Alamargot Chanquoy 2001

See
discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/280014349
Through the Models of Writing
Book · January 2001

DOI: 10.1007/978-94-010-0804-4
CITATIONS READS
199 1,916
2 authors:
Denis Alamargot Lucile Chanquoy

Université Paris-Est Créteil Val de Marne - Uni… University of Nice Sophia Antipolis
96 PUBLICATIONS 936 CITATIONS 86 PUBLICATIONS 762 CITATIONS
SEE PROFILE SEE PROFILE
Some of the authors of this publication are also working on these related projects:
ISEL [International Symposium on Educational Literacy] / SILE [Symposium International sur la

Litéracie à l'école] View project
TAO (Twictée pour apprendre l'orthographe) View project
All content following this page was uploaded by Denis Alamargot on 13 July 2015.
The user has requested enhancement of the downloaded file.

Alamargot, D., & Chanquoy, L. (2001). Through the models of writing. Dordrecht-Boston-
London: Kluwer Academic Publishers.
Usage interne - ne pas diffuser

PREFACE
ERIC ESPÉRET
University of Poitiers, France
Denis Alamargot and Lucile Chanquoy’s book offers a vivid and original presenta-
tion of main trends in the research field devoted to writing. First, it provides both
young and senior scientists with a comparative view of current theoretical models of
composition, with different levels of reading made available: each element of these
models is clearly situated in its historical context, and scrutinized in its further evo-
lution. Second, this well documented theoretical analysis of writing mechanisms is
checked against empirical data extracted from a lot of updated experimental studies;
and lack of necessary data is thought to be underlined and defined when noted.
Following the usual description of writing phases initially proposed by Hayes
and Flowers, the first part of this book presents planning, translating and revision
processes and compares them to other researchers’ conceptions (from Bereiter and
Scardamalia, to Kellogg or Galbraith). Such presentations of isolated models do ex-
ist in literature; but the present work really gives a good comparative analysis of
components inside each of models, in a clear and cumulative way; a fine-grained ob-
servation of differences between similarly-looking models is also performed.
Such an overview allows us to note a strong evolution in conception and modeli-
zation of writing activity: mainly descriptive and general at the beginning, models
become more and more functional, often more local, and try to define exact sub-
processes they comprise (for example, sub-processes of revising) and the relation-
ships linking these elements. They also take more and more into account cognitive
costs implied in managing such a complex activity, as well as individual differences
and developmental or environmental aspects. New global theoretical conceptions are
progressively invoked: connectionist or modularist views, for instance. Besides,
more and more experimental works are led in this field, because of an higher preci-
sion of models. Strict psycholinguistic aspects nevertheless remain the poor relation
in these approaches. All issues are questioned by authors caring of precision and
rigor.
The second part is entirely devoted to the cognitive functioning of writing. Deci-
sive issues are thus examined: are there sequential or parallel processes? How many
cognitive resources do they consume? Does there exist a specific control structure
managing different processes?
A particular chapter is devoted to a central component of most cognitive system
described in literature, Working Memory, whose role is not usually analyzed in clas-
E. Espéret. Preface. In G. Rijlaarsdam (Series ed. ) & D. Alamargot & L. Chanquoy, Studies
in Writing: vol 9. Through the Models of Writing, 1 –2 . © 2001 Kluwer Academic Publish-
ers. Printed in the Netherlands.
sical models of writing. A recent version of this approach, Kellogg’s one, put for-
ward the idea that the 3 classical writing processes are linked to the bringing to bear
of the 3 components in Baddeley’s model of Working Memory. The last chapter,
very interesting, is about expertise. It allows us to differentiate developing capacity
with age and acquisition (or learning) of writing with practice. Such a differentiation
is performed by analyzing evolution of each sub-process and checking contributions
from different developmental models (including neo-piagetian ones). And we cannot
evade the final question: does really exist a developmental model of writing? Is it
possible to get a model similar to those produced about spoken language, for which
a social institutionalized learning does not exist? While different possible models are
carefully examined, author is always keeping in mind the necessity of never forget-
ting functional aspects in change.
To conclude this fine presentation, in an interesting way, two prominent re-
searchers in this field, J.R. Hayes and R.T. Kellogg, have been invited to react to
analyses developed in the book and to complete the presentation of their own model.
At last, this book is pleasant to read, useful to people working in teaching of writing
or studying this specific human activity; it has to be included in students’ lists of
references.
GENERAL INTRODUCTION
A definition of writing and a presentation of the main models
1 WRITING A TEXT: A COMPLEX TASK

1.1 To elaborate a content, to write it, and to modify it
Writing a text is a complex task that needs a coordinated implementation of a large
set of mental activities. Writers have to clearly delimitate the nature, the goal and the
communicative function of the text. They also have to establish a precise representa-
tion about readers’ characteristics and expectations, in order to anticipate systemati-
cally what must, or can, be written. Writers have equally to control the text topic so
as to generate or to specify the most relevant ideas that will progressively constitute
the text content. In addition, they must sometimes clarify the message, reorganise,
modify and articulate ideas, while controlling the whole text coherence.
Surrounding, delimiting and adapting the text content constitute an important
writing phase. In addition, it is also necessary to put ideas into words, that is to for-
mulate them, throughout the writing process. This activity does not mean to simply
copy out some words or an isolated sentence, but to clearly formulate a set of coher-
ently articulated sentences, without any redundancy or, conversely, without too
many thematic ruptures. To realise these operations, it is necessary, at least, (1) to
choose the ‘appropriate words’ for each idea, (2) to use very strict syntactic, gram-
matical and orthographic rules, (3) to use correct punctuation and connection marks,
in order to translate, in terms of linguistic relations, the semantic relationships link-
ing these ideas.
These mental activities are still not sufficient to elaborate a text. A satisfactory
text is only very rarely produced during the first trial. It is often the result of an im-
portant number of drafts, corrections, scratches, additions, and so on. These succes-
sive versions can certify both surface modifications, such as orthographic correc-
tions, and deeper modifications, such as the reorganisation of the text organisation.
To correct or to modify a text supposes that the writer can evaluate the quality and
the pertinence of her/his own production. S/he has to read what has been written,
during all the writing process, in order to be able to continue to write as well as to
modify the previous text.
D. Alamargot & L. Chanquoy (2001). General introduction. A definition of writing and a

presentation of the main models. In G. Rijlaarsdam (Series ed. ) & D. Alamargot & L. Chan-
quoy, Studies in Writing: vol 9. Through the Models of Writing, 1– 29. © 2001 Kluwer Aca-
demic Publishers. Printed in the Netherlands.
2 ALAMARGOT & CHANQUOY
1.2 To manage writing constraints

There is no optimal method to conceive the text content, to elaborate its linguistic
form, to revise its content or its form and, consequently, to activate and to articulate
these three activities throughout the writing process.
The writer must take into account a great number of clues that come from, for
example, the writing context, imposed or posed goals, the state of writing, the text
already produced, etc. This set of clues can represent as much constraints as the
writer has to understand and to use in order to manage the necessary mental activi-
ties during writing and to control their efficiency. In other words, the writer has to
elaborate an efficient writing strategy.
However, these clues do not constitute the unique constraints for the choice of a
strategy. The writer’s capacities to simultaneously operate a more or less important
number of mental activities can equally play a central role. For example, the most or
less important limitation of processing capacities can constrain the writer in the
choice of appropriate mental activities, and perhaps in the activation of these activi-
ties. Thus, a writer not being able to think about the text continuation while writing,
must frequently alternate phases of content research and phases of handwriting.
When we look at this great number of mental activities to be realised, managed
and controlled, answerable to external as well as internal constraints, it is easy to
understand that the writing expertise could only develop very slowly in children, and
sometimes even to be very approximate in adults. To write good quality texts, both
linguistically and semantically, requires from the writer a lot of practice, training
and technique. Thus expertise certainly needs (1) the activation of writing strategies
more complex than those generally used by novices, (2) the possibility to simultane-
ously manage a greatest number of constraints, and/or (3) a sufficient writing prac-
tice to automatise letter transcription or spelling rules.
1.3 Text production: a possible definition in the context of cognitive psychology

The diversity of mental activities of text writing implies that its study is both attrac-
tive and difficult. The text production activity can thus be very relevant and interest-
ing for cognitive psychology because it supposes the activation of a great number of
processes, that do not only concern verbal production, but equally reading and com-
prehension. In this view, the study of text production offers the possibility to analyse
the functioning of numerous mental mechanisms that have often been defined and
studied in isolation in other psychology areas. It is thus possible to study the condi-
tions of process activation and interaction, processes ensuring, for example, (1) con-
tent retrieving and planning, (2) information encoding and maintaining in Working
Memory, or (3) text reading and understanding, in very specific contexts.
From a cognitive viewpoint, we think that text production can be defined as a
finalised and complex activity, because it supposes to process, by the implementa-
tion of several mental processes, and with a general goal – to write in order to com-
municate, for example – a great amount of knowledge. Thus, writing a text can be
compared to a problem-solving situation whose resolution implies complex cogni-
GENERAL INTRODUCTION 3
tive activities and abilities. In such a problem-solving context, the resulting product -
the written text - requires at least the uses of four types of knowledge: (1) Domain
knowledge (the conceptual domain to be expressed in the text), (2) Linguistic
knowledge (grammatical rules and lexical items that compose the text), (3) Prag-
matic knowledge (that allows the writer to adapt for the addressee the conceptual
content as well as the linguistic form of the text), and (4) Procedural knowledge (in
order to use the three preceding types of knowledge and to strategically process
them).
A general writing process fulfils the processing of each of these kinds of knowl-
edge. This process ‘transforms’ the domain knowledge into a (necessarily) linear
linguistic product, which must comply with a specific communicative goal. Due to
the limited processing capacity of the cognitive system, the global transformation
process is progressive and subdivided into a given number of sub-goals (i.e. to
elaborate, to write, to modify, etc.). Each sub-goal is realised by series of specific
processes (i.e. ordered sequences of mental operations) which are controlled by pro-
cedural knowledge (i.e. knowledge for the application of processes).
1.4 Some models to delimit and to define the necessary processes and knowledge
It is easy to understand, while reading this possible definition of text writing activ-
ity, that researchers’ important difficulty is to be able to identify, to study and to in-
tegrate, in a complex system, these different mental mechanisms (Cf. Hayes, 1989).
The complexity of this system is such that it is not possible to try this integration
without using a ‘model’ in order to delimit, to surround and to a priori define proc-
esses, knowledge and modes of processing necessary for the production of a text.
The term of model is here considered as a blueprint, a simplification, or an outline.
Text writing models allow researchers to focus on some dimensions of the writing
task, without forgetting that these dimensions belong to a complex system. Mainly
prospective, these models propose a relatively precise and analytic definition of the
writing activity, both concerning the process architecture (in terms of arrangement
of these processes in models as well as in terms of definitions of sub-processes or
operations that compose the processes) and functioning (in terms of process man-
agement rules, control and activation in Working Memory). These models lead to a
great number of experimental studies, but have still not reached a sufficient level of
formalisation in order that a computer simulation, for example, could be possible.
1980, 1987, 1989 and 1996 represent four very important years for those cogni-
tive psychologists who are interested in text production activity. It is in 1980 that
Hayes and Flower published the first general model of text writing. After that, Bere-
iter and Scardamalia elaborated the first developmental writing model, in 1987. Two
years later, in 1989, Levelt proposed a very precise modelisation of speaking activity
that has greatly influenced models and works in the area of writing. Finally, 1996 is
a prominent year. It corresponds to the modification, by Hayes, of the initial Hayes
and Flower’s model, and to the publication, by Kellogg, of a model of text writing
articulating processes and Working Memory, as conceived by Baddeley (1986).
2 THE MAIN MODELS OF VERBAL PRODUCTION

2.1 1980: Hayes and Flower’s model
Hayes and Flower’s model has been published in a collective book entitled ‘Cogni-
tive processes in writing’ and published by Gregg and Steinberg in 1980. It is proba-
bly one of the first books standing up for a cognitive approach to writing. The model
elaborated by Hayes and Flower is presented in two successive chapters in this book.
The first chapter, entitled ‘Identifying the organisation of writing processes’
proposed a description of writing processes and architecture while a second chapter,
entitled ‘The dynamic of composing: Making plans and juggling constraints’, writ-
ten by Flower and Hayes, rather focused on the functioning of the model, in terms of
constraints linked to the activation of processes.
On an architectural viewpoint, the model of Hayes and Flower (1980) is com-
posed of three main parts: the Task Environment, the writer’s Long Term Memory,
and the general writing process (Cf. Figure 1).
TASK ENVIRONMENT
WRITING TEXT
ASSIGNMENT PRODUCED
Topic SO FAR
Audience
Motivating Cues
THE WRITER’S PLANNING TRANSLATING REVIEWING

LONG TERM
Organizing Reading
MEMORY
Gene-
Knowledge of Topic rating
Goal
Knowledge Editing
Setting
of Audience
Stored Writing Plans
MONITOR
Figure 1: Hayes and Flower’s model, adapted from Hayes and Flower (1980). Copyright ©
1980 by Lawrence Erlbaum Accociates. Adapted with permission.
The ‘Task Environment’ comprises all that is outside of the writer and can influence
the performance. This environment is composed of:
• some writing instructions that determine (1) the general theme of the text to be
written (Topic); its communicative goal (Audience), and (2) some motivational
factors deriving from the writing situation (Motivating Cues),
• the text, that, gradually written (Text produced so far), is going to be used as a
reference for the writer, in order to progress as well as to revise the already writ-
ten text.
The ‘Writer’s Long Term Memory’ contains three knowledge areas concerning:
• the general text topic (domain knowledge – Knowledge of topic),
• the communicative act (pragmatic knowledge – Knowledge of Audience), and
finally,
• linguistic knowledge about specific text plans, for example, story grammars
(Stored Writing Plans).
The general writing process is composed of three processes that allows to transform
domain knowledge in a linguistic product (with their sub-processes and/or associ-
ated operations) and a process of control. These processes are:
• the ‘Planning’ process with three sub-processes: ‘Generating’, ‘Organising’ and
‘Goal-Setting’,
• the ‘Translating’ process,
• the ‘Reviewing’ process, with the sub-processes of ‘Reading’ and ‘Editing’,
• the management and control process, called ‘Monitoring’ and that defines the
order of activation of the three preceding processes.
More precisely, the main function of the Planning process is to establish a writing
plan from (1) domain knowledge retrieved from Long Term Memory, and (2) in-
formation extracted from the task environment. This plan guides text writing by de-
fining the main goal and the sub-goals. As seen above, this plan can also be re-
trieved from Long Term Memory, if it has been stored among the writer’s knowl-
edge (Stored Writing Plans). Otherwise, it has to be built through three sub-
processes: the retrieving (Generating) of the different pieces of knowledge stored in
Long Term Memory; their organisation (Organising) in a writing plan, and the
elaboration of criteria that will allow to judge the appropriateness between the writ-
ten text and the intentions (Goal Setting).
The Translating process runs under the control of the writing plan and translates
domain knowledge in language. According to Hayes and Flower (1980), the func-
tions of this process are: (1) to retrieve, from Long Term Memory, complementary
knowledge allowing (2) to develop each part of the writing plan before (3) translat-
ing the retrieved propositions in correct sentences (by means of lexical and gram-
matical processing).
Finally, the Reviewing process evaluates the appropriateness between the written
text and the linguistic, semantic and pragmatic particularities of the writing goal.
Two sub-processes carry out the revising activity: the analytical reading of the al-
ready written text (Reading) and its possible correction (Editing).
These three processes are managed by a control process – Monitoring – whose
function is, among others, to regulate the recursion of their application.
The methodology adopted by the authors, to elaborate and validate this model,
involved a verbal protocol analysis that was developed by Newell and Simon (1972)
to study problem-solving processes. Although elaborated only on the basis of the
protocol of one unique participant (speaking about his mental activities during ex-
pository text writing), the identification of processes and writing representations

proposed by Hayes and Flower (1980) is, still today, a prominent basis among all the
available models. It is nevertheless clear that this model has to be considered as a
framework tempting to identify the different processes, rather than as a procedural
model aiming at explaining the functioning and, more particularly, the relationships
between writing processes.
In fact, the formalisation proposed by these two researchers has allowed to pre-
cisely surround, for the first time, writing processes and representations. This limit
has very rapidly led to arouse a great number of experimental works whose objec-
tives concern as well the validation of the different processes such as proposed by
Hayes and Flower as their precision. However, the main interest of this model con-
cerns the fact that writing activity, at least in its technical aspects, can become a real
area of investigation, in the field of cognitive psychology. This activity can be ap-
proached as a complex human activity, as problem-solving or chess activities (Cf.
Newell & Simon, 1972). In other words, if researchers in cognitive psychology can
thus benefit from a new field of experimental studies, the writing activity can conse-
quently be considered as an important area, with a first possible definition in the
framework of the theory of information processing.
As established by the modelisation proposed and the concepts introduced by
Hayes and Flower (1980), the described processes and representations are, a priori,
present and efficient only in the most skilled and expert writers. A developmental
perspective of such an expertise in text writing activity will be described in 1987 by
Bereiter and Scardamalia.
2.2 1987: Bereiter and Scardamalia’s developmental model

Initiated in 1980 by Bereiter and published in its entirety in 1987, in a book entitled
‘The psychology of written composition’, Bereiter and Scardamalia’s developmental
model is made up of two main writing strategies. The first one, labelled ‘Knowledge
Telling Strategy’ is a novice strategy that globally consists of producing a text by
formulating ideas as they are retrieved or generated from Long Term Memory,
without reorganising the conceptual content or the linguistic form of the text.
The other strategy, more expert (and thus more frequent in teenagers and adults),
is called ‘Knowledge Transforming Strategy’. It needs to proceed to readjustments
of the text conceptual content according to both rhetorical and pragmatic goals and
the text produced so far. In others words, this strategy allows to re-elaborate the
conceptual content as well as the text-linguistic form, since they are perfectly appro-
priate to the writer’s intentions and communicative goals.
These two strategies must not be considered as two steps in the expertise devel-
opment, but rather as two extremes on a continuum. According to Bereiter, Burtis
and Scardamalia (1988), the development of writing expertise needs to be consid-
ered as a progressive change from Knowledge Telling to Knowledge Transforming,
via intermediate strategies.
2.2.1 The Knowledge Telling Strategy

The architecture of Knowledge Telling Strategy is articulated through three compo-
nents (Cf. Figure 2 ).
Mental representation
of assignment
Knowledge Telling
Process
Content Discourse
Know- Locate topic Locate genre Know-
ledge identifiers identifiers ledge
Construct
memory probes
Retrieve content
from memory
using probes
Run test Fail

of appropriateness
Pass
Write (note, draft, etc.)
Update mental
representation of text
Figure 2: Knowledge Telling Strategy,adapted from Bereiter & Scardamalia (1987). Copy-
right © 1987 by Lawrence Erlbaum Associates. Adapted with permission.
First, there is the mental representation of instructions (Mental Representation of as-

signment) allowing the writer to define the text topic and function. This representa-
tion is built in order to apprehend the task and to guide the whole writing activity.
The second component contains two types of knowledge in Long Term Memory,
necessary for the writing activity: topic knowledge and discourse knowledge. The
knowledge about the topic (Content Knowledge) allows the writer to elaborate the
text content. It globally concerns all the things that the writer knows about the topic
that must be developed in the text. The text knowledge (Discourse Knowledge)
mainly concerns linguistic knowledge (for example, lexical and syntactic informa-
tion) and knowledge about the type or the nature of the text (narrative, argumenta-
tive, expositive, etc.). The use of possible text schemas (as for example the narrative
frame) supposes that this schema belongs to the writer’s Discourse Knowledge.
The third component is the writing process, labelled ‘Knowledge Telling Proc-
ess’, maintains close relationships with the two other components. It is composed of
seven operations (or stages). The first two operations, called ‘Locate Topic Identifi-
ers’ and ‘Locate Gender Identifiers’ allow identifying the text topic (for the first
one) and type (for the second one). These two operations depend on the mental rep-
resentation of writing instructions, elaborated by the writer (Mental Representation
of Assignment). They guide the functioning of the third operation ‘Construct Mem-
ory Probe’. This operation consists of elaborating various memory probes to re-
trieve, via the fourth operation ‘Retrieve Content from Memory using Probes’,
knowledge relative to the content (retrieved from Content Knowledge) and to the
text nature (from Discourse Knowledge). Some tests called ‘Run test of Appropri-
ateness’ are carried out on these retrieved contents to examine their appropriateness
with the text topic and nature. If these contents are considered as inappropriate,
some memory probes are once again elaborated, controlled by the operations ‘Lo-
cate Topics Identifiers’ and ‘Locate Gender Identifiers’. If the retrieved knowledge
is coherent with the text topic and nature, they can be translated by the ‘Write’ op-
eration.
At the end of this handwriting stage, the seventh and last operation ‘Update men-
tal representation of text’ consists of a very local control of the writing activity by
allowing the writer to identify what has been written and what has to be written. It is
on the basis of this analysis that the elaboration of new memory probes can be again
engaged, by a new functioning of the ‘Construct Memory Probe’ operation.
Finally, the Knowledge Telling Strategy, partly analogous to those described by
Hayes and Flower (1980: 20) as ‘Get it down as you think of it…’, could consist of
using oral and conversational skills in a written task (Best, 1989). The text is thus
elaborated without a real (re)organisation of text content and, in return, of knowl-
edge domain implicated in this elaboration. The Knowledge Telling Strategy is eco-
nomic, but can only provide some local coherence between two ‘idea units’.
2.2.2 The Knowledge Transforming Strategy

The Knowledge Transforming Strategy corresponds to a rather expert writing and
that supposes, opposed to Knowledge Telling Strategy, to proceed to an important
planning of the text content, according to text rhetorical, communicative and prag-
matic constraints (Cf. Figure 3).
The Knowledge Transforming Strategy has the three main components of the
Knowledge Telling Strategy:
• The Mental Representation of Assignment component, allowing the writer to
understand writing instructions,
• The two knowledge areas stored in Long Term Memory: the Content Knowl-
edge and the Discourse knowledge,
The Knowledge Telling Process, whose function is, as for the Knowledge Telling
Strategy, the retrieval and the linguistic translation of the text content, according to
the text topic and genre.
Mental representation
of assignment
Problem analysis and

Goal Setting
Content Discourse
Knowledge Knowledge
Problem
Translation
Content Rhetorical
Problem Problem
Space Space
Problem
Translation
Knowledge
Telling
Process
Figure 3: Knowledge Transforming Strategy,adapted from Bereiter & Scardamalia (1987).

Copyright © 1987 by Lawrence Erlbaum Associates. Adapted with permission.
The main difference between the two strategies is the presence in the Knowledge
Transforming Strategy of a complex problem-solving system, operating between the
Mental Representation of Assignment component and the Knowledge Telling Proc-
ess. The interactive functioning of three specific components of the Knowledge
Transforming Strategy supports this problem-solving system:
• The ‘Problem analysis and Goal Setting’ component allows the expert writer to
realise, according to writing instructions, a complex analysis of (a) the task to
be carried out, (b) the objectives to reach, and (c) the necessary means to reach
this objective. This planning activity on the task allows to control two ‘sub-
problems’ solving, respectively linked to content planning (‘what to tell’) and to
the rhetorical process planning to express these contents (‘to who and how to
tell’).
• These two kinds of planning are managed in their own spaces. The ‘Content
Problem Space’, related with the Content Knowledge, transforms or modifies
the domain knowledge. The ‘Rhetorical Problem Space’, in relation with the
Discourse Knowledge, retrieves or modifies pragmatic or rhetorical processes.
A central point in the functioning principle of the Knowledge Transforming
Strategy concerns the fact that content and rhetorical problem-solving operate in
a close interaction.
• Thus, the ‘Problem Translation’ component transfers constraints or goals elabo-
rated in one specific space to another space. It is thus possible for the writer to
transform domain knowledge within the Content Problem Space, by taking into
account goals as well as rhetorical and pragmatic constraints defined within the
Rhetorical problem space (for example: ‘it is necessary to specify this content
for the reader’). Conversely, it is equally possible to modify, in the Rhetorical
Problem Space, pragmatic and rhetorical dimensions according to content con-
straints (for example: ‘even if this content is complex, it is absolutely necessary
to literally transcribe it in the text because it is central, and it is therefore neces-
sary to signal to the potential reader to carefully read this passage, even if it is
difficult’).
The totality of the problem-solving system in the Knowledge Transforming Strategy
must mainly elaborate contents before their formulation using the Knowledge Tell-
ing Process. In the Knowledge Telling strategy, these contents are directly exploited
from the Content Knowledge and the Discourse Knowledge, without any possibility
of transformation or preliminary adaptation. In the Knowledge Transforming Strat-
egy, the text content organisation does not reflect the direct domain knowledge or-
ganisation, or does not concern the realisation of a pre-elaborated text plan, as in the
Knowledge Telling Strategy. Conversely, it is a well-thought-out elaboration, taking
into account all rhetorical and content constraints of the writing task; such as it has
been specified in writing instructions.
The content transformation is thus realised through the control of two compo-
nents. The first one, as previously mentioned, is the Mental Representation of As-
signment component that manages the problem-solving activity. The second one
concerns the linguistic product effectively transcribed after the Knowledge Telling
Process. The analysis during text writing thus supplies, by feedback, the problem-
solving activity. The influence of the already written text is symbolized in the model
(Cf. Figure 3, p. 9) by the arrow from the Knowledge Telling Process and coming
back to the Problem-solving component. This feedback on problem-solving proc-
esses seems very important because it gives to the expert the possibility to compose
a text on the basis of planned intentions as well as on an analysis of the appropriate-
ness between linguistic product and intentions.
According to Bereiter and Scardamalia (1987), the possibilities to compose a text
with Knowledge Transforming Strategy are reserved to experts because it is neces-
sary to be able (1) to establish content plans as well as rhetorical plans while consid-
ering (2) all the constraints inherent to content and rhetoric spaces. Thus, for these
authors, the most plausible interpretation to explain the belated age to which a young
writer masters this strategy would be linked, in terms of maturation, to the increase
of the planning span (allowing to manage goal setting activities) and to Short Term
Memory span (allowing to maintain the constraints of the problem-solving activity
and the different knowledge units).
2.2.3 The interests linked to the developmental model elaborated by Bereiter and
Scardamalia
The ‘double’ model of Bereiter and Scardamalia (1987) has, at least for us, three
main interests. First, it allows integrating a developmental dimension in writing re-
search, by precisely defining processes and strategies whose implementation is pro-
gressive. Second, this model stresses the role, in the expertise development, of
pragmatic, communicative and rhetorical knowledge. Finally, the evolution of
Knowledge Telling Strategy to Knowledge Transforming Strategy questions the de-
termination of the expertise. Does the development of this expertise depend on gen-
eral factors (as the evolution, with the age, of capacities) or, conversely, on a par-
ticular expertise developing in a specific area such as writing ?
2.3 1989: Levelt’s speaking model

Hayes and Flower’s (1980) and Bereiter and Scardamalia’s (1987) models are
mainly general models because their authors have tried to define and to surround the
totality of processes and knowledge necessary for writing. Some years later, Levelt
has pursued the same general objective in the field of oral language. This model was
published in 1989 in a book written by Levelt and entitled: ‘Speaking: From inten-
tion to articulation’. Although the relationship between speaking and writing does
not seem very direct, this book constitutes a real contribution in writing research,
especially because this model offers to researchers the possibility to establish a theo-
retical comparison between oral and written processes, representations and types of
processing; or at least, to use the different types of processes proposed in speaking
to explain the management of writing.
2.3.1 From the Conceptualizer to the Articulator

Levelt’s (1989) model aims to describe speaking (oral language production) in
adults and follows the model of sentence production previously elaborated by
Garrett (1975, 1980, 1982, 1984, 1988), in a psycholinguistic perspective. The gen-
eral architecture of Levelt’s model is composed of four different components (Cf.
Figure 4).
The ‘Conceptualizer’ transforms domain knowledge retrieved from Long Term
Memory into a ‘preverbal’ message. This transformation is carried out during two
stages, respectively called ‘Macroplanning’ and ‘Microplanning’. Macroplanning
ensures both the realisation of the communicative goal and, the retrieval, the selec-
tion and the organisation of the domain knowledge necessary to reach these commu-
nicative goals. In this perspective, Macroplanning defines the ‘speech act’, or what
must be told and the way in which to tell it (in terms of commitment or politeness,
for example). Microplanning consists of appointing semantic functions to domain
knowledge units and to pragmatic parameters that constitute the preverbal message.
Microplanning allows to determine, for example, what unit can play the role of
agent, what unit has to be emphasised, etc. The content of microplanning of the fu-
ture sentence ensures thus an ‘informational perspective for an utterance’ by deter-
mining, among others, semantic structures as Given-New, Topic-Comment, the roles
of agent, patient, and so on.
CONCEPTUALIZER
Discourse model
Message Situation knowledge
Generation Encyclopedia, etc.
Monitoring
Parsed speech
Preverbal message
FORMULATOR SPEECH-
COMPREHENSION
Grammatical SYSTEM
encoding LEXICON
lemmas
Surface structure
forms
Phonological
encoding
Phonetic plan
(internal speech) Phonetic string
ARTICULATOR AUDITION
overt speech
Figure 4: Speaking production model, adapted from Levelt (1989). Copyright © 1989 by
M.I.T. Press. Adapted with permission.
The ‘Formulator’ allows lexical, grammatical and phonological encodings of the

preverbal message, in order to elaborate its linguistic form. This encoding produces
an entry information (input) for the following component. It transforms the output
provided by the Conceptualizer in a linguistic structure. Thus, to access words dur-
ing speaking, it is necessary to use two sub-processes: the ‘Grammatical Encoding’,
to select a lexically and grammatically appropriate item (the ‘lemma’), and the
‘Phonological Encoding’, to transform lemmas into lexemes, via the computation of
a correct phonetic program for the item in the context of the utterance (Levelt,
1992).
The ‘Articulator’ converts the phonetic plan (or ‘internal speech’), elaborated
through a syntactic structure and lexical forms, in speech, with the help of motor
programs. It allows the execution of the message.
Finally, Levelt includes in his model a control process, called ‘Self-Monitoring’,

assimilated to a comprehension system. The outputs elaborated by this process can
consist both of external or internal (mental) discourses, allowing to control all lan-
guage aspects. This process could globally correspond to the Revising process as de-
scribed in Hayes and Flower’s model (1980).
2.3.2 Adapting Levelt’s model to writing

This distribution of conceptual and formulation processes, conceived by Levelt
(1989) in the framework of speaking, has been adapted by van der Pool (1995) and
van Wijk (1999) in the case of text writing (Cf. also van der Pool & van Wijk,
1995).
Invent Comprehend
Conceptual Text record
processes Representations
Macro of text, intentions Macro
Processes and expectations Processes
Knowledge base
Micro Micro
Word knowledge
Processes Processes
Discourse knowledge
Preverbal message Parsed string
Encode Decode
Linguistic
Grammatical Lexicon Grammatical
processes
Lexical Lemmas Lexical
Phonological Lexemes Phonological
Morphological Morphological
Graphemic Graphemic
Graphemic plan Graphemic string
Execute Perceive
Physical
processes motor sensory
activities activities
Text
Figure 5: Model of written production inspired by Levelt’s model (1989),

adapted from van Wijk (1999). Copyright © 1986 by the Amsterdam University Press.
Adapted with permission.
For example, van Wijk’s model (Cf. Figure 5) shows more precisely and clearly than
Levelt’s (1989) framework, the different operations of each component, particularly
concerning the Self-Monitoring component. This component comprises of decoding
and comprehension activities. Some terminological differences can equally be no-
ticed. For example, the Conceptualizer is called here Invent and the Formulator is
called Encode. Nevertheless, these two components have the same processes that
those described by Levelt (1989): concerning conceptual processes of Macroplan-

ning and Microplanning or linguistic processes relying on the Lemmas principle.
The main point of this adaptation resides in the integration of operations, processes
or representations linked to writing, as the ‘Graphemic Encoding’, the ‘Graphemic
Plan Elaboration’ and the ‘Graphic Execution’ of this plan.
It is interesting to observe, in van Wijk’s model, that the adaptation of a speaking
model for writing can be conceivable on a strictly formal plan. Conversely, many
questions arise as soon as it is necessary to explain, on a functional viewpoint, the
very deep differences between speaking and writing activities. It is interesting to ob-
serve elsewhere that few authors consider such a comparison, while it could cer-
tainly contribute to better understand specificities and particularities of writing.
2.4 1996: A new architecture and a process model

1996 is the year of ‘The Science of Writing’. This collective work, published by
Levy and Ransdell, makes the point on theories and studies in text writing, in order
to measure its evolution, sixteen years after the appearance of Gregg and Steinberg’s
book. Two main developments are shown through this work. The first concerns the
general architecture of writing models, and the second concerns their modes of func-
tioning. On the plan of the architecture, Hayes proposes an important revision of
Hayes and Flower’s original model (1980), by modifying the definition and the bal-
ancing of the different processes as well as the nature of knowledge implied in the
activity of text production. Concerning the functioning of processes, a main over-
hang is proposed by Kellogg who considers a theoretical matching between a model
of text production and a Working Memory model, such as defined by Baddeley
(1986).
2.4.1 The revision of Hayes and Flower’s model (1980) by Hayes (1996)
As seen previously, the main interest of Hayes and Flower’s model is to have pre-
cisely defined, for the first time, processes and knowledge necessary in writing ac-
tivity, and to have considered them in a system of information processing. Neverthe-
less, this initial model is still incomplete and the authors themselves were aware of
the limits of this architecture that they qualified as ‘a target to shoot at’ (Hayes &
Flower, 1980: 29). In addition, many authors, since the publication of this model,
have pointed out and specified some of its limits. Hartley (1991), for example, has
addressed five critical points to the model:
• The context in which the model has been developed is not sufficiently specified,
• The model is purely descriptive,
• It does not specify what could be the effects of the writing medium (paper and
pen vs. computer) on the writing process,
• It only concerns the writer who works alone. The collaborative writing is thus
not taken into account while this kind of activity is increasingly frequent, nota-
bly with the appearance of electronic mails,
• It ignores inter-individual differences, mainly those concerning the writer’s

style, emotions and sentiments, and even sex differences.
Kintsch (1987) has also criticized Hayes and Flower’s model, mainly because these
authors do not consider the writer’s creativity. According to Kintsch, the creativity
allows the writer to go further than a simple recall of information retrieved from
Long Term Memory. It would be therefore necessary, according to Kintsch, to con-
sider a new component, allowing domain knowledge transformation, that could be
more general than the sub-process of knowledge organisation, and that allows the
elaboration of new ideas from retrieved information.
In addition, Kemper (1987) underlined that Hayes and Flower (1980) in their
model, but also Kintsch (1987) through his remarks, totally neglected the writer’s
goals and motivations, as well as the different types of texts and the potential ad-
dressee. According to Kemper, these different factors can constrain all the writing
processes (Cf. also Gregg, Sigalas, Hoy, Wisenbacker, & McKinley, 1996, concern-
ing the importance of the addressee).
Kellogg’s comments (1993, 1994) have mainly concerned the allocation of cog-
nitive resources to processes described in Hayes and Flower’s model. Indeed, some
researchers have shown that formulation and execution processes have a non negli-
gible cognitive cost, nevertheless less important than the planning cost (Daneman &
Green, 1986; Kellogg, 1987b; Glynn, Britton, Muth, & Dogan, 1982). Kellogg then
introduces the notion of cognition cycle. Through this cycle, some schemas would
model the writer’s knowledge on reader’s attitudes, beliefs and on the topic. These
schemas would also represent the writer’s rhetorical knowledge and would manage
the exploration of the environment. Finally, they could be modified, by a feedback,
by the information collected in the environment (for a study in children, Cf. Oliver,
1995). In addition, according to Kellogg (1993), the domain knowledge, and also in-
terindividual variables, would be key-factors influencing the manner in which writ-
ers write.
Finally, Brandt (1992), more than a critic, has proposed an adaptation of Hayes
and Flower’s model in a social perspective, so as to explain the role of the social
context and the social structure on individual acts of writing.
Most of these critics, numerous and various, are largely based. Thus, in order to
answer these remarks, Hayes (1996) has proposed the revision of his model.
Hayes’s model (1996). The model elaborated by Hayes (1996) comprises two main
dimensions: the individual and the task environment (Cf. Figure 6). In this way, the
author has named it: ‘an individuo-environmental model’ (Hayes, 1996: 5; Cf. also
Hayes, 1990).
THE TASK ENVIRONMENT
The Social Environment The Physical Environment
The audience The text so far
The composing
Collaborators
medium
THE INDIVIDUAL
MOTIVATION/AFFECT COGNITIVE PROCESSES
Goals Text Interpretation
Predispositions WORKING MEMORY
Phonological Memory Reflection

Beliefs and Attitudes
Visuo/Spatial Sketchpad Text Production
Cost/Benefit estimates
Semantic Memory
LONG-TERM MEMORY
Task schemas
Topic Knowledge
Audience Knowledge
Linguistic Knowledge
Genre Knowledge
Figure 6: The new model of Hayes, adapted from Hayes (1996). Copyright © 1996 by Law-
rence Erlbaum Associates. Adapted with permission.
The individual dimension (The Individual) comprises of the writer’s cognition, af-
fects and memory system. These three concepts work, in the model, by means of
four components:
• The component relative to motivations and affects (Motivation-Affects) com-
prises (a) representations about communicative goals justifying the act of writ-
ing a text (Goals) plus the writer’s attitudes and beliefs (Beliefs and Attitudes),
and (b) parameters concerning the fact that the writer will be engaged in a long-
term task (Predispositions) and the estimation, in terms of profitability, of the
cost of the writing investment (Cost/Benefit Estimates).
• The component relative to cognitive processes of text writing (Cognitive Proc-
esses) is composed of three processes. The ‘Reflection’ process federates a set
of mental activities allowing to transform pieces of knowledge in other pieces

of knowledge. These mental ‘reflective’ activities, certainly play a role in the
elaboration of the content of the text, and can also serve the reasoning, the in-
ference process or the problem-solving activity. The ‘Text Interpretation’ proc-
ess allows the writer to (re)read the already written text and to proceed to a
comprehension activity of the transcribed information. This process allows then
to continue the writing in formulating a coherent content with what has been
previously written or, conversely, to proceed to a revision of the conceptual
(content coherence, for example) and/or linguistic (appropriateness and accu-
racy of linguistic forms) characteristics of the text. The ‘Text Production’ proc-
ess underlies the elaboration of the text. It allows to transcribe the product of the
reflection in a text composed of linguistic information. According to Hayes
(p.13), this process can as well produce written, oral (in the case, for example,
of oral comments on what is written) or handwritten forms (in the case, for ex-
ample, of graphs or figures added in the text).
• The component relative to ‘Long Term Memory’ comprises a set of different
kinds of knowledge that ensure different functions within the writing activity.
These five types of knowledge concern: the type of text (Gender knowledge),
the addressee of the text (Audience knowledge), the linguistic components nec-
essary for the realisation of the text (Linguistic Knowledge), the area of the con-
tent of the text (Topic knowledge), and the procedures to guide and control the
effective realisation of the text production (Task Schemas).
These three components (Motivation/Affect, Cognitive Processes and Long Term
Memory) are linked together (for example, the degree of motivation can directly in-
fluence the more or less important mobilisation of cognitive processes) and also with
the fourth component that is Working Memory.
• Largely inspired by Baddeley’s (1986) model, the ‘Working Memory’ compo-
nent envisaged by Hayes (1996) is composed of three sub-registers respectively
dedicated to the maintaining and to the processing of phonological (Phonologi-
cal Memory), visuospatial (Visuo-Spatial Sketchpad) and (differing from the
original model of Baddeley) semantic (Semantic Memory) representations. This
last register would be more particularly used during translating (formulation) of
the text conceptual content. By its central position in the model, the Working
Memory register constitutes, according to the author, an interface by which are
matched the Cognitive Processes and the knowledge stored in Long Term
Memory, with the respective influence of Motivation and Affects.
The second dimension of the 1996 model is the environmental dimension (The Task
Environment). It takes into account both the writer’s social and physical environ-
ment. It concerns all the external factors that can influence the writer (‘… that lie
outside of the writer’s skin’, Hayes, 1996: 3). The social component (The Social en-
vironment) lists characteristics of the text addressee (The Audience) as well as those
about the writer’s possible partners (Collaborators), as in the case of collaborative
writing. The physical component (The Physical Environment) comprises the effec-
tively produced text (The Text so Far) and the nature of the used medium (The
Composing Medium) to specifically elaborate the written message, by means of

handwriting or typewriting, for example.
From the initial model to the revised model. Hayes (1996) emphasizes four main dif-
ferences to distinguish the initial model from his revision:
• the notion of Working Memory was absent in the 1980 model. Its role during
text production is now considered as central in the new model,
• more specifically, the introduction of the concept of Working Memory takes its
senses through the distinction of the processing of phonological, visuo-spatial,
and semantic representations,
• the writer’s motivation and affects, not much considered in the 1980 model,
find here a place that is as much important as for the 3 other components. Ac-
cording to the author, these two aspects linked to the writer would play a main
role in writing,
• finally, the most marked difference between the two models concerns the reor-
ganisation of the component inherent to cognitive processes. The Reviewing
process has been replaced by the activity of Text Interpretation; the elaboration
of the text plan (ensured in the 1980 model by the Planning process) belongs to
a more general process called Reflection and finally, the Translating process has
been included in an also more general process, Text Production.
These modifications, commented by Hayes (1996), certainly constitute the most
prominent differences between the two models. However some less important dif-
ferences can equally be described. Thus, in the revised model:
• the processes are voluntarily represented in identical sized boxes, to, according
to Hayes, not induce an effect of importance, as in the case of the 1980 model,
which showed a ‘big’ box for the Planning process, as if this process was able
to play a (too) determinant role,
• the writer’s Long Term Memory is differently located, so as to indicate that it
interacts with the three writing cognitive processes, and not only with the Plan-
ning process, as it was supposed in the 1980 model,
• the Monitoring process that appeared, in 1980, as a fourth writing process, with
the Planning, Translating and Reviewing processes, has disappeared in the re-
vised version. Now, the process management and control are ensured by the
Task Schemas that are stored in Long Term Memory, as procedural knowledge.
The 1996 revision mainly consists of a reorganisation of the different processes, also
with a certain number of precisions relative to definitions about writing activities. If
Working Memory is present and its registers specified in the new model, its narrow
relationships with the different processes are nevertheless very vague. Hayes pre-
sents more tracks concerning the role of Working Memory than a real procedural
model. This imprecision is compensated by the appearance of Kellogg’s (1996)
model, that proposes relatively fine theoretical hypotheses about the functioning of
writing processes in relation to Working Memory.
2.4.2 Kellogg’s (1996) model

The objective of Kellogg (1996) is to integrate, in a unique model, writing processes
and a system of information processing (Cf. Figure 7).
Formulation Execution Monitoring

Planning Translating Programming Executing Reading Editing
Visuo-Spatial Central Phonological

Sketchpad Executive Loop
Figure 7: Relationships between production and Working Memory components,adapted from

Kellogg (1996). Copyright © 1996 by Lawrence Erlbaum Associates. Adapted with permis-
sion.
The system is the model of Working Memory such as developed by Baddeley

(1986). The Working Memory architecture is composed of a Central Executive
globally dedicated to complex processes (i.e., reasoning, reflecting, etc.) and two
slave registers, the Visuo-Spatial Sketchpad and the Articulatory Loop, allowing to
maintain visual and phonological representations.
Concerning the architecture of writing processes, Kellogg (1996) resumes the
writing model elaborated by Brown, McDonald, Brown and Carr (1988). He distin-
guishes three components – the Formulation, the Execution and the Monitoring –
each comprising two basis processes.
• ‘Formulation’ is composed of two processes, the Planning and the Translating.
The first process allows to fix goals that have to be reached, to research relevant
domain knowledge units, according to these goals, and to organise the retrieved
information. The second one allows the transformation of ideas into linguistic
structures.
• ‘Execution’ allows the motoric elaboration (Programming) and the realisation
of the message (Executing). The output of the Formulation component is then
programmed before its transcription. The system of handwriting or typing con-
sists of graphically reproducing the message. The programming and the muscu-
lar executing vary as a function of the final mode of emission.
• ‘Monitoring’ implies to read the message and can lead to a possible new editing
of this message. The first process (Reading) allows to regularly reread/verify
the message, during and after its elaboration. The second process, Editing, com-
pares the writer’s intentions with the output of the different systems. All dis-
agreements between the two lead to a feedback, to a component (and its proc-
esses) or to preceding processes. The editing process allows to detect and to di-
agnose problems, then to edit a new version of the message, once the problems
have been solved.
Besides the definition of writing processes and the proposed architecture, the main
characteristic of Kellogg’s model is to very precisely describe the interactions be-
tween writing processes and the different components of Working Memory.
In terms of processing spreading in mnemonic registers, the Planning process
would necessitate the Visuo-Spatial Sketchpad to temporarily store conceptual rep-
resentations and the Central Executive for their processing. The Translating of an
idea or a concept in a correct sentence requires the Articulatory Loop. The phono-
logical representations of words, selected in a syntactic structure, are stored in the
short term buffer of the Phonological Loop. Moreover, the Translating process also
needs resources from the Central Executive, mainly when the writer has to choose
correct words and an appropriated syntactic structure. With respect to the Execution
process, Kellogg postulates that typing or graphic realisation of the message only
necessitates the Central Executive and that these activities are not costly when they
are automatised (Cf., on the Figure 7, the dotted lines between the central executive
and the execution component). The Reading process, that is the basis process of
Monitoring, solicits both the Articulatory Loop and the Central Executive.
Kellogg’s (1996) model is particularly interesting because it precisely locate the dif-
ferent writing processes in each of the Working Memory registers. Its ‘predictive
power’ is relatively important and allows to easily proceed to the elaboration of hy-
potheses and experimental paradigms in order to validate or not these predictions. In
a more general way, this model stresses the constraints linked to the different proc-
esses in writing (as previously developed by Flower and Hayes in 1980) and ques-
tions the dynamics within and through the different memory registers. Nevertheless,
it is still necessary that this theoretical link between writing processes and Working
Memory registers will be fully justified through experimental validations.
3 DIVERSITY OF WRITING MODELS: SOME POINTS OF COMPARISONS

To describe, as we have tried to do, the general models with an historical viewpoint
does not lead to forget that all these models are entirely compatible with current
theories.
They are interested in different but complementary aspects of verbal production,
concerning the definition of writing expertise (Hayes & Flower, 1980; Hayes, 1996),
the development of this expertise (Bereiter and Scardamalia, 1987), the appropriate-
ness of a comparison with the oral production (Levelt, 1989), or the integration of
memory models (for the revision of Hayes’s model and more particularly for Kel-
logg’s new model, both in 1996). This complementarity is often possible with the
adoption of common or at least relatively close principles, in the general idea of
processing and definition of processes.
In fact, Hayes and Flower’s (1980) model, constituting according to Levy (1997) a
notable progress in the area of writing research, represents still today a theoretical
basis that currently influences the other models. The models elaborated by Bereiter
and Scardamalia (1987), Hayes (1996), Kellogg (1996) and Levelt (1989), generally
linked to real time studies and/or verbal protocol analyses, are all built up in the
form of a global architecture and are aimed at describing the course of writing (or
speaking) as a whole. They conceive writing (or speaking) as composed of several
processes with specific relationships. These processes receive and transform infor-
mation, this being subordinated to an instance of control, which regulates the infor-
mation and evaluates the final product. Finally, they generally distinguish at least
three major processes, corresponding to the three main stages of verbal production:
(1) to plan the content, (2) to translate this content into a linguistic trace and (3) to
revise or to correct this content or this trace, as in the case of Hayes and Flower’s
(1980) model. In addition, whichever is the considered model, the activity of these
three processes is always considered in the framework of short term memory
(Flower & Hayes, 1980; Levelt, 1989; Bereiter & Scardamalia, 1987) or in the
framework of Working Memory (Hayes, 1996; Kellogg, 1996). The limitation of the
cognitive capacity supposes to fragment processing. The dynamic of the processes is
often postulated as sequential, with a possible recursion, according to the considered
phase of production. The management of recursion is generally ensured by a special-
ised process that can control the progress of processes or the appropriateness of writ-
ten texts with the writer’s or speaker’s intentions.
Finally, the common points between the main models of verbal production con-
cern relatively global dimensions, as the presence of a control (or monitoring) proc-
ess, the presence of three main production processes, or the consideration of con-
straints linked to the processing limited capacity, etc. Nevertheless, more than these
very general aspects, a fine analysis about descriptions and explanations proposed
by each model show deep and important differences. As underlined by Levelt
(1989), the analysis and modelling of a complex language processing system may
lead to varied segmentations of the processes. However, we think that such varia-
tions are not only terminological: they are often based on diverging theoretical con-
ceptions as regards to the nature and the functioning of operations which compose
the processes. These differences can be evaluated and analysed for each of the main
writing processes, concerning the architecture of these processes as well as the way
in which their functioning is defined.
3.1 Differences in the architecture

Considering their global architecture, the general models of verbal production can a
priori be distinguished by (1) the variety of the definitions proposed for a similar
process, according to their processing operations, and (2) the more or less important
interest allotted to one process to the detriment of the other processes.
3.1.1 The variety of process definitions.

Verbal production models do not inevitably propose, for a similar process, a congru-
ent definition, nor the same division of processing operations composing this proc-
ess. As an example, some characteristic and important differences can be found be-
tween Levelt’s (1989) model and Hayes and Flower’s (1980) model. At a superficial
analysis level, these two models appear relatively close because they both circum-
scribe processes or modules ensuring, on the one hand, the definition of contents
(Conceptualizer and Planning) and, on the other hand, the linguistic translation of
these contents (Formulator and Translating). This proximity is only apparent. In re-
ality, Levelt’s (1989) models do not propose the same sharing of processes and thus,
the same definition of conceptual and linguistic processes such as proposed by
Hayes and Flower (1980). In this last model, the Planning process function is only
and mainly dedicated to the elaboration of a text plan. Conversely, in Levelt’s
(1989) or van Wijk’s (1999) models, the Conceptualizer has not a unique function of
discourse plan elaboration (with the Macroplanning sub-process of the Conceptual-
izer), but can also establish an entire preverbal message (with the Microplanning
sub-process of the Conceptualizer). Then this message will be translated into the
Formulator. Conversely, Hayes and Flower (1980) clearly distinguish the elabora-
tion of the text plan, made by the Planning process, from the elaboration of the pre-
verbal message that is created by the Translating process.
This brief comparison is a good illustration of divergences in general models. In
the case of Hayes and Flower’s (1980) model, the Translating process ensures con-
ceptual processing (i.e., to define the text content on the basis of a plan) as well as
linguistic processing (i.e., to put into words the contents). In the case of Levelt’s
(1989) model, on the contrary, the Conceptualizer is specialised in the elaboration of
conceptual representations while the Formulator carries out linguistic processes.
This first point of divergence between general models can be explained by the
sharing of writing (or speaking) sub-processes (or operations) to define the different
processes. Another point of divergence notices the more or less important interest
that these models respectively give to the description of some particular processes.
3.1.2 Some processes are ignored, other are reduced

Although each of the four above presented models have a appealing features, none
fully explain the totality of the mental activities in writing or discursive tasks. This
fact is justifiable because one model, in focussing more strongly on one given di-
mension of the verbal production activity, can thus only decrease the description of
the other processes then considered as secondary (at least in the framework of this
model). For example, the ‘double’ model elaborated by Bereiter and Scardamalia
(1987), whose main objective is to detail the activity of reorganisation content, do
not clarify the linguistic translation process. They are only very briefly mentioned,
within the Knowledge Telling Process, through the Write operation. Conversely, the
description of the same process is particularly detailed in Levelt’s (1989) model.
This model distinguishes, within the Formulator, two processes ensuring a gram-
matical encoding and a phonological encoding. Moreover, an explanation about the

grammatical encoding is provided by using the notion of Lemmas.
Besides the fact to put more or less in obviousness some of their components,
general models can also entirely neglect some production dimensions. In this case,
Hayes and Flower’s (1980) and Bereiter and Scardamalia’s (1987) models do not
mention the existence of processes for the motor realisation of the text. This process
nevertheless systematically appears in the other models: in Levelt’s (1989) model,
the Articulator executes the phonological plan established by the Formulator or, in
Kellogg’s (1996), the Execution component is composed of two processes, one for
the programmation of the graphic execution (Programming), and another for its re-
alisation (Executing). It is finally interesting to note that the motoric execution proc-
ess is nevertheless present in the model revised by Hayes (1996), that, in addition,
proposes to distinguish the medium for the execution (handwriting or typewriting).
3.1.3 Local models in response to the limits of general models

The principal quality of general models is to provide clear definition and positioning
of writing knowledge and processes into an interactive system. However, these
models can rapidly be limited when, as we have just illustrated through linguistic
and motor processes, they have to account for the writing activity in its finest di-
mensions. One of the theoretical solutions to bring to this relative imprecision of
general models is to use local production models. These models, according to
whether they are interested more particularly or exclusively in a component of the
production activity, can be classified into two categories.
The first category brings together local models whose objective is to clarify one
specific part of a general model whose description is not very precise. For example,
in the case of text production, the local model elaborated by Galbraith (1999, Cf.
Chapter 5), called ‘Knowledge Constituting’, aims to account for mechanisms that
allow elaborating and creating new ideas during writing activity. In this way, this
model can represent a potential reply to the critical point addressed by Kintsch
(1987) to Hayes and Flower’s (1980) model (Cf. page 15).
The second category of local models does not aim to clarify some particular
components of verbal production general models. Its objective is to formalise one
mental mechanism or one specific process, independently of the complex activity in
which this mechanism or this process can be possibly implied. These local models
are finally numerous enough in the framework of verbal production and are suffi-
ciently precise and formalised to be or to have been the object of computer simula-
tions. They can concern as well an activity of isolated oral sentence production (Cf.
the models of Bock, 1996; Bock & Levelt, 1994; Dell, 1986 or Garrett, 1980) as the
installation and the functioning of graphomotoric or orthographic processes in
handwriting (van Galen, 1990).
Resorting to local models can constitute an interesting solution, on a theoretical
plan, to complete general models. For example, taking into account the more spe-
cific models of linguistic transfer (e.g., grammatical and lexical levels; Cf. Bock,
1995, 1996; Garrett, 1982) could indeed appear useful when describing some writ-
ing components. Nevertheless, a simple juxtaposition of models is really not suffi-
cient. Even if a local model belongs to one or another categories of models, it is nec-
essary to be able to specify in what and why this model has to be modified in order
to be included into a global model of verbal production. More specifically, it is
unlikely that the production of a sentence out of any context needs the same proc-
esses or the same modes of processing as the production of a series of connected
sentences, as is required when writing a text. However such a relation between local
and general models can sometimes be operated with a certain success as in the case
of Kellogg’s (1996) that establishes relationships between a general writing model
and a specific Working Memory model.
3.2 Differences of functioning

Beyond their architectural specificity, the way to consider modes of process func-
tioning can also considerably vary from one model to another. These divergences
are the most often due to different or opposite ideas about: (1) the nature and modes
of processing in the cognitive system, (2) the management and control principles of
the processes, or (3) the nature and the role of Working Memory for maintaining and
processing the different representations.
3.2.1 Some processing methods inspired by the general functioning of the cogni-
tive system
The modes of processing adopted by the different authors to account for the func-
tioning of their model often come from more general ideas about the functioning of
the cognitive system. Three dimensions globally determine the way to conceive the
functioning and the architecture of the cognitive system.
• The theoretical basis of a model can be guided by ‘modular’ ideas or, con-
versely, by ‘interactionist’ ideas about processing. In the first case, the process
realised by modules (as defined by Fodor, 1983), is considered as encapsulated
and autonomous and cannot be interrupted. In the second case, interactive and
permeable processes generally ensure processes. In this case, the functioning of
a process can be influenced, modified or interrupted by the functioning of an-
other process.
• The realisation of processing can be considered as ‘parallel’ (several processes
simultaneously operating), ‘sequential’ (processes sequentially applied) or both.
• These different processes can focus on ‘symbolic representations’ (about con-
cepts or knowledge units) or conversely on ‘non symbolic representations’ (as
in connexionist theory).
These three dimensions or these three axes, equally allow to characterise more spe-
cific models of verbal production by determining how the functioning, the modes of
processing or the circulation of processed information can be conceived. Levelt
(1989), for example, adopts a perspective that is largely modularist about speaking
processes. This theoretical choice, justified, according to the author, by the fluidity
and rapidity necessities of oral production, explains the specialisation and the exclu-
siveness of processing realised by the Conceptualizer and by the Formulator. In a
modularist framework, each module, concerned by the transformation of a represen-
tation into one another, autonomously processes one unique type of representation.
These general ideas on the nature and modes of processing have not only reper-
cussions on process or module definitions, but they also play a determining role
concerning the specification of process management and control in production mod-
els.
3.2.2 Different ways to consider process management and control

While all models agree about the necessity to temporarily manage the application of
processes and to control their correct progress, there is no real consensus for the
definition and the mode of functioning of the process ensuring this management and
this control.
In Hayes and Flower’s (1980) model, it is the Monitoring process that, in relation
with the Goal Setting sub-process, manages and controls the process realisation and
thus determines writing strategies the most adapted to the production context and
goals. A Monitoring process is equally found in Kellogg’s (1996) model where it
ensures, as in Hayes and Flower’s (1980) a control and a regulation of the sequence
of writing processes. Nevertheless, although the name and the function of these two
processes are the same, their functioning mode is totally different from one model to
another. In Kellogg’s model, the Monitoring is composed of two processes, Reading
and Editing, whose definition is very close to those provided by Hayes and Flower
(1980) to account for the Reviewing process and not for the Monitoring. It seems
that, for Kellogg, the management and control activity is confused with the revision
activity. This last one, by allowing the rereading and the editing of the already writ-
ten text, would control the activation of writing processes, necessary both for the
continuation of writing and for the modification of the already written text.
Conversely, in Hayes’s (1996) model, the process control and management do
not depend on the revising process that is a part of the Text Interpretation process.
However, contrary to Hayes and Flower’s (1980) model, the procedural knowledge,
labelled Task Schemas and stored in Long Term Memory, control the writing execu-
tion by recording its main stages.
Concerning the Monitoring or Task Schemas, these components share the power
to modify the progress of the writing activity by replacing the realisation of one
process by another. An entirely different way to consider the process management
and the control is developed by Levelt (1989). In his model, the Self-Monitoring
component is ensured by an activity of comprehension and analysis, realised by the
speaker, of her/his own discourse. This comprehension activity, by detecting possi-
ble formulation errors or the inappropriateness between the message and the
speaker’s intentions, do not lead to the interruption of one process, but to the reitera-
tion of the totality of processing, from conceptualisation to execution. The Self-
Monitoring represents therefore less of an authority of process management than a

control of the quality of processed products.
This diversity of process management and control modes can be mainly ex-
plained by the great dimensions chosen by the authors to define the general
functioning of their model. As seen above, the fact that, in Levelt’s (1989) model,
the detection of a speech error leads to the reiteration of a whole process cycle (from
the Conceptualizer to the Articulator) is strongly linked to the globally modularist
idea of this model. Thus, the processing encapsulation postulates that: (1) the
conscious product analysis is only carried out at the end of a complete process cycle,
and (2) the correction of a detected error, because of the automatic and successive
activation of modules, could only be realised by the whole reiteration of the cycle.
Finally, if the management and control authority of the verbal production can be
relatively different from one model to another, this process often depends on the na-
ture and modes of the process retained to describe the general model functioning.
Nevertheless, the diversity of underlying theoretical ideas, and their repercussions
on the way to consider process management and control, do not constitute the
unique sources of differences concerning the functioning of models. The definition
and the role of Working Memory in the realisation of the different processes must
also be discussed.
3.2.3 The role of Working Memory and the limitation of processing capacities in
writing activity
The notions of memory and limited capacities are always evoked in general writing
models. Flower and Hayes (1980) have thoroughly discussed the consequences of
Short Term Memory limitation on the generation of goals and the management of
constraints during writing. In addition, Bereiter and Scardamalia (1987) consider
that an increase of the Short Term Memory span is necessary to simultaneously
maintain a greatest number of knowledge units for the realisation of the Knowledge
Transforming Strategy. The difference between these two first models and the mod-
els elaborated by Hayes and Kellogg in 1996 concerns the evolution from the con-
cept of Short Term Memory to the notion of Working Memory, as defined by
Baddeley (1986). In this framework, the maintaining capacities are considered
within the different slave registers (dedicated to different kinds of representation,
phonological, visual and perhaps semantic), but concentrating on processing of these
representations are also explained in these registers. However, a certain number of
questions remain concerning the role of Working Memory in writing models. A first
question is to query whether the objective, by integrating a model of Working Mem-
ory in the activity of text (or speech) production, is to situate verbal production
processes in the different Working Memory registers (as proposed in Kellogg’s
model, 1996) or to analyse process constraints inherent to the limited capacity dur-
ing the realisation of processes, which represents another hypothesis. A complemen-
tary question is to query whether Baddeley’s model can be considered the most
adapted architecture for the study of writing. Indeed, other models or theories about
the structure and the functioning of Working Memory, as activation models (Ander-
son, 1983b; Cowan, 1993) can be more relevant to account for the on-line distribu-
tion of writing processes. It is thus necessary, concerning the functioning of a model
and its processes, to better define and justify the choice of a memory architecture, by
precisely clarifying writing phenomena that can be explained by such a specific
memory system.
3.3 Between architecture and functioning: the development of writing expertise

We have just quickly considered a certain number of points on which the main gen-
eral models of verbal production could be differentiated, both concerning the archi-
tecture of processes and functioning modes of these processes. In the framework of
the development of writing expertise, these theoretical divergences among models
play a central role because they can largely determine how it is possible to conceive
the access to expertise. The writing activity is indeed sufficiently complex to be
mastered only gradually and late. One of the questions, that is latent in writing mod-
els, is then to query whether the way to expertise is rather linked to: (1) a complexi-
fication of process architecture (appearance of new processes, for example) or, on
the contrary, (2) an intrinsic modification of the functioning mode of some already
existing process (automatisation of some processes, or more strategic use of a set of
processes, for example). In addition, if it is strongly possible that these two phenom-
ena co-occur, what relations do they maintain during the development?
These ideas of expertise development are related to the definition of the expertise
adopted by the authors of models. For example, Bereiter and Scardamalia’s (1987)
model can concern both children (who become experts with maturation and training)
and adults (who can increase their expertise by training; on this point, Cf. also
Flower, Schriver, Carey, Haas, & Hayes, 1989). There are also less generic models
whose objective is more strongly centered on the development of writing capacities
in very young writers. For example, the model developed by Swanson and Berninger
(1996) allows to account, more precisely than in Bereiter and Scardamalia’s (1987)
model, for the installation and the progressive elaboration of writing processes.
Whatever the objectives of the considered developmental model, a crucial problem,
in Bereiter and Scardamalia’s (1987) models as well as in Berninger and Swanson’s
(1994) proposition, concerns the proposed explanations of developmental factors.
When, for example, Bereiter and Scardamalia (1987) consider that the passage from
Knowledge Telling to Knowledge Transforming Strategy is linked to the develop-
ment of planning but also to the increasing of the memory span, the problem is not
entirely solved because it remains to analyse and explain the relations that can exist
between the evolution of the memory functioning and of the writing expertise.
4 THROUGH THE MODELS OF WRITING

Since these last 20 years, even if the area of research on writing has been strongly
structured, it has equally become strongly diversified, as it is the case for all theo-
retical areas in progression (Cf. Coirier, Gaonac’h, & Passerault, 1996; Fayol, 1997).
It can thus be difficult these days to apprehend all the dimensions, all the aspects and
all the elements of modelisations that have been, or that are currently, proposed.
Such a diversification and multiplication of writing models constitute a non negligi-
ble progress at a theoretical level, but, as a consequence, researchers are faced more
and more with the problem of the heterogeneity of these models.
As we have tried to demonstrate through some confrontations between the main
models, their internal architectures often present a low correspondence between the
concepts used and a vagueness of some components. A comparison of these archi-
tectures shows that some processes do not refer to the same operations or neglect
some operations or processes. Moreover, the models have not always been con-
ceived with the same general theoretical ideas, concerning processing modes, proc-
ess management and control or their realisation within Working Memory. More than
these functional and architectural differences, a great number of different concepts
are today introduced in the different text production models. For example, in the
model revised by Hayes (1996), some components about reading and comprehen-
sion are implied in activities concerning production, revision, as well as control. It
thus becomes necessary to theoretically clarify how very complex cognitive activi-
ties, such as reading and comprehension, can be integrated in production models. In
the same way, this reflection can certainly be made to explain the use of local mod-
els of verbal production to explain some components in general models.
Finally, the differences between writing models, and the interrogations that they
cause, are numerous and can be observed at all levels in the sequence of writing
processes. This fact does not allow to systematically confront these models. In addi-
tion, this does not allow critical points to be defined, from which precise hypotheses
could be raised and tested. It is the reason why the main goal of this book is to trig-
ger off an accurate comparison of current models. More precisely, our objective is to
offer an assessment (of course partial) about the main writing models by analysing
their similarities, their differences, their interests and especially their limits. We also
think that this analysis must allow considering, discussing and surrounding some
questions or hypotheses about possible research paradigms or about some unclear
facts. Of course, it is not possible to deal at the same time with all the above-
mentioned issues. Thus, we will only try to define some critical points: critical be-
cause they lead us to confront the different models on precise topics and also to take
into account topics where it would be possible to devise experimental paradigms,
leading to the gathering of relevant data and, at least we hope, to contribute, even
modestly to the progress from writing theories to computer implementation. These
different confrontations and comparisons of the models are carried out through three
main principles:
• Hayes and Flower’s (1980) model, due to its ‘initiating role’, will serve as the
main reference. All the other models will be compared to it, through their com-
mon points, their evolution and their diverging points, with regard both to ter-
minology and description. From this perspective, this book comprises of two
parts. In the first part, devoted to the architecture of processes in writing mod-
els, we will present and discuss, through different models, the varied definitions
of processes required for Planning (Chapter 1), Translating (Chapter 2) and Re-
vising (Chapter 3). In the second part, devoted to processing modalities and ex-
pertise development in writing models, we will discuss the way in which differ-
ent models conceive the management and the monitoring of these three proc-
esses (Chapter 4). This will lead to a closer look at the decisive role of Working
Memory (Chapter 5). Finally, we will try to look at the various formalisations
proposed to explain the development of writing expertise, in relation to the evo-
lution and the inter-relationships of process components as well as new concep-
tions of Working Memory (Chapter 6).
• While authors of experimental works often use the broadest definitions of proc-
esses (Planning, Translating, Revising, Monitoring), we propose, in the context
of this book, to examine the suggested architectures with a closest attention, by
describing the sub-processes as well as the precise operations they enclose. We
will also identify the contradictions occurring between one model and another,
and thus refine the issues raised by their comparisons.
• Finally, it appears central for us, and considering the main objective of this
book, not to discuss the different writing models and their evolution without re-
questing the authors of these main models to give their points of view about our
analysis. It is in this spirit that Ronald T. Kellogg and John R. Hayes have each
written specific comments that allows us to conclude this book.
Roll up, the textual mystery tour

Is hoping to take you away
(adapted from Lennon and McCartney, 1967)
AUTHORS' NOTE
We would like to thank Ronald Kellogg and John Hayes to have accepted to write
the concluding parts of this book. We also wish to sincerely express thanks to Eric
Espéret and Gert Rijlaarsdam for their very careful and critical reading and for their
many helpful and constructive comments during all the editing process of the book.
PART 1: ARCHITECTURE OF PROCESSES
IN WRITING MODELS
CHAPTER 1
PLANNING PROCESS
1 INTRODUCTION: PLANNING IN WRITING

When they have to write a text, writers have generally ‘to think’, ‘to anticipate’, ‘to
foresee’ what they will write. This anticipation can involve the content (what has to
be told) as well as the linguistic form (how to tell it) of the text. It is besides com-
mon to think that the longer and/or more intensive this reflection activity will be, the
easier to write and the higher the quality of the text will be. Consequently, this text
will not need any important modifications or corrections afterwards.
4.1 Planning: a general cognitive activity involved in text writing

‘Thinking before acting’ is not specific for text writing. Classically defined by
Miller, Galanter and Pribram (1960) as ‘Planning’, this activity is a very general
process that underlies every human task oriented by a goal, and which allows some-
one to differ in time every complex task, that cannot be simultaneously achieved be-
cause of human processing limited capacities. Planning can thus be considered as a
high level cognitive skill or as a mental activity driven by memory (Cf. Akyürek,
1992: 103). Thus, planning does not simply consist of selecting plans or information
stored in Long Term Memory but this activity can be more precisely defined as con-
taining deliberate start up strategic procedures and practical goals. The planning
process, conceived as ‘the schematisation of a complex situation that has to be
solved’ (Hoc, 1987: 37, our translation), appears to be the most effective strategy to
process different complex cognitive tasks and, in this way, it is involved in most
cognitive activities.
In the case of verbal production, the Planning process is conceived as a ‘main de-
terminer’ (Scardamalia and Bereiter, 1985). This process is costly in cognitive re-
sources (Bock, 1982; Kellogg, 1996) and could take, in a specific situation, two
thirds of the total production time (Gould, 1980). Within text writing activity, this
type of processing can be found within each writing level, from the elaboration of a
text plan to the elaboration of a graphic transcription plan. But more specifically,
different kinds of planning can be distinguished (Barritt and Kroll, 1978; Hayes and
Nash, 1996; Kaufer, Hayes and Flower, 1986). Hence, an important difference has
to be made between planning of processes (that would consist of ordering the execu-
tion of different writing processes in a strategic way), and planning of contents (that
would consist of delimiting, in organising and in adapting – according to the audi-
ence – the plan of the text – Cf. Piolat, 1987).
D. Alamargot & L. Chanquoy (2001). Planning process. In G. Rijlaarsdam (Series ed. ) & D.
Alamargot & L. Chanquoy, Studies in Writing: vol 9. Through the Models of Writing, 33– 65.
© 2001 Kluwer Academic Publishers. Printed in the Netherlands.
4.2 Different planning activities in text writing

Flower and Hayes (1980, 1981b) have distinguished different kinds of planning ac-
tivities by describing three types of processing that would underlie the elaboration of
plans of different natures, dedicated to different activities (Cf. Figure 8). Thus, for
the authors, there are three plans: a Plan To Do, a Plan To Say, and a Plan To Com-
pose.
Plan to Do
(rhetorical plan)
Plan to Say Plan to Compose

(product based) (process based)
Generate Produce
Knowledge Text
Figure 8: Different plans involved in writing activity, adapted from Flower and Hayes (1980).
The Plan To Do, that could also be called ‘Rhetorical Plan’, permits the manage-
ment of pragmatic constraints during the writing activity. It includes writer’s and
reader’s characteristics, as well as the general topic of the text. The Plan To Say is a
content plan (or a ‘Declarative plan’) that represents a summarised and simplified
version, or an abstraction, of the whole set of information that will be translated in
the text. The Plan To Say can be considered as the plan of the text. The Plan To
Compose is not a content plan but rather a process plan (Cf. Espéret, 1989, for a dis-
cussion the differences between process and content plan). It determines the order in
which the different writing processes and sub-processes will appear (as a ‘Proce-
dural plan’). This plan enables the transformation of domain knowledge (i.e., for the
authors, ‘Generate Knowledge’: generating and organising) as well as the linguistic
translation of the semantic content (‘Produce Text’).
Inspired by Flower and Hayes’ (1980) conception, Hayes and Nash (1996: 43-45)
have recently improved such a categorisation of the different planning types in-
volved in writing activity, by differentiating the nature of representations concerned
by each kind of planning (for instance, content, non content or text representations).
According to Hayes and Nash, planning in text writing could be decomposed into a
hierarchical set, regrouping different types of ’sub-planning’ activities (Cf. Figure
9). To describe planning in writing firstly requires distinguishing the planning of
processing and writing main goals (Process Planning) from the textual and linguistic
planning (Text Planning). While the first one is focused on the writer and the strat-
egy to comprehend and execute the task, the second one is centred on what is being
PLANNING PROCESS 35
written, that is to say the text content, its form and its awaiting impact upon the ad-
dressee.
Furthermore, according to Hayes and Nash (1996), the Text Planning is shared
into an Abstract Planning and a Language Planning. The first one generates ideas,
without specifying the language to be used. The second one is involved in the pro-
duction of a grammatically and syntactically correct text. This last type of planning
generally corresponds, in writing models, to the formulating stage (Cf. Hayes &
Flower, 1980; and Chapter 2). Finally, the Abstract Planning can be divided into two
parts. The Planning allows solving rhetorical problems – the evaluation of the ap-
propriateness between the text, the addressee and the writer’s goal. The Content
Planning generates a simplified version of information to be expressed. These two
kinds of Abstract Planning could respectively correspond to the ‘Planning to Do’
and the ‘Planning to Say’ sub-processes as defined by Flower and Hayes (1981b).
Planning Text
in Writing Production
Process Text
Planning Planning
Abstract
Planning
Process Non-content Content Language Text

Planning Planning Planning Planning Production
Figure 9: Taxonomy of different types of writing planning, adapted from Hayes and
Nash (1996). Copyright © 1996 by Lawrence Erlbaum Associates. Adapted with permission.
These various kinds of planning can equally be characterised by the nature of the re-
lated mental operations. Hayes and Nash (1996) distinguish three kinds of process-
ing involved in planning: ‘Planning by Abstraction’, ‘Planning by Analogy’ and
‘Planning by Modelisation’. Planning by Abstraction, consisting of manipulating
and ordering abstract concepts, would be more particularly involved in the case of
Content Planning – drafting an abstract content of the future text, independently of
the translation modalities of this content. Planning by Analogy allows generalising
knowledge involved in a specific activity, resulting in a similar or a close activity.
This kind of process would be used, in writing activity, each time that a text schema
or a previous text plan can be adapted and used to write a new text. Hence, the ge-
neric application of a narrative schema, in the case of narrative text writing, can be a
good example of Planning by Analogy, which is equally involved in content
planning and economically replaces the Planning by Abstraction. Planning by Mod-
elling can be distinguished from Planning by Abstraction in the sense that the first
one does not focus on gists or abstract ideas – like plan text content – but takes into
account the entire set of information needed to execute a given task. According to
Hayes and Nash (1996), this mode of planning could occur during the linguistic
formulation stage, when a sentence or a clause can be entirely and mentally planned
before its graphomotoric execution.
4.3 Planning to elaborate the plan of the text

The various definitions presented above seem to be a good demonstration of the fact
that planning, in writing activity, is a very complex notion and rests on various men-
tal operations, bearing on various representations. However, in general writing mod-
els, the concept of planning is mainly used to account for a ‘preparatory thought’
(Hayes and Nash, 1996), allowing the elaboration of a text plan (Hayes and Flower
1980; Hayes and Flower, 1986; Hayes and Nash, 1996). The plan, elaborated during
planning processing, globally depends on:
• both topic and objective of the production (i.e., to write a text to distract, to in-
form, to explain...),
• the addressee (i.e., to write a text for a known or unknown reader, for a friend,
for a professor, etc.; Cf. Caccamise, 1987; Ransdell & Levy, 1994; Rubin &
Piché, 1979),
• the text type (i.e., to write a text in order to explain, to describe, to argue, etc.;
Cf. Cooper & Matsuhashi, 1983; Martlew, 1983) and,
• knowledge available in Long Term Memory, as a function of the expertise level
concerning the domain knowledge of the text (Cf. Kellogg, 1987a; Martlew,
1983; Voss, Vesonder, & Spilich, 1980).
In the initial model of Hayes and Flower (1980), this process, simply labelled ‘Plan-
ning’ is introduced as one of the three main processes in writing activity (with
Translating and Reviewing processes). According to Hayes and Nash’s (1996) tax-
onomy, this planning process, dedicated to the content, could be ensured by a mode
of Planning by Abstraction. Planning process, as described by Hayes and Flower
(1980), is central in the organising of text content and in the processing of content
coherence. More precisely, according to the authors, the main role of Planning proc-
ess is to establish a writing plan by using (1) domain knowledge retrieved from
Long Term Memory and (2) information available in the Task Environment. If this
writing plan has not been previously stored in the writer’s memory (Stored Writing
Plans and, for Hayes, 1996, Planning by Analogy), it must be entirely produced by
means of three sub-processes: (1) Generating knowledge from Long Term Memory,
(2) Organising this knowledge into a writing plan, and (3) Goal Setting – establish-
PLANNING PROCESS 37
ing criteria to test the appropriateness between the written text and the communica-
tive goals.
The functioning of such a Content Planning process has been differently con-
ceived and formalised in the main models of verbal production. Here, the description
of content planning is globally guided by the organisation of the Planning process as
defined in Hayes and Flower’s (1980) model. Thus, we will first detail the function-
ing of Generating and Organising sub-processes, before considering, through Goal
Setting sub-process, the nature and the role of pragmatic processes during planning.
5 THE GENERATING SUB-PROCESS

5.1 Definition of the Generating sub-process
The sub-process labelled ‘Generating’ by Hayes and Flower (1980), consists of re-
trieving domain knowledge units from Long Term Memory and organising them
into a writing plan (Cf. General Introduction). A memory probe (according to the
classical definition of Raaijmakers & Shiffrin, 1981), initially elaborated on the ba-
sis of a writing instruction (defining the topic, the type of text and the audience),
would retrieve a first knowledge unit, serving in turn as a probe to retrieve the fol-
lowing unit, and so on. The initial probe would trigger off a ‘chain reaction’, ending
in a final product comparable to an associative chain of units (Cf. Figure 10).
During this sub-process of domain knowledge retrieving, a processing operation,
called ‘Evaluate Retrieved Element’, may interrupt this associative chain if contents
retrieved are not suitable regarding, for example, the theme of text. In this case, if
the writer’s objective is to generate contents, some new memory probes are elabo-
rated instead of the inappropriate probes. Finally, it is interesting to notice that the
Generating sub-process is not only dedicated to the mental content recovery but also
involves an operation named ‘Consider Notes’ that evaluates the possibility to lin-
guistically translate the knowledge unit via an operation labelled ‘Write Note’.
If we now consider this last point, the functioning of the Generating sub-process,
as conceived by Hayes and Flower (1980), is relatively similar to that described by
Bereiter and Scardamalia (1987), in the framework of the Knowledge Telling Strat-
egy generally adopted by younger and/or less skilled writers (Cf. General Introduc-
tion, page 7).
According to Bereiter and Scardamalia (1987), the content generation process
within the Knowledge Telling strategy is ensured by an iterative set (like a routine)
composed by three operations (Cf. Figure 2, page 7) called ‘Construct Memory
Probe’, ‘Retrieve content from memory using probes’ and ‘Run test of Appropriate-
ness’. Globally, these operations, under the control of the text theme and genre, re-
spectively underlie: (1) the elaboration of memory probes, (2) the effective retriev-
ing of domain knowledge units, before (3) the relevant evaluation of these units.
As in Hayes and Flower’s (1980) model, a first knowledge unit would be ini-
tially retrieved, via a memory probe, elaborated on the basis of minimal cues consti-
tuted by the text theme and genre. This first retrieved unit would contribute to the
elaboration of new memory probes allowing the retrieving of another unit, strongly
connected to the previous one. This chained retrieval, more precisely defined than in
Hayes and Flower’s (1980) model, would obey, according to Bereiter and Scar-
damalia (1987), to the complex principles of spreading activation in a knowledge
system (Cf. Anderson, 1983b and Chapter 5). The retrieval of a unit is completed
according to a gradual principle depending both on the availability in Long Term
Memory of this unit (activation potential principle) and on the degree of its relation-
ship with the unit that activates it (spreading activation principle). Furthermore, ac-
cording to these theoretical principles, one knowledge unit does not only activate
another connected unit but a set of the most strongly connected units.
Retrieve using Replace Current
Current Memory Memory Probe
Probe with New Probe
Succeed Fail
Retrieved Element Yes

> Current Memory Probe
Not
Evaluate Useful Goal =
Retrieved Generate ?
Element
No
Useful
Exit
Consider Note
Yes
Write No
Note
Yes Goal =
Generate ?
No
Exit
Figure 10: Structure of Generating sub-process,adapted from Hayes and Flower (1980).
5.2 Different conceptions of content generation

While the majority of theories or models agree that generating knowledge is first ini-
tiated by the interpretation of writing instructions defining the main textual topic,
there are diverging opinions as to the functioning nature of this process.
The first conception supposes that contents generated during writing are essen-
tially stored in Long Term Memory (i.e., domain knowledge), where they have to be
PLANNING PROCESS 39
recovered. This is the point of view adopted by Hayes and Flower (1980), van Dijk
and Kintsch (1983), or Bereiter and Scardamalia (1987, in the framework of their
Knowledge Telling Strategy). The question underlying this point of view is then to
determine whether the intrinsic functioning of this recovery is automatic (for exam-
ple: associative recovery, spreading activation) or controlled (for example: strategic
recovery, recovery with the help of stored plans).
The second conception proposes that generated contents would be less retrieved
from Long Term Memory than created during writing (creativity). That is, for ex-
ample, the point of view developed by Bereiter and Scardamalia (1987) in the
framework of their Knowledge Transforming Strategy, by Galbraith (1999) within
the Knowledge Constituting model, or by Hayes (1996) in the revised version of
Hayes and Flower’s model. It is also in this perspective that the different works on
the text epistemic effect can be found (Cf. Eigler, Jechle, Merziger, & Winter,
1991), that account for the domain knowledge increase after this knowledge has
been used in a verbal production task.
5.2.1 Generating as a recovery sub-process: automatic or controlled?
Automatic retrieval and a posteriori assessment of contents. When the Generating

sub-process is considered as a ‘recovery from Long Term Memory’ sub-process, its
functioning is frequently viewed as automatic; that is to say, as seen above, by asso-
ciation (according to Hayes and Flower, 1980), or by spreading activation (accord-
ing to Bereiter and Scardamalia, 1987).
It is important to consider that the automatic nature of the generating sub-process
assumes that the relevance of retrieved ideas is evaluated just after their effective re-
covery. Indeed, as the nature of the various knowledge units recovered would only
depend on their proximity within the knowledge network, nothing could a priori at-
test the appropriateness of these idea units to achieve the composition goal. This
evaluation is carried out by the ‘Run Tests of Appropriateness’ operation in the
Knowledge Telling Strategy elaborated by Bereiter and Scardamalia (1987),
whereas Hayes and Flower (1980) postulate the existence of two evaluation opera-
tions. The first one, called ‘Evaluate Retrieved Element’, is, as seen previously, a
constituent of the Generating sub-process (Cf. Figure 10, page 38). The second one,
labelled ‘Evaluate Usefulness of Topic’, arises from one of the first steps of the Or-
ganising sub-process, whose function is to reorganise, adapt or modify the contents
previously retrieved by the Generating sub-process (Cf. Figure 11, page 49).
Such an a posteriori evaluation of appropriateness between retrieved contents
and writing goals represents one of the main characteristics of an automatic recover-
ing conception. To try to describe the automatic functioning of the generating sub-
process, Torrance, Thomas and Robinson (1996) have analysed the ideas retrieved
during the initial prewriting phase of an essay composition, and during the composi-
tion phase itself. They found that the most part of the idea units would only be acti-
vated by the bias of memory probes elaborated from writing assignments or from
previously recovered units, during composition. For them, this characteristic, attest-
ing an automatic retrieving, would be confirmed, conversely, by the fact that con-
trolled memory probes from the text plan, or more precisely, from rhetorical con-
trolled processes, lead to the generation of very few ideas during writing, compared
to the number of idea units recovered by an automatic processing. These results may
seem to confirm a largely automatic functioning of the Generating sub-process.
However, Torrance et al. (1996) notice that such an automaticity could be stronger
when the text topic is usual and familiar for the writer. So, as they highlight, the
functioning of the retrieving sub-process can, however, be conceived as automatic or
as controlled, according to the type of writing task proposed.
Controlled retrieval and a priori assessment of contents: example of van Dijk and
Kintsch’s model. Within Van Dijk and Kintsch’s (1983: 300) model of text produc-
tion, for example, knowledge retrieval is not simply considered as an automatic op-
eration, but is conceived as a complex process whose ‘extracting’ operations, highly
automatic, would be applied under the control of an a priori (and no more a posteri-
ori) highly strategic and controlled structure. Strongly inspired by their own com-
prehension model (with macro- and micro-structural levels of processing), van Dijk
and Kintsch (1983) have proposed an explanation of the general writing process
functioning in terms of strategies depending on different processing levels.
Generally speaking, text writing would suppose the initial establishment of a
‘Macro-planning Strategy’ that carries out the production of speech acts and which
needs some pragmatic knowledge. This knowledge allows the elaboration of the text
communicative goals (1) by defining intention (to describe, to convince, to give in-
formation to the reader, etc.), and (2) by determining how (in terms of domain
knowledge and linguistic knowledge) to achieve this intention (Cf. also, on this last
point: Schmidt, 1979; Schneuwly, 1986). At the end of this Macro-planning Strat-
egy, would begin, according to van Dijk and Kintsch (1983), the first stage of do-
main knowledge transformation. This transformation would be ensured by a ‘Mac-
rostructural Planning Strategy’ whose product would be very similar to the result of
the Planning process defined by Hayes and Flower (1980). This product would be a
writing plan defining and ordering the main text ideas (Macrostructural units). Then,
in another instance, a ‘Microstructural Planning Strategy’ would develop this plan.
This strategy thus represents the first step of the text content propositional elabora-
tion and would extend each Macrostructural Unit by the recovering of Microstruc-
tural Units respectively associated. Finally, the linguistic translation of Microstruc-
tural Units could be achieved by a last strategy that would carry out the local seman-
tic coherence of sentences and the determination of syntactic and lexical forms (i.e.,
Local Coherence Strategies).
According to this model, the retrieving of knowledge units is considered to be in-
trinsically selective. This selectiveness can, for the authors, be supported by a rhe-
torical and pragmatic structure, called ‘Macro-plan’ and elaborated through the
Macroplanning Strategy (Cf. page 56 for a precise description of this strategy). The
different sub-goals of this structure guide both the retrieval of knowledge units and
PLANNING PROCESS 41
the subsequent drafting of the text macro- and microstructure. This concept of re-
trieval control by a knowledge structure (here, pragmatic) is relatively close, in its
main principles, to the theories developed by Reiser and Black (1982; Cf. also
Reiser, Black, & Abelson, 1985), assuming that knowledge recovery in Long Term
memory is controlled by schemas, scripts, or frames.
Automatic and controlled retrieval. Such a controlled recovering principle has

equally been considered by Bereiter and Scardamalia (1987) in their theory on
Knowledge Transforming Strategy (Cf. General Introduction, Figure 3, page 8).
However, in this strategy, the activity of the retrieving process seems to occur by an
increase and a diversification of memory probes. Mainly linked to the text topic and
type in Knowledge Telling Strategy, these probes seem to be produced on the basis
of writing assignments, text type, previously retrieved domain knowledge and
mostly on the writer’s pragmatic and rhetorical knowledge structures. When observ-
ing the interaction between content and rhetorical knowledge (via the Content Prob-
lem Space and the Rhetorical Problem Space), Scardamalia and Bereiter consider
the domain (or content) knowledge retrieval as a process both automatic (in terms of
domain aspects by spreading activation which would be ensured by the ‘Search’ op-
eration: Cf. Bereiter, Burtis, & Scardamalia, 1988) and controlled (in terms of rhe-
torical aspects by the activity of the ‘Delimit’ operation: Cf. equally Bereiter et al.,
1988).
This conception about retrieval seems relatively close to the idea supported by
Walker and Kintsch (1985). These authors hypothesise that all the components of a
schema or a plan (i.e., the different sub-goals), by determining the elaboration of
multiple memory probes, could manage and control the implementation of the re-
trieval process, which is intrinsically automatic.
In fact, it seems that an examination of knowledge retrieval principles in writing

cannot be carried out independently of an analysis of the broader theories of the re-
covery functioning developed in the more general area of cognitive psychology (Cf.
General Introduction). These theories rely on different modes of retrieving that can
be either controlled (‘compound-cue’ for Reiser and Black, 1982; Reiser, Black and
Abelson, 1985), automatic (‘spreading-activation’ for Anderson, 1983b; Collins and
Loftus, 1975), or mixed up (Walker & Kintsch, 1985; Cf. also Ratcliff & McKoon,
1994 or Baudet, 1988, for a discussion on the different concepts of knowledge re-
trieval in cognitive psychology).
5.2.2 Generating as a creative sub-process: what are the mechanisms?

There is no doubt, as seen above, that writing activity inevitably needs domain
knowledge stored in Long Term Memory. However, it is frequent that the writer do
not possess all the necessary domain knowledge, and, in this case, s/he needs to cre-
ate or to elaborate new knowledge units during the progress of writing activity. For
instance, Eigler, Jechle, Merziger and Winter (1991) have shown that writing activ-
ity has a ‘feedback’ effect on the initial domain knowledge (the extent of this
knowledge was evaluated by a questionnaire given before and after the writing pe-
riod). This effect, described as epistemic, is not specifically inherent to writing. It
has equally been described by these authors in the case of speaking. In the same
way, by comparing the number of ideas listed by a writer before and after writing a
text, Galbraith (1992, 1999) has shown that text writing leads to an increase of
knowledge stored in Long Term Memory after text writing activity. Thus this
(re)effect of verbal production would not be restricted to restructured knowledge,
but is also an increase of knowledge (the domain knowledge being enhanced in the
process).
These two examples show that the content generating sub-process can fulfil a
double function: on the one hand it helps to retrieve ideas and, on the other hand, it
allows creating some new pieces of knowledge. However, considering writing activ-
ity as one of the possibilities to acquire new knowledge units requires the identifica-
tion of the nature of the creativity operations, responsible for such an epistemic ef-
fect. Several non-contradictory hypotheses can be considered through the models
elaborated by Bereiter and Scardamalia (1987), Galbraith (1999) and Hayes (1996).
The role of pragmatic and rhetorical constraints in creativity: Bereiter and Scar-
damalia’s (1987) model. The first hypothesis, to explain creativity during writing,
comes from the Knowledge Transforming conception of Bereiter and Scardamalia
(1987), and more precisely, from Scardamalia and Bereiter’s (1991) theory. For
these authors, it would be less the generating activity that would be responsible for
knowledge creativity than the necessity to transform content knowledge because of
rhetorical and pragmatic constraints. In other words, some contents, activated in the
Content Space, can be modified or created under the influence of the Rhetorical
Space.
More precisely, in the Knowledge Transforming Strategy, the content creation,
during writing, would mainly come from the necessity of ‘reworking’ the text con-
tent in order to fulfil pragmatic constraints. This work on the content is achieved not
only under the influence of the Rhetorical Space but also from the Feedback of the
already written text trace to the Problem Space (Cf. General Introduction and Figure
3). Thus, according to Scardamalia and Bereiter (1991), the creation of new contents
via the Knowledge Transforming Strategy would be the result of an analysis, per-
formed by the writer, regarding the appropriateness between the text (via the Feed-
back) and the communicative intentions (via the Rhetorical space). Consequently, in
this perspective, if it is necessary to write (or to speak) in order to create new con-
tents, the rhetorical and pragmatic constraints carry out this creation. Nevertheless,
Galbraith’s (1999) model proposes on this last point a different explanation.
The role of linguistic constraints in creativity: Galbraith’s (1999) model. Another

explanation about creativity during writing comes from Galbraith (1999). He as-
sumes that linguistic processes, applied to activated content units, carry out the gen-
eration of new knowledge units. Hence, the translation of knowledge units in lan-
PLANNING PROCESS 43
guage would lead to the discovery or the creation of associated or ‘associable’ units,
that can be more easily translated into words. Galbraith (1999) has formalised this
mechanism in a writing model called ‘Knowledge Constituting’ which both de-
scribes the linguistic translation of ideas, contained in a sub-symbolic network (con-
nectionist network) and the nature of the feedback of linguistic processes on this
network (Cf. Chapter 5 for a detailed description of this model). According to this
model, the content elaboration would not be linked to pragmatic constraints or crite-
ria, as in Bereiter and Scardamalia’s (1987) model, but rather to linguistic con-
straints. ‘Translating ideas freely into words’ would lead to an evolution and/or a
modification of the nature of these ideas. Nevertheless, Galbraith (1999) gives few
details and little precise explanation on this mechanism and some questions remain
(Cf. Alamargot, Favart, & Galbraith, 2000). The simple fact to formulate knowledge
units into a linguistic code could ensure a deeper processing of representation. Is this
deeper processing responsible for the knowledge modification? If the translating
process plays an epistemic role, what is more precisely the involved process or op-
eration? Is it semantic processing, verbal translation, syntactic integration of various
ideas in the same sentence, or lexical knowledge associated to words? Does the ex-
pression of domain knowledge by other codes (drawing, schema or gestures, for ex-
ample) play the same epistemic role as a linguistic code?
The role of reasoning and comprehension activity in creativity: Hayes’s (1996)

model. Taking into account such a necessity or opportunity to create new knowledge
domain appears to lack in Hayes and Flower’s (1980) model. As considered in the
General Introduction, this weakness has been noted by Kintsch (1987), and more re-
cently by Hayes (1996). In his new writing model, it is possible to consider the crea-
tion of new contents according to two modalities: via the Reflection process or via
the Text Interpretation process.
By modifying Hayes and Flower’s (1980) model, Hayes (1996) has enclosed
Planning processes (and thus certainly the Generating sub-process) into a larger
‘Problem-solving’ process, itself included in a more general process labelled ‘Re-
flection’, whose function is to translate internal representations into other internal
representations. To ensure such a representation transformation, two processes, ‘De-
cision Making’ and ‘Inferencing’, complete the activity of the Problem-solving
process inside the more general Reflection process. If the Decision Making process
underlies the revising activity, the Inferencing process allows the writer to elaborate
a new representation from a former one. For Hayes (1996: 21), this process would
permit some extension of text content according to available knowledge units and to
the writer’s hypotheses about the addressee’s domain knowledge.
To complete the Reflection process, Hayes (1996: 13) proposes to include,
within the main writing activity, a Text Interpretation process. This process allows
the writer to create internal representations on the basis of graphical and linguistic
representations obtained from reading, listening and graphical revising activities.
Thus, by applying these processes to her/his own speaking or writing activity, the
writer could modify her/his own representations, after their linguistic translation.
Rereading, understanding, modifying the already written text could contribute creat-
ing new representations, as for a reading-comprehension activity.
5.3 Conclusion about the Generating sub-process

The definition of the Generating sub-process is far from being unique and its modes
of processing seem to allow content retrieval as well as some content creation during
the writing progress. Hayes and Flower’s (1980) model and Bereiter and Scar-
damalia’s (1987) Knowledge Telling Strategy put strongly (perhaps exclusively) in
obviousness the recovery function of this sub-process. In this framework, one of the
central questions is to query whether the functioning mode of this retrieving sub-
process is automatic (by association or spreading activation with the possibility of an
a posteriori control of the appropriateness of the retrieved information), or con-
trolled (with the help of plans enabling an a priori selection of retrieved contents).
However, in the Knowledge Transforming Strategy, this phenomenon is more com-
plex. Bereiter and Scardamalia (1987) postulate that content retrieval depends on
two functioning modes, one controlled and one automatic. They also consider that
the recovered contents can be modified to constitute new pieces of domain knowl-
edge. More generally, the creation of new contents during writing would depend on
several factors, concerning pragmatic constraints (in the case of Knowledge Trans-
forming Strategy of Scardamalia and Bereiter, 1991), linguistic constraints (for
Galbraith’s Knowledge Constituting model, 1999) or Reflection and Comprehension
activities (in Hayes’s model, 1996).
We think that two other factors influence the functioning of the Generating sub-
process, both for the retrieval and the creation of contents. These factors are (1) the
text type that has to be written, and (2) the initial domain knowledge that enables
constituting the text content.
5.3.1 Retrieval versus creation of contents according to the different types of text
It is possible to hypothesise, with Scardamalia and Bereiter (1991), that some types
of text could better ensure a high creativity, or, conversely, more information re-
trieval from Long Term Memory. Thus, the rhetorical constraints can be various ac-
cording to the type of writing to produce and, in this way, can give rise to various
kinds of domain knowledge transformations.
For example, according to Scardamalia and Bereiter (1985, 1991), the necessity
for the writer to strongly argue in the case of an essay (by using natural, sometimes
formal, thinking operations) can lead to the discovery of useful but initially non-
activated concepts. Comparatively, the constraints inherent to summary elaboration
can lead the writer to operate evaluative and critical analyses of her/his domain
knowledge. This effect can also be established during the writing of some transitions
between different sub-themes, which, by increasing the global text coherence, also
ensure the discovery of new conceptual links between domain knowledge units.
Conversely, the content of some procedural or descriptive texts has to be strongly
PLANNING PROCESS 45
delimited and circumscribed (when the writer needs to describe or explain a machine
functioning, for example). In this case, it is then possible to think that the use of
knowledge retrieved from Long Term Memory or of information from the Task En-
vironment, would be more essential than the use of creative processing. Finally, in
the case of narrative texts, two strategies could be opposed. It is possible to directly
retrieve a structure (as the narrative schema: Cf. Fayol, 1985) that needs to be devel-
oped, either by recovering knowledge units associated with schema slots (for well-
known or classical stories), or by creating content able to be integrated within the
various slots of the canonical schema.
In conclusion, approaching idea retrieval and creation in terms of different bal-
ances between these two activities and during the production of different types of
text, seems to be a very interesting and probably relevant issue to better surround the
functioning of the Generating sub-process. It is nevertheless necessary to precisely
define, on a methodological perspective, how it is possible to evaluate and to meas-
ure how the writer creates new contents.
5.3.2 The influence of domain knowledge on the Generating sub-process

Whatever the conception of Generating (recovery and/or creativity), the analysis of
the functioning of this sub-process in text writing activity cannot truly be pursued
without a prior analysis of the characteristics of the product to be generated. In this
perspective, it appears relevant to associate the study of the Generating sub-process
with a precise analysis of the structure of the domain knowledge. Indeed, when the
functioning of Generating is envisaged, for example, in the framework of the spread-
ing activation theory, it becomes obvious that: (1) the availability of knowledge
units (i.e., their respective potential of activation) in a semantic or conceptual net-
work, (2) the power of the links between these knowledge units, and (3) the more or
less hierarchical network architecture, can be considered as influent parameters for
the recovery of contents from Long Term Memory during writing. However, it is in-
teresting to remark that the consideration of the domain knowledge nature and struc-
ture is often neglected in the main models of verbal production.
The notion of Domain knowledge. The authors (for example, Hayes and Flower,
1980 or Levelt, 1989) often mention the presence, in Long Term Memory, of con-
ceptual systems, as ‘mental models’, ‘scripts’, ‘schemas’, ‘knowledge about the ad-
dressee’, ‘type of text’, and more recently with Hayes (1996), a set of knowledge
both declarative (Knowledge of Topic, of Audience, of Linguistic and of Genre) and
procedural (Task schemas). However, these researchers do not specify the perti-
nence, neither the structure nor the influence of these kinds of structure on writing
processes. In addition, in the more general area of cognitive psychology, the defini-
tion of the organisation in Long Term Memory is far from being homogeneous and
different formalisations can compete (as, for example, semantic networks, mental
models, a-linguistic conceptual storage, etc.) and bring researchers to produce dif-
ferent hypotheses to explain the functioning of Long Term Memory and retrieving
processes.
It is clear that writing models do not take into account these theoretical differ-
ences. They sometimes seem to evoke the presence, in Long Term Memory, of a
‘certain amount’ of useful ‘knowledge’, with various forms, without justifying the
choice of this or that kind of knowledge, in functional and structural terms, in the
framework of writing activity.
The influence of domain knowledge on writing processes. Some works, analysing

the role of domain expertise on writing processes, could offer a solution (even
though partly) to the problem of the domain knowledge structure and its effects on
recovery during writing. These works aim at considering both the knowledge avail-
ability (activation) and organisation (structure). Their results show that the nature
and the knowledge availability or structure can considerably facilitate or modify the
functioning of the retrieving sub-process and probably even the Organising sub-
process (Cf., for example, the experimental studies led by Alamargot, 1997; Cac-
camise, 1987; Heurley, 1994; Holland, 1987; Kellogg, 1987a; Savigny, 1995; Voss,
Vesonder & Spilich, 1980).
An experiment by Ehrich and Koster (1983), although carried out in oral lan-
guage, seems to illustrate modalities of domain knowledge influence on speaking
processes particularly well. The authors asked adult participants to precisely memo-
rise the nature and the location of furniture composing a miniature room (dollhouse
to 1/10th), before orally describing the content of this room, in order that a listener,
who has never seen this room, could easily imagine it. By manipulating the ar-
rangement of the furniture (that could be functional – i.e., the chairs are around the
table, etc. – or non functional – i.e., the chairs are around the bed, etc.), the authors
have shown that the functionality of the furniture arrangement led to the adoption of
different global and local description strategies (as: ‘Round About’, ‘Parallel Line’,
‘Grouping’ or ‘Sequencing’ strategies). Moreover, the effect of the functional ar-
rangement was not only limited to variations in the general organisation of informa-
tion contained in the discourse (macrostructural level), but equally influenced the
microstructural (semantic marking) and syntactic organisation. For example, the
adoption of the strategy ‘Grouping’ led the speakers to implement structures such as
‘Topic-Comment’ (according to the definition of Reinhart, 1981), where the object
reference was thematised and defined the set of the objects to be described that are
linked to it. The Sequencing strategy generated the establishment of structure such
as Given New (according to the definition of Clark and Clark, 1977), managing the
linking of both new information (object to describe) and already provided informa-
tion (already described object).
This experiment seems to perfectly illustrate the influence of the domain knowl-
edge organisation on the accomplishment of a discourse. This experiment equally
constitutes an interesting paradigm whose principles could be used to study the role
of the domain knowledge on the generating sub-process. The possibility ‘to imple-
ment’, in writers, some pieces of domain knowledge whose initial structure and/or
PLANNING PROCESS 47
modes of acquisition could be controlled, before the beginning of writing, would

certainly contribute to explain more precisely the relationships between these kinds
of knowledge and the functioning of the Generating sub-process (Cf. Alamargot,
Espéret, & Savigny, 1993). In addition, this can lead to an evaluation of the conse-
quences of these relationships on the other writing processes (Cf. Chanquoy, Foulin,
& Fayol, 1990; Dansac & Alamargot, 1999).
Finally, in Hayes and Flower’s (1980) model, the Generating sub-process has a cru-
cial position within the Planning process, because it determines and limits the con-
tents to be written. It can thus be considered as a kind of interface between the
writer’s knowledge in Long Term Memory and elaboration processing to apply to
the retrieved knowledge before the Translating process. Consequently, at the end of
each content generating step, it will probably be necessary for the writer to organise
these generated contents as a function of a communicative goal and a text plan. The
next part therefore aims to present and define these processes through different pro-
duction models.
6 THE ORGANISING SUB-PROCESS

6.1 Definition of the Organising sub-process
According to Hayes and Flower (1980), the contents retrieved from Long Term
Memory by the Generating sub-process are organised within the Organising sub-
process. The main function of this sub-process is to establish writing plans recording
the different topics formerly retrieved or created.
While the interpretation of the notion of ‘Topics’ varies from one author to an-
other (i.e., ‘Main Points’ for Bereiter, Burtis and Scardamalia, 1988 and ‘Macro-
structure’ for van Dijk and Kintsch, 1983), they all agree on the fact that the activity
of the Organising sub-process and its constitutive operations plays a central role in
achieving a strong textual coherence by deciding the order of appearance of topics
(or contents) in the text and especially by establishing conceptual or semantic rela-
tionships between these topics.
Kellogg (1988, 1990) or Flower, Schriver, Carey, Haas and Hayes (1989), for
example, showed that the general text quality was considerably improved when a
specific training was priorly provided on organisation (Cf. hereinafter). According to
Bereiter and Scardamalia (1987), this sub-process, by operating under the double
dependence of Content Knowledge and Discourse Knowledge, would mainly indi-
cate the Problem Analysis components of Knowledge Transforming Strategy. The
efficiency development of the Organising sub-process would represent one of the
main criteria for accessing writing expertise (Cf. Chapter 6).
6.2 The operations of the Organising sub-process

The Organising sub-process is often considered as central in the Planning process
and, more generally, in writing activity. However, it is surprising to note that the
definition of its constitutive operations is often only very cursory and can considera-
bly vary from one model to another, both in its terminology and in the functions of
its different operations.
6.2.1 The organising operations according to Hayes and Flower (1980) and
Bereiter, Burtis and Scardamalia (1988)
According to Hayes and Flower (1980), the organising sub-process is made up of a
series of operations undertaken both by the suppression of certain knowledge units
(via the ‘Evaluate usefulness of topic’ operation), and by the ordering and the hier-
archical categorising of retained units (Cf. Figure 11). This last aspect is thought to
be produced by the activity of a group of five organising operations, allowing to link
up the retained units (via the ‘Identify a possible first or last topic’ operation) and to
establish different types of semantic relationships between the knowledge units.
No Exit
Goal = Organize ?
Yes
Read next note

Not Useful
Evaluate usefulness
of Topic
Useful
Identify as Order with Search for Search for

respect Previously Previously Identify
a possible first to a noted topics noted topics
previously subordinate to superordinate a category
or last topic noted topic present topic of present topic
Succeed Fail
Organizational
Note
Figure 11: Organising sub-process framework,adapted from Hayes and Flower (1980).
These relationships can be direct (‘Order with respect to a previously noted topic’),
subordinated to a previously processed topic (‘Search for previously noted topics
subordinate to present topic’), superordinated (‘Search for previously noted topics
PLANNING PROCESS 49
superordinate to present topic’), or even hierarchically categorise’ (‘Identify a cate-

gory’).
Bereiter, Burtis and Scardamalia (1988) have succeeded in defining such organ-
ising operations in various ways. The authors maintain that four operations support
the drafting of the Main Points (or Topics). One unit (generated and selected from
Long Memory Memory by ‘Search’ and ‘Delimit’ operations) is evaluated to consti-
tute one Main Point by means of the ‘Fit’ operation, while the ‘Cohere’ operation
ensures the elaboration of the relationships between the different knowledge units.
The ‘Structure’ operation allows a textural organising structure to be adopted or
elaborated according with the referential organising structure (rhetorical aspect). Fi-
nally, the ‘Review’ operation can modify the nature of the Main Points related to
constraints imposed by writing, on the basis of the evaluation of the translated prod-
uct.
6.2.2 Organising operations: comprehension and problem-solving activities

Whatever their definitions, the Organising operations are fundamental in the writing
activity, because they enable reaching a strong coherence within the text content, by
establishing links between idea units. According to Hayes and Flower (1980), plan-
ning processes, and more precisely the Organising sub-process, would be particu-
larly costly in cognitive resources. This cost would mainly depend on the highly
controlled organising operations. These operations can be considered as operations
of comprehension and analysis of semantic properties of contents activated in Work-
ing Memory. It is probably in this meaning that Hayes and Flower (1980) call these
operations ‘Evaluation’, ‘Research’, or ‘Identification’, or more recently, that Hayes
(1996) includes these operations in the ‘Reflection’ process.
Conversely, Bereiter and Scardamalia (1987) consider Organising operations as
problem-solving operations. These operations allow to proceed to the installation of
(new) semantic relationships between activated knowledge and perhaps (as devel-
oped in the above section concerning the Generating sub-process) to create new
knowledge units (Cf. Burtis, Bereiter, Scardamalia, & Tetroe, 1983). These opera-
tions could then enable the elaboration of a text with a strongly structured content,
according to a hierarchical mode and whose Main Points, organised and connected,
would have a strong coherence. According to Bereiter and Scardamalia (1987), these
problem-solving operations, as previously defined, are the central components of the
architecture of their Knowledge Transforming Strategy of text production. These
operations are performed under the double constraint of domain knowledge charac-
teristics (Content Space) and pragmatic and linguistic possibilities (Rhetorical
Space). The content of the text would thus be organised according to a double crite-
rion: (1) ‘what I have to tell’, and (2) ‘how I must or can tell it’. In other words, the
Organising operations are fundamental in the Knowledge Transforming Strategy.
They would therefore mainly concern expert writers, mainly because their use would
be particularly costly in cognitive resources as they are highly controlled.
6.3 The processing cost of Organising sub-process

According to Bereiter and Scardamalia (1987), younger (or novice, inexperienced)
writers would not be able to manage organising or problem-solving operations be-
cause their planning and processing abilities are weaker and more limited, and also
because the load associated to the writing task is heavier than for experts, as they
have not yet automatised the low-level writing operations (such as graphic transcrip-
tion; Cf. Chapter 6, devoted to the development of writing expertise). Studying the
cost of Organising operations, in expert writers, supposes (1) to be able to effec-
tively evaluate the presence and the functioning of these operations, and (2) to de-
fine in which situations or conditions they are the more solicited (for example, as a
function of the text type).
6.3.1 To evalutate the presence and the functioning of Organising sub-process

The importance of the cost of the Organising sub-process has often been demon-
strated with the help of various experimental measures and paradigms.
Experimental measures. Two types of measures are generally used to surround the
activity of organising operations: (1) Off-line measures (i.e., product analysis) en-
able finding in the text the trace of the activity of Organising operations. An analysis
of the text local and global coherence, of the textual organisation of information, of
transitions, of the nature of links between different themes, sub-topics or different
parts of the text, may contribute inferring the nature of organising operations that
have been used (Cf. Chanquoy & Fayol, 1995; Dansac & Alamargot, 1999; van
Wijk, 1999); (2) On-line measures do not evaluate the presence of organisation op-
erations but the cognitive cost and/or the duration of their activation. Two types of
temporal parameters are generally used (Cf. Piolat & Pélissier, 1998): (a) The analy-
sis of variations of reaction times to a secondary task during the writing activity, al-
lows to infer variations of writing processing. The postulate is that the longer the re-
action time to the secondary task, obviously the more cognitive resources were
needed for the execution of the main task (writing). (b) The analysis of pause dura-
tions during writing allows to observe the temporal segmentation of the activity and
to notice periods during which the writer is engaged in a mental activity such that
graphic transcription is impossible (Alamargot, 1997; Foulin, 1993, 1995). The
claim is that the longer the pause duration, the more complex and/or numerous are
the ongoing processes. To surround the presence, the functioning and the cost of or-
ganising operations supposes usually to both analyse off-line and on-line variables
(Cf. Schilperoord & Sanders, 1997; Schilperoord & Chanquoy, 1999), by using spe-
cific paradigms.
Experimental paradigms. A first method to study the Organising sub-process is ask-

ing the writer to retrieve and organise ideas, before the starting to write (i.e, prewrit-
ing stage during which the writer has to write a draft or a plan; Cf. Bereiter, Burtis,
PLANNING PROCESS 51
& Scardamalia, 1988; Bereiter & Scardamalia, 1987; Kellogg, 1987b, 1988, 1990;
Scardamalia & Bereiter, 1985, 1987). The improvement in the quality of the final
product and, more particularly, a high local and global coherence within the text
would thus result from the processes linked to the organisation (elaboration of a
plan). This particular writing mode would allow the writer, during the task, to only
devote cognitive resources to (a) the elaboration of text content from the plan, (b)
the linearisation of content, and (c) the linguistic formulation (in other words, the
Translating process: Cf. Chapter 2).
A second method to put in obviousness the cost of the organising process con-
sists, for example, of the comparison of texts produced by experts and novices in
domain knowledge (Cf. above the conclusion on the Generating sub-process). Kel-
logg (1987a) has thus shown, while analysing reaction times to a secondary task,
that the domain expertise led to a decrease of cognitive resources needed for retriev-
ing ideas and then that these resources were used for organising these ideas. In addi-
tion, domain experts’ text content appears more coherent and organised than those
of novices.
These measures and these methods are not obviously exclusive to the study of
the Organising sub-process functioning, and can be (or are) used to study the activity
of the other writing components. However, the Organising sub-process, because it is
strongly controlled and strategic, can easily be studied using on-line methods, op-
posed to other writing processes which are more automatised, at least in experts,
such as the lexical retrieval or the graphic transcription of the retrieved ideas. Never-
theless, it is disturbing to observe, in spite of these methods, that there exists only
little progress in the modelisation or in the formalisation of the Organising sub-
process, as regards experimental data. Besides the evaluation of its presence in ex-
perts (i.e., in Knowledge Transforming Strategy) or the evaluation of cognitive re-
sources that it needs (Cf. Kellogg, 1987a), there are few investigations aiming to
specify the definition of its constitutive operations and their activation.
6.3.2 Organising operations and text types

On this last point, it is certainly important to note that the necessity to use, and the
cost of, the Organising sub-process can vary according to the type of text consid-
ered. For example, Matsuhashi (1981) has shown that pause durations significantly
varied as a function of the text type to be produced: the writers made shorter pauses
when they wrote stories than argumentative texts. In the same way, if the notion of
‘narrative schema’ has a certain psychological validity (Cf. Fayol, 1985), then, in the
case of a narrative, the writer would simply have to put into words the retrieved
ideas, that are, in this case, postulated to be already organised via this schema. In
this case, the cost associated to the written production would be reduced and its pro-
gress facilitated (Cf. McCutchen, 1987). Thus, the presence of schemas (or ‘plans of
text’ for Schneuwly, 1988; or ‘super-structures’ for Kintsch and van Dijk, 1978)
would decrease the cost of the organisational work linked to the production, since
the writer would only have to fill categories (or slots) with the retrieved information
(Bereiter, 1980; Matsuhashi, 1982). However, most often, the retrieved information
is not directly usable and has to be transformed and reorganised. In this case, the
Organising sub-process and its constitutive operations would have to be applied,
leading to a long, expensive and complex period of text content planning.
6.4 Conclusion on Organising sub-process: from specific to general operations

Finally, despite the numerous definitions of the Organising sub-process and the dif-
ferent methods to experimentally study this sub-process, there is still little precision
about the nature of its composing operations, either in the model of Hayes and
Flower (1980) or Bereiter and Scardamalia (1987). The most important fact is cer-
tainly that some authors currently hypothesise an equivalence between the Organis-
ing sub-process and processes of comprehension, analysis and inference on the con-
tents activated in Working Memory. For example, in Hayes and Flower’s (1980)
model, the Organising sub-process clearly focuses only on conceptual representa-
tions during planning (i.e., to elaborate and to organise main text ideas into a plan).
It is then possible to envisage that these operations are not specific to the text writ-
ing activity (notably, Hayes, 1996, situates the Planning process in the larger context
of Reflection). In this framework, one begins to wonder whether the organising op-
erations are specifically textual operations or can be considered as more general rea-
soning or comprehension operations that would be necessary during the planning ac-
tivity. This more complete definition of organising operations leads to several con-
sequences on how it is possible to consider the architecture of writing processes.
For instance, it should be noted that the Organising sub-process, in all the above-
presented models, is considered only as a component of the more general Planning
process. In this way, the Organising sub-process only suggests the drafting of a text
plan (according to Hayes and Flower, 1980) or even of a Macrostructure (in terms of
van Dijk and Kintsch, 1983). More precisely, according to Hayes and Flower’s
(1980) model, it is only during the Translating phase that the main ideas can be ef-
fectively developed (Cf. Chapter 2). Is it nevertheless possible to hypothesise that
these organising operations, if they are general, could also be applied during the
Translating phase? If this is possible, they could thus ensure not only the text global
coherence (i.e., text plan), but also its local coherence (i.e., for instance: coherence
between semantic propositions).
If, following Hayes (1996), we consider that organising operations are general
and non-specific for writing, they could be applied to all types of information proc-
essing, as the organisation of ideas in order to make a graph or a drawing, perhaps
even to solve a problem, etc. The difficulty is then to understand the relation be-
tween these general dimensions of organising and the writing activity. For example,
if considering the organising processes as a more general comprehension activity
seems very interesting, it still remains difficult these days to precisely define or de-
scribe what could be the nature and the role of such a comprehension activity inside
writing models. Moreover, comprehension is a very complex cognitive activity and
it seems difficult to extract some processes to integrate them in writing models (we
PLANNING PROCESS 53
will come back to this discussion about the relation between comprehension and
production activities in Chapter 3, dedicated to revising activity).
Finally, besides processes ensured by Generating and Organising sub-processes,
another processing must equally be considered as fundamental in the activity of
planning: the pragmatic process. This process contributes to the domain knowledge
recovery and the idea organisation into a text plan, by orienting the whole set of
planning processes to the addressee and the communicative goal.
7 THE PRAGMATIC PROCESSING

7.1 Definition of Pragmatic processing
The Pragmatic processing is a fundamental component of planning in writing activ-
ity because it determines the communicative quality of the produced text. The
Pragmatic processing functioning supposes that a representation of the potential
reader of the text is clearly established in the writer’s mind, so as to allow the writer
to adapt the text linguistically and semantically to this addressee. According to Trax-
ler and Gernsbacher (1992, 1993), the content of a text describing geometrical fig-
ures would be more precisely elaborated when the writer had the opportunity to be
confronted with the reading of his/her own text by a reader who had to redraw the
described figures afterwards (Cf. equally Holland, 1987, who studied this effect in
children, in the case of instructional texts).
Such a ‘strengthening’ of the communicative aims would arouse a more impor-
tant control of the writing activity, increasing the activity of Planning (through a
highest quality organisation), Translating (i.e., propositions could be more strongly
elaborated, lexical and grammatical choices more carefully carried out), as well as
Reviewing (by a modification of the text to be more adapted to the reader) proc-
esses. In this way, Ransdell and Levy (1994) have shown, with college students, that
the written text quality (in terms of syntactic complexity) was increased when the
communicative aims of the text included a potential reader well known by the writer
(to write for a peer). According to these authors, this pragmatic effect would not
only improve the linguistic quality of the written product, but would also facilitate
the functioning of the Organisation sub-process and Reviewing process.
Such a facilitation is not however systematic and it seems, a contrario, that the
writing of a text, for a ‘non familiar’ reader, could lead to a more important cogni-
tive cost of processes. This fact has been made obvious by Caccamise (1987), with
adults (psychology students) to whom it was required to transcribe the same domain
content for other adults or, what is less common, to children. The texts written for
children comprised more details (i.e., the elaboration was more important), but
fewer relationships between the main idea units. However this apparent simplifica-
tion in the text organisation was only relative because it seems than this mode of
writing is more constraining and interferes with the quality of the final text.
These experiments seem to clearly demonstrate how the consideration of com-
municative goals can, within particular conditions, modify some of the writing proc-
esses and their processing cost. This phenomenon would operate when the potential
reader’s representation is not (or little) integrated into the writer’s preliminary
pragmatic knowledge (notably in younger writers, Cf. Bracewell, Scardamalia, &
Bereiter, 1978) and must, consequently, be elaborated during writing. This could
thus explain the fact that, when the text addressee is a peer, the writing processes
(and therefore the text) would be facilitated, by limiting the elaboration cost of the
addressee’s representation. If the relevant representation of the reader’s characteris-
tics is a condition to pragmatic processes, these processes equally need rhetorical,
textual and linguistic knowledge to reach, during the Translating process, the com-
municative objectives and to guarantee the precision, the quality and the linguistic
orientation of the text. Similarly, the Pragmatic processing can necessitate an analy-
sis or a modification of the domain knowledge, so as to optimise the adaptation of
the text to the reader. Pragmatic processing therefore requires a fine exploitation and
articulation of the writer’s textual knowledge.
While all authors agree on the fundamental role of pragmatic knowledge for the
adaptation of the text to the reader, the description of their influence, although sys-
tematically present in the different models of verbal production, remains often very
cursory. After a closed examination of these cognitive models, it seems that the
pragmatic aspect arises less from a form of processing than from a control structure
influencing the components of the writing activity as a whole, by defining or rede-
fining their goals. However, the more complete description of pragmatic compo-
nents and their functions during production is not found in cognitive models but
rather in psycholinguistic models of language production. In this way, Bronckart,
Bain, Schneuwly, Davaud and Pasquier’s (1985) model will be carefully described.
7.2 Pragmatic processing in cognitive models of verbal production: from process to

control.
7.2.1 Pragmatic processing in Hayes and Flower’s (1980) model
In Flower and Hayes’ (1980) theoretical proposals, rhetorical or pragmatic processes
would be ensured by the ‘Planning to Do’ operation. However, in Hayes and
Flower’s architecture (1980), it does not clearly appear a specific process (or sub-
process) for pragmatic processing. It is nevertheless possible to consider, in the
framework of processes and representations of this model, two dimensions being
able to contribute to the execution of pragmatic processes: (1) the influence of the
Task Environment, and (2) one of the possible functions of the Goal Setting sub-
process.
The Task Environment, and more particularly one of its components labelled
‘Writing Assignment’ (i.e., topics, potential addressee, motivating cues), formalises
the writer’s external environment, by taking into account the communicative goal
and the reader. However, the influence of these pragmatic dimensions only concerns
the activity of the Generating sub-process. More precisely, this influence would ap-
pear during the elaboration of the first memory probe (on the basis of the Task As-
signment), and on the Evaluate Retrieved Element operation, during the associative
PLANNING PROCESS 55
recovery of knowledge units (Cf. above, the architecture of the Generating sub-
process).
The Goal Setting sub-process, the third component of the Planning process, es-
tablishes criteria to ensure the appropriateness between the written text and fixed
communicative goals. This sub-process, according to Hayes and Flower (1980; 15),
would not process domain knowledge but some writing criteria retrieved from Long
Term Memory. For example, the Editing sub-process, during the text reading and
revising, could use these criteria, maintained by the Goal Setting sub-process, in or-
der to improve the quality of the text. Is it then possible to consider that one poten-
tial function of Goal Setting is to establish, among others, pragmatic criteria? There
is not, however, a clear explanation on the function of pragmatic control led by the
Goal Setting sub-process. This sub-process could moreover play an important role in
the temporal management of processes (Cf. Chapter 4).
7.2.2 Pragmatic processing in van Dijk and Kintsch’s (1983) model

Van Dijk and Kintsch (1983), through the Macroplanning strategy, have proposed a
formalisation more explicit and complex than those proposed by Hayes and Flower
(1980), about the Pragmatic processing and its influence on writing. According to
these authors, any production, written or oral, must firstly be situated in a communi-
cative context relying on cultural and social conventions. The verbal production is
only one means among others to achieve a susceptible action to modify a situation,
if the goal is fixed by the speaker or the writer (Cf. Figure 12). In this context, the
Macroplanning strategy consists of evaluating, in the case of an oral interaction, for
example, (1) in what measure a situation can be modified (i.e., in what measure an
addressee can modify her/his opinion, for example), and (2) with what means a
situation can be modified (i.e., how it is possible to orally convince the addressee).
The elaboration of such a Macro-plan supposes, in one instance, that the producer of
the message has been able to elaborate a representation of the situation, the context
and the possible actions, as well as of her/his own motivations and goals (Plans for
global speech acts). In a second instance, the effective execution of this Macro-plan,
recording what van Dijk and Kintsch (1983) named the Macro-speech act, supposes
an analysis and a decomposition of this Macro-speech act in different Local Speech
Acts that must be related in the speech.
This analysis relies on the elaboration of a pragmatic Macro-plan, assuming a
systematic scanning of the situation, the addressee and the consideration of her/his
feedback.
According to van Dijk and Kintsch (1983), the Macroplanning strategy would be
particularly complex to instanciate because it requires an important number of
knowledge and situational parameters (Cf. Figure 12). For example, the speaker or
the writer must previously possess some general knowledge of linguistic interac-
tions, social and cultural rules, the preceding discourse, the interlocutor’s (or
reader’s) characteristics and beliefs, etc. These different kinds of knowledge must
then be adapted and interpreted regarding the real communication context. Indeed,
‘knowledge of goals and preferences’ will constitute some ‘beliefs about goals and
preferences’ in the precise context of the speech (or writing) act, and thus represent
one of the situation parameters in order to plan the Global Speech act. Conversely,
‘knowledge concerning co-operation situations’ allows to directly evaluate what the
Local Speech acts, planned shortly, will be (by the Plan for local Speech acts proc-
essing). Hence, the choice of the ‘Local Speech Acts’ (direct, indirect, preparatory,
component, terminatory, etc.) and their sequencing during verbal production, would
be determined by:
• a ‘Strategy for Evaluating Local Execution’ (globally relying on (a) the exploi-
tation of the addressee’s representation – his/her beliefs –, interest and on (b)
the knowledge about strategies relating local and global plans), and,
• by the execution of a ‘Plan for Local Speech Acts’ (relying on the analysis of
the effect of the previously accomplished speech act and on the addressee’s
mental spirit).
Knowledge about Interests Knowledge about

goals, preferences & values interaction
Belief about Beliefs about Beliefs about

goals, preferences what can be what can be
in context achieved achieved through
verbal interaction
Social & Knowledge about

Social & Plans for Pragmatic speech acts &
cultural cognitive global analysis of
knowledge speech acts appropriatness
context actual context conditions
Knowledge about Strategy for

Knowledge about
cooperation prin- evaluating local strategies relating
ciples, beliefs & execution local and global
interests of hearer Previous local plans
speech acts and
Plans for their consequences Memory represen-
local tations of previous
speech acts
Hypotheses about speech acts
actual state of
hearer
Execution of
speech act
Figure 12: Pragmatic production strategies,

adapted from van Dijk and Kintsch (1983). Copyright © 1983 by the Academic Press.
According to the authors, executing this Macroplanning strategy can be very costly
for the speaker or the writer because it simultaneously and quasi-instantaneously
manages all these parameters. In this way, the Macroplanning appears as a very
high-level strategy and would have a top-down influence on all the other processing
levels or strategies (from the retrieving of macrostructural knowledge units to the
PLANNING PROCESS 57
linguistic translation of clauses or sentences; Cf. above the description of van Dijk
and Kintsch’s model). However, the authors are not very precise concerning the ex-
act modality of influence of this pragmatic plan throughout the verbal production
processing.
7.2.3 Pragmatic processing in Bereiter and Scardamalia (1987) model

Some elements of precision, about the pragmatic knowledge influence on the other
components of the verbal production, can be found in the framework of the writing
Knowledge Transforming Strategy (Bereiter and Scardamalia, 1987). In this model,
the pragmatic knowledge is located in a Rhetorical Space that circumscribes the
problems relative to the text during writing. These problems can concern the com-
municative aims, the addressee (knowledge and pragmatic goals) or the linguistic
and rhetorical means (discourse knowledge) available to reach the communicative
objective of the text. According to Scardamalia and Bereiter (1991), in the case of
the Knowledge Transforming Strategy, the results of the domain knowledge re-
trieval are transmitted to the Rhetorical Space as goals. These goals are not simply
goals that allow to express these pieces of knowledge, but are rhetorical goals, re-
cording pragmatic modalities of this knowledge. In return, the execution of these
rhetorical goals supposes, as seen above, the modification of the domain knowledge.
This mechanism of goal transfer between the two knowledge spaces is interest-
ing because it relies on an interactive processing different from that proposed by van
Dijk and Kintsch (1983), who rather stipulated an influence mainly top-down of the
pragmatic knowledge on the other processing strategies. The Bereiter and Scar-
damalia’s approach (1987) also allows to hypothesise, during writing, the modifica-
tion of domain contents under the pragmatic knowledge influence and vice versa.
Nevertheless, Bereiter and Scardamalia (1987) are still vague concerning the nature
of problem-solving operations ensuring the interactive transfer of content and prag-
matic goals.
7.2.4 Pragmatic processing in Levelt’s (1989) model

It is then tempting to hypothesise that the interaction between domain knowledge
and pragmatic knowledge (or more globally rhetorical knowledge) could be formal-
ised by means of a classic production system (according to Anderson’s definition in
1983), relying on the matching of declarative (i.e., domain knowledge) and proce-
dural knowledge (i.e., production rules recording the rhetorical use of domain
knowledge, structured by the communicative goals). Such a principle has been pro-
posed by Levelt (1989), to account for the pragmatic knowledge influence within his
speaking model. By formalising the pragmatic knowledge as a system of production
rules, he stated that, stored in the register of a procedural memory, this system
played, as such, the role of a (pragmatic) controlling architecture for the application
of verbal production processes. However, the architecture and the control conditions
of such a production system are not precisely described by Levelt (1989) and this
notion remains conjectural.
7.2.5 Conclusion on pragmatic processing in cognitive models of verbal produc-

tion
Finally, for Bereiter and Scardamalia (1987), van Dijk and Kintsch (1983) or Levelt
(1989), few precise explanations are provided about operations, modalities and,
mainly, the determinism of the influence of pragmatic control structures on (1) the
other writing processes, (2) the products resulting from these processes (i.e., on laws
or rules of modifications brought on the semantic or linguistic product, according to
the modification of constraints or pragmatic goals). If it is still difficult to find a pre-
cise description of these aspects in cognitive models of text writing, a rather clear
explanation can be provided by psycholinguistic models, such as this elaborated by
Bronckart, Bain, Schneuwly, Davaud, & Pasquier (1985) (Cf. also Bronckart, 1977).
7.3 A psycholinguistic model to explain pragmatic dimensions

One of the most precise descriptions of the pragmatic dimensions governing speak-
ers’ or writers’ behaviour is provided by Bronckart, Bain, Schneuwly, Davaud and
Pasquier’s (1985) model. This model presents two main advantages. The first one is
to formalise the interaction between verbal production and social context. The sec-
ond is to consider both the oral and the written act of production. Different from
most models, Bronckart et al.’s model (1985) is not elaborated through the form of a
framework with ‘boxes’ (describing the writer, the writing process and the possible
interactions with the external world) and ‘arrows’ (underlining the relations among
the different boxes). Conversely, this model cannot (or very difficultly) be drawn,
but it concerns a very precise description of the different activities in which the
speaker or the writer is engaged in the course of her/his speech act, as well as the
whole social environment of the production.
The authors thus convey both the activities in which the speakers or writers are
engaged during the production and the social environment as a whole. On this last
point, the originality of Bronckart et al.’s (1985) model is to offer a formalisation
distinguishing the ‘Production Act Space’ (describing the writer/speaker, the inter-
locutors, the place and the moment of the act) from the Social Interactions Space,
representing the specific effects that the linguistic activity is supposed to produce on
the addressee (the social place, the speaker, the writer, the reader, the goal of the
discourse).
The strength of this model relies on its consideration of the connection between
Text and Context. The result of the processing of Extralinguistic parameters takes
the form of Linguistic units organised into text, through the Contextualisation (the
definition of a zone of social interaction, for example), Structuring (discursive an-
chorage, for example) and Textualisation (cohesion, connection, for example) opera-
tions. More precisely, this model has been elaborated from the analysis of oral and
PLANNING PROCESS 59
written discourses, in different contexts (or Extra-language for the authors). It only
describes the adult functioning that is to say of individuals who have learned the
rules of language.
Bronckart et al. begins by describing the Spaces before the Linguistic operations.
They distinguish two spaces. The first is the Production Act Space, described by
three parameters: (a) the Writer (or the speaker) to which is associated a mode of
production, oral or written; (b) the Interlocutors (or co-producers) who are human
beings physically present during the production activity (this is therefore only true
for oral); (c) the Time Space which is the place and the physical moment in which
the production is accessible. The second space concerns the ‘Social Interactions’.
According to the authors, the linguistic activity is both one of the aspects of the So-
cial Environment and the Structure of (oral or written) Productions. Inserting into all
human activities, the language is necessarily articulated to an infinite set of social
parameters. Here, four of these parameters are taken into account: the Social place,
the Addressee, the Speaker (or Writer) and the Goal. The Goal represents the spe-
cific effect that the language activity is supposed to produce on the addressee.
Then, several Linguistic operations constitute the central element of the model.
They account for the articulation of the Text with the Context and the Referential,
that is to say the processing of Extra-language parameters and the result of this proc-
essing in the form of Linguistic units organised in a text or in a discourse. These op-
erations are included in a network organised in three levels. The first level accounts
for the Contextualisation, that is the processes of representation and choices of val-
ues on all relevant extra-language parameters. The second level concerns the Struc-
turing, or the building of the text framework. The third level ensures the Textualisa-
tion, that is the effective organisation of the discourse, considering different aspects
such as connection, cohesion and modalisation. The authors define more precisely
operations from these three levels:
• The Contextualisation operations, concerning the processing of extra-language
parameters, allow to define a precise zone of social interaction and to create a
balanced pair of speaker/writer and addressee. They authorise the syntactic or-
ganisation of referential elements in the form of clause structures.
• The Structuring operations constitute a hinge allowing the articulation of lin-
guistic representation procedures of the context (contextualisation) to commu-
nicative organisation procedures of the textual chain (textualisation). They are
divided into three groups. The Discursive Anchorage operations define a par-
ticular text type. According to Bronckart et al., there are four fundamental dis-
course types, depending on the discursive anchorage: the ‘discourse in situa-
tion’, the ‘conversational discourse’, the ‘theoretical discourse’ and the ‘narra-
tive discourse’. The Discourse Locating operations (in French: repérage) define
the underlying text framework. These operations mainly concern the organisa-
tion of the general temporality of the discourse during the process; this tempo-
rality can be translated in the text surface by specific linguistic units (for exam-
ple, the verb tenses). The Discourse Planning operations constitute the text su-
perstructure. The word ‘planning’ is used here in a different meaning than the
way it is usually used for writing or speaking: Planning accounts for the fact
that texts are organised in distinct and organised parts, that integrates proposi-
tional microstructures. In order to accomplish these operations, the
speaker/writer has to manage one or several plans (or superstructures, Cf.
Kintsch & van Dijk, 1978).
• The Textualisation operations, as indicated by their name, concern the translat-
ing stage, that is the effective sequential organisation of linguistic units. These
operations translate constraints that result from the linearity of discourse. Ac-
cording to Bronckart et al., they accomplish the plan, previously carried out, by
integrating propositional structures into it and maintaining and guiding the ad-
dressee’s attention. The authors hypothesise that the textualisation is carried out
by three essential procedures: the Connection (or ‘connexity’, the exact word is
‘connexité’ in French), the Cohesion and the Modalisation. The Connection op-
erations accomplish the text organisation through a hierarchical fitting. Thus,
the connection rules the marking of the connection points of a text with specific
units (such as co-ordinating and subordinating conjunctions, interclause or in-
tersentence adverbs, some prepositions or specific adverbial phrases, etc.). The
Cohesion operations ensure the maintenance of the discourse unit and continu-
ity. The Modalisation operations constitute the communicative anchorage by
controlling the speaker/writer-addressee pair
Bronckart et al.’s model give an important weight to the relationships between so-
cial context of production and a large number of linguistic variables, such as the
type of discourse, the discursive frame, the surface organisation of speech, etc. Fur-
thermore, it describes very precisely all the variables that the speaker/writer must (or
should) account for during the elaboration and formulation of his/her discourse, in-
cluding contextual and pragmatic variables. It can therefore be considered more as a
psycholinguistic model, than a cognitive model that is taking into account more
‘procedural’ aspects of the production. Finally, its major interest resides in the preci-
sion of pragmatic operations, usually simply sketched in classic models of language
production.
7.4 Conclusion on Pragmatic processes

Although the procedural models of written production, or more generally of lan-
guage production, postulate the influence of pragmatic dimensions at all processing
levels (from planning to execution), these models only supply a relatively limited
description of this influence, in comparison with psycholinguistic models (Cf. the
description of Bronckart et al.’s model just above). It seems nevertheless absolutely
necessary (1) to furnish precise definitions of these central pragmatic components in
writing, (2) to understand their functioning rules, and, in order to do that, (3) to sur-
round and precisely evaluate these pragmatic processes, both on a methodological
and experimental perspective.
PLANNING PROCESS 61
7.4.1 Investigating Pragmatic processing definitions and rules

In the initial Hayes and Flower’s (1980) model, the influence of Pragmatic is not
very clearly defined and it is possible to consider that one function of the Goal Set-
ting sub-process and of the Task Environment, is, among others, to ensure pragmatic
processes. However, it is interesting to put in obviousness (Cf. above) that the influ-
ence of the Task Environment is only considered at the level of the Generating sub-
process, while this environment can influence, because of its pragmatic aspects, the
whole writing process. Indeed, Hayes (1996: 3), in his new models, proposes an ad-
justment of the 1980 model for the best legibility of this aspect and the restrictive re-
lationship between the Task Environment and the Generating sub-process no longer
exists. Despite this precision, there remains a certain number of questions. If the
cognitive model’s ‘haziness’ on this aspect is made obvious, how is it possible to in-
tegrate the elements formalised in Bronckart et al.’s (1985) model, or the complexity
of the Macroplanning process described by van Dijk and Kintsch (1983), in general
writing models? In the same way, if pragmatic knowledge can be described by a sys-
tem of production, as proposed by Levelt (1989), what are then the precise structure
and the conditional rules that underlie this knowledge?
A precise description of such pragmatic rules is important because we can ask
whether the pragmatic components would exercise a more specific and privileged
control over certain writing processes or operations. For example, one hypothesis
could suggest that evaluation and selection operations of domain knowledge units,
decisive in the drafting of a text content, could more strongly depend on pragmatic
dimensions. In addition, the pragmatic control could be more marked in certain
types of text as in the case, for example, of argumentative texts necessitating a more
precise accounting of the addressee’s characteristics (Cf., for example, Coirier, An-
driessen, & Chanquoy, 2000). Without any doubt, Hayes’ new propositions (1996)
about the writer’s environment, motivations, hypotheses concerning the reader, etc.
will permit partly explaining the functioning of pragmatic aspects. However, these
new theoretical proposals are still very global and cannot be immediately empiri-
cally tested.
7.4.2 Surrounding and evaluating pragmatic knowledge and processing: some

experimental propositions
To evaluate or to manipulate the nature or the representation of the addressee is a
possible way to study pragmatic components. Most works that have raised this ques-
tion have shown that, generally, the writer, especially when s/he is young and/or in-
experienced, does not take into account the real or potential addressee of her/his
production (Cf. for example: Hayes & Flower, 1986; Scardamalia & Bereiter, 1986).
Moreover, some authors deem that the consideration of the addressee is difficult, be-
cause the writer would have difficulties imagining how the reader will process the
text (Cf. Traxler & Gernsbacher, 1992). This would therefore create a supplementary
cognitive cost, impossible to manage if all resources are used for other processes.
These general remarks again raise several questions. Why do writers have diffi-
culty to considering the reader? What are the reasons for these difficulties? Is this
absence of consideration due to a lack of cognitive resources or to a deliberated atti-
tude from the writer, who would not see the necessity to consider a possible ad-
dressee?
In fact, few works are currently able to provide answers to these questions be-
cause the principal difficulty concerns the way in which it is possible to ‘measure’
pragmatic knowledge and to evaluate its efficiency. As exposed in the introduction
of this section, one of the usual methods (Cf. Caccamise, 1987, for example) con-
sists of varying the addressee’s characteristics through writing instructions (i.e., to
write for an adult versus to write for a child for example). Another method is the
possibility that the writer observes the reader of the text. According to Traxler and
Gernsbacher (1992), this feedback on production could lead writers to gradually in-
clude the reader in the goals they have fixed during the planning stage. Finally, a
closer possibility concerns the transformation of the writing act in a more interactive
situation and to propose collaborative writing sessions (Cf. Boniface & Pimet, 1992;
or Pontecorvo & Paoletti, 1991 quoted by Fayol, 1997; Veerman, 2000) implying an
immediate adaptation to the reader on the text produced so far.
All these methods seem to be relevant, but they are relatively indirect. A possible
track would be, as proposed by Galbraith (1996, 1999), to distinguish writers’ indi-
vidual characteristics. This could be done in the framework of Snyder’s work
(1986), which considers the impact on social relations, of a personality factor, la-
belled ‘the Self-Monitoring’. According to Snyder’s (1986) test, the participants
who are classified as ‘High Self-Monitoring’ are more aware of social environment
influences than the ‘Low Self-Monitoring’ participants. Galbraith (1996, 1999) has
adapted this characteristic to text writing activity. The distinction between ‘low’ or
‘high’ self monitors as writers mainly concern the consideration of the addressee.
Indeed, the low monitoring writers would be less able to consider the addressee dur-
ing writing activity. However, even if this conception can be an interesting research
track, it remains true that the interpretation of such personality factors is still very
difficult.
8 DISCUSSION AND CONCLUSION ABOUT PLANNING PROCESS

Beyond the limited remarks and points of discussion, we propose throughout this
chapter, that two major questions remain concerning the definition of the Planning
process in writing. The first one is about the potential interactions among the differ-
ent Planning sub-processes. The second questions, more general, concerns the nu-
merous definitions of the Planning process and the method that could be the most
well-adapted to differentiate them.
PLANNING PROCESS 63
8.1 Interaction between Planning sub-processes

Such as defined by Hayes and Flower (1980), the multicomponent process of Plan-
ning is a particularly complex process. Its main functions consist of establishing a
plan of content, but a great number of uncertainties reside concerning the function-
ing modalities of Generating and Organising activities, and also concerning the role
and the influence of Pragmatic processes. These imprecisions are not only found in
the Hayes and Flower’s initial model (1980). Indeed, in most models, the planning
components remain unclear and not really justified. Furthermore, it seems necessary
to carefully examine the relationships existing among the different Planning sub-
processes (Cf. Alamargot, Favart, Coirier, Passerault, & Andriessen, 1999). The
processing distribution between Generating and Organising sub-processes is not
very precise, especially when creativity in text generation is considered. For exam-
ple, is it possible to consider that the Generating sub-process could be partly influ-
ence by the Organising sub-process? This could notably be the case when this last
sub-process elaborates Main Points or Topics of the future text while contents are
not precise or must be inferred. Similarly, what kind of relationships do the Goal
Setting sub-process share with the other Planning sub-processes? How do they inter-
act within the Planning process?
Current works made the role and functions of these different sub-processes dur-
ing writing obvious (Cf., for example, van den Bergh & Rijlaarsdam, 1999). Never-
theless, even if most authors note the recursion of writing processes, this recursion is
not specified for each sub-process of the main processes. It is then only possible to
hypothesise that the Planning sub-processes (and this is certainly true for the other
writing sub-processes) (1) are recursive, (2) interact and (3) can influence each
other. However, the modalities of these interrelations and the possible feedback
among sub-processes are neither clearly defined, nor empirically validated (this as-
pect will be questioned more in Chapter 4).
8.2 Different categorisations of Planning types

In the introduction of this chapter, Hayes and Nash’s (1996) composite definition of
planning has been provided, showing that elaborating a plan of text was only a kind
of planning among the others. The authors distinguish at least five possible types of
planning processes during the writing activity. But other taxonomies have equally
been considered before Hayes and Nash (1996), notably those provided by Flower
and Hayes (1981b), Carey, Flower, Hayes, Schriver and Haas (1989), or Burtis,
Bereiter, Scardamalia and Tetroe (1983). Hayes and Nash (1996), through their re-
view of questioning, produced an inventory of the differences on definition and dis-
tribution of the main types of planning (Cf. Figure 13).
Hayes and Flower (1980)

Kaufer et al. (1986) Planning Translating
Planning Planning Planning

Flower and Hayes (1980)
to Compose to Do to Say
Burtis and al. (1983) Conceptual Content

Kozma (1990) Planning Generation
Abstract
Witte (1987) Pre-Text Transcribing
Planning
Rhetorical Content
Carey and al. (1989)
Planning Planning
Figure 13: Different ideas of the activity of planning in writing,adapted from Hayes and
Nash (1996). Copyright © 1996 by Lawrence Erlbaum Associates. Adapted with permission.
This inventory is interesting because it illustrates very clearly to what extent mean-
ings and definitions of the planning process can differ.
Another good example of these differences is provided by a differential analysis
of the definitions of (1) the Planning process in Hayes and Flower’s (1980) model
and (2) Levelt’s (1989) Conceptualiser in his speaking model. Often confused in the
literature, these two entities, dedicated to content planning, do not cover the same
sub-processes or modules. More precisely, as highlighted in the General Introduc-
tion, the main difference between Hayes and Flower’s (1980) and Levelt’s (1989)
models rely therefore on the position of Microstructural processes. Inserted in the
Conceptualiser by Levelt (1989) and van Wijk (1999), these semantic processes do
not concern the Planning process in Hayes and Flower’s (1980) and would constitute
one of the first stages of the Translating process. This implies that the Translating
process, opposed to Levelt’s Formulator, does not exclusively process linguistic
knowledge, but equally semantic and conceptual knowledge (Cf. Chapter 2).
Problems about heterogeneousness of definitions are justifiable and certainly
theoretically desirable. However, the remaining problem concerns the fact that the
choice of such repartition is not systematically discussed and argued in models. The
operations of the different planning forms are often a priori defined while it would
be important to experimentally confront these distributions and the categorisation of
operations in the different planning processes.
This leads to the problem of the evaluation and the measure of this (or these)
planning(s) for processes and contents. More particularly, is it possible to apprehend
them by a different method than by the verbal protocol method? This kind of proce-
dures is indeed subject to many controversies (Cf. Chanquoy, 1998; Piolat & Pélis-
sier, 1998), notably concerning their validity. For example, for expert writers, it is
possible to wonder whether some planning components, surely largely procedural-
ised, can be tracked with verbal protocols? Conversely, with novice writers, the ver-
bal protocol method can be considered as a secondary and competing task, therefore
PLANNING PROCESS 65
costly in cognitive resources. This second task would therefore not allow managing
such high level activities as planning and verbalisation.
Finally, it then appears necessary (1) to attempt to harmonise models concerning
the planning process and its sub-processes, (2) to homogenise, if possible, writing
and speaking models considering the planning process, and (3) to find methodologi-
cal means to validate Planning sub-processes and their role during writing. This last
point seems more crucial in the case of Pragmatic processing.
CHAPTER 2
TRANSLATING PROCESS
1 INTRODUCTION: FROM CONTENT ELABORATION TO GRAPHIC

EXECUTION
According to Hayes and Flower’s (1980) model, the writer would have built, as an
output of the Planning process (Cf. Chapter 1), a content plan. This plan, following
communicative and eventually rhetorical intentions, lists an organised structure of
the main themes or main ideas of the to-be-written text. However this content plan is
just a general plan. It can only permit to plan the general text content and to keep the
global coherence. It does not provide any detail about this content and, furthermore,
any information about the future ‘linguistic and physic aspects’ of the text, on paper
or on a computer screen. In other words, if the Planning process organises and plans,
it does not enable the effective composition and realisation of the text. Indeed, after
the Planning process, the writer has to (1) elaborate the text content by developing
each part of the plan that needs to be developed, and (2) translate this developed
content into a linguistic form. It is in this way that Hayes and Flower (1980: 15)
consider that ‘the function of the translating process is to take material from long
term memory under the guidance of the writing plan and to transform it into accept-
able written English sentences’. The general definition of the Translating process
they propose is underlined by the fact that these authors consider that the domain
knowledge is not linguistic nor linear: ‘We assume that material in memory is stored
as a proposition but not necessarily as language. By a proposition we understand a
structure […] where concepts, relations and attributes are memory structure, per-
haps complex networks or images, for which the writer may or may not have names
(p. 15)’. This claim highlights that the writer, during Translating, needs to bridge the
gap between hierarchically organised conceptual thoughts (networks, for instance)
and a linearly organised linguistic structure.
8.3 Translating process in Hayes and Flower’s (1980) model

In Hayes and Flower’s model (1980), the change from a conceptual structure to a
linguistic structure is carried out by six Translating operations, respectively labelled:
‘Get next part of writing plan’; ‘Plan next sentence: Retrieve propositions’; ‘Express
next proposition part’; ‘Repeat first part of sentence’; ‘Express proposition part’ and
‘Interrogative’ (Cf. Figure 14).
D. Alamargot & L. Chanquoy (2001). Translating Process. In G. Rijlaarsdam (Series ed. ) &
D. Alamargot & L. Chanquoy, Studies in Writing: vol 9. Through the Models of Writing, 67 –
98. © 2001 Kluwer Academic Publishers. Printed in the Netherlands.
Get next part of

writing plan
Plan next sentence

Retrieve propositions Fail
Probe
Express next
Repeat 1st part
proposition part
of sentence
Succeed
Express
Sentence done ? proposition part
No
Fail
Yes
Part done ? Interrogative

No
Yes
Plan done ? Express

proposition part
No
Yes
Exit
Figure 14: Architecture of Translating process,adapted from Hayes and Flower (1980).
According to the authors’ description (pp. 15-16), the ‘Get next part of writing plan’
operation would allow the writer to retrieve (and certainly to maintain in Short Term
or Working Memory) a new piece of the text plan. The elaboration of a sentence,
controlled by this activated piece of plan, would be carried out by the ‘Plan next sen-
tence: Retrieve propositions’ operation. This operation would have a double func-
tion: it would allow (1) to retrieve, from Long Term Memory, some propositions,
which can be considered as semantic structures (for example: ‘Concept: Horse’;
‘Action: To gallop’; ‘Attribute: Fearful’), and then (2) to relate these propositions
into a sentence plan (‘Horse = To gallop’ [because] ‘Horse = Fear’). This plan,
which can be considered as ‘syntactico-semantic’, would then be next linguistically
translated, part by part, by the ‘Express next proposition part’ operation.
The linguistic translation could be controlled in order to verify and to optimise
the local coherence or the sentences syntactic structure. To do so, the ‘Repeat first
part of sentence’ operation would allow to maintain in Short Term Memory the pre-
viously translated part of the sentence (i.e., ‘The fear…’) so as to retrieve the lin-
guistic form to better express the proposition that is semantically linked through the
‘Plan next sentence: retrieve proposition’ operation. During this research, the ‘Inter-
rogative’ operation would select more and more suitable linguistic forms (i.e., ‘…
TRANSLATING PROCESS 69
the fear, which causes the horse to gallop…’; ‘the horse is galloping because he is
afraid…’; etc.) while the ‘Express proposition part’ operation would mentally test
these forms before their effective transcription, via the ‘Express next part of proposi-
tion’ operation (‘The fear makes the horse gallop’).
8.4 Qualities and limits of Hayes and Flower’s (1980) model

8.4.1 Six Translating operations defined
The description Hayes and Flower (1980) have made concerning the nature and the
functioning of Translating operations shows that, conversely to Berninger and
Swanson’s (1994) objection, the Translating box is not so ‘desperately empty’, as it
is observed in the schema of the model. Thus, if Hayes and Flower do not develop
the Translating process as precisely as for the Planning process, they define six
processing operations. These operations carry out the transformation of a hierarchi-
cally organised structure of the domain knowledge into a linear linguistic structure.
In addition, it seems important to notice that the description of the nature of these six
operations shows that Translating processing does not only carry out the linguistic
(i.e., grammatical and lexical) translation, but also conceptual (i.e., non linguistic)
processing. This last process elaborates the content of the future text, under the con-
trol of the text plan, and transforms this content into a preverbal message that will be
linguistically translated. In other words, the different kinds of Translating processing
would concern conceptual representations (i.e., retrieving domain knowledge, syn-
tactico-semantic planning) as well as linguistic representations (i.e., grammatical
and lexical translating of planned propositions).
8.4.2 The necessity to explain the nature and the functioning of the Translating
process
Whether Hayes and Flower’s (1980) proposals on Translating operations seem inter-
esting and judicious, the description they made is nevertheless relatively superficial
and general. Many points must thus be clarified, concerning both 1) a precised defi-
nition and 2) the strategic and recursive functioning of these Translating operations.
To complete the definition of Translating operations: towards four processing

stages. The definition of Translating operations by Hayes and Flower (1980) raises a
certain number of questions or discussion points:
• The fact that they consider that some memory contents are retrieved, from Long
Term Memory (under the control of the text plan), supposes the presence of a
retrieving sub-process (or operation) which would function solely during Trans-
lating. However, Hayes and Flower (1980) do not explain this retrieving
mechanism. Can it be considered as the Generating sub-process, that the authors
define as belonging to the Planning process?
• In addition, it is not obvious to consider that the propositions, even if they are
retrieved under the control of the text plan, can be directly used to constitute the
future sentences. Despite this observation, no semantic elaboration or re-
elaboration mechanism of retrieved propositions is found.
• The model is also relatively imprecise regarding principles of proposition inser-
tion into a sentence plan. This processing would be carried out only by one op-
eration, that being labelled ‘Plan next sentence: retrieve proposition’. It is nev-
ertheless not so obvious that the proposition retrieval and the sentence seman-
tico-syntactic planning use the same kinds of processing. Another processing
distribution, sharing more clearly content retrieving and content planning, could
be discussed.
• Beyond the processing distribution, it is important to make it clear that linguis-
tic processes, which are actually the heart of the Translating process, are simply
identified through operations such as ‘Express next proposition part’ or ‘Ex-
press proposition part’ (Cf. Figure 14), with no details, once again, about the na-
ture of the underlying lexical and grammatical processes. It seems nevertheless
very important to explain how the matching between a conceptual structure
(some semantic propositions) and a linguistic structure (grammatical and lexi-
cal) can take place and with what kinds of criteria.
• Finally, Hayes and Flower (1980), opposed to Kellogg (1996, Cf. General In-
troduction) do not describe handwriting or typewriting operations that carry out
the physical output of linguistic processes, that is the transformation of the lin-
guistic mental product into a written trace.
These critical points concerning Hayes and Flower’s (1980) model show the neces-
sity to specify the definition of processing led by the Translating process. Some re-
sponses can be found in other models (as well global as local) of verbal (oral or writ-
ten) production. To describe and analyse the nature of Translating operations
through these different models, we propose to take into account our previous critics,
to distinguish four processing stages. They can be differentiated by the nature of
processing operations that compose them, or by the nature of representations that
they process. We label these stages ‘Elaboration’, ‘Linearisation’, ‘Formulation’ and
‘Execution’.
• The Elaboration stage consists of retrieving and elaborating the text content
from a piece or the totality of the text plan. Within Hayes and Flower’s (1980)
model, this step mainly concerns processing carried out by the ‘Get next part of
the plan’ operation and by a part of processing ensured by the ‘Plan next sen-
tence: retrieve propositions’ operation, which here focuses on the content re-
trieval.
• The Linearisation stage enables a first transformation of these contents in a lin-
ear semantico-syntactic structure. This transformation would correspond to the
other part of processing ensured by the operation ‘Plan next sentence: retrieve
proposition’, which is here the planning of the next sentence.
• The Formulation stage, mainly linguistic, puts into words and into a grammati-
cally correct form the previously planned semantico-syntactic structure. It con-
cerns a linguistic stage of sentence elaboration.
• The fourth stage, called Execution, carries out planning and graphic execution
of the linguistic product; it thus concerns handwriting as well as typewriting.
Surrounding the dynamics of the Translating process. Beyond imprecision concern-

ing the definition of Translating operations, few indications are provided of the acti-
vation, the progress or the temporal sequencing of these operations. The processing
recursion nevertheless appears as a fundamental necessity. The transformation of
ideas into text does not obviously operate during one unique process activity, but
more likely during recursive processing cycles. Throughout the writing activity,
some portions of plans are activated, developed, then expressed in linguistic terms.
This is probably realised on small units. The question is then to identify the nature of
these units. Hayes and Flower (1980) consider that a sentence is produced part by
part. This cycle of processing is however only postulated. Are there experimental re-
sults to validate this notion of processing units? The observation of on-line parame-
ters (such as pause duration, Cf. Chapter 1) can represent an interesting investigation
means to appreciate the temporal progress of processing cycles during content
elaboration and linguistic formulation.
In the following parts of this chapter, the nature of the Translating process is ana-
lysed by describing, through different models, each of the four processing stages,
which characterise the Translating process (Section 2). Then, the principles of dy-
namics and recursion of the Translating process are developed (Section 3).
9 THE NATURE OF TRANSLATING OPERATIONS:

FOUR PROCESSING STAGES
9.1 Stage one: elaboration of text content from the writing plan
During Translating, the writer has to retrieve all the sets of ideas about a topic or a
sub-topic, which is defined by a part of the text plan. According to Hayes and
Flower (1980: 15), domain knowledge units or structures, which are retrieved from
Long Term Memory, are formatted into a propositional structure, labelled ‘Argu-
ment-Predicate’, and built, for example, such as: <(Concept A) (Relation B) (Con-
cept C)> or <(Concept D) (Attributes E)>. This retrieval can be executed by using,
as memory probe, the main idea from a part of the text plan. The content, retrieved
under the control of the text plan, can thus be considered as an elaboration and a
specification of main ideas. This possible definition of the elaboration of the text
content raises two central questions as for the definition and principles of content
generation. The first question focuses on the possible specificity of the content gen-
eration during Translating as compared to the content generation during Planning.
The second focuses on the nature of operations that could evaluate and modify the
retrieved contents (or create new ones) if they do not correspond to referential or
communicative text criteria, for example.
9.1.1 Generating during Planning vs. Generating during Translating

As previously said, the content retrieval, associated with each part of the text plan,
needs a generation sub-process or operation. There are two possibilities: (1) the ac-
tivity of the Generating sub-process, described by Hayes and Flower (1980) as one
of the Planning sub-processes, is triggered once again during Translating; (2) an-
other Generating sub-process would be specifically dedicated to the Translating
process. Hayes and Flower (1980) do not provide information allowing to clearly
choose between these two possibilities. Nevertheless, consequences of the function-
ing of Planning and Translating processes are different.
In the first case, when a writer starts to elaborate the preverbal message (accord-
ing to Levelt, 1989) or the textbase (according to van Dijk and Kintsch, 1983), the
first step always is going back to the Planning process, and especially to the Gener-
ating sub-process. Concerning the processing dynamics, the phase of translating
would then be carried out by a permanent alternation between Planning and Trans-
lating.
In the second case, the whole set of Translating processing can be carried out
without the Planning process. This last functioning mode leads to distinguish two
content generation-elaboration sub-processes (or operations). The first one (Generat-
ing during Planning) would be solicited only during the elaboration of the text plan.
It would only concern general and abstract contents susceptible to constitute this
plan. The second one (Generating during Translating) would be solicited only to
specifically and precisely elaborate the text content (Cf. Dansac & Alamargot, 1999,
for a discussion of this viewpoint). Such a ‘double’ definition of content retrieval
can be found in van Dijk and Kintsch’s verbal production model (1983: 274-275)
that clearly distinguishes two content generation levels. In their proposal about dif-
ferent writing strategies (Cf. Chapter 1), Van Dijk and Kintsch differentiate two
strategies (or two levels) of content elaboration. The first strategy, labelled ‘Macro-
structural Planning Strategy’, elaborates, as Hayes and Flower’s (1980) Planning
process, the text plan or the ‘Main Points’ (according to Bereiter, Burtis and Scar-
damalia, 1988). The second strategy, labelled ‘Microstructural Planning’, subordi-
nate to the first one, is more local. It consists of developing macrostructural proposi-
tions, and thus elaborating a more precise and structured content with microstruc-
tural propositions (the textbase). More precisely, according to van Dijk and Kintsch
(p. 275), the elaboration processes would consist of retrieving in Long Term Mem-
ory (or eventually in building) microstructural units, which will then constitute a re-
finement of a more general and global macrostructural unit. It seems thus, according
to this model, that two levels of content retrieving or generation could be effectively
distinguished: the first one underlies the elaboration of the text macrostructure, the
other the elaboration of the associated microstructure.
9.1.2 Generating during Translating: content assessment and modification

During Translating, Hayes and Flower (1980) consider that contents are elaborated
according to the double a priori constraint of the text plan and text communicative
goals. It seems that the nature of these particularised contents could immediately ful-
fil all pertinence conditions, concerning both the text plan and/or the text pragmatic
dimensions. This implies that content generation would function in a controlled way
(through text plan) during the progress of Translating while it is defined as automa-
tised during the Generating sub-process of Planning (Cf. the discussion of Chapter
1). The authors do not refer, within the operation ‘Plan next sentence: Retrieve
propositions’, to some possible a posteriori criteria of evaluation for the generated
contents. This absence of a posteriori evaluation of retrieved products under the
control of the text plan is all the more surprising that such a processing operation is
considered in the case of the Generating sub-process of the Planning process (la-
belled ‘Evaluate Retrieved Element’ operation, Cf. Chapter 1). Indeed, there is no
reason to think that these products (in the same way as the main ideas of the text
during Planning) should not be rejected, or modified or re-elabored during Translat-
ing. This processing hypothesis is interesting because it supposes that the writer is
able to create, and perhaps to invent, new contents during the progress of Translat-
ing. This fact would reinforce Galbraith’s (1999) ‘Knowledge Constituting’ model,
which, as indicated in Chapter 1, accounts for the new content elaboration during the
linguistic translation phase (Cf. equally Chapter 5).
van Dijk and Kintsch’s (1983) model still provides an interesting explanation
about the modification and/or the creation of contents during the progress of Trans-
lating. According to the authors, during verbal production, analysis operations
would be applied on macrostructural propositions (that have the role of the text
plan). These mental operations would evaluate, select, and (more important to ac-
count for the idea creation) to modify microstructural propositions deriving from
macrostructural propositions.
To describe content elaboration operations, the authors resort to their theory and
model of text comprehension. During the comprehension of a text or a speech, the
reader or the listener has to summarise the numerous presented pieces of informa-
tion, by integrating several microstructural propositions within one macrostructural
unit. These integration operations most often consist of ‘generalising’, ‘putting to-
gether’, and ‘abstracting’ a set of microstructural propositions.
In the case of the verbal production, van Dijk and Kintsch (1983) hypothesise
that the content elaboration operations would be opposed to integration operations
that are used during comprehension (i.e., allowing to elaborate general ideas of the
text from the textbase). More precisely, in the case of the production, these opera-
tions are called ‘Strategies for the Microstructure’ and are applied to the macrostruc-
tural units. They would ensure ‘Addition’, ‘Particularisation’ (opposed to Generali-
sation), ‘Completion’ or ‘Analysing’ of the macrostructural units. It would thus be
possible to enumerate elements from a set, to analyse actions into component ac-
tions, to particularise general predicates, etc. To clarify the functioning of micro-
structural proposition elaboration operations, van Dijk and Kintsch (1983: 274),
quote the example of the description of a trip to Italy. Here, a possible macrostruc-
tural unit can simply be ‘My trip to Italy’. This macrostructural unit represents a
large and general conceptual set, playing here the role of a plan portion susceptible
to be developed. This development can then consist of particularising the concept
‘Italy’ in enumerating visited places by providing a procedural and analytical list of
actions to be accomplished (‘to take the train to Rome, you have to…’). It can also
consist of analysing what would have been necessary to modify during the trip
(‘…but do not take the night train, because…’). And so, according to the authors (p.
275), microstructural elaboration operations allow ‘going down from topic to sub-
topic and from there to individual semantic representations’.
9.1.3 Conclusion: from content elaboration to linearisation of content

There still remains an important number of questions about the definition of process-
ing sub-processes or operations susceptible to carry out the elaboration of the text
content, during Translating. It is thus necessary to explore the possible differences
between (1) generation in order to constitute a text plan and (2) content generation in
order to establish the actual text content. It equally remains, in this last framework,
to better define the processing reality and nature allowing to evaluate, to modify and
to create these contents during the linguistic translation. As we have tried to show,
van Dijk and Kintsch’s (1983) theoretical model can provide some interesting re-
plies to these different questions, because the authors propose a relatively precise
description of the elaboration and analysis of the text content, through macro- and
microstructural strategies. For them, the elaboration of microstructural propositions
is only one step in the definitive elaboration of the textbase (i.e., the text content or
the preverbal message) and another function of microstructural processing is to
equally carry out the ordering of semantic propositions. More precisely, according to
van Dijk and Kintsch (1983: 275), ‘what we also need, as part of the linearisation
problem, is an appropriate order for these semantics representations’.
9.2 Stage two: linearisation of a preverbal message

At the end of a phase of content elaboration, under the control of the text plan, the
writer would have a more precise and more detailed content of a part or some parts
of the text to-be-written. Nevertheless, this part of the text content, resulting more or
less directly from the domain knowledge system, can still be multidimensional or
hierarchically arranged. Therefore this elaborated textbase must be firstly linearised
before linguistic formulation can commence.
9.2.1 Linearisation in Hayes and Flower’s (1980) model

In Hayes and Flower’s (1980) model (Cf. p. 15 in their chapter), linearisation would
involve putting together at least two retrieved propositions (or knowledge structures)
in a unique ‘syntactico-semantic’ structure. For example, in Table 1, the two propo-

sitions A and B are integrated into a Sentence Plan.
Table 1: Example of the functioning of the ‘Plan next sentence: retrieve proposition’ opera-
tion, adapted from Hayes and Flower (1980). Copyright © 1980 by Lawrence Erlbaum Asso-
ciates. Adapted with permission.
1a - Proposition A: ((Rousseau) (showed) (some early promises))
1b - Proposition B: ((Rousseau) (did) (very little painting until 40))
2 - Sentence plan: ((Proposition A) but (proposition B))
The integration of at least two propositions into a syntactic plan has two important
consequences, which are not developed by Hayes and Flower (1980). On the one
hand, the syntactic planning notion supposes that a minimal order is defined in the
aggregation of propositions. This fact concerns a form of linearisation of proposi-
tions that will thus be organised ‘end to end’, in order to constitute the continuation
of ideas contained in a text sentence. On the other hand, the aggregation of proposi-
tions supposes to consider at least one semantic relationship between these proposi-
tions. In the above example from Hayes and Flower (1980), it concerns a relation-
ship of opposition (‘but’).
The linearisation processing can indeed be considered as a semantical (re)coding
of the generated propositions (or knowledge units) into one syntactic plan. This
processing stage is labelled ‘Shape’ by Levelt (1989), van der Pool (1995) or van
Wijk (1999). According to the authors, semantic choices, occurring during the lin-
earisation processing, determines, for example, which propositional unit will be po-
sitioned as ‘focus’ (defining a concept of object, place, or person), which new unit
should be supplied within a structure such as ‘Given-New’, or which unit can be
thematised in a structure such as ‘Topic-Comment’, etc. The roles of agents and pa-
tients, the definition of actions, etc., would, at this level, be defined within a predica-
tive structure. Thus the question is how these characteristics and this semantic order-
ing are chosen. In other words, how does a writer organise microstructural units to
constitute the textbase (or the preverbal message)?
9.2.2 Semantic Strategies of Order according to van Dijk and Kintsch (1983)
van Dijk and Kintsch (1983; page 275) describe at least three different strategies
(Ordering Strategies) that would carry out the linearisation of microstructural propo-
sitions:
• The simplest strategy involves respecting, in the proposition enunciation order,
the natural world order. Thus, the description of processes, events or actions fol-
lows a Temporal or Conditional Order (for example: ‘He comes to the car, -> he
opens the door, -> he sits down and -> starts’).
• Comparatively, the description of objects or persons most often follow an order

which is from more general to particular, from higher to lower, from sets to
elements, etc. (Categorising Order: ‘The car is red and big. -> It has four seats. -
> On the first seat, -> there is a road map. -> It is blue.’).
• van Dijk and Kintsch (1983) equally describe another organisation principle,
slightly different from the two preceding ones. This principle is, when knowl-
edge is not linear (as in the case of a conceptual network portion constituted of
several nodes connected by different bonds), to follow each network branch by
‘jumping’ or ‘going’ from one to the other. In this case, ‘the natural order of the
world’ would guide the exploration of the network less than (a) the structure of
the network and (b) the limited Short Term Memory span, which would not al-
low to apprehend all the network nodes (Cf., on this point, Levelt, 1981, 1982).
According to van Dijk and Kintsch (1983), these different linearisation strategies,
respecting a natural and/or functional order, can be used unless pragmatic, rhetori-
cal, cognitive or communicative constraints or decisions necessitate another order.
These deviations or these transformations are common and are insured by at least
four different ‘re-ordering’ strategies:
• Cognitive Reordering: According to this strategy, the information order in the
sentence, the text or the discourse does not depend on the world order, but on
the manner the speaker or writer perceives, interpret, or remember the world.
The utterances built via this strategy would equally comprise, in their surface,
some expressing verbs such as ‘to see’, ‘to think’, to ‘remember’, etc.
• Pragmatic Reordering: This information organising strategy amplifies the
pragmatic scope of an utterance or a message by positioning, for example, in the
beginning of the sentence, a request: ‘Can you please close the window?’ before
stating the cause of this request: ‘It is cold in here!’. In this example, the order
from cause to effect is inverted and the request is first produced, thus allowing
to strengthen the Speech Act effect, or the communicative effectiveness.
• Rhetorical Reordering: Rhetorical Reordering strategies are close to Pragmatic
Strategies but they have less of a communicative objective than aesthetic crite-
ria or styles to emphasise some information. For example, the sentence ‘It is a
frog that possesses delicious thighs’ can be re-elaborated into ‘These delicious
thighs belong to a frog’ to emphasise the edible quality of the inferior animal
members.
• Interactional Reordering. These strategies are largely linked to pragmatic strate-
gies but in a communicative and interactional context. For example, depending
on the properties of communication and interactional context, it is possible to
delay some information to respect principles of politeness, co-operation, etc. (it
is better to say ‘hello’ to friends, to worry about their health, before asking them
to lend some money, even if it concerns the main objective of the Speech
Act…).
These four reordering strategies are interesting but it seems difficult to clarify the
writer’s necessary strategic choices concerning the ‘semantic colouring’ of the mes-
sage. Indeed, it is not simple to confine what kind of factors lead to modify ‘the
natural order of the world’, and to accentuate one knowledge unit rather than an-
other. It is possible that the need to take the addressee and other discourse pragmatic
constraints into account (for example: to convince, to describe, to explain, to tell,
etc.) certainly plays an important role in these choices. However, current writing
models provide little information about the modalities of their influence and, more
globally, on the determinism of semantic choice. Van Dijk and Kintsch’s (1983)
proposals are speculative and the authors themselves wrote that
‘it goes without saying that both the normal ordering strategies and their systematic re-
orderings (…) need further specification, discourse analysis and experimental verifica-
tion.’ (p. 277).
One possible way of exploration of linearisation processes and application of a par-

ticular ordering strategy can be to determine and to experimentally vary the main pa-
rameters able to influence the choice of semantic strategies. The determination of
these semantic strategies could be influenced or could depend on several parameters,
such as:
• The text type: some text types could lead to more particular use of some seman-
tic strategies rather than others. Thus, in the case of narratives, it is possible to
create specific suspense or expectation situations in the reader by first present-
ing the description of a character before proposing her/his identification (for ex-
ample: ‘small but good swimmer thanks to its strongly muscular thighs, the frog
treats French people, this is told abroad’ – Rhetorical Reordering). In the case of
argumentative texts, thematisation can be used to accentuate the most important
arguments to convince the text addressee (for example: ‘Thighs characterise the
frog…’ instead of ‘the frog is characterised by its thighs…’ – Pragmatic Reor-
dering).
• The addressee type: the familiarity level between the writer (or the speaker) and
the addressee can, for example, lead to the use of different Pragmatic and/or In-
teractional reordering strategies, when it is or not necessary to respect conven-
tional rules, etc.
• The prior structure of the domain knowledge: this parameter can offer more or
less possibility to the semantic reordering of concepts, knowledge units or
propositions. Thus, a knowledge structure recording a temporal succession of
events or actions (for example: stages of starting and driving a car) can be reor-
ganised with difficulty for fear of the introduction of incoherence in this knowl-
edge structure (one can not drive a car without having started it). Conversely, a
mental image recording spatial information, for example, could eventually be
described without a fixed order (the description of a car can begin with the exte-
rior as well as by the wheels if particular risk at a rhetorical or pragmatic level
does not exist).
9.2.3 Conclusion: the elaboration of a preverbal message

The proposition linearisation stage, during Translating, allows to determine the mi-
crostructural propositions order as well as (and both are linked) their semantic func-
tions. According to Levelt (1989) or Bock (1995), the linearised content can be de-
fined as a ‘preverbal message’. Bock (1982) states that this semantic product plays a
central role in the verbal message generation, because it constitutes ‘an interfacing
representation’ between the thought (concepts) and the language (i.e., the transfor-
mation of concepts into words). When the preverbal message has been linearised
and semantically processed, it must be linguistically translated. The linguistic trans-
lation process is specifically devoted to the text (or discourse) formulation process-
ing. This process consists, according to Fayol and Schneuwly (1987), of (1) drafting
a grammatical and lexical plan, in order to translate the semantic content previously
drafted, while respecting the communicative textual goal. The product of this plan is
then (2) executed at a graphomotoric level.
9.3 Stage three: formulation of sentences from a preverbal message

During Formulation (that is to say the linguistic aspects of Translating), the writer
has available both the micro- (i.e., propositions and their semantic relationships) and
the macrostructural text units (i.e., each part of the text plan or each main point).
Thus an elaborate semantic representation of the text or the portion of text would al-
ready exist, just before its linguistic formulation. The problem is then to formulate
this textbase (Kintsch, 1987) or the preverbal message. In order to do so, it is neces-
sary – if the claim that thought has no linguistic form is accepted (Cf. above Hayes
& Flower, 1980) – to match preverbal message elements with linguistic elements. In
other words, the writer has to match a syntactico-semantic structure with grammati-
cal and lexical structures.
9.3.1 Limits of general writing models

As previously mentioned in the General Introduction, general writing models remain
relatively unclear concerning the formulation of sentences from a preverbal message
within the Translating process. The linguistic processing modalities are briefly men-
tioned and the principles of the passage from a conceptual to a linguistic structure,
through the choice of grammatical and lexical components, are quickly and very
globally described. That is the case, for instance, in Bereiter and Scardamalia’s
(1987, with the ‘Write’ operation), Kellogg’s (1996, through the Translating proc-
ess), or Hayes’ (1996, with the Text Production process) models, in which ‘the for-
mulation of ideas into words’ is too often rapidly evoked. These different models do
not therefore possess a sufficiently strong predictive power, which would allow re-
searchers to build claims or hypotheses. This relative ‘negligence’ of general writing
models is, without any doubt, linked to two facts. First, as Hayes (1996) has noticed,
the interest of experimental works, since the appearance of the initial Hayes and
Flower’s (1980) model has (too) strongly focused on the study of the Planning proc-
ess, certainly to the detriment of the Translating process. Secondly, the functioning
of this last process has been frequently defined as largely automatised in the writer’s
behaviour (at least, in expert writers: Cf. Chapter 6) and, consequently, considered
as having only a very little influence on processing and resource management, in

comparison to the Planning process.
9.3.2 Formulation in general speaking models

Conversely, in production models describing the activity of speaking, more precise
principles are described. It is the case, for example, in Levelt’s (1989) or
Grabowski’s (1996) models. This last model can however be considered as a mixed
general model, which concerns both writing and speaking.
Levelt’s (1989) model: the Formulator and the concept of Lemmas. Levelt (1989),
through the ‘Formulator’ module (equally labelled ‘Encoding’ by van der Pool, 1995
or by van Wijk, 1999), paves an interesting theoretical way to take into account
grammatical and lexical processes. According to this author, the entry product (In-
put) of the Formulator is made up of preverbal messages or fragments of preverbal
messages, which come from the ‘Conceptualiser’ (Cf. Figure 4, page 12). The For-
mulator only regards grammatical and lexical translations of a conceptualised mes-
sage. The propositional linearisation processing (Cf. supra) is consequently entirely
executed by the ‘Conceptualiser’. The linguistic formulation of these messages is
performed by the matching of linearised knowledge units to lexical units, called
‘Lemmas’, according to the initial definition of Kempen and Huijbers (1983). A
Lemma lists information related to the meaning of a word and to its syntactic envi-
ronment. On the basis of this information, the grammatical plan is subsequently
elaborated (Cf. also: Levelt, 1992; Levelt, Schriefers, Vorberg, Meyer, Pechmann, &
Havinga 1991). After this stage, the phonological encoding of the lemmas into lex-
emes is realised by computing a phonetic program appropriate to the item in the dis-
course context (Levelt, 1992).
Grabowski’s (1996) model: auxiliary systems to prepare a preverbal message to be

formulated. The formulation stage has been described slightly differently in the
model proposed by Grabowski (1996, Cf. Chapter 4 for an in-depth description of
this model and Figure 24, page 155). Grabowski, although strongly inspired by
Levelt (1989), does not integrate semantic processing within a Conceptualiser, but
gives them a specific place in the sequence of the linguistic production process,
within a process that he calls the Auxiliary Systems.
More precisely, the first steps of Translating, such as defined by Grabowski
(1996), are relatively close to those defined by Levelt (1989) for oral language. A
preverbal message, labelled the ‘Proto-input’, is elaborated by a ‘Central Control’
module in which three processes performed the Selection, the (re)Construction and
the Linearisation (Ordering) of information, retrieved from the environment or from
Long Term Memory. However, this message cannot be directly linguistically formu-
lated before a specifically linguistic processing executed by the Auxiliary Systems.
These systems have, among others, the function to prepare the proto-message for its
linguistic formulation, by the apposition of different semantic markers. These mark-

ers represent instructions, which carry out, by a deep semantic analysis of the proto-
message, the coherence, the perspective, the modal choices, the emphasis, etc. The
representation from the Auxiliary Systems (the Encoding input), is then processed
by the Encoding mechanism, whose function is to transform this Encoding Input
into a phonemic or graphemic sequence that could lead to the execution of a phono-
logical or graphomotoric sequence. The different kinds of processing carried out by
these encoding mechanisms mainly involve defining the grammatical structure of
the message (application of grammatical schemas) and completing it with segmenta-
tion or emphasis indication (in the case of oral language).
This model can be considered as generic because it can be applied both to oral
and written production. Grabowski (1996) has nevertheless added a differential
process in order to capture the differences of the two media. Thus, in the case of
writing, the ‘proto-input’ message sometimes needs specific processes/processing in
the Auxiliary Systems, just because the written encoding does not offer the same
possibility as the oral encoding to mark, for example, emphasis. The markers and
modalities of marking therefore differ between the media. For example, an oral em-
phasis can be converted by the semantic elaboration of a key sentence, by specific
punctuation marks, or by particular typographic marks (bold, italics, etc.). Besides
the specific sequencing of processing levels, Grabowski’s model also explains the
control and the regulation of processes. This aspect of his model is developed in
Chapter 4.
As previously described, it seems that general speaking models are more precise
than general writing models, concerning the description of processing linked to lin-
guistic formulation. Nevertheless, in both models, the approach is too descriptive
and not procedural enough. Linguistic translation criteria and conditions are rarely
clarified. Analysing the formalisations proposed by local models of oral sentence
production could nevertheless fulfil this relative absence of procedural formalisation
about linguistic formulation.
9.3.3 Local models of sentence production and formulation processing

Four local models of oral sentence production are briefly described. Two of them are
relatively old and have adopted very different perspectives – Garrett (1975, 1980,
1984)’s and Dell (1986)’s models. The other more recent models, have been elabo-
rated by Bock and Levelt (1994) and by Bock (1995, 1996), following Garrett’s
works.
Garrett’s (1975, 1980, 1984) model. For oral language, Garrett has elaborated a
model of sentence generation, mainly by studying the nature of errors in natural and
spontaneous discourse. He remarks that errors are mainly linked to exchanges and
substitutions of (1) sounds or morphemes and (2) words. The analysis of the location
of these two sets of errors shows that the environment of their respective occurrence
differs according to grammatical categories, phonological similarity and distances

between items in the sentence. Thus, phonological and morphological errors would
be more frequent when the items: (a) do not belong to the same grammatical cate-
gory, (b) are phonologically similar and (c) close within the sentence. Conversely,
errors on words would be more frequent between two items (a) belonging to the
same grammatical category, (b) that do not possess any phonological similarity and
(c) that are far from each other in the same sentence. According to Garrett (1980),
the origin of these errors would come from the simultaneous activation in Working
Memory of these items. This simultaneous activation would be explained by the fact
that these items belong to the same processing level. This hypothesis has led Garrett
(1984) to elaborate a model of sentence production that comprises two processing
levels.
The first level, labelled ‘Functional level’, operates on lexical items, semantic
(agent/patient structure, for example) and syntactic (as the choice of a grammatical
category) elements. At the end of this level, the syntactic structure of the sentence is
defined and the appropriate lexical items are selected. The errors of lexical substitu-
tion could then appear during the elaboration of this functional representation. In-
deed, these errors are sensitive, as seen previously, to the grammatical structure and
to the distance between items within the same sentence.
Once the Functional level is complete, the elaborated functional representation is
considered as an element for the next processing level, the ‘Positional level’. This
level ensures the phonological determination, the order of enunciation and the mor-
phological marking of lexical items. It would be during the execution of these proc-
esses that phonological and morphological errors could appear (i.e., these errors are
linked to the phonological and temporal proximity of items).
Garrett’s (1980) model clearly relies on a structural framework, with principles
of modular processes, which is to say, autonomous and non-interactive (Cf. Fodor,
1983). Dell (1986) has developed another proposal about the nature of processes for
sentence production. His model, opposed to Garretts, relies on functionalist princi-
ples, based on interactionist modalities of processes, thus relatively close to connec-
tionism theories (Cf. Chapter 4, for a discussion on modularity and process interac-
tivity).
Dell’s (1986) model. Dell’s (1986) objective was to explain, through a predictive
model, the passage from the meaning to the sound, during sentence production. Dell
(1986) has mainly focused on syntactic aspects and on phonological and morpho-
logical encoding. His main goal is to interpret, with an approach different from
Garrett (1980), language errors, such as ‘slips of the tongue’, or substitution and ex-
change errors of phonemes or words. Dell’s (1986: 290) model is relatively complex
concerning its functioning, but the described processing modalities are simple in
their principles. The originality of the author’s approach is to adapt the spreading ac-
tivation principle to sentence production in a conceptual network, earlier developed
by Anderson and Bower (1973) and by Lindsay and Norman (1972) to explain the
functioning of Long Term and Working Memory (Cf. Chapters 4 and 5 for further
details).
According to Dell (1986), the sentence production (i.e., the passage from a se-
mantic representation to an oral discourse, via syntactic, lexical and phonological
structures) would be made through numerous activations in a hierarchised network
with three levels: first a lexical level composed of lemmas (higher level), a morpho-
logical level (intermediate level), and a phonological level (lower level). The spread-
ing activation between the nodes of the different levels is considered as top-down
and bottom-up; moreover, the activation of a specific node by another one would not
operate as all or nothing, but would follow a continuum (Cf. Figure 15).
TACTIC FRAMES LEXICAL NETWORK
1 c 2
S Some Swimmer Sink Drown

NP VP 3
N
O N Plural V Plural Swim
(1) (2) (3) ?
SYNTAX
1 c
Word Word
Stem Some Swim er Plural Sink

So Sw Af1 Af2 Sv
So Sv Af1 Af2
(1) ?
MORPHOLOGY
SYL Sw A
On Nu
Rime
On Nu Co s w i m
? On On Nu Co
PHONOLOGY
Figure 15: Sentence production and activation spreading,

adapted from Dell (1986).Copyright © 1986 by the American Psychological Association.
According to these principles, a lexical item (that can be, for example, initially acti-
vated by the preverbal semantic message) can activate a set of morphological and/or
phonological components with which it can be associated. For Dell (1986), this
principle of spreading activation in a network can explain the specific relationships
between nodes of different levels. In addition, the activation pattern of the different
nodes is not sufficient, as such, to create the sentence. It is necessary to decide what
nodes, among the activated set, will be effectively used to elaborate a linguistic rep-
resentation of the sentence. The elaboration of this representation relies on two types
of rules: (1) the ‘Categorical Rules’, which build Frames of Syntactic, Morphologi-
cal and Phonological representations, and (2) the ‘Insertion Rules’ that specify what
nodes can be effectively inserted in the slots of these three Frames. Two Tags (or
markers), affixed on the nodes, allow this matching. Continuously and intrinsically,
each node is identified in the network, by a Tag that signals its category (Name,
Verb, Quantifier for syntactic categories, Stem, Affix for morphological categories,
etc.). This ‘Categorical Tag’ allows all the nodes of the same category to postulate
as potential candidates (if they are sufficiently activated) to be inserted in one of the
slots of the Frame corresponding to this category. The effective matching of a struc-
ture slot (Tactic Frame) with one of these candidate nodes operates through a second
contextual marking (‘Insertion Tag’) which therefore appoints the retained node to
the slot of the structure.
This model, in its main principles, suggests that the elaboration of the different
frames through Categorical and Insertion Rules operates in parallel and in close in-
teraction with the activation of the nodes of different levels. The different linguistic
errors, susceptible to appear during the interaction between the three frames, can re-
sult from a problem in the elaboration of the frames, or be linked to some bad
matching between nodes and slots of frames, which are actually correct (phoneme
anticipation, phoneme exchange, word substitution, etc.). In this last case, the errors
would be mainly linked to competition phenomena between several activated nodes.
For example, asking someone to ‘close the door’ can presuppose to have observed
that the door is opened. The lexical item ‘to open’ can thus be more activated than
the lexical item ‘to close’ and will be preferably matched in the slot of the syntactic
frame instead of ‘to close’ (Dell, 1986: 291).
Bock and Levelt’s (1994) model. Bock and Levelt’s (1994) model, aiming at
explaining the transformation of a preverbal message into a verbal message, is
strongly inspired by Garrett’s works (1980, 1982, 1988). It equally concerns a more
precise description of the functioning of the Formulator, previously described by
Levelt (1989). In their sentence production model, Bock and Levelt (1994)
distinguish four processing levels (Cf. Figure 16):
• The processing carried out at the ‘Message level’ seems relatively comparable
to those executed by the Conceptualiser in Levelt’s (1989) more general model.
The objective of this processing level is to record the conceptual content and the
speaker’s communicative intentions in a preverbal message, providing semantic
information for the next ‘Functional and Positional levels’ processing.
• The Functional level (Functional processing) is composed of two processes:
o the ‘Lexical Selection’ process identifies lexical concepts allowing
to translate into language the concepts contained in the message.
This identification is carried out through Lemmas that contain
grammatical information associated with lexical concepts (name,
verb, gender, preposition, adverb, etc.).
o the ‘Function Assignment’ process determines syntactic functions

and grammatical roles of each lexical concept. While the lemmas
circumscribe the grammatical nature of lexical concepts, the proc-
ess of Function Assignment appoints grammatical functions and
relationships to this lexical concept (for example: subject-
nominative or object-dative).
• The Positional level (Positional processing) gives the order of words within a
sentence. It creates a hierarchical schematic structure that determines and con-
trols:
o the position of words in the future sentence (via the process called
‘Constituent Assembly’),
o the morphology of these words (via the ‘Inflection’ process).
The elaboration of this hierarchical control structure relies on the interpretation
of functional clues (nature and grammatical function of each lexical concept).
The inflections are positioned at the end of the hierarchical structure, they belong
to a fine analysis of the constitution of a word (stem and affix in the case of a
verb, for example).
• The Phonological encoding attributes its phonological segmentation and pros-
ody to the future sentence.
MESSAGE
FUNCTIONAL
Lexical Function
Selection Assignment
PROCESSING
Grammatical
Encoding
POSITIONAL
Constituent Inflection
Assembly
PROCESSIN
PHONOLOGICAL
ENCODING
To output
Figure 16: Model of language production processes, adapted from Bock and Levelt (1994).
Copyright © 1994 by Academic Press. Adapted with permission.
According to the authors, the attribution of grammatical functions, important for the
structure of the future sentence, depends on two parameters, that are processed and
contained in the preverbal message: the ‘Event Roles’ and the ‘Attentional Roles’:
• the ‘Event Roles’ are functions or semantic instructions that enable determining
conceptual notions as the agent of an action, the patient or the theme, the bene-
ficiary, the instrument, the temporality, etc. These semantic notions can be
summarised in a more general way by means of primitive semantics (Bock and
Warren, 1985; Jackendorff, 1987), as notions of animacy or concreteness. The
influence of these event roles on the functional determination of the message
would be relatively direct notably because the agent of an action is often syntac-
tically encoded as the subject noun of the sentence and the same rule is applied
for a concrete and/or animate concept;
• the ‘Attentional Roles’ emphasise a concept compared to other concepts. Se-
mantic structures, like Topic-Comment or Given-New, carry out such a
perspective. Frequently, the concept benefiting from such an attentional focus,
labelled ‘Conceptual accessibility’ by Bock and Warren (1985), would more
likely have a function of subject in the sentence.
Finally, in Bock and Levelt’s (1994) model, the temporal sequencing of the different
processing levels is considered as incremental. This principle, illustrated by the dif-
ferent positions of arrows in Figure 16, supposes that the sentence is not globally
processed at each level, but that different and successive portions of the same sen-
tence can be processed in parallel in each of the four levels. This particular mode of
processing will be developed in Chapter 4.
Bock’s (1995, 1996) model. Bock (1995, 1996) proposes a model of oral production
of sentences (Cf. her schema p.185 for 1995 and Figure 1 p. 396 for 1996) that is
relatively close to Levelt’s (1989) Formulator and to Bock and Levelt’s (1994)
model. This model comprises three components that respectively designate (1) the
elaboration of a non-verbal message (what the speaker wishes to communicate, de-
pending on the meaning to be conveyed and its communicative aspect), (2) the draft-
ing of the grammatical message, and finally (3) the formulation of the sound of the
words (phonology and prosody: phonological encoder). Thus, as in the Bock and
Levelt’s (1994) or Levelt’s (1989) models, before verbal production, an initial non-
linguistic representation would exist that has to be elaborated lexically, syntactically
and phonologically during linguistic activity (Bock, 1996: 416). More precisely:
• the ‘Message component’ creates a non-verbal message that represents what the
speaker wishes to communicate, according to the meaning and to the communi-
cative goals;
• the ‘Grammatical component’ is a very central component for Bock (‘heart of
language production’, 1995: 184). It selects the semantically appropriate words
(i.e., identification of lexical entry – the ‘lemmas’ –, assignation of grammatical
roles) and integrates these words into an adequate syntactic structure. This
therefore allows (1) the lexical recovery, (2) the creation of a group of hierar-
chically structured words, and (3) the transformation of meaning into sound.
This component refers to the grammatical encoding in Levelt’s (1989) model.

More precisely, according to Bock, this component allows two processing
types: (1) a ‘Functional processing’ that associates elements of the message
both with grammatical functions and entries of the mental lexicon (i.e., identifi-
cation of lexical entries [the lemmas] and allocation of the grammatical roles to
the different phrases); (2) a ‘Positional processing’, that associates grammatical
functions and lexical entries to the structures of sentences and words (i.e., lexi-
cal recovery and creation of a set of hierarchically structured words);
• the ‘Phonological component’ is responsible for the execution of the message,
phonologically and prosodically.
In parallel with these three components, Bock (1995) postulates the existence of dif-
ferent exit (output) systems that would guide the production of utterances. She adds
to her model a process (Monitoring) that controls the output of the production sys-
tem and initiates eventual corrections. Thus, just as in Levelt’s (1989) model, there
would be a control-comprehension system that would allow, before the message
would be phonologically produced, an internal revision, leading to detect semantic
or grammatical errors.
9.3.4 Conclusion: local and general models of language production

Local models propose a relatively fine analysis of linguistic translation processes
that strongly contrasts with the vagueness of most writing models. Nevertheless, it
seems evident that a certain number of relations (or comparisons) can a priori be es-
tablished between local and general models.
For example, the principles and the rules described by Bock (1995, 1996) are
very close to those proposed by Levelt (1989) through the Formulator. Thus, Bock’s
model can easily be considered as a particular step in Levelt’s general model. How-
ever, Bock considers the production of sentences and not the production of dis-
course. This explains why Bock (1995) more strongly emphasises the grammatical
component, compared to Levelt (1989).
In the framework of writing models, while it is possible to postulate the exis-
tence of this same component during the writing process, it is worrying to observe
that such processes have never been precisely described. Even if it always appears
difficult to operate direct matching between general models and local models of sen-
tence generation, local models have a certain number of characteristics or principles
that could be relevant to consider (and eventually to test) in the framework of text
production. For example, matching processes between a preverbal message and
some ‘lemmas’ can be viewed as a constant rule in all models of oral sentence gen-
eration. The concept of lemmas, used both by Bock (1995, 1996), Bock and Levelt
(1994), Dell (1986) or Levelt (1989) could thus constitute an interesting theoretical
track to explain the linearisation of contents during writing.
In the same way, the four previously developed models offer an interesting per-
spective for the elaboration of grammatical structure and concerning the constraints
that load this elaboration. In the case of Bock’s (1995) and Garrett’s (1984) models,
linguistic processes are always considered through two levels: one functional and
the other positional. Is it possible to postulate the existence of these two levels in the
case of text writing? In the same way as for oral language, would a poor functioning
of one of these two levels lead to different kinds of errors? It is interesting to note
that such a qualitative analysis and such an interpretation of production errors has
been proposed for writing by Hotopf (1983), according to principles and procedures
relatively similar to those used by Garrett (1984). Hotopf made some differences
apparent concerning the nature of errors for speaking and for writing. These differ-
ences show that graphemic processes would be carried out during and after phono-
logical processes. This example illustrates the possible relationships, as experimen-
tal and theoretical, between speaking models and writing models, on the one hand
and between general models and local models, on the other hand.
Concerning the nature of processing, Dell’s (1986) model is clearly opposed to
Garrett’s (1984) and Bock’s (1995) proposals, because Dell has adopted some of the
principles of activation and inhibition, parallel processes, and there are no process-
ing levels in his model. Why can such parallel processing principles not be tested in
writing, to give an account of linguistic translation? Would the fact that, compara-
tive to the Planning process, the Translating process requires only few cognitive re-
sources, be linked to the existence of such modalities of processes, as proposed by
Dell (1986), allowing a rapid and costless processing? Besides its options of proc-
essing, another interest of Dell’s (1986) model involves the explanation of the rules
of syntactic structure elaboration: the ‘Categorical Rules’ and the ‘Insertion Rules’.
Again, could these rules be generalised to the functioning of the linguistic transla-
tion during writing?
Finally, another example of the interest of local production models, in the study
of general writing models, can be developed in the framework of Bock and Levelt’s
(1994) model. Although this model focuses first and foremost on the grammatical
production of a single statement, it appears to be one of the easiest to adapt to the
broader constraints of text production. Indeed, the essential interest of Bock and
Levelt’s (1994) model relies less on the description of linguistic formulation proc-
esses than on the analysis proposed on the determination of Function Assignment
processes by a certain number of semantic clues, which can constrain the lexical and
syntactic elaboration of a sentence. These clues (the Event Roles and the Attentional
Roles), compared to van Dijk and Kintsch’s (1983) Semantic Strategies of Order,
can be assimilated to processing instructions that must be interpreted for lexical or
grammatical components in order to linguistically indicate the semantic characteris-
tics of the message. Thus, the ‘natural order’ dependence, described by van Dijk and
Kintsch (1983, Cf. supra), is relatively close to that postulated by Bock and Levelt
(1994) between the Event Roles and the attribution of roles and grammatical func-
tions to concepts (Cf. above). The correspondence between semantic structure and
syntactic structure has however to be modulated, because (1) all languages do not
benefit from this characteristic in the same way (that would be particularly pregnant
in English), and (2) all sentences are not simple and direct, especially in expert
speakers or writers. In the case of complex sentence processing, it is necessary to
describe and to include in sentence production models some operations of comple-

mentary syntactic transformations. These operations would allow elaborating sen-
tence structures that do not inevitably correspond, word for word, to the semantic
structure, as it is the case, for example, of the passive voice. These grammatical op-
erations would however be applied only in an additional manner and probably under
strong specific pragmatic or rhetorical constraints (to adopt a particular linguistic
style, for example). However, it remains to determine to what extent the activity of
such grammatical transformations, for the production of one unique sentence, can be
generalised to the production of a text composed of a succession of sentences whose
syntactic structures are often varied and embedded.
These different remarks, concerning the possible integration of some elements of lo-
cal models of sentence generation in text writing models, constitute only some
tracks of reflection for future research. Such integration must first be justified, both
theoretically and empirically, and has a meaning only if it allows to elaborate new
hypotheses in the framework of text writing research. Nevertheless, it is clear that
these local models represent a source of particularly interesting and relevant data. It
remains now to evaluate how models that describe the generation of an isolated sen-
tence, although in the course of discourse production, can be relevant compared to
models that describe the production of several written sentences, linked together,
both semantically and syntactically. We will focus on these aspects at the end of this
chapter.
9.4 Stage four: graphic execution

Following the linguistic formulation step, the writer must effectively transcribe the
phrase, the sentence or the text that has been processed. The graphic transcription
can be defined as the programming and the execution of necessary motor move-
ments to produce a letter, a word or a text (Brown, McDonald, Brown and Carr,
1988). There are several possibilities to transcribe linguistic information. The most
classical one is to write the message, called handwriting. Nowadays, this execution
modality is frequently replaced, even by young or inexperienced writers, by writing
with a computer, called typewriting.
9.4.1 Graphic execution in general writing models

Frequently mentioned in general models of text production, the level of execution is
however rarely clearly defined, mainly because this step of Translating is frequently
considered as automatic.
Thus, despite its importance within the writing activity and the fact that it is nev-
ertheless the final writing stage, the Execution process, necessary for the physical
appearance of the message, has not been studied intensively in the area of writing
and is rarely defined and described in initial models of writing (for example: Bere-
iter and Scardamalia, 1987; Hayes and Flower, 1980). Berninger and her collabora-
tors have underlined, in this vein, that it was possible and important to integrate, in
the general Translating process, the processing concerning the graphic ‘Transcrip-
tion’ of the planned and formulated text segment (Cf. Berninger & Swanson, 1994;
Swanson & Berninger, 1996). Berninger and her collaborators (Berninger, 1994;
Berninger, Füller, & Whitaker, 1996; Berninger & Swanson, 1994; Berninger,
Whitaker, Feng, Swanson, & Abbott, 1996; Swanson & Berninger, 1996) propose to
complete Hayes and Flower’s (1980) model to specify the architecture of the Trans-
lating process, more particularly in order to explain the development of writing ex-
pertise (Cf. Chapter 6 for a more precise description of their proposals). For the au-
thors, such a ‘Transcription’ sub-process leads to the translation of linguistic repre-
sentations into the form of symbols written on the sheet of paper. Furthermore, the
‘Transcription’ processes phonological and orthographic coding and fine motor
skills.
On this last point, Kellogg (1987b), or more recently Graham (1990), have
shown that activities implied in making the message physical apparent, by consum-
ing a non-negligible part of cognitive resources, could be carried out to the detriment
of high-level writing processes (Cf. also Fayol, 1999). Such a consideration about
the influence of graphic execution appears more clearly today by means of a re-
focussing of new writing models to the transcription process or, more generally, of
execution of the message in various codes.
Hayes (1996: 13), for example, in his new model, presents Execution as a final
stage of the Text Production process, and distinguishes several possible modalities
to ‘write’ a text – handwriting, schemas or graphs, oral speech. Hayes (1996) also
distinguishes a mental execution and an effective execution. He focuses on the fact
that the effective execution of handwriting could be preceded by the mental execu-
tion (sub-vocal), perhaps effective (vocal), of an oral message.
A more precise description and explanation of the mode of functioning of the
Execution process, integrated within a general architecture of writing processes, has
been recently proposed by Kellogg (1996). In his model, the author clearly notices a
component of execution and distinguishes, as Hayes (1996), a mental and an effec-
tive realisation. These two stages are respectively carried out by the Programming
(planning the programming of the output) and Executing (performing the effective
execution of the programming) processes (Cf. Figure 7, page 20). The Execution
component occupies the same place as the components of Formulation (with the two
processes of Planning and Translating) and Monitoring (which is closer, with its
constitutive processes of Reading and Editing, to the Revising process as described
by Hayes and Flower, 1980; Cf. General Introduction). In Kellogg’s (1996) frame-
work, the important place given to the Execution, allows the author, in agreement
with Hayes’ (1996) recent proposals, to make a certain number of precise hypothe-
ses about the functioning of the different writing processes, by considering different
execution modes (oral dictation, handwriting, typewriting).
It seems therefore that the characteristics of the graphic output (handwriting or

typewriting, for example) are now integrated in the most recent writing models, thus
allowing to reconsider older works that have led to evaluate the cost of handwriting
comparatively to typewriting, at least in expert writers (Cf., for example, Norman &
Rumelhart, 1983). However, although the recent Hayes’ (1996) and Kellogg’s
(1996) models give a more important place to processes linked to execution, the au-
thors only propose a very global description of this activity.
9.4.2 Graphic transcription in local models

Actually, this fact is surprising. During the last 15 years, an important increase in the
amount of developmental research about graphic skills and processes has been ob-
served (Cf. van Galen & Teulings, 1983; Meulenbroek & van Galen, 1988; van
Galen, 1990; van den Plaats & van Galen, 1990; Zesiger, 1995). These works have
led to elaborate relatively complex procedural and developmental models about gra-
phomotoric processes, often related to neuropsychological data. For example, the
model of graphic activity, elaborated by van den Plaats and van Galen (1990), relies
on the principle according to which the real time progression and management of the
graphic transcription would be hierarchically ordered and would follow different
levels. The release of the transcription processing would first activate a lexical proc-
essor and this processor would provide abstract graphemic representations. These
representations would be put, as soon as they are activated, in a graphemic buffer in
memory. Then, an allographic processor would retrieve, letter by letter, the repre-
sentations stored in the graphemic buffer, and would translate them into allographic
codes, stored in a short-term motor buffer. The allographic codes would then be re-
trieved and would release the different motor programs necessary to write letters.
Thus, this short presentation of van Galen’s (1991) model shows to what extent
works dedicated to graphic processes are precise and predictive. Nevertheless, as we
previously discussed for local models of sentence production, they rarely take into
account the textual dimension, they just analyse the necessary motor skills and
movements to execute a letter or a word. In addition, a detailed study of the top-
down effects from the highest levels on the graphomotoric execution does not seem
to exist (Cf. nevertheless Fayol, 1999). In fact, it seems that writing research is more
strongly focused on the three (considered as) main writing processes (Planning,
Translating, Reviewing) than on the graphic transcription, that is however necessary
for the materialisation of the message. In the same way, works concerning with the
graphic transcription are led ‘separately’, that is to say outside of the area of text
writing (Cf. for two complementary discussions Dobbie and Askov, 1995; Graham
and Weintraub, 1996). Currently, we can only, once again, speculate, as in the case
of sentence formulation, the possibility to integrate these local models of graphomo-
toric execution into more general models of written production. It is only possible to
deplore the absence of interest for this process, nevertheless fundamental for the ap-
pearance of the message. Two reasons can explain this lack of interest: (1) the
handwriting is automatised in adult writers and writing models mainly concern the
production of expert writers; (2) the interest for high-level processes of writing, as
Planning and consequently the little interest, until now, for the whole process of
Translating (Cf. above).
9.5 Conclusion on the four stages of Translating

The Translating process can be divided into several steps in the complex transforma-
tion of units of activated, selected and organised knowledge, into a linear linguistic
trace. Four processing stages have been distinguished through the Elaboration, the
Linearisation, the Formulation and the Execution of the message. It appears hence-
forth possible, as we have done, to isolate each of these processing levels to clarify
their respective functioning. However, it remains to determine how sub-processes or
operations, associated with each process stage, could be linked, in a dynamic man-
ner, to allow the writer to translate semantic propositions in sentences during the
general process of writing. The interest for the dynamics of the Translating process
requires definition of (1) the modalities of recursion and processing cycles (the text
being formulated part by part) and thus (2) the nature of the units that will be proc-
essed during one cycle.
10 TRANSLATING RECURSION AND UNITS OF PROCESSING

If Hayes and Flower’s (1980, Cf. Figure 14 page 68) proposals are extended, the
production of a sentence would be carried out under the control of two information
sources. The first source is external and consists of the already written trace (the
sentence or the portion of the previously written sentence – Text Produced so far).
The second one is internal: it concerns the portion of the text plan elaborated by the
Planning process. Elaboration, Linearisation and Formulation processing would then
continue, recursively, on proposition parts (elaborated and linearised at a semantic
level) and on sentence parts (formulated at a linguistic level). The problem of the
Translating dynamics then concerns, during writing, the determination of: (1) pro-
positional contents that will allow to coherently continue the idea that has been for-
mulated, while respecting the topic or the subtopic fixed by the plan portion, as well
as (2) linguistic forms that will allow to formulate these propositional contents while
maintaining a cohesion with the previously written linguistic forms. The nature of
Translating processing and the principle of its recursion have been particularly well
described by van Dijk and Kintsch (1983) through the ‘Local Coherence Strategy’.
10.1 Local coherence strategy

According to van Dijk and Kintsch (1983), and, similarly to Hayes and Flower’s
(1980) principles, the choice for the modalities of linguistic formulation of a specific
semantic proposition would be carried out with two main constraints: (1) the con-
straint of surface clues that belong to the previously written sentence segment (lexi-
cal items, connectives, punctuation marks, etc.), and (2) the constraint of the macro-
structural proposition that controls the retrieval and the elaboration of the semantic
proposition (one part of the text plan, for example). Opposed to Hayes and Flower
(1980), van Dijk and Kintsch (1983) consider that the set of these processes are
equally executed under (3) a third constraint linked to the Short Term Memory span:
the different parts of the preverbal or propositional message would be temporarily
maintained in Short Term Memory, while waiting for their formulation (Cf. Levelt,
1989, on this last point). These three constraints (the previously written sentence
segment, the part of text plan and the Short Term Memory span) constitute the basis
of the Local Coherence Strategy (van Dijk and Kintsch, 1983: 278). According to
this strategy, the linguistic formulation would be executed on the basis of grammati-
cal clues included in the previously translated sentence (or part of the sentence) and
under the semantic control of the macrostructural proposition. More precisely, the
macrostructural proposition allows to choose, among all candidate microstructural
propositions, those that will maintain in the best way the local coherence of the sen-
tence (i.e., finding the most coherent continuation of the sentence, by respecting the
topic or the subtopic defined by the plan or a part of the plan). Grammatical forms
and lexical items, necessary to effectively express the chosen microstructural propo-
sition, are defined in agreement with the portion of the previously written sentence
in order to maintain local cohesion (by the linguistic analysis of previously produced
clues).
It is interesting to note that the idea of linguistic translation (Text Production
process) described by Hayes (1996) in his new model (idea elaborated from the con-
ception of Kaufer, Hayes and Flower, 1986) is relatively close to the principles of
Local Coherence Strategy defended by van Dijk and Kintsch (1983). Hayes (1996)
nevertheless provides a more precise description of the progress of this strategy, in
connection with Working Memory components and not simply, as done by van Dijk
and Kintsch (1983), in the framework of Short Term Memory (Cf. Chapter 5). De-
spite these remarks, the Local Strategy of linguistic formulation proposed by van
Dijk and Kintsch (1983) is interesting on two points. On the one hand, this strategy
seems to take into account the total set of functional parameters of the linguistic
translation (Short Term memory, the previously written text, the macroproposition
maintained in memory). On the other hand, this strategy offers a description of the
particular sequencing of the functioning of linguistic formulation processes. Thus,
the formulation could only be executed by the intermediary of frequent processing
cycles (from the elaboration of a proposition set to the graphic execution of a sen-
tence part). These cycles would concern, for each iteration, a limited set of linguistic
units (portions of sentences, for example). This allows to distinguish the linguistic
formulation from the conceptual planning which could be accomplished outside of
the handwriting activity, during long processing periods (i.e., during pauses) often
testifying the presence of a problem-solving activity (Cf., on this point: Dansac &
Alamargot, 1999 or Schilperoord & Sanders, 1999; and also Chapter 1).
Finally, the translation of a sentence seems to be performed, on a temporal plan,
portion by portion; this step by step processing can be illustrated by a cyclic idea of
the activity of the Translating process. This idea of linguistic processes thus raises
the problem of the nature and the size of the unit that would be processed and that
would determine the iteration of each linguistic translation cycle.
10.2 Processing units during translation cycles

As shown in the categorisation elaborated by Levelt (1989: 23), many and various
‘linguistic translation units’ have been examined in the literature. These units would
belong to the syntactic level (as the sentence or the word), as well as to the semantic
level (as the idea unit, the semantic proposition, or some parts of semantic proposi-
tions, etc.). If such a diversity of units probably testifies that authors have not sys-
tematically approached the verbal production at a same level of analysis (text, dis-
course, sentence, parts of sentences, etc.), it seems nevertheless, with the evolution
of these works, that the processing unit, firstly envisaged through the word (Cf.
Lounsbury, 1965) has been finally identified through the ‘Clause’, or again the ‘T-
unit’ of Bereiter and Scardamalia (1988).
Defined as a grammatical proposition (Clark and Clark, 1977), the pertinence of
the Clause, as the main unit processed during linguistic translation, has been estab-
lished on the basis of studies analysing the on-line characteristics of linguistic pro-
duction (Cf. Goldman-Eisler, 1968; Maclay & Osgood, 1959; Tannenbaum, Wil-
liams, & Wood, 1967; Butterworth, 1980a). Thus, Boomer (1965), Cooper and Pac-
cia-Cooper (1980), or Grosjean, Grosjean and Lane (1979), by interpreting a pause
duration in a sentence or a discourse, as the sign of the regeneration of a processing
cycle, have shown that the occurrences of these durations were more frequent to
grammatical boundaries of a same sentence than during the translation of each of its
grammatical components (Cf. also Chanquoy, Foulin, & Fayol, 1990, 1996). Adopt-
ing another methodology, Ford and Holmes (1978) have demonstrated that perform-
ances to a secondary task were more interrupted when the added task had to be exe-
cuted between the production of two clauses. The function of the clause, in the proc-
ess of translation, has been confirmed by Fodor, Bever and Garrett (1974), or by
Garrett (1975, 1980), through an analysis of the location of errors in discourse (Cf.
supra). Moreover, Levelt (1983) has shown that the detection of errors in discourse,
by the speaker (Self-Monitoring), was less frequent to clause boundaries, due to a
more important necessity at this moment to allocate cognitive resources, in order to
elaborate and to translate a new clause.
By approaching the formulation of a sentence as the matching of a syntactico-
semantic structure and a linguistic structure, the grammatical form of a clause (and
the difficulty of its elaboration) would be thus determined by the characteristics of
the semantic proposition to be translated. This relation of dependence has been made
apparent by Goldman-Eisler (1968, 1972), within a relatively varied corpus of lin-
guistic productions. One interesting result relies on the variation of the processing
time (pause duration), preceding a sentence composed of co-ordinate or subordinate
clauses. Even if, on a syntactic viewpoint, the elaboration of a subordinate clause
can generally be considered as more complex than the building of a co-ordinate, the
processing time of the subordinate is shorter than for the co-ordinate (Cf. also:
Piolat, 1983). This phenomenon, equally observed by Holmes (1984, 1988), would
be explained by the fact that the subordinate clause would translate, in a unique cy-
cle, a complex underlying proposition, within which several propositional parts (or
conceptual units) have been integrated beforehand (or linearised). Conversely, con-
cerning a sentence comprising co-ordinate clauses, the translation would be exe-
cuted by several cycles of processing, establishing a succession of clauses. The
maintaining of the local coherence would then suppose a more important planning
than for a subordinate clause, because this process has not been executed before.
According to this interpretation, the determination of the grammatical structure of a
clause and, globally, of a sentence (or several sentences), would depend on the se-
mantic structuring of the proposition to be translated (Butterworth, 1980b; Chan-
quoy, Foulin and Fayol, 1990; Foulin and Fayol, 1988).
10.3 Conclusion: the importance of the previously written text

Finally, the main problem confronted by the writer, in the course of Translating, is
to continue writing, by taking into account (1) what has been previously written
(necessary bottom-up influence for the establishment of cohesion and local coher-
ence), and (2) the portion of the plan that must be developed (necessary top-down
influence for the maintaining of global coherence). This double constraint is impor-
tant because it incites the consideration of the previously written text whose process-
ing is fundamental to continue writing the text. In the idea of Translating such as
conceived by van Dijk and Kintsch (1983) or Hayes and Flower (1980), the role of
the text and the previously written part of the sentence or text are emphasised.
Rummer and Grabowski (1993) have described the importance of the written trace
by showing that writing with invisible ink strongly increases the cognitive load
linked to the text production. The consideration of reading and interpretation of the
previously written text (that composes the task environment) is one of the central
dimensions of Hayes’ (1996) new model. It is probable that a more elaborated
analysis of this dimension is an interesting and original method to define more pre-
cisely the functioning of the Translating process (by means of the suppression of the
written trace, by introducing some constraints on the written trace – writing with
capital letters, for example – or by analysing the characteristics of occular move-
ments during writing – Cf. Kennedy & Murray, 1987; Pynte, Kennedy, Murray, &
Courrieu, 1988, for such on-line studies as in the case of reading and comprehension
of texts or sentences).
11 DISCUSSION POINTS
The Translating process is relatively difficult to describe because its function, archi-
tecture and processing are very complex. This complexity is largely linked to the
fact that the Translating process, such as defined by Hayes and Flower (1980), has
not only to elaborate conceptual and linguistic representations, but also to match
them. From a theoretical and experimental point of view, this complexity inevitably
leads to a certain number of unsolved questions or points of discussions. They con-

cern (1) the problem of the linguistic formulation at a textual level, and (2) the rela-
tionships between linguistic and conceptual processing.
11.1 The determinism of linguistic formulation for the whole text

According to the theories of sentence translation presented above (Bock and Levelt,
1994, for instance), the elaboration of the grammatical structure of a clause would
strongly depend on the structure of semantic propositions, processed before, during
the steps of elaboration and linearisation of the preverbal message. These relation-
ships of dependence have been particularly well described in local models of sen-
tence generation or in works concerning the dynamics of translating operations. The
question is then to know whether these dependence relationships and their principles
can be found, no more at the level of the sentence, but concerning a whole text or a
whole discourse.
Such a generalisation of processing principles for one sentence to processing
principles for one text (or again the generalisation of the matching between general
and local models of language production) supposes to stop considering the sentence
as an entity of analysis, but rather as a coherent set of sentences (Cf. Costermans &
Fayol, 1997). It is then necessary to define what the specificity of this ‘coherent set
of sentences’ is, as compared to an isolated sentence and also to state the constraints
linked to written and textual dimensions (global coherence, top-down effects of the
highest level of processing, for example). In others words, it is necessary to establish
criteria and clues that characterise the textual dimension at a semantic level, to be
able to define what constraints these clues (that would be comparable to ‘Textual
Roles’, in comparison to the ‘Event Roles’ and ‘Attentional Roles’ of Bock and
Levelt, 1994) can exert on linguistic translation components. It is clear that current
cognitive studies about text or speech production cannot truly describe with enough
precision such hypothetical Textual Roles. Some possible tracks must however be
signalled.
For example, a dependence relationship between semantic structure and linguis-
tic structure has been clearly showed, at a discourse level, by the experiment led by
Ehrich and Koster (1983, Cf. Chapter 1). This work appears to be the rare experi-
ment that has analysed the effects of the domain knowledge organisation until a
lexical and syntactic processing level. The authors have shown that a strong seman-
tic structuring (prototypical spatial relationships between some objects arranged in a
room that participants have to describe), via a semantic linearisation as ‘Topic-
Comment’, could lead to the adoption of more complex grammatical structures (co-
ordinate and subordinate clauses instead of simple sentences) and the use of specific
connectives. This experiment appears as a good idea to tempt one to precisely ana-
lyse, on a theoretical plan, what can be the transformation modes of domain knowl-
edge and the constraints that they can exert on linguistic processing levels, in the
case of a whole discourse and not only a single sentence.
In this way, experimental studies on the use of connectives or punctuation marks in

written productions can provide interesting results which tempt researchers to ana-
lyse how the writer proceeds textual relationships to carry out text coherence and
cohesion. Punctuation marks and connectives are generally used to indicate the de-
gree and the nature of links between states or events described in adjacent textual
segments (Bestgen and Costermans, 1994, 1997; Chanquoy, 1991, 1998; Coster-
mans and Bestgen, 1991; Fayol, 1986, 1997). The analysis of variations in the fre-
quency and use of punctuation marks and connectives, considered as textualisation
indices, can provide information on operations performed by the writer during the
formulation (Chanquoy and Fayol, 1995; Favart and Passerault, 1996; Schneuwly,
1988). Thus, their presence, their frequency and their variety can be considered as
visible marks of the mental conceptual elaboration or translation of information re-
trieved from Long Term Memory. For example, Foulin, Chanquoy and Fayol (1989)
have shown, with an analysis combining off-line and on-line measures (i.e., tempo-
ral analysis and text analysis; Cf. Chapter 1), a narrow relationship between the un-
derlying mental model of the text to be produced and the use of punctuation marks
and connectives. This type of approach is nevertheless very empirical and very de-
scriptive. The rules of use of these surface marks remain vague and have still not
been integrated in sentence generation models, nor in general text production mod-
els.
11.2 Relationships between linguistic and conceptual processing

More generally speaking, dealing with linguistic translating cannot avoid the prob-
lem of the frontier between conceptual and linguistic representations. Most of the
writing models have adopted the claim that domain knowledge is stored in the form
of non- (or pre-) linguistic conceptual items (Cf. General Introduction and Introduc-
tion of this Chapter). As such, it requires linguistic transformation. Just as there is no
current reason to neglect this claim, the question remains as to the definition of the
relationship between conceptual and linguistic representations.
While linguistic formulation is often seen as dependent on the underlying con-
ceptual or syntactico-semantic structure (top-down translating), linguistic constraints
could equally be exercised on the processing of the conceptual products (bottom-up
effect of linguistic characteristics on the content elaboration). It is interesting to ob-
serve that until now, such an interaction between conceptual and linguistic process-
ings has rarely been studied. This interactivity supposes that processes related to
these two types of knowledge could be executed both top-down and bottom-up. It is
not, indeed, inconceivable to think that writers, and mainly beginner writers, have to
modify processed propositional structures because they lack lexical, grammatical or
orthographic knowledge to translate them. In a more general way, the absence of
linguistic knowledge, concerning for example the marking of the global text coher-
ence by connectives, could constrain writers to choose a very iterative writing strat-
egy and lead to the elaboration of a text in which sentences would be simply juxta-
posed (Cf. for example: Favart & Chanquoy, 1999). The same remark can be stated
in the case of a writer’s low expertise concerning syntactic rules implemented during
the elaboration of (very) complex sentences, as subordinate or relative clauses.
The linguistic influence, in terms of choice and availability of connectives, for
example, could be more or less sginificant according to the text type. For example,
in the case of argumentative texts, the use of connectives and other complex markers
is fundamental to oppose, to put into perspective, or to focus on some arguments
(Coirier, Andriessen and Chanquoy, 2000). Researchers generally agree to tell that
children have a lot of difficulties with argumentations, oral as well as written
(Akiguet and Piolat, 1996; Akiguet, Roussey and Piolat, 1993; Andriessen, Coirier,
Roos, Passerault and Bert-Erboul, 1996; Gombert and Roussey, 1993; Bereiter and
Scardamalia, 1982; etc.). The writing of an argumentative text confronts the writer
to great difficulties and the control of this type of text appears belated and difficult,
even in adults (Andriessen, Coirier, Chanquoy and de Bernardi, 1997; Coirier, 1996,
1997; Coirier and Golder, 1993). Two reasons can be put forward to explain this
phenomenon: (1) the belated installation of pragmatic knowledge and of content
(re)organisation operations, as supposed by Bereiter and Scardamalia (1987); (2) the
absence of linguistic knowledge susceptible to translate, in a written form, complex
modes of argumentative reasoning. This last reason converges with studies showing
that young children’s argumentative capacities are relatively developed for oral lan-
guage and in a communicative context, compared to their expertise level for written
argumentative texts (Golder, 1996).
Finally, the questions concerning the belated appearance of the mastery of argumen-
tative texts show one of the current problems of writing models, even for the most
recent ones. Although the influence of the domain knowledge has been studied in
writing (Kellogg, 1987a, Caccamise, 1987) and although most of the models indi-
cates the pragmatic knowledge influence, the works and models concerning the in-
fluence of the availability and amount of linguistic knowledge are still too rare (ex-
cept the model elaborated by Berninger and Swanson, 1994). This kind of knowl-
edge is often neglected as an explanatory factor for some modes of written process-
ing. But doesn’t writing a text imply writing it with words?
12 CONCLUSION ON THE TRANSLATING PROCESS

The Translating process, as defined, for example, by Hayes and Flower (1980), does
not only affirm linguistic translation (sentence formulation), but also the retrieving
of more specific idea units (elaboration), as wells as (semantic) linearisation of these
units. In other words, the processing carried out by the translating process would be
concerned both with linguistic knowledge as well as semantic and conceptual
knowledge. In this framework, we have analysed and discussed the role and func-
tions ensured by the Translating process through four levels of processing: (1) the
content Elaboration under the control of each part of the plan, (2) the semantic and
propositional Linearisation of this elaborated content, (3) the Linguistic Formulation
(grammatical and lexical processing) of semantic propositions, and (4) the grapho-
motoric Execution of the linguistic product (i.e., the externalisation of the linguistic
product, coding the internal text into written form). The activation of these different
kinds of processing would operate through very limited recursive cycles whose unit
would be the clause. Three constraints seem important in their implementation: (1)
the Working Memory (or Short Term Memory) limited capacity, (2) the semantic
and linguistic characteristics of the portion of the already written text and (3) the
conceptual nature of the plan portion – or macrostructural unit – that controls the
proposition retrieval.
Due to its complexity, the functioning of the Translating process during the ac-
tivity of text writing certainly represents one of the least known components among
the writing processes. Furthermore, this fact is amazing because numerous works
and local models exist, for the linguistic translation as well as for the graphic execu-
tion (like handwriting models). Thus it could be interesting that general writing
models largely take ideas from the numerous works on the generation of oral sen-
tences, which have enabled authors like Garrett (1975, 1980, 1984), Dell (1986) or,
more recently, Bock (1996), to elaborate relatively precise and predictive formalisa-
tions about oral linguistic translation. Brought to the framework of written produc-
tion, these local models of sentence generation can of course be considered as re-
straining because they analyse the sentence out of context. But this restriction makes
some functioning rules apparent that would need to be taken into account in the gen-
eral models of writing.
CHAPTER 3
REVISING PROCESS
1 INTRODUCTION: REVISING PROCESS IN WRITING

For Fitzgerald (1987: 484) wrote:
‘revision means making any changes at any point in the writing process. It involves
identifying discrepancies between intended and instantiated text, deciding what could or
should be changed in the text and how to make desired changes and operating, that is,
making the desired changes. Changes may or may not affect meaning of the text and
they may be major or minor.’
The key elements of this definition concern the fact that revising involves making
some changes, whatever they are, not only in the text but also during the whole writ-
ing process. For the author, revision necessitates three main operations: identifying a
problem, deciding about any change and operating it. It also involves two objects
(the intended text and the instantiated text), two types of rules (conventions: what
should be changed, and rhetorical rules: what could be changed), and strategies (how
to implement the desired changes).
Despite the apparent simplicity of this definition, numerous other definitions of
revision are found in the literature and can be very different. This fact underlines the
ambiguity of the term ‘revision’, that designates both the realisation of a correction
and the different procedures (or processes) used to revise (Freedman, 1986).
It therefore appears important, in this chapter, (1) to clearly define the term of
revision, (2) to expose, through several models, how the activity of revision oper-
ates, and (3) to try to surround problems that prove to be important to solve when
proposing a procedural and functional model of revision.
12.1 Definition of the revising process

In Hayes and Flower’s (1980) model, the revising process, called ‘Reviewing’, is
conceived as being recurrently distributed throughout the writing activity. This
process is composed of two sub-processes, respectively labelled by the authors
‘Reading’ and ‘Editing’ (Cf. General Introduction). The Reading sub-process allows
the writer to detect errors and to evaluate the appropriateness of the written text in
relation with the communicative goals established during the Planning process. The
Editing sub-process appears as a system of production rules to solve problems.
While rereading the text, the writer can perceive dissonance between intention
and production (Scardamalia and Bereiter, 1983), which asks for or necessitates a
correction stage. This procedure of correction can, according to Bartlett (1982), be
D. Alamargot & L. Chanquoy (2001). Revising process. In G. Rijlaarsdam (Series ed. ) & D.
Alamargot & L. Chanquoy, Studies in Writing: vol 9. Through the Models of Writing, 99 –
123 . © 2001 Kluwer Academic Publishers. Printed in the Netherlands.
outlined in three main sub-processes: Detection, Identification and Modification of

the text portion considered as erroneous by the writer. However, the already written
text is not the unique object of revision. Revision is also a process of rethinking
thoughts, of evaluating and clarifying them (Chanquoy, 1997, 2001; McCutchen,
Francis and Kerr, 1997; McCutchen, Hull and Smith, 1987). In fact, the revision ac-
tivity can be considered as an activity of text reading or thought evaluation, fol-
lowed, in the case of dissonance(s), by some corrections. In this way, the revision
can then be defined as a set of processes, that consist, as previously defined by Fitz-
gerald (1987), of identifying a problem, deciding whether to change or not the part
of the text in which the problem has been found, and selecting a strategy and operat-
ing it. In addition, writers carry out revisions in order to verify and improve their
texts (Faigley and Witte, 1981; Hayes and Flower, 1986; Hayes, Flower, Schriver,
Stratman and Carey, 1987), but also to verify and eventually improve the mental
representation(s) elaborated during the writing activity. The revision can thus be
considered as a complex process that would heavily weigh on writer’s limited atten-
tional capacities (Beal, 1996; McCutchen, 1996).
12.2 Taxonomy of revisions

Another approach to define the revising process is to classify the different kinds of
revision that appear in texts. Some authors have tempted to categorise revisions, ac-
cording to certain types of criteria. For example, Sommers (1980) has specified four
revising operations: the ‘Deletion’, the ‘Substitution’, the ‘Shifting’ and the ‘Re-
organisation’, that can be carried out in four different linguistic levels: the word, the
phrase, the sentence, or the idea. Monahan (1984) has added or modified dimensions
to Sommers’ categorisation. The author proposed four dimensions: (1) the moment
during which the revision occurs (for example, on the draft or on the final copy), (2)
the linguistic level of the revision (to revise a word, a phrase, a clause, a sentence, a
paragraph, or the whole text), (3) the type of correction (addition, deletion, re-
arrangement, embedding) and finally, (4) the objective of the revision (visual pres-
entation of the text, spelling, style, emphasis of transitions, supplementary informa-
tion, etc.). The important point in this classification is that Monahan extended the
definition of revision, with a goal-related and a functional perspective. In a more
general way, two great types of modifications, or again two categories of revising
activities are taken into account by Chanquoy (1997): (1) the changes she labelled
‘Surface Revisions’, which comprise modifications of punctuation (addition of a
capital letter or a punctuation mark), the re-writing of a word or a text segment con-
sidered as hardly legible, spelling corrections, grammatical spelling (subject-verb
agreements, verb tenses, etc.), and (2) the ‘Deep’ or ‘Semantic Revisions’, that mod-
ify the sense of the text and comprise additions, deletions, shifting, substitutions and
transformations of words, phrases, clauses, sentences or longer text segments. Fur-
thermore, for each of these categories, the modifications can be considered as cor-
rect, erroneous or neutral (i.e., neutral corrections do not improve the quality of the
text, but do not decrease it; Cf. also Scardamalia & Bereiter, 1983).
REVISING PROCESS 101
Although there currently exists a relatively great variety of revision classifica-

tions, the taxonomy elaborated by Faigley and Witte (1981, 1984), even if it is now
outdated, is nevertheless still the only one that takes into account both syntactic and
semantic aspects of revision. Indeed, six types of operations and six linguistic levels
are distinguished. This classification considers, as the preceding proposal, the level
of text considered: revisions can be superficial (i.e., surface changes), or change the
meaning of the text (i.e., text-based – semantic – changes). In addition, this classifi-
cation proposes a certain number of revising subcategories. Thus, surface changes
can be formal (formal changes: spelling, punctuation, etc.), or can preserve the text
meaning (meaning-preserving changes: additions, deletions, etc.). Similarly, text-
based changes can be micro- (additions, deletions, etc.) or macro-structural (Cf.
Table 2).
Table 2: Taxonomy of revision, adapted from Faigley and Witte (1981)
Surface revisions
Formal Changes Meaning preserving changes (para-

(conventional editing revisions) phrases)
Spelling Addition
Tense Deletion
Number and modality Substitution
Abbreviation Permutation
Punctuation Distribution
Format Consolidation
Semantic revisions
Microstructural changes Macrostructural changes

(minor revisions) (major revisions)
Addition Addition
Deletion Deletion
Substitution Substitution
Permutation Permutation
Distribution Distribution
Consolidation Consolidation
According to this table, in surface changes that keep the meaning and in micro- and
macrostructural revisions that modify the meaning, the operations are the same, only
their depth level in the text can change.
Whatever the chosen taxonomy to classify revisions, it is always difficult to dis-
tinguish between a surface or a mechanical error and a meaning error (Cf. Chan-
quoy, 1998). The addition of an adjective can, for example, be considered as a sur-
face modification, because it does not modify the meaning of the text, or as a deeper
change, changing the meaning of the sentence. For example, if the sentence ‘the
boat was sailing on the sea’ is revised as ‘the boat was sailing on the blue sea’, the
modification is only a superficial one, and does not change the global meaning of
the sentence. Conversely, if this sentence is modified to ‘the boat was sailing on the
furious sea’, the meaning is radically changed.
Besides classifications exclusively concerning revisions of the already produced
text, some authors distinguish two modes of revision (revision here in the sense of
the activity of correction): an updating activity, automatically released and an activ-
ity of revision, controlled and intentional. The controlled revision activity would in-
tervene during precise moments of the writing process (Matsuhashi, 1987) and
would then be opposed to the editing activity (Editing sub-process of Hayes and
Flower, 1980), automatised and mechanical, able to intervene at any given time dur-
ing writing, and even to interrupt the other processes (Cf. Bridwell, 1980; Faigley &
Witte, 1981, 1984; Graves, 1975; Hayes & Flower, 1980; Kaufer, Hayes, & Flower,
1986; Perl, 1979).
The term of revision therefore designates both the complex process of changes
carried out during writing and the changes effectively realised by the writer. Up till
now, the authors have generally distinguished between effective revision on paper
and mental revision (Cf. supra). The first category has been classified, as previously
mentioned. Conversely, the second type, due to its mental aspect, is more difficult to
approach and categorise, except when researchers use, for example, verbal protocol
analyses. Thus, this definition of revision can be proposed: something (i.e., a word)
is done (i.e., added, deleted, etc.) to reach a certain goal (improving style, content),
at a certain text level and on a certain text (pretext, already written text), at a certain
moment (i.e., draft, final copy), with a certain effect (i.e., improvement, neutral, de-
creasing effect) and with a certain cognitive cost.
Even if, through the previous theoretical presentation of revision, it is possible to
roughly define the revising process, it is still quite difficult to explain how revision
works. It is therefore by means of models, presented hereafter, that the process of
revision will be described. The revising processes (or sub-processes), and external or
internal modifications to which they lead, are presented, through an analysis of the
evolution of revision models. This presentation would, in addition, allow specifying
the different sub-processes and operations in the progress of revision.
13 EVOLUTION OF REVISION MODELS: FROM REVIEWING PROCESS TO

SPECIFIC AND PROCEDURAL MODELS OF REVISION
13.1 Hayes and Flower’s first proposals (1980, 1983; Flower and Hayes, 1981c)
In the initial Hayes and Flower (1980) model, revision is viewed as an autonomous
process, called Reviewing, and composed of two sub-processes: Reading and Edit-
ing, that simply involve checking and correcting the text, already written or simply
mentally planned (Cf. Piolat & Roussey, 1991-1992 or Temple, Nathan, Temple, &
Burris, 1993 for a description of the revision activity). The Editing activity is con-
sidered as recursive, possibly appearing at any time during the writing process (Cf.
also: Collins & Gentner, 1980; Faigley, Cherry, Jollifre, & Skinner, 1985; Fitzger-
ald, 1987; Piolat, 1988, 1990; van den Bergh, Rijlaarsdam & Breetvelt, 1993) and,
consequently, may also interrupt the progress of the other activities. It can also ap-
pear after the text has been entirely written (Faigley & Witte, 1981, 1984). The
Reading sub-process is voluntarily released, so as to evaluate the text produced-so-
far. The Editing sub-process comprises production rules concerning possible correc-
tions (i.e., to correct semantic imprecision, to evaluate the precision for the reader,
etc.). This sub-process allows both automatic and controlled corrections (i.e., it can
interrupt any other writing process).
Thus, according to Hayes and Flower (1980), as soon as a discrepancy between
intended text and external text has been detected, the Editing sub-process would
function in an irrepressible and automatic manner. This puts nevertheless the prob-
lem of the correction of this error. According to these authors, once the error has
been detected, the writer would immediately (or almost immediately) have means
(or production rules) to solve it; it is thus possible to wonder why an error is com-
mitted…
The framework of Reviewing, such as proposed by Hayes and Flower (1980) in
their general writing model, enables distinguishing two sub-processes and specifying
that the revision has to be approached both as an internal (evaluation) and external
phenomenon (effective corrections). However, this formalisation is insufficient. This
is all the more paradoxical because the activity of revision is thereafter studied in a
particularly intensive manner, by these authors and other researchers, and has led to
increasingly complex architectures, resulting in formalisations sometimes more
complex than the general text production process itself. Besides, a year after the
1980 model, Flower and Hayes (1981c) have slightly modified the theoretical posi-
tion they had previously defended. Then, still two sub-processes are distinguished in
the Revising activity, the ‘Evaluation’ sub-process (that globally corresponds to the
Reading sub-process in the 1980 model; i.e., to read and to compare the already pro-
duced text with the intended text) and the ‘Revision’ sub-process (also similar to the
Editing sub-process, that roughly involves correcting the errors that can be found in
the text). However, opposed to the 1980 model, Revision is in this instance consid-
ered to be deliberate. Consequently, contrary to their preceding proposal, the Re-
viewing process and its two sub-processes are considered as controlled. More pre-
cisely, Revision sub-process is no longer limited to an automatic correction activity
that functions according to production rules.
In their publication of 1983, Hayes and Flower distinguish the activity of ‘Re-
viewing’ from the activity of ‘Revising’. This distinction relies on the ‘external’ or
‘internal’ characteristics of revisions that are carried out by the writer. Reviewing is
necessary to evaluate what is written or what has been planned, and it can lead to the
detection of discrepancies between the intended and the written or to be written text.
Thus, Reviewing is considered by the authors as a mental (or internal) activity
whereas Revising is an external activity, allocating the text and leading to physical
modifications, which can be traced in the text surface (external revisions; Cf. also,
for such distinctions: Fayol & Gombert, 1987; Flower, Hayes, Carey, Schriver, &
Stratman, 1986; Hayes & Flower, 1980; Scardamalia & Bereiter, 1986).
To summarise, the concept of revision in the Hayes and Flower models from
1980 to 1983 have become more and more detailed, and sub-processes are specified
and defined. Even if the 1983 model was clearly a model of revision, it can be con-
sidered as a framework to complete the proposals on the reviewing process made by
Hayes and Flower in their 1980 model. In parallel with these theoretical viewpoints
about revision, other models or theoretical ideas have been elaborated. These are
presented in the following paragraphs.
13.2 Other specific models of revision

13.2.1 Scardamalia and Bereiter’s (1983, 1985) model
Soon after Hayes and Flower’s (1980) general model about the writing process,
Scardamalia and Bereiter (1983, 1985) proposed a more complex and complete de-
scription of the revising activity. Before describing the authors’ perspective, we
would like to specify that their proposal cannot be considered as a real model of re-
vision, but rather, in an educational view, as a technique to help writers to revise.
Nevertheless, psychologists have used this architecture to understand the complexity
of the revising process.
According to Scardamalia and Bereiter, the functioning of revision would work
through a self-regulated procedure, composed of three recursive mental operations
that would chain cyclically during a stage of revision. These operations are respec-
tively labelled by the authors as ‘Compare’, ‘Diagnose’ and ‘Operate’, and the gen-
eral procedure is called ‘C.D.O. procedure’ (Cf. Figure 17).
1)
Composing Process
Plan
CDO process Operate
Representation Yes
Choose Generate
of Compare Diagnose Tactic Text
Intended text No Succeed
Change Change
plan ?
Mismatch Succeed
Representation Succeed
of
Actual text
No mismatch Fail Fail Fail
Return to
Interrupted
Process
2)
Figure 17: Model of the CDO process, adapted from Scardamalia and Bereiter (1983). Copy-
right © 1983 by John Wiley & Sons. Adapted with permission.
The three operations consciously intervene during specific revising cycles and allow
the writer to revise the text sentence by sentence. According to the authors, as
pauses are more frequent between two sentences, the revising procedure would
mainly occur during this text location. For Scardamalia and Bereiter, two types of
mental representations would be built and stored in Long Term Memory during the
composition of a text: a representation concerning the ‘Already Produced Text’ (ac-
tual text) and a representation of the ‘Intended Text’ (i.e., the planned text). It is
very important to stress this distinction. Indeed, for these authors, the text as it is
written is not in its actual form in the writer’s mind, but as a representation of the
text-so-far, which can be quite different from the actual written text. This fact could
explain a fact often noticed in works about revision: when revising their own texts,
writers do not read what they have written, but they read what they think they wrote.
Moreover, as it will be shown later in this chapter, it seems easier to revise a text
written by someone else than a text written by the reviser her/himself.
The C.D.O. procedure would be released when a conflict between these two rep-
resentations is observed and thus, in terms of processing control, it interrupts the
other processes. The three basic operations in the C.D.O. procedure are then carried
out in the following order:
• The Compare operation evaluates the discrepancy between the Already Written
Text and the writer’s Intended Text, and concludes to what extent there is an in-
adequacy between these texts.
• When an inadequacy is apparent by the Compare operation, the Diagnose opera-
tion determines the nature of the problem and its possible corrections.
• Then the Operate operation executes the correction with the help of two com-
ponents:
• the choice of a strategy to solve the problem (Choose Tactic), and
• the content generation (Generate Text Change), that means that the text is
changed by generating a new (piece of) content.
The modification of a text segment considered as problematic, leads to a change
when the writer has the necessary means to realise it. This leads to a new representa-
tion of the text and therefore to a new C.D.O. cycle in order to check whether the in-
tended text and the realised text match or not. However, the implementation of a cy-
cle of revision does not inevitably imply a revision visible in the surface of the text.
This is particularly true when writers do not know how to operate a modification.
Furthermore, writers need, according to Scardamalia and Bereiter (1983, 1985), a
certain number of sub-processes labelled ‘Evaluation’, ‘Tactic decisions’ (to delete
or to rewrite, for example) and ‘Executive control’ of the global revising process
(Cf. for similar and more recent approaches, within an educational psychological
perspective: Hacker, Plumb, Butterfield, Quathamer, & Heineken, 1994; Plumb,
1991; Plumb, Butterfield, Hacker, & Dunlosky, 1994).
Generally speaking, this model accounts for the different textual changes, con-
cerning both the surface and the meaning of the text. Indeed, Scardamalia and Bere-
iter (1983, 1985), as Hayes and Flower (1980) do, make the distinction between the
revision on paper and what they call ‘Reprocessing’, that concerns mental revisions
on the representation of intended text. The revision ‘on the paper’, that is on the rep-
resentation of actual text is a part of the Reprocessing, the visible part. The authors
emphasise that the C.D.O. procedure is not necessarily entirely performed, since
some failures can occur during all processing stages (or cycles) and result in stop-
ping the procedure. ‘Succeed’ and ‘Fail’ in Figure 17 indicate this. This enables
comparing the procedure described by the authors with a relatively natural activity
of revision, where a simple rereading does not inevitably lead to detect and correct
errors, whatever their nature. To specify the activity, different aspects of the writer’s
behaviour are envisaged: for example, the writer can change her/his text, or her/his
plan, or ignore the error (Cf. Bridwell, 1980).
Finally, the interest of this model is to authorise a more precise definition of re-
vising sub-processes: two sub-processes of evaluation are described (Compare and
Diagnose) and the modification of the text comprises two operations, as shown in
the Figure: the choice of a correction tactic and the changes generated in the text.
Another important feature of Scardamalia and Bereiter’s (1983) model is that it has
been experimentally validated, in a study aiming to test both the heuristic of the pro-
cedure and the implementation of a technique of procedural facilitation in education.
In their experiment, fourth (10 years of age), sixth (12 years) and eighth (14 years)
graders have written three short texts. To revise, writers used the C.D.O. procedure,
after each sentence, or at the end of the text, but always sentence by sentence. To
help them, participants had revision clues registered on cards, which corresponded
to the C.D.O. procedure. With this approach, children proved to make more revi-
sions. However, the global quality of their texts was not improved. It would there-
fore seem that this type of procedure causes young writers to revise their text, in a
more complete way. However, the problem of the improvement of the quality of the
text remains since, as the authors emphasise, the number of revisions increases,
while the quality remains stable. This is an important point in our final discussion at
the end of this chapter, since revision, in our definition, must lead to an effective re-
sult or effect.
The procedure described by Scardamalia and Bereiter (1983, 1985) can be con-
sidered more as a technique to facilitate revision than as a procedural model of revi-
sion. To specify the processing involved during revision, Flower and her colleagues
and Hayes and his colleagues have elaborated an even more complex model.
13.2.2 Flower, Hayes, Carey, Schriver and Stratman’s (1986) and Hayes,
Flower, Schriver, Stratman and Carey’s (1987) models
Flower et al. (1986) and Hayes et al. (1987) have elaborated a very precise and spe-
cific model of revision, conceived as architecture describing the revising cognitive
processes. It can be considered as a central framework among current revising mod-
els. This model describes revising processes, strategies and – this is new – the neces-
sary knowledge for revising. For these authors, revision is above all a deliberated
and strategic activity, which writers choose either to implement or not, according to
theirs goals, the text progression and their available knowledge (Cf. Figure 18).
Process Knowledge
Evaluate Task definition
Criteria for
plans and text
Read,
comprehend,
and criticize
Problem
representation
Select strategy
Revise Redraft
Modify
Modify Procedures for
plan and
text fixing text problems
text
Figure 18: A revision model,adapted from Hayes, Flower, Schriver, Stratman & Carey
(1987). Copyright © 1987 by Cambirdge University Press. Adapted with permission.
Processes and Knowledge. This model comprises three main processes, of which
one is composed of two sub-processes:
• the Text evaluation (‘Evaluate’) with (a) a Problem Detection and (b) a Problem
Diagnosis,
• the Selection of a revising Strategy (‘Select Strategy’) and,
• the Execution (‘Revise: Modify Text’ and/or ‘Redraft: Modify Plan and Text’)
These three processes of the model could globally correspond to the C.D.O. stages
in Scardamalia and Bereiter’s (1983, 1985) model. They interact with four kinds of
knowledge.
• the ‘Task Definition’,
• the reviser’s knowledge about the production task (‘Criteria for Plan and Text’
i.e., Linguistic knowledge and Domain knowledge),
• the ‘Problem Representation’,
• the reviser’s repertoire of revising strategies (‘Procedures for fixing text prob-
lems’).
Defining the task. The ‘Task Definition’ corresponds to the writer’s representation
of the task to accomplish. It enables management (that is controlled) of the distin-
guished revising processes and sub-processes. This Task definition is elaborated
through the writer’s metacognitive knowledge (i.e., the knowledge about the revi-
sion task), and mainly depends on the writer’s objectives (goals, a priori fixed plans,
and so on). The Task Definition allows the writer to define the general objective of
revision, the aspects of the text to be revised and the means to make use of to reach
this objective. This definition is fundamental, strategic and conscious; it will guide
the whole revision process and determine the sequence of processes during this ac-
tivity. It depends on the writer’s preliminary knowledge about revision, the features
of the text (genre, goal, theme, etc.), the context (instructions, task environment,
etc.) and the addressee.
Evaluating the product. ‘Evaluate’ is a complex process of text reading/scanning

(‘Read’, ‘Comprehend’ and ‘Criticize’). This process depends on a priori goals (for
example: to read for clarifying, for verifying, for evaluating the accessibil-
ity/comprehension of the text for the reader, etc.). This process is very costly in cog-
nitive resources because it is, for the authors, highly controlled. It is managed
through three main functions: to understand, to evaluate and to identify (or to define)
potential problems. It comprises two sub-processes: (a) a high level sub-process of
Problem Detection in the initial text, and (b) a sub-process of Identifying or Diag-
nosing the nature of the problem (Cf. also Bereiter & Scardamalia, 1986; Hayes &
Flower, 1986; Scardamalia & Bereiter, 1983). The Problem Detection results from
the comparison between the writer’s initial intentions (i.e., the intended text as a
function of constraints, goals and criteria fixed before writing commences) and the
text produced so far (Cf. also, Bridwell, 1980; Sommers, 1980). It concerns there-
fore a very complex sub-process, mainly because the writer must be able to carry out
the comparison between intention and text (Bartlett, 1982; Perl, 1979). Hayes and
his collaborators distinguish three evaluation levels: (1) comparison of intention and
text, (2) correspondence between the text plan and the writer’s goals, and (3) corre-
spondence between the text and the linguistic rules (spelling, grammar, etc.).
With a dynamic point of view, the first step of Revising, consists, via the ‘Evalu-
ate’ process, of detecting, roughly speaking, some errors. The detection of a disso-
nance between intentions and goals can therefore, according to Hayes, Flower,
Schriver, Stratman and Carey (1987), concern the text plan (that is to say the internal
representation of the text) and lead to an internal revision, or the already written text
and which would provoke an external revision. The identification of the problem –
or diagnosis – follows. From the identification of the problem, writers choose a
strategy adapted to the problem, given that this strategy is known and available. The
writer has therefore to diagnose the nature of the error and to correct it. The evalua-
tion can therefore be positive or negative. If it is positive, the writing activity con-
tinues, or ends, when the revising activity was the last activity of the writing proc-
ess. If it is negative, a representation of the problem must be elaborated. According
to the authors, this representation can be more or less precise and, consequently, can
lead to a more or less accurate/adequate solution.
Selecting a strategy according to problem representation. The selection of a Revi-

sion Strategy would mainly depend on the nature and the precision of the ‘Problem
Representation’ (Cf. also Hayes, 1996). For instance, dependent on the nature of the
problem, its detection in the text does not systematically release a modification visi-
ble in the written text. When the problem is too complex, the writer can indeed ig-
nore the problem and continue to write the text. Conversely, the writer can decide to
rewrite the portion of problematic text (Cf. also: Hayes & Flower, 1986; Hull, 1987).
The authors propose a framework to specify the nature of the Problem Representa-
tions. For them, these representations would be situated on a continuum going from
‘ill-defined’ to ‘well-defined’, that allow the writer to choose more or less sophisti-
cated ‘Procedures for Fixing Text Problem’ (detection and diagnosis procedures),
and hence, ‘Revising Strategies’.
Five Revising Strategies are defined. The first three strategies are called ‘strate-
gies of revising processes exploitation’ and lead to a come back of preceding stages;
the other two strategies concern some modifications of the text:
• to ignore the problem, because it is too superficial or, conversely, too complex,
• to decide to solve the problem later in the writing process,
• to search for more information in memory or in the text in order to understand
and define the problem better,
• to rewrite the text or the segment of the text while preserving the basic idea (this
strategy calls for Planning and Translating processes),
• to revise the text so as to greatly preserve the already produced text.
Executing revisions. The Execution process is the realisation of modifications in the

written text, when a Revising Strategy has been chosen, to solve the detected prob-
lem. Different typologies exist to categorise the nature of these modifications (Cf.
supra and also: Bridwell, 1980; Chanquoy, 1997; Faigley & Witte, 1981, 1984; Ma-
tsuhashi, 1987; Sommers, 1980). For Hayes et al. (1987), a revision can be charac-
terised by:
• the type of operation executed in order to sustain the revision: addition, deletion
or substitution of words, phrases, sentences, modification of a part of the text,
etc. (Cf. supra the paragraph about the taxonomy of revisions),
• the textual level in which the revision appears: text surface (mechanical revi-
sions) or text depth (semantic revisions: Cf. supra),
• the location of the revision in the text: at the beginning, the middle, or the end
of the text,
• its location during the writing activity: during the writing of a plan, a draft or
the final version of the text, during a specific revising stage, etc.
With respect to the preceding proposal elaborated by Scardamalia and Bereiter
(1983), this model is extremely precise since, besides the diagram of the model it-
self, the authors have frequently provided extra schemas to explain the different
choices in which the writer can be engaged during the revision. Thus, during each
stage of revision, the authors have specified many different options, which can be
chosen by the writer, and they have illustrated them as often as possible, both by ex-
amples and by production or procedural rules. Moreover, in Flower et al. (1986) and
Hayes et al.’s (1987) model, the knowledge stored in Long Term Memory has a
fundamental role in the releasing of the different stages. According to these authors,
revision would call for an ongoing interaction between the domain knowledge (the
knowledge about the theme), the linguistic knowledge, the pragmatic knowledge
(the writer’s goals and the knowledge about the addressee) and the processes defin-
ing the task, evaluating the text, detecting errors and selecting revising strategies.
To conclude, this model has very interesting frameworks among other very global
and very general models concerning writing. As previously specified, it integrates
the C.D.O. procedure elaborated by Bereiter and Scardamalia (1983), but it is, in ad-
dition, more precise and more explicit, since it specifies, for each operation of revi-
sion, necessary cognitive resources (or knowledge) and representations for the reali-
sation of the different revising sub-processes. Similarly, this model comprises an
important procedural component, by describing releasing conditions of the different
revising strategies. It therefore describes functional aspects of the revising process,
with hierarchically organised sub-processes, that serially appear, or that are subordi-
nated to other processes. It also demonstrated the great complexity of revision, func-
tioning in a cyclical way, with the help of various types of knowledge and many
processes. The revisions applied to the text are therefore the concrete result of a very
complex internal activity, in which many decision-making stages are involved,
which possibly result in a written correction. Thus, the revision appears as a largely
strategic activity (and therefore conscious, not automatised), necessitating from the
reviser to operate some choices through her/his available knowledge, to make use of
a certain number of very strategic options, while looking at the mental representa-
tions of the task. The revision could then be qualified as ‘a decisional activity that is
controlled at a metacognitive level’ (Piolat, 1990: 186, our translation).
The model elaborated by Flower et al. (1986) and Hayes et al. (1987) can then
be considered as the more complete and complex model, describing and explaining
as clearly as possible the revising activity. Hayes (1996) has nevertheless recently
discussed these proposals. More precisely, Hayes wished to show and to emphasise
the importance of reading and comprehension during the revising activity, because
these aspects were not precisely described/explained in preceding models, through
the evaluation of the text (comparison text-representation, verification of the text
surface or meaning, detection of errors and their diagnosis).
13.2.3 Hayes’ proposals (1996): Reading as a key-process

By emphasising reading and comprehension activities during revision, Hayes (1996)
has specified the function of the ‘Evaluate’ process, by considering it as a process of
reading and comprehension, adapted from the Just and Carpenter’s (1980) definition
of these activities. The reading comprehension process enables the writer to build a
representation of the text meaning, with the use of many kinds of knowledge. In ad-
dition, with the idea of detecting one (or some) problem(s) in mind, for Hayes, the
‘writer’s reading’ is different from the ‘reader’s reading’, because the writer does
not simply read to understand, but also to detect errors, problems or ambiguities.
Thus, Hayes considers revision as ‘a composite […] of text interpretation, reflection
and text production’ (p. 15). This definition implies that, to understand revision, it is
both necessary to consider how these processes function and in what order. Hayes
then proposes a new model of revision very close to a control structure.
This model is built with three entities:
• the control structure, or ‘Task schema’ for the author (with a goal, a set of revis-
ing activities, some revising criteria, etc.),
• the different ‘Fundamental Processes’ involved in revising, i.e.:
o Reflection (with Problem-solving and Decision Making),
o Text Processing (with Critical Reading), and,
o Text Production,
• the ‘Cognitive Resources’ (from Long Term Memory and Working Memory).
For Hayes (1996), the most important entity of the revising process is the control
structure that plays a great role for the nature and the quality of revising perform-
ance (Cf. Figure 22; Chapter 4, for a more precise description of this control struc-
ture). This model also emphasises the pregnant role of reading in revising proce-
dures. In this way, Hayes specifies that two other kinds of reading are fundamental
for writing. He calls them ‘Reading Source Texts’ and ‘Reading to Define Tasks’.
The second kind of reading could be necessary to decide what must be corrected in
the text. Thus, Hayes has not deeply modified the previous proposals of Flower et
al. (1986) or Hayes et al. (1987), but he has mainly completed these models by un-
derlining the fundamental status of reading in writing. For him, reading contributes
to writing performance in three ways: ‘Reading for Comprehension’, ‘Reading to
define the writing task’ and ‘Reading to revise’.
13.2.4 Conclusion on specific models of revision

All these previously presented models are finally relatively descriptive (even if
Flower et al.’s (1986) and Hayes et al.’s (1987) models are very complete, complex
and take into account a great number of operations and constraints), and do not en-
tirely explain how writers manage the revising process during the progress of the
writing activity. They give possible processes (and sub-processes) of revision, quite
numerous and quite complicated, without specifying what can prime them; why
most of the time writers do not revise; what are the necessary cognitive resources;
and so on. In addition, it seems that only Scardamalia and Bereiter (1983, 1985)
have provided an experimental validation of their procedure. Indeed, Hayes et al.
and Flower et al.’s model seem to be very difficult to experimentally validate. How
could it be possible to test such a complex model? How could it be possible to give
an account of so many possible strategies?
In addition, these models do not indicate precisely where the necessary pieces of
knowledge and the different representations that are built during the revising activity
are stored or maintained: in Long Term Memory or in Working Memory? If knowl-

edge and representations are stored in Long Term Memory, the models do not indi-
cate precisely how these pieces of information are retrieved (i.e. from Long Term
Memory to Working Memory in which they would be maintained and probably
processed).
Finally, two of the main criticisms to address the different models of revision are
the followings: (1) they do no locate the revising sub-processes in a processing
framework, and (2) the activating and progress modes of the different processing ac-
tivities are not explained. Butterfield, Hacker and Albertson (1996) have recently
developed this last procedural dimension of revising.
13.3 A procedural model of revision by Butterfield, Hacker and Albertson (1996)

13.3.1 Analysis of the model
The model built by Butterfield, Hacker and Albertson (1996) is not as precise as the
preceding ones, but can be considered as a modernised version of Flower et al.’s
(1986) model. It presents the advantage to show the importance and the role of Long
Term Memory, Working Memory, and also the importance of Metacognitive knowl-
edge during the revising activity. In addition, the authors have provided, for each re-
vising stage or for knowledge implied in revision, many convincing experimental
validations. This model takes into account a large number of dimensions. It distin-
guishes two main parts: the ‘Environment’ and the ‘Cognitive/Metacognitive sys-
tem’(Cf. Figure 19).
Environment Cognitive/Metacognitive System

REVISER’S LONG-TERM MEMORY
RHETORICAL PROBLEM
M METACOGNITION C
Topic Models of Knowledge Understanding of strategies
o Topic Thinking o
Audience
Importance n Language & Writing Reading n
Standards of Evaluation Writing
i t
t r
COGNITION
o Knowledge Strategies o
Topic Thinking
ACTUAL r Language & Writing Reading l
TEXT BEING REVISED Standards of Evaluation Writing
Format
Genre REPRESENTATION OF TEXT BEING REVISED
Lexical units
Syntactic units Format, Topic, Genre, Audience, Lexical and
Propositions Syntactic units, Propositions, Gist, Importance
Gist
REVISER’S WORKING MEMORY

Represent Rhetorical Problem, Plan, Read to Represent and Comprehend
and Standards of Evaluation for Text Actual Text
Detect and Diagnose problems in Select, Modify or Create Strategies
Represented Text for Revising Represented text
Translate Revisions from Represented text in Actual Text
Figure 19: Procedural model of revision, adapted from Butterfield et al.’s ( 1996)
The Environment comprises the ‘Rhetorical Problem’ and the ‘Actual Text being
Revised’. The Rhetorical Problem space allows the specification of the theme
(Topic) and the addressee (Audience), and the Importance of the text to be revised.
In this model, the text produced so far is a simple linguistic output: the authors in-
deed postulate that the revision mainly concerns a mental representation of the text,
even if it is effectively executed in the written text. In this way, this framework can
be distinguished from preceding models, that differentiated the ‘mental revision’
from the ‘revision realised on paper’ (i.e., external revision). Here, even though the
revision can be a visible trace on the sheet of paper as well as a change at any mental
level (i.e., concerning the plan or the text representation, for example), Butterfield et
al. consider that revisions are always mental, that is to say that they are elaborated
within an ‘internal’ level before – some of them – becoming apparent on the text
that is produced. It is probably for this reason that the authors have elaborated mem-
ory components in their model, as well as metacognitive components. In addition,
interactions between the different parts and/or between the different processing lev-
els of the model are always possible at any time during the progress of revision (Cf.
Figure 19).
In Butterfield et al.’s (1996) model, the Cognitive/Metacognitive system comprises

two entities: the reviser’s Long Term Memory and Working Memory.
Deliberated processes (i.e., controlled) take place in Working Memory; these
processes are rather similar to those described in Flower et al.’s (1986) model. They
respectively carry out: (1) the Representation of rhetorical problems and texts, (2)
the Detection and the Diagnosis of textual problems, (3) the Use of strategies to
solve these problems.
Long Term Memory is notably used to free resources in Working Memory: the
already revised text segment could, for example, be stored in Long Term Memory.
Moreover, Long Term Memory would be composed of two separated levels, la-
belled by the authors ‘Cognition’ and ‘Metacognition’ (Cf. also Butterfield, Albert-
son, & Johnston, 1995). Two components, labelled ‘Monitoring’ and ‘Control’ allow
the interactive functioning of these two levels. The two levels (i.e., cognitive and
metacognitive) are composed as following:
• At the ‘Cognitive level’, there is knowledge, strategies and representation of the
text being revised. Three categories of knowledge are described: (a) ‘Topic’
knowledge, (b) ‘Language and Writing’ knowledge (about linguistic rules and
conventions) and (c) ‘Standards of Evaluation’ knowledge. Three main strate-
gies would implement these different types of knowledge: (a) ‘Thinking’, (b)
‘Reading’ and (c) ‘Writing’. Butterfield et al. specify that when strategies are
automatised, they take place within Long Term Memory (that is to say that they
are directly retrieved, without supplementary costly processes), and do not need
any cognitive resources. Conversely, when these strategies are deliberated, they
take place within Working Memory whose resources are limited. In addition,
Plumb, Butterfield, Hacker and Dunlosky (1994) underline that the knowledge
stored in Long Term Memory is fundamental for revision. This kind of knowl-
edge indeed necessitates a constant interaction between the knowledge about

linguistic rules, the theme, the writer’s goals and the addressee, and between the
different processes which define the task, evaluate the text, detect errors and se-
lect strategies to solve these errors. This idea is very close to that previously de-
veloped by Hayes et al. (1987) .
• At the ‘Metacognitive level’, there are ‘Models of knowledge’ and ‘Understand-
ing of strategies’. In addition, the same kinds of knowledge would exist at the
Metacognitive level as at the Cognitive level: knowledge about the topic, lan-
guage and writing, and Standard of Evaluation. The understanding of strategies
would be distributed in three groups, as previously: (a) To think, (b) To read
and (c) To write.
Between Metacognitive and Cognitive levels, the authors distinguish Monitoring
and Control strategies.
• ‘Monitoring’ strategies consist of, for example, re-reading a difficult part of the
text, looking back to prior text, predicting the text-to-be-written, comparing
eventual solutions. The Monitoring activity supposes to proceed to a metacogni-
tive analysis of processing carried out at a cognitive level.
• ‘Control’ strategies aim at clarifying text information, correcting inaccurate
text, etc. Contrasted to the Monitoring activity, the Control activity is carried
out by metacognitive strategies on the cognitive level (Cf. Figure 19, page 115).
These metacognitive strategies would enable knowing when, where, how and
why using, evaluating and controlling strategies of the cognitive level, in the
Working Memory. The evaluation and the control-regulation would be auto-
matic in Long Term Memory (Cf. equally Hacker, 1994, 1997).
In summary, one of the advantages of this model, comparative to the preceding
frameworks, is to specify not only the role of Working Memory, but also the func-
tion of Long Term Memory during the revising activity. Furthermore, the authors
clearly distinguish (and this was only implicit in previous models) a cognitive and a
metacognitive level. Thus, this model, certainly currently the most complete, com-
prises, just like the general production model recently built by Hayes (1996), an En-
vironmental part and a Cognitive and Metacognitive part. This distinction is one of
most interesting aspects of Butterfield et al.’s model. With this distinction, the au-
thors have tried to formalise Flower et al.’s (1986) observation, according to
whether the revision would depend more on activated knowledge than on available
knowledge. The metacognitive level would thus guide the reviser in the use of the
necessary knowledge for implementing the revising activity. The writer could then
fail during revision, not because s/he does not possess the necessary knowledge for a
precise correction, but because s/he is not able to activate the right knowledge to
perform the adequate revision (Cf. infra, the experiment led by Plumb, Butterfield,
Hacker, & Dunlosky, 1994).
Finally, as mentioned previously, Butterfield et al. have not only elaborated a
very sophisticated model, but they have also provided empirical validations for each
component of their model. Thus, contrary to the previous frameworks, Butterfield
and his colleagues have tried to find, in the already existing literature, experimental
data to validate their model. Moreover, when this data does not exist, they have built
experimental paradigms to verify their claims. Some of these experiments are now
briefly presented. In addition, the interest of these studies is that they cannot only be
related to the model provided by Butterfield et al., but also, for some of them, to
Flower et al.’s (1986), Hayes et al.’s (1987), or Hayes’s (1996) frameworks.
13.3.2 Experimental validations
Research referring to the Environmental part. Concerning the Topic, the experi-
ments led by McCutchen, Francis and Kerr (1997) and Butterfield, Hacker and
Plumb (1994) show that a great amount of (or appropriate) knowledge about the
theme (Cf. our discussion in the conclusion of Chapter 1), needing less cognitive ef-
fort (Kellogg, 1987a), leads to more revisions. However, surface revisions (e.g.,
spelling, punctuation, etc.) are always more important than the corrections relative to
the meaning of the text. In the same way, experiments on the text genre showed that
it has no real impact on the nature and the frequency of revisions (Butterfield,
Hacker, & Plumb, 1994; McArthur & Graham, 1987; McArthur, Graham, &
Schwartz, 1991).
Relative to the audience, Traxler and Gernsbacher (1992, 1993; Cf. Chapter 1)
have shown the positive effect of the addressee’s feedback on the revision of de-
scriptive texts. In addition, Beason (1993) showed that the comments applied to a
first draft improved the revision. Similarly, a factor such as the authorship seems
important. Indeed, the revisions were more frequent and more errors were detected
in others’ texts than in the writer’s own texts, at least in young writers (Bartlett,
1982; Cameron, Edmunds, Wigmore, Hunt, & Linton, 1997) and in adults (Dane-
man & Stainton, 1993).
Specific works about the Cognitive and Metacognitive part. Domain knowledge, as-
sociated with knowledge about linguistic rules and revision, would enable more er-
ror detection and correction and, in addition, would facilitate the detection of surface
errors (Butterfield et al., 1994; Hacker et al. 1994 ; Plumb et al., 1994). Conse-
quently, a lack of revising knowledge or processing could explain why writers gen-
erally do not revise.
In a attempt to explain this question, Plumb, Butterfield, Hacker and Dunlosky
(1994) have proposed two hypotheses that are not exclusive: (1) writers do not re-
vise because they do not possess the necessary knowledge about revision, and/or (2)
writers do not revise although they have the necessary knowledge, they just do not
use it. The first hypothesis refers to a deficit at the Cognitive level (knowledge defi-
cit) or, more precisely, concerning Declarative (about the topic) and Procedural
(about the discourse) knowledge. Writers do not revise because they lack appropri-
ate knowledge to revise, knowledge concerning the discourse level (set of syntactic
and rhetoric rules and knowledge about the addressee and the goal, stored in Long
Term Memory) or the theme level. The second hypothesis refers to a deficit at the
Metacognitive level (processing deficit). It would concern a problem of processing

at the level of the control of comprehension, the management of the activity, and/or
the activation of the necessary knowledge. This deficit would be linked to a poor
comprehension of the task, or to the difficulty in applying a particular strategy. Ac-
cording to the authors, this second hypothesis is more complex to explain. Indeed,
the writers would fail to detect their errors by lack of efficient strategies; by an ab-
sence of necessary knowledge activation; by a bad perception of the task nature, or
by a lack of motivation.
The results of Plumb, Butterfield, Hacker and Dunlosky (1994) were globally in
favour of the hypothesis of a deficit at the processing level (Metacognitive level).
The topic and discourse knowledge would then be necessary, but not sufficient to
revise. The question now is, whether it is possible to guide these writers using the
knowledge they possess, so as to correct their texts. Thus, it would seem that, except
for spelling revisions, knowing how to correct an error is not sufficient for detecting
errors (Hacker, Plumb, Butterfield, Quathamer and Heineken, 1994). These authors
therefore assert that the position theoretically and empirically defended by Flower,
Hayes, Carey, Schriver and Stratman (1986), Bartlett (1982) and Fitzgerald (1987),
according to whether error detection and correction would depend on separated
knowledge and on different processes, is not valid. Their experiments have indeed
shown that the detection necessitates knowledge of how to correct and that this
knowledge is however not sufficient.
Conversely, Butterfield and his colleagues clearly propose fewer works aiming
to validate the metacognitive level of their model. Even if the metacognitive aspects
are largely developed in Butterfield et al.’s model, there is currently little research
about them. The authors simply underline the importance of these aspects during the
revising activity. It seems that a better control of thought results in a better metacog-
nitive comprehension and thus, better revising strategies (Hacker, 1997). This type
of comprehension appears very important for revision (Bracewell, 1983), and would
consequently need to be empirically validated. It remains now to find empirical
means so as to analyse these metacognitive components of the revising activity.
Despite the relative lack of studies concerning the metacognitive part of the
model, it is important, once again, to underline the fact that Butterfield, Hacker and
Albertson (1996) have carefully tested all components of their model, what largely
strengthens its validity compared to other revision models. These empirical valida-
tions specify the functioning of the activity, emphasise the factors important for re-
vising and enable the generation of new hypotheses that will have to be tested in the
future. For example, it seems very important to explain the minor impact of the text
type on the activity of revision, while this factor remains important for the global
writing activity. It is thus possible that experimental manipulations have not been
sufficiently accurate to make the effect of the text genre obvious.
14 POINTS OF DISCUSSION
14.1 Nature of revising activity: questions on type and functioning
If the models specifically concerning the revision process are now very complex, a
certain number of questions need to be clarified, especially those concerning the na-
ture of processes and their functioning. For example, with regards to Long Term
Memory, the role and the importance of the various pieces of stored knowledge must
be tested. In other words, it is possible to ask what the necessary resources for revi-
sion are. It seems that, when there is vast amount of knowledge about topic, ad-
dressee, text types and writing procedures, revision would be of better quality
(Hacker, Plumb, Butterfield, Quathamer and Heineken, 1994; Plumb, Butterfield,
Hacker and Dunlosky, 1994). However, the previous works have shown that these
different kinds of knowledge are necessary but not sufficient to revise. Indeed, as
shown by Plumb, Butterfield, Hacker and Dunlosky (1994), the stored knowledge is
not sufficient. It is necessary to activate items from storage correctly and when it is
useful in the course of revision. More research is needed on this point. In addition,
as shown previously, there is currently very little research about the role of meta-
cognitive knowledge on the revising activity. This knowledge is considered as im-
portant, but its precise function is not described or explained. Three questions are
then fundamental regarding the nature of the revision processing.
Internal vs external revisions. The first one concerns the distinction between two
types of revision. Is it relevant to distinguish the two types of revision, as suggested
by Scardamalia and Bereiter (1983), in dissociating the revision on the paper, that is
the external revision, and the reprocessing activity or internal/mental revision (Cf.
equally reviewing and revising in Hayes and Flower, 1983; or Fayol & Gombert,
1987; Flower, Hayes, Carey, Schriver, & Stratman, 1986; Hayes & Flower, 1980;
Scardamalia & Bereiter, 1986)? Indeed, it seems difficult to empirically distinguish
these two types of revision, both in the course of the writing activity and, more spe-
cifically, concerning the way to measure them. Currently, it is possible to postulate
that the internal revision would take place very early on during writing, for example
during the progress of the Planning process or during the first steps of the Translat-
ing process, while the external revision would mainly concern all the visible revi-
sions done on the draft or on the final copy. Is it then possible to consider that all re-
visions are mental and that their results could be ‘invisible’ or ‘apparent’ on paper?
It would not be the nature of revision (i.e., internal or external) that would be impor-
tant, but the level of representation, that is to say the mental text plan or, concerning
the already written text, the importance of the segment to be revised (paragraph, sen-
tence, clause, phrase, word, portion of word, etc.). Roughly, the revision of the text
plan would be necessarily mental, while the revision of the spelling of a word would
be inevitably external and visible on the sheet of paper.
Autonomous vs recursive process. The second question concerns the functioning of

revision. Is it possible to consider revision as a totally autonomous process, as it is
viewed in Hayes and Flower’s (1980) model or is it a recursive activity, with opera-
tions that can appear during the whole writing process (Collins & Gentner, 1980;
Faigley, Cherry, Jollifre, & Skinner, 1985; Fitzgerald, 1987; Piolat, 1988, 1990) and
stop the other activities (Cf. Chapter 4)? Here again, it seems that this question is
quite difficult to test. One of the unique empirical means would be to use verbal pro-
tocols, which are known to contain a certain number of problems, notably concern-
ing their validity. However, it would be equally possible to use a method of con-
straining participants to revise only once they have entirely written their text, or dur-
ing the course of the writing activity (Cf., for example: Chanquoy, 2001; Scar-
damalia & Bereiter, 1983). It would then be necessary to analyse the texts and the
quality of the revision produced. This paradigm would nevertheless not entirely
validate the above theoretical claims. These claims probably depend largely on the
nature of the writing activity: if this activity is, for a large part, automatised, then
some processes can be managed in parallel. When a particularly costly revision is
necessary, then it would be possible that a control process could interrupt or stop the
progress of the other processes, so as to allow the process of revision to take place.
In addition, it could be possible that revision could appear throughout the writing ac-
tivity. Revision being an autonomous process could interrupt all the other writing
activities. In the same way, revision considered as a recursive process could appear
in parallel with other writing processes. Thus, whatever the nature of revision, that is
autonomous or recursive, internal or external, it can appear throughout the writing
activity. In addition, revision, considered as a final checking, could just as well ap-
pear when the text has been entirely composed. It is thus reasonable to think that, at
least for expert writers, some surface revisions are executed in parallel with the other
writing processes, without any supplementary cognitive cost, and therefore without
the necessity of stopping the other processes. The revisions would notably involve
spelling checks and the addition-suppression of punctuation or capital letters. Some
semantic revisions could probably also occur as the other writing processes are per-
formed, but the majority of these revisions would probably need an interruption of
the other processes. Finally, if not executed during writing, surface or meaning revi-
sions would be operating during the final revision step, once the text has been writ-
ten. There again, this type of claim would need to be empirically validated. How-
ever, it is still difficult to validate such observations. The methodological means do
not seem to be sufficiently powerful to make the cognitive processes ‘visible’.
Automatic vs controled process. This second question brings about another one,
about the more precise nature of the revising process. Is, at least for expert writers,
the revision an automatic or controlled process? Indeed some researchers opposes a
deliberated revising activity (for example: Matsuhashi, 1987), to an automatic edit-
ing activity (Hayes & Flower, 1980), that can stop the other processes (Cf. Bridwell,
1980; Faigley & Witte, 1981, 1984; Graves, 1975; Hayes & Flower, 1980; Kaufer,
Hayes, & Flower, 1986; Perl, 1979). It is from then on possible to postulate, in rela-
tion to the preceding point, that revisions leading to surface corrections would be
automatically managed, at least for experts, while revisions concerning the meaning
of the text would be deliberated and would consequently necessitate more cognitive
resources and more control on one’s own production?
14.2 Revision and comprehension

Another point seems crucial for future research about revision: what are the relation-
ships between revision models and comprehension models?
Currently, many authors frequently point out the importance of reading for revis-
ing. For example, for Hacker (1994), revision necessitates reading and revision of
the meaning necessitates in addition the comprehension of what is read. He then
considers the comprehension monitoring as a writing process, that he defines as ‘un-
derstanding derived through reading’ (p. 146). In addition, Hayes’s new conception
of revision (1996) is mainly focused on reading. This aspect has been, until recently,
not very developed by authors. But, as a matter of course, the revision necessitates,
as it has been made apparent by the different models presented above, an attentive
reading of the text, so as to find some errors and so as to compare a mental represen-
tation of the intended text with the produced text.
Thus, some important aspects of revision depend on and begin with the reading
of the written text, so as to establish a comprehensible and coherent mental represen-
tation of the text to be revised, fundamental to process the text (Cf. Hacker, 1994;
for a complete review of works on comprehension and reading, necessary for revi-
sion). According to Daiute and Kruidenier (1985), the revision would consist of an
internal dialogue between the writer and the ‘writer-reviser’. The revision would de-
pend on the facility of text reading: the less costly the reading, the more the writer
would revise. Thus, the revision would be difficult or impossible when all cognitive
resources are used for reading (Hacker, 1994). However, the experiment led by
Cameron, Edmunds, Wigmore, Hunt and Linton (1997) is largely opposed to these
claims. These authors have shown that texts of increasing reading difficulty did not
lead to fewer revisions. McCutchen, Francis and Kerr (1997) add that meaning er-
rors would necessitate, in order to be detected, a more attentive reading than those
concerning the text surface would. Finally, Perl’s (1979) study reveals that writers
would not inevitably reread what they have written, but what they believe to have
written.
What is then the ‘best moment’ to revise, notably for beginner or novice writers,
which have still not automatised the low-level writing processes, nor those of read-
ing? (Cf. van den Bergh, Rijlaarsdam and Breetvelt, 1993 who show the importance
of the revision period between evaluating – mental process – and actually changing
– manual process). For example, Chanquoy (1997) has shown that the revision dur-
ing the production led to more surface modifications (spelling, punctuation), while
the revision made after the production, less expensive, entailed more semantic modi-
fications. Moreover, the corrections made afterwards are globally as efficient as
those made during writing. Thus, opposed to Scardamalia and Bereiter (1983), these
results show that the revision after the production is not more constraining but, con-
versely, it lightens the cognitive work and incites writers to a more attentive reread-
ing of their texts and perhaps even to a more elaborate reflection.
In fact, on the one hand, writers must be able to reread their texts and therefore
must have sufficient reading skills. On the other hand, writers must have sufficient
cognitive resources to manage this activity, in parallel with the revising process and
maybe with the other writing processes.
14.3 Revision and Working Memory

The last remark of the previous paragraph underlines, once again, the complexity of
the revision activity, that can heavily weigh on the writer’s limited processing ca-
pacities (Cf. Chapter 5). Many authors have already shown that limited Working
Memory capacities strongly load the writing process (for example in models built by
Hayes, 1996 and Kellogg, 1996 and in works led by Berninger & Swanson, 1994;
Swanson & Berninger, 1994). This seems equally true for the different Revising
sub-processes, that would largely depend on limited Working Memory resources
(Hacker, 1994). Today, increasingly numerous works are interested in the role of
limited Working Memory capacity on revision (Hacker, 1994; Hacker, Plumb,
Butterfield, Quathamer, & Heineken, 1994; McCutchen, 1994; Swanson & Bern-
inger, 1994). As for general writing processes, the different Revising sub-processes
can compete for limited Working Memory resources. During revision, it is indeed
necessary to read the text and to be able to detect an error (Cf. Beal, 1990, 1996) and
the detection of a problem appears as cognitively costly. Similarly, the diagnosis of
the detected problem depends on the simultaneous access both to a representation of
the written text and to a mental representation of the intended text. This sub-process
would be therefore equally costly, since it necessitates the capacity to maintain in
memory two representations of the text (McCutchen, 1996). Hacker (1994) then ex-
plains the fact that writers, whatever their expertise level, revise more the surface
than the meaning of the text, because surface revisions would consume less cogni-
tive resources (Cf. also: Hacker, Plumb, Butterfield, Quathamer, & Heineken, 1994;
McCutchen, 1994; Swanson & Berninger, 1994). Moreover, Fitzgerald and Mark-
ham (1987) underline that Working Memory would be more solicited for the detec-
tion of meaning or incoherence errors than for the detection of spelling or grammati-
cal errors. Consequently, this cognitive overload would not allow revision, all
writer’s resources being consumed for some activities considered as more important
(for example: Planning or Translating processes). It thus appears very important to
be able to evaluate the cognitive cost of revision. To do so, it would be necessary to
consider the cost of the different revising sub-processes (in the C.D.O procedure, or
in Flower et al.’s, 1986; or Hayes et al.’s, 1987 models). However, up until now, re-
searchers have more hypotheses or claims than empirical results enabling to pre-
cisely define the respective costs of the different revising sub-processes.
Finally, just as general writing processes, the revision sub-processes themself
can compete for limited Working Memory resources (McCutchen, 1996). To test
this kind of prediction, it is possible to envisage some techniques, as those used by

Levy (Cf. Lea & Levy, 1999 or Levy & Marek, 1999 in Chapter 5), consisting of
loading one of the two Working Memory slave systems or the central executive, so
as to examine the effects of this cognitive load on the activity and the management
of revision sub-processes. The revision must thus be conceived as a complex proc-
ess, consuming writers’ attention and memory capacities (Beal, 1996; McCutchen,
1996). We will come back to this aspect in Chapter 5.
14.4 Revision and text quality

Although we have honoured a more theoretical than empirical approach to revision,
it appears nevertheless important to emphasise that most researchers are unanimous
concerning the relationships between text quality and quantity of revisions. Appar-
ently, some precise links between the quantity and the efficiency of revisions in a
text and the quality of this same text would not exist. In other words, revision would
not systematically lead to improve the text quality. Nevertheless, a study by van den
Bergh, Rijlaarsdam and Breetvelt (1993) specifies this pessimistic observation.
These researchers have shown that, when an analysis of the revisions is differently
managed, it is then possible to observe a relationship between quality and revision.
These authors have attempted to differentiate more or less efficient revising proc-
esses in order to increase textual quality. By analysing revisions on texts concerning
two themes, produced by forty 9th graders (15-years-old), the authors have made it
apparent that three processes are most often linked to the quality of texts. They con-
cern the rereading of the last part of the text written so far, the evaluation of a text
passage and the change of a sentence or a segment of a sentence. Conversely, three
other processes are negatively correlated with the text quality: the rereading of the
entire text written so far, the evaluation of formal aspects and the change of words.
In a similar research, Breetvelt, van den Bergh and Rijlaarsdam (1994) have shown
that the revision is generally negatively related to the text quality when the act of re-
vision is in the first and second part (1/3 and 2/3) of the writing process, when this
process was divided in three equal parts. The quality was measured by four aspects
(orientation, structure and organisation, reader’s awareness, language and style).
These results, more complex than those they previously obtained and those of Scar-
damalia and Bereiter (1983), show the necessity to continue investigations aiming at
analysing relationships between text revision and text quality. Indeed, it seems evi-
dent that numerous and efficient revisions must increase the quality of the text.
Therefore, why is this not systematically the case?
14.5 Revision and development of expertise

Another very important point to be discussed, although neglected here, concerns the
developmental aspect of revision. However, some remarks relative to the differences
between expert writers and novice writers obviously imply that children belong to
this last category (Cf. Chapter 6). Briefly, authors agree that young writers do not
revise or few revise and that their revisions mainly involve on surface aspects of
text. This is explained by a lack of cognitive resources, all consumed for non auto-
matised high-level activities, as those of planning or translating. Thus, for young
writers, the revision emphasises the problem of self-evaluation and self-correction, a
very difficult problem for young and/or inexperienced writers (Beal, 1993; Piolat,
1988). Moreover, according to previous works, reading is one of the main sub-
processes of revising activity (Hayes, 1996). Consequently, to the management of
writing processes must be added the necessary reading of the text to be revised.
There again, children’s limited capacities would not allow a simultaneous manage-
ment of these different activities (McCutchen, 1996).
In the same way, Hayes and Flower (1986) have shown that, for revision as for
other writing processes, the difference between novices and experts was important.
While experts consider revision as an activity concerning the global text (‘whole-
text task’), novices would consider this activity at a more local level, which is at a
sentence level (‘sentence-level task’). Inexperienced or beginner writers define revi-
sion as consisting of changing words, suppressing errors and deleting parts of the
text (Sommers, 1980), and therefore make essentially low-level revisions, while ex-
perienced or expert writers have more sophisticated revision strategies (Faigley and
Witte, 1981; Sommers, 1980). These different results can equally be related to lim-
ited Working Memory capacity and the failure of revision is often explained in
terms of cognitive overload in Working Memory (Cf. McCutchen, Kerr, & Francis,
1994). This overload would result from the absence of automatisation of some cog-
nitive processes linked to writing. It would concern, for children or novice writers,
mainly low-level processes (such as handwriting), that would then prevent the high-
level processes (such as reviewing) to correctly occur. These different points will be
developed in the chapter devoted to the writing expertise (Chapter 6).
15 CONCLUSION ABOUT REVISING PROCESS

To conclude after these points of discussion, that seem theoretically fundamental,
some other points must equally be emphasised. These remarks are less thorough
than the preceding ones, but they should nevertheless be taken into account in the
evolution of revision models.
Most studies dealing with revision have essentially focused on the cognitive as-
pects of revision, conceived as a problem-solving activity, and have especially ana-
lysed the failures of the revising process, for which there are still more hypotheses
than solutions (Fitzgerald, 1987). The metacognitive aspects have therefore been, up
until now, very largely forsaken, as it has been underlined during the presentation of
Butterfield et al.’s (1996) model.
Also, an important point, currently very fashionable in cognitive psychology, has
not been envisaged here. The revision models have evolved to become real
‘autonomous’ (or specific) models, parallel to classic writing models. The impor-
tance of representations, knowledge and strategies stored in Long Term Memory and
the fundamental role of Working Memory are emphasised. However, one particular
aspect, underlined by Hayes (1996) in his new writing model is not very developed.
It concerns the role of motivation, affects, individual differences, situational factors,
etc. on revision. Indeed and more particularly for the process of revision, the motiva-
tion for example has to play a very fundamental role in the activating of this process.
PART 2: PROCESSING MODALITIES
AND DEVELOPMENT OF EXPERTISE
IN WRITING MODELS
D. Alamargot & L. Chanquoy (2001). Revising process. In G. Rijlaarsdam (Series ed. ) & D.
Alamargot & L. Chanquoy, Studies in Writing: vol 9. Through the Models of Writing, 99 –
123 . © 2001 Kluwer Academic Publishers. Printed in the Netherlands.
CHAPTER 4
NATURE AND CONTROL OF PROCESSING
1 INTRODUCTION: FROM THE NATURE OF PROCESSING

TO THE NATURE OF CONTROL
In the first part of this book, we have discussed the architecture of writing processes
(Planning, Translating and Revising), such as it can be described through different
models. Nevertheless, the structural description of processes implied in the activity
of text writing is not sufficient to account for the progress of this activity. The writ-
ing activity is indeed based on dynamism of processes. Thus, when composing a
text, writers have to continually shift between planning the main ideas, content
translating and text revising. This continuous shifting gives the writing activity its
cyclic nature and depends on a strategic and recursive ordering of different sub-
processes.
According to Breetvelt, van den Bergh and Rijlaarsdam (1994), such a temporal
management of writing process plays a central role in writing activity. For example,
if a given process is set up too late, or too early, this can affect the final product
quality, concerning both content organisation (planning process) and grammatical
aspects (linguistic formulation). These authors have carried out an analysis of the
temporal unfolding of processes underlying forty texts, accomplished by twenty par-
ticipants (15-years-old). Largely inspired by the architecture of Hayes and Flower’s
(1980) model, Breetvelt et al. (1994), like Caccamise (1987), have divided each sub-
ject’s total writing time in three periods. They have identified, within each period,
the verbal protocol nature collected throughout the writing activity. The protocols
have been classified in eleven categories, defined according to criteria determined
by Flower and Hayes (1983) or Swarts, Flower and Hayes (1984), and susceptible to
reflect the totality of writing sub-processes (Goal Setting, Generating, Organising,
etc.). The analyses of the distribution of the different verbal protocol categories,
considering the three periods, show that the first third of the writing time would be
devoted to the comprehension of the instructions and to the evaluation of the task
goal. The second period would be characterised by the realisation of Goal Setting,
Generating and Organising sub-processes, while the last period would be mainly de-
voted to the handwriting (Execution) and to the reading (Reviewing) of the text.
These results confirm processing sequencing (from Macroplanning to Execution and
Revision). They especially show that the quality of the final text would strongly de-
pend on adequately triggering the different processes, within a sequence of process-
D. Alamargot & L. Chanquoy (2001). Nature and control of processing. In G. Rijlaarsdam

(Series ed. ) & D. Alamargot & L. Chanquoy, Studies in Writing: vol 9. Through the Models
of Writing, 127 – 156 . © 2001 Kluwer Academic Publishers. Printed in the Netherlands.
ing. It seems, according to authors, that releasing a given process too early or too
late could lead to a negative effect on the quality of the final text.
This experiment clearly shows that the management and the co-ordination of the
different processes constitute a crucial problem for the writer (Fayol, 1994). In the
same perspective, Bruce, Collins, Rubin and Genter (1979) emphasize that the ne-
cessity of co-ordination represents one of the major difficulties of the writing activ-
ity. In Hayes and Flower’s (1980) model, such a co-ordination is carried out by a
control process called Monitoring (Cf. General Introduction). This entity would de-
termine the processes organisation and ordering, and would allow reiterating some
processes if the text characteristics do not respect the previous writing objectives
(Cf., about the notions of management and control: Bracewell, 1983; Espéret &
Piolat, 1991; Graham & Harris, 1994). Although minimum, this definition of
Monitoring is interesting because it supposes that the control of processes relies on
two activities. The first one is strategic: the writer needs to determine an optimal
processing organisation. However, this optimisation of processing itself depends on
a second activity, which involves evaluating the appropriateness between the proc-
essed and the fixed objective.
The writer has then to control not only the activation of processes but also the perti-
nence of the processed text, for a given stage in the progress of the writing activity.
Nevertheless, from one model of writing to another, and from speaking to writing
models, the conception and the definition of this control entity, managing these two
activities, can considerably vary. For example, while Hayes and Flower (1980) give
to the Monitor the function of processing control of Planning, Translating and Re-
viewing processes, Kellogg (1996) considers that this kind of control would be per-
formed by a component relatively close to the Reviewing process, as defined by
Hayes and Flower (1980). In the same way, Hayes (1996), in his new model, seems
to have renounced to a Monitoring entity and has replaced it with the notion of proc-
essing control through Task Schemas. This heterogeneity about control ideas is all
the more apparent when the oral production models are also considered. In Levelt’s
(1989) model, for example, two modes of control co-occur. A first type of process-
ing control would be intrinsically carried out. Each process component (here the
Formulator and the Articulator; Cf. General Introduction) would start only if the
products previously processed are in accordance with the processes ensured by the
component situated just after. A second mode of control would be executed by a
comprehension loop on the final verbal product. This comprehension system would
be able to engage the reiteration of processes if the message does not respect com-
municative goals, or comprises lexical or grammatical errors. It is clear that defini-
tions and formalisations to account for the principles of a control entity in oral pro-
duction models are very diversified. Nevertheless, if this fact certifies, to a certain
point, that this control entity is still relatively unknown, such diversity is finally only
apparent.
In fact, it seems that the different ideas about control, within the models, are
much linked to the more general theoretical foundations, adopted by authors, to ac-
NATURE AND CONTROL OF PROCESSING 129
count for the nature of processes that they formalise. The important point is here that
the nature and modalities of processing would strongly determine the mode of con-
trol of these processes. Indeed, while local models of sentence production adopt se-
quential modes of formalisation (for example: Garrett, 1980), as well as parallel (Cf.
Dell, 1986, and Chapter 2), global models of verbal production can also be distin-
guished as a function of theoretical foundations adopted to account for the nature of
processes that they formalise. For example, Levelt’s (1989) model is issued of
modularist ideas while the writing model developed by de Beaugrande (1984) relies
on interactive processes, most functioning in parallel. Thus, a modularist formalisa-
tion of processes often supposes the adoption of a control entity relying on a com-
prehension system, because processes would be impenetrable (i.e., they are consid-
ered as encapsulated; Cf. Fodor, 1983 and Cf. further). In a perspective close to con-
nectionist theories, the control of processes, here able to appear in parallel, would
rather be intrinsic, relying on activation and inhibition potentials that propagate in a
network.
Through these different examples, it clearly appears that an analysis of the nature
of the processing control within verbal production models cannot be dissociated
from an analysis of the nature of controlled processing. Before describing and dis-
cussing the control entities in different production models, the different conceptions
about processing will be analysed. Thus, the double question about (1) the nature of
writing processing and (2) the control nature of processes will be developed and dis-
cussed through different models, issued from cognitive psychology, speaking and of
course writing research.
16 THE NATURE AND MODALITIES OF PROCESSING

It seems that the components formalisation of verbal production models is currently
very diversified in the choice of more general cognitive models from which they are
inspired. They are different modes of representations and processes in the verbal
production models, and these different ideas about processing modalities can be de-
scribed according to two axes. The first axis concerns the nature of processes and it
globally opposes an interactionist idea to a modularist idea of processes. The second
axis, partly linked to the first, concerns modalities of processing implementation. It
opposes a sequential mode of processing to a semi-parallel or parallel mode of proc-
essing.
16.1 Processing in interactionist models of verbal production

The interactionist approach is currently the most classical in production models. In-
teractionist models can be classified into two categories: (1) computational symbolic
models, supposing globally that symbolic representations are transformed by proc-
esses ordering a continuation of operations; (2) connectionist models, characterising
by an absence of processes and symbolic representations and by the presence of im-
plicit processing, linked to modifications of activation potentials (state change) in a

sub-symbolic (or a-symbolic) network.
Although these two models (symbolic and connectionist) strongly differ, both
concerning the idea and the formalisation of processes, they nevertheless rely on the
same general interactionist principle. This principle supposes that the processes real-
ised for a given representation level (symbolic or sub-symbolic) can modify proc-
esses realised at a superior or inferior representational level. Processes are therefore
considered as interactive.
16.1.1 Computational symbolic models

Nature of processing. In the case of the symbolic computational approach, the inter-
actionist nature of processing relies on the following six postulations and principles:
• The processing is managed by interactive and recursive processes. These proc-
esses are composed by an organised set of operations, each playing a role in the
transformation of one symbolic representation into another.
• The process constitutive operations, as components of a more general cognitive
system, do not strictly depend on a given activity.
• The relatively generic character of an operation, and maybe of a process (as, for
example, the Planning process performed in most complex activities; Cf. Miller,
Galanter, & Pribram, 1960 and Cf. Chapter 1) allows to process different kinds
of representation, according to the nature of the activity for which this operation
or this process is solicited (i.e. visual, spatial or propositional representations).
• The implementation of a given operation, and, in a stronger way, of a given
process, necessitates, most of the time, an aware, conscious and attentional con-
trol, supposing the allocation of a pool of limited cognitive resources.
• The attentional control of the implementation of operations and processes im-
plies that processes can be interrupted at any given time and/or that the func-
tioning of an operation can be influenced by processing realised by another op-
eration. This permeability of processes, both top-down and bottom-up, mainly
underlies the interactionist principle (in other words, the interaction of proc-
esses).
• The permeable and generic characters of processing operations, and the neces-
sity to allocate them with cognitive resources propitious to their implementa-
tion, often suppose the existence of a superordinate control entity, managing the
temporal sequencing of these operations, as well as the allocation of cognitive
resources.
Applied on writing activity, these processing principles are the basis of the initial
model built by Hayes and Flower (1980) and Flower and Hayes (1980), in order that
the activity of a process can be released or interrupted according to the activity of
another process. Similarly, van Dijk and Kintsch (1983) have developed, in their
model, a particular mode of processing based on the top-down influence of levels of
symbolic representations, through the notion of embedded strategies (Cf. Chapter 1).
Finally, in Bereiter and Scardamalia’s (1987) Knowledge Transforming Strategy,
the formulating process can be interrupted to give place to rhetorical and organisa-
tional processes allowing to adjust contents to be formulated as well as the ‘linguis-
tic appearance’ of this formulation.
Concerning the non specificity of processes, Hayes (1996: 13) postulates that the
cognitive processes implied in writing are not exclusive to writing activity, but can
be used in other types of activities, such as reading (in the case of Text Interpretation
process, for example), problem-solving (in the case of Reflection process), or during
conversations or drawing execution (in the case of Text production process). Finally,
the idea of an allocation of cognitive resources to the different processes is equally
adopted by Hayes (1996) who considers that Working Memory and Long Term
Memory resources can be freely shared between writing processes.
Modes of processing: some mainly sequential processing. In computational sym-

bolic models, the necessity to allocate cognitive resources (furthermore limited) to
processes leads to the consideration that their respective implementation is sequen-
tial or serial. Such are equally the modalities of processes adopted by the most clas-
sical writing models, as Hayes and Flower’s (1980) or Bereiter and Scardamalia’s
(1987) ones. These authors globally consider that writing processes, because of their
respective cost, could only be applied separately. Nevertheless, a lesser necessity of
attentional control and allocation of resources can be reached if the functioning of
operations or processes is sufficiently automatised, as in the case of an important
expertise in a given activity (Anderson, 1983a). In this case, it is possible to envis-
age a parallel functioning of several processes.
More precisely, it could be considered that low level processes (as graphic plan-
ning or lexical access) could function at the same time as a high level process (plan-
ning one written segment during the graphic execution of another segment; Cf.
Chanquoy, Foulin, & Fayol, 1990, 1996; Power, 1986). But this parallel processing
mode is considered (as it will be developed in Chapter 6, devoted to the develop-
ment of expertise) only when low-level processes no longer require cognitive re-
sources. Unfortunately, non much experimental works exist concerning the validity
of such parallel processes during writing.
16.1.2 Connectionist models

Nature of processing. In the particular case of connectionist models, the interactivity
does not concern the process activity (that are non-existent to the extent that the sub-
symbolic system modifies itself by the game of activations and inhibitions propa-
gated between the nodes that compose it), but concerns different levels of sub-
symbolic representations in a hierarchical network (i.e. comprising several strata of
a-symbolic representations). The activation of a node in a given level propagates to
the related node(s) of similar, and low or high levels. By this game of propagation of
activations (and inhibitions), all modifications of one representation level (a stratum
in the network) leads to modifications within another level. In this system of proc-
essing and opposed to the symbolic computational approach, modifications of the
different representational strata operate in parallel until they find an interactive bal-
ancing state.
This principle of processing is especially interesting regarding the computation
power and economy that it offers, contrary to symbolic computational models. Nev-
ertheless, the main difficulty of a purely connectionist approach in the framework of
text writing, relies, in its original form (such as defined by Rumelhart and
McClelland, 1986), on its a-symbolic aspect. It becomes difficult to surround, for the
researcher, the processing nature and modes of the different intermediate mental rep-
resentations allowing transforming a multidimensional domain representation in a
linear written trace. Only the system input and output representations can be evalu-
ated in a connectionist model. Even if it is possible to consider, with Smolensky
(1992), the idea according to which connectionist processes would constitute an ‘in-
troductory’ stratum of processing, underlying and/or determining a higher symbolic
stratum, it is still difficult today to clearly establish a relationship between these sub-
symbolic and symbolic levels, only from activation and inhibition states in a connec-
tionist network.
This symbolic representation defect can be however bypassed by intermediate
approaches, proposing the implementation of modes of connectionist processes (ac-
tivation and inhibition principles) in symbolic networks (Cf. Anderson, 1990: 11-14,
for a discussion on this topic). Such theoretical approaches have been more particu-
larly developed within general models of knowledge acquisition, as production sys-
tems, whose ACT* model (and its evolution ACT-R: Adaptative Control of Though)
built by Anderson (1983a, 1993) can constitute a typical example.
Globally, the cognitive architecture of the ACT* system relies on the presence of
two registers in Long Term Memory, labelled Declarative Memory and Procedural
Memory, respectively storing Declarative Knowledge and Procedural Knowledge.
The Declarative Knowledge can be activated in Working Memory. The transit and
processing of information or knowledge, within and through these registers, are en-
sured by seven processes, whose definitions are classic in cognitive psychology. It
concerns processes of (a) information Encoding in Working Memory, (b) Storing of
this information in Long Term Memory, (c) knowledge Retrieving from Long Term
Memory, (d) Matching in Working Memory between Declarative and Procedural
Knowledge, (e) cyclic Application of the procedural system, (f) Execution of a re-
sponse in Working Memory, and finally (g) effective Performance of the previously
planned answer (Cf. Figure 20).
• The ‘Declarative Knowledge’ is formalised as a network of knowledge units
(Nodes). This knowledge can be encoded in the form of concepts, propositions,
mental images or temporal strings. Whatever their nature, these knowledge
units are connected together by semantic relationships (Links) and benefit from
Activation Potentials, facilitating more or less their recovery in Working Mem-
ory, via the principle of Spreading Activation, propagating all along the links
(Anderson, 1974; Anderson and Bower, 1973).
• The ‘Procedural Knowledge’ is formalised by Production Rules such as ‘If …
Then …’ that represent autonomous processing operations, composed of their
conditions of implementation (‘If ...’) and their processing instructions (‘Then

...’). This production system is organised by the final task goal. The implemen-
tation order of different rules is thus defined by a control structure composed of
processing sub-goals allowing to reach the final goal. The elaboration of this
procedural framework (production rules – operations –; sub-goals; final goal)
allows to plan the realisation sequencing of the different production rules (proc-
essing planning).
• The process of ‘Matching’ carries out the interaction between Procedural
Knowledge and Declarative Knowledge. The role of this process is to insert the
corresponding units of Declarative Knowledge in the slots of production rules
(If – declarative unit X – then – processing Y–). It thus allows the processing, in
Working Memory, of declarative representations by the production system and
the production of a response adapted to the goal fixed by the task.
Declarative Application Production

Memory Memory
Retrieval Match
Storage Execution
Working
Memory
Encoding Performances
Ouside World
Figure 20: General architecture of the ACT* system,adapted from Anderson (1983a). Copy-
right © 1983 by Harvadr University Press. Adapted with permission.
According to this model, the knowledge acquisition and the expertise development,
concerning a precise task or a given activity, relies on a progressive restructuring of
the Procedural Knowledge (automatisation), that, in return, will lead to modify the
content and the hierarchical organisation of knowledge units contained in the De-
clarative Memory. It is certainly because of this specificity in knowledge acquisition
that the ACT* system represents a particularly well adapted framework to account
for learning by action (by the practice of an activity).
The ACT* model formalises knowledge acquisition through connectionist processes

applied on symbolic representations. Thus, most of the processes of the ACT* gen-
eral processes, concerning both the Declarative Knowledge retrieval or the Matching
process between Declarative Knowledge and Procedural Knowledge, rely on the
same functional principles as a connectionist system: a network composed of nodes
and links, that are characterised by activation and inhibition potentials. Anderson in-
deed considers that the ACT* system has been implemented on neurological basis
(Cf. Anderson, 1990: 2-13).
The relationships between connectionist systems and some symbolic production
systems come from a common problematic aiming to clarify storing and retrieving
modes of knowledge in Long Term Memory. This is illustrated by Collins and Quil-
lian’s (1969), or Bower’s (1970) first generation works. Defending the notion of
network as a mode of structuring in Long Term Memory, the second generation of
researchers in this area (Anderson and Bower, 1973; Rumelhart, Lindsay and Nor-
man, 1972) has been divided from 80’s by continuing the study of the symbolic
processing or by exploring modes of a-symbolic processes, like the connectionism.
In both approaches, computation systems are often equivalent (i.e., they both use
differential equations). However the connectionism does not take into account, op-
posed to the symbolic approach, the constraints linked to characteristics of knowl-
edge units (mental image, temporal string, semantic proposition, Cf. above). For ex-
ample, the difference between the ACT* system and a connectionist network mainly
resides in the symbolic definition of nodes in the ACT* system, and in the fact that
processing procedures of these nodes are not intrinsically linked to the valences of
the different network potentials. They are contained in a procedural mnemonic regis-
ter where they are stored as production rules.
The ACT* model is very heuristic through the distinction it makes between De-
clarative Knowledge and Procedural Knowledge, and also the formalisation of prin-
ciples of Spreading Activation. In the framework of text writing, this model is often
quoted, both to explain the formalisation of Pragmatic processing, as in the case of
Levelt’s model (1989; Cf. Chapter 1), or to specify principles of Spreading Activa-
tion as a mode of domain knowledge retrieval, as in the case of Bereiter and Scar-
damalia’s (1987) model. In addition, Schilperoord (1996) or van den Bergh and Ri-
jlaarsdam (1999; Cf. Chapter 5) recently proposed some experimental validations to
modelise the writing activity by using the principles and architecture of the ACT*
system.
Modes of processing: parallel processing modalities. Concerning connectionist sys-

tems or intermediate frames (i.e., ACT*), models are mainly characterised by the
presence of parallel processes. Dell’s (1986) research on generation of oral sen-
tences can be situated in this framework (Cf. Chapter 2). In addition, some models
or elements of models for writing, try to integrate modalities of purely parallel proc-
essing. Such is the case, for example, in Galbraith’s theory (1999) that explain, by
means of a connectionist system, the generation and the creation of new domain
knowledge in the course of the linguistic translation activity (Cf. Chapters 1 and 5).
However, such an integration of purely connectionist and parallel processes in a

text-writing model is still a theoretical assumption. In addition, Galbraith’s (1999)
model of Knowledge Constituting cannot really account for the pieces of knowledge
that are created or modified during writing activity, or for the modalities of knowl-
edge elaboration or transformation.
Consequently and surely due to the a-symbolic character of connectionism, writ-
ing models relying entirely and explicitly on these principles, do not seem to cur-
rently exist, contrarily, for example, to the sentence production model developed by
Dell (1986). In the framework of text writing, the possible interest for the connec-
tionist approach seems to mainly concern processing modalities inherent to a net-
work (activation and inhibition principles) and not principles of a-symbolic knowl-
edge representation defended by researchers in this area. Thus, parallel-processing
principles of several representation levels (by reciprocal activation and inhibition)
could constitute an entirely relevant approach in the case of writing activity. Writing
indeed supposes to process various important types of knowledge (domain, linguis-
tic and pragmatic) that can constrain each other. In this way, parallel processing
models could be particularly adapted for reconsidering the processes underlying, for
example, the Knowledge Transforming strategy that needs, according to Bereiter
and Scardamalia (1987), an interaction between two spaces of Content Knowledge
and Rhetorical knowledge.
The proposals made by de Beaugrande (1984) can be considered as a theory in-
cluding some of these parallel principles. De Beaugrande (1984) has proposed a
writing model called Parallel Stage Interaction. It is a multilevel model that empha-
sises different mental units being manipulated during writing. The main interest of
this model is to show that the different processes of symbolic construction function
more or less simultaneously and can strongly interact. Thus, the result of a level
processing can modify the state of knowledge in the other levels. This model implies
that the writer (here an expert writer) could have a flexible access to different mental
representations of the future and already written text, as well as to conditions about
textual plans and to a highly sophisticated control structure that co-ordinates proc-
esses during these different stages (interaction of processing levels).
16.1.3 Conclusion on interactionist models

In the framework of the interactionist theory, the nature of processing can be modi-
fied as a function of the result of a process of higher (top-down effect) or lower level
(bottom-up effect). In this perspective and more particularly in the framework of a
computational approach, the processing carried out by a particular process can be in-
terrupted or postponed, for example, if the realisation of the activity needs the im-
plementation of another process. This principle of interactivity of processes or repre-
sentation levels is not however found in all general cognitive models. A diametri-
cally opposite approach, in its principles, can be represented by the modularist the-
ory, that relies on a principle of non interactivity of processes. In the framework of
verbal production, these modularist principles, explained hereafter, are often adopted
to formalise the speech production (Cf. Levelt, 1989).
16.2 Processing in modularist models of verbal production

The modularist approach fundamentally differs from the interactionist approach. In
this theory, mainly developed for oral comprehension by Chomsky (1980) and Fo-
dor (1983, 1985), processes are not accomplished by processes but by modules. The
distinction between a ‘process’ and a ‘module’ is not only terminological but relies
on a different conception about processing modalities. In the case of modules, the
processing modalities are as follows (Fodor, 1983) :
• The modules are specifically and exclusively linked to a given activity.
• They are very rapidly executed.
• This execution does not assume a conscious control or a particular allocation of
cognitive resources.
• The non-allocation of cognitive resources allows a nearly parallel functioning of
all the modules.
• The modules can be identified as specific zones in the cerebral system and
would be genetically determined; this last aspect would explain, for example,
why human beings are the unique species able to use language.
• The processes executed by each module are considered as encapsulated.
This last characteristic is central in the modularist approach. The encapsulation of
processes supposes that processes are (1) released only by the presence of a highly
specific representation (a maximum input), and (2) not hampered by the outputs of
the other modules (minimally affected by other output for other components). Be-
cause of this encapsulation, the processes associated with a module, once engaged,
are irrepressible and cannot be interrupted during the entire activity. Thus, a module
cannot be subjected to the direct influence of another module and the organisation
goes essentially from the highest to the lowest levels. The encapsulation of modules
limits the possible interaction between them during the progress of the different
processing.
16.2.1 Nature of processing

This principle of modularist processing has been considered by Levelt (1989: 14), in
his speaking model, even if Levelt prefers to use the term of Component instead
Module. According to Levelt, each processing component (among others, the For-
mulator and the Articulator; Cf. General Introduction for a description of this model)
would be relatively autonomous and specialised. The activity of a module would be
released only by the output of the preceding module. One of the author’s arguments
to justify this choice is based on on-line characteristics of oral production. The
speech has to be rapid, correct and fluent. Another argument relies on the fact that,
opposed to writing, speaking traces are very transitory, and do not offer the speaker
the possibility to revise his/her speech.
While Levelt’s (1989) various arguments are receivable in the case of speaking,
such an idea of processes seems nevertheless difficult to adapt to writing models. On
the one hand, in written production, the writer’s activity often consists of gradual
and strategic writing, by frequently alternating Planning, Translating and Revising
activities. In addition, many experimental studies have shown, for instance concern-
ing the revising activity, that writing processes could be stopped even in the course
of processes by the revision process (Cf. Chapter 3). On the other hand, the
encapsulation of processes and the automaticity of their progress do not correspond
with the generally observed characteristics of the implementation of the Planning
process. Indeed, during planning, the writer’s activity is strongly segmented on a
temporal plan because of the execution of problem-solving operations, like
inferences, reflection, adjustment of goals, etc. (Bereiter and Scardamalia, 1987;
Scardamalia and Bereiter, 1991; Hayes, 1996). Finally, the automatic functioning of
modules is often associated with an absence of attentional focalisation or cognitive
resources allocated to constitutive operations. Again, many studies of text writing
obviously show that the Planning process, for instance, necessitates a strong
implication of the writer’s cognitive resources (Kellogg, 1987b, 1988).
Although these three arguments strongly militate an interactionist idea of writing
processes, the modularist hypothesis cannot be totally rejected because it is not in-
conceivable that some writing processes could be encapsulated. Indeed, if the proc-
ess interactivity is a main characteristic of the writing activity, all the processes can-
not demand the same degree of interactivity. Linguistic processes (i.e., syntactic,
lexical and graphemic) could certainly be modelled according to modularist princi-
ples, as defended by Levelt (1989) for speaking. Conversely, conceptual processes
(i.e., Generating, Organising, Goal Setting and Rhetorical-Pragmatic processes)
need a high attentional control and the modularist approach thus appears more un-
certain. Levelt (1989: 21), although adopting a modularist viewpoint, nevertheless
admits that the functioning of the Conceptualiser and the Self-Monitoring (1) can
operate under the dependence of the Working Memory capacity, (2) necessitates im-
portant cognitive and attentional resources, and (3) needs a conscious control. Ac-
cording to the author, this necessity of control would be exceptionally linked to
communicative situations where pragmatic and/or conversational knowledge (al-
though largely automatised by the experience) cannot be directly retrieved and used
from Long Term Memory. One of the specific characteristics of the Conceptualiser
would indeed be sensitive to the conversational and communicative context. This
relative absence of encapsulation would explain why the Conceptualiser must some-
times modify or strategically adapt conversational knowledge initially automatised.
This idea seems close to Fodor’s (1985) theory that admits that some modules
can sometimes require cognitive resources to function. This relative permeability of
a module constitutes one of the evolutions of the modularist approach as compared
to the idea initially developed by Fodor (1983). Indeed, in 1983, Fodor presented the
‘modularity of mind’ as a formalisation essentially dedicated to the low-level proc-
esses, thus inadequate to account for conceptual processes. This evolution has been
systematised by Jackendorff (1992). This author defends the idea according to which
the cognitive system would contain a particularly important number of modules.

Each of these modules could be specialised (by specific computational procedures
that compose them) in the processing of information that could be linguistic, as well
as visual, as auditory, as spatial and (that is an evolution) as social (Cf. equally
Sperber, 1994). The extension of the modularity to the processing of a largest set of
representations allows this approach to account for higher-level activities, needed in
communication.
By resuming this argument, Schumacher and Ma (1999) have recently defended
the interests of modularity to clarify the nature of domain representation structures
elaborated during writing activity. According to these authors, writers would gener-
ate, during writing, specific domain representations whose structural organisation
would differ according to the nature of the writing activity and to the text communi-
cative objective. Such differences in the structure of representations would be linked
to exclusive and differential writing processes, varying as a function of the task de-
mands (i.e., the communicative aims). Schumacher and Ma (1999) defend the idea
that a fine analysis of representation differences would allow to specify, in a modu-
larist perspective, the nature and the functioning of processes underlying the writing
activity.
This approach, even if it is still very exploratory, illustrates relatively well the
possible contributions of the modularist theory. The pertinence of modularism, in
the framework of writing activity, would reside less in its processing principles than
in the necessity to recentre on the characteristics of processed representations; these
representations being required as necessary conditions for processing releasings (Cf.
on this point: Stanovitch, 1990). In the study of text production, such a careful
analysis of the structure of representations would surely be profitable. It is clear, that
current writing models give only little importance to processed representations and
knowledge. For example, concerning the Planning process functioning, the Domain
Knowledge description, on which this process applies, is often vague and relatively
general (Cf. Chapter 1). It hardly allows to hypothesise about modes of processing
of this type of knowledge. Also, without rejecting interactionist principles of proc-
esses, it appears relevant to consider that the integration, in writing models, of
modularist principles could bring researchers to pose new hypotheses on the deter-
minism of the implementation of processes according to the nature of representa-
tions to be processed. More precisely, the consideration of different types of repre-
sentations, as propositional, temporal, spatial, visual, or auditory in a writing activity
(underlying the written synthesis of a set of information from multimedia docu-
ments, for example) supposes to explore the nature of the processed representation
and the influence of this nature on processes. In the same way, the recent integration
of a Working Memory modularist model (Baddeley, 1986) in writing models can
lead to more, stronger focussed research on the nature of the processed representa-
tion nature than on the nature of these processes (Cf. Chapter 5).
16.2.2 Modes of processing

The principles of a module functioning, which is released only if the output of the
preceding module is processed and consumes no or few cognitive resources, lead to
envisage the modalities of processing according to a principle called ‘Incremental
Processing’. This idea, that considers the temporal distribution of processes, has
been described by Levelt (1989) for oral production. According to this author, the
different modules of the Conceptualiser, the Formulator and the Articulator would
function in parallel, each of them processing different discourse segments or parts. It
thus concerns an incremental processing mode whose functioning relies on a system
of buffers storing the different products at the different stages of their processing
(Cf. Figure 21).
Sentence : “John played in Amsterdam last week”
Event Place Time

Conceptualizing
Formulating
Articulating
Figure 21: Incremental processing in verbal production, adapted from Levelt (1989). Copy-
right © 1989 by M.I.T. Press. Adapted with permission.
According to Levelt (1989), this principle, inspired by Kempen and Hoenkamp’s

(1982, 1987) works and equally resumed by Bock and Levelt (1994; page 949; Cf.
equally Chapter 2), would characterise the extreme speed and fleetingness of oral
language, that creates a message part by part (or clause by clause).
As regard to writing, Kellogg (1996: 59) adopts an intermediate viewpoint be-
tween a parallel approach and a sequential approach (Cf. General Introduction and
Chapter 5). Writing processes of Formulation, Execution and Monitoring could be
carried out in parallel, as long as the Working Memory capacity (here the Central
Executive) is not overloaded. Conversely, a sequential implementation of Formula-
tion and Monitoring components would be necessary when the Working Memory
capacity is exceeded: the Execution component would be able to continue to func-
tion in parallel if it is sufficiently automatised. This idea of parallel processes, being
replace by sequential processes (and vice versa), according to the availability of
cognitive resources, appears very interesting because it allows to envisage interindi-
vidual differences for processes, differences that could be directly linked to the
processing limited capacity in Working Memory. Nevertheless, Kellogg (1996) does
not describe or clarify such parallel processing modes for the three components of
Formulation, Execution and Monitoring of his model.
16.3 Conclusion on the nature and modalities of processing

The choice of such a basic approach to processing modalities – an interactionist,
symbolic, connectionist, modularist, and so on – poses the more global problem of
the nature of the functioning and the organisation of the general cognitive system. In
the more specific case of text writing activity, it is tempting to hypothesise that some
processes could be parallel, by nature or by training (as linguistic processes, for ex-
ample), while others would be (and stay) sequential because they are relatively stra-
tegic, or not automatised. Similarly, the conception of the automatic functioning of
the Generating sub-process, that is based on the principle of Spreading Activation
(Anderson, 1983b), could equally use connectionist principles of parallel processing,
while this sub-process is integrated in a sequential model (as in the case of Bereiter
and Scardamalia’s model, 1987).
In what measure can such an hypothesis of mixed nature of processes be sus-
tained and justified? Unfortunately, satisfying theoretical responses do not seem to
exist, because this choice depends on the researcher’s theoretical options, that must
be motivated by experimental arguments. Nevertheless, the experimental validation
of processing nature and modalities, mainly for those adopted in writing models, is
far from being solved. It is currently necessary to a priori define a model, with a
precise structure and modes of functioning, rather than to construct an a posteriori
model of writing on the basis of different experimental results. Of course, this fact is
certainly due to the great complexity of the writing activity, concerning both the
number of implied processes and the diversity of the knowledge transformed by
these processes. In this way, Hayes and Flower’s (1980) model has been particularly
heuristic because it proposed, for the first time, a framework (or a first summary) of
this complex activity. Nevertheless, the main part of experimental works based on
this model has mainly tried to validate the writing process described (and this is
more particularly true for the Planning process), instead of verifying the principles
of the economy and general functioning of this model. It is partly for this reason,
without any doubt that one of the current problems, in this area of research, concerns
the dynamics of processes and their control.
17 THE CONTROL OF PROCESSES

As shown in the introduction of this chapter, the notion of processing control could
considerably differ according to whether models are situated in a modularist or in-
teractionist perspective and in a parallel or sequential function of processing. After a
presentation of the nature and the different processing modes underlying verbal pro-
duction models, the different modalities of control will be considered in order to ac-
count for the strategic triggering of processes. In this presentation, the control prin-
ciples of verbal production models, relying on an interactionist approach will be dis-
tinguished of control principles inspired by the modularist theory. Thus, essentially
linked to the activity of a ‘supervisor process’ in interactionist models, the control
would usually be executed by a ‘comprehension system’ of the executed product in
modularist models. In addition, if, as for these two examples, the control is per-
formed by a specific entity, there equally exist modes of ‘intrinsic control’ of proc-
esses that are ‘self-regulated’.
17.1 Nature of control entities in interactionist models

17.1.1 Intrinsic control of processes
In interactionist models, the most evident example of an intrinsic control of proc-
esses is provided by the connectionist approach. In this type of architecture (Cf.
above), the different representational strata counterbalance each other in the pro-
gress of their transformations, via the concept of Spreading Activation. A connec-
tionist system does not inevitably need a specific control entity, because it functions
within a mode of self-regulation and equilibration of representations. Thus, accord-
ing to Berg (1986), for example, the organisation of oral production processes would
be ensured by a production system intrinsically recording the order of implementa-
tion of these processes, by listing their conditions of execution. The priority of proc-
ess implementation would then be determined by activations and inhibitions be-
tween Declarative and Procedural Knowledge used during the production activity.
The processes would be interactively controlled, by top-down (from conceptual
processing to articulatory processing) and bottom-up (from articulatory processing
to conceptual processing) influences as well as by processing stages (goals and sub-
goals) depending on the intrinsic architecture of the production system. This pro-
posal about the control and management of processes is relatively close to those
adopted by Anderson (1983a) in his ACT* model, through the structuring of proc-
essing goals and sub-goals within the set of Procedural Knowledge. This control
system is intrinsically linked to the sequencing of processes that, by activating and
inhibiting each other, are self-regulated.
In the framework of computational models, van Dijk and Kintsch (1983) suggest
a kind of control which could be related, in its main principles, to a mode of control
based on a self-regulation of different representation levels. Thus, the control would
not be executed by a separated entity but intrinsically, by a series of embedded plans
whose high level, the most influential, would be constituted by the Macroplanning
strategy. According to the authors, a top-down management mainly performs the
monitoring of the different processes, from highest to lowest processing levels (em-
bedded strategies). For example, Macroplanning (i.e., pragmatic) strategy influences
the set of subordinate processing like Macrostructural and Microstructural strategies
(Cf. Chapter 1). This kind of top-down control would possibly occur during the lin-
guistic formulation process. Hence, van Dijk and Kintsch (1983) describe local co-
herence strategies. According to these last strategies, the choice of microstructural
units depends both on the already written linguistic segment and on macrostructural
units rehearsed in Short Term Memory (Cf. Chapter 2).
17.1.2 Control of processes by a specific entity

Besides this example from van Dijk and Kintsch’s (1983) theory, in most computa-
tional writing models, the modes of processing control are often seen through an en-
tity specially dedicated to this task. The Monitoring process, initially defined by
Hayes and Flower (1980), is a good example of such a control system. In this case,
the control of processes itself is often considered as a controlled process, where the
notion of control is closely linked to that of strategy.
The Monitoring process in Hayes and Flower’s (1980) model. For Hayes and
Flower (1980), a general control system operates through the Monitoring that would
manage, in the course of writing activity, the temporal sequencing of processes and
sub-processes by defining rules of priority. According to the authors, if a general
priority were given to the handwriting (in order to empty the content of the mne-
monic register) and to retrieving the information from Long Term Memory (to sup-
ply the system with new knowledge units), the interruption of these two processes
would systematically release the implementation of the Organisation sub-process.
These general processing modes allow the elaboration of different strategies linked
to different kinds of process recursion and interaction. Hayes and Flower (1980)
have distinguished four recursive prototypic strategies linking different processes,
respectively labelled: ‘Depth first’, ‘Get it down as you think of it, then review’,
‘Perfect first draft’, ‘Breadth first’.
These strategies could be found in more or less expert and costly writing produc-
tion management (Cf. Table 3). The 3rd to 6th production rules (in italics), called
‘Goal Setting productions’, are considered as the really strategic part of the Monitor-
ing. They consist, for the writer, of choosing processing goals that characterise the
adopted writing strategy. As seen in Table 3, one of these strategies, for example,
consists in entirely and correctly elaborating the text content before handwriting and
revising it (Breadth first). Conversely, it is equally possible for the writer to ‘throw
ideas on the paper’, as soon as they are retrieved, then to proceed to the content or-
ganisation in revising the draft produced so far (Get it down as you think of it, then
review). If it is therefore possible that the Domain knowledge processing (from re-
trieving to linearising) could be entirely performed before realising the linguistic
translation, this strategy would be costly and only used by expert writers.
According to Wason (1980), one of the possible strategies would be to segment
the writing process in several cycles of processing, performing each time on a lim-
ited set of knowledge units. In this case, however, the absence of a global organisa-
tion of the retrieved contents would limit the possibility to evaluate the coherence of
the whole text, contrary to the preceding strategy.
Table 3: Different types of writing strategies, adapted from Flower and Hayes (1980)
Configuration 1 (Depth first)

1. (Generated language in STM -> edit)
2. (New information in STM -> generate)
3. (New element from translate -> (goal = review))
4. (New element from organise -> (goal = translate))
5. (New element from generate -> (goal = organise))
6. (Not enough material -> (goal = generate))
7. ((Goal = generate) -> generate)
8. ((Goal = organise) - > organise)
9. ((Goal = translate) -> translate)
10.((Goal = review) -> review)
Configuration 2 (Get it down as you think of it, then review)
1. (Generated language in STM -> edit)
3. (New element from generate -> (goal = organise))
4. (New element from organise -> (goal = translate))
6. (Enough material -> (goal = review))
10. ((Goal = review) -> review).
Configuration 3 (Perfect first draft)
1. (Generated language in STM -> Edit)
4. (Enough material, plan not complete -> (goal = organise))
5. (New element from translate -> (goal review))
6. (Plan complete -> (goal translate)
10.((Goal = review) -> review)
Configuration 4 (Breadth first)
1. (Generated language in STM -> Edit)
3. (Not enough material -> (goal generate))
4. (Enough material, plan not complete -> (goal = organise)
5. (Plan complete -> (goal translate))
6. (Translation complete -> (goal review))
10. ((Goal = review) -> review).
Control by a Metacognitive entity, according to Berninger and Swanson (1994). The

idea of a Monitoring process imposing different processing priorities, underlying
different writing strategies, appears entirely relevant in the case of text writing.
However, Flower and Hayes (1980) give only few details about Monitoring and,
consequently, about the determinism of writing strategies.
First, it can be hypothesised, within Hayes and Flower’s (1980) framework, that
the Goal Setting sub-process, during planning, could constitute a first control, able
to evaluate the appropriateness between text and previous goals, releasing, and guid-
ing the activity of the Monitor (by means of goal setting productions). Another pos-
sible interpretation of the functioning of the Monitor can be found in Berninger and
Swanson’s (1994) proposals. These authors have resumed and adapted Hayes and
Flower’s (1980) model in a developmental perspective (Cf. Chapter 6). They do not
include, in their own model, Hayes and Flower’s (1980) monitoring because they
consider that this entity does not only concern writing activity, but should be consid-
ered as a general system, regulating both writing processes and the functioning of
Working Memory and Long Term Memory. For Berninger and Swanson (1994), the
control would be performed by a Metacognitive process, allowing writers to proceed
to a reflective analysis of their text and to modify or systematise their writing strate-
gies as a function of this mental analysis (Cf. also, in Chapter 3, the procedural
model of revision by Butterfield, Hacker, & Albertson, 1996). Thus, the Metacogni-
tive knowledge would be neither exclusive, nor restricted to the three main writing
processes (Cf., for a close viewpoint: Hacker, 1997; or Saada-Robert, 1995).
Control by Task Schemas or Procedural Knowledge, according to Hayes (1996).

Hayes (1996), in his revised version of Hayes and Flower’s (1980) model, hypothe-
sises that the releasing and organising of writing processes would be performed by
different Task Schemas (Cf. General Introduction), acquired through the practice,
stored in Long Term Memory and dedicated to different components of text produc-
tion (Task Schema for revising, for editing, for writing a specific type of text, etc.).
A task schema would list knowledge as well as (1) the goals of the task, (2) the
choice and the sequencing of processes to reach these goals, and (3) the evaluation
criteria of the result of these goals.
For example, concerning the revising process, Hayes (1996) postulates that the
expert schema dedicated to the realisation of this process comprises a goal (to im-
prove the text), a set of processes to realise this goal (evaluative reading, problem-
solving, writing of a segment of text), some attentional sub-goals (what important is
to be attentive of in the text to be revised, what kinds of errors are important to
avoid, etc.), quality criteria (about the choice of the style, for example) and finally
strategies to correct specific problem categories concerning the text (Cf. Figure 22
and Chapter 3).
The retrieving and the releasing of these task schemas would operate as a whole, on
the basis of the environment information processing (the text produced so far, for
example), and/or through a reflective analysis throughout the course of the writing
activity. By its close nature, according to Hayes (1996, p. 16), of a production sys-
tem listing conditions of process execution, this task schema would exert an essen-
tially top-down control on the totality of processes during a given task. This would
be true for Reflection processes, as well as for processes concerning the Interpreta-
tion, or the Production of the text.
Control Revision
Structure Task
Schema
Reflection
- Problem solving
Fundamental - Decision making
Processes
Text Processing
Text Production
- Critical reading
Resources Working Long term

Memory Memory
Figure 22: Control of the revising activity by a Task Schema,adapted from Hayes (1996).
The notion of Task Schema allows describing different kinds of control of the same
processes according to the task in which these processes are involved. Nevertheless,
its schematic character, relying, by definition, on an all-or-nothing retrieval of the
totality of parameters, cannot really account for eventual inter- or intra-individual
differences between the adopted writing strategies. Similarly, Hayes (1996) does not
really specify the dynamic functioning of this control system. He does not explain
neither how the writer must organise the different mental activities during writing,
nor how the interpretation of environmental data can influence the processing com-
ponents.
To clarify these dynamic characteristics and the influence of the text produced so
far, van den Bergh and Rijlaarsdam (1999), pursuing Breetvelt, van den Bergh and
Rijlaarsdam’s (1994) work, have elaborated a control model inspired by the princi-
ples developed by Hayes (1996). In addition, this model introduces a monitoring en-
tity that could be considered as an interface ensuring a co-ordination between text
information (during its realisation) and writing instructions of the Task Schema. The
authors approach the dynamics of the writing process control through the notion of
management of different constraints whose nature could evolve in the progress of

writing. Thus, according to the phases of the activity, some processes, or more glob-
ally, some mental activities, would be more strongly solicited than others (Cf. Figure
23).
Cognitive
Knowledge
Monitor
Cognitive Cognitive
Activity C Activity A
Cognitive
Activity B
Figure 23: A dynamic model of cognitive activities,

adapted from van den Bergh and Rijlaarsdam (1999). Copyright © 1999 by the Amsterdam
University Press. Adapted with permission.
The choice of the mental activity in which the writer is engaged, in a given writing
period, would not be random but would depend on two factors. The first factor, de-
scribed as internal, would come from the writer’s ‘Procedural Knowledge’ about the
manner in which to realise and organise the different mental activities for managing
the task. This Procedural Knowledge, constituted by the writer’s previous experi-
ence and susceptibility to be taught, would determine the writer’s personal writing
strategies. These kinds of knowledge are relatively close, in their definition, to the
notion of Task Schema such as Hayes (1996) described. The second factor, external,
would come from the ‘Task Environment’ and would be mainly constituted of the
text-produced-so-far. The influence of this second factor is expressed in terms of
distance or deviation between the elaborated text and the writing objective. Accord-
ing to van den Bergh and Rijlaarsdam (1999), the choice and the order of implemen-
tation of the different mental activities would depend on the matching between Ex-
ternal Information (i.e., the text) and Procedural Knowledge. This matching would
be executed by a Monitoring entity that would manage, on the basis of this double
determination, the activation of some necessary activities, and as a consequence the
Inhibition of the other activities.
The important characteristic of this mode of control is that it relies on a probabil-
istic idea about the functioning of writing processes. The implementation of a given
activity increases the probability of realisation of the next activity. For example, the
start in a reading activity would strongly increase the probability that an editing or
correcting activity will follow. In the framework of this control system, a given writ-
ing process has not a strictly determined function but can, conversely, ensure differ-
ent functions according to the realisation context during the progress of the activity.
In this way, van den Bergh and Rijlaarsdam (1999) give four different functions to
the Generating sub-process, determined by its releasing conditions: (1) the genera-
tion of ideas as a consequence of instruction reading (Assignment Driven Genera-
tion), (2) the generation of ideas as a result of the rereading of the text already pro-
duced (Rereading Text Driven Generation), (3) the generation of ideas as a conse-
quence of the linguistic translation (Translation Driven Generation), and (4) the gen-
eration of ideas resulting from the generation itself (Generation Driven generation).
17.1.3 Conclusion on control in interactionist models

Through these different models or theories, it is obvious that control modalities can
be numerous and varied, especially in interactionist models. However, concerning
Monitoring, as well as Procedural Knowledge (Task Schema), and Metacognitive
knowledge, there is no a priori pre-established order to execute processes in interac-
tionist models. These models necessitate most of the time a specific control entity
that plays a central role in strategic choices and in the management of process inter-
action.
In converse, in the case of modularist models, where modules are encapsulated
and the processing organisation is more strongly constrained, the control entity is
rather intrinsic to processes. Furthermore, in these models, a specific control entity
would be mainly used to evaluate the products rather than to elaborate them.
17.2 Nature of control entities in modularist models

17.2.1 Intrinsic control of processes
In the case of purely modularist models, the main part of processing control is di-
rectly linked to the adequate nature of the representation that can be used as input,
by a module. Indeed, in modularist models, the encapsulated character of modules
lets little ascendancy to a voluntary, direct and conscious control of the progress of
processes. According to modularist principles, the activity of a module can be en-
gaged only if the product processed by the preceding module is consonant with
specifications of the new module. Hence, in Levelt’s (1989) model of speaking, the
information is top-down driven (there would be no direct feedback from the Formu-
lator to the Conceptualiser), and stored in many different and successive buffers to
allow the management of speaking (Cf. above). The mode of processing control
would therefore be intrinsic to the framework of modules and implicitly linked to
the nature of these modules. In that way, it would concern a very local mode of con-
trol, relying on the successive ‘filters’ that constitute the input conditions of each
module, leading to stop some processes (by error or failure of an input) and to reiter-
ate non validated processes.
17.2.2 Control by a comprehension system: a contribution of modularist theories

The nearly step-by-step control system such as described previously, does not allow
a voluntary and conscious evaluation of processes. However, some errors in verbal
production can occur, that concern local errors, as slips of the tongue, lexical term
inversions in a sentence (Cf. Garrett, 1988), or more global errors, in relation with
the content and its coherence. Instead of directly intervening processes, the neces-
sary correction of these errors requires an a posteriori analysis concerning the qual-
ity of the product or the appropriateness between the final product and the commu-
nicative goals.
In Levelt’s speaking model (1989), such a very global control is described
through a comprehension system (the Self-Monitoring). This comprehension system
allows the speaker to analyse his/her speech and to consciously decide (i.e., in a con-
trolled way) to reiterate the sequence of processes leading to the generation of a new
message (Cf. Figure 4, page 12). According to Levelt, a speaker is at the same time
her/his own listener and has access both to internal language (inner speech) and to
external language (oral discourse). Furthermore, Levelt postulates that two levels of
control are available: (1) a control from internal speech (represented in Working
Memory), via the discourse comprehension system and (2) a control from the al-
ready produced speech, via the audition and then the comprehension system.
The Self-Monitoring system can therefore control what is going to be told and
what has already been told through to two perceptual loops ensuring both a pre- (in-
ternal loop) and post-articulatory (external loop) control. The regulation can be car-
ried out before or after the Articulator component and is done through these two
loops. The first loop directly operates on the phonetic plan (internal language) pro-
duced by the Formulator. The second loop intervenes the external language, that is
to say on the articulated message and allows self-corrections. This control would be
cognitively costly and would draw its resources from Working Memory.
According to Levelt (1989), this control would act on the phonological represen-
tation of a linguistic segment already elaborated and stored in a mnemonic buffer,
until the oral execution. The control would be then put forth before or in parallel
with speaking, on the internal language at a phonetic level (phonetic plan), due to
the language comprehension system. The analysis of this final product can focus on
all its dimensions, local (lexicon, syntax) as well as global (coherence, pragmatic
aims). The detection of errors is signalled to the Conceptualiser that must entirely
plan a new message taking into account the necessary corrections.
This principle of the intervention of a comprehension system, in interaction with an

oral production system, can easily be extended to writing. It suffices, indeed, to pos-
tulate the existence of a pre- or post-graphic (during revision) control, as Scar-
damalia and Bereiter (1983) have already proposed when they analysed the revising
process (Cf. Chapter 3).
The integration of a Comprehension loop in writing models seems to be one of
the major evolutions of the most recent writing models, considered as modularist or
not. Such is the case in van der Pool’s (1995) and van Wijk’s (1999) models, this
last one (Cf. General Introduction) being considered as a rather direct adaptation for
writing of Levelt’s (1989) speaking model. Furthermore, according to Hayes (1996),
writing processes could be controlled, among others, by an evaluation of the result-
ing product, which supposes that writers read and analyse their text. This compre-
hension system (Text Interpretation process), if it is associated with a correction sys-
tem of possibly detected error, would then play a role relatively similar to that of a
revision entity, which it could replace.
Kellogg (1996; Cf. the description of his model in the General Introduction and
in Chapter 5) has systematised this approach by integrating Reading and Editing in a
global Monitoring component, thus attributing a function of control to these proc-
esses rather than a function of revision, as this term is usually defined. The interest-
ing point of this model, besides this structural characteristic, is that Kellogg (1996)
describes different releasing conditions of Monitoring processes. This description
takes into account the nature of the product that must be revised, thus resuming, in
terms of processing priorities, the ideas developed by Flower and Hayes in 1980,
concerning the different strategies of Monitoring (Cf. above). Kellogg specifies
these conditional relationships through a system of control flow between the differ-
ent components and the different processes (Cf. Figure 7; page 20). Thus, a bi-
directional flow between the Formulation and the Monitoring allows the Editing
process to appear both (1) before the Programming and the Execution of a sentence,
and/or (2) after the writer has read the sentence, the paragraph or a larger text unit
that has already been written. Kellogg (1996) resumes here the principle of a possi-
ble double control, both on a mental product and on a physical product. According
to the author, the internal language, preceding the execution, would play a central
role because it would offer an important possibility to finalise corrections before the
production (control before execution).
The possibility of the existence of such types of control could be strengthened by the
irregularity and/or the non-systematic errors, especially for expert writers or speak-
ers (i.e., errors being able to occur in a particular linguistic context, or when the
writer is cognitively overloaded; Cf. for example Negro & Chanquoy, 1999). The
role of the control entity would then be to detect the risk of error, according to the
linguistic configuration temporarily stored in a buffer, and to suppress the ambiguity
before transcription (Cf. Largy, 1995). Kellogg (1996) nevertheless states that con-
trary to speaking, writing offers the possibility to realise approximate physical pro-
ductions such as drafts or elements of the future text. This ‘memory of text’ allows
the monitoring to be executed no more from an internal speech, but on a written
trace of this internal speech. This possibility would allow proceeding to controls and
economic adjustments, in terms of cognitive load, provided the product to be con-

trolled is not necessarily maintained in memory.
18 DISCUSSION POINTS
The nature, modalities and control of processes are multiple in language production
models. But this multiplicity, as we have tried to illustrate, is often linked to theo-
retical choices adopted by authors, before elaborating their models, to formalise
processes. In the case of writing models, the computational interactionist approach,
inherent to the initial Hayes and Flower’s (1980) model, currently seems the most
privileged theory. In addition, the main part of experimental works plans most often
to validate the existence and the functioning of each writing process. Nevertheless,
the crucial problem, in the study of text writing, is less the description of the func-
tioning of a given process – which is however a necessary stage in the comprehen-
sion of writing activity – than the explanation of modalities of activation of these
processes. A control is then all the more an indispensable necessity as processing er-
rors can occur in the course of the production. It is thus mainly necessary for the
writer, throughout the activity, to supervise and regulate the totality of processes un-
til the physical and effective production of the text is completely satisfactory. If the
choice of modes of control that manages the dynamics of these processes is the most
often coherent with a general economy of writing models, the crucial problem re-
mains with the determinism of this control. While all computational models agree on
the presence of a process control, intrinsic or linked to the activity of a specific en-
tity, it is difficult today to explain how and why this control is carried out. More pre-
cisely, in computational models, it is difficult to identify releasing criteria of this
control. In this way, according to Berninger, Whitaker, Feng, Swanson and Abbott
(1996), when Flower and Hayes (1980) describe a recursion among processes, they
do not take into account the clues used by the Monitoring entity to manage the se-
quencing and the implementation of processes. On what conditions and according to
which clues does the writer control writing processing? Two types of ‘release
mechanisms’, at least, can be described: (1) the first mechanism would be linked to
the text being written, whose analysis provides clues about the nature of processes to
be executed; (2) the second type of mechanism would be linked to the constraints,
due to the limitation of cognitive capacities that would necessitate using the most ef-
ficient and economic writing strategies.
18.1 Control and analysis of the processed product: relations with the Reviewing
process
One indication that can be used by the control to manage the activity of processes is
obviously the final product, that is to say the linguistic written trace, whose analysis
must allow to determine whether the text is in accordance with writing goals fixed
during the planning process. This type of analysis can be realised by a reading com-
prehension activity on the previously written text. Nevertheless, the increasingly
marked integration, in production models, of comprehension systems that play an

important role in the control of processes, raises the question concerning the rela-
tionships between Reviewing process and Monitoring process. In most of theoretical
and empirical works, the fundamental notion of control is never compared to the no-
tion of revision. However, this could concern two very close cognitive phenomena.
For example, in Hayes and Flower’s (1980) model, a writing process is special-
ised to perform the activity of text revision, and another process carries out the
monitoring of the totality of the writing process. This monitor could therefore, in
this case, be conceived as a metacognitive entity (Cf. Berninger & Swanson’s pro-
posals, 1994).
In addition, in oral production models, as Levelt’s (1989) for example, there
also exists a control entity, that could both serve as a ‘production regulator’ and as a
‘local checker’. Consequently, it appears important to ask the following question:
which entity controls the production? Two responses are then possible:
• The Monitoring controls both the progress of the totality of the production
process and the final product. In this case, it is necessary to consider a specific
process for the Revision. This process could be conceived as a ‘mini-cycle’ of
production, once the error, whatever its importance, its nature and its localisa-
tion, has been detected by the monitor.
• The Reviewing process, with its different stages, controls the production; this
process itself is under the control of the Monitor that, in the meaning of
Butterfield and al. (1996), would allow evaluating (at a metacognitive level; Cf.
Chapter 3) if activity could be performed.
From this fundamental question, another interrogation emerges: what are the clues
that release the Reviewing process? According to models and proposals presented in
Chapter 3, the perception of an ‘error’, in a broad meaning, would release this proc-
ess. However, to spot the error, the writer must be engaged in an activity of control
of his/her production. Thus, how is this activity then impulsed? This question is
close to those relative to the automatism or control of the activity. Rapidly speaking,
in research on spelling, concerning both oral (for example: Bock and Cutting, 1992;
Bock and Eberhard, 1993) and written language (for example: Largy, Fayol and
Lemaire, 1996; Negro and Chanquoy, 1999), most authors frequently refer to a pre-
articulatory or pre-graphic control process, that would be sensitive to some seman-
tic, phonological or syntactic indices to detect errors before they are orally produced
or written on the paper. Consequently, these authors do not speak about ‘Revision’,
in the meaning evoked up until now, but about ‘Control’ that would result in avoid-
ing spelling errors (more specifically in the above mentioned works, subject-verb
agreement errors). Thus, is it possible to consider this Control as an internal Revi-
sion? It seems that the answer must be positive. However, the researchers who are
working on spelling exclusively explain their results in terms of Control, that could
or could not function as a function of available resources in Working Memory. Con-
sequently, these researchers, even though they are working on written language,
seem to completely ignore models of written production and refer, mainly by fol-
lowing works led by Bock, to oral production models as Levelt’s (1989), Bock’s
(1996) or Garrett’s (1980, 1982). This appears to be an important research track to

continue. It would be effectively important to link researches on spelling, even if
their first objective is to understand how the production is performed through the
analysis of errors, with works on writing revision and control, whose aims are possi-
ble more ‘global’.
18.2 Control and limited capacity

One of the release mechanisms of the control activity would come from the Task
Environment and would be mainly constituted by information extracted from the
text during handwriting. The interpretation of these clues by the control entity would
lead to the deliberated activation of adequate processes to continue writing activity.
However, it can be postulated that the control of processes is less of a deliberated
strategy than a necessary strategy to efficiently manage the pool of limited cognitive
resources, concerning as well resources inherent to the used knowledge, as cognitive
resources dependent on processing constraints. The limited capacity could be
equally considered as one of the main release mechanisms of this control. This idea
is far from being original since it underlies Flower and Hayes’s (1980) theory. In-
deed these authors consider that the elaboration of different strategies of processing
would always be linked to the necessity to manage a great number of constraints
loading on the writing activity.
Following this theory, Grabowski (1996) describes control and regulation mo-
dalities of the verbal production process, as including them in a general system of
information processing, largely inspired by Baddeley (Baddeley, 1986; Gathercole
and Baddeley, 1993). In this system, all language production processes are distrib-
uted. This last aspect confers to Grabowski’s (1996) model the possibility to clarify
both the control of speaking and the control of writing. The process regulation oper-
ates internally, by a flow of information established between the different entities.
Different modes of control and regulation can be realised according to cognitive re-
sources attributable to processes and to the nature of knowledge implied in the writ-
ing activity (type of texts, for example).
The architecture of linguistic production processes, such as proposed by
Grabowski (1996), is relatively close to those of Levelt (1989; Cf. General Introduc-
tion and Chapter 2). Nevertheless, the control and the regulation of the different
processing stages do not operate via a comprehension system, as in Levelt’s, but via
an exchange of information between three production systems: the Central Control,
the Auxiliary Systems and the Encoding Mechanism (Cf. Figure 24).
The Central Control is composed of two subsystems. The first sub-system (Focus
Memory) allows to attentionally focus on Declarative Knowledge (as domain
knowledge, for example), but equally on production goals, social conventions of text
or discourse, or on the addressee’s model. This first attentional subsystem can be
considered as equivalent to a zone of Working Memory allowing to activate all the
necessary representations for the verbal production. The second sub-system (Central
Executive) would be relatively close to Baddeley’s (1986) Central Executive. It
would be dedicated to Procedural processes (production system), performed on De-

clarative Knowledge activated in the attentional focus (Focus Memory). Three proc-
esses allowing the Selection, the Elaboration and the Linearisation of domain con-
tents carry out these processes, strongly controlled and depending on the processing
limited capacity. Finally, linearised contents are then semantically processed (shape)
by the Auxiliary Systems, before being linguistically translated by the Encoding
Mechanism. According to Grabowski (1996), the functioning of these last two com-
ponents, opposed to the Central Control, would be largely automatised.
Perception LTM
Central Control
Focus Memory
Central Executive
(selection, construction, linearisation)
Proto Input
Auxiliary Systems
adjustment
of Auxiliary (language dependant planning
Systems marking the Proto-Input and
generating the Encoding input)
Vertical
Encoding Input Feedback
adjustment
Encoding Mechanism
of Encoding
Mechanism (generation of grammatically ordered
phoneme sequences with supplementary
information: stress, segmentation)
Phoneme sequence with

supplementary information
Figure 24: Regulation of written and oral language production,adapted from Grabowski
(1996). Copyright © 1996 by Lawrence Erlbaum Associates. Adapted with permission.
The Central Control, by the double intervention of the attentional focus and of the
central executive, plays a central role in the system of regulation. The procedural
characteristics of the central executive allow it to consciously control the appropri-
ateness between production objectives and the effective result of each processing
stage. Thus, the Central Control exerts a top-down influence on the two other enti-
ties (Auxiliary systems and Encoding mechanism), by modulating and adjusting

their functioning (Hierarchical control). This mode of control stays top-down be-
cause the two subordinate entities cannot directly regulate a superordinate entity, but
only transmit to the superordinate level some pieces of information about the pro-
gress of their functioning (Vertical Feedback). The three systems of production and
their respective regulation flow, thus form ‘a hierarchical system with vertical feed-
back’ (Grabowski, 1996: 79). The feedback of the two subordinate registers has to
be interpreted by the Central Control, which will then directly control the function-
ing of these two registers.
According to Hermann and Grabowski (1995), this idea of regulation between
the Central Control and the two subordinate systems supposes that two control levels
co-exist, or maybe are opposed, in the progress of the verbal production (oral or
written). The first type of control, defined as ‘Global’, would be rather dedicated to
Conceptual Planning processes. It would involve maintaining the goals and commu-
nicative intentions of the production act activated, in Working Memory, under the
Attentional Focus. The second control, defined as ‘Local’, would be rather con-
cerned by Textual Planning and Linguistic Translating processes (shape). It would
execute the transfer between propositional linearised representations to linguistic
formulation operations.
The double interest of Hermann and Grabowski’s (1995) formalisation is that (1)
one these two modes of control can supervise the other (Trade Off between modes
of control), and (2) the determinism of this predominance would rely on the dis-
course or text type to be produced. Thus, the Global Control would be more strongly
solicited when the verbal production relies on highly conventional text schemas,
that, available in Long Term Memory and activated in the attentional focus, would
guide the totality of the production. This kind of Global Control can be maintained
until the adoption of linguistic utterances remains adapted to discourse conventions
or text schematic structures. Conversely, when the verbal production does not rely
on such specific schemas, the Local Control would take the place of the global con-
trol. This Local Control would operate through automatic and non-conscious opera-
tions of linguistic formulation. This control would thus be particularly costly for the
Central Control because the generation and the linguistic execution of text content
have to be determined by an elaborated analysis of the addressee, of environmental
changes and of discourse or text elements.
The regulation of writing processes, in such a system, would therefore be intrin-
sically performed, without the recourse to a specific and independent control entity,
as the Monitor or the Task Schema postulated in Hayes and Flower’s (1980) and
Hayes’s (1996) models. Indeed, if the Central Control plays a regulation role, its
mode of functioning is intimately linked to the feedback sent by the subordinate reg-
isters. In addition, the top-down and bottom-up exchanges allow the system to find a
point of balancing in its functioning, thus going to a self-regulation of the implied
processes, according to the parameters linked to attentional resources, to limited ca-
pacities and to the nature of necessary knowledge.
19 CONCLUSION ON NATURE AND CONTROL OF PROCESSING

At the end of this chapter, it appears important to us to give a reminder that text
writing is mainly a temporal (on-line) activity. The text, progressively written, is
evolving temporally and spatially during the progress of the writing activity. This
writing dynamics both leads and supposes a dynamics of writing processes. The cen-
tral question is then to define, in verbal production models, what are the principles
of processing management and control. Several factors seem very important in the
formalisation of these principles. Two of them are discussed here.
The first factor concerns the necessity for a theoretical coherence. Thus, the gen-
eral idea of the cognitive functioning adopted by authors to elaborate their own
model (modularism versus interactionist theory; parallel versus sequential process-
ing) implicitly lead to specific modes of control. The control of processes frequently
relies, in interactionist (and more specifically computational) models, on a specific
entity that supervises the process interaction and intervention. In modularist models,
an intrinsic control is linked to the fact that the releasing of a module can only oper-
ate if the output of the preceding module can be considered as its input. This intrin-
sic control, quasi-automatic, is associated (in the case of Levelt’s model, for exam-
ple) with a control external to processes that evaluates the final product. This second
mode of control, highly controlled, is close to a comprehension and/or revision ac-
tivity. This idea about control underlines the interpretation of the written (or oral)
product as an activator and a manager of verbal production processing. A theoretical
problem remains with this conception, as seen previously, concerning the relation-
ships between the notions of control, comprehension, revision and metacognition.
Another crucial factor in the determinism of processing control, beyond the
characteristics of the text produced so far, concerns Working Memory processing
and/or maintaining span. It is important to remember, with reference to Flower and
Hayes (1980) that there are mainly the limited capacities that require the writer to
juggle with constraints and to more or less appropriately articulate the realisation of
the different processes. In chapters 5 and 6 we will re-introduce this last aspect more
intensively.
Finally, as seen in the introduction of this chapter, the control process would rely, by
definition, on the implementation of two different activities: the first one regards the
evaluation of the product as a function of the writing objectives; the second activity
enables reaching these goals by determining an adequate processing strategy. Is it
thus possible to hypothesise the existence of two control systems, which would not
operate at the same level? A higher entity, as suggested by Berninger and Swanson
(1994), would control not only the written production but also the writer’s entire
cognitive system. It would be used for the writing activity, solving problems, chess
playing, and all the activities considered as very strategic. Inherent to the general
cognitive system, this global control would combine all the available cognitive re-
sources to efficiently manage not only writing processes, but also the functioning of
mnemonic registers (Working Memory, Long Term Memory), and would thus carry
out the recovery (and the storing) of the knowledge implied in writing activity. The
decision-making of this entity would be realised on the basis of information pro-

vided by the second system of control. This other ‘subsystem’, specialised for writ-
ing, would be more particularly dedicated to the evaluation of the text appropriate-
ness with writing goals and with the quality, in terms of surface and meaning, of the
text. As considered above, this mode of local control would maintain close relation-
ships with revising processes, or could be ensured by revising processes of diagnos-
tic, or text comprehension processes.
CHAPTER 5
WORKING MEMORY IN WRITING
1 INTRODUCTION
1.1 Importance of Working Memory in text writing activity

Writing processes are multiple and focus on different kinds of knowledge and repre-
sentation (domain, linguistic and pragmatic). This multiplicity needs, as seen in
Chapter 4, the implementation of management, control and regulation of processing.
Defined as a complex activity, the text writing assumes that the writer could manage
a great number of constraints (Flower and Hayes, 1980; Scardamalia, 1981). In addi-
tion, the choice of a given strategy (added to the simultaneous consideration of this
set of constraints) supposes the management of various processing levels (Graham
and Harris, 1996). Thus, one of the fundamental problems of language production
consists in the capacity to manage simultaneously and/or serially different types of
processing, while taking into account Working Memory limited capacity (Bereiter
and Scardamalia, 1984; Glynn, Britton, Muth and Dogan, 1982).
One of the main factors in the determination of writing strategies seems to be
linked to Working Memory span. The Short Term Memory (or the temporary storing
capacity of Working Memory) has been considered very early on as one of the cen-
tral factors in writing activity, as well as in Flower and Hayes’s (1980), Bereiter and
Scardamalia’s (1987), or van Dijk and Kintsch’s (1983) models. More precisely, the
role of Short Term Memory has been mainly considered through processing con-
straints, imposed by the memory span limitations. While these constraints are often
only briefly noticed by Flower and Hayes (1980), or Bereiter and Scardamalia
(1987), van Dijk and Kintsch (1983) have proposed a more precise explanation of
writing strategies, interacting with the Short Term Memory functioning. Such is the
case, for example, with the Local Coherence Strategy necessary for sentence plan-
ning and linguistic translation (Cf. Chapter 2). This supposes the simultaneous up-
holding of three different knowledge units (the macrostructural unit, the proposition
to be translated and the surface form of the previously translated proposition).
Penningroth and Rosenberg (1995) have made obvious the evaluation of the cost
to maintain a more or less important number of knowledge units. The authors, in-
spired by Tetroe (1984), have varied the amount of information simultaneously
maintained in Short Term Memory, to proceed to the writing of a narrative text. To
perform of which the authors proposed to forty students to write a text whose last
sentence was imposed. This sentence comprised a more or less important number of
D. Alamargot & L. Chanquoy (2001). Working memory in writing. In G. Rijlaarsdam (Series
ed. ) & D. Alamargot & L. Chanquoy, Studies in Writing: vol 9. Through the Models of Writ-
ing, 157 – 186. © 2001 Kluwer Academic Publishers. Printed in the Netherlands.
arguments, representing units of information that it was necessary to maintain and

organise, during the whole text production, to respect the meaning of the last sen-
tence. The nature of writing processes and their cost have been evaluated through an
analysis of performances to a secondary task, combined with an analysis of verbal
protocols collected during writing (i.e., close to Kellogg’s method, 1987a). The au-
thors have observed that the increase of the number of units to maintain led to a de-
crease in the quality of the text, both for its coherence and cohesion. This alteration
was all the more marked that the number of arguments exceeded the Short Term
Memory span (as defined by Miller, 1956 or Simon, 1974). Providing more than
seven arguments to organise would strongly increase the mental resources necessary
for the Planning process, and would lead to a qualitative change in the temporal
management of the writing processes. Hence, the Planning process is not correctly
managed during the production and the Reviewing process is then more solicited at
the end of the period of handwriting. However, according to Penningroth and
Rosenberg (1995), this modification of the writing strategy does not allow to main-
tain the quality of the text content.
It seems therefore, through this experiment, that the role of Short Term Memory
limitation, in interaction with the number of constraints to manage, could be an im-
portant factor to determine the quality of a text. Jeffery (1996) and Jeffery and Un-
derwood (1996) have also demonstrated the influence of Working Memory on writ-
ing at a conceptual level. They have shown that the more or less important manipu-
lated cognitive load could more or less facilitate ideas of recovering and organising.
Finally, for McCutchen (1996), the capacity to manage the three processes of writ-
ing (such as defined by Hayes and Flower, 1980) depends largely on the efficiency
of these processes and on storage constraints that they impose to Working Memory
(Cf. Chapter 6; Cf. also Kellogg, 1994). The ‘orchestration’ of the three writing
processes would be thus constrained by the attentional limits of Working Memory
(McCutchen, Covill, Hoyne, & Mildes, 1994).
Two important facts emerge from these works. On the one hand, the role of
Short Term and/or Working Memory has been evoked very early on or considered in
the activity of text writing, even if there is currently, in the literature, an increasing
number of works dealing with this phenomenon. On the other hand, in these works,
there are as many definitions of Working Memory as works on it. While Hayes and
Flower (1980) mainly speak of the maintaining span of Short Term Memory, Pen-
ningroth and Rosenberg (1995) are interested in the mental load in Short Term
Memory and McCutchen et al. (1994) focus on the notion of Working Memory and
attentional cost. This diversity of approaches and theoretical conceptualisation about
Working Memory, in the study of writing is certainly due to the fact that several
definitions of Working Memory have been proposed and currently coexist, in the
more general framework of cognitive psychology.
WORKING MEMORY IN WRITING 159
1.1 Diversity of the definitions of Working Memory

Historically, the concept of Working Memory in cognitive psychology has been
characterised by important theoretical evolution. Currently, there is not one precise
definition of this concept. Logie (1996), through a review of questions, has distin-
guished seven different definitions of Working Memory (or Short Term Memory).
According to this author, the philosophical concept of memory, conceived as (1)
‘Comptemplation’ (Locke, 1690), has been replaced by a first psychological descrip-
tion of (2) ‘Primary Memory’ by James (1905). Then Waugh and Norman (1965)
have defined this concept (as opposed to the permanent secondary memory) by its
(3) ‘Limited Capacity of Storing’ and temporary maintaining of information. A more
complete definition has been provided by Atkinson and Shiffrin (1968) who,
through the concept of Short Term Memory, have introduced some control processes
(strategy for coding and retrieving information) added to temporary information up-
holding processes (rehearsal). According to this idea, Short Term Memory is con-
sidered as a ‘unique and flexible system’, ensuring the storage as well as the strate-
gic processing of information. A fourth idea of Working Memory, conceived by
Craik and Lockhart (1972) and opposed to the previous one, relies on the notion of
(4) ‘Processing Depth’. According to Logie (1996), these authors have rather ap-
proached Working Memory as a processor that more or less emphasises the proc-
esses than as a real structure of memory. A fifth definition of Working Memory has
been more recently proposed by Just and Carpenter (1992), who consider Working
Memory as specialised in language processes. According to this ‘Capacity Theory’,
Working Memory would be mainly constituted by a (5) ‘Pool of Cognitive Re-
sources’ (an Activation pool) that it is necessary to share between the different proc-
esses, for example, of reading and comprehension. The central point of this ap-
proach is that the allocation of resources to a given process would operate to the det-
riment of resources allocated to another process. A sixth idea of Working Memory
relies on narrow relationships that this concept maintains with Long Term Memory.
Working Memory can be conceived as an (6) ‘Activated Zone of Long Term Mem-
ory’ (Cowan, 1993; Anderson, 1983a and b), or as completed by an interface be-
tween Long Term and Working Memory, which allows Working Memory to strate-
gically extract from Long Term Memory knowledge necessary for a given task. This
interface, recently discussed by Ericsson and Kintsch (1995) is called Long Term-
Working Memory. Finally, Baddeley (1986, 1990) has provided a seventh and last
definition of Working Memory. It is different from the preceding ones since it ap-
proaches Working Memory as a (7) ‘Composite System of Modules’, with three
main components, dedicated to process different pieces of information (visuo-spatial
and auditory), and each characterised by specific storing and/or processing capaci-
ties.
Even if, today, the concept of Working Memory is often associated with
Baddeley’s conception (1986), the two preceding conceptions (Working Memory as
a pool of specialised cognitive resources and Working Memory as an Activated
Zone of Long Term Memory, or completed by the concept of Long Term-Working
Memory) are always investigated and can be considered as competing approaches
with Baddeley’s theory (1986). We will have the opportunity, in this chapter as well
as in Chapter 6, to describe more precisely these three different conceptions and to
envisage in what measure each of them can be relevant in the framework of the
study of text writing.
1.2 Precising the definition and the role of Working Memory in text writing
Models that attempt to define and/or to specify the role and the impact of Working
Memory in text production can sometimes show a certain heterogeneity concerning
the definition to characterise Working Memory. These differences are mainly due to
the fact that these models are largely inspired by the theoretical ideas of Working
Memory in the more general area of cognitive psychology. The problem (or, con-
versely, the interest), as previously highlighted, is that these ideas are themselves
heterogenous and are very rapidly evolving. The question is then to determine which
memory model is the most adapted and/or the most heuristic to explain the specific
characteristics of written production. It seems that the consideration, in the first
models of text writing (Hayes and Flower, 1980; van Dijk and Kintsch, 1983) of the
role of Short Term or Working Memory has mainly been approached in terms of
limitation of cognitive resources and storing capacity. The role of Working Memory
in text writing has stayed relatively obscure for a long time and close to a ‘black
box’ whose processing and storing modalities, although evoked, were not systemati-
cally clarified. By the number and the diversity of processes that underlie the writing
activity, a more differentiated approach was required. This has notably enabled dis-
tinguishing process costs, according to their conceptual or linguistic nature, and to
specify constraints that weigh on the execution of these processes, according to the
nature of the processed knowledge (i.e., phonological, visuo-spatial or semantic). It
was indeed necessary to go from a passive and descriptive approach of Working
Memory constraints to a dynamic and explanatory approach of the functioning of
these constraints (Cf. Torrance & Jeffery, 1999).
Until the beginning of the 90s, such research relative to the dynamic role of
Working Memory in language activities was limited to the analysis of comprehen-
sion and reading (Cf., for example, Just & Carpenter, 1980, 1987). It is only very re-
cently that writing researchers have been interested in Working Memory and in the
different constraints that it makes weigh on writing, according to the different proc-
esses. This recent approach focuses more on the definition of the nature of mne-
monic processes than on limited Working Memory cognitive resources. This more
qualitative orientation of studies on the role of Working Memory in the activity of
text production has been enhanced by the recent models proposed by Hayes (1996)
and more especially by Kellogg (1996). This one has elaborated an architecture
combining a writing model (Brown, McDonald, Brown, & Carr, 1988) and a Work-
ing Memory model, as conceived by Baddeley (1986).
2 INTEGRATION OF BADDELEY’S MODEL OF WORKING MEMORY IN

MODELS OF TEXT WRITING
2.1 Hayes’s proposals (1996)
According to Hayes (1996), Working Memory plays a central role in the activity of
text writing and occupies, symbolically, the central position in his revised model
(Cf. General Introduction, page 17). Within this model, all writing processes are in
relation with Working Memory, with knowledge stored in Long Term Memory
(Domain and Linguistic knowledge, knowledge about the Genre of the Text, etc.),
and equally with the writer’s motivations and affects. Hayes (1996: 8) postulates
that the potential framework of Working Memory, that plays the role of a maintain-
ing and processing interface of these different kinds of knowledge, can be relatively
close to that defined by Baddeley.
2.1.1 Baddeley’s (1986) Working Memory model

Baddeley and Hitch (1974), as well as Baddeley (1986, 1990), have described
Working Memory as a temporary storing and processing system of information, im-
plied in most of cognitive tasks and composed of three major components: the ‘Cen-
tral Executive’ and two slave systems, the ‘Articulatory Loop’ and the ‘Visuo-
Spatial Sketchpad’.
• The Central Executive is the most important system of Working Memory, even
if it still remains vague and underdeveloped on a theoretical viewpoint (Cf.
Baddeley, 1996; Ehrlich & Delafoy, 1990). Its functions essentially concern the
spreading of attentional resources. It is in this view very close to that of the Su-
pervisory Attentional System (S.A.S.) described by Norman and Shallice
(1980). It equally controls the regulation of the flow of information circulating
in Working Memory, the retrieval of information situated in other systems (as
Long Term Memory), the information processing and storing. Globally, the role
of the Central Executive is to manage some activities and to inhibit some others,
so as to avoid exceeding the limited capacities of the processing system. To
help it in these different tasks, the Central Executive has two slave systems that
it controls and co-ordinates:
• The Phonological-Articulatory Loop is specialised in the processing of verbal
information, and allows to store and to manipulate acoustic and verbal equip-
ment (storing and maintaining by rehearsal, Cf. Gathercole & Baddeley, 1993).
It is built with a unit of phonological storing, able to contain linguistic informa-
tion and a process of articulatory rehearsal (or recapitulation), relying on the
subvocal self-rehearsal, that allows to maintain verbal information by repetition
(Saito, 1997) and that is equally used to phonologically encode non phonologi-
cal entries (Cf. Gathercole & Baddeley, 1993).
• The Visuo-Spatial Sketchpad allows the short term storing of visual and spatial
information, as well as the elaboration and the manipulation of mental images
(Cf. Logie, 1995 for a more detailed description of this system).
The Working Memory can thus be considered as a limited capacity system, allowing
both the upholding and the processing of information of different natures: visuo-
spatial and phonological (for slave registers), conceptual and semantic (in the case
of one of the functions of the Central Executive).
2.1.2 Hayes’ (1996) integration of Baddeley’s conception of Working Memory

While globally postulating the existence of such an architecture of mnemonic proc-
essing as a basis for writing activity, Hayes (1996) nevertheless underlines the ne-
cessity to proceed to some modifications (Cf. Figure 6 page 17) of Baddeley’s model
(1986). On the one hand, Hayes does not regard processes like planning and deci-
sion-making as constituents of the Central Executive (what could be the case follow-
ing Baddeley’s reasoning, 1986), but as elements of the more general process called
Reflection, external to Working Memory. On the other hand, the author includes,
contrarily to Baddeley (1986), a supplementary register in Working Memory whose
function is to ensure a semantic storing (Semantic Memory). According to Hayes,
this component would be more particularly used in the case of the Translating proc-
ess and more precisely of linguistic formulation. This semantic buffer would allow
storing the semantic preverbal message, while lexical and grammatical structures
would be stored in the Articulatory Loop. The matching of these two semantic and
linguistic representations would lead to the elaboration of a verbal message (inner
speech) in the Articulatory Loop. This message will be graphically executed if it is
judged relevant.
Thus, the integration of Baddeley’s (1986) Working Memory model in Hayes’s
(1996) writing model allows the elaboration of relatively fine hypotheses on the
management of processes devoted to Translating and formulation, that are still quite
unknown or not very experimentally explored (Cf. Chapter 2). Nevertheless, this in-
tegration still remains a hypothesis or a speculation, and it would be necessary to
empirically justify the integration of a supplementary Semantic component (Cf. for
this last point: Baddeley, 2000). If Hayes’s proposals (1996) nevertheless provide
interesting tracks, the model developed by Kellogg (1996) comprises a more predic-
tive ambition. It aims at explaining the functioning of the totality of the most impor-
tant writing processes within Baddeley’s (1986) Working Memory .
2.2 Kellogg’s model (1996)

2.2.1 Some principles of the model
As described in the General Introduction of this book, Kellogg (1996), inspired by
the model of written production elaborated by Brown, McDonald, Brown and Carr
(1988), distinguishes three writing components, each comprising two basic proc-
esses: Formulation (with Planning and Translating), Execution (with Programming
and Executing) and Monitoring (with Reading and Editing). These different proc-
esses are related to different Working Memory registers, as a function of their re-
spective processing functions (Cf. Figure 7, page 20 in General Introduction). In his

model, the Planning process necessitates the Visuo-Spatial Sketchpad (for storing)
and the Central Executive (for processing). The Translating et the Reading processes
requires the Articulatory Loop (for maintaining) et the Central Executive (for proc-
essing). The Programming and Editing processes necessitate, according to Kellogg,
the Central Executive only if their functioning is not sufficiently automatised.
Thus, according to the author, five out of the six writing processes (Planning,
Translating, Programming, Reading and Editing – excluding Executing) inevitably
need one, two and sometimes all of the three Working Memory components. The
spreading of writing processes in mnemonic registers can be summarised in the fol-
lowing Table 4.
Table 4: Resources of Working Memory used by the six basic processes of writing,
adapted from to Kellogg (1996).
Working Memory Resource

Basic Processes Visuo-Spatial Central Phonological
Sketchpad Executive Loop
Planning 9 9
Translating 9 9
Programming 9
Executing
Reading 9 9
Editing 9
The Central Executive plays a central role and intervenes in the execution of practi-
cally all writing processes (except the execution process, for the expert writers).
Conversely, and very interestingly in terms of experimental predictions, the storing
capacity of the Articulatory Loop is, for Kellogg, only solicited during the activity
of Translating and Reading processes, that is to say of processes that need (at least)
some linguistic knowledge. In opposition, the Planning process is the unique process
to use storing capacities of the visuo-spatial sketchpad. Kellogg (1996) justifies the
recourse to this register by postulating the fact that creating ideas and retrieving
them from Long Term Memory can necessitate mental imagery.
According to Kellogg, all the writing processes can function in parallel as long
as processing limited capacities or available cognitive resources are not overloaded
(Cf. Chapter 4). However, in terms of cognitive load, the different writing compo-
nents or processes have not the same cost. The highest demand does not come from
Reading, but from the Formulation component, that is be more particularly expen-
sive for the Central Executive. The Formulation represents the heaviest cost for the
three components of Working Memory because, in the architecture of Kellogg’s
(1996) model, the Formulation covers the Planning as well as the Translating proc-
ess. This process would thus ask more attention and would consume more cognitive
resources than the Execution of the message. Kellogg postulates that this last proc-
ess is less expensive (at least with adults), which permits other processes to appear
simultaneously to the graphic transcription (Cf. the empirical results of Chanquoy,

Foulin, & Fayol, 1990 and Power, 1986).
Kellogg’s (1996) integration of Baddeley’s (1986) model of Working Memory in
the architecture of writing models is finally relatively simple in its principles. This
should not hide two main interests of this work:
• On a theoretical perspective, the distinction between visuo-spatial and verbal
processing (respectively ensured by the Visuo-Spatial Sketchpad and the
Phonological Loop) allows to differentiate, during writing, conceptual
processing (Planning process) and linguistic processing (Translating process).
• On an experimental viewpoint, the paradigm of interfering tasks, largely used
by Baddeley (1986), can be easily adapted to the study of writing. This para-
digm, possibly associated with an analysis of performances to secondary tasks,
allows to make obvious and to describe the temporal management of these lin-
guistic and conceptual processes.
2.2.2 Experimental validation of Kellogg’s (1996) model

This double interest has led to a growing amount of research. These works currently
try to validate the model elaborated by Kellogg (1996), for all the writing compo-
nents as well as for Working Memory slave systems and the Central Executive. In
search for the respective processing cost in Working Memory of the different main
writing processes, these works nevertheless suppose to firstly evaluate writers’ writ-
ing, speaking or reading spans. The processing and storing span test generally used
in cognitive psychology is the test developed in the case of text reading and compre-
hension by Daneman and Carpenter (1980, the ‘Reading Span Test’). This test is
now equally used in writing research.
Levy and Ransdell’s experiments. The experiments led by Levy and Ransdell
(1996), and Ransdell and Levy (1999), can be situated in this framework. According
to Levy (1997), Kellogg’s model is one of the first theoretical models that it is pos-
sible to test since at least ten years. However, this model is still too recent to be, as
Levy, so assertive, but the approach that he and his collaborators are using appears
particularly relevant to make the important role of Working Memory in writing ap-
parent. This method consists of varying the nature of interfering tasks, that can load
one (or more) of the three components of Working Memory, in order to specify the
writing temporal management and the cost of processing in Working Memory. By
manipulating the nature of the writing tasks, Levy and his colleagues plan to estab-
lish a real ‘cartography’ of writing processes, according to their respective cost.
For example, Ransdell and Levy (1996a) have, through a review of questions,
shown some relationships between different Working Memory span measures
(speaking and writing span tests) and different measures of writing, notably the
quality of productions, evaluated through the Six-Subgroup Quality Scale (SSQS,
Ransdell and Levy, 1996a: 102-105). They have equally shown that a competing
task allocating the Central Executive decreases the fluency (decline of 40%) and the
quality of productions.
More precisely, to analyse the role of the Articulatory Loop and the Central Ex-
ecutive during the written production, Ransdell and Levy (1996b) have observed the
effects of several added tasks on different textual and temporal parameters. Their
main objective was to provide a clear overview of the manner in which adult writers
allocate resources to the different writing processes, by referring to the architecture
of Kellogg’s model (1996). In three experiments, Ransdell and Levy have postulated
that individual differences in Working Memory capacity (measured by different
scores in writing span tests) could predict the fluency and the quality of written pro-
ductions. In addition, they have precisely analysed how some supplementary tasks
can disturb the writing processes and modify the progress of writing (Cf. also Rans-
dell & Levy, 1996a). In order to verify these claims, the authors have asked adult
participants to write down, in a limited time, argumentative texts, while performing,
or not, a secondary task.
• In the first experiment, participants had (or had not) to listen to an irrelevant
speech, while they were writing their argumentation. The main hypothesis was
that the writers had to slow their writing down in order to maintain the quality
of their text while they were subjected to hearing a speech throughout the pro-
duction.
• In the second experiment, while they were writing, participants had to carry out
competing task (to perceive and to remember a series of six digits).
• In the third experiment, the adults were writing while they were listening to an
irrelevant speech, in which there were target words. They had to then take three
decisions (phonological, semantic or spatial) concerning the characteristics of
these words. Moreover, they had to repeat target words after their decisions. As
in the second experiment, the secondary task was expensive and heavy.
The results of the first experiment showed, contrary to the authors’ hypotheses, that
there was no effect of the irrelevant speech on the text quality. However, a decrease
of the production speed occurred during the irrelevant speech. Thus, this irrelevant
speech would modify the progress of the writing process, notably by decreasing the
speed and increasing the pause duration (increase of 34.8%). According to the au-
thors, an overload of the Articulatory Loop would disturb some writing processes,
especially Planning and Translating processes. The results of the second experiment
revealed that the cost associated with the digits disorganised the writers’ activity
more than the irrelevant speech did. This result is interpreted as reference to the
Central Executive. This secondary task would load the Central executive even more
and, consequently, would more importantly disturb the progress of the writing proc-
ess. The added task of the third experiment was equally very expensive in cognitive
resources. In addition, the writers with a high Working Memory span have not pro-
duced the best quality texts compared to writers with low or moderate spans, and the
results of the writing span test were not correlated with the text quality, but only
with their fluency (Cf., for opposed results, Jeffery, 1996).
Levy and Marek’s experiments. The principle of these first experiments carried out
by Ransdell and Levy (1996a) was recently used by Levy and Marek (1999). These
authors have proposed a series of experiments to analyse the role of the Phonologi-
cal Loop or, more generally, of a verbal sub-system of Working Memory during
writing. For Levy and Marek, the Phonological Loop may influence Translating and
Reading processes but not Planning, Programming, Execution and Editing processes
(Cf. Kellogg, 1996). In their experiments, writers were listening to irrelevant speech
while they were performing different tasks focusing on the different writing proc-
esses mentioned above.
• The first experiment aimed at studying the Execution component. For Kellogg
(1996), the Execution has no direct link with the Phonological Loop. The results
indeed showed that the irrelevant speech had no effect on writers’ performances
during an execution task.
• In the second experiment, the authors wanted to study the Monitoring compo-
nent. For Kellogg, the Phonological Loop is used for Reading and not for Edit-
ing. In this experiment, the participants’ task was to highlight errors in a text
and to specify their category. The results showed that the same percentage of er-
rors was detected with or without irrelevant speech.
• The third experiment has been led to analyse the Formulation component. For
Kellogg, the Formulation component places the heaviest demands on Working
Memory because it needs the Phonological Loop, the Visuo-Spatial Sketchpad
and the Central Executive. The participants had to make a sentence with 5
words. In the irrelevant speech condition, participants used less words and made
sentences of lower quality.
• In the last experiment, the authors have replicated the third experiment with an
extra condition: the irrelevant speech can be a story or the words used in the
story but randomly presented. The two irrelevant speech conditions had the
same effect.
All these results are consistent with Kellogg’s (1996) model: the Formulation com-
ponent is assigned by irrelevant speech and it is not the case for the other processes.
However, Chanquoy (1998), replicating these experiments using irrelevant speech
with children and adults, did not find these results: the irrelevant speech condition
never led to a decrease in writing measures.
Lea and Levy’s experiments. In another series of experiments, Lea and Levy (1999)
have tried to clarify the respective functioning and role of the Articulatory Loop and
the Visuo-Spatial Sketchpad. In order to reach this objective, the authors have clia-
med that the two Working Memory slave registers are entirely autonomous and
function with two mutually inaccessible pools of resources for verbal and visuo-
spatial processing and storage. To verify this postulate, three experiments were
elaborated.
In the first experiment, the authors wished to verify if the two slave systems are
engaged in a binary fashion or in a graded fashion, by using a dual-task paradigm
with varying degrees of difficulty for the secondary task (two types of irrelevant
speech). Indeed, their results showed that the phonological loop can be considered as
a pool of resources that can be used in a graded fashion.
The second and third experiments investigated the contribution of visuo-spatial
and verbal resources for writing in dual-task paradigms. The second experiment
aimed at showing the contribution of the Phonological Loop with another type of
secondary task: this task consisted of presenting participants with large alphabetical
or numerical characters and participants had to signal when two identical characters
successively appeared, while writing an essay. Results showed that, with the dual-
task condition, the quality was lower and the number of words produced per minute
strongly decreased.
The last experiment analysed the Visuo-Spatial Sketchpad. Instead of presenting
digits and letters, participants were presented with arrows that pointed in different
directions which they had to remember. As in Experiment 2, the performances sig-
nificantly decreased from single to dual-task conditions, for all the measures (quality
and words produced per minute).
According to these results, the two slave systems could operate in a continuous
or graded fashion and thus must not be considered as all-or-none mechanisms. In
addition,
‘when the loading tasks were performed at the same time that participants engaged in
text production, the phonological task clearly interfered more than the visuo-spatial
task.’ (Lea and Levy, 1999: 80).
Conclusions on validations of Baddeley’s model. Globally, the experiments led by

Levy and his collaborators reveal that Working Memory contributes both to the flu-
ency of written production (due to the Phonological Loop register) and to the global
text quality (via the Central Executive register). Indeed, when these different Work-
ing Memory components are solicited by a secondary task, whatever it is, perform-
ances associated with the written production strongly decrease.
However, although their results are very interesting and promising, these ex-
periments are still far from accounting for the location of all writing processes in the
different Working Memory registers, and still farther from envisaging the possible
competitions between these processes. It thus remains difficult to clearly define the
links between each writing component and process, and the three Working Memory
registers. This difficulty can be greatly linked to Kellogg’s theoretical proposals to
define his model. For example, situating Planning processing in the Visuo-Spatial
Sketchpad supposes that the generated and (re)organised representations are both
visual and spatial. This claim is not so evident. Indeed, if it could be proven in spe-
cific text types (such as descriptive texts), it does not allow to account for the plan-
ning of propositional or conceptual knowlegde units. For the processing of these
kinds of knowledge, the integration in Baddeley’s (1986) model, as suggested by
Hayes (1996), of a Semantic register seems more judicious (we will see later that
this integration is now also postulated by Kellogg, 1999). Another limitation of Kel-
logg’s (1996) model, that is able to explain that there is still no clear and powerful
validation, is the relative imprecision concerning the definition of the Central Execu-
tive role. Indeed, if most writing processes (except for the Executing process) are
carried out by the Central Executive, as it is postulated by Kellogg (1996), it appears
more important to describe processing modes led by this component, than to explain
how phonological and visuo-spatial representations are maintained in slave registers.
Finally, if the Working Memory concept, such as defined by Baddeley (1986), al-
lows or will allow to precisely describe the nature of representations processed dur-
ing writing, this concept is however limited concerning the management of cognitive
resources during the activity (in the Central Executive).
Thus, if we consider that Baddeley’s (1986) model is only one of the possible
ideas about Working Memory, it is fundamental to surround the advantages and the
limits of these different theories, by analysing the hypotheses they generate to ex-
plain the functioning of writing processes. It is in this framework that the choice of
Baddeley’s (1986) model in Kellogg’s (1996) writing theory can be discussed.
2.3 Discussion on the choice of Baddeley’s model (1986)

Although relevant on many points, the choice of Baddeley’s model in the framework
of text writing is far from finding a theoretical justification. As stated by Kellogg
(1996: 71) to conclude the presentation of his model, ‘Baddeley’s approach to
Working Memory has born fruit in understanding memory, reading, comprehension
and production of speech. Extending it to writing is a plausible step’. The plausibil-
ity of such an ‘extension’ has been strengthened by the possibility to adapt to the
study of text writing, and this without important changes, the methodology devel-
oped by Baddeley to test his model (paradigm of interfering tasks, irrelevant speech,
simultaneous articulations, tracking of spatial information, etc.). As nevertheless un-
derlined by Kellogg (1996: 71), ‘(one) of ultimate interest to the field of cognitive
psychology is whether Baddeley’s version of Working Memory is superior to com-
peting approach’. This remark underlies an epistemological question: (1) Is it
important, in current research, to validate Baddeley’s model (1986) in a particular
activity as text writing, or (2) is it necessary to explore the cost and the dynamics of
writing processes? The consequences linked to one or another of these choices are
completely different.
In the first case, the consideration of Baddeley’s theory supposes to focus on reg-
isters functioning in order to (a) validate them in the framework of writing research,
before (b) analysing the writing processing nature that could be more specifically
carried out in one or more of these registers. In Levy and colleagues’ experiments,
the reality and the functioning mode of writing processes are admitted and not dis-
cussed. The question is then how Working Memory registers are implied in writing
processes. Then the most frequent experimental procedure consists of overloading
either one or the two slave registers, so as to observe some global effects on writing
activity.
Conversely, in the second case, it is necessary to focus on the analysis of one
specific writing process to surround its processing modalities within Working Mem-
ory. This approch can be illustrated by McCutchen (1994, 1996) who, before Kel-
logg’s (1996) model, was interested in the relations between written production and
Working Memory limited capacities (Cf. also Daiute, 1984). McCutchen’s objec-
tives were not to surround Working Memory functioning in the framework of writ-
ing activity but, conversely, to explain the functioning of writing processes, knowing
that Working Memory has limited resources. For example, in focusing their research
on the functioning of the Translating process, McCutchen, Covill, Hoyne and Mildes
(1994) have shown that the more or less important accessibility of linguistic knowl-
edge, indispensable during the activation of this process, interacts with verbal mem-
ory span (measured with Daneman and Carpenter’s test, 1980), and could consid-
erably influence the quality of children’s written texts (in terms of content organisa-
tion, syntax and lexicon variation, length of sentences and so forth). To interpret this
effect, McCutchen did not use Baddeley’s (1986) model but focused on the ‘Capac-
ity Theory’ developed by Daneman and Carpenter (1980) and Just and Carpenter
(1992, Cf. the introduction of this Chapter, Chapter 6 and, for a review, McCutchen,
1996, who discusses the interest of this theory to explain the development of writing
expertise). According to this theory, the automatisation of some operations (such as
lexical retrieval) would free cognitive resources that can thus be used for other op-
erations (such as grammar processing) or for other processes (Revising, for exam-
ple). However, this ‘exchange’ within cognitive resource sharing is constrained by
the general pool of available cognitive resources of each person (evaluated through
the Speaking Span Test, for the Translating Process). The execution of one particu-
lar writing process would thus depend both on: (a) limited capacities of cognitive re-
sources and (b) resources released by other writing processes.
Finally, this opposition between two different approaches, concerning both the idea
and the role of Working Memory in writing, is interesting because it brings another
fundamental problem. This problem is how to know which can be the most efficient
model of memory, on a theoretical plan, to explore this or that dimension of the
writing activity. It is probable that the answer is not all-or-nothing and that the dif-
ferent frameworks or conceptions about Working Memory (and maybe concerning
the whole architecture of the cognitive system), could be fruitful in explaining proc-
ess levels or different aspects of text writing. In other words, different models of
Working Memory could be combined to study the processing nature and to precisely
analyse the role of cognitive resources.
3 DIFFERENT APPROACHES TO WORKING MEMORY

IN TEXT PRODUCTION MODELS
Baddeley’s (1986) model implements a Working Memory system that is independ-
ent of Long Term memory. The architecture is multi-component (the Central Execu-
tive and the two slave registers; Cf. supra), and the different available cognitive re-
sources would depend on the processed areas (visuo-spatial, semantic or phonologi-
cal). These theoretical options are nevertheless not the unique possible choice. In
fact, as seen at the beginning of this chapter (Cf. Introduction part), different ideas of
Working Memory and Short Term Memory coexist since the beginning of the 70’s,
and they are always debatable (Cf. for a review: Richardson, Engle, Hasher, Logie,
Stoltzfus, & Zacks, 1996; Gaonac’h & Larigauderie, 2000).
Historically, the principle of a sharing of the cognitive system in one register
specialised in the temporary storing-processing of information and in another regis-
ter for Long Term storage (i.e. the principle adopted by Baddeley & Hitch, 1974),
has been initially proposed by Miller, Galanter and Pribram (1960) and by Atkinson
and Shiffrin (1968). Nevertheless, in the same period, Norman (1968) rejected this
separation by postulating that Short Term Memory and Long Term Memory were
two different aspects of the same storing system. The Short Term system would al-
low a temporary activation of knowledge units that are stored in a permanent way in
Long Term Memory. The theoretical and experimental investigation on this alterna-
tive proposal has led to the elaboration of new models, labelled ‘Activation Models’.
These models can be described through two slightly different trends, but they glob-
ally opposed to Atkinson and Shiffrin’s (1968) and Baddeley’s (1986) approaches.
The first trend was based on Lindsay and Norman’s (1972) and Anderson’s
(1972) proposals, that modulated Norman’s (1968) initial position, by hypothesising
that Working Memory would be a register in the cognitive system, but that the con-
tent of this register would be composed of directly activated contents from Long
Term Memory (Logie, 1995: 126). Working Memory can thus be considered as a
‘window’ allowing to ‘read’ conceptual knowledge from Long Term Memory.
The second trend, initiated by Anderson and Bower (1973) is close to Norman’s
(1968) idea. These authors consider that there exists no separation between Working
Memory and Long Term Memory. Consequently, only the activated elements in
Long Term Memory can become conscious.
By their different ways to envisage the functioning of Working Memory, the Ac-
tivation models enable the approach and study of some aspects of text writing dif-
ferent from those that Baddeley’s model (1986) gives access to. While this last
model mainly allows to make some hypotheses about the nature of processes in
Working Memory, the Activation models allow to generate hypotheses concerning
the dynamics and the releasing conditions of these processes. The theoretical princi-
ples of these models will first be defined and then the implications of the use of
these models on the study of text writing will be discussed.
3.1 The Activation models as an alternative

3.1.1 Principles of Activation models
The Working Memory models relying on the notion of Activation comprise a certain
number of specificities, concerning the definition of the memory span, the atten-
tional management, and equally the conception of the processing limited capacity.
Spreading Activation and General Potential of Activation. The modelisation, devel-

oped by Anderson (1983a, 1993) assimilates Working Memory to an activated zone
of a conceptual network in Long Term Memory (Cf. also the description of Ander-
son’s ACT* model in Chapter 4). The degree of activation and the size of this acti-
vated zone depend on the Potential of Activation of each node (or concept) of the
system. Thus, a node or a concept activated in a semantic network propagates this
activation (Spreading Activation), all along the links, to other nodes or concepts.
This propagation is larger and more rapid than surrounding concepts are already pre-
activated (this phenomenon is able to explain the ‘priming’ effect). This spreading
therefore depends on the potential of activation of each knowledge unit and the total
of activation would provide to the system its ‘General Potential of Activation’ (Cf.
Anderson, Reder and Lebiere, 1996). Such an idea, by relying on the double princi-
ple of spreading and potentials of activation, leads to several consequences, impor-
tant on a theoretical viewpoint:
• According to Anderson (1983a), Working Memory, the place where Declarative
and Procedural Knowledge matches, has to be rather envisaged as a portion of
activated units in Long Term Memory and not as an independant mnemonic
register. In other words, according to Anderson (1993: 52), the content of the
Declarative Memory is the Working Memory and the activation of units (data)
determining what chunk is going to be processed before the other ones. The
elements are not in Working Memory if they are not sufficiently activated to be
matched with one production rule.
• The size of this activated zone would depend on the total sum of activation of
the cognitive system. This last aspect is fundamental because it supposes, ac-
cording to Anderson (1993), to no longer define Working Memory span, as it is
nevertheless classic to do so, by a limited number of information units suscepti-
ble to be simultaneously maintained and processed, but rather by a maximal
value of the General Potential of Activation. The value of this general potential
would be mainly determined (within the ACT-R model) by the sum of Activa-
tion potentials inherent to each memory probe in play in an activity plus the ini-
tial potential of the units that are going to be processed.
• The phenomenon of Spreading Activation leads to the activation of an impor-
tant number of knowledge units that are not systematically pertinent to be used
in a given task. According to Anderson (1983a), these knowledge units would
be de-activated by a process of active inhibition during their matching (Match-
ing process), with production rules enacting their use (Procedural Knowledge).
Thus, according to this principle, an activated node in a hierarchised system can
activate the superordinated or subordinated nodes, while inhibiting the compet-
ing nodes of a same conceptual level. This principle of inhibition (named ‘lat-
eral inhibition’, in this case) plays a role as important as the principle of activa-
tion on a functional plan.
Cantor and Engle (1993) have confirmed, by using the ‘Fan Effect’ paradigm de-
fined by Anderson (1983a), the pertinence of the Spreading Activation principles as
a possible idea of Working Memory. By varying the more or less strong thematic
and conceptual structuring of elements composing the same sentence (more or less
strong links), these authors showed that the recognition speed for previously learnt
sentences depended on the interaction between the Working Memory span (evalu-
ated by a span test) and the conceptual structure of the sentence, implementing
knowledge units close or not in the conceptual network. The further the concepts of
same sentence are from each other (in term of links or semantic relationships), the
longer the recognition time of the sentence is. This effect is more important when
Working Memory span is weak (i.e., when it is impossible for participants with
weak spans to simultaneously activate two too distanced concepts in a network).
Activation and attention. Lapointe and Engle (1990) and Cowan (1988, 1993),
equally consider Working Memory as an activated zone of Long Term Memory.
However, these authors have specified the functional characteristics of this activated
zone by distinguishing, within it, a pre-activated peripheral zone and a more strongly
activated central part, called ‘Attentional Focus’. In this hierarchical idea of Work-
ing Memory, the elements that are out of the focus are only pre-activated.
Cowan (1988) has carefully described the modalities of attentional management
in his model (Cf. Figure 25). According to this author, the activated elements can be
placed under the attentional focus in an active or in an automatic manner.
a : Voluntarily attended
CENTRAL EXECUTIVE
(directs attention and b : Habitued
Attention can be directed
outward, to stimuli, or controls voluntary processing) c : Habitued
inward, to long-term d : Dishabitued
memories
Controlled actions
*No “filter” is needed
physically unchanged a
Focus Automatic actions
stimuli do not elicit b of
attention (b,c), with the c d Attenti on
possible exception of Long-term storage of some
significant signals. coded features occurs
Activated memory
Unchanged stimuli can automatically (b,c). Attentive
(short-term store)
enter the focus of processing (a,d) results in more
Long Term elaborate encoding (critical for
attention through
voluntary means (a) Store voluntary retrieval, episodic
storage)
*Initial phase of sensory storage lasts only several

Unchanged a Brief
hundred milliseconds (left). Second phase is one type of
b Sensory
Stimuli c activated memory (above). Both sensory and semantic
Novel d Store
activation may last some seconds.
Stimulus
Figure 25: Principle of Working Memory by activation and management of the attentional fo-
cus, adapted from Cowan (1988). Copyright © 1988 by the American Psychological Associa-
tion. Adapted with permission.
In the first case, a component labelled ‘Central Executive’ (that equally controls de-
liberated processes) directs the attentional focus to a voluntarily attended element. In
the second case, the attentional focus can be automatically directed to a non attended
element if this element is unusual and has not made the object of a preliminary train-
ing (the training consists of neglecting, in the flow of stimuli, those that are not rele-
vant: for example, a surrounding noise, the blinking of a neon light when one writes
a text, etc.). This last aspect is interesting because the elements that have been habi-
tued (unchanged stimuli) can only be voluntarily positioned under the attentional fo-
cus by the Central Executive. The habituation would thus play an inhibitive role by
limiting the number of information to process. This mechanism, close to a filter, is
however more general and global than the lateral inhibition described by Anderson
(1983a).
Unique limited capacity. Besides the notions of spreading activation, all the Activa-
tion models agree on the fact that the storing and processing capacity is unique in
the cognitive system. It concerns one of the main differences between this kind of
model about Working Memory and Baddeley’s model (1986).
According to the famous Miller’s (1956) paper, the immediate memory span
would be limited by the number of information units or, according to this author,
‘chunks’ that can be simultaneously maintained by the cognitive system. This limita-
tion would be due to a limited number of slots of approximately seven (plus or mi-
nus two) elements. If this limitation is still admitted by researchers, its explanation
has evolved since its original formulation by Miller. The interpretation of the limita-
tion of the Short Term Memory capacity (or Working Memory) relies more today on
a general problem of allocation of cognitive resources, managed by a controlled
mechanism, than on a structural problem concerning a number of slots. In
Baddeley’s (1986) model, the Central Executive would carry out this management.
Thus, the limited storing capacity would be attributed to the two slave registers: the
Articulatory Loop and the Visuo-Spatial Sketchpad (Baddeley, 1990: 74-79). In the
case of Activation models, this type of limited storing capacity does not exist be-
cause the Working Memory register is not considered as a specific structure. Con-
versely, the limited capacity would be linked to some characteristics of the Activa-
tion (in terms of decay and spreading) inherent to Long Term Memory. The current
level of activation then controls the rate of information processing and the memory
span would be linked to the sustained capacity of Working Memory. In other words,
the most important difference between these two theories is due to the fact that sev-
eral limited capacities would coexist in Baddeley’s (1986) model through (1) the
two storing capacities of Visuo-Spatial and Articulatory registers and (2) the proc-
essing capacity of the Central Executive, while only one unique general processing
capacity in activation models would exist.
More precisely, according to Anderson, Reder and Lebiere (1996), the Activa-
tion amount is the General Potential of Activation of the cognitive system. This gen-
eral potential can vary as a function of the level of expertise for the accomplishment
of a specific task – the knowledge in Long Term Memory could be more or less ac-
tivated and able to be activated. Playing the role of cognitive resources, this poten-
tial may or may not authorise the application of more or less complex processes.
Anderson et al. (1996) have shown via an original experimental paradigm that the
simple retrieving of a knowledge unit could be strongly compromised if the task (in
which this unit is nevertheless implied) is complex and requires a great part of the
General Potential of Activation available within the cognitive system (thus limiting
the share of Activation necessary for the recovery of the unit).
To test the effect of the General Potential of Activation on the functioning of the
recovery process, the authors have asked students to solve mathematical equations
(as the main task), immediately after the memorisation of a series of digits (as the
secondary task, in a limited time). They were then asked, at the end of the problem-
solving, to orderly recall the list of digits. The authors have varied (1) the number of
digits to recall (from two to eight), (2) the complexity of the equations (in terms of
number of digits to activate to find the unknown ‘x’), but equally and especially, (3)
the necessity to use or not to use the first two digits of the list (respectively substi-
tuted by ‘a’ and ‘b’ in the equation). The performances have been mainly measured
by the analysis of the number of errors associated with the recall of the list of digits
and with the equation solving. When the difficulty of each of the two tasks in-
creased, performances decreased for both tasks. The most important point of these
results concerns the length of the list to be recalled: the length had a more constrain-
ing effect when it was necessary to use its first two digits. According to Anderson et
al. (1996), this triple interaction effect would be linked to the fact that the simulta-
neous increase of the complexity of equations and of the number of digits supposes
to activate, from Working Memory, a more important number of digits. The increase
of recall errors would mean that the General Potential of Activation, strongly in-
volved in these complex competing tasks, would not be sufficient to allow a good
recall of the list of digits.
For the authors, the results of these experiments testify that, in the cognitive sys-
tem, a General Potential of Activation would exist that must be shared to retrieve the
different units necessary for the execution of different tasks. The execution of more
complex or more numerous tasks could only be carried out to the detriment of the
possible recovery of units, which are nevertheless useful to accomplish the task(s).
Then, due to the too weak potential of activation for the recovery of a unit, this one
could be only partially matched (partial matching) with production rules. This fact
would lead to errors as for the characterisation of the unit. The constraints of Work-
ing Memory would therefore exert their effects (via the General Potential of Activa-
tion that is or is not available) more particularly on the recovery of units by restrain-
ing the activation. This restriction would mainly depend on the necessity to share a
unique source of Activation between different activities, more or less complex.
3.1.2 Activation models and verbal production activity

It is possible to find a certain number of works or theories of verbal production in
the literature that refers to the principles of Activation models to give an account of
the processing modalities in this kind of cognitive activity. More precisely, if the
Spreading Activation is the most often used to explain the cognitive functioning, the
role, in certain models, of Inhibition must be equally analysed.
Notion of activation. The use of Activation models is nevertheless more frequent in

local models of sentence production than in general models of text or speech pro-
duction.
In the case of the sentence production, the model elaborated by Dell (1986, Cf.
Chapter 2 for a description of this model) relies clearly on principles of activation
(and inhibition) between linguistic units of different levels. The modalities of sen-
tence elaboration directly depend on the Activation Potentials of each of the units
and can lead, via the Spreading principle, to the generation of errors classically
listed in oral production (displacement, slips of the tongue, etc., Cf. Garrett, 1980).
In the case of text production, the principle of Spreading Activation has been
adopted very early on in models (for example in Bereiter and Scardamalia’s model,
1987), to explain how the knowledge units are retrieved from Long Term Memory.
However, as previously discussed (Cf. Chapter 1), these principles, often mentioned
in classic models, are not really developed. Moreover, the precise functioning and
the influence of a mode of recovery by Spreading Activation within the writing ac-
tivity are rarely analysed and tested. It concerns more a claim than a choice justified
by precise hypotheses or experimental evidences. However, more than the simple
description concerning the modes of domain knowledge recovery, the principles of
Activation can also allow to account for the dynamics of writing processes during
this specific activity. The interest of Activation models, through the parameters of
activation and decay that characterise them, then becomes evident. Indeed, if we ac-
cept with Anderson (1983a) that the use of Declarative Knowledge and/or Proce-
dural Knowledge in a given task brings about, in return, a restructuring of this
knowledge, it is clear that all the writing processes, including those inherent to the
recovery of information, are going to modify the amount and the nature of Domain,
Linguistic and Pragmatic knowledge. In addition, the variation of the potentialisa-
tion of these knowledge units will change, via some modifications of the General
Potential of Activation, the possibility and/or the strategies of execution of these
writing processes.
It should be noted that this conception of the writing dynamics has recently been
adopted, via the principle of activation evolution, by van den Bergh and Rijlaarsdam
(1999), to acknowledge the functioning of processes during writing. According to
these authors, this principle, due to Activation models (and more particularly to
ACT* model, proposed by Anderson, 1983a), supposes that the execution of writing
processes would depend both on (1) potentials of activation, (2) instructions, (3)
possible organisation schemas, but equally on (4) the text produced so far. The text,
always in evolution, would permanently modify the Potentials of Activation of the
knowledge network and, consequently, the attentional state of Working Memory.
This principle leads to focus not only on the functioning of writing processes as
such, but rather on the context of their releasing, through an analysis of activities
that have immediately preceded this releasing. As already mentioned in the preced-
ing chapter (Cf. Chapter 4), it is in this framework that van der Bergh and Rijlaars-
dam (1999) have shown that the releasing of the recovery process was dependent on
different situational contexts (for example: Assignment Driven Generation; Reread-
ing Text Driven Generation; Transfer Driven Generation), and that the evolution of
these contexts during the progress of the writing activity is dynamic.
This example of adoption of Activation principles by van der Bergh and Ri-
jlaarsdam (1999) mainly concerns the activity of knowledge recovery that has been
stored in Long Term Memory beforehand. Otherwise, Galbraith (1999) has recently
elaborated, from the same activation principles, a model called ‘Knowledge Consti-
tuting’ (Cf. Figure 26). The objective of his model is to account for the mechanism
underlying the epistemic effect of the text (i.e., creation of new domain knowledge
during the writing activity – Cf. Chapter 1).
D
C A
TOPIC + TASK SPECS
Figure 26: Principles of the Knowledge Constituting model,

adapted from Galbraith (1999). Copyright © 1999 by the Amsterdam University Press.
The framework proposed by Galbraith (1999) is composed of a connectionist net-

work recording sub-symbolic nodes and links between these nodes (Cf. Chapter 4
for an explanation about general connectionists principles). The knowledge units are
not directly stored in this type of network, but the emergence of an idea is linked to a
pattern of activation in a set of nodes. An identical set of nodes can then bring about
the emergence of different ideas if the patterns of activation of this set are different.
A component called ‘Topic and Task Specs’ allows, on the basis of the interpreta-
tion of the writing instructions, to mobilise Rhetorical Knowledge and to proceed to
an explicit planning, for example, of the text plan. These processes, close to prob-
lem-solving processes, would be relatively similar to those that characterise Bereiter
and Scardamalia’s (1987) Knowledge Telling Strategy. This component provides the
first necessary input to the connectionist network, that modifies, as a consequence,
the activation pattern of the nodes and produces ideas under the double constraint of
(1) the Topic and Task Specs and (2) the initial activation of each of the nodes in the
network. This idea will be processed by a Module of Language (labelled ‘A’ in

Figure 26), whose functioning is envisaged as relatively similar to those of Dell’s
model (1986), that is to say a linguistic network translating the activations of the
nodes in linguistic propositions (‘B’ in the figure). The fundamental characteristic of
this model is that this linguistic output plays the role of a Feedback (‘C’ in the fig-
ure). This Feedback will modify the activation pattern of the nodes, creating thus a
new idea more or less close to the precedent, that has underlied the execution of the
first message.
If this new idea is acceptable for the Topic and Task Specs component, it can be
translated and executed (‘D’ in the figure). This process, by continuing, leads to the
writing of a text based on the interactive creation of knowledge units. Indeed, in this
model, a textualised idea intrinsically leads to the creation (and not the recovery) of
a new idea. The feedback of a translated idea does not only increase or modify the
activation of nodes, but it equally plays an inhibitive role whose function is to de-
crease the activation of the nodes having generated the textualised idea. This
mechanism allows the writer to limit redundancies and not to always generate,
throughout the writing cycles, the same idea.
Another possible exploitation of the characteristics of Activation models in writ-
ing refers to the notion of General Potential of Activation, such as described by
Anderson et al. (1996). According to this principle, the Activation Potential of a
network is assimilated to available cognitive resources at a given moment in the
cognitive system and would allow the functioning of more or less numerous and/or
more or less complex processes. Alamargot (1997, 2000) has proposed to adults and
teenagers to regularly write (during several sessions, one per week) a text describing
a set of complex geometrical figures. The writers’ expertise, in this imposed domain
area, has been measured before the beginning of each writing session, by an analysis
of their performances in memorising and drawing the figures from memory (analy-
ses of memorisation duration, of drawing errors, of graphomotoric spans, etc.). In-
terpreted in the framework of Anderson’s ACT* model (1983a), the analysis of
these non linguistic performances has revealed that the participants’ domain exper-
tise progressed in the course of sessions and could be interpreted by an increase of
the Potential of Activation in a knowledge network. Most important is that this evo-
lution of non-verbal performances would largely determine, when the participants
write texts to describe the figures, the complexity of content organisation. More pre-
cisely, the decrease in the learning times for the geometrical figures, related to the
evolution of domain expertise, was strongly correlated with the increase of the num-
ber of semantic relationships inserted in descriptive texts. According to the author,
the evolution of the expertise in a given domain field, by leading to an increase of
the Potential of Activation of the knowledge network, would allow the writer to
have more resources to establish a strong text coherence. The activation and the exe-
cution of the Organising sub-process (as defined by Hayes and Flower, 1980) would
thus be facilitated.
These examples show the possible integration of Spreading and Potential princi-
ples of Activation in models or experimental works of the writing activity. They
equally show that this approach is still very exploratory, but that it however con-
cerns a very interesting theoretical track to try to explain as well the dynamic man-
agement of processes during writing, that the influence, in terms of modification of
activation potentials, of domain knowledge on the writing process.
Notion of inhibition. The mechanism of Inhibition, in the area of language produc-

tion, has mainly been studied in the case of comprehension. For example, Gerns-
bacher (1989) and Gernsbacher and Faust (1991) have tempted to explain the proc-
esses functioning during reading. For these authors, the comprehension activity
would consist, among others, of managing inhibition and activation mechanisms of
the text concepts and themes. These mechanisms could be principally released by
different linguistics marks (for example, the anaphora, being able to play the role of
processing instructions and to lead, during reading, to the re-activation of the re-
ferred theme in Working Memory and the inhibition of the preceding theme).
As previously mentioned, such a mechanism would equally be involved, accord-
ing to Galbraith (1999), in text production. During the formulating phase, an inhibi-
tion mechanism would allow the writer to deactivate domain knowledge units that
have been immediately translated in order to activate new knowledge units neces-
sary for the pursuit of writing. However, inhibition mechanisms are not exclusive to
the Translating process. In the case of processing control, for example, it is neces-
sary, as underlined by Grabowski (1996, Cf. Chapter 4), to inhibit some irrelevant
processes in order to activate more adapted processes. Nevertheless, in Galbraith’s
(1999), as well as in Grabowski’s (1996) models, the presence of inhibitions is
claimed, although no experiments tested its functioning. However, for writing proc-
esses and for the control of these processes, this claim seems particularly relevant
and it seems evident that some inhibition processes could play a role during the writ-
ing activity. It now remains to define the experimental conditions of such a valida-
tion.
3.1.3 Conclusion on activation models and text production

Some activation models seem to be particularly useful in the framework of the study
of writing processes. These models provide a very relevant explanation of the dy-
namics and conditions of processes, in a functional perspective. The characteristics
of writing processes are, conversely, difficult to interpret in the strict framework of
Baddeley’s (1986) model, that is rather implicit, notably concerning the functioning
of the Central Executive, theoretically responsible of the dynamic application of
processes.
If the Activation models seem relevant to account for the dynamics of writing
processes, it would be nevertheless interesting and certainly necessary, by replicat-
ing the procedure used by Cantor and Engle (1993), to evaluate in what the dynam-
ics, aroused by variations of activation of units in a network, can be constrained by
the writers’ General Potential of Activation (in others terms, by their memory span).
Indeed, in the case of the approach developed by van den Bergh and Rijlaarsdam
(1999), is it possible to wonder if and how such a General Potential of Activation

could constrain the functioning and the sequencing of writing processes. Thus, a
writer with a limited memory span could not fully process the feedback that consti-
tutes the previously written text. Consequently, the adaptation of her/his writing
processes, according to the temporal evolution of the task, would be particularly lim-
ited. In the same way, in the framework of Galbraith’s model (1999), it is possible to
hypothesise that the possibility of new contents creation, in the course of writing,
would be more or less constrained by the value of the activation potential inherent to
the sub-conceptual network. This value would therefore more or less limit the acti-
vation of the patterns of nodes and, consequently, the nature and the diversity of the
generated ideas. The integration of activation models in research on text writing is
very recent. These models are more invoked as a factor of explanation or interpreta-
tion of a general process functioning rather than as models susceptible to lead to the
elaboration of hypotheses on this functioning. It is very probable that, in the future,
such questions will appear and, hopefully, be solved.
Finally, the activation models, due to the notion of activation, equally possess
some limits to account for some phenomena linked to the expertise in a specific task.
Indeed, one main characteristic of the activation is the decay, that is a temporary
phenomenon, useful to explain the knowledge mobilisation during an activity, but
insufficient to account for the modification of knowledge in Long Term Memory,
after this activity has been acoomplished. In others words, one of the major criti-
cisms on a definition of Working Memory as an activated zone of Long Term Mem-
ory relies on the rapid decline of the activation that generally characterises the
spreading principle (Anderson, 1983b). According to Ericsson and Kintsch (1995),
even though the activation (and the rapid decline) can perfectly account for the tem-
porary aspect of the information maintained in Working Memory (at least in the ab-
sence of rehearsal processing; Baddeley, 1986), it is however not adapted enough to
explain that experts can, for example, interrupt and continue on (after an important
period) the execution of a complex task, without a decrease in their performances.
To explain this type of phenomena, Ericsson and Kintsch (1995) have elaborated
a new idea about the cognitive system organisation, by distinguishing, within Long
Term Memory, a specific component, labelled ‘Long Term-Working Memory’, that
plays the role of an expert interface during the transfer of information between Long
Term Memory and Working Memory. Kellogg (1999) has recently considered how
to integrate this component into the framework of his 1996 writing model.
3.2 The Long Term Working-Memory as another alternative

3.2.1 Principles of Long Term-Working Memory, according to Ericsson and
Kintsch (1995)
The model, developed by Ericsson and Kintsch (1995) relies on the classic distinc-
tion between Working Memory and Long Term Memory. With the concept of Long
Term-Working memory, Ericsson and Kintsch (1995) suppose that experts in a
given domain knowledge can both acquire knowledge about (1) this domain (De-
clarative Knowledge), (2) procedures of use of this knowledge (Procedural Knowl-

edge), but equally (3) ‘methods’ of storing and retrieving of knowledge units. This
structure of clues, called Long Term-Working Memory, is established as a function
of the task demands and instituted in Long Term Memory. It allows to strategically
operate storing and retrieving knowledge (Ericsson, 1985: ‘Skilled Memory’), and
would thus lead to an extension of Working Memory capacities (Cf. Figure 27).
Retrieval structure
Retrieval cues Cue 1 Cue 2 Cue 3 Cue 4
Association
Encoded i j k l
Information
Figure 27: Principle of the Long Term-Working Memory,

adapted from Ericsson and Kintsch (1995). Copyright © 1995 by the American Psychological
Association. Adapted with permission.
These retrieval clues are strategically associated with knowledge units stored in
Long Term Memory. The simple activation, in Working Memory, of the ‘Retrieval
Structure’, would economically activate all the pieces of knowledge (Encoded In-
formation) necessary for a given task. The Long Term-Working Memory is thus not
generic and can be used only for a given activity whose accomplishment has become
expert. It is important to note that this new and relatively original concept of Long
Term-Working Memory, about the functioning of Working Memory, has not yet
been tested via experimental validations. Nevertheless, it is interesting to wonder if
it is possible to integrate this new concept, more specifically dedicated to expertise,
in current writing models.
3.2.2 Long Term-Working Memory and text production activity

Kellogg (1999) has recently reconsidered the choices he made to elaborate his 1996
model, by confronting the interest of the different ideas about Working Memory in
the framework of text production. According to Kellogg (1999), the choice of a spe-
cific model of Working Memory, or the decision to modify an existing model of
memory, has to rely on the consideration of three phenomena, strongly marked in
text writing:
• The first parameter concerns the limited capacity of the Central Executive. This
limitation would play a fundamental role in writing, which particularly needs
the Central Executive for Formulation, Execution and Monitoring components
(Cf. Kellogg, 1994; Piolat, 1998; Piolat, Roussey, Olive, & Farioli, 1996).
• The second parameter supposes to account for, in the activity of text writing, the
different codes of processed knowledge units: linguistic, conceptual, and/or se-
mantic. A multi-component model could be relevant here and must comprise
(but is not limited to) verbal, spatial and executive components.
• Finally, as third parameter, the memory system must explain the fact that ex-
perts in a domain write best quality texts, more strategically and more effi-
ciently than novices in the same referential domain (Cf. Chapter 1). Kellogg
(1999) has indeed considered in what measure it would be important to add, as
recommended by Hayes (1996: 8), a Semantic Memory dedicated to the storage
and processing of propositional codes. More precisely, this component would
become central in text production, concerning both the recovery and the proc-
essing of domain knowledge.
According to Kellogg (1999), writers with expertise in a particular domain knowl-
edge may expend their transient Working Memory capacity through reliable re-
trieval of knowledge from Long Term Memory. The particular memory component
that would account for this phenomenon could be the Long Term-Working Memory
such as defined by Ericsson and Kintsch (1995). Indeed, as described previously, the
Long Term-Working Memory concerns all the mechanisms that support reliable ac-
cess to knowledge that is needed to perform a task. This structure, highly strategic,
could belong to a Working Memory system and account for conceptual and semantic
processing differences between experts and novices in a specific domain. The strate-
gic retrieval of domain knowledge, during the execution of a task, would explain (in
a different manner than activation models, Cf. supra) the writing facilitation effects
generally noticed when experts in a domain write a text concerning this domain (Cf.
the works of Voss, Vesonder, & Spilich, 1980; Kellogg, 1987a; Cf. also Chapter 1).
3.2.3 Conclusion on Long Term-Working Memory and text production

This theoretical matching, proposed by Kellogg (1999), between Ericsson and
Kintsch’s (1995) Long Term-Working Memory model and Kellogg’s (1996) model
of text writing, could account for some phenomena linked to the writer’s expertise.
Furthermore, Long Term-Working Memory could not only concern the domain
knowledge recovery, but equally the retrieval of pragmatic, linguistic and procedural
knowledge in the progress of writing. The principle of Long Term-Working Mem-
ory could then be considered as a relevant explanation of the development of writing
expertise, in terms of highly strategic recovery of the four knowledge areas implied
in this activity. Nevertheless, the concept of Long Term-Working Memory, rela-
tively recent, is still very theoretical and has not been largely empirically tested. Its
integration in a writing model supposes the creation of particular paradigms allow-
ing to explain (1) the constitution of encoding and retrieving strategies, specific to
Long Term-Working Memory (in others words, the Long Term-Working Memory
content), and (2) the modalities of the influence of these strategies of knowledge
management on the functioning of the writing process.
4 DISCUSSION POINTS
4.1 A finer analysis of some writing activities: the Translating and the Revising
processes
The integration of a model of text production with Baddeley’s model (1986) of
Working Memory is recent, but has already led, impulsed by Kellogg (1996), to a
consequent number of works aiming to validate this theoretical approach. Only a
limited number of processes are involved in these studies. They focus mainly on the
role of the Articulatory Loop in text production (Cf. above, the different experiments
led by Levy and his collaborators). Moreover, the adopted definition of processes
(through Planning, Translating, Programming, etc.) remains very global while it
could be envisaged that the different sub-processes or operations maintain specific
and close relationships with the different registers in Working Memory. In the case,
for example, of the Translating process (according to Hayes and Flower’s 1980
model), linguistic and conceptual processes can operate alternately, going from the
elaboration of a detailed content of a text plan to the elaboration of the lexical and
grammatical structure of a sentence (Cf. Chapter 2). In the framework of Kellogg’s
(1996) model, it would be particularly relevant to specify, as suggested by Hayes
(1996), the mnemonic components that could play a role in the alternation of lin-
guistic (in relation with the Articulatory Loop) and conceptual (in relation with the
Visuo-Spatial Sketchpad) processes. However, neither current works, nor methods
already used have this degree of precision.
In a more general way, this same remark can be made concerning the Reviewing
process whose processing architecture and implied sub-processes seem to be very
complex. Indeed, the Reviewing process is considered as a very complex process,
heavily loading on writers’ attention and memory capacities (Beal, 1996;
McCutchen, 1996). Currently, increasingly numerous works show interest in the role
of Working Memory limited capacity on the revising processes (Cf., for example:
Chanquoy, 2001; Hacker, 1994; Hacker, Plumb, Butterfield, Quathamer, & Heine-
ken, 1994; McCutchen, 1994; Swanson & Berninger, 1994). In addition, the last
model of revision, elaborated by Butterfield, Hacker and Albertson in 1996, specifi-
cally mentions the main role of Working Memory as an important co-ordinator of
sub-processes during this activity (Cf. Chapter 3). Is it, in the light of these works,
possible to consider that the different Reviewing sub-processes compete for the lim-
ited Working Memory resources? It would thus be necessary to query the cost of the
different revising sub-processes (Cf. the discussion of Chapter 3).
4.2 Problems concerning the investigation of the memory span

Most of the research aiming to evaluate the role of Working Memory in writing ac-
tivities often relies on the necessity of a preliminary evaluation of the writer’s mem-
ory span, either to constitute a systematic factor opposing writers with ‘high span’
and those with ‘low span’ (Cf. McCutchen, Covill, Hoyne, & Mildes 1994, for ex-
ample), or to establish a control factor (Cf. Ransdell & Levy, 1996a, for example).
Central in these types of research, the evaluation of the Working Memory span(s) is
not without posing a certain number of theoretical and methodological problems.
4.2.1 Surrounding what is measured by a span test

The memory span is often evaluated by different kinds of tests, which are sometimes
difficult to interpret what they exactly measure. A good example of this problem can
be provided through the numerous debates about the ‘Reading Span Test’, elabo-
rated by Daneman and Carpenter (1980). The principle of this test is simple: partici-
pants must first read a series of two complex sentences, then recall the last word of
each sentence. This procedure is repeated by increasing each time the number of
sentences, until they are no longer able to recall. The number of sentences constitut-
ing the series preceding the failure to recall provides the size of the ‘Reading Span’
in Working Memory.
According to Daneman and Carpenter (1980), this test is supposed to measure a
storing and processing span in a Working Memory specifically dedicated to linguis-
tic processing, such as reading and comprehension. Because of this linguistic speci-
ficity, the more or less important limitations of this span would only constrain the
execution of linguistic activities as reading or comprehension, and consequently not
of non linguistic activities (this is moreover one of the founder principles of Just and
Carpenter’s ‘Capacity Theory’, 1992). Nevertheless, a lot of research has shown that
this postulated linguistic specificity, measured by this test, was not so inevitably es-
tablished (Cf. Logie, 1996 for a discussion of this test), and that strong relationships
existed between the results of this verbal test and of performances to tasks of digit
memorisation (counting span: Case, Kurland and Goldberg, 1982), solving of simple
arithmetical operations (Turner and Engle, 1989), or reading of digit sequences
(Yuille, Oakhill and Parkin, 1989). According to Turner and Engle (1989), if the
Reading Span Test really measures the capacity of sentence processing (i.e., mainly
reading activity), it would equally measure, and in a confusing way, a more general
storing capacity, concerning any piece of information, whatever its nature (thus ex-
plaining the relationships between the reading span test and non linguistic activities).
This rapid debate concerning the interpretation of a span test, illustrates the diffi-
culties that currently remain to measure storing and processing capacities in Work-
ing Memory. In this view, it is interesting to note that Daneman and Green (1986)
have complexified Daneman and Carpenter’s (1980) Reading Span Test by adding a
verbal production measure. In this new test, labelled ‘Speaking Span Test’, partici-
pants have to produce a list of sentences from a list of words (i.e., one sentence per
word, with, in addition, constraining instructions). This test is largely used today, in
the framework of studies on writing. At present there are researchers who try to mul-
tiply the used tests: Reading Span Test plus Speaking Span Test, plus Writing Span
Test, whose procedure is identical, except for the medium – speaking vs. writing –
of the ‘Speaking Span Test’). However, still in most experiments, when these differ-
ent span measures are used for writing research, they often concern a speaking or a
reading span, and that does not allow to clearly relate speaking performances to
writing performances.
4.2.2 Adapting the chosen span test to the model of Working Memory to be tested
All these interrogations on the pertinence of span tests and the reality of their meas-
ures are all the more crucial than the study of writing processes as approached in the
framework of Baddeley’s (1986) multicomponent Working Memory. Indeed, as
writing processes, according to Kellogg (1996), would need, in a differentiated
manner, all the three different registers of Working Memory, it becomes necessary
to evaluate both the storing capacities of the Articulatory Loop and the Visuo-
Spatial Sketchpad, as well as the processing capacity of the Central Executive. Such
an expansion in the span measures (verbal, visuo-spatial and processing spans) cer-
tainly raises a great number of theoretical and methodological problems. This last
remark can be equally formulated in the case of the elaboration of secondary tasks,
built to load either one or other Working Memory component. As seen with the de-
scription of Levy and Ransdell’s (1996) experiments, the preferred methodology to
validate Kellogg’s (1996) model involves asking writers, while they are writing, to
perform different kinds of secondary tasks that are supposed to respectively over-
load one out of the three Working Memory components. The difficult aspect is nev-
ertheless to be sure that a given secondary task only overloads one component. This
difficulty is sharp in the case of the Central Executive: when a writer is asked to
solve an arithmetical problem (Cf. Levy & Ransdell, 1996), it is possible to wonder
if the Articulatory Loop is not implied in the temporary storing of digits.
All these difficulties linked to the complexity of Working Memory do not simply
concern a Working Memory system, such as defined by Baddeley (1986). In the per-
spective of activation models, it is equally necessary to evaluate the General Poten-
tial and the Spreading of Activation.
The experimental testing of activation models appears less easy to achieve than
in the framework of Baddeley’s (1986) model. It is necessary to use complex para-
digms with Priming Effects (in order to analyse the modalities of spreading activa-
tion and pre-activation states of concepts in a network), Fan Effects (to measure the
relative distancing of two concepts in a knowledge network), and also secondary
tasks (overloading cognitive resources and needing the general potential of activa-
tion). On this last aspect, Anderson, Reder and Lebiere (1996) rely on a very rele-
vant methodological approach to evaluate effects of the activation potential on proc-
ess releasing (Cf. supra). This type of paradigm is interesting by the secondary task
method linked to the main task that it proposes (i.e., to use, in order to solve the
main task, information from the secondary task).
4.2.3 Span test and expertise development in a specific domain or for a specific
task
The exploration of memory capacity is necessary in works that deal with relation-
ships between writing processes and Working Memory. Nevertheless, span meas-
ures or secondary tasks are relevant only when the memory capacity is considered as
(a) limited (in terms of storing, cognitive and attentional resources, or general poten-
tial of activation) and (b) independent of a given task.
Conversely, in the case of Ericsson and Kintsch’s (1995) Long Term-Working
Memory, the notion of limited span does not exist because the expertise more or less
developed for a given task would lead to an important increase of storing capacities
in Working Memory. In addition, the increase of capacities would be contextually
linked to the task and not transferable to another type of task, in which the partici-
pant is not an expert. In this conception, the evaluation of a memory span, with clas-
sic tests (i.e., to recall digits, words, or figures, for example) is useless because it
does not systematically account for processing capacities during the task. This could
be equally said in the case of expertise in a writing task. If Ericsson and Kintsch’s
(1995) conception is adopted, it is possible to hypothesise that memory span –
measured by classic span tests – and writing processes are related when the writer is
relatively novice and that memory span, to perform writing processing, has only ‘ba-
sic’ mnemonic capacities. Conversely, for an expert writer, highly strategic process-
ing capacities would not be systematically related with span measures, obtained
through simpler activities.
5 CONCLUSION ON THE ROLE OF WORKING MEMORY IN WRITING

Finally, the analysis of the relationships between writing processes and Short Term
Memory, although initially envisaged by Hayes and Flower (1980) or by van Dijk
and Kintsch (1983), has only begun to truly develop since the last few years. This
orientation to account for Working Memory components, devoted to storing and
processing, in writing research is certainly linked to Hayes’s (1996) new proposals
and to the appearance of Kellogg’s (1996) model that must be validated. This field
of research is therefore far from being worn-out because, as seen previously, there
are still many questions, theoretical as well as methodological, concerning both the
justification of the choice of memory models to account for writing processes
(Baddeley’s model, Activation models, Long Term-Working Memory model) and
the modes of evaluation of the influence of the memory capacity on these writing
processes (nature of memory span tests).
If, as underlined by Flower and Hayes (1980), one of the main writing difficul-
ties involves simultaneously managing (or juggling with) a great number of con-
straints, current works still do not really describe these constraints, within a func-
tional viewpoint. Similarly, Kellogg’s model (1996), although very heuristic, is nev-
ertheless a relatively static model that can only difficultly account for the role of
Working Memory in the dynamics of processes, in the determinism of writing
strategies and, more strongly, in the development of writing expertise.
CHAPTER 6
DEVELOPMENT OF EXPERTISE IN WRITING
1 INTRODUCTION: EXPERTISE IN WRITING

5.1 Specificity of expertise development in writing
According to Simon (1995, p.507), ‘the very complex behaviour of the human expert
are produced by a small number of rather simple processes’. More precisely, ex-
perts would essentially use recognition skills (based on a great amount of knowledge
on the activity to process), that they would combine with heuristic research skills
(based on a set of research techniques). This would then explain their rapidity and
their efficiency in most cognitive activities. According to Ericsson and Lehmann
(1996), experts are distinguished from novices by their capacity to adapt themselves
in an optimal and maximal manner to constraints linked to the task, whatever it is.
Although these definitions of expertise lead to assume that complex behaviour is
finally the product of some relatively simple processes, it is nevertheless true that
the acquisition of expertise in a domain such as writing is not easy to reach, even by
experienced writers. According to Bialystok (1986), there are important differences
of writing management between adults and children and/or between novice and ex-
perienced writers. Following Scardamalia and Bereiter (1991), writing expertise can
be defined as a state in which different pieces of knowledge would enable to easily
execute activities that non-experts could not execute or only with difficulty. Hence,
on a developmental viewpoint, the installation and the efficiency of the writing ac-
tivity are particularly long and progressive. For example, the composition of certain
text types, such as argumentative texts, by a demanding and complex management
of writing processes, would not be really efficient before the age of 15 or 16 years
(Bereiter and Scardamalia, 1987). Furthermore, such texts are often not mastered,
even by adults (Cf. Coirier, 1996).
As noticed by Torrance (1996), resuming arguments advanced by Scardamalia
and Bereiter (1991), the expertise development, in a given domain, is often charac-
terised by a speeding up and a more powerful efficiency of processes. However, the
case of writing activity is very specific because it seems that writing expertise de-
velopment operates on the contrary by a lengthening of process duration and by the
use of more complex and more sophisticated writing processes. In others words, ex-
pert writers paradoxically spend more time on their text and operate more complex
processes than novices. The specificity of writing expertise development relies on
progressive (and the more or less rapid) installation and construction of: (1) main
D. Alamargot & L. Chanquoy (2001). Development of expertise in writing. In G. Rijlaarsdam

of Writing, 187 – 220. © 2001 Kluwer Academic Publishers. Printed in the Netherlands.
writing processes (such as Planning, Translating and Reviewing), and (2) different
pieces of knowledge (i.e., domain, linguistic and pragmatic knowledge).
Concerning Planning and Translating processes, while novice writers most often
produce a unique formulation and preserve it throughout the production, experts
have a longer reflection on their plan and do not hesitate to modify their previous
plan or text (Bereiter and Scardamalia, 1987). Therefore, there would be no or little
anticipation in children or novice writers who, as soon as the instructions have been
given, begin very rapidly to write the text (McCutchen, 1988). Conversely, adults or
expert writers plan and elaborate their text more deeply. They do not execute a sim-
ple orderly enunciation of available knowledge, but they proceed to a domain
knowledge reorganisation according to the task constraints. For instance, Scar-
damalia and Bereiter (1982, 1986, 1987; Cf. also: Bereiter, Burtis, & Scardamalia,
1988; Bereiter & Scardamalia, 1987, 1988) consider that inexperienced writers
would rather have a step by step processing of their production (i.e., Knowledge
Telling Strategy), without systematically considering the addressee and the context
(Cf. Cannella, 1988). This results in texts appearing as juxtapositions of utterances,
without clear links. Conversely, adults and/or expert writers would use more com-
plex procedures, notably in re-elaborating the nature and the structure of information
retrieved from memory, as a function of goals and addressee of the production (i.e.,
Knowledge Transforming Strategy). With respect to the translating stage, different
experimental results show that the Translating process is not distinguished from the
Planning process in novice writers (Cf. for example, Berninger & Swanson, 1994;
McCutchen, 1988). Thus, when novice writers are asked to make a draft during
Planning, the text produced during translating simply consists of a recopy of the
draft, without any change. Conversely, experts’ final texts present a form and a vo-
cabulary very different in comparison to the draft notes. It seems that expert writers
clearly distinguish the stage of Planning from the stage of Translating (Kellogg,
1990, 1993). Concerning the Reviewing process, most research works have shown
that the revision on the text is rare in novice writers who generally carry out a rapid
re-reading of their text (Piolat, 1990) and therefore only perform superficial ‘groom-
ing’ (Fayol and Gombert, 1987) or ‘cosmetic’ changes (Hague and Mason, 1986).
Consequently, it is often impossible for these writers to detect errors. Conversely, in
expert writers, the revision activity would not only be more thorough but also more
frequent. In addition, the revisions about text form and content can take place during
the entire writing process, from planning to graphic transcription and even after, dur-
ing a specific revising stage (Cf. Chanquoy, 1997). The revision, to be efficient,
would therefore have to be largely automatised, at least in its simplest aspects (e.g.,
spelling or punctuation corrections). However, it seems that, even for experts, this
activity is never considered as simple or costless (Chanquoy, 1998).
Concerning the different pieces of knowledge involved in writing, it seems that the
increasing efficiency of pragmatic knowledge, by an increasingly important adapta-
tion of the text to the addressee, constitutes an important component in writing ex-
pertise. Hence, Martlew (1983) considers that writers, until 13 years of age, would
have many difficulties taking the potential reader into account (Cf. also in Chapter 1,
DEVELOPMENT OF EXPERTISE IN WRITING 189
works of Traxler and Gernsbacher, 1992, 1993). Young or beginner writers would
only translate spoken language into written language, without considering the ad-
dressee. Indeed, while experts would be able to consider the potential reader of their
text (reader-based prose), novices would focus more largely on low-level aspects of
writing (writer-based prose), such as spelling, punctuation or grammar (Hartley,
1991). Thus, the difficulties linked to writing in young writers would be mainly due
to two factors (Cf. Martlew, 1983): (a) a lack of awareness about writing problems
(such as the consideration of a writing aim, for example), and (b) the failure of the
automatisation of appropriate procedures and recognition of errors.
5.2 From a progressive installation to a complexification of writing processes: the

role of developmental factors
These last examples about the evolving of writing expertise clearly show that this
evolution can be approached in two non contradictory ways. (1) In the first case, the
evolution is conceived as a progressive installation of the different writing proc-
esses; (2) in the second case, it is considered, at a precise age, as an increase or a
complexification of writing expertise. The distinction of these two viewpoints is im-
portant and lead to several consequences:
• In the first case, cross-age and longitudinal studies allow analysing this installa-
tion. Most often, it involves the emergence of the first writing skills of the
Translating process, that, in the absence of the Planning and Reviewing proc-
esses, enables the writing of a text. In the second case, writing processes are
considered present in the writer and the expertise development is considered as
to complicate the functioning of these processes, those becoming then increas-
ingly efficient. This form of evolution of the expertise can be analysed by stud-
ies of inter- (or intra-) individual differences (experts vs. novices, for example).
It more particularly concerns the Planning process whose implementation is be-
lated and relatively complex and costly (Cf. above and Chapter 1).
• The distinction between these two forms of writing expertise development
means not inevitably assimilating, on the one hand, expert writer and adult
writer and, on the other hand, novice writer and beginner writer. Even if chil-
dren are, by their knowledge level and their processing capacity, more inexperi-
enced than adults, writing strategies adopted by adults can be, under certain
conditions, of a more novice nature than those of children (Cf. McCutchen,
1996).
• Finally, distinguishing different evolution modes of expertise can cause to won-
der about the determinism of this or these development(s). The developmental
factor that is generally evoked, mainly by Bereiter and Scardamalia (1987, Cf.
next sections), as an ‘engine’ of the writing expertise, is the increase of Work-
ing Memory capacity. However, this increase can be envisaged as being linked
(a) to a general maturation of the cognitive system (and, here, to maintaining
and processing capacities), or (b) to a practice and training effect (in terms of
automatisation and/or more strategic implementation of different kinds of proc-
esses). While the hypothesis about the maturation effect appears probable for
the youngest writers, the hypothesis about the training effect could explain the
access to expertise or the systematisation of this expertise in teenagers and
adults.
Focussing on these three remarks, (1) firstly, the main models concerning the devel-
opment of writing expertise are described by distinguishing (a) those whose objec-
tive is to account for the installation of writing processes, and (b) those that aim at
clarifying the processes complexication. (2) Secondly, the different ideas and possi-
ble theories concerning the role of Working Memory as factor of development of
writing expertise are discussed. The effects of the variation of Working Memory ca-
pacities are considered (a) in the framework of the maturation and (b) in the frame-
work of the apprenticeship.
6 DIFFERENT MODELS WHICH ACCOUNT FOR THE DEVELOPMENT OF

WRITING EXPERTISE
The rapid overview about the development of writing expertise for the main proc-
esses shows that it concerns a phenomenon that is complex to analyse. The different
models, aiming to circumscribe this development, are still rare but each of them ap-
pear as relatively heuristic and consequently being able to arouse a number of ex-
perimental works in the field of development. As mentioned in the introduction of
this chapter, two categories of models can be distinguished: the first category de-
scribes the installation of the first writing skills in children (from ‘Translating’ to
‘Planning and Translating’: Martlew, 1983 and Berninger and Swanson, 1994); the
second one describes the complexication of these skills in teenagers and adults
(Planning complexification: Bereiter and Scardamalia, 1987 and Flower, Schriver,
Carey, Haas and Hayes, 1989).
6.1 Development of expertise as a progressive installation of processes: from

‘Translating’ to ‘Planning and Translating’
6.1.1 Martlew’s (1983) model: from speaking to writing
Martlew’s (1983) model, relatively general, proposes a global overview of necessary
knowledge to reach writing expertise, as compared with speaking expertise. Martlew
lists three important aspects of writing: (1) the appropriate cognitive procedures, (2)
the cognitive awareness about the activity, and (3) the different writing processes
(Cf. Figure 28).
The model is composed of an important number of dimensions listing knowledge,
processes and metaknowledge whose respective mastery will be indispensable to the
evolution of the writing expertise. Three components can be assimilated to necessary
resources for written production.
• The ‘Memory’ component contains information and knowledge about writing as
well as strategies for these pieces of information and knowledge.
• The ‘Goal/Purpose’ component determines the text Topic, the characteristics of

the Reader, the Discourse mode and Style.
• The ‘External Aids’ component is largely close to the Task Environment com-
ponent described by Hayes and Flower (1980; Cf. General Introduction). It con-
tains writing Instructions, potential Resources (books, for example) as well as
Procedures that have been possibly taught to write.
WRITING PROCESS
MEMORY GOALS/PURPOSE
Information Topic
COGNITIVE Knowledge Reader COGNITIVE
PROCEDURES Strategies Discourse mode AWARENESS (Flavell)
Style
Recognize need to act
Recognize EXTERNAL AIDS Realize how to act
Select Instructions Sustain this action
Compare Procedures over time
Organize Books, etc. Integrate with other
procedures
Evaluate
Edit PLANS
Revise Global
Local
LINGUISTIC EXPRESSION
Semantic
Syntactic
PRODUCE TEXT
Spelling
Punctuation, etc.
Handwriting
Figure 28: Writing processes while a child is writing,adapted from Martlew (1983). Copy-
right © 1983 by John Wiley & Sons. Adapted with permission.
Three processes carry out the text production:

• The ‘Plans’ process ensures the elaboration of the text plan (both global and lo-
cal).
• The ‘Linguistic Expression’ proceeds syntactic and semantic aspects during lin-
guistic translation.
• Finally, the text as such is produced by means of the third process, labelled
‘Produce Text’, which carries out linguistic processing concerning Spelling,
Punctuation, Handwriting, etc.
The Memory and Goal /Purpose components maintain direct relationships with the
Plans process, but the whole set of processes and resources composing the Writing
Process equally benefit from two other more general components:
• The first, labelled ‘Cognitive Procedures’, is constituted by various mental op-

erations susceptible to playing a role in the writing process. These operations,
respectively named Recognise, Select, Compare, Organise, Evaluate, Edit, and
Revise, would intervene during the elaboration of the plan as well as during the
linguistic expression processing. Conversely, only the last three operations
(Evaluate, Edit and Revise) would equally participate in the text elaboration
process (Produce Text).
• The second component, labelled ‘Cognitive Awareness’, lists the writer’s meta-
cognitive knowledge, necessary in order to recognise, for example, the necessity
to act, the manner in which to act, etc.
On a developmental perspective, Martlew (1983) mainly bases her model on the op-
position between spoken language and written language. First of all, she underlines
an important phenomenon, even if it can be considered as trivial: when a child be-
gins to write, s/he already knows how to use spoken language. Thus, compared to
oral language, the acquisition of writing seems largely more complex, slow and ef-
fortful. Indeed, children have to cope with phoneme-grapheme transformation, sen-
tence segmentation, connection and demarcation and, of course, handwriting and
spelling. For the author, both cognitive and linguistic demands of writing generate
difficulties greater than those encountered in spoken language (Cf. her chapter, pp.
302-303). In addition, writing necessitates a different way of thinking than speaking.
In the framework of her model, Martlew points out what the writing expertise is for
her:
‘Writing proficiency is advanced when writers successfully automate what could be
termed the lower-order skills relating to the transcription aspects of writing. When the
majority of words used can be run off without having to focus on spelling and clause
boundaries which are automatically demarcated by appropriate punctuation and capitali-
sation, the writer is relieved from having to devote conscious attention to these opera-
tions. The writer can then focus on the higher-level operations needed to maintain an
awareness of overall global aims, either in accordance with a prespecified plan or inte-
grating new ideas which evolve in the course of composition.’ (p. 305).
In fact, Martlew adds that one of the important ingredients of expertise relies on the
capacity to think about and to evaluate one’s own language (i.e., metacognitive
awareness). Moreover, the young writer must not only be aware of her/his written
language, but equally of the reader. This last aspect is in fact considered as the most
problematic in writing.
Thus, with less than a complete model of writing development, Martlew mainly
points out the main problems of writing, compared with speaking. These problems
finally remain, for the young writer, to determine the specificity of writing compared
to speaking, and her framework provides a good overview to understand the difficul-
ties children are faced with when they begin to learn how to write. Nevertheless, two
remarks emerge from Martlew’s (1983) idea:
• On the one hand, the opposition between speaking and writing is perhaps not so
important as underlined by the author because, as noticed by de Beaugrande
(1984), Berninger and Swanson (1994), Grabowski (1996), or McCutchen
(1996), the processes (or sub-processes) of text generation (such as idea retriev-
ing, lexical and syntactic choices) also exist orally (discourse generation) or, at
least, share common mechanisms with those of writing. In young writers, the
main difference between speaking and writing would therefore be a matter of
the necessity and the difficulty to graphically translate a text conceived as an
oral speech, rather than of the implementation of specific writing processes.
• On the other hand, Martlew’s (1983) framework, by delimiting the different
necessary components for the beginning of writing, does not give a precise de-
scription of the installation of these different components, contrary to the more
recent model developed by Berninger and Swanson (1994).
6.1.2 Berninger and Swanson’s (1994) model: the development of the Translat-
ing process
One of the currently most complete proposals relative to writing development has
been proposed by Berninger and her collaborators from Hayes and Flower’s (1980)
model (Cf. Berninger, 1994; Berninger, Füller, & Whitaker, 1996; Berninger &
Swanson, 1994; Swanson & Berninger, 1996). Properly speaking it does not concern
a model explaining the development of written language. However, their idea can be
considered as an attempt to include, in the framework developed by Hayes and
Flower (1980), a modification aiming to show how the different writing processes
can develop and more particularly the Translating process (Cf. Chapter 2). Berninger
and Swanson’s (1994) proposals are interesting for several reasons. On the one
hand, they represent a synthesis of the currently available theoretical frameworks, as
well as in speaking as writing and, on the other hand, they specify the nature of
Translating sub-processes or operations. But mainly, this model enables the consid-
eration of the writing of ‘really’ beginner or novice writers. In other words, it con-
cerns a model more aiming to account for the installation of Translating sub-
processes than for their ulterior development.
Modification of Hayes and Flower’s (1980) model for a consideration of develop-

mental aspects of writing. By resuming the tripartite distribution (i.e., Planning,
Translating and Reviewing processes) of Hayes and Flower’s (1980) model, as well
as the principle of process recursion throughout writing, Berninger and her collabo-
rators propose, in different papers (Cf. references above), several modifications,
which mainly consist of additions or precisions allowing to consider the develop-
ment of the different writing processes. These modifications aim at redefining the
Translating process and at specifying processing modalities of writing processes
(and specifically their relationships with different memory systems), the regulation
of these processes or sub-processes by a metacognitive system and, finally, the con-
sideration of interindividual differences in the activity of text production.
According to Berninger and Swanson (1994), the writing activity would be com-
posed of three processes: Planning, Translating and Reviewing. As for Hayes and
Flower (1980), the Planning process would allow Goal-Setting, Generating and Or-
ganising the content, while the Reviewing process would operate through text Read-
ing, error Detection and Correction. However, Berninger and her colleagues high-
light that the ‘box’ of the Translating process is ‘empty’ in the model elaborated by
Hayes and Flower (Cf. nevertheless, to discuss this point of view, our analysis about
the Translating process within Hayes and Flower’s model in Chapter 2). They thus
propose to divide this process into two sub-processes:
• The Text Generation sub-process involves the transformation of retrieved ideas
into linguistic representations in Working Memory. This sub-process is dedi-
cated to the processing of word, sentence, paragraph and text.
• The Transcription sub-process carries out the translation of representations in
Working Memory into written symbols on the sheet of paper. This sub-process,
by allocating both operations of phonological and orthographic coding, but also
fine motor skills, allows spelling words, punctuating and graphically executing
the text.
Berninger et al. equally specify the importance of memory systems during writing.
First of all, it would be necessary to include in their model not only Long Term
Memory, as did Hayes and Flower (1980), and Working Memory, but also a specific
Short Term Memory. The authors indeed distinguish between Working Memory and
Short Term Memory. While Working Memory would play a crucial role in high-
level skills, Short Term Memory would be essentially solicited during Graphic Tran-
scription activity (Cf. Swanson & Berninger, 1994).
The Monitoring process, which controls the releasing of the three writing proc-
esses in Hayes and Flower’s (1980) model, does not feature in this model. In fact,
according to Berninger and her collaborators, the Monitoring would not only con-
cern writing, but must be considered as a general metacognitive control process.
Consequently, the authors judge the control proposed by Hayes and Flower (1980)
as inadequate. They state state it is limited to writing processes, while it would, for
them, not only allow to regulate the whole writing process, but also the memory sys-
tems intervening in the writing progress. The authors therefore specify that meta-
cognitive knowledge on writing is not exclusive to the three writing processes. Thus,
the control system (Monitoring) of Hayes and Flower’s (1980) model could be
qualified as a general metacognitive system that regulates all the cognitive activities.
Berninger et al. insist on the fact that it is necessary to take into account writers’
individual differences, which are not considered in Hayes and Flower’s model.
However, Berninger and Füller (1992) have shown, concerning the Translating
process, that these differences constrain more transcription than text generation (Cf.
equally: Engelhard, Gordon & Gabrielson, 1992). These authors have shown, for
example, that girls were faster than boys during text handwriting.
Finally, the authors underline that non cognitive variables such as affects, moti-
vations and social context (mainly the addressee and the text communicative func-
tion) play a role in the Task Environment (Cf. Oliver, 1995, who has shown the im-
portant role of instructions on young writers’ productions). To go further than Bern-
inger et al., Chin (1994) proposes to integrate, in the context of writing, the writer’s
individual personality and the historical circumstances (that is to say political and
economic) in which the written production appears. According to her, the context
must henceforth be considered as a far larger object. It is finally interesting, in this

view, to show that Hayes (1996) has taken into account these different objections (as
well as several others) by integrating, in the reformulation of the 1980 model, a
component labelled ‘Motivation/Affect’ (Cf. General Introduction).
Progressive installation of writing processes. Besides their modifications and com-

plements to the initial Hayes and Flower’s (1980) model, Berninger and Swanson
(1994) propose a very relevant analysis of the installation and the development of
the different writing processes and sub-processes (Cf. Figure 29).
The opposition between beginner and expert writers is therefore important for
them. For expert writers, the three writing processes are mature and can interact; for
beginner writers, they are developing and would have specific development speeds.
The authors, by means of the following three frameworks, illustrate a description of
the installation of these different processes.
According to Berninger and Swanson (1994), Planning, Translating and Review-
ing processes would not be operational from the beginning of writing apprenticeship
but would appear in a specific order during the course of its development. More spe-
cifically, the Translating process could appear before Planning or Reviewing proc-
esses.
• Concerning the Translating process, its two sub-processes (Text Generation and
Transcription, according to Berninger and Swanson, 1994) could develop either
at the same time, or with different speeds. Indeed, as Berninger, Yates, Cart-
wright, Rutberg, Remy and Abbott (1992) have highlighted, the young writers
would not have any problem retrieving ideas, but could have difficulties to lin-
guistically express them. More precisely, following Berninger, Cartwright,
Yates, Swanson and Abbott (1994), two components would play a central role
in the development of the Translating process. These components are spelling
and graphic realisation (handwriting). Spelling development has been largely
studied and Templeton and Bear (1992) or, more recently, Martlew and Sorsby
(1995) have observed that spelling is, at the beginning of the learning of writ-
ing, very costly in cognitive resources. Graphic execution has been mainly stud-
ied in comparing oral and written productions (Bereiter and Scardamalia, 1987;
McCutchen, 1987; or more recently, Bourdin and Fayol, 1994, 1996). The re-
sults of these different works show that, in children, written texts are globally
always shorter than oral productions. According to Bereiter and Scardamalia
(1987), this difference would be linked to (a) the slowness of graphic execution,
which can entail information loss (by memory delay), and/or (b) the interfer-
ence between the different writing processes, which is a consequence of the non
automatisation of execution in young writers leading to a heavy processing
load. Bourdin and Fayol (1994, 1996) equally defend this last interpretation, in
order to explain writing and speaking differences in young writers.
• The Planning would be first very local (sentence by sentence), before being
both more global and concerning larger text portions. Thus, Text Generation
sub-process would first appear at a word level, then at a sentence level and fi-
nally at a textual level. In addition, at the beginning of writing apprenticeship,

planning would appear throughout writing, while it would then appear before
translating and graphic execution. This fact is analogous to Bereiter and Scar-
damalia’s (1987) Knowledge Telling Strategy (Cf. paragraph 2.2.1. thereafter)
or from McCutchen’s (1996) proposals about children’s writing. For all these
authors, for children or novice writers, planning would mainly consist of
retrieving ideas whereas the next process would immediately translate these
• ideas.
Similarly, the Reviewing process would first appear at a word level before at a
sentence level, then at a paragraph level and finally at a text level.
1- Model of Writing Development in the Primary Grades :

Relative emergence of Component and Sub-Component Processes
Translating Note : The higher a component or sub-

component in the figure, the earlier it
emerges. Once it appears it continues to
develop.
Transcription
Text Generation
Reviewing/Revising
Word Some on-line revision
of Transcribed words
Sentence Occasional on-line
(clause) revision of sentence
Paragraph
On-Line Planning (multiple
clauses)
On-line algorythms
and schemata for
planning the next
sentence (local planning)
2- Model of Writing Development in the Intermediate Grades :

Continues development and relative emergence of Component and Sub-Component processes
Working Memory
Translating
Transcription
Automatized Text Generation
or Becoming Posttranslation
Automatized Reviewing/Revising
word sentence paragraph

paragraph
discourse
structures
preplanning literacy
genre
3- Model of Writing Development in the Junior High Grades :

Continued development of component processes, and Working Memory and
Metacognitions about Writing
Working Memory
Translating
transcription text generation

Posttranslation
Revising
Preplanning word sentence text
word sentence text
Figure 29: The development of writing processes,

adapted from Berninger and Swanson (1994). Copyright © 1994 by Elsevier Science.
Finally, according to Berninger and al., these three processes would be constrained
by writers’ Working Memory span and metacognitive knowledge about the different
processes. These would be mainly these two dimensions (Working Memory span
and Metacognitive Knowledge) that would be responsible for writing development
and inter-individual differences (Cf. above).
6.1.3 Conclusion: from Translating to Planning expertise

In Martlew’s (1983) and Berninger and Swanson’s (1994) models, the expertise de-
velopment is often approached in terms of a very slow installation of new writing
processes, which will allow the writer to elaborate texts (1) with a more strongly or-
ganised, structured and coherent content, (2) whose communicative objective would
be particularly well adapted to the reader, and (3) using complex linguistic processes
of cohesion to mark the text coherence. In parallel with these changes, the expertise
would equally lead to an automatisation of low-level processes (such as graphic
transcription, for instance), allowing the writer to progressively devote more atten-
tion to high-level processes (such as planning). The model of Berninger and Swan-
son (1994, and parallel proposals made by Berninger and collaborators) appears es-
pecially interesting because it attempts to explain, on developmental and differential
perspectives, the installation of the different Translating sub-processes in connection
with the whole cognitive system (Working Memory, Long Term Memory, Metacog-
nition, etc.).
Berninger and Swanson’s (1994) model has however some limitations, which are
linked to the authors’ main objective. Focussing on the development of writing ex-
pertise, this model does not really explain the appearance and the development of
the Planning process. The authors did not precisely explain the installation and the
complexication of processing linked to Planning, to account for writing expertise.
Such an evolution of Planning efficiency has been more particularly studied by
Bereiter and Scardamalia (1987) in teenagers, and by Flower, Schriver, Carey, Haas
and Hayes (1989), in adult writers.
6.2 Development of expertise as a complexification of the Planning process

On a general point of view, the development of planning activity, according to
Baker-Sennett, Matusov and Rogoff (1993, Cf. also Hudson & Fivush, 1991),
mainly consists (1) of learning to plan in a more efficient way, (2) of adapting
strategies according to the task environment, circumstances and constraints, and (3)
of being more and more capable to anticipate constraints linked to the specific char-
acteristics of the task.
In the framework of writing activity, the Planning process allows a global
(re)organisation and adaptation of the text content (Cf. Chapter 1). According to
Berninger, Füller and Whitaker (1996), it can therefore be considered as a process
susceptible to apprenticeship and, consequently, which can develop. According to
Flower and Hayes (1981a and b), expert and novice writers would essentially differ
by (1) the nature and the quality of goals that they determine during planning, and
equally by (2) the more or less important skill to use planning to guide the whole
writing process. More precisely, according to Bereiter and Scardamalia (1987), or
Burtis, Bereiter, Scardamalia and Tetroe (1983), the expertise in planning would im-
ply the possibility to plan both before and during the composition at a global level
(for example, the text plan) as well as a local level (such as the connection between
two sentences).
6.2.1 Bereiter and Scardamalia’s (1987) model: from Knowledge Telling Strat-
egy to Knowledge Transforming Strategy
As previously underlined in the General Introduction of this book, Bereiter and
Scardamalia (1987) have described two strategies in the framework of a double
model, distinguishing Knowledge Telling Strategy from Knowledge Transforming
Strategy.
The Knowledge Telling Strategy. The writing strategy generally used by young
and/or beginner writers has been described within the Knowledge Telling Strategy.
This strategy necessitates translating knowledge units while they are retrieved, with-
out operating a real (re)organisation of the text content. Consequently, the Knowl-
edge Telling Strategy, even if economic, allows ensuring only weak local coherence
between two ‘idea units’. Although elementary, this strategy nevertheless assumes
that the writer could (1) have acquired knowledge about the text type (through the
operation labelled ‘Locate Gender Identifiers’), (2) respect writing instructions (with
the ‘Mental Representation of Assignment’ operation), (3) identify the text topic or
theme (using the ‘Locate Topic Identifiers’ operation), and (4) finally judge a min-
ima the appropriateness of retrieved contents, according to writing instructions. This
appropriateness is achieved by the ‘Run Test of Appropriateness’ operation that al-
lows the selection of domain knowledge units as a function of the objectives of the
writing task (Cf. in Chapter 1, the architecture of the Generating sub-process).
Compared with Berninger and Swanson’s (1994) proposals about the ‘very first’
writing development, the Knowledge Telling Strategy is relatively ‘more expert’. It
suggests that writers have acquired certain kinds of text knowledge in Long Term
Memory (i.e., Discourse Knowledge). It is certainly in this meaning that, according
to Bereiter and Scardamalia (1987), this writing strategy would only appear around
9-10 years of age. At this period, the manipulation of language and the notion of text
would already be functional, at least in their first forms. The functioning of the
Knowledge Telling Strategy can easily be compared to those of the Translating
process such as described by Hayes and Flower (1980) for adult writers. It would
require a text produced without the use of the Planning process by the writer. In this
way, McCutchen (1988) considers that children’s writing process can be summa-
rised as a ‘think it, write it’ strategy. This corroborates previous results that she ob-
tained by providing children with the first and last sentences of a text that they had
to complete (Cf. McCutchen, 1986, 1987). Her observations and conclusions seem
to verify the model developed by Bereiter and Scardamalia (1987) that characterises
novice writers. These writers would ‘directly’ transform their thoughts into words.
However, this strategy, necessarily used by novice or beginner writers, could also be
used by experts, especially when they are confronted with a very easy writing task,
such as writing Christmas cards, for instance (Cf. McCutchen, 1994).
The Knowledge Transforming Strategy. The Knowledge Transforming Strategy, op-

posed to the Knowledge Telling Strategy, allows the writer to elaborate texts with a
more strongly organised and coherent content, by taking into account both Content
and Rhetorical constraints (Cf. also General Introduction). In a more precise way,
the Knowledge Transforming Strategy includes the Knowledge Telling Process,
which then becomes a final processing stage, being related to a Translating stage,
leading to linguistically translate retrieved knowledge units. Contents that are com-
municated to the Knowledge Telling Process are however not simply retrieved from
Long Term Memory and then evaluated, but they are also reorganised (transformed)
by a system of problem-solving and establishment of goals (Problem Analysis and
Goal-Setting), which modifies (reorganisations and/or creations) the domain con-
tents according to rhetorical criteria or constraints. This content modification is car-
ried out by a series of operations allowing, among others, to inter-connect knowl-
edge units, to select relevant units, etc. (Cf. in Chapter 1, the architecture of the Or-
ganising sub-process).
In addition, as Scardamalia and Bereiter (1991) have underlined, the certainly
most crucial difference between the Knowledge Telling Strategy and Knowledge
Transforming Strategy is the presence, in the second strategy, of a Feedback be-
tween the two components labelled Knowledge Telling Process and Problem Analy-
sis (Cf. also General Introduction and Chapter 1). This feedback is absent in the
Knowledge Telling Strategy, indicating that, in novice writers, some possibility to a
posteriori analyse the written product, and therefore some possible feedback on the
retrieved content to modify it according to Rhetorical constraints, would not exist.
The presence of this feedback, in the Knowledge Transforming Strategy, would lead
expert writers to consider or to reconsider contents to be translated in relation with
previously translated contents. This could thus release the activity of problem-
solving operations. In this way, according to Scardamalia and Bereiter (1991), the
Knowledge Transforming Strategy is more than a simple expert strategy of text writ-
ing, it equally represents a model of knowledge acquisition through the writing ac-
tivity (Cf. Chapter 1). According to these authors, the expert writer would be taken
through a process leading her/him to continuously perfect his/her domain or written
expertise. Writing domain knowledge would thus lead to reorganise and strengthen
this knowledge. Becoming more expert in one specific area would facilitate textual
composition about this area.
Progressive evolution from Knowledge Telling to Knowledge Transforming Strat-

egy. On a developmental side, the passage from Knowledge Telling Strategy to
Knowledge Transforming Strategy is very progressive and would only be efficient
around the age of 16 years, at least for complex text types, such as argumentative
texts. Intermediate strategies between Knowledge Telling and Knowledge Trans-
forming would exist, showing the progressive appearance of problem-solving opera-
tions, able to elaborate the textual ‘Main Points’.
Bereiter, Burtis and Scardamalia (1988) have studied, in the framework of
Knowledge Telling and Knowledge Transforming strategies, the nature of processes
in each of these strategies. They have collected and analysed verbal protocols from
participants of different levels (9, 11, 13 years old and adults). They have asked
them to write an argumentative text with the instruction to ‘plan’, as deeply as they
are able, the content before beginning to write (the instructions prompt them to take
notes and explain the potential organisation of a text, etc.). The verbal protocols ac-
count for the functionality of six problem-solving operations, which characterise the
Knowledge Transforming (i.e., ‘Search’, ‘Delimit’ the content and ‘Fit’, ‘Cohere’,
‘Structure’ and ‘Review’ this retrieved content; Cf. Chapter 1). The quality of the
text content organisation has been evaluated through three levels of elaboration of
‘main ideas’ (Main Points). Level 1 is characterised by Main Points limited to the
statement of the general text topic while Level 3 lists not only the statement of the
theme but equally some subordinated Main Points (organisation of the text content
according to different hierarchised sub-goals). Level 2 is thus considered as an in-
termediate level.
By analysing the nature of verbal protocols, according to these three levels of
content structuring, the authors have shown that the Main Points of the first level
would result from a writing process typical of Knowledge Telling. Indeed, these
Main Points would exclusively and directly depend on the text topic and its literary
gender, according to the writing instruction. They are directly retrieved from Long
Term Memory and would be constituted by knowledge units initially contained in
the knowledge network (Burtis, Bereiter, Scardamalia and Tetroe, 1983). The writ-
ing mode of Level 1 would therefore not particularly rely on problem-solving opera-
tions, opposed to Level 3 (Knowledge Transforming). The results of this experiment
confirm that Knowledge Telling and Knowledge Transforming Strategies are carried
out by different modes of processing. Moreover, the authors have shown that the
evolution of the effective use of problem-solving operations was in connection with
participants’ age. Older participants (13 years of age and students) could write texts
with a more important number of Main Points, while this processing activity, under
the double constraint of Rhetorical and Content Spaces, could not be used by the
youngest children (9-11 years). The latter used a text content organisation which
would only rely on a limited number of Main Points, with a relatively low hierarchi-
cal level and depending directly on the main theme of the text. However, Scar-
damalia and Bereiter (1987) specify that the writing expertise must not be consid-
ered as a passage ‘in all or nothing’ from the Knowledge Telling Strategy to the
Knowledge Transforming Strategy. These two writing modes must rather be envis-
aged as two opposed processing strategies between which there are intermediate
strategies, which would represent many possible evolutions from the Knowledge
Telling Strategy to the Knowledge Transforming Strategy; and the composition of
texts with an increasingly organised content.
Waters and Lomenick (1983) have led one of the clearest demonstrations of this
evolutive aspect. These authors have proposed, to adults (students) and children (8
to 11 years old), to produce a descriptive text on the basis of key words whose or-
ganised presentation suggests the main theme of the text and its different parts and
subparts. In addition, the participants could insert these key words in their text in the
order they want. The complexity of the organisation of the content has been evalu-
ated with a scale of seven levels. The lowest level corresponds to an absence of rela-
tionships between described facts. Conversely, the highest level corresponds to a
maximal hierarchical organisation, within which all the different facts are related
and organised in different subtopics, themselves connected through the main theme
of the text. The results obtained by Waters and Lomenick show two important phe-
nomena. On the one hand, they confirm the fact that the hierarchy of information
contained in a text is more important in adults than in children. On the other hand,
they show that the complexication of this organisation evolves with age (from 8 to
11 year olds). The evolution of the content structuring would mainly operate by in-
creasingly marked relationships between knowledge units, allowing to unify these
units within higher-level conceptual categories. It is interesting to observe that the
evolution of this processing mode could also be described as a progressive effi-
ciency of the operation ‘Identify a Category’, composing the Organising sub-process
in Hayes and Flower’s (1980) model (Cf. Chapter 1).
The Knowledge Transforming Strategy and adult writers: the necessity of a deeper
description. Scardamalia and Bereiter (1987) have essentially adopted a develop-
mental perspective to evaluate the relevance of their two writing strategies. It gener-
ally seems that adults are more able to adopt a Knowledge Transforming Strategy
than children (Schneuwly, 1997). Scardamalia an Bereieter, on the basis of various
experimental results, consider the age of 16 years as a limit under which the adop-
tion of a real Knowledge Transforming Strategy would be difficult. This however
does not imply that an adult is always capable to adopt a Knowledge Transforming
Strategy to adequately write a text.
Thus, according to Bereiter and Scardamalia (1987), these two strategies
(Knowledge Telling and Knowledge Transforming) would allow the writer to de-
velop from a very local planning to an increasingly global planning of the text con-
tent. Nevertheless, the authors give only little information about this complexifica-
tion and the manner whose organisation operations of main points, for example,
evolve. Only some intermediate forms between Knowledge Telling and Knowledge
Transforming are evoked by Bereiter, Burtis and Scardamalia (1988), without being
really described and clarified. The complexication of Planning processing, as one of
the main characteristics of the development of writing expertise, has been more pre-
cisely described by Flower, Schriver, Carey, Haas and Hayes (1989) in adults. Ac-
cording to these authors, such an evolution of the expertise would portray the possi-
bility to build and to elaborate an increasingly complex system of plans, including
content as well as processing plans.
6.2.2 The notion of Constructive Planning (Flower, Schriver, Carey, Haas and
Hayes, 1989)
According to Flower and Hayes (1980), the general activity of planning, in the pro-
gress of writing, would elaborate several types of plans: plans ‘To Do’, ‘To Com-
pose’ and ‘To Say’ (Cf. Chapter 1). The complexication of the ‘To Compose’ part of
planning has been more particularly described by Flower, Schriver, Carey, Haas and
Hayes (1989), so as to explain the development of writing expertise, for the writing
of expository texts (Cf. also: Carey, Flower, Hayes, Schriver, & Haas, 1989). Ac-
cording to these authors, expert writers could be distinguished from novices by their
more important competence, not only in elaborating pragmatic (‘To Do’) and con-
tent (‘To Say’) plans, but especially in integrating the construction of these different
content plans within a unique structure (Constructive Planning). This structure
would control the whole progress of the writing process (i.e., for the authors, Net-
work of Working Goals).
Before reaching such a mode of processing (i.e., Constructive Planning), novice
writers would operate on the basis of more economic writing strategies, mainly be-
cause of their limited capacity resources. The elaboration of the text content would
be intrinsically guided by the domain knowledge structure (Knowledge Driven) or,
at least, by the application of a text plan (Schema Driven). It is interesting, on this
last point, to note that the writing strategies described by Flower and her collabora-
tors, to account for the difference between different expertise levels, seem to be
therefore relatively similar to Scardamalia and Bereiter’s Knowledge Telling
(Knowledge Driven) and Knowledge Transforming (Constructive Planning) strate-

gies.
• The first strategy, labelled ‘Knowledge Driven’, would be, according to Flower
et al. (1989), the simplest for the writer because it only requires a good knowl-
edge of the domain area whose internal organisation, logical links in the knowl-
edge network, hierarchical structures, etc., are assumed to largely guide the
elaboration of the text content.
• The second strategy, named ‘Schema Driven’, is considered as more complex
because it assumes that the writer (a) possesses knowledge (under possible
schemas) on how to manage different writing tasks, (b) identifies the schema
susceptible to carry out the execution of a given writing task, and (c) could par-
ticularise and specify this schema as a function of the effective task demand. In
these strategies, the writer’s activity is guided by a knowledge structure that
could be textual or relative to the domain.
• The third strategy, ‘Constructive Planning’, supposes that the writer manages
five types of activity to plan the text without resorting to preliminary knowledge
structures. Thus, the writer must be able:
o to elaborate her/his own initial representations (goals which guide
planning) about the task. In order to do so, s/he has to explore
components of the writing task, to retrieve contents, to take into
account the addressee and the text type. This analysis, particularly
costly in cognitive resources, would end to the elaboration of gen-
eral and abstract writing goals that would guide and constrain the
whole activity,
o to decompose abstract goals in more local subgoals (i.e. generating
a network of working sub-goals),
o to integrate these local sub-goals in a global coherent strategy (i.e.
integrating plan, goals and knowledge),
o to particularise in a plan taking into account constraints and lin-
guistic conventions (i.e. instantiating abstract goals), and, finally,
o to solve all conflicts generated by this very expert type of planning
(i.e. resolving conflicts).
According to Flower et al. (1989), differences between novice and expert writers
would appear in the ability with which experts could proceed to the processing of
‘Constructive Planning’, opposed to applying the ‘Knowledge and Schema Driven
Planning’. Being more voluntarily involved in this mode of complex planning than
novices, experts would be more able to generate a more elaborated network of work-
ing sub-goals, and these sub-goals could be more numerous, more hierarchised and
more strongly connected. The control of the writing activity and the consideration of
all linguistic, pragmatic and domain constraints would then be deeper and more effi-
cient. Finally, expert writers would be better in managing conflicts occurring in the
progress of the different writing processes.
6.3 Conclusion: models of writing expertise development are mainly descriptive

The objective of the main models of written production, described in this chapter,
was to account for the development of writing expertise. Martlew’s (1983) model,
relatively general, describes the different processes and the different pieces of
knowledge whose development is indispensable in order that writing not be consid-
ered as a simple translation of oral discourse. Berninger and Swanson’s (1994)
model aimed to complete Hayes and Flower’s (1980) model, so as to give it a devel-
opmental perspective. The strategies elaborated by Scardamalia and Bereiter (1987)
have shown two specific architectures as a function of the writer’s expertise. Flower
et al.’s (1989) proposal, which does not really describe a model, has nevertheless
precisely analysed the planning process in expert writers. All these models are, by
their own objectives, relatively complementary and offer a good overview about the
complexity of the development of writing expertise. Generally speaking, two re-
marks can be formulated, concerning the impact and the limits of these works:
• These models focussed either: (a) on the emergence and the elaboration of the
first writing skills by studying more particularly the evolution of the Translating
process, such as in Martlew’s (1983) or Berninger and Swanson’s (1994) mod-
els, or (b) on the more belated appearance and the complexity of the Planning
process, in the case of Bereiter and Scardamalia’s (1987) or Flower et al.’s
(1989) models. It is then interesting to observe that proper developmental mod-
els of the whole writing activity do not truly exist or, conversely, developmental
models concerning more specifically Reviewing or Monitoring processes. This
observation can be linked to the fact that these two components of the writing
activity are relatively complex in themselves and can also signify that each of
the writing processes do not benefit from the same interest in research about de-
velopment.
• The main models or element of models that approach the development of Trans-
lating and Planning processes are interesting because they propose a relatively
fine analysis of the installation of these processes as well as increasingly elabo-
rated and efficient strategies. However, these models remain relatively descrip-
tive and are not very clear concerning the determinism or the developmental
factors that underlie the modification and the development of processing archi-
tectures.
This last remark seems very fundamental. Indeed, the study of development does not
simply involve listing different development stages, but rather in accounting for this
development. In others words, a model can be considered as developmental only if it
proposes a certain number of developmental variables that can be responsible for the
on-line evolution of the functioning modalities of processes and/or of the nature of
processed knowledge. In the main model of writing expertise development, the gen-
erally evoked determining factor is mainly the increase of Working Memory capaci-
ties, as it is described in the next part of this chapter.
7 WORKING MEMORY TO EXPLAIN THE DEVELOPMENT

OF WRITING EXPERTISE
As described in Chapter 5, maintaining and processing spans in Working Memory
(as defined by Baddeley, 1986) play a central role in writing activity. They both in-
fluence (1) the simultaneous activation and progress of a more or less important
number of processes, and (2) the parallel and temporary maintaining of a more or
less large set of representations.
It is then coherent to think that the more the Working Memory capacities are in-
creased (in terms of processing or storage; Cf. Chi, 1976), the more the writer is able
to use complex and expert writing strategies, by simultaneously managing increas-
ingly numerous constraints. According to Bereiter and Scardamalia (1987), Bern-
inger and Swanson (1994), or McCutchen (1996), the evolution of Working Mem-
ory capacities would be an explanatory factor of both interindividual differences
(i.e., expert writers’ processing capacities would be more important and more effi-
cient than those of novice writers) and the progressive evolution of writing expertise
(i.e., adults would possess processing capacities more powerful than those of chil-
dren). Many experimental studies and theories tend to confirm this idea.
7.1 Relations between Working Memory and writing expertise

Concerning individual differences, Benton, Kraft, Glover and Plake (1984) have, for
example, shown that the opposition between good and poor writers would mainly
come from differences in general and elementary information processing programs.
These programs would be more efficient in good writers than in poor writers. More-
over, good writers would manipulate more rapidly and more efficiently pieces of in-
formation in Short Term Memory. Jeffery and Underwood (1996) have confirmed
these results in children by showing that results of general memory span tests (con-
cerning words, digits and reading) were strongly correlated with the possibility, for
participants (12, 14 and 17 years of age), to co-ordinate several ideas within a same
sentence and, via this co-ordination, to write a text necessitating an integration of
high conceptual level knowledge units.
On a developmental point of view, Stein and Miller (1992) have postulated that
the development of the memory span (as well as those of reasoning capacities)
would play a fundamental role in the writing of argumentative texts by allowing to
evaluate and to put in relationships larger sets of idea units. This conception has
been defended by Bereiter and Scardamalia (1987) early on, who consider that the
most important factor to explain writing development (and, for them, the access to
Knowledge Transforming Strategy) concerns the evolution of processing (through a
planning span) and storing (through Short Term Memory span) capacities in chil-
dren. In other words, according to Bereiter and Scardamalia (1987), the evolution
from Knowledge Telling to Knowledge Transforming Strategy would be mainly
subordinate to the Working Memory span increase. For these authors, indeed, the
fact that storing and processing span in memory is more limited in children (or nov-
ice writers) than in adults (or experts) is one of the possible explanations for the
adoption of the Knowledge Telling Strategy in order to produce a text, instead of the
Knowledge Transforming Strategy. The series of following experiments, led by
Bereiter and Scardamalia (1987), as well as by other authors, aims to validate this
hypothesis.
7.2 Some data to describe relationships between writing expertise and Working
Memory capacities.
For Bereiter and Scardamalia (1987), the implementation of the Knowledge Telling
Strategy necessitates a minima, two knowledge units to be activated in Working
Memory during the writing of a text. These units are: (1) the thematic text represen-
tation, and (2) the content unit previously translated. They would be sufficient to en-
sure the local coherence of the text (McDonald, 1980; Bereiter, 1980). However, in
the case of a mode of writing as Knowledge Telling, the cognitive cost of composi-
tion processing would be reduced because of the maintaining of only two informa-
tion units in Working Memory. Conversely, the Knowledge Transforming Strategy
would suppose, according to Yau and Bereiter (1978 – quoted by Bereiter and Scar-
damalia, 1987), the costly maintaining of a higher number of information units in
Working Memory. More precisely, Bereiter and Scardamalia’s (1987: 155-176)
works show that the access to the Knowledge Transforming Strategy supposes the
maintaining of four or five chunks (or knowledge units).
Following the idea of these works, van Wijk (1999) has recently specified the
number and the nature of these units. This author has proposed a new analysis of the
texts used by McCutchen and Perfetti (1982), to describe the evolution with age of
the content hierarchical structuring, by means of a software labelled ‘TRACE’
(Text-Based Reconstructions of Activities by the Conceptual Executive). This soft-
ware, according to the text content structure, interprets the processing operations
underlying the linguistic message (Cf. van der Pool, 1995 for a detailed description
of this method of content analysis).
van Wijk (1999) shows, in the case of argumentative texts, that the evolution of
structuring modes of text contents (on a macrostructural plan as well as on a micro-
structural plan) would depend on the number of information units that a writer is
able to simultaneously maintain in Working Memory during text writing. According
to the author, four knowledge units are necessary to produce a text whose complex
structure can testify the application of the Knowledge Transforming Strategy (in
terms of global and local coherence and motivated stylistic choices): (1) the text
topic, (2) a local concept (i.e., the microproposition that has to be textualised, as in
the local coherence strategy of van Dijk and Kintsch, 1983, Cf. Chapter 2), (3) the
global design of the text structure, and (4) a set of rhetorical considerations. In addi-
tion, according to van Wijk (1999), these knowledge units can be characterised by
their stability (static – the topic of a text must be maintained in memory – or dy-
namic – a local concept is only temporarily activated –), and their accessibility (di-
rect – the topic or the local concept can be automatically activated by spreading ac-
tivation – or indirect – the controlled choice of a rhetorical set or of a global design
of the text would be necessary). Thus, the fact that some of these units are dynamic
and have to be controlled, in their retrieving, can increase the cost (or the load) in
Working Memory and could therefore heavily weigh on novice writers’ more lim-
ited capacities. According to van Wijk (1999), this could explain why beginner
and/or novice writers have to use the Knowledge Telling Strategy, simply because
they can not maintain, in a dynamic and controlled way, more than two knowledge
units during writing activity.
According to these experimental works, it seems necessary, in order to use the
Knowledge Transforming Strategy, to be able to maintain in Working Memory a
more important number of knowledge units, that can then be assimilated to con-
straints. In this way, McCutchen and Perfetti (1982) have simulated, by means of
Anderson’s (1976) ACT* model, the role of these constraints, to explain develop-
mental differences in texts written by different level graders (2nd grade, 4th grade,
pupils at the beginning of college). The characteristics of these texts have been imi-
tated by simply varying, in the ACT* model, the number of simultaneously activated
constraints. This simulation has confirmed that the youngest graders could manage
only one constraint, while the oldest children could simultaneously manage at least
three constraints. The authors have therefore shown that an increase of the number
of constraints, in the Working Memory of the ACT* model, entailed the system to
produce simulated texts corresponding to texts written by more aged writers.
These various results, from experiments or simulations, confirm that there is a
relationship between memory span and the access to expert writing strategies. More
precisely, it seems that a difference of memory span (from a writer to another),
and/or the evolution of the span (in a same writer) could constitute central factors for
writing expertise. The question thus concerns the determinism of information main-
taining and processing span variations. Two responses, classic in cognitive psychol-
ogy and in developmental psychology, can be formulated: either (a) these variations
are consequences of a maturation effect and the greater efficiency of the cognitive
system, or (b) they are linked to an effect of practice and training for a given activ-
ity.
These different ways to consider the span differences or increase lead to different
viewpoints in the area of writing research. These viewpoints are more precisely de-
scribed in the following section.
7.3 Increase in Working Memory with age: effect of maturation

Many researchers have tried to explain mechanisms underlying the Working Mem-
ory span development. The review of papers made by Dempster (1981) shows a very
interesting statement of the explanatory power of several types of ‘non strategic’ and
‘strategic’ processes (Grouping, Chunking, Rehearsal, Search Rate, etc.), whose in-
stallation and/or evolution could be responsible for interindividual and developmen-
tal differences characterising mnemonic performances. According to Dempster
(1981), one of the most relevant factors to explain the development, with the age, of
the memory span would be related to the increase of the identification speed of in-
formation during the encoding process (Cf. also on this point, Huttenlocher &
Burke, 1976).
7.3.1 Increase in the processing speed within Working Memory
For example, Chi (1977), through a memory span task, has shown that (1) the neces-
sary duration to encode and to name items (familiar faces or not) is always longer in
children of 5 years that in adults, (2) while a strong experimental decrease in the du-
ration of item presentation for adults decreases the difference between their recall
performances and those of children. These two results confirm the fact that adults
identify stored information more rapidly than children do. According to Kail (1979,
1991), the increasing speed of ‘non strategic’ processes could be responsible for im-
portant developmental differences. It is certainly in this way that, as seen above,
Benton, Kraft, Glover and Plake (1984) have shown that good and poor writers
could be differentiated by the speed in which they process information in Working
Memory. It therefore concerns a very important aspect, that would be necessary to
integrate in future theories relative to writing development and to writing expertise.
7.3.2 Processing modification in Working Memory

The increase in information processing speed in Working Memory is not the unique
phenomenon that characterises the development of the cognitive system. It is indeed
possible to describe some modifications in the installation of memory components
and in the nature of general processing performed in Working Memory. For in-
stance, Berninger and Swanson (1994) consider that the different components of
Working Memory should gradually become increasingly efficient. More precisely,
the elaboration of memory registers seems to be only progressive in young children,
more particularly in the case of the phonological register. According to Logie
(1989), the phonological buffer would develop after the visuospatial sketchpad, and
its structure would strongly depend on children’s language acquisition and develop-
ment. Such an interdependency between general cognitive development and memory
capacities development has been very precisely described by neo-piagetian re-
searchers, and more particularly by two of them: Pascual-Leone (1970, 1987) and
Case (1985).
Constructive operators and processing increase for Pascual-Leone. Pascual-
Leone’s (1970, 1987) theory suggests that children’s cognitive system is made up
of:
• a set of cognitive, affective and individual schemes (schemes according to the
piagetian definition) that would be stored in Long Term Memory and that could
be activated independently or in groups;
• a set of seven ‘silent’ operators that comprises, for instance: (a) the constructive
operator ‘M’, which activates, transforms and co-ordinates schemes, and, (b)
the constructive operator ‘I’, which, conversely, inhibits non-appropriate
schemes (Cf. for a review: Roulin & Monnier, 1996).
More precisely, the operator ‘M’ would represent the child’s possibility to activate,
from Long Term Memory, relevant structures as compared to the execution of a
given task, while the operator ‘I’ would correspond to the child’s capacity to inhibit
or to interrupt the activation of inappropriate structures for the execution of the same
task. The idea of the operator ‘M’ is therefore very close to the Working Memory
definition, and is defined as a processing and storing space (Processing Space M),
whose capacity is limited in number of schemes or groups of schemes being able to
be simultaneously manipulated.
The important point is that, because of the maturation, the power of ‘M’ (that is
to say, generally, the cognitive capacity) would increase with the age (one unit every
two years), and in parallel, the power of ‘I’ would equally increase. Thus, with the
age, the development of the two capacities of activation and inhibition would allow
the child to commit no more errors by activating inadequate schemes and could ex-
plain, according to Pascual-Leone (1987), the evolution of thought modes as de-
scribed in the different piagetian stages.
Case’s (1985) model of cognitive development. According to Case (1985), each

child has a cognitive system composed of:
• figurative schemes in order to have representations about the world,
• operative schemes transforming some representations in order to produce new
ones,
• an internal plan of problem-solving, voluntarily elaborate, labelled ‘Executive
Control Structure’, with three components: (a) the representation of the problem
situation, (b) the representation of the objective to reach (these two representa-
tions are composed by figurative schemes), and (c) the representation of the
strategy (composed of operative schemes). This Executive Control Structure
therefore comprises all the necessary basic components for any activity in re-
sponse to the environment.
In Case’s theory, the concept of cognitive resources is defined as ‘Executive Proc-
essing Space’ or ‘Total Processing Space’. It can be defined as the maximal number
of independent schemes than children can simultaneously maintain activated when
they are solving any problem. This attentional space would be composed of an ‘Op-
erational Space’ and a ‘Short Term Storing Space’. Thus, the double function of in-
formation storing and processing shares the same cognitive space and needs the
same resources.
In order to explain children development, Case hypothesises that the Short Term
Storing Space would be extended with one unit at each new piagetian sub-stage, due
to the increase in the efficiency of cognitive operations (or schemes), which charac-
terises a given stage. However, Case postulates that the total processing space would
remain constant in the progress of development, and, consequently, the increase in
the capacity of the Short Term Storing Space would be compensated by a decrease
of the Total Processing Space. Thus, the passage from one piagetian stage to another
is due to the increase in the capacity of Short Term Storing of Information, resulting
from a greatest efficiency of operations, and partly due to the biological maturation.
7.3.3 Conclusion: maturation and experience

Bereiter and Scardamalia (1987), defending an increase in memory span as one of
the possible factors of development in children, suggest that children’s general cog-
nitive development plays an important role in the evolution of writing expertise.
This hypothesis can be justified by following Case’s and Pascual-Leone’s theories.
For these authors, the storing and processing capacity in each stage is limited, and
thus constrained the cognitive development by restricting the cognitive structure
complexity being able to be elaborated at a given age.
These two theories, both close and opposed on certain points, are very interesting
to explain relationships between children’s general cognitive development and the
evolution of processing capacities. It is therefore very surprising to observe that re-
searchers in writing, and more particularly those interested in the development, ab-
solutely never refer to these neo-piagetian models, which are nevertheless relative to
the development and, moreover try to explain the Working Memory evolution. This
disinterest could probably be due to the fact that it is theoretically very complex to
adapt neo-piagetian models to efficiently account for the development of writing
skills.
Finally, and in a more general way, this perspective about maturation does not
account for the evolution or the systematisation of writing expertise in adults. This
last phenomenon is more precisely explained through a contextual increase of mem-
ory capacities, as a consequence of practice, training, and experience.
7.4 Increase in Working Memory capacities as a function of a specific activity: ef-

fect of practice and training
In the preceding conception of Working Memory development, the evolution is
nearly considered in terms of general maturation, entailing the possibility to apply
more rapid processes, to maintain or process more information, or to process infor-
mation differently (according to the piagetian stages). An opposing idea is that the
increase in the span would be only relative, that is to say contextually linked to the
evolution of the expertise in a given task. It can then operate by two mechanisms:
(1) the automatisation of some processes allowing to allocate more cognitive re-
sources to other processes necessary for a given task, and (2) the installation of
processing strategies (storing and retrieving, for example) in order to maintain and
thus to process more information in Working Memory, in a given activity. These
strategic processes can be illustrated by the implementation of a particular knowl-
edge structure recording complex storing and retrieving procedures: the ‘Long
Term-Working Memory’ (Cf. Chapter 5).
7.4.1 Automatisation of processes with practice and an adaptation of the Capac-

ity Theory by McCutchen (1996)
The concept of automatisation, which comes from research in cognitive psychology
concerning the acquisition of knowledge, has been more particularly developed by
Anderson (1983a, 1990, 1993), or by Logan (1988) and Logan and Etherton (1994).
These works show that the automatisation of a process appears by means of a regu-
lar practice of an activity. An automatised process theoretically does not need any
conscious control, and, thus, it does not require cognitive resources to be released
and to execute processing.
Automatisation and text writing. According to Daiute (1984), this phenomenon of

process automatisation would largely explain the opposite differences between nov-
ice and expert writers. For instance, Glynn, Britton, Muth and Dogan (1982), in the
case of argumentative text writing, have shown that the lightening of the cost of
low-level processes (i.e., graphic transcription and linguistic translation) could lead
to a considerable improvement in the quality of recovery and text content organisa-
tion. These authors, by distinguishing four processing levels in the elaboration of
this type of text (retrieval of arguments, logical organisation of these arguments,
sentence generation and establishment of punctuation, and spelling), have built an
experimental situation within which they have gradually increased the processing
costs by increasing the number of constraints the writer has to manage. The most
expensive writing condition suggests that participants take into account the four
processing levels (from recovery to spelling), while the less constraining condition
only necessitates the simple statement of a list of arguments, without particular at-
tention to their organisation or to the syntax. Establishing thus a four-points scale of
constraints, the authors have observed that the elimination of the mental load im-
posed by low-level processing (spelling, syntax) leads to an increase in the number
of produced arguments and also a better text organisation. These results reveal that
the decrease in the cognitive cost, imposed by linguistic translation levels, allows al-
locating more resources for knowledge transformation (Planning level).
Some writing processes can be more easily automatised. As shown by Glynn,
Britton, Muth and Dogan’s (1982) experiment, some Translating sub-processes or
operations would be more particularly permeable to the automatisation, via an inten-
sive and long practice. It can concern, as previously mentioned in the framework of
Martlew’s (1983) and Berninger and Swanson’s (1994) models, graphic execution
processing or grammatical and lexical processing (Berninger, Vaughan, Graham,
Abbott, Abbott, Rogan, Brooks and Reed, 1997).
Concerning graphomotoric processing, the most important point certainly relies
on the particularly long duration of handwriting development in children. According
to van Galen (1991), the graphic production is, compared to the other activities that
need an apprenticeship during childhood (for example: to walk, to speak, to draw or
to practice a sport), mastered very late. A child begins to write at 6 years of age, but
s/he acquires an adult’s speed of handwriting only around 15 years of age. It there-
fore involves a very long apprenticeship, compared to other activities. In this way,
Abbott and Berninger (1993) have evaluated the decrease in the cost of transcription
processes in 6 to 11 year olds, by comparing, in speaking and writing, some clues
able to account for the quality of different processes (from graphomotoric to text as-
pects). For these authors, the constraints of graphic planning would only constrain
higher-level processes in 6 year old children. Bourdin and Fayol (1994) have con-
firmed the age of the decrease of graphic constraints, by showing that serial recall
performances were more important for spoken language than for writing, only in 6
year olds. At this level, such a difference could be explained by the difficulty to op-
erate both an orthographic and a graphic transcription of lexical items. It seems
therefore that the transcription cost gradually decreases from 6 to 11 years old,
probably due to the automatisation of these processes.
The effect of graphomotoric constraints, due to a non-automatisation or a belated
automatisation of these types of processing, has been made apparent by Graham
(1990) in 9 to 11 year old children with learning difficulties, but without any motor
or sensory handicap. It appears, for these children, that the necessity of a strong con-
trol of processes involved in the graphic transcription suggests an important amount
of cognitive resources, leading to a decrease in text quality, concerning both idea or-
ganisation, number of transcribed ideas and global content coherence. Conversely,
when these children are asked to slow down their writing speed, the competition (in
terms of allocation of cognitive resources) between levels is considerably reduced.
According to Graham (1990), the cognitive resources necessary for graphic tran-
scription would lead to an impossibility to maintain in Working Memory numerous
knowledge units and to proceed to their organisation. If such constraints appear even
in older children with learning difficulties, they would however rapidly disappear in
children without any particular difficulty.
Concerning lexical and grammatical processing, McCutchen, Covill, Hoyne and
Mildes (1994) have experimentally analysed the functioning of sentence generating
and lexical retrieving in children who were more or less skilled in writing activity.
They have shown that clever writers (1) perform the sentence generation sub-
process better, (2) have higher reading and speaking spans (evaluated with Reading
Span and Speaking Span Tests), and (3) are more rapid and more precise in lexical
decision tasks than the less skilled children. Thus, for these authors, certain Translat-
ing sub-processes or operations, due to their automatisation, are more easily acti-
vated because they only need few cognitive resources. This could lighten the Work-
ing Memory load during the production and free cognitive resources for other proc-
esses. Indeed, the Generating sub-process is therefore equally expensive in cognitive
resources and can compete with the Translating process (McCutchen, 1994;
McCutchen, Covill, Hoyne and Mildes, 1994). This ‘balancing’ among the sharing
of cognitive resources for the different processes is constrained by the respective
need of each process (and sub-process), as a function of their automatisation, and by
the Working Memory span, which delimits the general pool of cognitive resources.
The Capacity Theory for writing (McCutchen, 1994, 1996). McCutchen (1994,
1996) has proposed an explanatory theory of relationships between memory span,
process automatisation and writing expertise development. Her proposal globally
concerns an adaptation to writing of Just and Carpenter’s (1992, Cf. Chapter 5)
‘Theory of Capacity’, which has been elaborated to account for reading and com-
prehension expertise. According to this theoretical adaptation, the writing expertise
would depend on:
• Working Memory span and cognitive resources, which are permanent and de-
termined for a given writer,
• the necessity, because of this limited processing capacity, to operate, in a per-
manent manner, a sharing of cognitive resources between the different writing
processes, and
• the automatisation, which notably allows to reduce the cognitive cost of the dif-
ferent processes in Working Memory.
These three principles lead to two important consequences.
• Firstly, Working Memory processing and maintaining capacities does not in-
crease due to the practice, but the writing process cost in cognitive resources
decreases (due to the automatisation). In this case, the writer can simultaneously
release more writing processes and maintain a greatest number of representa-
tions or constraints.
• Secondly, as Berninger and Swanson (1994) and McCutchen (1996) underline,
the three main writing processes (Planning, Translating and Reviewing) do not
simultaneously develop and do not require the same needs in terms of cognitive
resources during the writing expertise development. Thus, when Working
Memory capacities are intrinsically weak and when the writing process (or sub-
process) functioning is not sufficiently automatised, the different writing proc-
esses would function as if they were encapsulated (Cf. Fodor, 1983 and Chapter
4) and not interactive. According to McCutchen (1996), this processing encap-
sulation, which characterises novice writers, would be an artifact due to limita-
tions of their cognitive resources. These limitations hinder the possibility to
proceed, in an interactive and fluid way, to the implementation of several proc-
esses (Cf. also Cossu & Marshall, 1990).
In the framework of the Capacity Theory adapted for writing, the limits of informa-
tion processing capacity would constrain novice writers more heavily because most
writing processes are not yet automatised but are all under the writer’s control. In-
deed, beginner writers, often only being able to execute one unique operation at one
time, are frequently cognitively overloaded. Conversely, in expert writers, the man-
agement of writing ‘mechanical’ aspects (such as transcription, spelling, etc.) would
be entirely or almost entirely automatised. They could then focus their attention on
higher level activities, such as those related with idea organisation and expression,
as a function of text type, topic and potential addressee (McCutchen, 1996).
Conclusion: does automatisation only concern low-level processing? These different

works or theories globally indicate that the automatisation of low-level processing
can facilitate high-level writing processes because some cognitive resources are then
available for these kinds of processing. However, if, in expert writers, grammatical
and lexical processing could be largely automatised (Bock, 1982), it appears diffi-
cult to consider an automatisation of the Planning process, whose functioning seems
to be highly strategic and controlled.
For non-automatic processing, it seems that processing capacities can be in-
creased, on a functional level, by using expert and controlled processing strategies.
7.4.2 Installation of strategic processing with training: the Long Term-Working

Memory
Ericsson and Kintsch (1995) have proposed a new idea of the memory component,
labelled ‘Long Term-Working Memory’ (Cf. Chapter 5, for a description of this
memory). The Long Term-Working Memory can be considered as a structure of
strategic storing and retrieving of information from Long Term Memory. The main
role of Long Term-Working Memory is, according to Ericsson and Kintsch (1995),
to increase the processing span in a given task. Thus, as indicated by the authors
(and also by Newell, 1991), although the classic limitation, defined by Miller
(1956), of ‘7 + or – 2’ chunks is surely relevant in the case of the execution of sim-
ple memory tasks (such as the memorisation and the recall of letters, digits, or
words), the memory capacity, during the execution of more complex tasks, seems to
have to be far more important (Chase and Ericsson, 1981). This could be the case,
for example, of the practice of different cognitive activities by experts in areas as
various as a chess game (Chase & Simon, 1973; Chi, 1978; Cf. Charness, 1991 for a
review), reading (Farnham-Diggory & Gregg, 1975), or recalling of complex elec-
tronic circuits (Egan & Schwartz, 1979).
Taking into account the text writing as one of these complex cognitive activities,
it could be postulated that the development of writing expertise is linked to the de-
velopment of a structure like Long Term-Working Memory. This structure would al-
low to efficiently store, for example, strategies and necessary knowledge for text
writing. These strategies could be completed, in the precise framework of text writ-
ing, by linguistic and pragmatic knowledge. Such a modality about the development
of expertise would be contextual (strategies to write a text) and fundamentally linked
to the experience. The concept of Long Term-Working Memory would enable ex-
plaining the development of writing capacities with age, as well as the increase of
expertise in adults. It would equally explain the emergence and the systematisation
of writing strategies specific to each writer and linked to the elaboration of her/his
own Long Term-Working Memory register.
However, the Long Term-Working Memory component has still not been suffi-
ciently experimentally validated to account for the development of writing perform-
ances. Its recent evocation by Kellogg (1999) does not concern a developmental as-
pect, but the possible explanation of the facilitating effect of domain expertise on the
Planning process. Experts in a specific knowledge domain would have built a Long
Term-Working Memory concerning this domain, which would allow them, during
the writing of a text concerning this domain, to retrieve knowledge more easily and
without costly cognitive resources. This decrease of the retrieving cost would allow
allocating more cognitive resources to the organisation of the text content (Kellogg,
1987a ; Cf. also Chapter 5).
It nevertheless remains to account for this concept of a developmental viewpoint,
by specifying the role and conditions of installation of such a mnemonic structure
and the nature of storing and retrieving strategies. Similarly, it is necessary to spec-
ify what type of knowledge (domain, pragmatic, linguistic, etc.) is retrieved and/or
stored by this structure throughout the development of the writing activity. Such a
theoretical consideration has been recently developed by McCutchen (2000) in order
to account for the development of writing expertise. More precisely, for the author,
the presence of such a Long Term-Working Memory component in expert writers,
could allow a better and more fluent management of the different pieces of knowl-
edge involved in writing and stored in Long Term memory (like Genre knowledge
and Topic knowledge). In return, the fluency of Text Generation and Transcription
processes would be then enhanced because expert writers, via the Long Term-
Working Memory component, could ‘link developing sentences to extensive knowl-
edge stored in Long Term Memory’ (McCutchen, 2000: 21).
7.5 Conclusion: Working Memory span increase as a possible explanation of writ-

ing expertise
On a developmental perspective, the ‘natural’ increase of memory span (Cf. the re-
view of Dempster, 1981; the developmental works led by Case, Kurland and Gold-
berg, 1982; or the book of Schneider and Pressley, 1989), could constitute one of the
main factors for the development of expertise concerning the possibilities (1) to
maintain more information in memory (Bereiter and Scardamalia, 1987; McDonald,
1980; Yau and Bereiter, 1978), (2) to process it more rapidly (Benton, Kraft, Glover
and Plake, 1984; Chi, 1977), and/or (3) to process this information in a different way
or by using different strategies or operations, according to the intellectual maturation
(Case, 1985; Pascual-Leone, 1970, 1987). Alternatively, the memory capacity in-
crease due to practice or training would lead to (1) the automatisation of some writ-
ing processes, and/or (2) the installation of mnemonic processing strategies, by the
instantiation of Long Term-Working Memory that records these kinds of strategies.
However, defining and explaining the role of memory span variations in writing
expertise are complex because a central question, discussed by McCutchen (1996)
must be asked. For a given expertise level, do the differences of memory span condi-
tion the execution of processes? Or conversely, would less efficient processes, by
overloading Working Memory and consuming all cognitive resources, lead ‘to de-
crease’ the available size of this span? This double interrogation hides a more theo-
retical question concerning the development of writing expertise. Are there several
types of expertise, according to considered cognitive areas, or, conversely, is there
just one unique expertise, with specific aspects according to each of the cognitive
areas? In other words, is the development of expertise linked to the general devel-
opment of global cognitive capacities of the human system, or is it specific to a par-

ticular activity? In the framework of development factors, this debate becomes cen-
tral because it leads to two different viewpoints concerning the relationships be-
tween the evolution of memory capacities and the evolution of writing expertise.
• According to the first point of view, the increase in memory capacities would be
especially linked to some phenomena of maturation, progressive structuring of
memory registers, notably for the phonological register, or rather, more glob-
ally, to the evolution of the general cognitive functioning. In this way, the
memory development would affect all cognitive activities and would concern
all achievable cognitive tasks.
• According to an opposing viewpoint, the increase in memory span would be
relative and linked to an increasing experience in a given area, for a given task.
This would lead to a more efficient and strategic functioning of processes. This
efficiency would be then underlied by the automatisation of some processes,
that would free cognitive resources and/or by the implementation of more stra-
tegic processes, that would limit the cost of controlled cognitive processes.
In the heart of this debate, Swanson and Berninger’s (1996) works, more particularly
focussed on the importance of inter-individual differences, aim to verify whether, in
children, individual differences concerning Working Memory span and written pro-
duction are linked or not to a general system or to specific processes. These authors
want to test if the Working Memory capacity can predict individual variations in
writing. Their study analyses correlation between several measures of Working
Memory and phonological storage and writing and reading measures. According to
Swanson and Berninger, the Working Memory capacity is strongly linked to some
aspects of written language, concerning writing as well as reading. Nevertheless,
there are currently more works about the relationships between Working Memory
and reading or comprehension (Daneman and Carpenter, 1980, 1983; Just and Car-
penter, 1992) than works analysing links between writing and Working Memory. As
writing is a complex activity, it can rapidly lead to an overload in Working Memory,
especially in children. More precisely, the text quality would depend on the writer’s
limited resources. However, the authors underline that many questions relative to the
relationships between Working Memory and written production currently remain
without any answer:
• Is the influence of Working Memory general to any cognitive system or specific
to the writing system (Cf. supra)? To answer this question, it is necessary to ver-
ify if Working Memory measures can predict writing abilities.
• Do Working Memory processing and storage components independently con-
tribute to writing? Are they two distinct systems? In this case, according to the
authors, Working Memory would be a more important factor than Short Term
Memory to explain individual differences, especially for writing high-level
processes. Short Term Memory would be therefore more linked to low-level ac-
tivities (such as graphic transcription), while Working Memory would be linked
to high-level activities like Planning (i.e., Generating and Organising sub-
processes).
• Are the relationships between Working Memory and reading identical to the
relationships between Working Memory and writing? To reply to this question,
it is firstly necessary to clearly differentiate Short Term Memory and Working
Memory. Consequently, in Swanson and Berninger’s (1996) work, Short Term
Memory tasks only involve storing information while Working Memory tasks
necessitate processing, inference, transformation and storing.
To reply to these different interrogations, Swanson and Berninger test three groups
of 100 participants (4th, 5th and 6th graders) in order to take measures of Working
Memory, visuo-spatial working, Short Term Memory, writing, transcribing, vocabu-
lary and reading. Their results first show that Working Memory and Short Term
Memory are independent. Then, measures of Working Memory reflecting the execu-
tive processing are correlated with measures of writing relative to the text generation
and to the comprehension during reading, while the Short Term Memory measures
mainly predict performances during transcription. For them, children’s individual
differences in writing essentially reflect a system with a specific capacity, different
from the reading system (Cf. also former research of Bryant and Bradley (1983), to
which authors do nevertheless not refer, that showed that children manage reading
and writing differently after two or three years of learning). In a similar work,
Brügelmann (1996, 1997) equally underlines the importance of individual differ-
ences concerning reading and writing studies. His objective is then to be able to de-
termine individual characteristics to predict these differences and the heterogeneity
in learning.
8 DISCUSSION POINTS
8.1 Differences between novices and experts exceeding Working Memory capacity
According to most models and works previously presented, a very important relation
in children (but also in adults) would exist between Working Memory span and writ-
ing process efficiency. For McCutchen (1996, 2000), Working Memory limited ca-
pacity would have two main consequences on writing. On the one hand, it would
limit the number of processes that it would be possible to simultaneously carry out;
on the other hand, it could more deeply determine the nature of processes than can
be activated. This would then explain why strategies, as direct knowledge translation
(Knowledge Telling), would persist even in adult writers.
However, if the increase in Working Memory capacities is a central factor, it is
certainly not the unique factor. Besides, we think that the opposition between nov-
ices and experts in text writing can be demonstrated at three other levels, summa-
rised as follows:
• The first level concerns Declarative Knowledge about the text topic. Both pro-
duction facility and text quality would be influenced by the quantity of available
knowledge relative to the production theme. The recovery and the organisation
of ideas would be carried out more easily and more rapidly when the writer
knows the theme better about which s/he must write a text. The writing process
would be conversely more costly when the writer has little knowledge about the
topic to write about (Cf., for example, Caccamise, 1987; Kellogg, 1987a; Chap-
ter 1).
• The second level concerns the necessary linguistic resources. The knowledge
about text organisation would facilitate the production (Cf. Boscolo, 1995;
Chanquoy & Fayol, 1991, 1995). For example, some characteristics associated
with particular text types would more or less facilitate the translation into a text
of retrieved information (Fayol, 1997; Schneuwly, 1988). More specifically, the
nature and the quantity of orthographic knowledge, punctuation, connectives,
syntactic organisers and cohesion marks would influence the final text (de
Weck and Schneuwly, 1994). The nature of lexical knowledge can equally in-
fluence the text progress and quality. The age and/or the level of knowledge can
also play a role in the use of this knowledge. Thus, the degree of complexity of
the writing management would depend both on basic writing mechanisms (such
as spelling, lexical and syntactical knowledge, etc.), on more general knowledge
concerning text types and themes, on the writer’s own characteristics, and on
the task environment.
• The third level concerns available strategies (or procedural knowledge) to man-
age the complexity of the activity. The implementation of adapted strategies fi-
nally appears as essential for an efficient management of the whole writing
process. These strategies would consist of using, in a controlled and finalised
way, already known procedures in order to reach a priori fixed goals and sub-
goals. These goals and sub-goals would correspond to a hierarchised distribu-
tion of the task so as it could be manageable. Scardamalia and Bereiter (1982)
underline the fact that the difficulty of the written production depends on goals
that the writer a priori defines, as a novice or as an expert. Using a Knowledge
Telling Strategy is less costly, while using a Knowledge Transforming Strategy
is more costly, because it necessitates the respect of a very precise set of hierar-
chised goals. Thus, experts would be distinguished from novices by their capac-
ity to deal with, in an optimal and maximal manner, constraints linked to the
task (Ericsson and Lehmann, 1996; Schraagen, 1993; Simon, 1995).
8.2 Is it possible to learn how to become an expert?

According to previously developed models and theories, it appears possible to help
children (or novice writers) to become expert and competent writers. This aspect is
however not very developed in the fundamental research about writing, aiming es-
sentially to analyse and to understand cognitive processes carried out in the progress
of writing. Thus, works analysing or proposing training strategies so as to teach
writers to manage their writing activity better are mainly found in applied research
in the field of pedagogy or educational psychology. It is nevertheless possible to de-
fine what could be ‘the ideal expert writer’. For Boscolo (1995: 354), a strategic
writer has to be both a thinking planner, a coherent organiser, a careful reviser and
an audience-sensitive message sender. To reach such a level of competence, it is ob-
viously necessary to develop writing activities to make young writers sensitive to

these different aspects of writing. These activities would then concern each writing
process and sub-process, from planning to revising. According to Hayes (1996), who
has studied, in a more general way, expertise development in activities such as
painting and music (Hayes, 1985; Cf. equally Whisbow, 1988 for the same type of
study in the area of literary writing), a period of 10 years of intensive practice is
generally necessary before being able to become an expert.
8.2.1 Learning to use writing processes

In the writing area, Graham and Harris (1997) think that the use of ‘self-regulation’
procedures is a necessary condition (but not sufficient) to become an expert writer,
and sufficient to become a competent writer. Moreover, it would be possible to teach
children these types of procedures. Zimmermann and Risemberg (1997) add that a
skilled production depends on the level of writers’ self-regulation because the writ-
ten production is typically an autonomous and intentional activity. While the first
authors think that self-regulation is not sufficient to allow a writer to become an ex-
pert, the second authors, conversely, defend the idea according to which this strat-
egy, if it is taught to beginner writers, allows them to become more rapidly and more
efficiently expert. On a fundamental perspective, Kellogg (1987b) has examined
drafting strategies in writers composing a letter. According to him, elaborating a
writing plan on a draft is a cognitively very costly activity. The results of Kellogg’s
(1987b) experiment show both that the preparation of a draft increases planning du-
ration (and also translating duration) and improves the quality of the production.
Similarly, Kellogg (1990) has shown that the creation of a draft before writing activ-
ity increases the text quality. Thus, learning how to produce drafts and to plan before
writing the final text surely provides a fundamental basis to become an experienced
writer.
8.2.2 Environments to learn how to use writing processes

Currently, the development of computer software, functioning as tutorial or didacti-
cial, is very fashionable in applied psychology. These computer softwares aim to
help very young children while they are beginning to learn how to write, or children
in the course of learning, or even children with learning difficulties, in order to pro-
vide a tool to find a remedy for these difficulties (Cf. Bradfort DeCosta, 1992; Dai-
ute, 1992; McArthur, Schwartz, & Graham, 1991; Mahach, Boehm-Davis, & Holt,
1995; Overbaugh, 1992; Sharples & Evens, 1991; Snyder, 1993a and b, 1994, 1995;
Williams, 1992, etc.). In summary, these works consider that the computer tool, and
more specifically the word processor software, can be considered as an ‘instructional
tool’ (Bangert-Drowns, 1993: 71). Is it, from then on, possible to conceive applica-
tions or interactive models guiding the novice during writing? This passage from
theoretical models to the modelisation of instructional software is an interesting
track that is still not much explored (Cf. although, see Mangenot's review, Man-
genot, 1996). It is however not inconceivable, as seen in the model of knowledge

acquisition ACT-R (Adaptative Control of Thought) built by Anderson (1983a,
1990, 1993), that underlies principles of tutorials in computer programming or in
arithmetic (called then ACT: Advanced Computer Tutor), to conceive such tutorials
for text writing. If this elaboration is far from being improbable, it however previ-
ously requires a very precise determination, on a theoretical plan as well as on an
experimental basis, of all ‘sensitive’ components for the development of the activity
of text writing.
9 CONCLUSION ON WRITING EXPERTISE DEVELOPMENT: THROUGH

A PROCEDURAL MODEL
It seems that a precise theory of writing development still lacks in the general theory
of writing. There are, of course, many studies comparing writing skills within differ-
ent ages or school levels. However, most of the developmental studies about written
production have mainly studied narrative texts and have mainly shown a complexi-
fication of the narrative competence with age. Experiments on other text types are
rarer. Concerning this last report, Boscolo’s (1995) advice could be followed by re-
searchers who are interested in writing development. According to this author, the
development of writing must be investigated according to three main dimensions
(Cf. p. 353):
• Research concerning learning how to write is important, it would thus be neces-
sary to be able to fulfil the gap that exists between works on early literacy and
writing.
• It would be important to analyse interactions between developing competencies
in children and the environment in which they learn how to write.
• Finally, it would be necessary to carry out experiments on perceptions and rep-
resentations that children have about writing, so as to be able to relate these per-
ceptions and representations with young writers’ competencies.
These three points are fundamental and can guide the implementation of new ex-
periments in order to describe and to explain the development of writing expertise.
On a theoretical level and concerning modelisation, it seems however necessary to
integrate current writing models in a larger architecture, being able to show the ex-
pertise development and the processing constraints of the cognitive system. It would
be thus important and necessary to situate a writing model in a more general model
of cognition, in taking into account procedural and declarative knowledge (Model
ACT-R: Anderson, 1993), and in elaborating a more powerful procedural model of
writing (Fayol, 1991, 1994). However, it still seems very difficult to elaborate such a
model, because all the constraints and all the developmental factors are still not well
surrounded. Indeed, the unique explanatory factor of the writing expertise develop-
ment currently concerns the Working Memory span, that this span is conceived in
maturation or experience (practice) terms. This explanation is obviously too restrain-
ing and restrictive, but current writing models do not, unfortunately, propose other
interpretations.
GENERAL CONCLUSION
COMMENTARY ON PART II:

PROCESSING MODALITIES AND DEVELOPMENT OF
EXPERTISE IN WRITING
RONALD T. KELLOGG
University of Missouri - USA
The chapters in Part II address interwoven questions regarding the control of writing
processes, their use of Working Memory resources and changes in writing processes
with growing expertise. All these questions about writing have been raised in the
broader context of Working Memory theories and expertise and my objective here is
explicitly link the two endeavors. There are important insights to be gained in our
understanding of writing by looking at the points of consensus and contrast among
Working Memory theories (Miyake and Shah, 1999). At the same time, there are
also useful additions to be made to the broader debates from what we now know
about writing, a highly complex example of cognitive control and expertise.
1 THE BROAD CONTEXT

Miyake and Shah (1999) posed eight questions to Working Memory theorists in an
effort to compare and contrast the diverse models proposed in the literature. As they
noted, the term Working Memory is used in different ways by different researchers
across behavioral neuroscience, cognitive neuroscience and cognitive psychology.
Even within cognitive psychology, the relation of Short Term Memory to Working
Memory, the relation of Working Memory to Long Term Memory and the unitary
versus domain-specific nature of Working Memory are unclear. The eight questions
they posed are as follows, together with their relevance to writing:
• ‘Basic Mechanisms and Representations’. How is information encoded into,
maintained and retrieved from Working Memory and what is the format of the
representations? As described in the chapters here, Planning, Translating and
Reviewing each make specific demands on Working Memory storage and re-

222 KELLOGG
trieval of a variety of representations, ranging from vague feelings to proposi-

tions to concrete images.
• ‘The Control and Regulation of Working Memory’. Is the information in Work-
ing Memory controlled and regulated by a central executive or by some non-
central means? When a writer is planning, what determines which idea will be
maintained in Working Memory for further development and which will be ig-
nored? What determines when one or more propositions have been planned suf-
ficiently and are ready for translation into sentences? What determines that an
episode of Reading and Editing should give way to further Planning or Translat-
ing to correct errors or revise the ideational content or organization of a text?
• ‘The Unitary Versus Non-Unitary Nature of Working Memory’. Is Working
Memory best conceived as a single system or a multicomponent system? Writ-
ing models have adopted both the unitary perspective (McCutchen, 1996) and
the multicomponent system (Kellogg, 1996), but it is not yet clear the empirical
evidence on writing supports one over the other approach.
• ‘The Nature of Working Memory Limitations’. Is the capacity of Working
Memory constrained by activation limits, processing speed, inhibition, skills or
some other factor? In writing research, the activation viewpoint has received the
most attention (van den Bergh and Rijlaarsdam, 1999), although a role for inhi-
bition has also been suggested (Galbraith, 1999).
• ‘The Role of Working Memory in Complex Cognitive Activities’. What role
does Working Memory play in language comprehension, reasoning, problem-
solving and other complex cognitive acts? Plainly, the writing task invokes not
only language production and comprehension, but also thinking in all its forms
(Kellogg, 1994).
• ‘The Relationship of Working Memory to Long Term Memory and Knowl-
edge’. Are Working Memory and Long Term Memory structurally distinct or
the former the activated subset of the latter? Besides being a task of language
and thought, writing is inherently a task of memory in all its forms (Kellogg,
1994). Writing draws on the procedural system of Long Term Memory (e.g.,
handwriting skill), semantic memory (e.g., conceptual content to be expressed
linguistically) and episodic memory (e.g., events recalled or imagined that com-
prise a narrative or support an argument). The writer must retrieve information
from Long Term Memory into Working Memory and then creatively apply that
knowledge to solve the rhetorical and content problems posed by the task at
hand. Both novices and experts alike encounter the difficulties of juggling mul-
tiple processes and representations within the limits of short-term Working
Memory. Experts may gain access to Long Term Memory representations
through retrieval structures for knowledge within their domain of expertise
(Kellogg, 1998).
• ‘The Relationship of Working Memory to Attention and Consciousness’. Is at-
tention the same construct as Working Memory or perhaps one aspect of it and
are we necessarily conscious of representations in Working Memory? Writing
studies have depended heavily on verbal protocols, directed forms of introspec-
tion and in some cases the use of secondary task reaction time (RT) to measure
COMMENTARY 223
attentional engagement in the primary writing task. The models of writing have
been built on the assumption that the writer has conscious access to at least
some of the representations and processes that are involved in the task. The va-
lidity of introspection is always a key question in cognitive psychology (Erics-
son and Simon, 1978) and writing research is no exception (Janssen, van Waes
and van den Bergh, 1996). Borrowing from the attention literature (Kahneman,
1973), a high degree of interference in secondary task RT relative to a baseline
control indicates that Working Memory capacity is substantially engaged in one
or more writing processes, leaving little spare capacity for the RT task (Kellogg,
1994).
• ‘The Biological Implementation of Working Memory’. How do the recent find-
ings of cognitive neuroscience, including tests of brain-damaged patients and
neuroimaging studies, inform models of Working Memory? Although there is
an extensive literature on acquired dysgraphia and other forms of aphasia, writ-
ing models have not been substantially informed by such work. Researchers in
the writing community have focused more on the planning and reviewing of
discourse and less on the linguistic task of generating well-formed sentences
and linking these into coherent text structures. Sentence generation (Bock and
Levelt, 1994) and the processing of interclausal relationships (Costermans and
Fayol, 1997) are no less important and will increasingly be illuminated by the
findings of cognitive neuroscience. Thus, work on planning and monitoring
problems experienced by individuals with frontal lobe lesions (Shallice, 1988)
is clearly relevant to our understanding of writing control and regulation, as are
the case studies of acquired dysfunctions of the phonological and orthographic
representations involved in written and oral spelling (Caramazza, 1991). Simi-
larly, the recent neuroimaging studies of Working Memory ought to help nar-
row the range of plausible models of complex tasks such as writing (e.g., Jon-
ides and Smith, 1997).
10 IMPLICATIONS OF WRITING RESEARCH

As brought out in the chapters of Part II, we are beginning to see interesting, albeit
tentative conclusions about how writers control what they do, how they manage the
limits of Working Memory and how expertise in a domain alters text production.
Because writing brings together memory, language and thinking, it may well be an
ideal tool for making basic discoveries about complex cognition in general, particu-
larly in other production tasks such as choreography, musical composition, painting,
sculpture, graphic design and software design (Kellogg, 1994). Further, the work on
writing research needs to be linked directly to the broader concerns of cognitive
psychologists, such as the eight points outlined in the first section. In this next sec-
tion, my focus will be on three observations about writing discussed by Alamargot
and Chanquoy that indeed appear to have important general implications. The first is
that writing appears much more amenable to interactionist models than to modular
models of cognitive processing. The second is that writing processes appear to de-
mand multiple components of Working Memory rather than a unitary resource. The
224 KELLOGG
third is that the control of writing processes and regulation of Working Memory re-
sources does not appear to be entirely under central control. Instead, writers seem to
adapt to the demands of the task at the moment and to the limits of Working Mem-
ory and available knowledge.
10.1 Interactionist versus Modular Models

In chapter 5, the contrast between interactionist and modular cognitive models was
drawn.
Alamargot and Chanquoy suggest that Levelt’s (1989) speech production model
is essentially modular in that the Formulator, Articulator and other components act-
ing effortlessly, rapidly, accurately and independently of other components. Al-
though this may fit well with the on-line nature of speech production, written text
production appears to be entirely different. Instead of rapid production with little ed-
iting as in speech, writing involves slow and gradual production with many opportu-
nities for reading and editing the evolving text. In fact, McCutchen (1988) has ar-
gued that it is precisely the extensive interaction among the Planning, Translating
and Reviewing processes seen in skilled writers that rules out a modular account of
writing.
In planning a text, an experienced writer sets rhetorical goals that are shaped by
the intended audience. Such high-level planning has consequences for the selection
of ideas to be included and even the choice of words in a given sentence. It is not the
case that the writer plans the ideas of the text and then automatically translates them
into sentences. Similarly, when experienced writers review already generated text,
the processes that monitor how the audience will react can directly intervene in the
translation process, prompting a change in word choice. Although the motor execu-
tion of handwriting or typing may become automatized with practice (Fayol, 1999),
it is not apparent how the formulation and monitoring processes can similarly be
automated in skilled writing, according to McCutchen. Of interest, it is mature,
skilled writers who are more likely to engage in extensive high level processes in the
formulation and monitoring of texts compared with novices (Bereiter and Scard-
malia, 1987).
A second compelling argument against modularity is that writing processes are
largely controlled rather than automatic. Planning and Reviewing in particular re-
quire large investments of Working Memory resources. This can be seen in work us-
ing secondary task reaction time (RT) during text production. An auditory probe oc-
curs at random times during composition and RT to detect each occurrence is meas-
ured and compared against a baseline control RT. The more interference in probe
RT, the more momentary cognitive effort the writer is devoting to the primary task
(Kahneman, 1973). After probe detection the writer presses a key to indicate
whether Planning, Translating, or Reviewing was the focus of attention at the time
of the interruption. The highest levels of RT interference were associated with Plan-
ning and Reviewing, but even translating was highly effortful, especially relative to
learning and reading tasks (Kellogg, 1994). Other researchers have confirmed that
writing processes are generally effortful, although the absolute levels of interference
COMMENTARY 225
observed depends on the frequency with which the secondary probes are presented
(Piolat, Roussey, Olive and Farioli, 1996).
Translating an idea into a sentence is sometimes regarded as automatic, but it,
too, can require Working Memory resources. Fayol, Largy and Lemaire (1994)
found that writers make subject-verb agreement errors in when transcribing orally
presented sentences in French, if at the same time they retained five words in Work-
ing Memory. Specifically, the writers incorrectly inflected the verb to a plural form
if an object of a preposition closest to the verb was plural and the more remote sub-
ject noun was singular. Thus, when verbal and plausibly executive resources are en-
gaged in maintaining five words in Working Memory, grammatical encoding at
times fails with respect to subject-verb agreement.
The motor programming involved in handwriting, typing or dictating also de-
mands response scheduling and other executive functions, although these demands
are minimal when the skills are well-practiced. For young children, however, these
demands are formidable; just the transcription of words in handwriting is laborious
(McCutchen, 1996; Bourdin and Fayol, 1994). In a large scale correlational study
with 600 students in the primary and intermediate grades, Graham, Beringer, Ab-
bott, Abbott and Whitaker (1997) found that handwriting ability positively affects
fluency and quality. In their words:
‘Beginning and developing writers who have to devote more attentional resources to the
mechanical skills of rapid, automatic letter production will have less attentional re-
sources to devote to the planning, translating and reviewing and revising processes of
composition’ (p. 11).
It seems unlikely, then, that a modular model can account for the control of writ-
ing processes given the interaction among Planning, Translating and Reviewing and
their relatively heavy demands on Working Memory. Alamargot and Chanquoy are
correct to note that it may be useful to think of Long Term Memory as having a
modular structure, however. The knowledge representations retrieved and used dur-
ing the course of Planning, Translating and Reviewing may be domain specific.
Schumacher and Ma (1999) have explored some of the consequences of this point of
view.
But from a processing standpoint, writing plainly illustrates the importance of in-
teraction among diverse processes. Modular approaches have been successful in un-
derstanding the processes involved in face recognition and speech comprehension
(Fodor, 1983). One possibility is that writing and other production tasks are not
amenable to modular principles in the same way that perception appears to be.
10.2 Multicomponent Working Memory

It is reasonable to suppose that the widely diverse processes of writing compete for a
common resource of Working Memory. The central executive of Baddeley’s (1986)
model provides a general purpose resource. As Jonides and Smith (1996) suggested,
the central executive might better be viewed as an array of executive functions, such
as focusing attention, inhibiting alternatives and selecting responses. Evidence sup-
226 KELLOGG
porting the idea of shared central resource comes from studies showing trade-offs
between different processes.
For example, Bourdin and Fayol (1996) demonstrated trade-offs between idea re-
trieval and sentence generation, on the one hand and motor transcription, on the
other. Oral output is easier for young children than handwriting, but not for adults.
In one study, they compared a speaking span task in which subjects read a series of
words and then orally generated a sentence for each word with a writing span ver-
sion of the same task. They measured the number of sentences that were success-
fully generated from the words the subjects could remember. The retrieval and gen-
eration output should be greatest when attention is not diverted to motor execution.
They expected such a diversion when children must use written output. They ob-
tained the predicted mode by age interaction in which oral output only exceeds writ-
ten output for young children.
Another result showed that when demands on one process are reduced, the freed
Working Memory capacity is then shared by other processes (Kellogg, in press).
Narrative, descriptive and persuasive texts were written by college students in long-
hand or on a word processor. The writers concurrently detected auditory probes cu-
ing them to retrospect about whether they were planning ideas, translating ideas into
sentences, or reviewing ideas or text at the moment the probes occurred. Narrative
planning and longhand motor execution were presumably heavily practiced, freeing
capacity for rapid probe detection. Of interest, spare capacity was distributed equally
among all three processes, judging from probe reaction times, when planning de-
mands were low in the narrative condition. When motor execution demands were
low in the longhand condition, however, reviewing benefitted more than planning.
The results indicate that Planning, Translating and Reviewing processes in writing
compete for a common, general purpose resource of Working Memory.
Besides the attentional functions of the central executive, writing may also re-
quire components of Working Memory are dedicated to the storage and processing
of code-specific representations. Both Hayes (1996) and Kellogg (1996) theorized
that verbal, visual and spatial components of Working Memory serve specific func-
tions. And, some evidence is beginning to support this position, as noted by Ala-
margot and Chanquoy in Chapter 6. For example, in planning it is known that adults
report using imagery more often in writing definitions of concrete words compared
with abstract words and these reports are accompanied by faster sentence initiation
and better definitions (Sadowski, Kealy, Goetz and Paivio, 1997). One interpretation
of these data is that only the concrete words gain access to visual and spatial Work-
ing Memory, permitting a faster and better response.
In an interesting test of the multicomponent view of Working Memory in writ-
ing, Lea and Levy (1999) designed a visuospatial secondary task that was about
equally difficult when performed in isolation as a phonological task. The former in-
volved updating the location of an imaginary object in an x, y coordinate system,
where the movement of the object was altered by presenting arrows on a computer
screen that pointed up, down, left, or right. They were to respond whenever the ob-
ject moved away from either axis and then returned. The phonological task involved
maintaining the category of a verbal stimulus (a letter or a digit) and responding
whenever the category remained the same from one presentation to the next. When
COMMENTARY 227
combined with a primary writing task, Lea and Levy found that the phonological
task caused a reduction in writing fluency of 3.4 words per minute compared with
only 1.9 for the visuospatial task. Furthermore, performance on the secondary task
itself declined 60% for the phonological task compared with 44% for the visuospa-
tial task. Thus, the results indicate that phonological or verbal Working Memory is
more critical for writing than visual or spatial resources, although the data also sug-
gest that the latter are not unimportant.
Translating ideas into sentences involves lexical access and phonological proc-
essing (Bock and Levelt, 1994). The phonological representations of sentence con-
stituents are often briefly stored in verbal Working Memory, giving rise to the sub-
jective experience of inner speech. The writer may maintain these representations in
verbal Working Memory for the purpose of covert editing prior to their execution.
Neuropsychological evidence indicates that patients who have lost the phonological
storage associated with inner speech make phonemic and other linguistic mistakes in
speech (Baddeley and Wilson, 1985) and are typically unable to produce complex
written expressions (Friederici, Shoenle and Goodglass, 1981). Further, Levy and
Marek (1999) used irrelevant speech to occupy verbal Working Memory as writers
viewed five words that could be rearranged and combined with function words gen-
erated as needed to form a sentence. They found that the irrelevant speech decreased
the percentage of target words remembered and used in their exact forms in sentence
generation, but it had no impact on a control task of transcription typing.
10.3 Processing Control

Writing processes appear to be highly interactive and depend in part on central ex-
ecutive resources of Working Memory. However, a separate issue is whether a cen-
tral executive must decide when a particular process should be engaged and the de-
gree of cognitive effort to allocate to it.
As Alamargot and Chanquoy discuss in Chapter 5, the text produced thus far and
the degree to which Working Memory is currently overloaded may trigger the acti-
vation of specific processes. The allocation decision need not involve central execu-
tive functions if the task environment can control the sequence of processes. Here
we will examine the possibility that control is partly determined by the central ex-
ecutive and partly by the task. Specifically, the strategy followed by a writer in at-
tending first to one process and then to another is postulated to be centrally deter-
mined. However, the degree of effort or the fraction of available Working Memory
resources allocated at a given moment is assumed to be dependent on the demands
of the task and is not under volitional control.
The notion that momentary cognitive effort is controlled by the task rather than the
individual can be traced back to Kahneman (1973). He illustrated this point by con-
trasting the task of mentally multiplying 83 by 27 with the task of remembering four
digits, 7, 2, 5 and 9. One cannot decide to allocate as much cognitive effort to retain-
ing four digits as is dictated by the mental multiplication task.
In the case of writing, several studies have now been conducted using the triple
task technique to assess cognitive effort and processing time (Kellogg, 1994; Piolat,
228 KELLOGG
Roussey, Olive and Farioli, 1996; Piolat and Olive, in press). As noted earlier, the
slower the RT to the probe (relative to a baseline control), the more cognitive effort
was momentarily allocated to the reported process. The time given to each process
can be estimated by calculating the percentage of responses for each of the three
processes as a function of the first, second, or third phase of total composition time.
An early study using a variation of this method indicated that the strategy of outlin-
ing affects only processing time and not cognitive effort (Kellogg, 1988).
In the first experiment, participants retrospected when interrupted regarding
whether they were planning, translating, or reviewing (RT was not collected). Half
of the participants immediately began composition, whereas the other half spent five
minutes preparing an outline. Furthermore, in each group, half of the participants
were asked to compose a rough draft and the and the other half a polished draft. The
results showed that the prewriting and drafting strategies controlled the way time
was allocated during composition. Without outlining in advance, time was spent
about equally on Planning, Translating and Reviewing, although initially the focus
was on planning and later in the composition, reviewing dominated. But for those
who already outlined, they were able to focus Working Memory resources on trans-
lating for over 50% of the time during all phases of composition. The polished draft
strategy similarly caused writers to devote more time to Reviewing throughout all
phases compared with the rough draft strategy.
In Experiment 2, the triple task technique was used in three groups of writers
(no-outline vs. written outline vs. mental outline), allowing cognitive effort as well
as processing time to be assessed. The results showed that the outlining task, regard-
less of whether it was mental or written, again influenced the time allocations as ob-
served in the first experiment. However, the pre-writing task had no effect on the
writers' allocations of momentary cognitive effort. These findings suggested that
writing strategies that are presumably invoked by central executive processes have
control of processing time but not momentary effort.
Further studies have shown that differences between tasks and individual differ-
ences among writers in domain-specific knowledge and Working Memory capacity
influence the degree of momentary cognitive effort allocated, but do not necessarily
the distribution of time devoted to writing processes. For example, narrative and
persuasive texts impose different task demands because of the degree of a priori
knowledge about each kind of text. Narrative schemas are acquired at an early age
and are heavily practiced throughout childhood in listening to and telling stories
(Mancuso, 1986). In contrast, schemas for argumentation and persuasion are weakly
developed in all but the highest ability college students (Britton, Burgess, Martin,
McLeod and Rosen, 1975). It might be expected then that the narrative task would
demand less effort for planning and possibly other processes relative to the persua-
sive task.
Kellogg (in press) observed that the RT interference score associated with plan-
ning was indeed reliably lower when writing a narrative (M = 232 ms) compared
with a persuasive text (M = 363 ms). The Working Memory resources freed by nar-
rative composition was not devoted exclusively to Planning, however. Instead, the
cognitive effort devoted to Translating and Reviewing also was reliably lowered in
narrative relative to persuasive composition. Thus, the degree of total Working
COMMENTARY 229
Memory capacity momentarily allocated to writing processes varied markedly de-

pending on the kind of text written. In contrast, the distributions of time allocated to
the Planning, Translating and Reviewing varied little with the type of task. Although
the interaction of process type and text type was reliable, reviewing time was essen-
tially the same for narrative (M = 21%) and persuasive texts (M = 20%). Moreover
the differences that were observed for planning time (M = 23% versus M = 26%, for
narrative and persuasive texts respectively) and translating time (M = 48% versus
50%) were slight. Thus, it appears that the the task itself dictates how much momen-
tary effort must be given, but that the time allocated to processes varies little with
the task. It may be that the time allocation is a strategic decision made by the writer
and is not controlled by the task environment.
The effect of topic knowledge on the allocation of time and cognitive effort in
writing was first investigated by Kellogg (1987a). In his experiments, participants
were divided into two groups of low- and high-knowledge according to their results
on a test of knowledge about the topic of the assigned text. Results showed that
high-knowledge writers engaged less cognitive effort in the writing than low-
knowledge writers did. On the other hand, the allocation of processing time to the
writing processes was not affected by the writers' level of knowledge. In a recent ex-
tension of this line of research (Kellogg, 1998), participants wrote narratives about a
half-inning of baseball and were divided in high and low knowledge conditions us-
ing Voss, Vesonder and Spillich’s (1980) procedure. The RT interference scores
were reliably lower for individuals with a high degree of domain-specific knowledge
compared with low-knowledge writers. This result held true for writers with a low
degree of linguistic skill or verbal ability as well as those with a high degree. Of in-
terest, the time devoted to planning, translating and reviewing again was the same
for high and low domain-knowledge writers.
So, we have now a case in which a strategic decision to outline affects process-
ing time but not cognitive effort whereas task differences and domain knowledge
differences affect only cognitive effort. There is no reason why both strategic and
task or knowledge driven control could not both operate. One possible example of
this can be found in work by Piolat, Roussey and Rous (1996). In oral as in written
language, different parameters of the communicative context ought to affect the in-
vestment of cognitive effort. One of these parameters is the social status of the audi-
ence. Piolat, Roussey and Rous (1996) manipulated this by asking participants to
write a letter either to a friend or to the University's President about an increase of
university fees. Their results revealed that writers allocated more cognitive effort
when the audience held high social status rather than low. Presumably the task was
more difficult and dictated a higher investment of effort. However, the writers also
allocated more time to reviewing their text when the social status of the addressee
was high. According to the view explored here, the extra time allocated to reviewing
reflected a central, strategic decision on the part of the writer. In contrast, the in-
creased cognitive effort allocated is likely a task driven difference in Working
Memory allocations.
230 KELLOGG
11 CONCLUSION
The eight issues brought to the fore by Miyake and Shah (1999) are fundamental to
understanding cognitive control, Working Memory limitations and expertise in hu-
man cognition. Writing is an excellent task to investigate these matters. To date,
only some of the eight issues have received much investigation in the writing area
and the others deserve greater attention from researchers. Conversely, to date the re-
sults of writing research have not informed the broader debate about Working Mem-
ory to the extent that they should. Three specific points were raised here. First, writ-
ing and presumably other related acts of creative thinking found in choreography,
musical composition, painting, sculpture, graphic design and software design are
more amenable to interactionist modeling than modular. Second, it appears neces-
sary to postulate a multicomponent model of Working Memory to account for such
complex tasks. Third, it may be that volitional, strategic control mediates how time
is spent on a task but not the fraction of the available Working Memory resources al-
located.
COMMENTARY ON THE BOOK: THROUGH THE
MODELS OF WRITING
JOHN R. HAYES
University of Carnegie Mellon – USA
The authors of this volume, Alamargot and Chanquoy, have performed a very valu-
able service to the scientific community. They have searched out a very large num-
ber of theoretical contributions relevant to writing from a variety of fields including
cognitive psychology, composition research, speech and linguistics. They have
shown where these theoretical contributions complement each other and where they
differ. In addition, they have identified topics that require theoretical elaboration on
which the available theories are either silent or vague. This distillation of the re-
search literature will help to focus the efforts of researchers already in the field and
should be an invaluable aid to those just entering the field.
As is inevitable in such a comprehensive effort, the reader may not agree with all
of the judgments that the authors make or may feel that some research has been in-
appropriately omitted. Indeed, I disagree with the authors on a few points and feel
that some studies have been omitted that do have important implications for our
field. I mention this not to criticize the authors, who I feel have done a superlative
job, but rather to illustrate the value of this text for promoting discussion.
In this chapter, I will first review several studies that I feel should be included in
the discussion because they provide interesting perspectives on the issues treated in
this volume. These include studies on children’s rhetorical abilities, on revision and
on aspects of expertise in writing. Finally, I will discuss some general issues that
may be important for the continuing development of our field.
1 CHILDREN’S RHETORICAL ABILITIES

It is widely observed that young children are more successful in adapting their lan-
guage to their audience in conversation than in writing. It is not that young children
lack rhetorical skills. Indeed, the authors note that there is substantial evidence that
young children have rhetorical skills that they use in communicating with audiences.
Rather, the issue is that children seem to have specific difficulty using their rhetori-
cal skills in writing. Moffett (1968), Britton, (1970) and Bereiter and Scardamalia
(1987) have hypothesized that the critical difficulty for children is that during con-

232 HAYES
versation the audience is present but during writing it is not. Bereiter and Scar-
damalia (1987) put it this way:
‘Generating content is seldom a problem in oral discourse because of the numerous
kinds of support provided by conversational partners. Without this conversational sup-
port, children encounter problems in thinking of what to say, in staying on topic, in pro-
ducing an intelligible whole and in making choices appropriate to an audience not im-
mediately present’(p. 7).
The reason that an absent audience poses a problem specifically for children, ac-
cording to Bereiter and Scardamalia (1987), is that adapting language to the needs of
an absent audience requires more cognitive capacity than children possess. They
suggest that children are limited to knowledge-telling in their writing because they
lack the cognitive capacity to ‘hold constraints in mind while searching for contents
to meet them’ (p. 343).
A study by Littleton (1995) tested the notion that young children are unable to
adapt their language to an absent audience. The study involved ten children between
the ages of 5 and 9 in a daycare facility. In the study, the children learned ‘magic
tricks’ that they then taught to peers. Each child taught tricks both to present peer
and, by tape recorder, to absent peers. Littleton found that the children in her study
did adapt their language to the absent audience in a number of ways. In the audience
absent condition, the children were significantly more likely to articulate the steps
involved in the trick (e.g., ‘Rub the balloon on your head’), to make helpful remarks
about timing (e.g., ‘Before you tape it, you have to…’), to list the materials needed
for the trick, to make descriptive references to locations (‘the cards inside the enve-
lope’), features (‘the blue cups’) and qualities (‘shake it really fast and
hard’)involved in the tricks and to make disambiguating speech repairs. Overall it
appeared that the children in Littleton’s study did adapt their language to the absent
audience in ways that took into account the fact that the absent audience could not
see the materials or the actions that were being described. Littleton’s results require
us to reject the idea that young children don’t have the cognitive capacity to adjust
their language to an absent audience. In speaking to an absent audience, they clearly
do have that capacity. We will need to explore further to find the reasons for chil-
dren’s failure to adapt their language to their audiences in writing.
12 METACOGNITION IN REVISION
In their chapter on revision, Alamargot and Chanquoy assert that metacognitive fac-
tors in revision have largely been ignored except in the theorizing of Butterfield et
al. (1996) . However, they assert that there is currently little research concerning
these factors. In making this assertion, I believe that the authors have missed the
work of Wallace and his collaborators (Wallace and Hayes, 1991; Wallace, Hayes,
Hatch, Miller, Moser and Silk, 1996). The Wallace studies directly address the ques-
tion of why college students often fail to revise globally. Wallace and Hayes (1991)
outlined three possible reasons for this failure. First, students may lack essential re-
vision skills. For example, they may fail to notice the global problems or when they
do notice them they may not have the skills to fix them. Second, students may have
COMMENTARY 233
all the low-level skills needed for global revision (that is, noticing and fixing skills),
but may not have the executive procedures needed to coordinate those skills. This
suggestion is parallel to a position proposed by Scardamalia and Bereiter (1983). A
third possible source of difficulty is that the students may have an inappropriate task
definition for revision. By task definition, Wallace and Hayes (1991) meant ‘the
writer’s understanding of what he or she is supposed to do’ when faced with the task
of revision. The writer’s task definition (a cognitive structure that Hayes (1996)
called a task schema) is a metacognitive structure that controls the execution of the
task. Wallace and Hayes (1991) reasoned that if the cause of students’ failure to re-
vise was largely either a lack of basic skills or the lack of executive procedures, then
simply telling the students that they ought to revise globally would have little effect.
However, if the failure to revise globally was, at least in part, the result of an inap-
propriate task definition, then relatively brief task instructions might have a substan-
tial impact.
To test this hypothesis, Wallace and Hayes devised eight minutes worth of in-
struction in which they outlined for students the major differences between local and
global revision. The experimental group was given this instruction and was asked to
globally revise a text they were provided. The control group did not receive this in-
struction and was simply asked to make the provided text better. The revisions were
analyzed for evidence of global revision and independently, for overall text quality.
The results were that writers in the experimental group did more global revision and
produced revisions of higher quality than did those in the control group. Because of
the evidence that somewhat different processes may be involved when writers revise
other people’s text than when they revise their own texts, Wallace and his collabora-
tors (Wallace et al., 1996) decided to study the same question in writers who were
revising their own texts. They designed a study parallel to that of Wallace and Hayes
(1991) and found similar results. That is, writers given instructions to ‘revise glob-
ally’ did more global revision and improved the quality of their text more than did
writers who had not been given these instructions.
These studies by Wallace and his colleagues indicate clearly that metacognitive
factors play an important role in the revision practices of competent college writers.
The writer’s task definition influences the revision activities they carry out. Further
and important for writing pedagogy, the studies also show that that college writers’
task definitions can be readily changed by instruction even if that instruction lasts
only eight minutes. These results suggest significant potential for metacognitive
training in writing.
13 ASPECTS OF EXPERTISE
Alamargot and Chanquoy have appropriately focussed their discussion of writing
expertise on developmental issues. After all, the most dramatic changes in writing
expertise occur during childhood and adolescence. Below, I review several studies
of various aspects of adult writing expertise that can add to the picture of expertise
that the authors provide.
234 HAYES
13.1 Perceiving that a Reader will have Difficulty Understanding a Text

Writers who have a lot of knowledge about a topic often have difficulty explaining
the topic to less informed readers. User manuals written by engineers are frequently
sighted as examples of this phenomenon which I have called the ‘knowledge effect.’
Hayes, Schriver, Spilka and Blaustein (1986) explored the knowledge effect experi-
mentally in two studies. In both studies, adults were asked to read a set of four one-
page texts. The four texts included:
• a naturally occurring unclear text,
• the same text but rewritten to make it clear,
• a naturally occurring clear text and,
• the same text rewritten to make it unclear.
The clear text was rewritten to be unclear by introducing ambiguous language and
technical vocabulary. The unclear text was rewritten to be clear by removing ambi-
guities and technical language.
Two studies using different participants were carried out with these texts. In the
first study, participants were asked to explain the text line by line. This allowed the
researchers to determine what was clear to readers in each text and what was not. In
the second study, readers were asked to underline those parts of the text that they be-
lieved another reader who was unfamiliar with the topic would have difficulty un-
derstanding. The only variable in these studies was the order in which the partici-
pants read the texts. As expected, the first study showed that readers understood
some parts of the unclear texts better if they had read the clear texts first. In the sec-
ond study, participants who read the clear versions of the texts before they read the
unclear versions were less accurate in identifying what would confuse readers of the
unclear texts than were participants who had not read the clear versions first. Fur-
ther, it was just those aspects of the unclear texts that were better understood if the
clear text was presented first that were predicted less accurately by those who saw
the clear texts first. Thus, it appears that the participant’s knowledge of the subject
matter influenced their ability to judge whether or not a text was clear, even though
they had acquired that knowledge less than five minutes before making their judg-
ments.
This study implies that one aspect of expertise, topic knowledge, can actually
create a difficulty for the writer. It also suggests that another aspect of expertise in
the writer is the ability to understand what will and what will not be clear to readers
who do not share the writer’s topic knowledge.
13.2 Helping Writers Understand their Readers’ Needs

Schriver (1987) carried out a novel training study designed to help writers under-
stand and predict their readers’ needs. Schriver had noticed that writers who used
think-aloud protocols to evaluate their documents reported that the experience im-
proved their ability to write clearly. To test this observation, she constructed a set of
ten lessons designed to engage the writer with think-aloud reader protocols. Each
lesson involved a problematic text and a protocol of a reader trying to understand the
COMMENTARY 235
text. The problematic texts were first drafts of instructions for operating a word
processing system intended for a lay audience. The texts did not contain spelling or
grammatical errors, but rather had poor definitions, unclear procedures, missing ex-
amples, ambiguities and other ‘above the word- or phrase-level’ problems. The
reader protocols were collected from members of the intended audience and re-
vealed a variety of comprehension problems. The lessons were carried out as fol-
lows: First, the participants read an problematic text and predicted where in the text
the readers would experience difficulty. Then, the participants read a protocol of a
real reader trying to understand the text. Finally, the participants revised their
predictions.
To evaluate the effectiveness of the lessons, Schriver used a pre-test, post-test,
control group design. The control group consisted of students in technical writing
courses who were trained in traditional audience analysis and peer response meth-
ods. The pre- and post-tests consisted of popular science texts for which reader prob-
lems had been identified by examining reader protocols.
The result of the study were quite dramatic. The treatment group detected 62%
more of the problems on post-test than pre-test. In contrast, the control group de-
tected slightly fewer problems on post-test than pre-test. These results suggest that
writers can increase their sensitivity to the needs of their readers through exposure to
reader feedback.
13.3 The Role of Persona in the Perception of Text

In addition to conveying topic information, a text can create the impression that the
writer is modest, arrogant, biased, or stupid. A text may be perfectly clear but may
not be persuasive if it causes readers to form negative impressions of the writer.
Thus, an important aspect of skill in writing is the writer’s ability to control the im-
pression of themselves that the text projects. I will briefly review a series of studies
that explore this skill.
Hatch, Hill and Hayes (1993) studied students’ letters of application to college
and established that these texts do convey consistent impressions of the writer’s per-
sonality. That is, they found that independent judges agreed well beyond chance in
attributing personality traits to the writers of these texts. In a second study, they
found that these impressions of the writer’s personality could influence the decisions
of college admissions personnel. Positive admissions decisions were significantly
correlated with independently measured positive impressions of the writer’s person-
ality inferred from texts. By analyzing the texts, Hatch et al. (1993) were able to
identify several rhetorical actions, such as ‘acknowledging the contributions of oth-
ers’ and using pretentious language, which appeared to influence persona judgments.
In their third study, the authors rewrote the texts to change these rhetorical actions
while making minimal changes in content. For example, they replaced pretentious
language with more commonplace language or replaced ‘we’ references with ‘I’ ref-
erences. Then they obtained personality judgments of the authors based on both the
original and the altered texts. The results indicated clearly that by rewriting these
236 HAYES
features of the texts, readers’ judgments of the writer’s personality could be changed
in highly predictable ways.
The texts that Hatch et al. (1993) studied were written by teen-age writers to an
adult audience. Schriver, Hayes and Steffy (1996) (described in Schriver, 1997)
studied texts written by adults for teen-agers, namely drug-use prevention brochures.
The authors found that teen-agers could understand the messages contained in the
brochures quite well. However, they often rejected the messages of the brochures
because they formed rather unfavorable impressions of the writers. Some readers felt
that the writers were talking down to them. Others viewed the writer’s as old-
fashioned or out of touch with the reality of the reader’s life. Sometimes the writers
attempted to use slang, humor, or cartoons to make the texts appealing. These efforts
backfired if the readers regarded the slang, the humor, or the style of the cartoons as
not current. As in the case of the Hatch et al. (1993) study, the readers’ impressions
of the writers’ personalities influenced how the readers responded to the content of
the writers’ messages.
14 SOME GENERAL ISSUES

Seeing these theoretical contributions to writing research laid side by side, as Ala-
margot and Chanquoy have done for us, reminds us that in some ways the modeling
of writing behavior is still in a very early stage of development. In our field, models
have not yet achieved the level of explicitness often found in cognitive psychology.
Indeed, some psychological models are explicit enough to be embodied in runable
computer programs that can be used to identify the model’s predictions..
Generally, models in our field are not specific enough to be embodied in com-
puter programs. Indeed, they often seem quite under-specified. Typically, writing
process models consist of more or less vaguely described elements such as the fol-
lowing:
• a set of processes,
• a set of connections indicating interactions between or among the processes
and,
• a set of resources (e.g., types of knowledge, basic processes, etc).
A skeptic might criticize such models on the following grounds:
• The choice of the particular processes represented in a model is rarely fully jus-
tified. The fact, for example, that I decided to replace revision with reading in
my 1996 revision of the 1980 Hayes and Flower model suggests that there can
be argument about the appropriateness of the processes chosen for inclusion in a
model.
• In many models, the connections between processes are represented by lines
(sometimes with direction and sometimes without) in graphic representations of
the model. In the worst cases, these lines suggest little more than ‘these two
processes are related in some unspecified way.’ In some models, the connec-
tions are intended to indicate the sequence in which the processes are activated.
However, the suggested sequences are often uncontroversial (e.g., generating
COMMENTARY 237
ideas occurs prior to translating the ideas into language) and hence not of great
interest.
• The sets of resources are often simply uncontroversial lists of items everyone
would agree to (e.g., Long Term Memory contains knowledge of topic, audi-
ence and writing plans).
Bereiter and Scardamalia (1987) have defended their own models and, by implica-
tion, other such models as follows:
‘But the models of the kind we have proposed are not intended to describe. They are
more like representations of design concepts. They are intended to capture core ideas
which can be elaborated in different ways to correspond to real-world variability’ (p.
29).
I think that it is useful to describe models as representations of design concepts but I

also think that the critic I described above would not be satisfied. He or she might
reply in something of the following vein:
It has been some time since cognitive psychologists recognized the fallacy of using av-
eraged data to represent the processes of individuals. If some individuals approach a
task in one way while others approach it in a different way, then the averaged data may
represent no one. Ideally, in accounting for data, we should try to identify the various
approaches that are employed and to try to model each approach. A specific instantia-
tion of a model (say, in a computer program) should act like a member of the group
adapting that approach. Adjusting a few of the model’s parameters should allow the re-
searcher to fit the behavior of each individual. If adjusting the parameters of the model
does not succeed in fitting the behavior of all members of the group, then more models,
or design concepts, are needed.
The problem with the Bereiter and Scardamalia defense of models as design con-
cepts, the critic might conclude, is that current writing models are so under-specified
that it doesn’t make sense to try to fit them to data from human writers. There are
just too many unspecified parameters in the models. Sure, human data may not vio-
late the models but then no reasonable data would. Here, the critic goes too far. De-
spite its shortcomings, the current enterprise of modeling writing behaviors is still a
valuable one. Although current models are not fully explicit, neither are they com-
pletely vague. Predictions can be derived from them and tested. For example, it is
clear that the 1980 Hayes and Flower model does not fit the behavior involved in
free writing because free writing involves no revision. Perhaps the best known ex-
ample of prediction and testing of writing models is provided by Bereiter and Scar-
damalia’s (1987) extensive research on their knowledge-telling and knowledge-
transforming strategies. Also, the studies I summarized above provide other clear
examples of empirical testing of prediction. Littleton showed that contrary to prior
theoretical speculations, young children can adapt their language for an absent audi-
ence. Wallace and his colleagues predicted and then demonstrated that college-age
writers can modify their revision behaviors in response to brief instructions. This re-
sult supports the notion that a modifiable task schema guides revision. Schriver pre-
dicted and demonstrated that reader feedback can have a powerful influence in shap-
ing the writer’s perceptions of audience needs in text. Hatch et al. predicted and
tested the effects of rewriting on personality judgments derived from text.
238 HAYES
Clearly the field of writing research is not as well developed, either theoretically or
empirically, as the field of psychology but that is understandable. Scientific psy-
chology has had a considerable head start—more than a hundred years—over scien-
tific writing research. Further, writing is a very complex task compared to many of
the tasks studied by psychologists. Writing researchers can be pleased with the re-
search they have carried out and the models they have constructed even though they
are not as yet as complex or detailed as those of older, better established fields. How
can we promote the theoretical development of our field? I think that three direc-
tions need to be pursued:
First, we need to continue vigorous empirical research programs to identify em-
pirical phenomena, such as Bereiter and Scardamalia’s ‘knowledge telling’’ or
Hayes et al.’s ‘knowledge effect’ and to explore the parameters of these phenomena.
By vigorously pursuing such research, we can gain the knowledge that will allow us
to better specify our currently under-specified models.
Second, we need to maintain a strong connection between research and applica-
tion. I believe that when writing researchers become actively involved in helping to
solve practical communication problems, whether in schools or government or in-
dustry, they not only bring new ideas to the task but they also discover new ideas
that they can take away. I believe, therefore that a major goal of researchers in this
field should be to apply what is learned through research to improve the quality of
writing by studying writing in practical settings, by working to improve writing
pedagogy and by improving the tools that facilitate writing (e.g., writing software).
Involvement in such practical enterprises can not only enrich the researcher’s per-
spective on the nature of writing but can also help to create a constituency in the
community that values and supports writing research. A corollary of this idea is that
evaluation of the quality of writing is critical. If a major research goal is the im-
provement of writing, then we must be able to evaluate the quality of the written
product. That is, we must ask, ‘Does this text achieve its intended purpose?’ whether
that purpose is to inform, persuade, or entertain a specified audience or to teach the
writer. Answering these questions is not easy. For example, Hayes, Hatch and Silk
(2000) found that holistic evaluation of essays provided very unreliable information
about writing skills. A major goal for our field must be the identification of better
means to evaluate the quality of writing.
Third, we need to take a broad view of our field and to work toward integrative
theories that will tie together the results of both theoretical and applied research on
writing. Alamargot and Chanquoy’s admirable review of research and theory in our
field will be very helpful in promoting this goal.
REFERENCES
Abbott, R.D., & Berninger, V.W. (1993). Structural equation modeling of relationships among develop-
mental skills and writing skills in primary -and intermediate- grade writers. Journal of Educational
Psychology, 85 (3), 478-508.
Akiguet, S., & Piolat, A. (1996). Insertion of connectives by 9 to 11 year old children in an argumentative
text. Argumentation, 10 (2), 271-282.
Akiguet, S., Roussey, J.Y., & Piolat, A. (1993). The use of schema argumentation in 9-11 years old chil-
dren. In M. Gilly, & J.P. Roux (Eds.), Abstract volume of the fifth E.A.R.L.I. European Conference
(pp. 236). Aix en Provence, 31 août- 5 septembre.
Akyürek, A. (1992). On acomputational model of human planning. In J.A. Michon, & A. Akyürek (Eds),
Soar: A cognitive architecture in perspective (pp. 81-108). Dordrecht: Kluwer Academic Publishers.
Alamargot, D. (1997). Processus de récupération et d’organisation dans l’activité de rédaction de texte:
effets de l’acquisition de connaissances référentielles. Unpublished doctoral dissertation. University
of Poitiers, France.
Alamargot, D. (2000). Models of Activation and Text Production: some results and perspectives. EARLI -
Writing Conference 2000. Verona – September 7-9, 2000.
Alamargot, D., Espéret, E., & Savigny, A. (1993). Planning processes in writing directions: structure of
procedural knowledge and planning patterns in two knowledge domains. In Oral presentation:
EARLI Conference. Aix en Provence - France, 31 August - 5 September.
Alamargot, D., Favart, M., & Galbraith, D. (2000). Evolution of idea generation in argumentative writing:
writing as knowledge constituting or knowledge transforming process ? EARLI - Writing Conference
2000. Verona – September 7-9, 2000.
Alamargot, D., Favart, M., Coirier, P., Passerault, J.M., & Andriessen, J. (1999). Argumentative text:
planning and translating. In M.F. Crété and E. Espéret (Eds.). Proceedings of the 1998 European
Writing Conference – Writing and learning to write at the dawn of the 21st century. Editions du
C.N.R.S. (pp. 149-156)
Anderson, J.R. (1972). FRAN: a simulation model of free recall. In G.H. Bower (Ed.). The psychology of
learning and motivation: advances in research and theory (vol 5, pp. 315-378). New York: Aca-
demic Press.
Anderson, J.R. (1974). Retrieval of propositional information from Long Term Memory. Cognitive Psy-
chology, 6, 451-474.
Anderson, J.R. (1976). Language, memory and thought. Hillsdale, N.J.: Erlbaum.
Anderson, J.R. (1983a). The architecture of cognition. Cambridge M.A.: Harvard University Press.
Anderson, J.R. (1983b). A spreading activation theory of memory. Journal of Verbal Learning and Ver-
bal Behavior, 22, 261-295.
Anderson, J.R. (1990). The adaptative character of thought. Hillsdale, N.J.: Erlbaum.
Anderson, J.R. (1993). Rules of the mind. Hillsdale, N.J.: Erlbaum.
Anderson, J.R., & Bower, G. (1973). Human associative memory. Washington: Winston Sons.
Anderson, J.R., Reder, L. M., & Lebiere, C. (1996). Working memory: activation limitations on retrieval.
Cognitive Psychology, 30, 221-256.
Andriessen, J., Coirier, P., Chanquoy, L., & Bernardi, B. de (1997). Argumentative text writing: From
planning to translating. Paper presented at the Vth European Congress of Psychology: Invited Sym-
posium on ‘Psychology and the study of written text production’, Dublin (Ireland), July, 6th-11th.
Andriessen, J., Coirier, P., Roos L., Passerault J.M., & Bert-Erboul A. (1996). Thematic and structural
planning in constrained argumentative text production. In G. Rijlaarsdam, H. van den Berg, & M.
Couzijn (Eds.). Theories, Models and Methodology in Writing Research (pp.237-251). Amsterdam:
Amsterdam University Press.
Atkinson, R.C., & Shiffrin, R.M. (1968). Human memory: a proposed system and its control process. In
K.W. Spence (Ed.), The psychology of learning and motivation: Advances in research and theory
(Vol.2, pp. 89-195). New York: Academic Press
Baangert-Drowns, R.L. (1993). The word processor as an instructional tool: A meta-analysis of word
processing in writing instruction. Review of Educational Research, 63(1),69-93.
Baddeley, A.D. (1986). Working memory. Oxford, U.K.: Clarendon.
240 HAYES
Baddeley, A.D. (1990). Human memory. Theory and practice. Boston, London, Sydney, Toronto: Allyn
and Bacon.
Baddeley, A.D. (1996). Exploring the central executive. The Quarterly Journal ofExperimental Psychol-
ogy, 49A(1), 5-28.
Baddeley, A.D. (2000). The episodic buffer as a new component of Working Memory ?. Trends in Cogni-
tive Sciences, 4(11). 417-423.
Baddeley, A.D., & Hitch, G. (1974). Working memory. In G.A. Bower (Ed.). Recent advances in learn-
ing and motivation. Vol. 8 (pp. 47-90). New York: Academic Press.
Baddeley, A.D., & Wilson, B. (1985). Phonological coding and Short Term Memory in patients without
speech. Journal of Memory and Language, 24, 490-502.
Baker-Sennett, J., Matusov, E., & Rogoff, B. (1993). Planning asdevelopmental process. Advances in
Child Development and Behavior, 24, 253-281.
Barrit, L., & Kroll, B.M. (1978). Some implications of cognitive developmental psychology for research
in composing. In C.R. Cooper, & L. Odell (Eds.). Research on composing: Points of departure (pp.
49-57). Urbana: N.C.R.E.
Bartlett, E.J. (1982). Learning to revise: Some component processes. In M. Nystrand (Ed.). What writers
know. The language, process and structure of written discourse, (pp. 345-363). New York: Academic
Press.
Baudet, S. (1988). Récupération de l'information sémantique en mémoire: recouvrement des connaissan-
ces. European Journal of Psychology of Education, 3 (2), 163-176.
Beal, C.R. (1990). The development of textevaluation and revision skills. Child Development,61, 247-
258.
Beal, C.R. (1993). Contributions of developmental psychology to understanding revision: Implications
for consultation with classroom teachers. School Psychology Review, 22, 643-655.
Beal, C.R. (1996). The role of comprehension monitoring in children’s revision. Educational Psychology
Review, 8(3), 219-238.
Beason, L. (1993). Feedback and revision in writing across thecurriculum classes. Research in the Teach-
ing of English,27(4), 395-422.
Beaugrande, R. de (1984). Text production. Toward a science of composition. Norwood, N.J.: Ablex.
Benton, S.L., Kraft, R.G., Glover, J.A., & Plake, B.S. (1984). Cognitive capacity differences amoung
writers. Journal of Educational Psychology, 76, 820-834.
Bereiter, C. (1980). Development in writing. In L.W. Gregg, & E.R. Steinberg (Eds). Cognitive processes
in writing,(pp. 73-93). Hillsdale, N.J.: Erlbaum.
Bereiter, C., & Scardamalia, M. (1982). From conversation to composition: the role of instruction in a de-
velopmental process. In R. Glazer (Ed.). Advances in Instructional Psychology (pp. 1-64). Hillsdale,
N.J.: Erlbaum.
Bereiter, C., & Scardamalia, M. (1983). Levels of inquiry in writing in writing research. In P. Mosenthal,
S. Walmsley, & L. Tamor (Eds.), Research on writing: principles and method. New York, Longman.
Bereiter, C., & Scardamalia, M. (1984). Information processing demands of text composition. In H.
Mandl, N. Stein, & T. Trabasso (Eds). Learning and comprehension of texts. Hillsdale, LEA.
Bereiter, C., & Scardamalia, M. (1986). Levels of inquiry into the nature of expertise in writing. Review
of Research in Education, 13, 250-281.
Bereiter, C., & Scardamalia, M. (1987). The psychology of written composition. Hillsdale, NJ: Lawrence
Erlbaum Associates.
Bereiter, C., & Scardamalia, M. (1988). The psychology of written composition. Journal of Memory and
Language, 27, 261-278.
Bereiter, C., Burtis, P.J., & Scardamalia, M. (1988). Cognitive operations in constructing main point in
written composition. Journal of Memory and Language, 27, 261-278.
Berg, J. (1986). The problem of language control. Editing, monitoring and feed-back. Psychological Re-
search, 48, 133-144.
Berninger, V.W. (1994). Reading and writing acquisition. A developmental neuro-psychological perspec-
tive. Dubuque, IA: Brown and Benchmark.
Berninger, V.W., & Füller, F. (1992). Gender differences in orthographic, verbal and compositional flu-
ency: Implications for assessing writing disabilities in primary grade children. Journal of School Psy-
chology, 30, 363-382.
Berninger, V.W., & Swanson, H.L. (1994). Modification of the Hayes and Flower model to explain be-
ginning and developing writing. In E. Butterfield (Ed.). Advances in cognition and Educational Prac-
COMMENTARY 241
tice. Vol. 2. Children’s writing: toward a process theory of development of skilled writing (pp. 57-
82). Greenwich: CT: JAI Press.
Berninger, V.W., Cartwright, A.C., Yates, C.M., Swanson, H.L., & Abbott, R.D. (1994). Developmental
skills related to writing and reading acquisition in the intermediate grades. Reading and Writing: An
Interdisciplinary Journal, 6, 161-196.
Berninger, V.W., Füller, F., & Whitaker, D. (1996). A process model of writing development across the
life span. Educational Psychology Review, 8(3), 193-217.
Berninger, V.W., Vaughan, K.B., Graham,S., Abbott, R.D., Abbott, S.P., Rogan, L.W., Brooks, A., &
Reed E. (1997). Treatment of handwriting problems in beginning writers: transfer from handwriting
to composition. Journal of Educational Psychology, 89(4), 652-666.
Berninger, V.W., Whitaker, D., Feng,Y., Swanson, H.L., & Abbott, R.D. (1996). Assessment of planning,
translating and revising in junior high writers. Journal of School Psychology, 34(1), 23-52.
Berninger, V.W., Yates, C., Cartwright, A., Rutberg, J., Remy, E., & Abbott, R. (1992). Lower-level de-
velopmental skills in beginning writing. Reading and Writing: an interdisciplinary Journal, 4, 257-
280.
Best, J.B. (1989). Cognitive Psychology. St Paul, MN: West Publishing Company.
Bestgen, Y., & Costermans, J. (1994). Time, space and action: exploring the narrative structure and its
linguistic marking. Discourse Processes, 17, 421-446.
Bestgen, Y., & Costermans, J. (1997). Temporal markers of narrative structure: studies in production. In
J. Costermans, & M. Fayol (Eds.). Processing interclausal relationships. Studies in the production
and comprehension of text (pp. 201-218).Mahwah, NJ: L. Erlbaum Associates.
Bialystok, E. (1986). Factors in the growth of linguistic awareness. Child Development, 57, 498-510.
Bock, J.K. (1982). Toward a cognitive psychology of syntax information processing contributions to sen-
tence formulation. Psychological Review, 89 (1), 1-47.
Bock, J.K. (1995). Sentence production: From mind to mouth. In J.L.Miller, & P.D. Eimas (Eds.). Hand-
book of perception andcognition. Vol. 11. Speech, language and communication (pp. 181-216). Or-
lando, FL: Academic Press.
Bock, J.K. (1996). Language production: methods and methodologies. Psychonomic Bulletin and Review,
3(4),395-421.
Bock, J.K., & Cutting, J. C. (1992). Regulating mental energy: performance units in language production.
Journal of Memory and Language, 31, 99-127.
Bock, J.K., & Eberhard, K.M. (1993). Meaning, sound and syntax in English number agreement. Lan-
guage and Cognitive Processes, 8(1), 57-99.
Bock, J.K., & Levelt, W.J.M. (1994). Language production: grammatical encoding. In M. A. Gernsbacher
(Ed.), Handbook of Psycholinguistics (pp. 945-978). San Diego: Academic Press, California.
Bock, J.K., & Warren, R.K. (1985). Conceptual accessibility and syntactic structure in sentence formula-
tion. Cognition, 21, 47-67.
Boniface, C., & Pimet, O. (1992). Les ateliers d’écriture. Paris: Retz.
Boomer, D.S. (1965). Hesitation and grammatical encoding. Language and Speech, 8, 148-158.
Boscolo, P. (1995). The cognitive approach to writing and writing instruction: a contribution to a critical
appraisal. Current Psychology of Cognition, 14(4), 343-366.
Bourdin, B., & Fayol, M. (1994). Is a written language production more difficult than oral language pro-
duction ? A Working Memory approach. International Journal of Psychology, 29 (5), 591-620.
Bourdin, B., & Fayol, M. (1996). Mode effects in a sentence production span task. Cahier de Psychologie
Cognitive/Current Psychology of Cognition, 15(3), 245-264.
Bower, G.H. (1970). Organizational factors in memory. Cognitive Psychology, 1, 18-46.
Bracewell, R. J. (1983). Investigated the control of writing skills. In P. Mosenthal, L. Tamor, & S.
Walmsley (Eds.). Research on writing: principles and methods. New-York: Longman.
Bracewell, R. J., Scardamalia, M., & Bereiter, C. (1978). The development of audience awareness in writ-
ing. Ressources in Education, 12, 154-433.
Bradford DeCosta, S. (1992). Sociological findings in youngchildren’s word-processed writings. Com-
puters in Human Behavior, 8, 17-5.
Brandt, D. (1992). The cognitive as the social. An ethno methodological approach to writing process re-
search. Written Communication, 9(3), 315-355.
Breetvelt, I., van den Bergh, H., & Rijlaarsdam, G. (1994). Relations between writing processes and text
quality: When and how ? Cognition and Instruction, 12(2), 103-123.
242 HAYES
Bridwell, L.S. (1980). Revising strategies in twelfth grade students’ transactional writing. Research in the
Teaching of English, 14, 197-222.
Britton, J. (1970). Language and Learning. University of Miami Press. Florida.
Britton, J., Burgess, T., Martin, N., McLeod, A., & Rosen, H. (1975). The development of writing abilities
(11-18). London: Macmillan Education.
Bronckart, J.P. (1977). Théories du langage. Une approche critique. Bruxelles: Mardaga.
Bronckart, J.P., Bain, D., Schneuwly, B., Davaud, C., & Pasquier, A. (1985). Le fonctionnement des dis-
cours. Un modèle psychologique et une méthode d’analyse. Neuchâtel, Paris: Delachaux et Niestlé.
Brown, J.S., McDonald, J.L., Brown, T.L., & Carr, T.H. (1988). Adapting to processing demands in dis-
course production: the case of handwriting. Journal of Experimental Psychology: Human Perception
and Performance, 14(1), 45-59.
Bruce, B., Collins, A. M., Rubin, A. D., & Genter, D. (1979). A cognitive science approach to writing. In
C. H. Fredericksen, M. F. Whiteman, & J. D. Dominic (Eds.), Writing: the nature, development and
teaching of written communication. Hillsdale, N.J.: L.E.A.
Brügelmann, H. (1996). Modèles empiristes, constructivistes ou interactionnistes de l’acquisition du lan-
gage écrit. Projekt OASE.
Brügelmann, H. (1997). Curriculum concepts and research evidence on language experience in open
classrooms. Projekt OASE.
Bryant, P.E., & Bradley, L. (1983).Psychological strategies and the development of reading and writing.
In M. Martlew (Ed.), The psychology of written language. Developmental and Educational Perspec-
tives, (pp. 163-178). New York: Wiley and Sons.
Burtis, P.J., Bereiter, C., Scardamalia, M., & Tetroe, J. (1983). The development of planning in writing.
In B. M. Kroll, & G. Wells (Eds.). Explorations in the development of writing (pp. 153-174). New-
York: Wiley.
Butterfield, E.C., Albertson, L.R., & Johnston, J. (1995). On making cognitive theory more general and
developmentally pertinent. In F. Weinert, & W. Schneider (Eds.), Memory, performance and compe-
tence: Issues in growth anddevelopment (pp. 181-205). Hillsdale, NJ: L. Erlbaum Associates.
Butterfield, E.C., Hacker, D.J., & Albertson, L.R. (1996). Environmental, cognitive and metacognitive in-
fluences on text revision: assessing the evidence. Educational Psychology Review, 8(3), 239-297.
Butterfield, E.C., Hacker, D.J., & Plumb, C. (1994). Topic knowledge, linguistic knowledge and revision
processes as determinants of text revision. In E.C. Butterfield (Ed.), Advances in cognition and Edu-
cational Practice. Vol. 2: Children’s writing: Toward a process theory of the development of skilled
writing (pp. 83-141). Greenwich, CT: JAI Press.
Butterworth, B. (1980a). Introduction: A brief review of methods of studying language production. In B.
Butterworth (Ed.). Language production. Vol.1: Speech and talk (pp.1-17). NewYork: Academic
Press.
Butterworth, B. (1980b). Evidence from pauses in speech. In B.Butterworth (Ed.), Language Production.
Vol.1: Speech and talk (pp. 155-177). New York: Academic Press.
Caccamise, D.J. (1987). Idea generation in writing. In A. Matsuhashi(Ed.). Writing in real time, (pp.224-
253). Norwood, N.J.: Ablex.
Cameron, C.A., Edmunds, G., Wigmore, B., Hunt, A.K., & Linton, M.J. (1997). Children’s revision of
textual flaws. International Journal of Behavioral Development, 20(4), 667-680.
Cannella, G.S. (1988). The effect of environmental structure on writing produced by young children.
Child Study Journal, 18(3), 207-221.
Cantor, J., & Engle, R. W. (1993). Working-memory capacity as Long Term Memory activation: an indi-
vidual differences approach. Journal of Experimental Psychology: Learning, Memory and Cognition,
19 (5), 1101-1114.
Caramazza, A. (1991). Issues in reading, writing and speaking: A neuropsychological perspective.
Dordrecht, the Netherlands: Kluwer Academic.
Carey, L., Flower, L., Hayes, J. R., Schriver, K. A., & Haas, C. (1989). Differences in writer's initial task
representations (N°35). Center for the Study of Writing. Carnegie Mellon University. Pittsburgh,
P.A.
Case, R. (1985). Intellectual development: Birth to adulthood. New York: Academic Press.
Case, R., Kurland, D. M., & Goldberg, J. (1982). Operational efficiency and the growth of short term
memory span. Journal of Experimental Child Psychology, 33, 386-404.
COMMENTARY 243
Chanquoy, L. (1991). Ponctuation et connecteurs: Acquisition et fonctionnement. Etudes comparatives

chez l’enfant et l’adulte (Méthodes off-line et on-line). Unpublished doctoral dissertation. University
of Bourgogne. France
Chanquoy, L. (1997). Thinking skills and composing: examples of text revision. In J.H.M. Hamers, & M.
Overtoom (Eds.) Inventory of European Programmes for teaching thinking (pp. 179-185). Utrecht:
Sardes.
Chanquoy, L. (1998). La rédaction de textes: mémoire de travail et niveaux de traitement. Habilitation à
Diriger les Recherches. Université de Provence.
Chanquoy, L. (2001). How to make a revising task easier ? A study of text revision from 3rd to 5th grade.
British Journal of Educational Psychology, March 2001.
Chanquoy, L., & Fayol, M. (1991). Etude de l’utilisation des signes de ponctuation et des connecteurs
chez des enfants (8-10 ans) et des adultes. Pratiques, 70, 107-124.
Chanquoy, L., & Fayol, M. (1995). Analyse de l’évolution de l’utilisation de la ponctuation et des
connecteurs, dans deux types de texte. Etude longitudinale du C.P. au C.E.2. Enfance,2, 227-241.
Chanquoy, L., Foulin, J. N., & Fayol, M. (1990). Temporal management of short text writing by children
and adults. Cahiers de Psychologie Cognitive, 10 (5), 513-538.
Chanquoy, L., Foulin, J.N., & Fayol, M. (1996). Writing in adults: A real time approach. In G. Rijlaars-
dam, H. van den Bergh, & M. Couzijn (Eds.). Current trends in writing research: Theories, models
and methodology (pp. 36-43). Amsterdam: Amsterdam University Press.
Charness, N. (1991). Expertise in chess: the balance between knowledge and search. In K. A. Ericsson, &
J. Smith (Eds.). Toward a general theory of expertise: prospects and limits (pp. 39-63). New York:
Cambridge University Press.
Chase, W.G., & Ericsson, K.A. (1981). Skilled memory. In J. R. Anderson (Ed.), Cognitive skills and
their acquisition. Hillsdale, N.J.: Erlbaum
Chase, W.G., & Simon, H. A. (1973). The mind's eye in chess. In W. G. Chase (Ed.), Visual information
processing. New-York: Academic Press.
Chi, M.T.H. (1976). Short term memory limitations in children: capacity or processing deficits ? Memory
and Cognition, 4, 559-580.
Chi, M.T.H. (1977). Ages differences in memory span. Journal of Experimental Child Psychology, 23,
266-281.
Chi, M.T.H. (1978). Knowledge Structures and memory development. In R. S. Siegler (Ed.), Children's
thinking: What develops ? Hillsdale, N.J.: Erlbaum.
Chin, E. (1994). Redefining « context » in research on writing. Written Communication, 11(4),445-482.
Chomsky, N. (1980). Rules and representations. NY. Columbia University Press.
Clark, H.H., & Clark, E.V. (1977). Psychology and Language. New-York: Harcourt Brace Javanovich.
Coirier, P. (1996). Composing argumentative texts: cognitive and/ortextual complexity. In G. Rijlaars-
dam, H. van den Bergh, & M. Couzijn(Eds.). Current trends in writing research: Theories,models
and methodology (pp. 317-338). Amsterdam: Amsterdam University Press.
Coirier, P. (1997). Des types de textes ? Un eapproche de psychologie cognitive. Document de Synthèse
pour l’Habilitation à Diriger des Recherches. Université de Poitiers.
Coirier, P., & Golder, C. (1993). Writing argumentative text: A developmental study of the acquisition of
supporting structure. European Journal of Psychology of Education, 8(2), 169-181.
Coirier, P., Andriessen, J., & Chanquoy, L. (2000). From planning to translating: the specificity of argu-
mentative writing, in G. Rijlaarsdam & E. Espéret (Series Eds.) & J. Andriesse, & P. Coirier (Vol.
Eds.) Studies in Writing: Vol 5. Foundations of argumentative text processing (pp. 1-28), &. Amster-
dam: Amsterdam University Press.
Coirier, P., Gaonac’h, D., & Passerault, J.M. (1996). Psycholinguistique textuelle. Approche cognitive de
la compréhension et de la production des textes. Paris: Colin.
Collins, A.M., & Gentner, D. (1980). A framework for a cognitive theoryof writing. In L.W. Gregg, &
E.R. Steinberg (Eds.), Cognitive processes in writing, (pp. 51-71). Hillsdale,NJ: L. Erlbaum Associ-
ates.
Collins, A.M., & Loftus, E.F. (1975). A spreading-activation theory of semantic memory. Psychological
Review, 82, 407-428.
Collins, A.M., & Quillian, M.R. (1969). Retrevial time from semantic memory. Journal of Verbal Learn-
ing and Verbal Behavior, 8, 240-247.
Cooper, C.R., & Matsuhashi, A. (1983). A theory of the writing process. In M.L. Martlew (Ed.). The psy-
chology of written language (pp. 3-39). New York: John Wiley and Sons.
244 HAYES
Cooper, W.E., & Paccia-Cooper, J. (1980). Syntax and speech. Cambridge, M.A.: Harvard University
Press.
Cossu, G., & Marshall, J.C. (1990). Are cognitive skills a prerequisite for learning to read and write ?
Cognitive Neuropsychology, 7(1), 21-40.
Costermans, J., & Bestgen, Y. (1991). The role of temporal markers in the segmentation of narrative dis-
course. Cahiers de Psychologie Cognitive / European Bulletin of Cognitive Psychology, 11(3), 349-
370.
Costermans, J., & Fayol, M. (1997). Processing interclausal relationships: Studies in the production and
comprehension of text. Mahwah, NJ: Lawrence Erlbaum Associates.
Cowan, N. (1988). Evolving conceptions of memory storage, selective attention and their mutual con-
straints within the human information processing system. Psychological Bulletin, 104 (2), 163-191.
Cowan, N. (1993). Activation, attention and Short Term Memory. Memory and Cognition, 21 (2), 162-
167.
Craik, F.I.M. et Lockhart, R.S. (1972). Depth of processing: a framework for memory research. Journal
of Verbal Learning and Verbal Behavior, 11, 671-684.
Daiute, C.A. (1984). Performance limits on writers. In R. Beach, & L. S. Bridwell (Eds.). New Directions
in composition research (pp. 205-224). New-York: Guilfort Press
Daiute, C.A. (1992). Multimedia composing: extending the resources of kindergarten to writers across the
grades. Langage Arts,69, 250-260.
Daiute, C.A., & Kruidenier, J. (1985). A self-questioning strategy to increase young writers’ revising
processes. Applied Psycholinguistics, 6, 307-318.
Daneman, M., & Carpenter, P.A. (1980). Individual differences in Working Memory and reading. Jour-
nal of Verbal Learning and Verbal Behavior, 19, 450-466
Daneman, M., & Carpenter, P.A. (1983). Individual differences in integrating information between and
within sentences. Journal of Experimental Psychology: Learning, Memory and Cognition, 9, 561-
584.
Daneman, M., & Green, I. (1986). Individual differences in comprehending and producing words in con-
text. Journal of Memory and Language, 25, 1-18.
Daneman, M., & Stainton, M. (1993). The generation effect in reading and proof-reading. Reading and
Writing: An Interdisciplinary Journal, 5, 297-313.
Dansac, C, & Alamargot, D. (1999). Accessing referential information during text composition: when and
why ? In G. Rijlaarsdam, & E. Espéret (Series Eds), & M. Torrance, & D. Galbraith (Vol. Eds), Stud-
ies in Writing: Vol 4. Knowing what to write: cognitive processes in the generation, selection and de-
velopment of ideas during text production. Amsterdam. Amsterdam University Press (pp. 79-97).
Dell, G.S. (1986). A spreading activation theory of retrieval in sentence production. Psychological Re-
view, 93, 283-321.
Dempster, F. N. (1981). Memory span: sources of individual and developmental differences. Psychologi-
cal Bulletin, 89 (1), 63-100.
Dobbie, L., & Askov, E.N. (1995). Progress of handwriting research in the 1980s and future prospects.
The Journal of Educational Research, 88(6), 339-351.
Egan, D. E., & Schwartz, B. J. (1979). Chunking in recall of symbolic drawings. Memory and Cognition,
7 (2), 149-158.
Ehrich, V., & Koster, C. (1983). Discourse organization and sentence form: the structure of room descrip-
tion in Deutsch. Discourse Processes, 6, 169-195.
Ehrlich, M.F., & Delafoy, M. (1990). La mémoire de travail: Structure, fonctionnement, capacité.
L’AnnéePsychologique, 90, 403-428.
Eigler, G., Jechle, T., Merziger, G., & Winter, A. (1991). Writing and Knowledge: effects and re-effects.
European Journal of Psychology of Education, 4 (2), 225-232.
Engelhard, G., Gordon, B., & Gabrielson, S. (1992). The influences of mode of discourse, experiential
demand and gender on the quality of student writing. Research in the Teaching of English, 26(3),
315-336.
Ericsson, K.A. (1985). Memory skills. Canadian Journal of Psychology, 39, 188-231.
Ericsson, K.A., & Kintsch, W. (1995). Long-term Working Memory. Psychological Review, 102 (2), 211-
245.
Ericsson, K.A., & Lehmann, A.C. (1996). Expert and exceptional performance: evidence of maximal
adaptation to task constraints. Annual Review of Psychology, 47,273-305.
Ericsson, K.A., & Simon, H.A. (1978). Verbal reports as data. Psychological Review,87, 215-251.
COMMENTARY 245
Espéret, E. (1989). Writing texts of different types: a changes in planning processes or knowledge
bases ?. In Third European Conference for Research on Learning and Instruction. Madrid, September
Espéret, E., & Piolat, A. (1991). Production, planning and control. In G. Denhière, & J. P. Rossi (Eds.),
Texts and Text processing (pp. 317-331). Amsterdam: North Holland.
Faigley, L., & Witte, S.P. (1981). Analysing revision. College Composition and Communication,32, 400-
414.
Faigley, L., & Witte, S.P. (1984). Measuring the effects of revisions on text structure. In R. Beach, &
L.S.Bridwell (Eds.). New directions in composing research (pp. 95-108). New York: Guilford Press.
Faigley, L., Cherry, R.D., Jollifre, D.A., & Skinner, A.M. (1985). Assessing writers’ knowledge and
processes of composing. Norwood, N.J.: Ablex.
Farnham-Diggory, S., & Gregg, L. W. (1975). Short-term memory function in young readers. Journal of
Experimental Child Psychology, 19, 279-298
Favart, M., & Chanquoy, L. (1999). How children ensure cohesion in written texts with or without con-
nectives. In M.F. Crété, & E. Espéret (Eds.), Proceedings of the 1998 European Writing Conference
– Writing and learning to write at the dawn of the 21st century (pp. 213-218). Editions du C.N.R.S.
Favart, M., & Passerault, J.M. (1996). Functionality of cohesion devices in the management of local and
global coherence: Two studies in children’s written production of narratives. In G. Rijlaarsdam, H.
van den Bergh, & M. Couzjin (Eds.), Studies in Writing: Vol 1. Theories, models and methodology in
writing research (pp.349-365). Amsterdam: Amsterdam University Press.
Fayol, M. (1985). Le récit et sa construction. Paris: Delachaux, & Niestlé.
Fayol, M. (1986). Les connecteurs dans les récits écrits. Pratiques, 49, 101-113.
Fayol, M. (1991). From sentence production to text production: Investigating fundamental processes.
European Journal of Psychology of Education, 6, 99-117.
Fayol, M. (1994). From declarative and procedural knowledge to the management of declarative and pro-
cedural knowledge. European Journal of Psychology of Education, IX(3), 179-190.
Fayol, M. (1997). Des idées au texte. Paris: Presses Universitaires de France.
Fayol, M. (1999). From on-line management problems to strategies in written production. In M. Torrance,
& G.C. Jeffery (Eds), The cognitive demands of writing. Processing capacity and Working Memory
effects in text production (pp. 13-23). Amsterdam: Amsterdam University Press.
Fayol, M., & Gombert, J.E. (1987). Le retour de l’auteur sur son texte: bilan provisoire des recherches
psycholinguistiques. Repères, 73, 12-25.
Fayol, M., & Schneuwly, B. (1987). La mise en texte et ses problèmes. In J. L. Chiss, J. P. Laurent, J. C.
Mever, H. Romian, & B. Schneuwly (Eds.), Apprendre/Enseigner à produire des textes écrits (pp.
223-240). Bruxelles: De Boek
Fayol, M., Largy, P., & Lemaire, P. (1994). When cognitive overload enhances subject-verb agreement
errors. The Quarterly Journal of Experimental Psychology, 47A, 437-464.
Fitzgerald, J. (1987). Research on revision in writing. Review of Educational Research, 57(4),481-506.
Fitzgerald, J., & Markham, L.R. (1987). Teaching children about revision in writing. Cognition and In-
struction, 4(1), 3-24.
Flower, L.S., & Hayes, J.R. (1980). The dynamic of composing: Making plans and juggling constraints.
In L.W. Gregg, & E.R. Steinberg (Eds). Cognitive processes in writing, (pp.31-50). Hillsdale, N.J.:
Erlbaum.
Flower, L.S., & Hayes, J.R. (1981a). The pregnant pause: an inquiry into the nature of planning. Research
in the Teaching of English, 15(3), 229-243.
Flower, L.S., & Hayes, J.R. (1981b). Plans that guide the composing process. In C.H. Frederiksen, & J.F.
Dominic (Eds), Writing: The nature, development and teaching of written communication.Vol. 2:
Writing: Process, development and communication (pp. 39-58). Hillsdale, NJ: Laurence Erlbaum As-
sociates.
Flower, L.S., & Hayes, J.R. (1981c). A cognitive process theory of writing. College Composition and
Communication, 32, 365-387.
Flower, L.S., & Hayes, J.R. (1983). A cognitive model of the writing process in adults. Final Report.
Pittsburgh, P.A. Carnegie Mellon
Flower, L.S., Hayes, J.R., Carey, L., Schriver, K., & Stratman, J. (1986). Detection, diagnosis and the
strategies of revision. College Composition and Communication, 37, 16-55.
Flower, L.S., Schriver, K., Carey, L., Haas, C., & Hayes, J.R. (1989). Planning in writing: the cognition
of a constructive process (Tech. Report N°34). Center for the study of language. Berkeley, CA, &
Pittsburgh, P.A.
246 HAYES
Fodor, J.A. (1983). The modularity of mind. Cambridge, MA: MIT Press
Fodor, J.A. (1985). Precis of the modularity of mind. Behavioral and Brain sciences, 8, 1-42.
Fodor, J.A., Bever, T.G., & Garrett, M.F. (1974). The psychology of language. New-York: McGraw-Hill.
Ford, M., & Holmes, V. M. (1978). Planning units and syntax in sentence production. Cognition, 6, 35-53
Foulin, J.N. (1993). Pause et débit: les indicateurs temporels de la production écrite. Etude comparative
chez l’enfant et l’adulte. Unpublished doctoral dissertation. University of Bourgogne, Dijon - France.
Foulin, J.N. (1995). Pauses et débits: les indicateurs temporels de la production écrite. L’Année Psycholo-
gique, 95, 483-504.
Foulin, J.N., & Fayol, M. (1988). Etude en temps réel de la production écrite chez des enfants de 7 à 8
ans. European Journal of Psychology of Education, 3 (4), 461-475.
Foulin, J.N., Chanquoy, L., & Fayol, M. (1989). Approche en temps réel de la production des connecteurs
et de la ponctuation: vers un modèle procédural de la composition écrite. Langue Française, La-
rousse, 81, 5-20.
Freedman, S.W. (Ed.) (1986). The acquisition of written language: Response and revision. Norwood, NJ:
Ablex Publishing Corporation.
Friederici, A. D., Shoenle, P. W., & Goodglass, H. (1981). Mechanisms underlying writing and speech in
aphasia. Brain and Language, 13, 212-222.
Galbraith, D. (1992). Conditions for discovery through writing. Instructional Science, 21, 45-72.
Galbraith, D. (1996). Self-monitoring, discovery through writing and individual differences in drafting
strategy. In G. Rijlaarsdam, H. van der Bergh, & M. Couzijn (Eds.), Studies in Writing: Vol 1. Theo-
ries, Models and Methodology in Writing Research. (pp 121-141). Amsterdam: Amsterdam Univer-
sity Press.
Galbraith, D. (1999). Writing as a knowledge constituting process. In G. Rijlaarsdam & E. Espéret (Series
Eds), & M. Torrance & D. Galbraith (Vol. Eds). Studies in Writing: Vol 4. Knowing what to write:
conceptual processes in text production (pp. 139-160). Amsterdam. Amsterdam University Press.
Gaonac’h, D., & Larigauderie, P. (2000). Mémoire et fonctionnement cognitif: la mémoire de travail.
Armand Colin. Coll. U.
Garrett, M.F. (1975). The analysis of sentence production. In G.H. Bower (Ed.), The psychology of learn-
ing and motivation. New York: Academic Press.
Garrett, M.F. (1980). Levels of processing in sentence production. In B. Butterworth (Ed.). Language
production. Vol.1. Speech and talk (pp.177-220). New York: Academic Press.
Garrett, M.F. (1982). Production of speech: observations from normal and pathological use. In A. Ellis
(Ed.), Normality and pathology in cognitive functions (pp. 19-76). London: Academic Press.
Garrett, M.F. (1984). The organization of processing structure for language production: Applications to
aphasicspeech. In D. Caplan, A.R. Lecours, & A. Smith (Eds.). Biological perspectives on language.
Cambridge, Ma.: M.I.T.Press.
Garrett, M.F. (1988). Processes in language production. In J.F.J. Newmeyer (Ed.), Language: psychologi-
cal and biological aspects. Vol. 3 (pp. 69-96). Cambridge: Cambridge University Press.
Gathercole, S.E., & Baddeley, A.D. (1993). Working memory and language. Hillsdale, NJ: L. Erlbaum
Associates.
Gernsbacher, M.A. (1989). Mechanisms that improve referential access. Cognition, 32, 99-156
Gernsbacher, M.A., & Faust, M.E. (1991). The mechanism of suppression: a component of general com-
prehension skill. Journal of experimental psychology: learning, Memory and cognition., 17, 245-262.
Glynn, S.M., Britton, B.K., Muth, K.D., & Dogan, N. (1982). Writing and revising persuasive documents:
cognitive demands. Journal of Educational Psychology, 74(4), 557-567.
Golder, C. (1996). Le développement des discours argumentatifs. Neuchâtel: Delachaux et Niestlé (col-
lection: Actualités Pédagogiques et Psychologiques).
Goldman-Eisler, F. (1968). Psycholinguistics: experiments in spontaneous speech. Londres: Academic
Press.
Goldman-Eisler, F. (1972). Pauses, clauses, sentences. Language, & Speech, 15 (2), 103-113.
Gombert, A., & Roussey, J.Y. (1993). Computer-assisted training effects on argumentative text writing
skills in children. In G. Eigler, & T. Jechle (Eds). Writing:Current trends in European research
(pp.183-196). Freiburg: Hochschul Verlag.
Gould, J.D. (1980). Experiments on composing letters: some facts, some myths and some observations. In
L.W. Gregg, & E.R. Steinberg (Eds), Cognitive processes in writing, (pp.98-127). Hillsdale, N.J.: L.
Erlbaum Associates.
COMMENTARY 247
Grabowski, J. (1996). Writing and speaking: common grounds and differences toward a regulation theory
of written language production. In M. Levy, & S. Ransdell (Eds.). The science of writing (pp 73-91).
Hillsdale, NJ: Erlbaum.
Graham, S. (1990). The role of production factors in learning disabled students compositions. Journal of
Educational Psychology, 52 (4), 781-791.
Graham, S., & Harris, K.R. (1994). The role and development of self-regulation in the writing process. In
D. Shunk, & B. Zimmerman (Eds), Self-regulation of learning and performance: Issues and educa-
tional applications (pp. 203-228). Hillsdale, NJ: Erlbaum Associates.
Graham, S., & Harris, K.R. (1996). Addressing problems in attention, memory and executive functioning.
An example from self-regulated strategy development. In L. Reid, & N. Krasnegor (Eds). Attention,
memory and executive function. Baltimore: Paul Brookes.
Graham, S., & Harris, K.R. (1997). Self-regulation and writing: where do we go from here ? Contempo-
rary Educational Psychology, 22, 102-114.
Graham, S., & Weintraub, N. (1996). A review of handwriting research: progress and prospects from
1980 to 1994. Educational Psychology Review, 8(1), 7-87.
Graham, S., Berninger, V.W., Abbott, R.D., Abbott, S.P., & Whitaker, D. (1997). Role of mechanics in
composing of elementary school children: a new methodological approach. Journal of Educational
Psychology, 89(1), 170-182.
Graves, D.H. (1975). An examination of the writing processes of seven year old children. Research in the
Teaching of English,9, 227-241.
Gregg, N., Sigalas, S.A., Hoy, C., Wisenbaker, J., & McKinley, C. (1996). Sense of audience and the
adult writer: a study across competence levels. Reading and Writing: an Interdisciplinary Journal, 8,
121-137.
Grosjean, F., Grosjean, L., & Lane, H. (1979). The patterns of silence: performance structures in sentence
production. Cognitive psychology, 11, 58-81.
Hacker, D.J. (1994). Comprehension monitoring as a writing process. Advances in Cognition and Educa-
tional Practice, 6, 143-172.
Hacker, D.J. (1997). Comprehension monitoring of written discourse across early-to-middle adolescence.
Reading and Writing:an Interdisciplinary Journal, 9, 207-240.
Hacker, D.J., Plumb, C., Butterfield, E.C., Quathamer, D., & Heineken, E. (1994). Text revision: detec-
tion and correction of errors. Journal of Educational Psychology, 86(1),65-78.
Hague, S.A., & Mason, G.E. (1986). Using the computer’s readability measure to teach students to revise
their writing. Journal of Reading, 3, 14-18.
Hartley, J. (1991). Psychology: Writing and computers: a review of research. Visible Language, 25(4),
339-375.
Hatch, J.A., Hill, C.A., & Hayes, J.R. (1993). When the messenger is the message: Readers' impressions
of writers' personalities. Written Communication, 10(4), 569–598.
Hayes, J.R. (1985). Three problems in teaching general skills. In S. Chipman, J. Segal, & R. Glaser (Eds).
Thinking and learning skills. Hillsdale, NJ: Laurence Erlbaum Associates.
Hayes, J.R. (1989). Writing research: The analysis of a very complex task. In D. Klahr, & R. Kotovsky
(Eds.), Complex information processing: The impact of H.A. Simon. Hillsdale, NJ: L. Erlbaum Asso-
ciates.
Hayes, J.R. (1990). Individuals and environments in writing instruction. In B.F. Jones, & L. Idol (Eds.),
Dimensions of thinking and cognitive instruction (pp. 241-263). Hove: Laurence Erlbaum Associates.
Hayes, J.R. (1996). A new framework for understanding cognition and affect in writing. In C. M. Levy, &
S. Ransdell (Eds.). The science of writing: Theories, methods, individual differences and applications
(pp. 1–27). Mahwah, NJ: Lawrence Erlbaum Associates.
Hayes, J.R., & Flower, L.S. (1980). Identifying the organization of writing processes. In L. W. Gregg, &
E.R. Steinberg (Eds.), Cognitive processes in writing: An interdisciplinary approach (pp. 3–30).
Hillsdale, NJ: Lawrence Erlbaum.
Hayes, J.R., & Flower, L.S. (1983). Uncovering cognitive processes in writing: an introduction of proto-
col analysis. In P. Mosenthal, S.Walmsley, & L. Tamor (Eds), Research on writing: principles and
methods, (pp. 206-219). New York: Longman.
Hayes, J.R., & Flower, L.S. (1986). Writing research and the writer. American Psychologist, 41, 1106-
1113.
Hayes, J.R., & Nash, J.D. (1996). On the nature of planning in writing. In C.M. Levy, & S. Ransdell
(Eds.). The science of writing (pp. 29-55). Mahwah, NJ: L. Erlbaum Associates.
248 HAYES
Hayes, J.R., Flower, L.S., Schriver, K.A., Stratman, J., & Carey, L. (1987). Cognitive processes in revi-
sion. In S. Rosenberg (Ed.). Advances in psycholinguistics. Vol. 2: Reading, writing and language.
Cambridge: Cambridge University Press.
Hayes, J.R., Hatch, J. A., & Silk, C.M. (2000). Does holistic assessment predict writing performance? Es-
timating the consistency of student performance on holistically scored writing assignments. Written
Communication, 17(1), 3-26.
Hayes, J.R., Schriver, K.A., Blaustein, & Spilka, R. (1986). If it's clear to me, it must be clear to them:
How knowledge makes it difficult to judge. Paper presented at the annual meeting of the American
Educational Research Association, San Francisco, CA.
Herrmann, T., & Grabowski, J. (1995). Pre-terminal levels of process in oral and written language. In
U.M.Quasthoff (Ed.). Aspects of oral communication, (pp. 67-87). Berlin, New York: Walter de
Gruyter.
Heurley, L. (1994). Traitement des textes procéduraux. Etude de psycholinguistique cognitive des proces-
sus de production et de compréhension chez des adultes non experts. Unpublished doctoral disserta-
tion. University of Bourgogne, Dijon – France.
Hoc, J.M. (1987). Psychologie cognitive de la planification. Grenoble: Presses Universitaires de Greno-
ble.
Holland, V.M. (1987). Processes involded in writing procedural instructions. Unpublished doctoral dis-
sertation. Graduate School, Pennsylvania State University, USA.
Holmes, V.M. (1984). Sentence planning in a story continuation task. Language and Speech, 27, 115-134.
Holmes, V.M. (1988). Hesitations and sentences planning. Language and Cognitive Processes, 3 (4),
323-361.
Hotopf, W.H.N. (1983). Slips of the pen and tongue: What they tell us about language production. In B.
Butterworth (Ed.), Language production. Vol. 2. New York: Academic Press.
Hudson, J.A., & Fivush, R. (1991). Planning in the pre-school years: the emergence of plans from general
event knowledge. Cognitive Development, 6, 393-415.
Hull, G.A. (1987). The editing process in writing: a performance study of more skilled and less skilled
college writers. Research in the Teaching of English, 21, 8-29.
Huttenlocher, J., & Burke, D. (1976). Why does memory span increase with age ? Cognitive Psychology,
8, 1-31.
Jackendorff, R. (1987). Consciousness and the computational mind. Cambridge, MA: MIT Press.
Jackendorff, R. (1992). Languages of the mind: essays on mental representation. Cambridge, MA: MIT
Press.
James, W. (1905). Principles of psychology (vol.1). London. Methuen.
Janssen, D., van Waes, L., & van den Bergh, H. (1996). Effects of thinking aloud on writing processes. In
C.M. Levy, & S. Ransdell (Eds.), The science of writing: Theories, methods, individual differences
and applications (pp. 233-250). Mahwah, NJ: Lawrence Erlbaum Associates.
Jeffery, G.C. (1996). Combining ideas in written text. Cognitive aspects of a writing process. Unpub-
lished doctoral dissertation. Nottingham University. England. April.
Jeffery, G.C., & Underwood, G. (1996). The role of Working Memory in the development of a writing
skill: learning to co-ordinate ideas within written text. In G. Rijlaarsdam, H. van der Bergh, & M.
Couzijn (Eds.), Studies in Writing: Vol 1 .Theories, models and methodology in writing research (pp.
268-282). Amsterdam: Amsterdam University Press.
Jonides, J., & Smith, E. E. (1997). The architecture of Working Memory. In M. D. Rugg (Ed.). Cognitive
Neuroscience (pp. 243-276). Cambridge, MA: MIT Press.
Just, M.A., & Carpenter, P.A. (1980). A theory of reading: From eye fixations to comprehension. Psycho-
logical Review, 4, 329-354.
Just, M.A., & Carpenter, P.A. (1987). The psychology of reading and language comprehension. Boston,
M.A.: Allyn and Bacon.
Just, M.A., & Carpenter, P.A. (1992). A capacity theory of comprehension: Individual differences in
Working Memory. Psychological Review, 99(1), 122-149.
Kahneman, D. (1973), Attention and effort. Englewood Cliffs, NJ: Prentice-Hall.
Kail, R. (1979). The development of memory in children. New-York: Freeman.
Kail, R. (1991). Developmental change in speed of processing during childhood and adolescence. Psy-
chological Bulletin, 109, 490-501.
Kaufer, D.S., Hayes, J.R., & Flower, L.S. (1986). Composing written sentences. Research in the Teach-
ing of English, 20(2), 121-140.
COMMENTARY 249
Kellogg, R.T. (1987a). Effects of topic knowledge on the allocation of processing time and cognitive ef-
fort to writing processes. Memory and Cognition, 15 (3), 256-266.
Kellogg, R.T. (1987b). Writing performance: effects of cognitive strategies. Written Communication, 4
(3), 269-298.
Kellogg, R.T. (1988). Attentional overload and writing performance: Effects of rough draft and outline
strategies. Journal of Experimental Psychology: Learning, Memory and Cognition,14(2), 355-365.
Kellogg, R.T. (1990). Effectiveness of prewriting strategies as a function of task demands. American
Journal of Psychology, 103(3), 327-342.
Kellogg, R.T. (1993). Observations on the psychology of thinking and writing. Composition Studies, 21,
3-41.
Kellogg, R.T. (1994). The psychology of writing. New York: Oxford University Press.
Kellogg, R.T. (1996). A model of Working Memory in writing. In C.M. Levy, & S. Ransdell (Eds.). The
science of writing: Theories, methods, individual differences and applications (pp. 57-72). Mahwah,
NJ: L. Erlbaum Associates.
Kellogg, R.T. (1998). Long-term Working Memory in text production. Paper presented at the annual
meeting of the American Psychological Society, Washington, DC. May.
Kellogg, R.T. (1999). Components of Working Memory in text production. In G. Rijlaarsdam & E Espé-
ret (Series Eds.), & M. Torrance, & G. Jeffery (Vol. Eds),. Studies in Writing: Vol 3. The cognitive
demands of writing: processing capacity and Working Memory effects in text production (pp. 43-61).
Amsterdam: Amsterdam University Press.
Kellogg, R.T. (in press). Competition for Working Memory among writing processes. American Journal
of Psychology.
Kempen, G., & Hoenkamp, E. (1982) Incremental sentence generation: implication for the structure of a
syntactic processor. In J. Horecky (Ed.). Proceedings of the Ninth International Conference on Com-
putational Linguistics. Amsterdam: North Holland.
Kempen, G., & Hoenkamp, E. (1987). An incremental procedural grammar for sentence formulation.
Cognitive science, 11, 201-258.
Kempen, G., & Huijbers, P. (1983). The lexicalization process in sentence production and naming: indi-
rect election of words. Cognition, 14, 185-209.
Kemper, S. (1987). Constrains on psychological processes in discourse production. In H.W. Dechert, &
M. Raupach (Eds), Psycholinguistic models of production (pp. 185-188). Norwood, NJ: Ablex.
Kennedy, A., & Murray, W.S. (1987). Spatial coordinate and reading: comments on Monk (1985). The
Quarterly Journal of Experimental Psychology, 39, 649-656.
Kintsch, W. (1987). Psychological processes in discourse production. In H.W. Dechert, & M. Raupach
(Eds), Psycholinguistic models of production, (pp. 163-180). Norwood, N.J.: Ablex.
Kintsch, W., & van Dijk, T.A. (1978). Toward a model of text comprehension and production. Psycholo-
gical Review, 85,363-394.
Lambert E., & Espéret E. (1997). Le début du langage écrit: les premières productions grapho-motrices.
Arobase, 2 (revue électronique; http://www.liane.net/arobase).
Lapointe, C.B., & Engle, R.W. (1990). Simple and complex word spans as measures of Working Memory
capacity. Journal of Experimental Psychology: learning, Memory and Cognition, 16, 1118-1133.
Largy, P. (1995). Production et gestion des erreurs en production écrite: le cas de l’accord sujet/verbe.
Etude chez l’adulte et l’enfant. Unpublished doctoral dissertation. University of Bourgogne, Dijon.
France
Largy, P., Fayol, M., & Lemaire, P. (1996). The homophone effect in written French: the case of
verb/noun inflection errors. Language and Cognitive Processes, 11(3), 217-255.
Lea, J., & Levy, C.M. (1999). Working memory as a resource in the writing process. In G. Rijlaarsdam &
E. Espéret (Series Eds), & M. Torrance & G. Jeffery (Eds.), Studies in Writing: Vol 3. The cognitive
demands of writing: Processing capacity and Working Memory in text production (pp. 63-82). Am-
sterdam: Amsterdam University Press.
Levelt, W.J.M. (1981). The speaker's linarization problem. Phil. Trans. Roy. Soc. London, B295, 305-
315.
Levelt, W.J.M. (1982). Cognitive styles in the use of spatial direction terms. In R. J. Jarvella, & W. Klein
(Eds.), Speech, place and action (pp. 251-267). Chichester, England: John Wiley and Sons.
Levelt, W.J.M. (1983). Monitoring and self repair in speech. Cognition, 14, 41-104.
Levelt, W.J.M. (1989). Speaking: from intention to articulation. Cambridge, London: M.I.T. Press.
250 HAYES
Levelt, W.J.M. (1992). Accessing words in speech production: stages, processes and representations.
Cognition, 42,1-22.
Levelt, W.J.M., Schriefers, H., Vorberg, D., Meyer, A.S., Pechmann, T., & Havinga, J. (1991). The time
course of lexical access in speech production: a study of picture naming. Psychological Review, 98
(1), 122-142.
Levy, C.M. (1997). The ‘R’ that psychology forgot: Research on writing processes. Behavior Research
Methods, Instruments and Computers, 29(2), 137-142.
Levy, C.M., & Marek, P. (1999). Testing components of Kellogg’s multicomponent models of Working
Memory in writing: The role of the phonological loop. In G. Rijlaarsdam & E. Espéret (Series Eds),
& M. Torrance & G. Jeffery (Vol. Eds), Studies in Writing: Vol 3. The cognitive demands of writing.
Processing capacity and Working Memory effects in text production (pp. 25-41). Amsterdam: Am-
sterdam University Press.
Levy, C.M., & Ransdell, S. (1996). Writing signatures. In C.M. Levy, & S. Ransdell (Eds.), The science
of writing: Theories, methods and applications, (pp. 149-161). Mahwah: Laurence Erlbaum Associ-
ates.
Lindsay, P.H., & Norman, D.A.(1972). Human information processing: an introduction to psychology.
New York: Academic Press.
Littleton, E. B. (1995). Emerging rhetorical knowledge for writing: The effect of audience-presence on
five- through nine-year-olds' speech. Carnegie Mellon University.
Locke, J. (1690). An essay concerning human understanding. London. Thomas Bassett.
Logan, G.D. (1988). Toward an instance theory of automatization. Psychological Review, 95(4), 492-527.
Logan, G.D., & Etherton, J.L. (1994). What is learned during automatization? The role of attention in
constructing an instance. Journal of Experimental Psychology: Learning, Memory and Cognition,
20(5), 1022-1050.
Logie, R.H. (1989). Characteristics of visual Short Term Memory. European Journal of Cognitive Psy-
chology, 1 (4), 275-284.
Logie, R.H. (1995). Visuo-spatial Working Memory. Hillsdale, N.J., Laurence Erlbaum
Logie, R.H. (1996). The seven ages of Working Memory. In Ridcharson, T.E., Engle, R.W., Hasher, L.,
R.H. Logie, Stoltzfus, E.R., & Zacks, R.T. (Eds.). Working memory and human cognition. (pp31-65).
Counterpoints: Cognition, Memory and Language.
Lounsbury, F.G. (1965). Transitional probability, linguistic structure and systems of habit-family hierar-
chies. In C.E. Osgood, & T.A. Sebeok (Eds.), Psycholinguistics: a survey of theory and research
problems (pp. 93-101). Bloomington: Indiana University Press.
Maclay, H., & Osgood, C.E. (1959). Hesitation phenomena in spontaneous speech. Word, 15, 19-44.
Mahach, K.R., Boehm-Davis, D., & Holt,R. (1995). The effect of mice and pull-down menus versus
command-driven interfaces on writing. International Journal of Human-Computer Interaction, 7(3),
213-234.
Mancuso, J. C. (1986). The acquisition and use of narrative grammar structure. In T. R. Sarbin (Ed.). Nar-
rative psychology: The storied nature of human conduct (pp. 91-110). New York: Praeger.
Mangenot, F. (1996). Les aides logicielles à l’écriture. L’ingénierie Educative. CNDP.
Martlew, M. (1983). Problems and difficulties: Cognitive and communicative aspects of writing. In M.
Martlew (Ed.), The psychology of written language (pp. 295-333). London:Wiley and Sons.
Martlew, M., & Sorsby, A. (1995). The precursors of writing: Graphic representation in pre-school chil-
dren. Learning and Instruction, 5(1), 1-20.
Matsuhashi, A. (1981). Pausing and planning: the tempo of written discourse production. Research in the
teaching of English, 15 (2), 113-134
Matsuhashi, A. (1982). Explorations in the real-time production of written discourse. In M. Nystrand
(Ed.), What writers know: the language and structure of written discourse (pp.269-289). New-York:
Academic Press.
Matsuhashi, A. (1987). Revising the plan and altering the text. In A.Matsuhashi (Ed.), Writing in real
time (pp. 197-223). Norwood, N.J.: Ablex.
McArthur, C.A., & Graham, S. (1987). Learning disabled students’ composing under three methods of
text production: Handwriting, word processing and dictation. Journal of Special Education, 21(3),
22-42.
McArthur, C.A., Graham, S., & Schwartz, S. (1991). Knowledge of revision and revising behavior among
students with learning disabilities. Learning Disability Quarterly,14, 61-73.
COMMENTARY 251
McArthur, C.A., Schwartz, S., & Graham, S. (1991). A model of writing instruction: integrating word
processing and strategy instruction into a process approach to writing. Learning Disabilities Re-
search and Practice, 6, 230-236.
McCutchen, D. (1986). Domain knowledge and linguistic knowledge in the development of writing abil-
ity. Journal of Memory and Language, 25, 431-444.
McCutchen, D. (1987). Children's discourse skill: form and modality requirements of schooled writing.
Discourse Processes, 10, 267-286
McCutchen, D. (1988). « Functional automaticity » in children’s writing: a problem of metacognitive
control. Written Communication, 5,306-324.
McCutchen, D. (1994). The magicalnumber three, plus or minus two: Working memory in writing. In
E.C.Butterfield (Eds), Advances in cognition and EducationalPractice. Vol. 2: Children’s writing:
Toward a process theory ofthe development of skilled writing, (pp. 1-30). Greenwich,CT: JAI Press.
McCutchen, D. (1996). A capacity theory of writing: Working Memory in composition. Educational Psy-
chology Review,8(3), 299-325.
McCutchen, D. (2000). Knowledge, Processing and Working Memory: implications for the theory of
writing. Educational Psychologist, 35(1), 13-23.
McCutchen, D., & Perfetti, C.A. (1982). Coherence and connectedness in the development of discourse
production. Text, 2, 113-139.
McCutchen, D., Covill, A., Hoyne, S.H., & Mildes, K. (1994). Individual differences in writing: implica-
tions of translating fluency. Journal of Educational Psychology, 86(2), 256-266.
McCutchen, D., Francis, M., & Kerr, S. (1997). Revising for meaning: effects of knowledge and strategy.
Journal of Educational Psychology, 89(4), 667-676.
McCutchen, D., Hull, G.A., & Smith, W.L. (1987). Editing strategies and error correction in basic writ-
ing. Written Communication, 4, 139-154.
McCutchen, D., Kerr, S., & Francis, M. (1994). Editing and revising: effect of knowledge of topic and er-
ror location. Paper presented at the Annual Meeting of the American Educational Research Associa-
tion. New Orleans, LA.
McDonald, J. D. (1980). Memory capacity and inferential abilities in five and seven year old children.
Unpublished master's thesis, York University, Toronto.
Meulenbroek, R.G., & van Galen, G.P. (1988). The acquisition of skill handwriting: discontinuous trends
in kinematic variables. In A. M. Colley et J. R. Beechs (Eds.). Cognition and action in skilled behav-
ior. Amsterdam: North-Holland.
Miller, G.A. (1956). The magical number seven, plus or minor two: some limits on our capacity for proc-
essing information. Psychological Review, 63, 81-97.
Miller, G.A., Galanter, E., & Pribram, K.H. (1960). Plans and structures of behavior. Londres, Holt:
Rinehart and Winston.
Miyake, A., & Shah, P. (1999). Models of Working Memory: Mechanisms of active maintenance and ex-
ecutive control. Cambridge: Cambridge University Press.
Moffett, J. (1968). Teaching the universe of discourse. Boston: Houghton Mifflin.
Monahan, B.D. (1984). Revision strategies of basic and competent writers as they write for different au-
dience. Research in the Teaching of English, 18(3), 288-304.
Negro, I., & Chanquoy, L. (1999). Subject-verb agreement errors: Phonological and semantic control in
adults. In G. Rijlaarsdam & E. Espéret (Series Eds.), & M. Torrance & G. Jeffery, & (Vol. Eds), Stu-
dies in Writing: Vol 3. The cognitive Demands of Writing: processing capacity and Working Memory
effects in text production (pp. 83-98). International Series on the Research of Learning and Instruc-
tion of Writing. Amsterdam: Amsterdam University Press.
Newell, A. (1991). Unified theories of cognition. Cambridge, MA: Cambridge University Press.
Newell, A., & Simon, H.A. (1972). Human problem-solving. Englewood Cliffs, NJ: Prentice Hall.
Norman, D.A. (1968). Toward a theory of memory and attention. Psychological review, 75, 522-536.
Norman, D.A., & Rumelhart, D.E. (1983). Studies of typing from ILK research group. In W.E. Cooper
(Ed.). Cognitive aspects skilled tapewriting (pp.45-65). New-York: Sringer-Verlag.
Norman, D.A., & Shallice, T. (1980). Attention to action: willed and automatic control of behavior. Uni-
versity of California at San Diego, CHIP Report 99.
Oliver, E.I. (1995). The writing quality of seventh, ninth and eleventh graders and college freshmen: Does
rhetorical specification in writing prompts make a difference? Research in the Teaching of English,
29(4), 422-450.
252 HAYES
Overbaugh, R.C. (1992). Word processors and writing-process software: introduction and evaluation.
Computers in Human Behavior, 8, 121-148.
Pascual-Leone, J. (1970). A mathematical model for the transition rule in Piaget’s developmental stages.
Acta Psychologica, 32, 301-345.
Pascual-Leone, J. (1987). Organismic processes for neo-piagetian theories: a dialectical causal account of
cognitive development. International Journal of Psychology, 22, 531-570.
Penningroth, S. L., & Rosenberg, S. (1995). Effects of a high information-processing load on the writing
process and the story written. Applied Psycholinguistics, 16 (2), 189-210.
Perl, S. (1979). The composing processes of unskilled College writers. Research in the Teaching of Eng-
lish, 13(4), 317-336.
Piolat, A. (1983). Localisation syntaxique des pauses et planification du discours. L'Année Psychologique,
83, 377-394.
Piolat, A. (1987). Planification et contrôle de la mise en texte. In M. Fayol (Ed.). La psychologie du lan-
gage (pp. 37-49). Dijon: Cahiers du C.R.D.P.
Piolat, A. (1988). Le retour sur le texte dans l’activité rédactionnelle précoce. European Journal of Psy-
chology of Education, 3(6), 449-459.
Piolat, A. (1990). Vers l’amélioration de la rédaction de texte. Apport des technologies nouvelles pour la
recherche et l’apprentissage. Synthèse pour l’Habilitation à Diriger des Recherches. Université de
Provence, Aix-en-Provence. France.
Piolat, A. (1998). Evaluation and Assessment of written texts. The encyclopedia of language and educa-
tion. Vol. 7. In C. Clapham (Ed.). Language Testing and Assessment. Dordrecht: Kluwer Academic
Publishers.
Piolat, A., & Olive, T. (in press). Comment étudier le coût et le déroulement de la rédaction de textes? La
méthode de la triple tâche: Un bilan méthodologique [How to study the cost and unfolding of writ-
ing? The triple task technique: A methodological assessment]. L’Année Psychologique.
Piolat, A., & Pélissier, A. (1998). Etude de la rédaction de textes: Contraintes théoriques et méthodes de
recherches. In A. Piolat, & A. Pélissier (Eds.), La rédaction de textes: Approche cognitive (pp. 225-
269). Lausanne: Delachaux et Niestlé.
Piolat, A., & Roussey, J.Y. (1991-1992). A propos de l’expression ‘stratégie de révision’ de texte en psy-
chologie cognitive. Texte en Main, 10/11, ‘Lis tes ratures’, 51-64.
Piolat, A., Roussey, J.Y., & Rous. R. (1996). Effects of attention directed to different texts on the mental
effort related to writing processes. Paper presented at the European Writing Conference, Barcelona.
Piolat, A., Roussey, J.Y., Olive, T., & Farioli, F. (1996). Charge mentale et mobilisation des processus
rédactionnels:examen de la procédure de Kellogg. In A. Tricot, & L. Chanquoy, (Eds). Numéro spé-
cial sur la charge mentale. PsychologieFrançaise, 41(4), 339-354.
Plumb, C. (1991). The effects of domain-specific knowledge and revising expertise on the number and
type of changes made during revision. Doctoral dissertation of the University of Washington, Seattle
(U.S.A.).
Plumb, C., Butterfield, E.C., Hacker, D.J., & Dunlosky, J. (1994). Error correction in text. Testing the
processing-deficit and knowledge-deficit hypotheses. Reading and Writing: An Interdisciplinary
Journal, 6, 347-360.
Pontecorvo, C., & Paoletti, G. (1991). Planning story completion in a collaborative computer task. Euro-
pean Journal of Psychology of Education, 6, 199-213.
Power, M.J. (1986). A technique for measuring processing load during speech production. Journal of
Psycholinguistic Research,15, 371-380.
Pynte, J., Kennedy, A., Murray, W.S., & Courrieu, P. (1988). The effect of spatialisation on the process-
ing of ambiguous pronominal reference. In G. Luer, U. Lass, J. Shallo-Hoffmanm (Eds.) Eye Move-
ment Research, New York: Hogrefe, 214-225.
Raaijmakers, J.G.W., & Shiffrin, R. M. (1981). Search for associative memory. Psychological Review, 88
(2), 93-134.
Ransdell, S.E., & Levy, C.M. (1999). Writing, reading and speaking memory spans and the importance of
resource flexibility. In G. Rijlaarsdam & E. Espéret (Series Eds), & M. Torrance & G. Jeffery (Vol.
Eds). Studies in Writing: Vol 3. The cognitive demands of writing: Processing capacity and Working
Memory in text production, (pp. 99-113), Amsterdam University Press.
Ransdell, S.E., & Levy, C.M. (1994). Writing as process and product: the impact of tool, genre, audience
knowledge and writer expertise. Computers in Human Behavior, 10 (4), 511-527.
COMMENTARY 253
Ransdell, S.E., & Levy, C.M. (1996a). Working memory constraints on writing quality and fluency. In
C.M. Levy, & S. Ransdell (Eds.). The science of writing: theories, methods and applications, (pp. 93-
105). Mahwah: L. Erlbaum Associates.
Ransdell, S.E., & Levy, C.M. (1996b). The effects of irrelevant speech and concurrent digit load on writ-
ing quality and fluency. Paper presented at the E.A.R.L.I. S.I.G. Writing Conference. University of
Barcelona, Spain, October 23-25, 1996.
Ratcliff, R., & Mc Koon, G. (1994). Retrieving information from memory: spreading-activation theories
versus compound-cue theories. Psychological Review, 101 (1), 177-184.
Reinhart, T. (1981). Pragmatics and linguistics: an analysis of sentence topics. Philosophica, 27, 53-94.
Reiser, B. J., & Black, J. B. (1982). Processing and structural models of comprehension. Text, 2, 30-45
Reiser, B. J., Black, J. B., & Abelson, R. P. (1985). Knowledge structures in the organization and re-
trieval of autobiographical memories. Cognitive Psychology, 17, 89-137.
Richardson, J.T.E., Engle, R.W., Hasher, L., Logie, R.H., Stoltzus, E.R., & Zacks, R.T. (1996). Working
Memory and Human Cognition. New York-Oxford. Oxford University Press.
Roulin, J.L., & Monnier, C. (1996). La mémoire de travail. In F. Eustache, & B. Chevalier (Eds.), Mé-
moire et amnésies, (pp. 65-83). Bruxelles: De Broeck.
Rubin, D.L., & Piché, G.L. (1979). Development in syntactic and strategic aspects of audience adaptation
skills in written persuasive communication. Research in the Teaching of English, 13(4), 293-316.
Rumelhart, D.E., & McClelland, J.L. (1986). Parallel Distributed Processing. Explorations in the micro-
structure of cognition. Vol. 1: Foundations. Cambridge, MA: M.I.T. Press.
Rumelhart, D.E., Lindsay, P.H., & Norman, D.A. (1972). A process for Long Term Memory. In E. Tulv-
ing, & W. Donaldson (Eds.), Organisation of memory (pp. 197-245). London: Academic Press.
Rummer, R., & Grabowski, J. (1993). Speaking, writing and Working Memory load. Paper presented at
the 5th European Conference EARLI. Aix en Provence. France. September, 1-5.
Saada-Robert, M. (1995). Effective activities in learners managing cognitive load in 2nd grade school-
writing. Paper presented for the 6th European Conferencefor Research on Learning and Instruction.
Biennial Meeting. University of Nijmegen, The Netherlands, August 26-31, 1995.
Sadowski, M., Kealy, W. A., Goetz, E. T., & Paivio, A. (1997). Concreteness and imagery effects in the
written composition of definitions. Journal of Educational Psychology, 89, 518-526.
Saito, S. (1997). When articulatory suppression does not suppress theactivity of the phonological loop.
British Journal of Psychology, 88, 565-578.
Savigny, A. (1995). L'organisation des connaissances et la production de textes procéduraux (études off-
line et on-line). Unpublished doctoral dissertation – Psychology, University of Poitiers – France.
Scardamalia, M. (1981). How children cope with the cognitive demands of writing. In C.H. Frederiksen,
& J.F. Dominic (Eds), Writing: the nature, development and teaching of written communication.Vol.
2: Writing: Process, development and communication (pp. 81-104). Hillsdale, NJ: L. ErlbaumAsso-
ciates.
Scardamalia, M., & Bereiter, C. (1982). Assimilative processes in composition planning. Educational
Psychologist, 17(3), 165-171.
Scardamalia, M., & Bereiter, C. (1983). The development of evaluative, diagnostic and remedial capabili-
ties in children’s composing. In M. Martlew (Ed.), The psychology of written lan-
guage.Developmental and educational perspectives, (pp. 67-95).New York: Wiley and Sons
Scardamalia, M., & Bereiter, C. (1985). The development of dialectical processes in composition. In D.
Olson, N. Torrance, & A. Hildyard (Eds.). Literacy, language and learning: the nature and conse-
quences of reading and writing (pp. 307-329). New-York. Cambridge University Press.
Scardamalia, M., & Bereiter, C. (1986). Research on written composition. In M. Wittrock (Ed.), Hand-
book of research on teaching, (3rd ed., pp. 778-803). New York: McMillan.
Scardamalia, M., & Bereiter, C. (1987). Knowledge telling and knowledge transforming in written com-
position. In S. Rosenberg (Ed.), Advances in applied psycholinguistics (pp. 142-175). Cambridge:
Scardamalia, M., & Bereiter, C. (1991). Litterate expertise. In K.A. Ericsson et J. Smith (Eds). Toward a
general theory of expertise. (pp.172-194). Cambridge University Press.
Schilperoord, J. (1996). It’s about time. Temporal aspects of cognitive processes in text production. Am-
sterdam, Atlanta (GA): Rodopi.
Schilperoord, J., & Chanquoy, L. (1999). Analyse inter-langue de narrations et de descriptions produites
par des enfants français et hollandais. Communication pour le Congrès National de la Société Fran-
çaise de Psychologie, Aix-en-Provence, 25-27 Mai 1999.
254 HAYES
Schilperoord, J., & Sanders, T. (1997). Pauses, cognitive rhythms and discourse structure: an empirical
study of discourse production. In W.A. Liebert, G. Redeker, & L. Waugh (Eds.). Discourse and per-
spective in cognitive linguistics. Current issues in linguistic theory. Amsterdam/Philadelphia: Benja-
mins Püblishing Compagny.
Schilperoord, J., & Sanders, T. (1999). How hierachical text structure affects retrieval processes: implica-
tions of pause and text analysis. In G. Rijlaarsdam, & E. Espéret (Series Eds), & M. Torrance, & D.
Galbraith (Vol. Eds). Studies in Writing: Vol 4. Knowing what to write: cognitive processes in the
generation, selection and development of ideas during text production. Amsterdam. Amsterdam Uni-
versity Press (pp 13-30).
Schmidt, W. (1979). Aufgaben und probleme einer funktional-kommunikativen sprachbeschreibung.
Zeitschrift für Phonetik, Sprachwissenschaft und Kommunikation, 32, 123-134 (in Schneuwly, 1986).
Schneider, W., & Pressley, M. (1989). Memory development between 2 and 20. New-York: Spinger-
Verlag.
Schneuwly, B. (1986). Activité langagière ou action langagière complexe. Discussions et recherches al-
lemandes. Bulletin de Psychologie, 38, 595-606.
Schneuwly, B. (1988). La construction du langage écrit chez l’enfant. Neuchâtel, Paris: Delachaux and
Niestlé.
Schneuwly, B. (1997). Textual organizers and text types: Ontogenetic aspects in writing. In J. Coster-
mans, & M. Fayol (Eds.), Processing interclausal relationships. Studies in the production and
comprehension of text (pp. 245-263). Mahwah, NJ: L.Erlbaum Associates.
Schraagen, J.M. (1993). How experts solve a novel problem in experimental design. Cognitive Science,
17, 285-309.
Schriver, K.A. (1987). Teaching writers to anticipate the reader's needs: Empirically-based instruction.
Unpublished doctoral dissertation, Carnegie Mellon University, Department of English, Pittsburgh,
PA.
Schriver, K.A. (1997). Dynamics in document design: Creating texts for readers. New York: John Wiley
and Sons.
Schriver, K.A., Hayes, J.R., & Steffy, A. (1996). ‘Just say no to drugs’ and other unwelcome advice: Ex-
ploring the creation and interpretation of drug education literature (Tech. Rep. ). University of Cali-
fornia at Berkeley and Carnegie Mellon University, National Center for the Study of Writing and Lit-
eracy: Berkeley, CA and Pittsburgh, PA (Also published in K. A. Schriver, 1997, Dynamics in docu-
ment design: Creating texts for readers, pp. 167-203. New York: John Wiley and Sons).
Schumacher, G.M., & Ma, C. (1999). Representations in writing: a modularity perspective. In G. Ri-
jlaarsdam and E. Espéret (Series Eds), & M. Torrance, & D. Galbraith (Vol. Eds). Studies in Writing:
Vol 4. Knowing what to write: cognitive processes in the generation, selection and development of
ideas during text production. Amsterdam. Amsterdam University Press (pp. 173-190).
Shallice, T. (1988). From neuropsychology to mental structure. Cambridge: Cambridge University Press.
Sharples, M., & Evans, M. (1991). Computer support for the development of writing abilities. Cognitive
Science Research Reports of the University of Sussex, England.
Simon, H. A. (1974). How big is a chunk ? Science, 183, 482-488.
Simon, H.A. (1995). The information-processing theory of mind. American Psychologist, 50(7), 507-508.
Smolensky, P. (1992). IA connexionniste, IA symbolique et cerveau. In D. Andler (Ed.), Introduction aux
sciences cognitives (pp.77-105). Paris: Gallimard.
Snyder, I. (1993a). Writing with word processors: a research overview. Educational Research, 35(1), 49-
68
Snyder, I. (1993b). The impact of computers on students’writing: A comparative study of the effects of
pens and word processors on writing context, process and product. Australian Journal of Education,
37(1), 5-25.
Snyder, I. (1994). Writing with word processors: The computer’sinfluence on the classroom context.
Journal of Curriculum Studies, 26(2), 143-162.
Snyder, I. (1995). Toward electronic writing classrooms: The challenge for teachers. Journal of Informa-
tion Technology forTeacher Education, 4(1), 51-65.
Snyder, M. (1986). Public appearances, private realities: the psychology of self-monitoring. New-York:
W.H. Freeman and Compagny.
Sommers, N. (1980). Revisions strategies of student writers and experienced writers. College Composi-
tion and Communication, 31, 378-387.
COMMENTARY 255
Sperber, D. (1994). The modularity of thought and the epidemiology of representation. In L. Hirschfeld S.
Gelman (Eds). Mapping the mind: domain specificity in cognition and culture (pp.39-67). New York:
Stanovich, K.E. (1990). Concepts in developpemental theories of reading: cognitive ressources, auto-
maticity and modularity. Developmental review, 10, 72-100.
Stein, N.L., & Miller, C. A. (1992). The development of memory and reasoning skills in argumentative
contexts: evaluating, explaining and generating evidence. In R. Glaser (Ed.). Advances in Instruc-
tional Psychology (pp. 285-335). Hillsdale, N.J.: Erlbaum.
Swanson, H.L., & Berninger, V.W. (1994). Working memory as a source of individual differences in
children’s writing. In E.C. Butterfield (Ed.). Advances in cognition and Educational Practice.Vol. 2:
Children’s writing: toward a process theory of the development of skilled writing (pp. 31-56).
Greenwich,CT: JAI Press.
Swanson, H.L., & Berninger, V.W. (1996). Individual differences in children’s Working Memory and
writing skills. Journal of Experimental Child Psychology, 63, 358-385.
Swarts, H., Flower, J.S., & Hayes, J.R. (1984). Designing protocol studies of the writing process: an in-
troduction. In R. Beach, & L. S. Bridwell (Eds.). New directions in composition process (pp. 53-71).
New-York: Guilford Press.
Tannenbaum, P.H., Williams, F., & Wood, B.S. (1967). Hesitation phenomena and related encoding
characteristics in speech and typewriting. Language and Speech, 10, 203-215.
Temple, C., Nathan, R., Temple, F., & Burris, N.A. (1993). The beginnings of writing. Boston: Allyn and
Bacon. (3rd Edition).
Templeton, S., & Bear, D.R. (1992). (Eds). Development of orthographic knowledge and the foundations
of literacy. Hillsdale, NJ: L.E.A.
Tetroe, J. (1984). Information processing of plot construction. Paper presented at the meeting of the
American Educational Research Association. New Orleans.
Torrance, M. (1996). Is writing expertise like other kinds of expertise ? In G. Rijlaarsdam, H. van der
Bergh, & M. Couzijn (Eds.), Studies in Writing: Vol 1. Theories, models and methodology in writing
research (pp. 3-9). Amsterdam: Amsterdam University Press.
Torrance, M., & Jeffery, G.C. (1999). Writing processes and cognitive demands. In G. Rijlaarsdam & E.
Espéret (Series eds), & M. Torrance & G. Jeffery (Vol. Eds.). Studies in Writing: Vol 3. The cogni-
tive demands of writing. Amsterdam: Amsterdam University Press. (pp. 1-12)
Torrance, M., Thomas, G.V., & Robinson, E.J. (1996). Finding something to write about: strategic and
automatic processes in idea generation. In C. M. Levy, & S. Ransdell (Eds.), The science of Writing:
theory: methods, individual differences and applications (pp. 199-205). Mahwah, N.J.: Lawrence Er-
baum Associates.
Traxler, M.J., & Gernsbacher, M.A. (1992). Improving written communication through minimal feed-
back. Language and Cognitive Processes, 7, 1-22.
Traxler, M.J., & Gernsbacher, M.A. (1993). Improving written communication through perspective tak-
ing. Language and Cognitive Processes, 8 (3), 311-334.
Turner, M.L., & Engle, R.W. (1989). Is Working Memory capacity task dependent ? Journal of Memory
and Language, 28, 27-54.
van den Bergh, H., & Rijlaarsdam, G. (1999). The dynamics of idea generation during writing: an on-line
study. In G. Rijlaarsdam, & E. Espéret (Series Eds), & M. Torrance, & D. Galbraith (Vol. Eds), Stud-
ies in Writing: Vol 4.. Knowing what to write: cognitive processes in the generation, selection and
development of ideas during text production. Amsterdam. Amsterdam University Press (pp. 99-120).
van den Bergh, H., Rijlaarsdam, G., & Breetvelt, I. (1993). Revision process and text quality: an empiri-
cal study. In G. Eigler, & T. Jechle (Eds). Writing: Current trends in European research (pp. 133-
147). Freiburg: Hochschul Verlag.
van den Plaats, R.E., & van Galen, G.P. (1990). Effects of spatial and motor demands in handwriting.
Journal of motor behavior, 22, 361-380.
van der Pool, E.M.C. (1995). Writing as a conceptual process: a text-analytical study of developmental
aspects. Nederlands: Els van Der Pool.
van der Pool, E.M.C., & van Wijk, C. (1995). Process and strategy in a psycholinguistic model of writing
and reading. Tijdschrift voor Onderwijsresearch, 20, 200-214.
van Dijk, T.A., & Kintsch, W. (1983). Strategies of discourse comprehension. New York: Academic
Press.
256 HAYES
van Galen, G.P. (1990). Phonological and motoric demands in handwriting: evidence for discrete trans-
mission of information. Acta Psychologica, 74, 259-275.
van Galen, G.P. (1991). Handwriting: a developmental perspective. In A.F. Kalverboer, R.H. Geuze, & B.
Hopkins (Eds.), Motor development in early and later childhood: Longitudinal approaches (pp. 143-
172). Cambridge: Cambridge University Press.
van Galen, G.P., & Teulings, H.L. (1983). The independent monitoring of form and scale factors in
handwriting. Acta psychologica, 54, 9-22.
van Wijk, C. (1999). Conceptual processes in argumentation: a developmental perspective. In G. Ri-
jlaarsdam, & E. Espéret (Series Eds), & M. Torrance, & D. Galbraith (Vol. Eds), Studies in Writing:
Vol 4. Knowing what to write: conceptual processes in text production (pp.31-50). Amsterdam. Am-
sterdam University Press.
Veerman, A. (2000). Computer-supported collaborative learning through argumentation. Utrecht: Ruiter.
Voss, J.F., Vesonder, G.T., & Spilich, G.T. (1980). Text generation and recall by high-knowledge and
low-knowledge individuals. Journal of Verbal Learning and Verbal Behavior, 17, 651-667.
Walker, W.H., & Kintsch, W. (1985). Automatic and strategic aspects of knowledge retrieval. Cognitive
Science, 9, 261-283.
Wallace, D., & Hayes, J. R. (1991). Redefining revision for freshmen. Research in the Teaching of Eng-
lish, 25(1), 54-66.
Wallace, D., Hayes, J.R., Hatch, J.A., Miller, W., Moser, G., & Silk, C.M. (1996). Better revision in eight
minutes? Prompting first-year college writers to revise globally. Journal of Educational Psychology,
88(4), 682-688.
Wason, P.C. (1980). Specific thoughts on the writing process. In L. W. Gregg, & E. R. Steinberg (Eds.),
Cognitive processes in writing (pp. 129-137). Hillsdale, N.J.: Erlbaum.
Waters, H.S., & Lomenick, T. (1983). Levels of organisation in descriptive passages: production, com-
prehension and recall. Journal of Experimental Child Psychology, 35, 391-408.
Waugh, N.C., & Norman, D.A. (1965). Primary memory. Psychological review, 72, 89-104.
Weck, G. de,, & Schneuwly, B. (1994). Anaphoric procedures in four texttypes written by children. Dis-
course Processes, 17,465-477.
Whisbow (1988). Studies of creativity in poets. Unpublished doctoral dissertation. Carnegie Mellon Uni-
versity. Pittsburgh – USA.
Williams, N. (1992). A hypertext open learning system for writers. Instructional Science, 21, 125-138.
Yuille, N., Oakhill, J., & Parkin, A. (1989). Working memory, comprehension ability and the resolution
of text anomaly. British Journal of Psychology, 80, 351-361.
Zesiger, P. (1995). Ecrire. Approches cognitive, neuropsychologique et développementale. Paris: Presses
Universitaires de France.
Zimmerman, B.J., & Risemberg, R. (1997). Becoming a self-regulated writer: a social cognitive perspec-
tive. Contemporary Educational Psychology, 22, 73-101.
INDEX BY REFERENCE
Abbott and Berninger (1993), 224 Benton, Kraft, Glover and Plake (1984),
Akiguet and Piolat (1996), 109 217; 220; 227
Akiguet, Roussey and Piolat (1993), 109 Bereiter (1980), 64; 218
Akyürek (1992), 45 Bereiter and Scardamalia (1982), 109
Alamargot (1997), 58; 62; 189 Bereiter and Scardamalia (1983), 122
Alamargot (2000), 189 Bereiter and Scardamalia (1984), 169
Alamargot, Espéret and Savigny (1993), Bereiter and Scardamalia (1986), 120
59 Bereiter and Scardamalia (1987), 14; 18;
Alamargot, Favart and Galbraith (2000), 21; 22; 33; 34; 35; 36; 39; 40; 49; 50;
55 51; 53; 54; 55; 56; 59; 61; 62; 63; 64;
Alamargot, Favart, Coirier, Passerault and 69; 70; 90; 101; 109; 142; 143; 146;
Andriessen (1999), 75 147; 149; 152; 169; 187; 188; 199;
Anderson (1972), 182 200; 202; 207; 208; 210; 211; 214;
Anderson (1974), 144 216; 217; 218; 222; 227; 236; 243;
Anderson (1976), 219 244; 249
Anderson (1983a), 143; 144; 145; 153; Bereiter and Scardamalia (1988), 105;
171; 183; 185; 187; 189; 223; 232 200
Anderson (1983b), 40; 50; 53; 152; 171; Bereiter, Burtis and Scardamalia (1988),
191 13; 53; 59; 60; 61; 63; 84; 200; 212;
Anderson (1990), 144; 146; 223; 232 214
Anderson (1993), 144; 183; 223; 232 Berg (1986), 153
Anderson and Bower (1973), 93; 144; Berninger (1994), 101; 205
146; 182 Berninger and Füller (1992), 206
Anderson, Reder and Lebiere (1996), Berninger and Swanson (1994), 40; 81;
185; 186; 189; 196 101; 109; 132; 155; 156; 163; 167;
Andriessen, Coirier, Chanquoy and de 200; 202; 204; 205; 207; 209; 211;
Bernardi (1997), 109 216; 217; 220; 223; 225
Andriessen, Coirier, Roos, Passerault and Berninger, Cartwright, Yates, Swanson
Bert-Erboul (1996), 109 and Abbott (1994), 207
Atkinson and Shiffrin (1968), 171; 182 Berninger, Füller and Whitaker (1996),
Baangert-Drowns (1993), 231 101; 205; 210
Baddeley (1986), 7; 27; 32; 39; 150; 164; Berninger, Vaughan, Graham, Abbott,
165; 171; 172; 173; 174; 176; 179; Abbott, Rogan, Brooks and Reed
180; 181; 182; 185; 190; 191; 194; (1997), 223
196; 197; 217 Berninger, Whitaker, Feng, Swanson and
Baddeley (1990), 171; 173; 185 Abbott (1996), 101
Baddeley (1996), 173 Berninger, Whitaker, Feng, Swanson and
Baddeley (2000), 174 Abbott (1996), 162
Baddeley and Hitch (1974), 173; 182 Berninger, Yates, Cartwright, Rutberg,
Baddeley and Wilson (1985), 239 Remy and Abbott (1992), 207
Baker-Sennett, Matusov and Rogoff Best (1989), 16
(1993), 210 Bestgen and Costermans (1994), 108
Barritt and Kroll (1978), 45 Bestgen and Costermans (1997), 108
Bartlett (1982), 111; 120; 127; 128 Bialystok (1986), 199
Baudet (1988), 53 Bock (1982), 45; 90; 226
Beal (1990), 132 Bock (1995), 37; 90; 92; 97; 98; 99
Beal (1993), 134 Bock (1996), 36; 37; 92; 97; 98; 110; 164
Beal (1996), 112; 132; 133; 194 Bock and Cutting (1992), 163
Beason (1993), 127 Bock and Eberhard (1993), 163
258
Bock and Levelt (1994), 36; 92; 95; 96; Chanquoy (1991), 108
97; 98; 99; 107; 151; 235; 239 Chanquoy (1997), 112; 121; 131; 200
Bock and Warren (1985), 97 Chanquoy (1998), 76; 108; 113; 178; 196;
Boniface and Pimet (1992), 74 200
Boomer (1965), 105 Chanquoy (2001), 112; 130; 194
Boscolo (1995), 230; 232 Chanquoy and Fayol (1991), 230
Bourdin and Fayol (1994), 207; 224; 237 Chanquoy and Fayol (1995), 62; 108; 230
Bourdin and Fayol (1996), 207; 238 Chanquoy, Foulin and Fayol (1990), 59;
Bower (1970), 146 105; 106; 143; 176; 199
Bracewell (1983), 128; 140 Chanquoy, Foulin and Fayol (1996), 105;
Bracewell, Scardamalia and Bereiter 143; 199
(1978), 66 Charness (1991), 226
Bradfort DeCosta (1992), 231 Chase and Ericsson (1981), 226
Brandt (1992), 28 Chase and Simon (1973), 226
Breetvelt, van den Bergh and Rijlaarsdam Chi (1976), 217
(1994), 133; 139 Chi (1977), 220; 227
Bridwell (1980), 114; 118; 120; 121; 130 Chi (1978), 226
Britton (1970), 243 Chin (1994), 206
Britton, Burgess, Martin, McLeod and Chomsky (1980), 148
Rosen (1975), 241 Clark and Clark (1977), 58; 105
Bronckart (1977), 70 Coirier (1996), 109; 199
Bronckart, Bain, Schneuwly, Davaud and Coirier (1997), 109
Pasquier (1985), 66; 70; 73 Coirier and Golder (1993), 109
Brown, McDonald, Brown and Carr Coirier, Andriessen and Chanquoy
(1988), 32; 100; 172; 174 (2000), 73; 109
Bruce, Collins, Rubin and Genter (1979), Coirier, Gaonac’h and Passerault (1996),
140 41
Brügelmann (1996), 229 Collins and Gentner (1980), 114; 130
Brügelmann (1997), 229 Collins and Loftus (1975), 53
Bryant and Bradley (1983), 229 Collins and Quillian (1969), 146
Burtis, Bereiter, Scardamalia and Tetroe Cooper and Matsuhashi (1983), 48
(1983), 61; 75; 210; 213 Cooper and Paccia-Cooper (1980), 105
Butterfield, Albertson and Johnston Cossu and Marshall (1990), 225
(1995), 125 Costermans and Bestgen (1991), 108
Butterfield, Hacker and Albertson (1996), Costermans and Fayol (1997), 107; 235
124; 126; 128; 156; 163; 194; 244 Cowan (1988), 184
Butterfield, Hacker and Plumb (1994), Cowan (1993), 40; 171; 184
127 Craik and Lockhart (1972), 171
Butterworth (1980a), 105 Daiute (1984), 223
Butterworth (1980b), 106 Daiute (1992), 231
Caccamise (1987), 48; 58; 65; 74; 109; Daiute and Kruidenier (1985), 131
139; 230 Daneman and Carpenter (1980), 176;
Cameron, Edmunds, Wigmore, Hunt and 181; 195; 228
Linton (1997), 127; 131 Daneman and Carpenter (1983), 228
Cannella (1988), 200 Daneman and Green (1986), 28; 195
Cantor and Engle (1993), 183; 190 Daneman and Stainton (1993), 127
Caramazza (1991), 235 Dansac and Alamargot (1999), 59; 62; 84;
Carey, Flower, Hayes, Schriver and Haas 104
(1989), 75; 214 de Beaugrande (1984), 141; 147; 204
Case (1985), 220; 221; 227 De Beaugrande (1984), 147
Case, Kurland and Goldberg (1982), 195; de Weck and Schneuwly (1994), 230
227
INDEX BY REFERENCE 259
Dell (1986), 36; 92; 93; 94; 95; 98; 99; Flower, Schriver, Carey, Haas and Hayes
110; 141; 146; 147; 187; 189 (1989), 40; 59; 202; 210; 214; 215;
Dempster (1981), 219; 227 216; 226
Dobbie and Askov (1995), 102 Fodor (1983), 37; 93; 141; 148; 149; 225;
Egan and Schwartz (1979), 226 237
Ehrich and Koster (1983), 58; 107 Fodor (1985), 148; 149
Ehrlich and Delafoy (1990), 173 Fodor, Bever and Garrett (1974), 105
Eigler, Jechle, Merziger and Winter Ford and Holmes (1978), 105
(1991), 51; 53 Foulin (1993), 62
Engelhard, Gordon and Gabrielson Foulin (1995), 62
(1992), 206 Foulin and Fayol (1988), 106
Ericsson (1985), 192 Foulin, Chanquoy and Fayol (1989), 108
Ericsson and Kintsch (1995), 171; 191; Freedman (1986), 111
192; 193; 197; 226 Friederici, Shoenle and Goodglass
Ericsson and Lehmann (1996), 199; 230 (1981), 239
Ericsson and Simon (1978), 235 Galbraith (1992), 54
Espéret (1989), 46 Galbraith (1996), 74; 146
Espéret and Piolat (1991), 140 Galbraith (1999), 36; 51; 54; 55; 56; 74;
Faigley and Witte (1981), 112; 113; 114; 85; 147; 188; 190; 191; 234
115; 121; 130; 134 Gaonac’h and Larigauderie (2000), 182
Faigley and Witte (1984), 113; 114; 115; Garrett (1975), 22; 92; 105; 110
121; 130 Garrett (1980), 22; 36; 92; 93; 95; 105;
Faigley, Cherry, Jollifre and Skinner 110; 141; 164; 187
(1985), 114; 130 Garrett (1982), 22; 37; 95; 164
Farnham-Diggory and Gregg (1975), 226 Garrett (1984), 22; 92; 93; 99; 110
Favart and Chanquoy (1999), 108 Garrett (1988), 22; 95; 160
Favart and Passerault (1996), 108 Gathercole and Baddeley (1993), 164;
Fayol (1985), 57; 63 173
Fayol (1986), 108 Gernsbacher (1989), 190
Fayol (1991), 232 Gernsbacher and Faust (1991), 190
Fayol (1994), 140; 232 Glynn, Britton, Muth and Dogan (1982),
Fayol (1997), 41; 74; 108; 230 28; 169; 223
Fayol (1999), 101; 102; 236 Golder (1996), 109
Fayol and Gombert (1987), 115; 129; 200 Goldman-Eisler (1968), 105
Fayol and Schneuwly (1987), 90 Goldman-Eisler (1972), 105
Fayol, Largy and Lemaire (1994), 237 Gombert and Roussey (1993), 109
Fitzgerald (1987), 112; 114; 128; 130; Gould (1980), 45
134 Grabowski (1996), 91; 92; 164; 165; 166;
Fitzgerald and Markham (1987), 132 190; 204
Flower and Hayes (1980), 34; 39; 46; 66; Graham (1990), 101; 224
142; 155; 161; 162; 164; 167; 169; Graham and Harris (1994), 140
197; 214 Graham and Harris (1996), 169
Flower and Hayes (1981a), 210 Graham and Harris (1997), 231
Flower and Hayes (1981b), 46; 47; 75; Graham and Weintraub (1996), 102
210 Graham, Beringer, Abbott, Abbott and
Flower and Hayes (1981c), 114; 115 Whitaker (1997), 237
Flower and Hayes (1983), 139 Graves (1975), 114; 130
Flower, Hayes, Carey, Schriver and Gregg, Sigalas, Hoy, Wisenbacker and
Stratman (1986), 115; 118; 122; 123; McKinley (1996), 28
124; 126; 127; 128; 129; 132 Grosjean, Grosjean and Lane (1979), 105
Hacker (1994), 126; 131; 132; 194
Hacker (1997), 126; 128; 156
260
Hacker, Plumb, Butterfield, Quathamer Jeffery (1996), 170; 177

and Heineken (1994), 117; 127; 128; Jeffery and Underwood (1996), 170; 217
129; 132; 194 Jonides and Smith (1997), 235
Hague and Mason (1986), 200 Just and Carpenter (1980), 122; 172
Hartley (1991), 27; 201 Just and Carpenter (1987), 172
Hatch, Hill and Hayes (1993), 247; 248 Just and Carpenter (1992), 171; 181; 225;
Hayes (1985), 231 228
Hayes (1989), 6 Kahneman (1973), 240
Hayes (1990), 28 Kail (1979), 220
Hayes (1996), 27; 28; 30; 31; 33; 34; 36; Kail (1991), 220
38; 39; 41; 48; 51; 54; 55; 56; 57; 61; Kaufer, Hayes and Flower (1986), 45;
64; 73; 90; 101; 102; 104; 106; 121; 104; 114; 130
122; 123; 126; 127; 131; 132; 134; Kellogg (1987a), 48; 58; 63; 109; 127;
135; 140; 143; 156; 157; 158; 161; 170; 193; 227; 230; 241
166; 172; 173; 174; 179; 193; 194; Kellogg (1987b), 28; 63; 101; 149; 231
197; 207; 231; 238; 245 Kellogg (1988), 59; 63; 149; 240
Hayes and Flower (1980), 7; 8; 10; 11; Kellogg (1990), 59; 63; 200; 231
16; 22; 25; 27; 33; 34; 35; 36; 38; 41; Kellogg (1993), 28; 200
47; 48; 49; 50; 51; 52; 55; 56; 57; 59; Kellogg (1994), 28; 170; 193; 234; 235;
60; 61; 64; 67; 73; 75; 76; 79; 80; 81; 237; 240
82; 83; 84; 85; 86; 87; 90; 101; 103; Kellogg (1996), 31; 32; 33; 34; 36; 37;
104; 106; 109; 111; 114; 115; 116; 38; 39; 45; 82; 90; 101; 102; 132; 140;
117; 129; 130; 139; 140; 142; 143; 151; 161; 172; 174; 175; 176; 177;
152; 154; 156; 162; 163; 166; 170; 178; 179; 180; 181; 193; 194; 196;
172; 189; 194; 197; 205; 206; 207; 197; 198; 234; 238
211; 213; 216; 248; 249 Kellogg (1998), 234; 241
Hayes and Flower (1983), 114; 116; 129 Kellogg (1999), 179; 191; 192; 193; 226
Hayes and Flower (1986), 48; 73; 112; Kellogg (in press), 238; 241
120; 121; 134 Kempen and Hoenkamp (1982), 151
Hayes and Nash (1996), 45; 46; 47; 48; Kempen and Hoenkamp (1987), 151
75; 76 Kempen and Huijbers (1983)., 91
Hayes, Flower, Schriver, Stratman and Kemper (1987), 27
Carey (1987), 112; 118; 119; 120; Kennedy and Murray (1987), 106
121; 122; 123; 125; 126; 127; 132 Kintsch (1987), 27; 36; 55
Hayes, Hatch and Silk (2000), 250 Kintsch and van Dijk (1978), 63; 72
Hayes, Schriver, Spilka and Blaustein Lambert and Espéret (1997), 102
(1986), 246 Lapointe and Engle (1990), 184
Hermann and Grabowski (1995), 166 Largy (1995), 161
Heurley (1994), 58 Largy, Fayol and Lemaire (1996), 163
Hoc (1987), 45 Lea and Levy (1999), 133; 178; 179; 238;
Holland (1987), 58; 65 239
Holmes (1984), 106 Levelt (1981), 88
Holmes (1988), 106 Levelt (1982), 88
Hotopf (1983), 99 Levelt (1983), 105
Hudson and Fivush (1991), 210 Levelt (1989), 22; 24; 25; 26; 33; 34; 35;
Hull (1987), 121 36; 38; 39; 57; 69; 70; 73; 76; 84; 87;
Huttenlocher and Burke (1976), 220 90; 91; 95; 97; 98; 104; 105; 140; 141;
Jackendorff (1987), 97 146; 148; 149; 151; 159; 160; 161;
Jackendorff (1992), 149 163; 164; 236
James (1905), 171 Levelt (1992), 91
Janssen, van Waes and van den Bergh Levelt, Schriefers, Vorberg, Meyer,
(1996), 235 Pechmann and Havinga (1991), 91
INDEX BY REFERENCE 261
Levy (1997), 34; 176 Negro and Chanquoy (1999), 161; 163
Levy and Marek (1999), 133; 178; 239 Newell (1991), 226
Levy and Ransdell (1996), 176; 196 Newell and Simon (1972), 11; 12
Lindsay and Norman (1972), 93; 182 Norman (1968), 182
Littleton (1995), 244 Norman and Rumelhart (1983), 102
Locke (1690), 171 Norman and Shallice (1980), 173
Logan (1988), 223 Oliver (1995), 28; 206
Logan and Etherton (1994), 223 Overbaugh (1992), 231
Logie (1989), 220 Pascual-Leone (1970), 220; 227
Logie (1995), 173; 182 Pascual-Leone (1987), 220; 221; 227
Logie (1996), 171; 195 Penningroth and Rosenberg (1995), 169;
Lounsbury (1965), 105 170
Maclay and Osgood (1959), 105 Perl (1979), 114; 120; 130; 131
Mahach, Boehm-Davis and Holt (1995), Piolat (1983), 106
231 Piolat (1987), 45
Mancuso (1986), 240 Piolat (1988), 114; 130; 134
Mangenot (1996), 231 Piolat (1990), 114; 122; 130; 200
Martlew (1983), 48; 200; 201; 202; 203; Piolat (1998), 193
204; 205; 209; 216; 223 Piolat and Olive (in press), 240
Martlew and Sorsby (1995), 207 Piolat and Pélissier (1998), 62; 76
Matsuhashi (1981), 63 Piolat and Roussey (1991-1992), 114
Matsuhashi (1982), 64 Piolat, Roussey and Rous (1996), 241
Matsuhashi (1987), 114; 121; 130 Piolat, Roussey, Olive and Farioli (1996),
McArthur and Graham (1987), 127 193; 237; 240
McArthur, Graham and Schwartz (1991), Plumb (1991), 117
127 Plumb, Butterfield, Hacker and Dunlosky
McArthur, Schwartz and Graham (1991), (1994), 117; 125; 126; 127; 128; 129
231 Pontecorvo and Paoletti (1991), 74
McCutchen (1986), 211 Power (1986), 143; 176
McCutchen (1987), 63; 207; 211 Pynte, Kennedy, Murray and Courrieu
McCutchen (1988), 200; 211; 236 (1988), 106
McCutchen (1994), 132; 181; 194; 211; Raaijmakers and Shiffrin (1981), 49
224; 225 Ransdell and Levy (1994), 48; 65
McCutchen (1996), 112; 132; 133; 134; Ransdell and Levy (1996a), 176; 177;
170; 181; 194; 201; 204; 208; 217; 178; 195
223; 225; 227; 229; 234; 237 Ransdell and Levy (1996b), 177
McCutchen (2000), 227; 229 Ransdell and Levy (1999), 176
McCutchen and Perfetti (1982), 218; 219 Ratcliff and McKoon (1994), 53
McCutchen, Covill, Hoyne and Mildes Reinhart (1981), 58
(1994), 170; 181; 195; 224 Reiser and Black (1982), 53
McCutchen, Francis and Kerr (1997), Reiser, Black and Abelson (1985), 53
112; 127; 131 Richardson, Engle, Hasher, Logie,
McCutchen, Hull and Smith (1987), 112 Stoltzfus and Zacks (1996), 182
McCutchen, Kerr and Francis (1994), 134 Roulin and Monnier (1996), 220
McDonald (1980), 218; 227 Rubin and Piché (1979), 48
Meulenbroek and van Galen (1988), 102 Rumelhart and McClelland (1986), 144
Miller (1956), 170; 185; 226 Rumelhart, Lindsay and Norman (1972),
Miller, Galanter and Pribram (1960), 45; 146
142; 182 Rummer and Grabowski (1993), 106
Miyake and Shah (1999), 233; 242 Saada-Robert (1995), 156
Moffett (1968), 243 Sadowski, Kealy, Goetz and Paivio
Monahan (1984), 112 (1997), 238
262
Saito (1997), 173 Temple, Nathan, Temple and Burris

Savigny (1995), 58 (1993), 114
Scardamalia (1981), 169 Templeton and Bear (1992), 207
Scardamalia and Bereiter (1982), 200; Tetroe (1984), 169
230 Torrance (1996), 199
Scardamalia and Bereiter (1983), 111; Torrance and Jeffery (1999), 172
112; 116; 117; 118; 119; 120; 121; Torrance, Thomas and Robinson (1996),
123; 129; 130; 131; 133; 161; 245 51; 52
Scardamalia and Bereiter (1985), 45; 56; Traxler and Gernsbacher (1992), 65; 73;
63; 116; 117; 118; 119; 123 74; 127; 201
Scardamalia and Bereiter (1986), 73; 115; Traxler and Gernsbacher (1993), 65; 127;
129; 200 201
Scardamalia and Bereiter (1987), 61; 63; Turner and Engle (1989), 195
200; 213; 216 van den Bergh and Rijlaarsdam (1999),
Scardamalia and Bereiter (1991), 54; 56; 75; 146; 157; 158; 187; 191; 234
69; 149; 199; 211; 212 Van den Bergh and Rijlaarsdam (1999),
Schilperoord (1996), 146 159
Schilperoord and Chanquoy (1999), 62 van den Bergh, Rijlaarsdam and Breetvelt
Schilperoord and Sanders (1997), 62 (1993), 114; 131; 133
Schilperoord and Sanders (1999), 104 van den Plaats and van Galen (1990), 102
Schmidt (1979), 52 van der Bergh and Rijlaarsdam (1999),
Schneider and Pressley (1989), 227 187; 188
Schneuwly (1986), 52 van der Pool (1995), 25; 87; 91; 161; 218
Schneuwly (1988), 63; 108; 230 van der Pool and van Wijk (1995), 26
Schneuwly (1997), 214 van Dijk and Kintsch (1983), 51; 52; 59;
Schraagen (1993), 230 64; 67; 68; 69; 70; 73; 84; 85; 86; 87;
Schriver (1987), 246 88; 99; 103; 104; 106; 142; 143; 153;
Schriver (1997), 248 154; 169; 172; 197; 218
Schriver, Hayes and Steffy (1996), 248 Van Dijk and Kintsch (1983), 52; 67; 153
Schumacher and Ma (1999), 150; 237 van Galen (1990), 36; 102
Shallice (1988), 235 van Galen (1991), 102; 223
Sharples and Evens (1991), 231 van Galen and Teulings (1983), 102
Simon (1974), 170 van Wijk (1999), 25; 26; 35; 62; 76; 87;
Simon (1995), 199; 230 161; 218; 219
Smolensky (1992), 144 Veerman (2000), 74
Snyder (1986), 74 Voss, Vesonder and Spilich (1980), 48;
Snyder (1993a), 231 58; 193; 241
Snyder (1993b), 231 Walker and Kintsch (1985), 53
Snyder (1994), 231 Wallace and Hayes (1991), 244; 245
Snyder (1995), 231 Wallace, Hayes, Hatch, Miller, Moser and
Sommers (1980), 112; 120; 121; 134 Silk (1996), 244; 245
Sperber (1994), 150 Wason (1980), 154
Stanovitch (1990), 150 Waters and Lomenick (1983), 213
Stein and Miller (1992), 217 Waugh and Norman (1965), 171
Swanson and Berninger (1994), 132; 194; Whisbow (1988), 231
206 Williams (1992), 231
Swanson and Berninger (1996), 40; 101; Yau and Bereiter (1978), 218; 227
205; 228; 229 Yuille, Oakhill and Parkin (1989), 195
Swarts, Flower and Hayes (1984), 139 Zesiger (1995), 102
Tannenbaum, Williams and Wood Zimmermann and Risemberg (1997), 231
(1967), 105
SUBJECT INDEX
ACT*, 132; 133; 134; 141; 171; 175; 163; 165; 170; 173; 181; 182; 183;
177; 207 184; 185; 187; 189; 192; 202; 209;
activation 210; 213; 214; 215; 216; 217; 218;
potential, 38; 130; 178; 179; 184 222; 223; 226; 228; 229; 232; 240;
spreading, 82 242; 243; 245; 249; 250; 251; 252;
activation, 2; 3; 4; 5; 25; 26; 27; 38; 39; 253; 256
41; 44; 45; 46; 50; 51; 71; 81; 82; 83; central control, 224
87; 98; 116; 128; 129; 130; 131; 132; central executive, 20; 121; 153; 222; 225;
134; 135; 146; 150; 152; 169; 170; 226; 227; 228; 240
171; 172; 173; 174; 175; 176; 177; child, 62; 191; 192; 209; 211; 232
178; 179; 180; 181; 184; 185; 205; clause, 36; 59; 93; 94; 95; 98; 100; 117;
206; 209; 222; 227; 239; 242; 244; 139; 192
253 cognition, 15; 16; 220; 223; 230; 239;
ACT-R, 132; 171; 220 241; 242; 246; 247; 248; 250; 251;
addressee, 3; 15; 17; 35; 36; 43; 45; 53; 252; 253; 255
54; 55; 56; 57; 58; 59; 60; 61; 62; 77; cognitive capacity, 21; 209; 232
108; 110; 113; 114; 115; 117; 152; cognitive load, 94; 121; 150; 158; 163;
154; 188; 189; 194; 203; 213; 230 253
adult, 46; 59; 62; 90; 165; 189; 198; 199; cognitive model, 54; 58; 60; 61; 129; 135;
201; 202; 211; 217; 234; 236; 247 224; 246
affect, 99; 127; 216; 229; 248 cognitive resources, 1; 15; 33; 49; 50; 51;
allocation, 15; 86; 130; 131; 136; 159; 62; 65; 87; 89; 93; 108; 110; 111; 113;
173; 212; 227; 229; 249 119; 120; 122; 130; 131; 136; 137;
allographic, 90 139; 152; 155; 159; 160; 163; 165;
argument, 136; 138; 165; 222; 224; 228; 168; 169; 173; 177; 184; 195; 203;
236; 239; 256 209; 210; 211; 212; 213; 214; 215;
argumentative, 8; 51; 61; 77; 97; 165; 216
187; 200; 205; 206; 211; 239; 243; coherence, 1; 8; 17; 36; 40; 44; 47; 49;
247; 255 50; 51; 52; 67; 68; 80; 91; 92; 94; 95;
assignment, 7 96; 141; 142; 148; 155; 158; 177; 197;
a-symbolic, 130; 131; 132; 134; 135 198; 206; 212; 245
attention, 29; 60; 121; 163; 172; 182; communicative goal, 3; 5; 6; 11; 16; 37;
192; 197; 211; 213; 222; 223; 224; 40; 47; 53; 54; 55; 57; 73; 85; 99; 128;
226; 230; 244; 247; 250; 252; 253 148
automatisation, 27; 122; 133; 169; 189; component, 1; 7; 8; 9; 10; 12; 13; 15; 16;
195; 197; 210; 211; 212; 213; 214; 17; 18; 19; 20; 23; 25; 52; 53; 55; 56;
215; 216 73; 85; 86; 89; 110; 114; 128; 136;
auxiliary system, 79 139; 148; 149; 162; 163; 166; 167;
behaviour, 58; 78; 106; 187 168; 169; 172; 176; 177; 179; 181;
bottom-up, 82; 94; 96; 130; 135; 141; 154 184; 188; 190; 191; 192; 195; 214;
buffer, 20; 90; 148; 149; 162; 208; 240 215; 240; 247
C.D.O. procedure, 104; 105; 106; 110 comprehension
capacity, 2; 3; 21; 26; 98; 120; 122; 137; activity, 17; 25; 43; 44; 52; 150; 178
139; 152; 153; 157; 158; 160; 162;
264
comprehension, 2; 13; 14; 17; 25; 28; 40; development, 6; 7; 11; 20; 27; 29; 47; 50;
43; 44; 49; 52; 53; 73; 86; 94; 108; 74; 89; 121; 131; 133; 169; 181; 185;
110; 116; 119; 127; 128; 129; 136; 186; 187; 189; 190; 192; 193; 195;
140; 148; 149; 150; 151; 152; 155; 197; 198; 199; 202; 204; 205; 207;
156; 159; 160; 164; 168; 178; 183; 208; 209; 210; 211; 213; 214; 215;
213; 216; 217; 222; 225; 235; 240; 216; 219; 220; 222; 231; 236; 238;
241; 244; 247; 249; 250; 253; 254; 240; 241; 242; 243; 244; 246; 247;
256; 257 249; 251; 252; 254; 255; 256
computational, 129; 130; 131; 132; 135; diagnosis, 108; 109; 110; 120; 246
138; 141; 142; 150; 155; 248 discourse, 7; 22; 25; 46; 55; 57; 58; 59;
computer, 3; 14; 23; 28; 67; 88; 219; 220; 60; 76; 77; 78; 79; 80; 82; 86; 88; 93;
227; 236; 237; 247; 253; 255 95; 115; 116; 139; 148; 152; 154; 193;
conceptual, 3; 6; 13; 14; 17; 20; 22; 44; 204; 223; 232; 240; 242; 244; 245;
45; 47; 52; 64; 67; 69; 70; 74; 76; 78; 247; 249; 251; 252; 254; 256
81; 83; 85; 92; 94; 95; 96; 97; 98; 137; dynamics, 20; 71; 72; 91; 95; 140; 145;
141; 158; 160; 162; 164; 167; 170; 150; 155; 168; 170; 175; 178; 186;
171; 172; 179; 181; 182; 201; 205; 256
222; 246; 256 editing, 19; 20; 25; 101; 102; 118; 144;
connectionism, 81; 134; 135 147; 224; 227; 248
constituent, 39; 162; 227 elaboration, 5; 8; 10; 15; 17; 18; 19; 20;
constraint, 49; 73; 91; 92; 94; 176; 201; 22; 27; 33; 34; 36; 37; 40; 41; 43; 44;
207 47; 49; 51; 53; 54; 55; 60; 67; 68; 70;
content generation, 37; 38; 71; 72; 73; 74; 71; 72; 73; 74; 77; 80; 81; 83; 84; 85;
105 86; 87; 92; 93; 95; 96; 97; 133; 135;
context, 1; 2; 3; 12; 14; 15; 24; 25; 29; 142; 152; 161; 162; 170; 175; 179;
52; 55; 56; 58; 59; 60; 76; 79; 97; 98; 182; 184; 191; 192; 200; 202; 203;
108; 137; 147; 149; 175; 188; 194; 204; 208; 211; 214; 220; 231
221; 229; 243; 244; 255 encapsulation, 26; 136; 137; 213
contextualisation, 59 encoding, 2; 12; 23; 79; 80; 81; 84; 86;
control, 1; 2; 3; 5; 8; 9; 10; 13; 17; 18; 21; 181; 208; 225; 241; 255
24; 25; 26; 28; 37; 40; 41; 44; 53; 54; epistemic, 39; 42; 43; 176
55; 57; 58; 61; 69; 70; 71; 73; 74; 80; error, 26; 101; 103; 106; 108; 115; 116;
84; 86; 91; 92; 97; 105; 111; 114; 116; 120; 148; 149; 151; 194; 251
118; 119; 127; 128; 129; 130; 131; evaluation, 35; 37; 39; 47; 49; 51; 61; 64;
133; 135; 136; 137; 140; 141; 142; 73; 100; 103; 106; 107; 108; 110; 114;
144; 145; 146; 147; 148; 149; 150; 121; 122; 127; 144; 148; 149; 155;
151; 152; 153; 154; 155; 156; 157; 156; 157; 183; 185; 238; 252
159; 178; 183; 194; 202; 203; 211; event role, 85
212; 213; 221; 223; 224; 225; 227; executing, 19; 56; 194
228; 229; 230; 233; 235; 239; 240; execution, 13; 15; 20; 23; 25; 26; 33; 36;
241; 245; 251; 252 50; 54; 55; 56; 57; 60; 67; 71; 80; 81;
control of processes, 128; 129; 140; 141; 86; 88; 89; 90; 92; 98; 131; 136; 137;
142; 147; 150; 151; 152; 155; 212 141; 144; 148; 149; 154; 160; 163;
creation, 42; 43; 44; 45; 73; 85; 86; 134; 166; 169; 174; 175; 177; 179; 181;
176; 177; 179; 181; 219; 255 183; 195; 196; 203; 209; 211; 214;
creativity, 15; 39; 42; 43; 44; 45; 63; 257 215; 224; 226; 227; 233
cyclic, 92; 127; 132 expert, 6; 8; 9; 10; 49; 50; 64; 78; 87; 90;
detection, 26; 93; 103; 108; 109; 110; 91; 118; 121; 122; 135; 142; 144; 149;
115; 116; 120; 148; 224; 226; 247 163; 179; 180; 185; 187; 188; 189;
development 195; 198; 199; 200; 202; 203; 204;
expertise, 121; 131; 214; 215 205; 207; 211; 213; 214; 215; 218;
219
SUBJECT INDEX 265
feedback, 10; 15; 19; 42; 43; 55; 62; 63; 215; 217; 218; 221; 222; 235; 238;
115; 147; 154; 177; 179; 200; 235; 239; 240; 241; 243; 244; 248; 250;
238; 256 252; 253; 255; 256
fluency, 165; 167; 215; 225; 227; 241; inhibition, 87; 129; 132; 134; 135; 171;
251; 253 173; 175; 178; 209; 222
formulation, 10; 13; 15; 17; 25; 60; 78; inner speech, 148; 162; 227
79; 80; 85; 90; 92; 93; 96; 97; 131; instructions, 5; 7; 8; 9; 10; 38; 62; 80; 85;
162; 173; 188; 224; 241; 249 87; 108; 127; 133; 145; 175; 176; 178;
frame, 8; 60; 83 184; 188; 194; 198; 199; 200; 233;
functional, 1; 2; 14; 28; 46; 76; 81; 84; 235; 237; 248
85; 87; 92; 99; 100; 110; 134; 171; interactionist, 24; 81; 129; 130; 135; 136;
172; 178; 186; 199; 214 137; 138; 140; 141; 147; 150; 155;
generating, 34; 37; 38; 39; 42; 46; 47; 223; 224; 230
105; 203; 212; 223; 237; 255 interactive, 9; 23; 24; 57; 62; 81; 113;
generation, 26; 40; 42; 63; 72; 74; 78; 80; 129; 130; 132; 177; 213; 219; 227
86; 88; 95; 96; 98; 116; 134; 147; 148; internal speech, 13; 148; 149
154; 175; 192; 193; 194; 211; 212; irrelevant speech, 165; 166; 167; 168;
217; 223; 226; 227; 239; 242; 244; 227; 253
249; 254; 255; 256 knowledge
goal setting, 10; 144 audience, 253
grammatical, 1; 3; 5; 12; 23; 24; 53; 69; declarative, 220
70; 78; 79; 80; 81; 83; 84; 85; 86; 87; discourse, 7; 57; 116
88; 92; 93; 94; 95; 96; 97; 100; 120; domain, 3; 5; 10; 11; 12; 15; 19; 34;
127; 128; 162; 182; 211; 212; 214; 36; 37; 38; 39; 40; 41; 42; 43; 44;
225; 235; 241 45; 46; 49; 51; 53; 54; 55; 57; 61;
graphomotoric, 23; 36; 78; 80; 90; 98; 67; 69; 71; 74; 77; 95; 96; 97; 110;
177; 211; 212 134; 152; 175; 176; 178; 180; 181;
handwriting, 2; 8; 18; 19; 23; 70; 71; 88; 188; 199; 200; 202; 229
89; 90; 92; 98; 122; 127; 142; 152; linguistic, 5; 7; 40; 54; 64; 96; 97;
158; 192; 194; 195; 211; 222; 224; 110; 163; 169; 242; 251
225; 226; 241; 242; 244; 247; 251; metacognitive, 108; 117; 192; 194;
256 197
hierarchical, 34; 45; 48; 49; 60; 84; 131; pragmatic, 5; 40; 54; 57; 61; 62; 97;
133; 154; 172; 201; 203; 206
110; 188; 214
idea, 2; 1; 8; 20; 26; 39; 40; 41; 45; 49;
procedural, 3; 18; 25; 57; 181; 218;
53; 71; 73; 91; 92; 93; 94; 95; 97; 100;
239; 245
109; 111; 114; 129; 130; 131; 132;
rhetorical, 11; 15; 41; 57; 250
137; 138; 139; 143; 146; 152; 154;
155; 159; 169; 170; 171; 172; 176; topic, 7; 229; 234; 249
177; 179; 192; 193; 198; 205; 206; unit, 11; 24; 26; 37; 38; 39; 40; 41; 42;
209; 210; 212; 213; 214; 219; 222; 43; 45; 48; 49; 55; 56; 75; 77; 79;
225; 226; 232; 238; 239; 256 132; 133; 134; 142; 157; 170; 171;
inference, 17; 52; 217 172; 174; 175; 176; 177; 178; 180;
information, 2; 5; 6; 12; 14; 15; 17; 19; 181; 198; 199; 201; 205; 206; 207;
21; 24; 33; 34; 35; 36; 40; 44; 45; 46; 212
50; 51; 52; 67; 72; 73; 76; 77; 79; 83; knowledge, 2; 3; 5; 7; 8; 11; 14; 15; 17;
88; 91; 96; 100; 109; 112; 114; 132; 23; 24; 26; 35; 36; 37; 38; 39; 40; 41;
138; 142; 143; 144; 145; 147; 152; 42; 43; 45; 46; 47; 48; 49; 54; 55; 56;
154; 156; 157; 158; 159; 161; 162; 57; 61; 64; 74; 75; 76; 77; 79; 91; 96;
168; 170; 171; 173; 175; 179; 183; 97; 106; 107; 108; 110; 111; 112; 113;
185; 188; 190; 195; 201; 202; 205; 114; 115; 116; 117; 122; 132; 133;
206; 207; 208; 209; 210; 213; 214; 134; 135; 137; 138; 140; 142; 144;
266
146; 147; 152; 154; 155; 157; 159; short term memory, 21; 243
160; 161; 167; 170; 171; 172; 173; memory probe, 8; 37; 39; 40; 41; 54; 71;
174; 175; 176; 177; 178; 179; 180; 171
181; 182; 184; 187; 188; 189; 190; memory span, 27; 157; 169; 170; 173;
198; 199; 200; 201; 203; 204; 205; 178; 179; 183; 185; 205; 207; 208;
206; 207; 210; 211; 212; 214; 215; 210; 213; 215; 216; 243; 248; 253
217; 218; 220; 222; 224; 225; 228; message, 1; 11; 13; 18; 19; 20; 22; 25; 55;
229; 232; 234; 236; 237; 238; 239; 76; 78; 79; 80; 82; 83; 85; 86; 87; 88;
242; 243; 245; 246; 248; 251; 253; 89; 90; 91; 92; 128; 139; 148; 162;
254; 255; 256 163; 177; 206; 218; 247
knowledge driven, 229 method on-line, 51
learning, 2; 133; 177; 195; 198; 212; 217; microplanning, 12
219; 220; 225; 239; 240; 244; 245; microproposition, 206
246; 247; 249; 250; 251; 254; 256; modalisation, 59
257 model
lemmas, 12; 79; 82; 84; 85; 86 activation, 27; 173; 178; 179; 181; 184
lexical, 3; 5; 7; 12; 13; 24; 40; 43; 51; 53; global, 24; 129
69; 70; 78; 79; 81; 82; 83; 84; 85; 86; interactionist, 129; 135; 140; 141;
87; 90; 91; 92; 95; 96; 97; 128; 131; 147; 223; 230
137; 148; 162; 169; 182; 193; 211; local, 23; 24; 28; 80; 86; 87; 88; 90;
212; 214; 218; 227; 250 95; 98; 129; 175
limited capacities, 26; 33; 122; 154; 155; modularist, 136; 138; 140; 147; 155
161; 163; 169; 173; 207 psycholinguistic, 54; 58; 60; 256
limited resources, 169; 216 model, 2; 3; 4; 6; 10; 11; 12; 13; 14; 15;
linearisation, 51; 74; 75; 76; 77; 79; 86; 16; 17; 18; 19; 20; 21; 22; 23; 24; 25;
95; 97 26; 27; 28; 29; 36; 37; 38; 39; 40; 42;
linguistic formulation, 36; 51; 71; 74; 78; 43; 44; 47; 48; 52; 54; 55; 57; 58; 59;
79; 80; 87; 88; 91; 92; 95; 96; 127; 60; 61; 63; 64; 67; 69; 70; 72; 73; 74;
141; 154; 162 78; 79; 80; 81; 83; 85; 86; 87; 89; 90;
linguistic processing, 22; 78; 79; 95; 96; 92; 94; 96; 97; 99; 102; 103; 104; 105;
164; 183; 191 106; 107; 109; 110; 111; 112; 113;
long term memory, 67 114; 115; 116; 118; 122; 123; 127;
macroproposition, 92 128; 129; 130; 132; 133; 134; 135;
main point, 14; 78; 202; 240 136; 139; 140; 141; 142; 144; 145;
maintaining, 2; 17; 24; 26; 60; 91; 94; 146; 147; 148; 149; 150; 151; 152;
154; 155; 158; 159; 161; 163; 189; 155; 160; 161; 162; 163; 164; 165;
205; 206; 207; 213; 225; 227 166; 167; 168; 169; 170; 171; 172;
management, 3; 5; 11; 18; 21; 24; 25; 26; 173; 175; 176; 177; 178; 179; 180;
28; 29; 34; 55; 79; 90; 107; 116; 121; 181; 182; 184; 185; 186; 190; 192;
122; 127; 128; 141; 142; 146; 147; 193; 194; 195; 197; 198; 199; 200;
155; 157; 158; 162; 164; 168; 170; 201; 204; 207; 209; 220; 224; 225;
172; 173; 178; 182; 187; 213; 215; 226; 230; 236; 237; 239; 241; 249;
218; 243; 245 250; 251; 252
matching, 14; 57; 70; 79; 83; 86; 93; 95; module, 25; 79; 136; 137; 139; 147; 148;
146; 162; 171; 174; 181 155
maturation, 10; 27; 189; 190; 207; 209; monitor, 151; 224
210; 215; 216; 220 monitoring, 21; 29; 62; 119; 141; 144;
meaning, 49; 59; 79; 81; 85; 88; 99; 101; 145; 149; 151; 223; 224; 240; 246;
102; 105; 110; 115; 118; 119; 120; 247; 256
151; 156; 158; 199; 251 morphological, 81; 82; 83
medium, 14; 17; 23; 184 motivation, 17; 18; 116; 123; 239; 240;
memory 246
SUBJECT INDEX 267
motor, 13; 23; 88; 89; 90; 194; 212; 224; 78; 91; 92; 108; 117; 176; 182; 191;
225; 226; 256 198; 202
narrative, 8; 36; 45; 51; 59; 157; 220; planning
222; 226; 228; 229; 241; 244; 251 content, 9; 36; 37; 64; 70
narrative schema, 36; 45; 51 planning, 1; 2; 8; 9; 10; 15; 27; 33; 34;
neo-piagetian, 2; 208; 210; 252 35; 36; 37; 40; 49; 50; 52; 53; 59; 60;
network, 39; 43; 45; 59; 76; 81; 82; 83; 62; 63; 64; 65; 69; 70; 71; 75; 89; 92;
129; 130; 131; 132; 134; 135; 171; 94; 122; 127; 131; 137; 144; 150; 157;
172; 175; 176; 177; 178; 179; 184; 162; 167; 176; 188; 196; 197; 198;
201; 203 202; 203; 204; 205; 212; 219; 222;
node, 82; 83; 131; 171 223; 224; 225; 226; 228; 229; 239;
novice, 6; 50; 64; 119; 121; 122; 185; 241; 242; 243; 245; 246; 248; 251;
187; 188; 189; 193; 196; 198; 199; 254
200; 202; 203; 205; 207; 211; 213; positional, 87
218; 219 practice, 2; 133; 144; 189; 207; 210; 211;
operation 213; 214; 215; 219; 220; 224; 225;
organising, 48; 49; 50; 52 226; 228; 240
operation, 8; 22; 37; 39; 40; 41; 48; 49; pragmatic, 5; 6; 8; 10; 11; 12; 34; 37; 40;
54; 56; 68; 69; 70; 72; 73; 75; 76; 78; 41; 42; 43; 44; 49; 53; 54; 55; 57; 58;
105; 109; 110; 130; 198; 199; 201; 60; 61; 62; 73; 76; 77; 88; 97; 110;
213 135; 137; 141; 148; 157; 181; 188;
oral, 8; 11; 17; 20; 23; 25; 46; 55; 58; 59; 202; 203; 214; 215
70; 79; 80; 82; 85; 86; 87; 89; 97; 98; preverbal message, 12; 22; 69; 72; 74; 75;
128; 134; 136; 139; 141; 148; 151; 77; 78; 79; 83; 85; 86; 95; 162
152; 153; 154; 155; 175; 192; 193; problem-solving, 2; 3; 6; 9; 10; 11; 17;
195; 204; 223; 226; 229; 232; 241; 49; 50; 57; 92; 122; 131; 137; 144;
248 174; 176; 199; 200; 201; 209; 222;
organising, 33; 34; 36; 37; 47; 48; 49; 50; 252
51; 52; 76; 144; 158 procedural, 3; 6; 18; 25; 44; 45; 57; 60;
orthographic, 1; 23; 89; 96; 194; 212; 74; 80; 90; 99; 102; 106; 110; 112;
218; 223; 241; 255 132; 133; 134; 144; 153; 181; 218;
pause, 50; 51; 71; 93; 165; 246; 254 220; 222; 239; 245; 248; 249
pause duration, 50; 51; 71; 93; 165 process, 2; 1; 2; 3; 4; 5; 8; 9; 13; 14; 15;
phonological, 12; 17; 18; 19; 20; 23; 26; 16; 17; 18; 19; 20; 21; 22; 23; 24; 25;
79; 80; 81; 82; 84; 85; 87; 89; 148; 26; 27; 28; 29; 33; 34; 35; 36; 37; 38;
151; 160; 161; 162; 165; 167; 168; 39; 40; 41; 42; 43; 44; 45; 46; 47; 48;
169; 194; 208; 216; 223; 226; 227; 49; 50; 51; 52; 53; 54; 55; 58; 59; 61;
250; 254 62; 63; 64; 65; 67; 68; 69; 70; 71; 72;
phonological loop, 167; 250; 254 73; 78; 79; 80; 81; 83; 84; 86; 87; 88;
plan, 5; 13; 14; 21; 22; 23; 33; 34; 36; 37; 89; 90; 91; 92; 93; 94; 97; 98; 99; 100;
40; 41; 50; 51; 52; 55; 56; 57; 60; 63; 102; 103; 104; 105; 108; 109; 110;
67; 68; 70; 71; 72; 74; 75; 78; 79; 91; 111; 117; 118; 119; 120; 121; 122;
92; 94; 95; 97; 98; 106; 109; 113; 133; 123; 127; 128; 130; 131; 133; 134;
137; 143; 148; 164; 169; 171; 188; 135; 136; 137; 138; 140; 141; 142;
192; 198; 200; 203; 206; 209; 219; 143; 144; 145; 147; 149; 150; 151;
220; 251 152; 155; 158; 159; 160; 161; 162;
plan and content, 9; 10; 34; 36; 37; 64; 163; 164; 165; 167; 168; 169; 171;
67; 70; 202 173; 174; 175; 177; 178; 179; 182;
plan and processing, 133; 202 184; 187; 188; 189; 191; 192; 193;
plan and text, 5; 10; 18; 22; 33; 36; 40; 194; 195; 196; 198; 199; 200; 201;
47; 52; 53; 68; 69; 70; 71; 72; 73; 74; 202; 204; 208; 210; 211; 212; 213;
214; 215; 217; 218; 219; 224; 226;
268
227; 228; 229; 236; 239; 240; 241; 171; 174; 175; 178; 180; 181; 182;
242; 244; 246; 247; 248; 249; 250; 188; 190; 191; 193; 194; 204; 211;
251; 252; 253; 255; 256 212; 216; 217; 218; 219; 220; 222;
processing 223; 224; 225; 239; 240; 241; 242;
conceptual, 22; 95; 96; 141; 164 244; 245; 246; 247; 248; 249; 250;
incremental, 139 251; 252; 253; 254; 255; 256; 257
linguistic, 22; 78; 79; 95; 96; 164; programming, 19; 88; 89; 220; 225
183; 191 proposition, 27; 67; 68; 69; 70; 73; 75;
modalities, 29; 81; 129; 135; 136; 140; 77; 91; 92; 93; 94; 98; 134; 157
168; 174; 193 prosody, 84; 85
parallel, 87; 131; 134; 135; 139; 140 psycholinguistic, 1; 11; 54; 58; 60; 256
pragmatic, 54; 58 punctuation, 1; 80; 91; 96; 100; 101; 115;
sequential, 131; 155 118; 119; 188; 189; 192; 211; 218
processing, 2; 3; 5; 6; 11; 14; 17; 18; 19; reader, 10; 15; 34; 36; 40; 53; 54; 55; 58;
20; 21; 22; 24; 25; 26; 28; 29; 33; 34; 61; 62; 73; 77; 103; 108; 111; 121;
35; 36; 37; 40; 43; 44; 45; 47; 50; 52; 188; 189; 192; 197; 231; 234; 235;
53; 54; 55; 56; 57; 58; 59; 60; 61; 63; 236; 238; 254
65; 69; 70; 71; 72; 73; 74; 75; 78; 79; reading, 1; 2; 3; 5; 28; 43; 44; 53; 55; 94;
80; 81; 83; 84; 85; 86; 87; 88; 89; 90; 100; 108; 110; 111; 114; 119; 120;
91; 92; 93; 94; 95; 96; 97; 98; 105; 122; 127; 131; 144; 147; 150; 159;
106; 112; 113; 114; 115; 116; 117; 160; 164; 168; 178; 183; 184; 188;
120; 127; 128; 129; 130; 131; 132; 205; 212; 213; 214; 216; 217; 224;
133; 134; 135; 136; 138; 139; 140; 225; 236; 241; 242; 243; 244; 249;
141; 142; 143; 144; 145; 147; 149; 253; 254; 255; 256
150; 152; 153; 155; 157; 159; 160; recovery, 37; 39; 41; 44; 45; 46; 53; 55;
161; 162; 163; 164; 166; 167; 168; 85; 86; 132; 155; 174; 175; 176; 177;
169; 170; 173; 174; 178; 179; 181; 181; 211; 217
182; 183; 184; 185; 188; 189; 192; recursion, 5; 21; 63; 71; 91; 142; 150; 193
193; 194; 195; 198; 199; 201; 202; reflection, 17; 28; 33; 88; 111; 120; 137;
203; 204; 205; 206; 207; 208; 209; 188
210; 211; 212; 213; 214; 215; 216; representation
217; 220; 222; 223; 224; 225; 226; mental, 7; 8; 100; 105; 110; 113; 119;
227; 228; 229; 235; 239; 240; 241; 120; 132; 135; 248
242; 243; 244; 245; 246; 248; 249; representation, 1; 7; 25; 26; 43; 53; 54;
250; 251; 252; 253; 255 55; 56; 59; 61; 78; 80; 81; 82; 83; 85;
production 105; 106; 107; 108; 109; 110; 113;
activity, 2; 3; 23; 59; 141; 180 117; 120; 130; 131; 132; 135; 136;
rules, 57; 99; 103; 133; 134; 142; 171; 138; 141; 147; 148; 157; 206; 209;
174 251; 255
system, 57; 58; 86; 132; 133; 134; resource, 79; 169; 224; 225; 226; 250;
141; 144; 148; 152; 153 253
production, 1; 2; 3; 11; 12; 13; 14; 17; 18; retrieval, 9; 11; 38; 39; 40; 41; 44; 45; 51;
19; 20; 21; 22; 23; 24; 25; 26; 28; 33; 57; 70; 71; 72; 92; 98; 134; 145; 161;
35; 36; 40; 45; 47; 49; 51; 53; 54; 55; 169; 180; 181; 211; 222; 226; 239;
56; 57; 58; 59; 60; 61; 62; 70; 73; 79; 244; 253; 254; 256
80; 81; 82; 83; 84; 85; 86; 87; 88; 90; structure, 222
91; 93; 94; 95; 96; 98; 99; 103; 107; reviewing, 102; 104; 117; 122; 223; 225;
110; 111; 114; 119; 120; 128; 129; 226; 228; 229; 230
131; 132; 133; 134; 135; 136; 137; revising, 1; 5; 25; 43; 53; 55; 99; 100;
138; 139; 141; 142; 144; 148; 149; 101; 102; 104; 105; 106; 107; 108;
150; 151; 152; 153; 154; 155; 157; 109; 110; 111; 112; 114; 115; 116;
158; 160; 162; 165; 167; 168; 169; 117; 118; 119; 120; 121; 122; 127;
SUBJECT INDEX 269
137; 142; 144; 145; 149; 156; 182; spelling, 2; 100; 101; 108; 115; 116; 117;
188; 219; 225; 233; 241; 243; 244; 118; 119; 120; 151; 152; 188; 189;
247; 251; 253 192; 194; 195; 211; 213; 218; 223;
revision 235
external, 103; 108; 113; 117 storage, 45; 117; 158; 159; 166; 170; 181;
internal, 86; 108; 117 205; 216; 221; 226; 227; 244
revision, 1; 14; 15; 17; 18; 20; 25; 28; 99; storing, 132; 134; 139; 155; 157; 159;
100; 101; 102; 103; 104; 105; 106; 160; 161; 162; 163; 164; 170; 173;
107; 108; 109; 110; 111; 112; 113; 180; 183; 184; 185; 205; 209; 210;
114; 115; 116; 117; 118; 119; 120; 214; 215; 217
121; 122; 123; 137; 144; 148; 149; strategy
151; 152; 155; 182; 188; 231; 232; local coherence, 206
233; 236; 237; 238; 240; 242; 243; writing, 2; 96; 142; 158; 198; 199
245; 246; 247; 248; 251; 253; 256 strategy, 2; 6; 10; 33; 34; 37; 40; 41; 46;
schema, 8; 35; 41; 43; 45; 51; 69; 85; 55; 56; 72; 75; 76; 77; 91; 92; 100;
111; 144; 203; 239 105; 108; 109; 116; 135; 141; 142;
secondary task, 50; 51; 93; 158; 164; 165; 152; 155; 157; 159; 198; 199; 200;
166; 167; 174; 184; 185; 223; 224; 203; 209; 219; 227; 228; 244; 246;
226; 227 247; 251; 256
selection, 11; 44; 61; 109; 199; 224; 244; structure, 1; 12; 13; 15; 20; 27; 40; 41;
254; 255; 256 45; 46; 49; 54; 61; 67; 68; 69; 70; 71;
self-monitoring, 255 75; 76; 77; 78; 80; 81; 83; 84; 85; 86;
self-regulation, 141; 154; 219; 247 87; 88; 93; 94; 95; 96; 111; 121; 133;
semantic, 1; 5; 12; 17; 18; 26; 34; 40; 43; 135; 138; 140; 159; 172; 173; 180;
45; 46; 47; 48; 49; 52; 58; 64; 68; 69; 181; 182; 188; 202; 203; 206; 208;
70; 74; 75; 76; 77; 78; 79; 80; 81; 82; 210; 214; 215; 225; 233; 239; 240;
83; 85; 86; 87; 88; 91; 92; 93; 94; 95; 241; 242; 243; 244; 245; 246; 249;
96; 97; 98; 101; 103; 109; 118; 119; 250; 251; 254; 255
132; 134; 151; 160; 162; 165; 169; sub-goal, 3; 5; 40; 41; 133; 141; 144; 200;
171; 172; 177; 181; 191; 222; 244; 203; 218
252 sub-theme, 44
sentence, 1; 11; 12; 20; 23; 24; 36; 43; super-structure, 51
67; 68; 70; 71; 73; 75; 76; 80; 81; 82; surface, 1; 59; 60; 76; 91; 96; 101; 103;
83; 84; 85; 86; 87; 88; 90; 91; 92; 93; 105; 109; 110; 115; 118; 119; 120;
94; 95; 96; 97; 98; 100; 101; 102; 105; 122; 156; 157
106; 117; 121; 122; 129; 135; 148; symbolic, 24; 43; 129; 130; 131; 132;
149; 157; 158; 166; 172; 175; 182; 134; 135; 140; 176; 244
183; 192; 194; 195; 196; 205; 211; syntactic, 1; 7; 13; 20; 40; 43; 46; 53; 59;
212; 224; 225; 226; 227; 241; 244; 68; 70; 71; 75; 79; 81; 82; 83; 84; 85;
245; 246; 247; 249; 253 87; 88; 93; 95; 97; 101; 115; 137; 151;
slave system, 121; 161; 164; 166; 167 191; 193; 218; 241; 249; 253
social interaction, 58; 59 syntax, 148; 169; 211; 241; 246
speaker, 21; 25; 55; 56; 58; 59; 60; 76; system, 1; 3; 9; 10; 13; 14; 16; 19; 21; 23;
77; 83; 85; 93; 136; 148; 250 24; 27; 38; 57; 61; 74; 86; 99; 112;
speaking, 3; 6; 11; 12; 13; 14; 21; 22; 25; 113; 128; 129; 130; 131; 132; 133;
40; 42; 43; 46; 57; 60; 64; 65; 79; 80; 134; 136; 138; 139; 140; 141; 142;
87; 96; 105; 108; 128; 129; 136; 137; 144; 145; 147; 148; 149; 152; 153;
147; 148; 149; 151; 152; 164; 184; 154; 155; 156; 159; 161; 162; 166;
190; 192; 193; 195; 204; 212; 226; 169; 170; 171; 173; 174; 177; 179;
232; 243; 247; 253 181; 184; 189; 193; 194; 197; 199;
speech act, 11; 40; 55; 56; 58 202; 207; 208; 209; 216; 217; 220;
222; 226; 235; 239; 244; 257
270
task environment, 5; 15; 94; 108; 198; 41; 42; 43; 44; 45; 47; 50; 51; 53; 54;
218; 227; 229 55; 56; 58; 59; 60; 61; 62; 67; 68; 71;
task schema, 144; 233; 238 72; 73; 74; 75; 76; 77; 78; 88; 91; 94;
text 96; 97; 99; 100; 102; 103; 105; 106;
produced so far, 6; 62; 108; 113; 144; 107; 108; 109; 110; 111; 114; 115;
145; 155; 175 119; 120; 128; 135; 137; 142; 145;
type, 36; 41; 44; 50; 51; 59; 77; 97; 146; 149; 150; 151; 155; 157; 161;
116; 117; 154; 167; 187; 198; 200; 177; 178; 179; 181; 183; 184; 185;
203; 213; 218; 220; 229; 254 189; 192; 194; 197; 198; 199; 200;
text composition, 240; 244 202; 203; 204; 205; 206; 207; 211;
text content, 1; 2; 7; 8; 9; 10; 22; 35; 36; 213; 214; 216; 217; 218; 219; 222;
40; 42; 43; 44; 49; 51; 52; 61; 67; 70; 223; 224; 227; 229; 230; 233; 234;
71; 72; 74; 142; 154; 158; 198; 200; 235; 236; 238; 243; 247; 248; 253;
201; 202; 203; 206; 211; 215 257
text interpretation, 111 writing, 1; 2; 1; 2; 3; 4; 5; 6; 7; 8; 9; 10;
text quality, 47; 121; 165; 167; 212; 216; 11; 13; 14; 15; 16; 17; 18; 19; 20; 21;
217; 219; 233; 242; 256 22; 23; 24; 25; 26; 27; 28; 29; 33; 34;
textualisation, 59; 60; 96 35; 36; 37; 38; 39; 40; 41; 42; 43; 44;
theme, 5; 37; 85; 108; 110; 113; 114; 115; 45; 46; 47; 48; 49; 50; 51; 52; 53; 54;
178; 199; 200; 201; 217 55; 56; 57; 58; 60; 61; 62; 63; 64; 65;
top-down, 56; 57; 82; 90; 94; 95; 96; 130; 67; 68; 71; 72; 77; 78; 79; 80; 86; 87;
135; 141; 144; 147; 154 88; 89; 90; 91; 94; 96; 97; 98; 99; 100;
topic, 1; 5; 7; 8; 9; 15; 34; 36; 37; 38; 40; 102; 103; 104; 108; 109; 110; 111;
41; 48; 49; 71; 74; 91; 92; 114; 115; 114; 116; 117; 118; 119; 120; 121;
116; 117; 132; 199; 200; 206; 213; 122; 123; 127; 128; 129; 130; 131;
217; 218; 229; 232; 234; 235; 237; 132; 134; 135; 136; 137; 138; 139;
249; 251 140; 142; 143; 144; 145; 146; 147;
training, 2; 27; 47; 140; 173; 189; 190; 148; 149; 150; 151; 152; 154; 155;
207; 210; 214; 215; 218; 233; 234; 156; 157; 158; 160; 161; 162; 163;
247 164; 165; 166; 167; 168; 169; 170;
translating, 1; 5; 17; 43; 60; 67; 69; 72; 175; 176; 177; 178; 179; 180; 181;
95; 96; 97; 122; 127; 177; 188; 196; 182; 183; 184; 185; 186; 187; 188;
198; 219; 225; 226; 228; 229; 237; 189; 190; 191; 192; 193; 194; 195;
239; 241; 243; 251 196; 197; 198; 199; 200; 201; 202;
typewriting, 18; 23; 70; 71; 88; 89; 90; 203; 204; 205; 206; 207; 208; 210;
255 211; 212; 213; 214; 215; 216; 217;
verbal production, 2; 4; 20; 21; 22; 23; 218; 219; 220; 221; 222; 223; 224;
24; 26; 27; 28; 33; 37; 39; 42; 45; 54; 225; 226; 227; 228; 229; 230; 231;
55; 56; 57; 58; 72; 73; 85; 93; 129; 232; 233; 234; 235; 236; 237; 238;
136; 139; 140; 148; 152; 154; 155; 239; 240; 241; 242; 243; 244; 245;
174; 183 246; 247; 248; 249; 250; 251; 252;
verbal protocol, 6; 21; 64; 102; 118; 127; 253; 254; 255; 256
158; 200; 222 writing assignment, 39; 41; 248
visuo-spatial, 18; 159; 160; 162; 163; writing constraints, 2
164; 166; 167; 168; 169; 184; 217 writing skills, 189; 190; 204; 210; 220;
visuo-spatial sketchpad, 163 238; 239; 241; 247; 255
word, 59; 60; 79; 83; 84; 88; 90; 93; 100; written text, 3; 5; 10; 17; 21; 25; 37; 42;
102; 117; 183; 184; 194; 195; 196; 44; 53; 55; 67; 92; 94; 98; 99; 100;
219; 224; 226; 235; 239; 242; 250; 102; 103; 105; 108; 109; 113; 117;
251; 255 119; 120; 135; 150; 169; 179; 195;
writer, 1; 2; 3; 4; 5; 6; 7; 8; 9; 10; 15; 16; 224; 239; 245; 249; 252
17; 18; 19; 20; 21; 34; 35; 36; 37; 40;
View publication stats

Alamargot Chanquoy 2001

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

Alamargot Chanquoy 2001

Загружено:

Авторское право:

Доступные форматы

See

Through the Models of Writing

Book · January 2001

Denis Alamargot Lucile Chanquoy

SEE PROFILE SEE PROFILE

ISEL [International Symposium on Educational Literacy] / SILE [Symposium International sur la

TAO (Twictée pour apprendre l'orthographe) View project

The user has requested enhancement of the downloaded file.

Usage interne - ne pas diffuser

University of Poitiers, France

A definition of writing and a presentation of the main models

1 WRITING A TEXT: A COMPLEX TASK

D. Alamargot & L. Chanquoy (2001). General introduction. A definition of writing and a

1.2 To manage writing constraints

1.3 Text production: a possible definition in the context of cognitive psychology

2 THE MAIN MODELS OF VERBAL PRODUCTION

THE WRITER’S PLANNING TRANSLATING REVIEWING

pository text writing), the identification of processes and writing representations

2.2 1987: Bereiter and Scardamalia’s developmental model

2.2.1 The Knowledge Telling Strategy

Run test Fail

First, there is the mental representation of instructions (Mental Representation of as-

2.2.2 The Knowledge Transforming Strategy

Problem analysis and

Figure 3: Knowledge Transforming Strategy,adapted from Bereiter & Scardamalia (1987).

2.3 1989: Levelt’s speaking model

2.3.1 From the Conceptualizer to the Articulator

The ‘Formulator’ allows lexical, grammatical and phonological encodings of the

Finally, Levelt includes in his model a control process, called ‘Self-Monitoring’,

2.3.2 Adapting Levelt’s model to writing

Preverbal message Parsed string

Graphemic plan Graphemic string

Figure 5: Model of written production inspired by Levelt’s model (1989),

those described by Levelt (1989): concerning conceptual processes of Macroplan-

2.4 1996: A new architecture and a process model

• It ignores inter-individual differences, mainly those concerning the writer’s

THE TASK ENVIRONMENT

The Social Environment The Physical Environment

The audience The text so far

Phonological Memory Reflection

of mental activities allowing to transform pieces of knowledge in other pieces

Composing Medium) to specifically elaborate the written message, by means of

2.4.2 Kellogg’s (1996) model

Formulation Execution Monitoring

Visuo-Spatial Central Phonological

Figure 7: Relationships between production and Working Memory components,adapted from

The system is the model of Working Memory such as developed by Baddeley

3 DIVERSITY OF WRITING MODELS: SOME POINTS OF COMPARISONS

3.1 Differences in the architecture

3.1.1 The variety of process definitions.

3.1.2 Some processes are ignored, other are reduced

matical encoding and a phonological encoding. Moreover, an explanation about the

3.1.3 Local models in response to the limits of general models

3.2 Differences of functioning

3.2.2 Different ways to consider process management and control

Monitoring represents therefore less of an authority of process management than a

3.3 Between architecture and functioning: the development of writing expertise

4 THROUGH THE MODELS OF WRITING

Roll up, the textual mystery tour

1 INTRODUCTION: PLANNING IN WRITING

4.1 Planning: a general cognitive activity involved in text writing

4.2 Different planning activities in text writing

Plan to Say Plan to Compose