 SPC 22

@% atama' atama atama Lo -Lo-Lo Lo Hi Hi


FIGURE 3 b

#i

ra'si ra'si rasi' -Lo -Lo Hi Hi

&i %@ &i i Lo Lo

Rule C1a Convention21 SPC 22

@% atama atama' Lo -Lo -Lo


Lo Hi Hi

#i

ra'si ra'si -Lo Lo Hi Lo

&ku % na' &i %@ ku na' i Lo -Lo Lo Lo Mid Lo

FIGURE 3 C

contain one or more minor phrases; and (b) that if there is a [-Lo] mora, then that is necessarily [Hi] in the leftmost minor phrase and [Mid] in the other minor phrases. Thus SPC's 20 and 22 together state phonotactic constraints and phonetic redundancies that govern Japanese pitch accent.
(22) SPC
FOR PITCH SHAPES OF MAJOR PHRASES:

@%(Lo) -Lo
Hi Hi

Loo % ((Lo) -Lo

Loo %)o

Mid Mid One may argue that SPC 20 and the convention given in 21 would not give a unique pitch to a monomoraic word: by convention, [aLo] specification ranges over [+Lo] and [-Lo]. But the actual situation is exactly what SPC 20 and the convention say. I.e., pitch is relative, and unless a monomoraic word in question is preceded or followed by another mora with a different pitch, one cannot tell whether it is high pitched or low pitched. Thus the Greek variable representation in 20 captures the actual situation. In summary, let us apply our non-cyclic rules to McC's examples of derivation,

NATURE OF JAPANESEACCENTUATION THE NON-CYCLIC

595

and see whether our re-analysis involves any greater complication. The examples kdbutti mitdrd 'if (I) were to try putting on (a hat)', dtdmd rdsii 'like a head', and dtdmdrdsiki nai 'not like a head' (McC 173, 176) are derived in Figure 3a-c.14 As one can see by comparing Fig. 3a with McC's derivation given in Fig. 2 and in 14, the non-cyclic derivations not only work effectively, but are also simpler and much more straightforward. When a non-cyclic analysis is as effective as a cyclic one, there is no need to introduce transformational cycles. And the Japanese case is even more serious, in that the cyclic analysis does not work.'5 REFERENCES 1968.The soundpatternof English.New York:Harper& CHOMSKY, and M. HALLE. N., Row. -- , -- , and F. LUKOFF. On accentand juncturein English.For RomanJakob1956. son, 65-80. The Hague: Mouton. to HALLE,M. 1963. On cyclicallyorderedrules in Russian.Americancontributions the Fifth InternationalCongressof Slavists, 113-32. The Hague: Mouton. P. KIPARSKY, 1966. ttber den deutschenAkzent. Studia Grammatica7.69-98. Berlin: AkademieVerlag. J. LARSON, 1970. A redefinitionof the terms 'tone language' and 'pitch language'. Presentedat the AnnualMeetingof the LinguisticSociety of America,Washington, D.C. LEE,G. 1969.Englishwordstress.Papersfromthe 5th RegionalMeetingof the Chicago LinguisticSociety, 386-406. T. LIGHTNER, 1965. Segmentalphonologyof Moder StandardRussian. Doctoral dissertation,MIT. S. MARTIN, 1967. On the accent of Japaneseadjectives.Lg. 43.246-77. LinguisticSociety of America. -- . 1968.The phonological of of component a grammar Japanese.The Hague:Mouton. -- . 1970. Some tonal systems that come close to being pitch accent systems but don't quite make it. Papers from the 6th Regional Meeting of the Chicago Linguistic Society, 526-31. K. of OKUDA, 1970. Review of The phonological componentof a grammar Japanese,by J. D. McCawley.Lg. 46.736-53. of MIT. Ross, J. R. 1969.A re-analysis Englishwordstress.Unpublished, Project on LinguisticAnalysis Reports (2nd series), 13. Berkeley: University of
J. MCCAWLEY, 1964. What is a tone language? Presented at the Summer meeting of the

SHIBATANI, M. 1971. The role of surface phonetic constraints in generative phonology.

S. SCHANE, 1968. French phonology and morphology. Cambridge: MIT Press.

California, Phonology Laboratory. [Revised version to appear in LANGUAGE.] STANLEY, R. 1970. Boundaries in phonology. Unpublished. Berkeley: University of

California. [Received14 May 1971]


14 McC

Since Japanese does not involve any accent reduction rules in his sense, it must be grouped with true tone languages according to one of his typological criteria. Another criterion of McC's has to do with the amount of information needed in the lexicon. According to this criterion, Japanese cannot be grouped with true tone languages, which means that McC's criteria fail to group Japanese either with English and Russian, or with true tone languages. McCawley 1970 tries to improve some of the inadequacies in his earlier criteria, but his revisions still raise a number of problems (cf. Larson 1970for further discussion).

16 Note that this conclusion has a serious impact on McC's typological criteria (1964).

(176) treats the phrase atama rasii as containing an internal word boundary, [#i].

Вам также может понравиться