ED 258 204

AUTHORTITLEINSTITUTION

SPONS AGENCYREPORT NOPUB DATENOTE

PUB TYPE

EDRS PRICEDESCRIPT0g5

DOCUMENT RESUME

CS 208 968

McCutchen, DeborahWriting as a Linguistic Problem.Pittsburgh Univ., Pa. Learning Research andDevelopment Center.National Inst. of Education (ED), Washington, DC.LRDC-1984/448417p.; Reprint from "Educitional Psychology," Volume19, Number 4, p226-238, 1984.Viewpoints (120)

MF01/PC01 Plus Postage.*Cognitive Processes; Language Skills; LanguageUsage; *Linguistics; *Problem Solving; *Writing(Composition); *Writing Processes; Writing Skills

ABSTRACTThe current focus on high level planning and abstract

goals runs the risk of misrepresenting the contributions thatcognitive psychology could make to the study of writing if itneglects important linguistic features that distinguish the writingof natural language from other problem solving tasks. An alternativeperspective on writing emphasizes psycholinguistic processes involvedin generating sentences and linking them into coherent text. Theproblem solving models of writing, developed out of work in cognitivescience, are hierarchical in nature. While some hierarchical planningis done in writing (such as working from an outline), writerscommonly begin with something and follow it to a preplanned nextidea, or to a newly discovered thought or to a dead end. Wordsalready written can drastically affect what follows them. Then, theinteraction between the text-level processes and the planning-levelprocesses can be seen. Understanding the linguistic text features,such as the "top down" approach to reading (an emphasis on thereader's knowledge and its schematic organization) and the "bottomup" approach (an emphasis on lower level linguistic processes) isimportant to the writer's effectiveness in communicating ideas to thereader and in solving writing problems. (EL)

************************************************************************ Reproductions supplied by EDRS are the best that can be made ** from the original document. *

*************************************L*****************************

LEARNING RESEARCH AND DEVELOPMENT CENTER

1984/44

WRITING AS A LINGUISTIC PROBLEM

DEBORAH MCCUTCHEN

2

WRITING AS A LINGUISTIC PROBLEM

Deborah McCutchen

Learning Research and Development CenterUniversity of Pittsburgh

1984

Reprinted by permission from Educational Psychologist, 19 226-238. Copyright 1984 byDivision 15 of the American Psychological Association.

Preparation of the material herein was supported by the Learning Research and Develop-ment Center, funded in part by the National Institute of Education (NIE), U.S. Departmentof Education. The opinions expressed do not necessarily reflect the position or policy ofN1E, and no official endorsement should be inferred.

Educational Psychologist19144 Vol 19. No 4. 226-- 238

Writing asa Linguistic Problem

Deborah McCutchenUniversity of Pittsburgh

This article presents a psycholinguistic view of writing that focuses on theprocesses of translating concepts into sentences. The current research em-phasis on planning is discussed in terms of its theoretical roots in artificialintelligence models of planning and the limited applicability of suchmodels for writing. A case is made for viewing writing not only as a plan-ning problem but as a linguistic problem that can benefit from work inreading and speech production.

As writing has become an area of cognitiveinquiry, the related research has acquired a

peculiar character. Some of the best knowncognitive work compares the composing vo-cess with problem solving (Collins & Gentner,1980; Hayes & Flower, 198C; Nold, 1981), andwithin the problem solving framework, writinghas become yet another domain in which theimportance of high level planning can bedemonstrated. Thus, writing is clustered withphysics and other problem solving domainsand separated from linguistic processes such as

speech production and reading to which it in-tuitively seems related. Much of the cognitivework on writing has focused on high levelplanning and abstract goals (e.g., Burtis,Bereiter, Scardamalia, & Tetroe, in press;Flower & Hayes, 1980, 1981a; Matsuhuhi,1982; Scardamailia & Bereiter, 1982), to theextent that problem solving heuristics are ad-vocated to students as ways to improve theirwriting (Flower, 1981).

This article will argue that the current focuson high level planning and abstract goals runsthe risk of misrepresenting the contributionsthat cognitive psychology could make to thestudy of writing if it neglects importantlinguistic features that distinguish the writingof natural language from other problem solv-int, tasks. Planning as it is frequentlydiscussed seems to end just where much ofthe real problem of writing begins, and littleattention is given to the on-line processes

Preparation of this manuscript and the researchpresented here was supported by the Learning Researchand Development Center. which is supported in part bythe National Institute of Education. The author alsowishes to thank Charles A. Perfetti for discussions of theasues presented here and for helpful comments on anearlier version of this manuscript.

The address of Deborah McCutchen is: Learning Re.search and Development Center; University of Pittsburgh;3939 O'Hara Street: Pittsburgh, PA 15260.

involved in linear sentence generation. Leftunspecified are the processes that make writinga unique problem: the generation of extended,coherent language.

Contributions and Limitationsof Current Approaches

The problem solving framework has enabledimportant insights into writing as a process, andperhaps the most important of those insightshave concerned the interactive nature of thewriting process (Hayes & Flower, 1980). Somuch recent empirical work, however, hascentered on planning that other processes whichinteract with planning have been neglected.This is not to say that these other processes aretotally ignored by researchers. In much of thework cited earlier and in other cognitive work(e.g., E. J. Bartlett, 1982; Beaugrande, 1982;Bereiter & Scardamalia, 1981; Flower & Hayes,1984; Shuy, 1981), there are numerousacknowledgements of the importance oflanguage-based processes in writing. Flower(1981) allots eight pages of her tutorial onwriting to such linguistic concerns. However,the problem solving perspective and its empha-sis on planning are so dominant in the popularperception formed by treatments such asFlower's (1981) that little attention is givento the varisdways in which cognitive science caninform the study of writing. Planning is certain-ly important in writing, but a well-planned textis not necessarily a well-written one.

In what follows, an alternative perspectiveon writing is developed one that em-phasizes psycholinguistic processes involved ingenerating sentences and linking them intocoherent text. However, before that perspec-tive is presented in detail, the potential

Copyright 1984 by Division II of the American Psychological Association, Inc.

226

227 WRITING AS A LINGUISTIC PROBLEM

limitations of too narrow a focus on planningneed to be understood. We begin by tracingthe theoretical roots of the problem solving ap-

proach, in order to explain its somewhatnatural emphasis on planning, and by examin-ing the applicability of planning models towriting.

Problem Solving Approaches

Theoretical roots

The problem solving models of writing havedeveloped out of work in cognitive science.These and other studies of human complex-problem-solving behavior (e.g.: Hayes-Roth &Hayes-Roth, 1979; Jeffries, Turner, Poison, &Atwood, 1981; Larkin, McDermott, Simon, &Simon, 1980; Voss, Greene, Post, &Penner,1983) have examined aspects of humanoehavior for which quantifiable specificationspi.eviously had been few. They generally havefollowed the example of artificial intelligence(Al) models of problem solving, which em-phasize problem decomposition, with plan-ning as an especially important subprocess.(See Cohen & Feigenbaum, 1982, for a detail-ed summary.) The importance of planning inmost complex tasks is well recognized, and itscentral role in writing is emphasized not onlyin the work of cognitive researchers but intraditional rhetoric texts as well (e.g., Skwire,(Inwood, Ackley, & Fred1'ian, 1975).However, when the nature of the planning in-volved in writing is compared with the plan-ning done by Al models, some interesting dif-ferences emerge.

How well does writing fit typical planningmodels? To answer this question, let us ex-amine first some Al models in which theunderlying process assumptions are well laidout. Then we will examine how well these Almodels describe human problem solving invarious domains, and finally how well theydescribe writing.

Al planners. The most well-specified modelsare those planning systems developed in theuea of Al, and these planners differ accordingto the number of levels of abstraction permit-ted in the problem representation. That is,how much of the problem solving is done inthe abstract, before local details are specified?In this respect, there are important implica-tions for models of human problem solving,

and for writing especially. While Al modelsare not typically intended to be simulations ofhuman processing, their feasibility as such isinteresting to examine because. ultimately.any model of the writing process should be aswell specified as these Al models.

AI planners conventionally described as

nonhierarchical (e.g., STRIPS, HACKER, IN-TERPLAN; See Cohen & Feigenbaum, 1982)represent the problem at a single level andthus do no abstract planning. They begin solv-

ing the initial subgoal and continue workinglinearly, on the assumption that early decisionsare independent of later ones. Because of theirlack of foresight, they are generally less power-ful problem solvers. In these systems, criticalsteps in the solution process are notdistinguished from trivial ones, and a con-siderable amount of work can be dome (i.e.,many subgoals created and satisfied) beforethe planner confronts a critical goal that can-not be achieved because of a trivial early deci-sion. This requires a good deal of extra process-ing: backtracking to a critical choice point un-doing and then redoing the solution process.

Other Al planning systems described ashierarchical planners (e.g., NOAH,MOLGEN; See Cohen & Feigenbaum, 1982;Stefik, 1981a, 1981b) avoid backtracking byworking at multiple levels of abstraction intheir initial plans. Planners of this type followthe least commitment principle, postponingdecisions about details until a proposedabstract solution is shown not to result in in-terference among important decisions. Poten-tial conflicts in the plan can thus be detectedearly and corrected before the costly work ofsolving subproblems in detail is done. Theseprocedures, together with extensive knowledgeof the specific problem domain (see especiallyMOLGEN, by Stefik, 1981a, 1981b), makehierarchical planners very powerful and thusquite popular in AI research.

Human problem solving. Only a few ex-amples of human problem solving, however,seem as hierarchical in nature as the morepowerful Al planners. Experts in sof:waredesign (Jeffries et al., 1981) and physics(Larkin et al., 1980) seem to decompose com-plex problems into classifiable problem typeswith recognizable solutions (recognizable atleast to these experts). Novices, however, lack-ing the rich knowledge base of the experts, aretypically less successful in decomposing andclassifying the problems and seem to be forced

6

MORAH MtLU I Llit.N

into nonhierarchical planning, forced simplyto begin wherever they can and see how theirsolutions work out.

In other problem domains, eve') expert prob-lem solving looks less hierarchical. Voss et al.(1983) found that the solutions generated bytheir social science experts did not alwaysemerge neatly out of hierarchical refinements ofabstract representations. Often subgoals wereunexpectedly encountered and solved duringthe evaluation of implications of anothersubgoal. Such multidirectional, "oppor-tunisitic" problem solving characterizes otherhuman problem solving performance (Hayes-Roth & Hayes-RJth, 1979). t I these descrip-tions, human solutions evolve incrementallyfrom subgoals at various levels rather than fromorderly problem refinement at any given level,with low-level constraints sometimes beingdealt with before more abstract ones.

Apphcationi to Writing

Planning in writing. What is the nature ofthe planning done by writers? Many tradi-tional composition texts present writing as atask d',ectly amenable to hierarchical plan-ning. with multiple levels of problem decom-position (e.g., Skwire et al., 1975). This isreflected in the frequent suggestion to the stu-dent to first create an outline of the composi-tion. A topic is to be chosen, and the paperdivided into introduction, body, and conclu-sion. The body of the paper is further sub-divided into paragraphs of thesis support, eachmaking a main point stated in the paragraph'stopic sentence and supported in its body. Allthat then remains is the "fleshing out" of theoutline. (See also Emig, 1971, for a relateddiscussion of rhetoric texts.)

There are limits, however, to the ap-propriateness of hierarchical models forwriting. Recall that hierarchical plannersoperate according to the least commitmentprinciple keeping variables unspecified for asbng as possible. Only so much planning of acomposition, however, can be done in theabstract, even by skilled writers. Relatively ear-ly the writer is forced to define variables (i.e.,to actually write a sentence or a few words),and this often occurs before every paragraph isfully planned and waiting to be "dressed" inthe appropriate words. Words already writtencan drastically affect what follows them; infact. they must if smooth transitions and local

228

coherence are to be maintained (Halliday &Hasan, 1976) With such early constraints onvariables, the writer loses the power of thehierarchical planners. The writer is forced, atsome point in the actual generation ofsentences, to follow the linearity assumptiontypical of nonhierarchical planners, choosing tobegin with something and following it,sometimes to a preplanned next idea,sometimes to a newly discovered thought, andsometimes to a dead end. The nonhierarchicalaspects of writing are all too many, as data fromHayes and Flower's (1980; Flower & Hayes,1981 b) protocol studies reveal. Low level editingfrequently interrupts the planning andgenerating processes, and in most research onwriters' actual composing behavior, emphasis isplaced on the interactive nature (in thepsychological sense of multiple informationsources) of the subprocesses of writing (E. J.Bartlett, 1982; Beaugrande, 1982; Burtis et al.,in press; Emig, 1971; Hayes & Flower, 1980;Matsuhashi, 1982; Nold, 1981; Shuy, 1981).

The interactive nature of writing. The interac-tion between text-level processes and planning-level processes is well illustrated in the followingexcerpts from the protocol of a writer of anewspaper wine column. This writer began theprotocol with a well formed plan concerning au-dience and style, even specifying the structure ofhis column about a tasting of wines fromChateau Latour: "The general structure has gotto be, we've got to give them some informationabout Chateau Latour, make it kind of real tothem, give them something to chew on, andthen we'te going to go through the tasting notes. . . ." Even with this plan, however, text-leveldecisions had to be made, as his protocol shows.

In Figure 1, the writer knows the content hewants to express, but it is the expression, inlinearly structured sentences, that gives himtrouble. (The section of text on which thewriter is working is presented on the right sideof the figure, and the writer's comments onthe left.)

The writer's plan had specified that he gavesome information about Latour, specificallythat 80% of the Latour vineyards are plantedin cabernet grapes and that this is the source ofthe wine's longevity. However, an appropriatesentence structure coordinating those two ideasdoes not just fall out of his semantic plan, andwe see the writer try one alternative afteranother. Constructing appropriate sentences ispart of the writing task.

7

229 WRITING AS A LINGUISTIC PROBLEM

PROTOCOL

That doesn't read too well at all, but it's theright idea. So why not break it up . . .

Now we can get the 80% in, OK? (reads)"80% of the grapes? Cabernet? 80%" 1want to say it's the vineyard that's cabernet,but we just said it's the vineyard, and thatgets boring with too many "vineyards." . . .

So why not just make it a nice run-onsentence . . . That way we don't have torepeat.

TEXT

. . . Probably one of the major reasons forthe longevity of the wines lies in the 80%cabernet sauvignon grapes

(edits) . . . lies in the vineyards of Latour.80%

(edits) . . lies in the vineyards of Latour.80% of which are given over to the cabernetgrape.

Figure 1. Adult writer's protocol and text as the writer works out sentence syntax.

PROTOCOL

Well now, how would one describe thegrape? Wildness? See, what we have to donow is tell them why it is that cabernetsauvignon gives it the longevity. Why doesit? Because it is a hard grape. It takes a longtime to come around. OK . . . (reviews)"This is the grape that . . ." ah, "providesbackbone . . ."

. . . the vineyards of Latour, 80% of whichare given over to the cabernet sauvignongrape. (types) This is the grape used in thehardest

(edits) This is the grape that providesbackbone . . .

Figure 2. Adult writer's protocol and text as the writer uses the text to refine conceptual plan.

The protocol continues in Figure 2, and herewe can see that the writer's linear generationprocesses have outrun his plan for content. Hethen uses the text he has written to better for-mulate his idea and help retrieve content.

It is probably not a coincidence that theword "hard," appearing first in the writtentext, is used as a prompt in a memory searchfor better descriptors. Here the processes of

writing connected sentences has led the writerto a point where his high level plans were notwell specified, and his written sentences ac-tually help achieve the gual by providing aprompt, a jumping-off point from which tobegin some new semantic planning. In thesetwo excerpts we see different types of text -levelprocessing, and the interaction among them isstriking. In Figure 1 the writer knew oh:

8

DEBORAH McCUTCHEN

concepts that he wanted to express and evenin:my of the specific lexical items; however, hehad not worked out the syntactic frames forthose lexical items. In the course of the protocolwe set the writer work out an appropriate syntax, guided by constraints from the lexical level,constraints against too much word repetition.

In Figure 2 something slightly different oc-curs. There is a general semantic plan for thesection (i.e., to explain the connection be-tween cabernet grapes and a wine's longevity),but the writer has neither the specific syntax,concepts, nor the lexical items to express them.As we see in the sentence fragment that endsthe fi.:t segment of text, the writer seemssimp'y to begin the new sentence, building iton roe sentence just completed, and he goes asfar as he can with it. His first pass at thesentence is not in the form he will actually use,but that first expression gets him into the ap-propriate semantic field and enables furtherrefinement of the semantic plan.

In this second excerpt we see an example of"text-based" writing used to its fullest advan-tage by an expert writer, and it illustrates howtruly interactive writing can be. This writerbegan with a well-formed plan of the generalstructure of the column and of the audiencewho would read it. Even experts, however,cannot plan in advance every elaborative detailthat might become appropriate as the textdevelops. As we see in Figure 2, the text itselfcan influence phrasing and can even promptthe writer to pursue an idea that was notsalient in the initial plan.

Thus, to accurately describe the behavior ofwriters, planning systems such as thosedeveloped within Al must be altered substan-tially. The hierarchical models that are powerfulenough to solve relatively complex problems donot fit very well the task of generating coherenttexts. Even descriptions of human experts per-forming the superficially similar task of softwaredesign (Jeffries et al., 1981) seem quite dif-ferent from expert writing performance. Thesolution processes of software design expertsseem somewhat hierarchical in nature, whilethose of writers are much more interactive. Thedifferences are probably due to the symboliccodes required in the two tasks. Writing com-puter code is not like writing natural languagebecause the syntax of computer code is fixed.Once the general semantics of a computer pro-gam have been worked out, translation intocode may be a rather trivial problem, at least forprogrammers of reasonable skill. For the writer

230

of natural language, however, syntax is not fix-ed. There are a variety of lexical and syntacticforms that can be used to express the samegeneral semantic concepts (as our columnistdemonstrated). Most important, those varioussyntactic forms render the semantic conceptsno longer exactly equivalent. A passive construction, for example, signals a differentsentence focus than does an active construc-tion. Because natural language permitsnuances of meaning that computer codes donot, rearranging syntax can result in subtlechanges in theme or foregrounding that can af-fect the reader's comprehension (Chafe, 1972;Halliday, 1967; Halliday & Hasan, 1976;Lesgold, Roth, & Curtis, 1979).

In writing, unlike some tasks, decisions atthe most detailed level of word choice andsentence construction can have large effects onabstract goal outcomes such as tone, perspec-tive, and audience. This is because those goalsare fully achieved only at the most local level.For example, our wine columnist proposed tocontinue his explanation of wine's longevitywith a discussion of esters and aldehydes.Reconsidering his purpose and audience, hechose instead to refer to "smells and flavors,"because he wanted only a brief reference tothose concepts. The concepts themselves werenot ruled out by his general plans, but thewriter had to decide on which aspect of theconcepts to focus (on their chemical basis ortheir perceptual qualities) and on the cor-respc'iding lexical labels ("esters andaldehydes" or "smells and flavors"). Inap-propriate choices at that final level of specification could have undermined his plans concern-ing audience and purpose.

Language-Based Approaches

Because of the distinctive features ofnatural language production. linguistic tasksmay be much more like one another than theyare like other problem solving tasks. Thus,work in reading and speech production mayprovide additional models for studyingwriting. In work on reading comprehension,two perspectives have emerged, and while theyare often viewed as adversative, they areactually rather complementary. The "top-down" approach emphasizes the importanceof the reader's knowledge and its schematicorganization, while the "bottom-up" ap-proach emphasizes lower level linguistic

2; I WRITING AS A LINGUISTIC PROBLI:NI

processes./ The model emerging from readingresearch reconciles these approaches by em-phasizing interactions among processes at alllevels. Thus, both top-down and bottom-upprocesses have been shown to contribute tosuccessful comprehension, and there may becorresponding contributions, as well as interac-tions, at each level in the case of writing.

Top down Processes

Top down approaches to reading. Top-downinformation plays an important part in reading,for example the role of schemata in guidingcomprehension. A schema is a hypothesizedknowledge structure that connects events orconcepts in some organized arrangement.Popular examples of such schemata are the col-lections of events that typically make up a story(setting, characterization, complication, andresolution) or a visit to a restaurant (ordering,eating, paying, and leaving).

The usefulness of the organizational proper-ties of schemata for comprehension and recallhas been repeatedly demonstrated. Textsordered according to typical narrative schemataare consistently better recalled than those withunusual orders (Stein & Nezworski, 1978:Thorndyke, 1977), and recall of discourse thatdoes not follow the ideal schematized ordertends to be restructured more in accordancewith that order than the actual order of input(Bower, Black & Turner, 1979; Mandler, 1978;Stein & Nezworski, 1978; Thorndyke, 1977).Recall also suffers when appropriate schemataare unavailable (F. C. Bartlett. 1932;Bransford & Johnson, 1972), and recall is bet-ter when events in stories are logically relatedrather than loosely temporally ordered (Ander-son, Spiro, & Anderson, 1977, in Anderson,1978; Black & Bern, 1981; Brown, 1976; Kint-sch, Mandel, & Kozminsky, 1977).

In addition, a concept's probability of recallor inclusion in a summary increases when thatconcept is judged (by various indices) to bemore important or more central to the schema(Brown & Smiley, 1977; Johnson, 1970;

Omanson, 1982; Rumelhart, 1975). Similarly,the schema instantiated during comprehensioncan have dramatic effects on which conceptsare recalled and on their interpretation(Anderson, Reynolds, Schallert, & Goetz,1977; Pichert & Anderson, 1977. in Anderson,1978). Comprehension has, in fact, been

defined by some as instantiating the

appropriate schema and mapping the incom-ing information unto the various slots (Collins,Brown & Larkin, 1980; Rumelhart & Ortony,1977: Schank & Abelson, 1977). While others

argue that there is more to comprehension andreading skill that' top-down knowledge(Perfetti, in press; Perfetti & Roth, 1981,

Stanovich, 1981), it is generally acknowledgedby researchers from all perspectives thatschemata are quite useful in organizing newinformation, in relating it to the reader'sgeneral knowledge during reading, and in ac-cessing that information during recall.

Schemata in composition: applications andlimitations. Tilt role of schemata in memoryaccess and retrieval during reading suggeststhat schemata may also be useful in writing.There is some suggestion that schemata can actas regulators for the arrangement of textelements in original text generation, as well asrecall (Paris, Scardamalia, & Bereiter, 1980, inBereiter & Scardamalia, 1981; Stein & Glenn,1979; Waters, 1980). Meehan (1981) hasfound schemata to be necessary knowledgecomponents of his story-generating computerprogram, and Black, Wilkes-Gibbs, arid Gibbs(1982) describe how schemata (and deviationsfrom them) can help a writer determine an ap-propriate level of detail to focus interest, createdrama, and hold interest.

Certainly a key role for schemata duringwriting is the activation of relevant schematiccontent. Schemata may be very much in-volves:, therefore, when writing becomes aprocess of discovery of ideas rather than meretranscription, and there may be something tothat common observation, "I don't know whatI think until I write it down." Schemata mayaid in memory search, since they containpointers to yet unaccessed information in thewriter's memory and thus facilitate retrieval oftopic relevant information.

II-he bottom-up approach to reading has often beencharacterized as primarily emphasizing decoding and otherword-level processes. And much of the work on individualdifferences in reading ability has indeed focused on theimportance of such low level processes (e.g . Hunt.Lunneborg. & Lewis, 1975; Perfetti 8c Lesgold, 1979).'This, however, has led to a misconception7 that in thebottom-up approach. decoding is all there is to comprehension. On the contrary, these lower points in the ver-bal processing chain are emphasized only as potential proc-essing weaknesses that can. if not automated. drain

ognitive resources away from the textlevel, integrativeprocesses that are critical for comprehension (Perfetti. inpress: Perfetti & Lesgold. 1977).

0

DEBORAH Want-1EN 232

Once relevant semantic content is activated.there still remains much for the writer to do interms of translating dusters of semanticknowledge into actual ICVI. An attractivefeature of schemata is that they deal withsemantic information at an abstract level.However. the processes of translating semanticconcepts and relations into grammaticalnatural language sentences are not wellspecified in the schema-oriented work. Seman-tic relations arc well expressed by propositions

relations between predicates and nounsbut even when relations among semantic con-cepts have been irganized into lists of proposi-tions. there is still no natural language text.The utility of a propositional representation isthar it can be mapped into a variety oflinguistic expressions, all paraphrases of eachother (see Kintsch. 1974: Kintsch & van Dijk,1978). Generating those linguisticallyspecified Alternatives And choosing the onemost appropriate in a given linguistic contextcomprise much of the writer's job, and thatjob is often not an easy one.

As our wine columnist found, it is notalways easy for the writer to choose the linearsyntactic arrangement that best expresses theconceptual relations and still honors local con-straints, especially since the writer is rarelygenerating a single sentence in isolation. Thewriter usually tries to generate a connecteddiscourse and is thus foced to deal with howextended texts "work," that is, how thespecific wording of the message places someconcepts in the foreground, others in thebackground. and integrates them all. Thus, inaddition to the insights into writing gainedfrom schemaoriented, top-down models ofcomprehension, there is much to learn fromwork that focuses on how meaning depends onthe specific wording of texts and how specificwordings can affect processing.

Bottom up Processes

Role of linguistic text features in reading.The bottom-up approach to reading com-prehension has focused on the text itself andhas emphasized many concepts developed inlinguistics. For example, linguistic ideas ofsentence perspective have been discussed atlength by Halliday (1967; Halliday & Hasan,1976). Halliday (1967) distinguished severalrelated concepts: information focus, whichis indicated by tonal groups in speech;

themaination, realized I ty order of clause con-stituents; identification, realized by specialmarkings of "identified and identifier" incleft and pseudocleft constructions; andgiven/new, which is based solely on whether ornot specific information has been previouslypresented in the discourse. Many of these ideashave been incorporated into theories of com-prehension (e.g., Just & Carpenter, 1980;Perfetti & Lesgold, 1977), and they have beenthe subject of empirical investigation.

In one such line of research, Clark andHaviland (1977) proposed a model of con-nected sentence understanding which they callthe given/new strategy. The listener or readerattempts to match the given information ineach sentence with some information alreadyin memory. If that match is successful, the newinformation is added to memory. lf, however,the match is unsuccessful, added processing isrequired to make a bridging inference orrestructure the original given/ new assignmentsin the sentence. Reading times lend plausibili-ty to such a hypothesis. Reading times werefound to be shorter when syntactically in-dicated sentence parsings were appropriate tothe given/ new semantics of the passage.Similarly, Hornby (1974) found indicationsthat cognitive processing was influenced by thelinguistic presuppositions of the sentence syn-tax, and Sanford and Garrod (1981) have pro-posed a model of comprehension that deals, atthe level of specific wording, with such text-based processes as inferencing and assigningpronominal reference.

Chafe (1972) discussed the related conceptof foregreunding, which entails the linguistic"staging" of certain lexical items and allowstheir being treated as given in the followingutterance. Translated into cognitive processingterms, foregrounding helps to mark some lex-ical items for inclusion in STM while others arebackgrounded. Thu::, reading times shoulddecrease for foregrounded information, andthis was found to be the case (Lesgold et al.,1979; Sanford & Garrod, 1981).

Lingusitic text features: implications forwriting. The implications of this work forwriting are two-fold. First, the writer shouldwant to create texts that most effectively com-municate ideas to the reader. Thus thoselinguistic features of text that affect a reader'sprocessing should be important to the writer aswell. It may not be the case, for stylistic orother reasons, that the writer consistently

11

2 WRITING AS A LINGUISTIC PROBLEM

makes the reader's job as easy as possible.However, as the writer searches for various syn-tactic constructions, he or she should be con-tinually ware, at some level, of the subtlechanges in thematization or information focusthat are cued by alternative syntactic construc-tions. A writer's control over local text featurescan thus affect the quality of the written pro-duct, and analyses of local text coherence hasshown that such local control seems to contri-bute to developmental differences observed inwriting by children (McCutchen & Perfetti,1982) and to perceived quality differencesin writing by college students (Witte &Faigley, 1981).

The second implication of these linguistictext features concerns the processing of thewriter rather than the reader. The writer, afterall, becomes a reader during the repeatedcycles of generating, translating. and reviewingthat comprise the process of writing (Hayes &Flower, 1980), and thus the writer can be af-fected by many of the same text features.Pertetti and Goldman (1975) observed thatreaders' preferences for syntactic alternativeswas indeed influenced by the syntacticthematization of sentences that precededthem. We also saw evidence of this interactionbetween the developing text and the writer'smore general semantic plans in the writing ofour wine columnist (see Figure 2).

The writer's realization of how a text isworking linguistically can be very useful. Withthis information, the writer can understandthe syntactic reasons why a text seems to be"going nowhere" or even going somewherethe writer does not intend. Understanding thesyntactic reasons for the problems, the writermay then know better how to solve them.

The ability to ultimately solve such writingproblems may critically depend on the writer'sfluency in the processes of linear sentencegeneration: encoding concepts into actual lex-ical items, formulating clause-level syntacticarrangements, and then morphologicallymanipulating the lexical items to fit the syn-tactic frames. It is fluency in linear sentenceproduction that aids manipulation ofsentences and thus ideas. Just as the imposi-tion of high level schemata may organize infor-mation in interesting and sometimes unex-pected ways, lexical and syntactic manipula-tions at the local text level may also result infresh juxtapositions of concepts that the writercan then evaluate for style, clarity, direction,or even plausibility.

Only with reasonable fluency and cognitiveefficiency in processes at the local text level.however. can the writer afford to play such ex-perimental linguistic games with the text."Writing as discovery" is simply too cngnitive-ly expensive for the writer with limited fluencyin linear sentence processing. Young writersmight be at a special disadvarage here not on-ly because of their limited syntactic fluency butalso because of their limited syntactic reper-tory. Even when children can recognize flawsin their writing, they often cannot proposealternative constructions that remedy the prob-lems (E. J. Bartlett. 1982: Bereiter &Scardamalia. 1981). The writer may even, insome sense, know alternative word chokes orsyntactic constructions, but when sentenceproduction is cognitively inefficient thegeneration of sentences may proreed on a"first come, first served" basis,regardless of appropriateness within thespecific linguistic context. This toocan be problematic, especially for the youngwriter, since studies by Bracewell andScardamalia (reported in Bereiter & Scar -damalia, 1981) showed that children have par-ticular trouble linguistically recasting sentenceswhen alternative linguistic forms of thesentences are present. Thus, if text-level pro-cesses are not well under control, the writermay simply not risk local manipulations, and ifthey are attempted, faulty local processing mayresult in the errors so typical of problem writers(see Bartholomae, 1980; Daiute, 1981;Shaughnessy, 1977).

It is the very fluency of most writers' linearsentence processing, successful or not, thatmay make it difficult to identify their impor-tance in the writing process. Our wine colum-nist was extraordinarily verbal about some ofhis text-level decisions, but this was not true inmuch of his protocol, nor in the protocols ofmany other skilled writers. In a study of pausesduring writing, Flower & Hayes (1981b) foundthat higher level, rhetorical goals correlatedbetter with pause-bordered episodes than didlocal, sentence-level decisions. This is not sur-prising. Especially for the adult writers in thatstudy (several of whom were classified as expertwriters), one might expect that sentence-leveldecisions would not account foe large propor-tions of pause time, compared with rhetoricaldecisions. These writers may be so fluent withlocal text manipulations that those sorts ofdecisions are very rapid and not as availablefor report.

12

DEBORAH WU:I-CHEN 234

PROTOCOL

That doesn't sound good at all! . . . I'll haveto start the sentence with diffetent, urn, dif-ferent words because, um, it says "There'smany different places to skate" and it reallydoesn't fit right there. "There's also manydifferent places to skate."

TEXT1) Roller-skating is fun and exciting.2) Bechuse you can skate with slot of people.3) There's many different places to skate.4) To roller -skate you use tennis shoes withwheels.

(edits)

3) There's also many different places toskate.

Figure .c Fourth grade writer's protocol and text as the writer explicitly coordinatesadjacent sentences.

Text level processes in young writers. Forwriters with less skill and less experience,however, such sentence-level decisions are notso fluent. and those decisions are observableparts of the writing process. A close look at aspontaneous editing session of a fourth gradergives some insights into her writing processand into aspects of performance not observablein more fluent writers.

In Figure 3 and 4 we see, on the right, thetext produced so far in the writing session and,on the left, the writer's comments at thatpoint. The sentences of the text have beennumbered here for ease of reference.

In Figure 3, the writer has reread her firstfour sentences and is dissatisfied. She is strug-gling with local coherence problems betweensentences 2 and 3, and she solves them with astrictly local, $entence-level change. She insertsthe word "also" into sentence 3 to explicitlymark the coordination of ideas betweensentences 2 and 3. As her protocol reveals,text-level decisions take much of this writer'sattention as she tries to generate sentences that"fit" with their neighboring sentences.

In Figure 4, the writer has deleted sentence4 altogether, after several attempts to reword itand "start it out different." Her commentsshow that she has decided what semantic con-tent she would like to include in her nextsentence, but the local decisions of "how towrite it" are very difficult for her. She hasdecided to extend her discussion of "places toskate" by mentioning streets and hills, but she

struggles at the level of phrasing and questionshow explicit she must make the link between"places" and "streets and hills."

For this young writer, whose linear sentenceprocessing is not fluent, text-level decisions arevery prominent aspects of the writing process.In addition, young writers are notorious fortheir lack of high love: planning (Burtis et al.,in press; Scardamalia & Bereiter, 1982), andthus their text-level decisions might be evenmore difficult, operating without the guidanceof surnrordinate plans. For ter, more fluentwriters, such text-based processing may nolonger be prominent aspects of writing, ob-servable in protocols, but they certainly mustremain important parts of the process ofsentence generation and thus important pansof writing.

Speech Production Prormes

The writer's job, in some respects, is notunlike the speaker's job: The goal of both is togenerate a linguistic expression. Of course,unlike the speaker who produces a transientacoustic signal, the writer produces an endur-ing written transcript that can be reexaminedand edited to improve its fit within a givencontext. With the luxury of revision, the writercan alter the text so as to most effectively com-municate with the reader.

Like speech, however, the writing of sen-tences requires encoding semantic concepts

13

235 v;'ItITING AS A LINGUISTIC PROBLEM

PROTOCOL TEXT1) Roller-skating is fun and exciting.2) Because you can skate with alot of people.3) There's also many different places toskate.

l'ra trying to figure out how write ithowto put it downto fit with, um See, it'll fitmore with, urn, "many different places toskate." It'll fit with that, like, 'cause, um,flat hills are different placeswell streets aredifferent places, hills are different places . .

. I'm trying to get this sentence well, it'llfit this sentence, but should I write um,"Hills are steep. and they're saucy"? Wouldthat make senseto makewith thissentence? Or just write "Hills areHills andstreets are different places to skate"?

hi tire 4. Fourth grade writer's protocol and text as the writer works out syntactic frame for chosenconce ts.

into actual lexical items and arranging them ingrammatical sequences that best express theirsemantic relations. Research cis speech produc-tion shows that the process is not one of directtranslation. Linear sentence generation re-quires much interplay among semantic, syn-tactic, and lexical levels, as evidenced byspeech error data (Fromkin, 1973, 1980;Garrett, 1981; Levelt. 1983).

Various kinds of speech errors suggest thatthere are multiple stages in the process ofsentence production. For example, word ex-changes, as in sentences (1) and (2) (fromGarrett, 1981), tend to be between words ofthe same grammatical category, suggestingsome syntactic framing had occurred prior tothe point at which lexical items were insertedinto the frame.(1)Older men choose to tend younger

wives.(intended: tend to choose)

(2)Write a request for tickets at two for thebox office.(intended: tickets for two at the box of-fice)

Other errors, called stranding errors, suggestthat the bound morphemes marking gram-nmical function are partly independent of thelexical items with which they are paired andare ,perhaps connected more intimately withthe syntactic frame itself. In these errors, wordstems exchange places but leave behind,"stranded" in the original syntactic position,

the bound morpheme that serves as the gram-matical marker. Further, as shown in sentence(3) (from Garrett, 1981), vccy late in the pro-duction process those stranded morphemes areaccommodated to their lew phonological en-vironment.

(3) It waits to pay.Is/

(intended: pays to wait)/z/

Such errors have prompted theories ofspeech production that involve several rapidlyexecuted stages in which general semantic con-tent is chosen first, then individual concepts or"lemmas." Then clause-level syntactic framesare specified, individual lexical items retrieved(corresponding to the semantic lemmas), andfinally some morphological adjustments madeto fit the words into the specified frame.

Protocols from writers such as our wine col-umnist also suggest the existence of severallevels in the sentence production process. Inthe excerpt in Figure 2, the writer was search-ing for semantic concepts and words to capturethem, while in Figure 1, he had the concepts,words, and much of the syntax but was work-ing on subtle syntactic refinements that avoid-ed word repetition.

Although we do not usually see word ex-changes and stranding errors occurring inwriting, speech production models have beenapplied to writing ,vith some interesting results.

14

DEBOkAII McCUICIWN 236

For example, one theory of speech production(Beyer, Carroll. & Hurtig. 1976) explainsmany syntactic errors, such as sentence (4), interms of overlapping syntactic frames that,when combined, result in an ungrammaticalmerging of two potentially grammatical se-quences. Sequences (4a) and (4b) ar- bothgrammatical, but when connected via theircommon segment. "1 understand." theyviolate syntactic rules.

(1) I really enjoyed flying in an airplane thatI understand how it works.

(4a) I really enjoyed flying in an airplane thatI understand.

(4b) I understand how it wurks.

Daiute (1981) has applied similar analyses tostudents' written sentence errors and was ableto account for a large proportiJn of thestudents' syntactic errors.

Conclusions

The argument presented here has someprecedents. (See Bracewell, 1980, for a relateddiscussion.) In fact. a similar point has beenmade by Flower and Hayes (1981b), majorproponents of the problem solving approach towriting whose recent work focuses ptimarily onthe role of planning in writing skill:

an important pan of being a skilled writer is know.mg not only how to do this rhetorical planning. buthow to embed sentente.leyel planning within ithow to turn intentions and knowledge into text.

So much empirical attention, however, hasbeen focused on the planning cemponent oftheir problem solving model that the translatingcomponent often seems trivial in comparison.Studies in speech production and reading com-prehension remind us that generatinglanguage, even with the help of appropriateplans, is a nontrivial task and that linguisticfeatures of the text affect processing in impor-tant ways. Thus the translating of plans andgoals into text is an important part of thewriting process, and the interaction betweenhigher-level plans and linguistic features of thedeveloping text are a worthy research focus.

A focus on the linguistic nature of thewriting process will prompt research to addressquestions that are differenct from those posedin a planning-oriented view. For instance, does

sentence generation during writing follow thecourse hypothesized in speech productionstudies? It may be the use that the slowedpace of writing and the reflection it permits,combined with the written transcript it leavesbehind, alter the process. In the less transientenvironment of written text, sentences occur-ing earlier in the discourse may affect on-lineproductions in ways that spoken sentences can-not. On the other hand, writing might notsubstantially change the process; it mightsimply make it easier to track and thus helprefine theories of sentence production.

When the focus is on how semantic contentis translated into language, issues also arise asto how (and how well) coherence among thesemantic concepts is represented throughcoherence in the text itself. Such issues includehow do linguistic devices maintain textualcoherence and how intimately are suchlinguistic devit tied to the semantics of thecontent. Since .tilled writers seem to be ableto transfer at le: It part of their skill acrossknowledge domains, one might be tempted tohypothesize that SLIM aspects of coherentwriting are independent of content. This implies that, in an effort to create a coherent text,the good writer somehow recognizes areas ofignorance and (a) either avoids or "writesaround" them, or (b) clears them up in theprocess of writing. The second alternative isclearly the most interesting. Many writers havehad the experience of cry.tallizing ideas onlyonce hey begin to write them down, andthe role that language generation itselfplays in this process is a most in-triguing question.

Also interesting is the development of theability to view language as separate from thecontent it expresses. How and when does thewriter, or the language user in general, beginto represent language as opaque, as somethingthat can be crafted to better express givensemantic concepts rather than just atransparent window on those concepts? For thenovice language user, the emphasis is usuallyon the message, but the writer must focus onthe linguistic expression of that message aswell. Understanding how "what is said" dif-fers from "what is meant" is a critical partof writing.

Thus, focusing on writing both as a textdriven linguistic task and as a planningtask, writing researchers may begin to get amore comprehensive understanding of thewriting process.

15

237

References

WRITING AS A LINGUISTIC PROBLEM

Anderson, R. C. (1978). Schemadirected processes inlanguage comprehension. In A. M. Lesgold, J. W.Pellegrino, S. D. Fokkema, & R. Glaser (Eds.),Cognitive psychology and instruction. New York:Plenum.

Anderson. R. C., Reynolds. R. E., Schallert, D. L., &Goetz, E. T (1977). Frameworks for comprehendingdiscourse. Amerman Edmcational Research Journal 14.367.382.

Bartholomae, D. (1980). The study of error. CollegeComposition and Commamication. 31, 253.269.

Bartlett, E. J. (1982). Learning to revise: Some componentprocesses. In M. Nystrand (Ed.), What writers know.New York: Academic Press.

Bartlett, F. C. (1932). Rentembenrag. Cambridge: Cam.bridge University Press.

Beaugrande, R. de. (1982). Psychology and composition:Past, present, and future. In M. Nystrand (Ed,), WhatWWII know. New York: Academic Press.

Bereiter, C., & Scardamalia, M. (1981). From conversation to composition: The role of instruction in adevelopmental process. In R. Glaser (Ed.), Advances inartructional psychology (Vol. 2)'. Hillsdale, NJ:

Erlbaum.Bever, T. G., Carroll, J. M., & Hunig, R. (1976). Analogy

or sequences that are utterable and comprehensible arethe origins of new grammars in language acquisitionand linguistic evolution. In T. G. Bever, J. J. Katz, &D. T, Langendoen (Eds.), An integrated theory oflanguage behavior. New York; Crowell.

Black, J. B., & Bern, H. (1981). Causal coherence andmemory for events in narratives. Jommal of VerbalLamm and Verbal Behavior, 20, 267-275.

Black, J. B., WilkesGibbs, D., & Gibbs, R. W. (1982).What writers need to know that they don't know theyneed to know. In M. Nystrand (Ed.), What writersknow. New York: Academic Press,

Bower, G., Black, J. B., & Turner, T. J. (1979). Scripts inmemory for texts. Cognitive Psychology, 11, 177.200.

Bracewell, R. J. (1980). Writing as a cognitive activity.Visible Language, 14, 400-422,

Bransford, J. D., & Johnson, M. K. (1972). Contextualprerequisites for understanding: Some invest; ationsof comprehension and recall. Joarmal of Verbi LearnMg and Verb's! Behavior, 11, 717.726.

Brown, A. L. (1976). The construction of temporal succes-sion in preopctational children. In A. D. Pick (Ed.),Minnesota tymposimm on child psychology (Vol. 10).Minneapolis: University of Minnesota.

3rown, A. L., & Smiley, S. S. (1977). Rating the impor-tance of structural elements of prose passages: A prob-lem of metacognitive development. Child Developmew. 48, 1.8.

Burns, P. J., Bereiter, C., Scardamalia, M., & Tetra, J.(in press). The development of planning in writing. InC. G. Wells & B. Kroll (Eds.), Exploration ofchild,-en's development in writing. Chicester,England: Wiley.

Chafe, W. I.. (1972). Discourse structure and humanknowledge. In J. B. Carrol & R. 0. Freedle (Eds,),&visage comprehension and the acquisition ofknowledge. Washington, D. C.: Winston.

Clark, H. H., & Haviland. S. E. (1977). Comprehensionand the givennew contract. In R. 0. Freedle (Ed.),Discosirre prodriction and comprehension. Norwood,NJ Ablex.

Cohen, P. R., & Feigenbaum, E. A. (1982). Planning andproblem solving. In The handbook of artificial intelligence (Vol. 3). Los Altos, CA: Kaufman.

Collins, A., Brown, J. S., & Larkin, K. M. (1980). Infer-ence in text understanding. In R. J. Spiro, B. C. Bruce,W. F. Brewer (Eds.), Theoretical nines in reading cornprehension. Hillsdale, NJ: Erlbautn.

Collins, A., & Gentner, D. (1980). A framework for acognitive theory of writing. In I. W. Gregg & E. H.Steinberg (Eds.), t'ognitive processes in writing.Hillsdale, NJ: Erlbaum.

Daiute, C. A. (1981). Psycholinguistic foundations of thewriting process. Research in the teaching of English./3, 5.22.

Emig, J. (1971). The composing processes of twelfthgraders Research Report No. 13. Champaign, IL: National Council of Teachers of English.

Flower, L. (1981). Problem solving strategies for writing.New York: Harcourt Brace Jovanovich.

Flower, L. S., & Hayes, J. R. (1980). The dynamics ofcomposing: Making plans and juggling constraints. InL. W. Gregg & E. R. Steinberg (Eds.), Cognitive pmasses in writing. Hillsdale, NJ: Erlbaum.

Flower, L., & Hayes, J. R. (1981a). Plans that guide thecomposing process. In C. H. Frederiksen & J. F.

Dominic (Eds.), Writing: The :mum development.and teaching of written corn vs mirication (Vol. 2).Hillsdale. NJ: Erlbaum.

Flower, L., & Hayes, J. R. (1981b). The pregnant pause:An inquiry into the nature of planning. Research inthe Teaching of English, 15. 229.244.

Flower, L., & Hayes, J. R. (1984). Images, plans. andprose: The representation of meaning in writing. Irmiten Communication, 1, 120.160.

Fromkin, V. A. (Ed.) (1973). Speech errors as lingua/K.evidence, The Hague: Mouton.

Fromkin, V. A. (Ed.) (1980). Errors in linguistic perfOrmance. New York: Academic Press.

Garrett, M. F. (1981). Levels of processing in sentenceproduction. In B. Butterworth (Ed.), tannage proaction. New York: Academic Press.

Halliday, M. A. K. (1967). Notes on transitivity andtheme in English, Part 2. Journal of Lingruitics, 3,199.244.

Halliday, M. A. K., & Hasan, R. (1976). Cohesion inEnglish. London: Longman.

Hayes, J. R., & Flower, L. S. (1980). Identifying theorganization of writing processes. In L. W. Gregg & E.R. Steinberg (Eds.), Cognitive processes in writing.Hillsdale, NJ: Erlbaum.

HayesRoth, B., & HayesRoth, F. (1979). A cognitivemodel of planning. Cognitive Science, 3, 275.310.

Hornby, P. A. (1974). Sarface structure and presupposition. Journal of Verbal Learning and Verbal Behavior,13, 530.538.

Hunt, E., Lunneborg, C., & Lewis, J. (1975). What doesit mean to be high verbal? Cognitive Psychology, 7,194.227.

Jeffries, R., Turner, A. A., Poison, P. 0 & Atwood, M.E. (1981). The processes involved in designing soft-ware. In J. R. Anderson (Ed.), Cognitive skills andtheir acquisition. Hillsdale, NJ: Erlbaum.

Johnson, R. E. (1970). Recall of prose as a function of thestructural importance of the linguistic units. Journal ofVerbal Learning and Verbal Behavior, 9, 12.20.

Just, M. A., & Carpenter, P. A. (1980). A theory of read-ing from eye fixations to comprehension. PsychologicalReview, 87, 329.354.

16

DEBORAH Mt CM F 238

Kintsch. W (19,'4) 7br reprriemiattou meamirtg inmentors Hillsdale. NJ blbaum

Kintsch. W , & van DIX T A (19'8). Toward a modelof text comprehension and production hycholograReview. 83. 363.394.

Kintsch. W . Mandel. T S.. & Kozminsky. E (1977)Summariting scrambled stories. Memory & Cogintros.3. 547552

Larkin. J H., McDermott.) . Simon. D P.. & Simon. HA (1980) Expert and novice performance in solvingphysics problems. Science. 208, 1335.1342

Lesgold, A. M , Roth. S. F.. & Curtis. M E. (1979).Foregrounding effects in discourse comprehension.

Journal Verbal tearrtiug and Verbal Behavior. ;8.291. 308.

W J M (1983). Monitoring and selfrepair inspeech Cogmatov. 14. 41.104.

Mandler. J. M. (1978) A code in the node. The use of astory schema in retrieval. &MAIM Prucerrer, 1. 14.35

Matsuhashi. A. (1982). Explorations in the real timeproduction of written discourse. In M. Nystrand (Ed.).What wirier, brow New York: Academic Press.

McCutchen. D . & Peden'. C A. (1982) Coherence andconnectedness in the development of discourse production Text 2. 1 1 5.139

Meehan. J 11981) Talespin. In R C Schank & C. K.Riesbeck (Eds ). !utile computer muientworinig.

NJ ErlbaumNold. E W (1981) Revising. In C. H Frederiksen & J. F.

Dominic (Eds.). 11nitsg: The nature. developmest.and se.khrug 01 unties communcutom (Vol. 2).Hillsdale. NJ Erlbaum

Onianson. R C 11982). An analysis of narratives:Identifying central. supportive, and distracting con.tent Dryostrie Procerier. 3. 195.224.

Perfetti. C. A (in press) Reading Jetility. OxfordPress

Perfetti, C A & Goldman. S. R. (1975). Discoursefunctions of thematization and topicalization. fusirmalref Pocholtuguntic Research. 4. 257.272.

Perfetti. C. A . & Lesgold. A. M. (197') Discoursecomprehension And sources of individual differences.In M A Just & P. A Carpenter (Eds ). Cuguitive procenet rs comprehemuom Hillsdale. NJ: Erlbaum.

Peden'. C. A , & Lesgold. A M. (1979). Coding andcomprehension in skilled reading and implications forreading instruction In L. B. Resnick & P. Weaver(Eds ). Theory asd practice is early reading. Hillsdale,NJ Erlbaum.

Peden!. C A.. & Roth. 5 (1981). Some of the interactiveprocesses in reading and their role in reading skill. InA M Lesgold & C. A. Perfetti (Eds.), Interactive pmcener rs ream. Hillsdale. NJ: Erlbaum.

Rumelhart. D. E. (1975). Notes on a schema for stories. InD (i Bobrow & A M Collins (Eds ). Representationold mudentuatmg Didier is squirm science. NewYork: Academic Press.

Sumelliati. U. E., & Ortony, A. (1977). Representationof knowledge. In R. C. Anderson, R. J. Spiro. & W. E.Montague (Eds.), Schooling Jul the acquisitors of4mowledge. Hillsdale, NJ: Erlbaum.

Sanford. A. J., & Garrud, S. C. (1981). Usensonlimgroutes lamgmage. New York: Wiley.

Scardamalia, M., & Serener, C. (1982). Assimilativeprocesses in composition planning. EilmcairomalPsychologist, 17, 165.171.

Schank, R. C., & Abelson, R. (R77). scnpts, Plum,rush, asd annientamdisg. Hillsdale, NJ: Erlbaum.

Shaughnessy. M. (1977). Errors mod expeciatiost: A guilefor the teacher of basic whin's. New York: Oxford.

Shuy, R. W. (1981). Toward a developmental theory ofwriting. In C. H. Frederiksen & J. F. Dominic (Eds.),Irwin: The "awe, developments, ma seachimg ofMettles C011esemmiceirom (Vol. 2). Hillsdale, NJ:Erlbaum.

Skwire, D., Chitwood, F.. Ackley. R., & Fredman. R.(1975). Simi/ores book of college &s/ash, BeverlyHills. CA: Glencoe.

Stanovich, K. E. (1981). Attentionsl and automatic con-text effects in reading. In A. M. Lesgold & C. A.Perfetti (Eds.), heierative processes in restarts.Hillsdale, NJ: Erlbaum.

Stefik, M. (1981a). Planning with constraints (MOLGEN:Part 1). Artificial luelligemce, 16, 111.140.

Stefik . M. (1981b). Planning and metaplanning(MOLGEN: Part 2). Artificial Intelligence, 16.141.170.

Stein, N. L.. & Glenn. C. G. (1979). An analysis of storycomprehension in elementary school children. In R. 0.Freedle (Ed.), Advances is flacons, processes (Vol. 2).Hillsdale, NJ: Erlbaum.

Stein, N. L.. & Nezworski. M. T. (1978). The effect oforganization and instructional set on story memory.Dircomne 1. 177.193.

Thorndyke, P. (1977). Cognitive structures in comprehen-sion and memory of narrative discourse. CavitiesPsychology, 9, 77.110,

Voss, J F., Greene. T. R., Post, T. A.. & Penner, B. C.(1983). Problem solving skill in the social sciences. InG. H. Bower (Ed.), The psychology of lear"' Nedmotivation: Advances is research theory (Vol. 17).New York: Academic Press.

Waters. H. S. (1980). "Class News": A singlesubjectlongitudinal study of prose production and schema for-mat ion during childhood humid of Verbal Laarmargand Verbal Behaysor, 19. 152.167.

Witte. S. P.. & Faigley, L. (1981). Coherence, cohesion,and writing quality. College Composition asd Comessmicarooe. 32, 189.204.