Writing as Problem Solving Strategic Decisions

J ohn R. Hayes and Linda S. Flower O ur approach proceeds from five strategic decisions wh ich we made
about how to conduct our research . Briefly these decisions were:
l. ro focus on rhe act of writing;
2. ro rry for a process m odel of writing;
A top down approach employing protocol analysis can yield valuable data about . . 3. to model individual writers;
writing processes. The main featttres of composition appcwent 111 the data are: wntmg 4. to work wholistically or "top down" ; and
is goal directed, writing processes are hierarchically orgamzed, some ofthe processes 5. to divide the writing task into parts fo r easier analysis.
may intermpt others, rewrsion is possible, and writing go~!J may be modified as the As we will see below, these decisions are genuine ones in rhe
result ofwriting. The first four of these features are embod1ed 111 a process model sense rhar we could reasonably have made other choices. Alternative
ofcomposition. approaches to the study of writing do proceed from different decisions
on these same issues.

It's very exciting ro be doi ng research on composition just now because 1. Our firsr and most important decision was ro focus on the act of
a great deal is happening. Theorists are exploring new ideas and writing-that is , ro attend to whatever iris that writers do when they
experimenters are providi ng us with new empirical results at an produce a rexr. Thus, we viewed writing primarily as a process rather
unprecedented rate . Currently, there are at least four major theoretical than as a product. We fe lt that by far the richest source of information
viewpoints guiding the work of researchers who are trying ro u nderstand about wri ting would be to observe step by step how the writer had
composition: act ually created rhe essay. H owever, we did not intend ro ignore the
1. The psycholinguistic viewpoint, represented by Kinrsch ( 1974), product. Wherever possible, we looked ro the writer's essay for evidence
Rurnelharr ( 1975), and others; to confirm or elaborate the more direct observations of process.
2. The ling uistic viewpoint, represented by You ng (1970), Cooper and To observ~ writers in action we have employed process tracing
Odell (1977), de Beaugrande (1979), and others; methods borrowed from cog nitive psychology. In our studies a typical
3 . The developme ntal viewpoint, represented by Bereiter, Scardamalia, experiment p roceeds as fo llows: subjects appear at the experimental
and Bracewell (1979), Graves (1975), and others; and session knowing that they will be assigned a ropic on which to write an
4. The cognitive processing viewpoint , represented by Collins and essay and rhar rhe whole procedure will rake about an hour. Further,
Gentner ( 1979), Nold (in press), and H ayes and Flower (1980). they know that they will be asked to "think aloud" while writing. The
A casual observer could easily find this m ultiplicity of approaches subject is seared in a quiet office with a desk, pencil, and paper, and the
confusing . Even researchers active in the fie ld may som~tim:s find it rape recorder is t urned on. The experimenter then gives the subject an
difficult to characterize their own research. Their attentiOn IS llkely to envelope containing t he writing assig nment- that is, the topic and rhe
be directed at the subject matter they are trying to understand rather intended audience. The subject then busily sets to work writing and
than at the assumptions underlying t heir research methods. In this commenting roughly as follows: " W ell , open up the magic envelope.
paper we will stand back from the research we have been doing for the OK. Whew! This is a killer. Write about abortion pro and con for
last several years and attempt ro say what ir is we think we have Catholic Weekly. Ok, boy ! How am I goi ng ro handle this?", ere. This
been doing. . . continues for about an hour until t he subject says someth ing like,
There are two major facrors which have shaped our work on wrltlng. " W ell , that's it. G ood bye, tape recorder (click)." The data of the study
First, in conducting our research we have made a number of strategic consist of a verbatim t ranscript of the rape recording (with all the
decisions about what is interesting and about how best to proceed. To "urn's" and pauses and expletives undeleted) together with rhe essay and
pur it plainly, these decisions are the incarnat ion of our scientific biases. all of t he notes the writer has generated along the way. The transcript is
Second, our research has been shaped by a refracrory world wh ich has called a protocol. These materials are then examined in considerable detail
insisted that we attend to certain salient facts about t he writi ng fo r evidence which may re-nal something of t he processes by wh~he
processes. In what fo llows we w ill describe the strategies which we have writer has created the essay. In general , the data are very rich in such'
chosen ro g uide our research and then the facts which nature has evidence. Subjects typica lly g ive many hints about their plans and goals 1
imposed on us. e.g. , 'Til just jot down ideas as they come rome"; about strategies for
dealing with the audience, e.g. , 'Til write this as if I were one of
them "; about cri teria for editing and evaluation , e .g., "For 10-year-olds,
Visible La11guage, XIV 4, pp. 388-399. . . .
Hayes's address: Communications Design Center, Carnegie-Mellon Un1vcrs•t y, Pmsburgh, we better keep this simple"; and so on. The analysis of th is data is
PA 152 13 called protocol analysis.
0022-2224/80/l000-0388S02.00/0© 1980 Visible Language, Box 1972 CMA, Cleveland,
OH 44106.
389 Hayes & Flower I Writiltg as Problem Solving

..
 2. To understand the writing act, we certainly need to identify the describe individuals rather than averages of g roups . The disadvantage of
processes involved-but this is not enough. We also need to know how this approach is that it may be expensive. In the worst case, each individual
these processes are organized to produce a text. That is, we need to know may require a separate model. With better luck, m odels of individual
how the processes are sequenced, how one process is terminated and how the writers wi ll rurn out to be variants of a small number of model types. The
one which follows is chosen, how errors are detected, etc. Further, we want advantage of this approach is that it is more likely than a model of the
to know how simultaneous processes interact. When writers construct average to capture the behavior of actual (rather than idealized) writers.
sentences, we want ro know how they hand le such multiple constraints as Our third strateg ic decision, then, was to model the behavior of
the requirement for correct g rammar, appropriate rone , accuracy of individual writers rather than the average behavior of groups of writers.
meaning, and smooth transition. In short, we want a model which specifies
the processes involved in writing and accu rately describes their organization 4. In studying writing, we might well have started with processes which
and interaction . psychologists and psycholing uists have already identified as fundamental
A model is a m etaphor for a process: it 's a way to describe something, ones-processes such as short-term memory, grammatical categorizat ion,
such as the composing process, which refuses to sit still for a portrait. and lexical marking . We might then have attempted to synthesize more
People bui ld models in order to understand how a dynami c system works , complex processes using these fu ndamental processes as building blocks.
and to describe the functional relationsh ips among its parts. In addition, if a This synthetic or bottom 11p approach is a very familiar one in science and has
model is really to help us understand more, it should speak to some of the frequently been used with g reat success. Geometry and Newtonian physics
critical questions in the fie ld of writi ng and rhetoric. It should help us see are perhaps the best known examples.
things in a way we didn 't see them before. However, research often proceeds in the opposite direction; that is,
Our second strategic decision was to di rect our research toward the wholistically, or from rhe top down . Chemistry provides a good example
construction of such a model. Ideall y, the model should be capable of telling of top down research. Chemical research often starts with a complex
us how writers go about producing a text when they are g iven a writing compound and then looks for the elementary components and their
assig nment. It should tell us what p rocesses are involved , in what order relations. T he top down approach is the one we have chosen to apply in our
they occur, and at what points the writer will experience difficu lty. At writi ng re~arc h. We have started from the top with rhe complete wri ting
present, of course, we must be satisfied with a model which is much less act and have attempted to analyze it first into a few relatively complex
complete than the ideal. The ideal defines where we would like ro go, subprocesses. As the analysis proceeds , the complex subprocesses are
but-alas!-not where we are now. analyzed further into progressively simpler subprocesses. Ultimately, we
hope that this top down analysis will make contact with the fund amental
3. It is apparent that not all writers write in the same way. For example , processes which psychologists and psycholinguists have already identified.
some writers plan their essays from beginning to end before they write a Thus, the top down and bottom up approaches may be viewed as
si ngle word of text, whi le others never seem to look beyond the next complementary.
sentence. Further, some writers seem to wri te with their readers constantly The advantage of the bottom up approach is that it is rooted in
in mind , checking frequently to be sure that they have taken the reader's fundamental processes. The advantage of the top down approach is that its
knowledge and att itudes into account. Others appear serenely unaware that results are almost certain to be relevant to real writing siruarions .
an audience could fail to understand what they, in good fait h , have intended
to say. 5. Our fina l strategic decision was to divide the writing task into three
In modeling we can deal with such d ifferences in ei ther of two ways. We parts:
can choose to construct a model of the "average" writer and delay until some A. The writer's long-term memory;
more propitious time the description of d ifferences among writers. This B. The writing processes- that is, the writer excluding the writer's
approach has the merit of simplicity. Further, if things work out well , a long-term memory; and
model of an average writer might be useful in characterizing individual C. The task environment-that is, the world outside the writer's skin. The
differences. Thus , models for individual writers might prove to be minor relevant parts of the task environment are assumed to be: (1) The
variants of the average model. H owever, this approach may have rhe rhetorical situation~that is, the specifications of topic an :audience to
d isadvantage rhar averages sometimes suffer from-the average may be which a writer must respond ; and (2) The text which the writer nas
representative of no one. Thus, we sincerely hope that no one has the produced so far. This text becomes an increasingly important parr of the
average number of children- two and a half- nor would we want anyone to task environment as writing proceeds.
have to eat an average course at dinner, which might be a compromise We chose this division because it is an especially convenient one for
between appetizer and dessert such as oysters with chocolate sauce. psychological analysis and modeling. Transfers of information between the
An alternative approach is to construct models which are in tended to task environmen t and the writer are usually marked clearly by overt acts of

390 Visible Language XIV 4 1980 39 1 Hayes & Flower I Writing as Problem Solving
reading or writing. Further, information retrieval from long-term memory
is frequently detecrable by examining the verbal protocol. Thus, the
,....--- ~
boundaries we have chosen divide the writing cask into parts whose (!)
z (!)
z (!)
interactions are relatively easy to observe. § z .....--
0<(
Bitzer's analysis of the rhetorical situation ( 1968) focuses on the w
> w ~ i=
0
w
-
importance of the task environment. Lowes' class ic study of Coleridge 0
w w a: ....__
a: a: 1-
( 1927) focuses on the importance of the writer's long-term memory. Our 1- () <(
X ::::l u.. (/)
own research has focused on the writi ng processes. w 0
0 (/)
1- 0 (/) w
a: ()
a.. 0
a: (!)
Our Model 1- a.. z a:
Whi le we don't wane to present our model in great detail (that has been zw ~ i= 0
<( t:
done elsewhere; see Hayes & Flower, 1980), we do want to show enough
to illustrate how it had been shaped both by fact and by our strategic
::E
z
0
'II
(!)
......
(/)
z<(
- z
0
::E
a: z a:
decisions. >
z
E
a:
1- ~
~
Figure 1 shows the overall structure of the model. That we have a model w ~
of the writing act at all, of course, illustrates our first two strategic :.::: 1-
(/)
z '""(3""
decisions: to focus on the writing act and to model it.
<(
1-
w z (!)
::E
"' !Si ......
The effect of our fifth decision-to divide the writing task into task
environment, long-term memory, and writing process-is also evident
in Figure l.
z
(!)
Ci)
(/) Q)
Q)
~
()
C>
c
(!)
z
zz
z<(
(!)
<(
0
(!)
z
i=
1-
w
(/)
--
~
<( 0
c :; <(
Figure 2 shows the subprocesses of the writing process. Figure 3 shows
the inner structure of one of the subprocesses-the monitor. The
(!)
z
E
0
·a.
0
1-
.~
"0
~
<(
.:::
0
::E
......
a..
L
~N I1Vl::l3 N 3 ~
""'I
progression from Figure 1 to Figu re 3 parallels the progression of our top a:
~
down approach. At fi rst we analyze the writing act only into its largest,
most evident components. Then , as research proceeds and as data allow, we
analyze these major components successively into more refined subparts.
T he function of the monitor (see Figure 3) is to control the sequence of
writing processes. For example, it determines when idea generation wi ll
stop and organizing processes will beg in. The middle section of the
monitor- rules 3 through 6-is variable in form co allow for differences
::E
among wri ters in rhe way writing processes are sequenced. Figure 4 shows a:
w Q)
0
four alternative forms for rules 3 through 6. The first form represents a 1-I c
writer who polishes each sentence before considering rhe next. The fourth (!)
z
0
·a.
Q)
:0 "'c
form represents a writer who plans rhe entire essay before writing rhe first 0
......
0
1-
~
<(
"'
il:
C>
sentence. (/)
>- 0 0 c
0: a: Q) Q)
W e know, of course, chat there are many more differences among writers w ~
0 C> C>
1- "0 "0
::E
chan the few we have discussed. Our strategic decision to model individual a: w
::E
Q)
j j
Q)
"0
~
writers will lead us co search for ocher variations in rhe model which will ~ 0
c
0
c 0
allow us ro describe other individual differences. We expect, for example, w :.::: :.::: Ci)
I
1-
that some novice writers may entirely omit one or more of che major
ft;>
processes employed by competent writers. W e have frequently observed
writers who omit the review process- char is perhaps rhe commonest
~
problem found in papers handed in at the lase minute-and we have found
ar least one writer who showed no trace of an organizing process.

Figure 1. Structure of the writing model

392 Visible Lang11age X IV 4 1980

 ill I
1. (Generared language in STM ~ edir)
2. (New informacion in STM ~generate)
I- 3.-6. Goal sening producrions
~ X
w w 1- (T hese vary from wrirer ro wrirer: see Fig. 4.)
z
>
w
a:
0
<(
X
w
1-
- 7. ((goal = gene rare) ~ gene rare]
8. ((goal = organize) ~ organize]
w
a: 9. [(goal = rranslare) ~ translate]
10. ((goal = review) ~ review]

S3!::1n03:::l01::1d
31::1nliV:I
I Figure 3. Monitor.

C) _I
z <(
Configuration 1 (Depth first)
i= u...
<(
_I 3. [New element from rranslare ~ (goal = review)]
en en I- z 4. [New element from organize ~
en z<( X
w 0 a:
(goal = translare)]
w I- w 5. [New element from generare ~
a: X u... C!l > 6 en z i= 0 (goal = organize)}

~~~--w
()
I- w 0 z w 6. [Not enough material ~
0
a: z I- i= a:1- a:0a_ t5 en Ci5 !:::
z
(goal = generare)]
a: i=-?~a_ 0a_ <( -
a_ a: a_ X 0
C!l ti:i ~ ~ w
a:
a_ Ci5 0 ~ Corifiguration 2 (Get it down as you think of it, then review)
z C!l a:
i=
X
w
a_ 3. [New eTement from generate ~ (goal = organ ize)]
a: 4. [New element from organize ~ (goal = translate)]
~ 5. [Nor enough material ~ (goal = generate)]
II 6. [Enough material ~ (goal = review)]
1n:13sn C01zjiguration 3 (Perfect firJt draft)
C!l ~lON~ fl: ~ 3. [Not enough marerial ~ (goal = generare)]
z W W <( W C!l
E:i I- I- I-N z 4. [Enough material,
z 0
z , , :::l
<(
_I
:::l z z _I
<( i= plan nor complere ~ (goal = organize)]
a:o
<(
C!l 0 ........,
<t
_I
<t
Gi
u:..w
f--i..U--;J
en
::::~
~
w a:
ul o
<(
C!l h
01-
C!l w
en
....._
5. [New element from rranslare
6. [Plan complete
~ (goal
(goal
= review)]
= rranslate)]
I
w ~

C!l
..... ~ L- Configuration 4 (Breadth firJt)
zz ~'L-'--~======~----~------~ ~
z 3. [Not enough material ~ (goal = generate)}
3BOI::Id 3:::>Vld31::1_r~u---n-:~"""
1 3 sn
4. [Enough marerial,
<(
_I
a_
plan nor complere ~ (goal = organize)]
C!l
lON 5. [Plan complete ~ (goal = translate)]
z w 0 w
6. [Translation complere ~ (goal = review)]
i= Gi en m w 0
<( w 1-
z w
a: w 1-

..
I-
a: a: Gi w ~
0 <( > z
w
w ~ 0 a_ :::l
a:wI- Figure 4. Alternate configuration for the monitor.
~ ~ ~
z ~
w _I
~ ~
() w
w _I w <(
w > w _I
C!l
Ta: w m
z w a: w

Figure 2. The writing process

395 Hayes & Flower I \'(lriting as Problem Solving

ft 11
 In the same way, the writer who said that he was "simply being
The Facts descriptive" elaborated his goal as follows: "I think what I really want is to
Over several years, as we applied our research method to the analysis of p resent maybe one (point) with a lot of illustrations." He then went on to
writers in action, we were forced to a number of conclusions about the state the point and to develop a list of eight illustrations . In many cases,
writing process. Among these conclusions, the most important in shaping then,_ wnter_s tell us in their " thinking aloud " protocols that their goals
our model of the writing process are these: are h1erarch1cally structured. Even if the writers didn't tell us explicitly
1. Writing is goal d irected. though, there would still be plenty of evidence that writing processes are
2. Writing processes are hierarchically organized. hierarchically organized. For example, many writers start the writing
3. Some writing processes may interrupt other processes over which they seSS!on With a penod of planning in which they try to develop an outline to
have priority. write from. To do this _they may first t ry to generate ideas freely. When they
4. Writing processes may be organized recursively. feel they have enough 1deas, they try to organize them into an outline.
5. Writing goals may be modified as writing proceeds. Generating and organizing are part of p lanning, and planning in turn is
part of writing. Clearly these p rocesses are hierarchically organized.
1. Writing is goal directed. Evidence that writing is goal directed is easy to The model reflects the hierarchical organization of writing processes in
find in the protocols. Typically, writers comment on their major goals early two ways. First, the structure of processes in the model is intended to match
in the wri ting session. For example, one writer who was asked to write the hierarchical structure of processes observed in writers. Second,
about a woman's role fo r a hostile audience , said: " If an audience were operators within the ORGANIZE p rocess allow the model to construct a
hosti le the worst thing to do would be to defend yourself- so I would try to hierarchical arrangement of goals.
humor them- to make them-uh-more sympathetic maybe .. .. "A
second writer assigned this same topic said: ''I'm trying to decide 3. Priority intermpts. Editing appears to take precedence over all other
whether ... I want to convince my audience of someth ing specific writing processes in the sense that editing may interrupt the other processes
about- uh- fo r instance the Equal Rights Amendment or whether at any time. The generating process appears to be second in order of
something general about women should have the same rights as men ... and precedenc~since it interrupts any process except editing. Here are two
I also need to decide if I want to actively convince my audience or simply examples of edits (in italics) which interrupt the writer whi le he is
state my point of view .... " A few lines later she decides: "I'll try to generati ng new ideas: "The p roblem is to make the uses more general and
convince them of what it's like to-not to have certain rights .... "A third acceptable--that's the tvrong word- ! mean important seeming"; "Basically
writer said , ''I'm not really trying to persuade these people of anything, I'm the idea is that if one has a special marker in a bui lding that-
simply being descriptive . ... I'm saying this is the way the world is . .. . " which-means stop .... "
Goals enter into the model at several places. First, the goal-setting Interrupts by the ed iting p rocess often appear quite abruptly. The
part of the planning process (see Fig ure 2) uses information from the task writer no sooner has the wrong word out than the editing process leaps on it
environment about the rhetorical situation and information from the in the middle of a sentence and changes it. Interrupts by the generating
writer's long-term memory about the topic and the audience to establish process typically wait for the end of a sentence. However, they also appear
goals for the essay. Examples of such goals are seen in one writer's attempt rather abruptly. For example, while one writer was busily generat ing
to make her audience more sympathetic to a point of view and another sentences according ro his writing plan, he unexpectedly said,
writer's recognition that her essay for 10-year-olds must be made " . . . possibility of a p leasantry I suppose at that point ... . "
"very interesting. The mechanism for priority interrupts in the model is located in the
Second, the monitor (see Figures 3 and 4) sets goals for carrying out moniror (see Figure 3). The first two rules in the monitor control editing
writing processes. Such goals are reflected in statements such as, "Let's and idea generation. T he fact that these two rules come before the goal
organize this mess ," and, "OK, let's get it down on paper. " setting rules (3 -6) g ives these processes priori ty over all others.

2. Wt·iting is hierarchically organized. When writers have identified their 4. Remrsiveprocesses. The term "recursive" is used here in the mathematical
major goals, e.g., the particular aspect of the topic they want to discuss and sense. A recursive procesi;ais one which can contain itself as a part. Perhaps
their general approach to the aud ience, they frequently identify subgoals on the best way to make this term clear is to start by discussing re~ve
the route to these major goals. Indeed, the subgoals may in turn have their definitions and to proceed by analogy to recursive processes.
own subgoals. A writer who had as his main goal to write about the A term is said to be defined recursively when its definition contains the
"worries" of a particular group set up subgoals to write about the subropics term being defined as a part. At first g lance, this situation may seem to ·!
"the political issue" and "the philosophical issue." Under each of the involve an unacceptable circularity, but as we will see, everything works out
subtopics he specified a list of three or fo ur sub-subtopics. Thus, his all right. Let's take an example from linguistics. Consider the following
major goal was expanded into a hierarchy of subgoals. (incomplete) definition of a sentence:

397 Hayes & Flower I Writil1g as Problem Solving

396 Visible Language XIV 4 1980
 Sentence = noun phrase +predicate, or back to it by the translating and organizi ng processes. As yer we don't have
sentence+ "and" +sentence. sufficient data to specify how rhe mechanism works.
This is a perfectly workable definition even though the t~rm being defined
appears in rhe definition. Consider rhe problem of deCidmg whether or nor
rhe following srring of words is a sentence:
References
" Frank has warts and Berry has hiccoughs."
Since rhe string doesn't have rhe form "noun phrase + predicate," we rest to Bereiter, C., Scardamalia, M. , & Bracewell, R. J. An applied cognitive-
developmental approach co writing research . Paper presenred at the American
see if ir has rhe form "sentence+ 'and' +sentence." To do rhis, we have ro
Educational Research Association Meeting, San Francisco, April, 1979 .
show rhat rhe strings before rhe "and" and after the "and" are sentences.
Birzer, L. The rhetorical situation. Phi!oJopbyand Rhetoric, 1968, I, 1-14.
They turn our to be, since both have rhe form "noun phrase+ predicate." Coll ins, A., & Genrner, D. A framework fora cognitive theory of writing.
Just as the idea of a recursive definition contains no intraccable Unpublished manuscript . Cambridge, Mass.: Bolt Beranek & Newman, 1979.
circularity, neither does the idea that a recurs ive process can contain itself as Cooper, C. R., & Odell, L. Evaluating writing : describing, measuring , judging.
a parr contain any inrracrable circularity. To illusrrare a recursive process, Urbana, Ill.: N ational Counci l ofTeachers of Engl ish, 1977 .
we will consider the Wendy protocol. In her first draft, Wendy wrote de Beaugrande, R. T he processes of invenrion: association and recombination . College
sentence 1 of rhe final draft and rhen followed ir directly by sentence 7 of rhe CompoJitionandComnumication, 1979, 30(3), 1-12 .
fin al draft. When she was editing (a parr of the writing process), Wendy G raves, D. H . An examination of the writing processes of seven-year-old children.
decided rhar readers would have trouble with the transition between Ruearchinthe TeachingofEngliJh, 197 5, 9(3), 227 -241.
H ayes, J . R. , & Flower, L. S. ldenrifying the organization of writing processes. In
sentences 1 and 7. As a result she called on rhe whole writing process ro
L. W. Gregg and E. Steinberg (eds.) Cognitilll! proceJJeJ in writing. H illsdale, N .J. :
insert a small essay inside her larger essay. The whole writing process rhen
Lawrence Erlbaum , 1980.
was used as parr of editing, and rhus as a parr of itself. Kinrsch , W. The reprmntation ofmeaning in memory. Nrw York: Lawrence Erlbaum ,
In rhe model, the mechanism underlying recursion is hidden in the fine 1974.
structure of the editing process. We believe rhar when rhe editing process Lowes, J. L. The road to Xanadu. Boston : Hough ron Mifflin , 1927.
identifies a major fault in rhe rexr, e.g., lack of context or poor organization, Nold, E. Re~sing. In C. H. Frederiksen , M.S. Whiteman, andJ. F. Dominic
it may employ the whole writing process in the effort ro fix the fault. (eds.), \flriting: the natm·e, developme/11, and teaching ofwrillen communication.
Hillsdale, N.J .: Lawrence Erlbaum, in press.
Rumelharr, D . Notes on a schema for srories. In D. Bobrow and A. Collins (eds .),
5. Dynamic modification ofgoals . When writers choose goals, rhey are by no
Repmentationandtmdemanding. New York: Academic Press, 1975 .
means stuck with them. When a goal is difficult ro meet, the writer can
Young, R. E., Becker, A. L. , & Pike , K. E. Rhetoric: diJcoveryandchange. New York:
respond to the difficulty by modifying or abandoning the goal. Consider, Harcourt, Brace, and World, 1970.
for example, the writer who was assigned the task of writing about
"abortion: pro and con, for a hostile audience." Early in the session, she
decided ro incl ude as one of the " pro" tOpics the point that "a woman Funding for this paper was supported in parr by Grant No . NIE-G-78-0 195 from the
should have the right to limit the size of her family." About half-way National Inst itute ofEducation, Washington, D. C.
through the session, she came back ro rhis point and re-evaluated it as
follows: "A woman should have the rig ht ro limit the size of her
fami ly-urn-and I think I'm not going ro mention that- because that is
nor a view that is shared by the audience."
The writer mentioned earlier who was consideri ng "a pleasantry,"
explored the issue in some detail, decided ir wouldn't work well , and
abandoned it. This same writer had establ ished goals ar various times in
the session to write about rhe issues "individ ual freedom," "the impacc of
technology," and "attitudes toward scientific analysis ." After considerable
work, he discovered relations among the three and tied them together under
a single heading as "the philosophical issue." Thus, three independent
objectives were modified to become subgoals of a single higher level goal.
As yet, the model has no mechanism to account for rhe modificat ion
of goals. We assume that the mechanism must reside in the goal setting
processes and that it m ust use information about processing fai lures fed

398 Visible Lang11age XIV 4 1980 399 Hayes & Flower I Writing as Problem Solving