WRITING PROCESS THEORY: A FUNCTIONAL DYNAMIC APPROACH HUUB VAN DEN BERGH* & GERT RIJLAARSDAM */** *Utrecht University & **University of Amsterdam 0. PREAMBLE In this chapter, we present key elements of a theory of writing processes, based on our observations in the empirical studies we conducted over the last ten years. The theoretical framework rests upon two paradigms in cognitive psychology. One paradigm is the writing process model introduced by Hayes and Flower (1980; see also Hayes’s 1996 revisions), which is assumed in this chapter as known. The second paradigm is parallel distributed processing (Rummelhart, McClelland & the PDP Research group, 1999). Connecting both paradigms, we propose a functional dynamic system as the basic structure of writing processes. The empirical data we present have three common features: 1) We adapted a weak novice/good novice paradigm. Much writing process research has been carried out using an expert-novice paradigm. However, experts can be defined in different ways (Torrance, 1996). Experts may excel in some fields of writing, because of their subject and/or genre knowledge. How experts became experts and on which dimensions they differ from novices is not considered in most analyses. Hence, differences in 1 Corresponding author: Gert Rijlaarsdam, Graduate School of Teaching and Learning, University of Amsterdam, Wibautstraat 2-4, 1091 GM Amsterdam. [email protected]. 2 VAN DEN BERGH & RIJLAARSDAM writing processes between these extreme groups may have many causes. In order to circumvent this fallacy, just one group should be considered, whether (relatively) novices or experts (in one subject area). In our studies, we investigate the writing process of writers of about 15 years old, and study the natural variance within this group. 2) Students in our studies wrote two argumentative, documented essays, within a peer audi- ence oriented contextual frame, while having access to documentation on the topic (clippings from newspapers and journals, tables and figures). Writing time varied from 60 to 103 minutes. 3) Students wrote under think aloud conditions; protocols were fragmented into cognitive activities, and a jury evaluated the quality of the resulting text written by the student. During our excursion to reach the final destination (section 7), we will visit six observation posts. These observation posts serve as landmarks for a theory in development, and can be considered as calibration points for a writing process theory. In the figure captions we refer to the original studies. 1. OBSERVATION 1: WHAT CONSTITUTES A WRITING PROCESS? To determine the constituting elements of writing processes, one may observe processes, and identify and categorize mental activities. One problem is how to define and select the mental activities? What should one distinguish, what should one leave out as irrelevant? Our solution was partly a functional approach: all process elements that together predict the resulting text quality are taken into account, and partly a theoretical one, as we started with a fine grained scheme based on Hayes. This approach resulted in two findings: WRITING PROCESS THEORY: A DYNAMIC & FUNCTIONAL APPROACH 3 1) The frequencies of eleven (broad) categories of cognitive activities explain 76% of the (true score) variance in product quality (Breetvelt, Van den Bergh, & Rijlaarsdam, 1994, p. 116); categories were derived from the Hayes and Flower model (generating, translating, revising etc.). 2) This result holds only when the factor time or ‘moment in the process’ is included in the prediction. We distinguished three equal parts in the process. It turned out that none of the activities were effective during the whole process; some contributed during one or two phases in the same direction (positively or negatively), some in reverse directions: positively in one phase, negatively in another (see Appendix). 3) We concluded that differences become apparent only when the moment activities are em- ployed is taken into account. If the moment an activity is employed is left out of the analysis, hardly any relation could be found between cognitive activities on the one hand and text quality on the other hand. Therefore, the moment a cognitive activity takes place is of crucial importance. 2. OBSERVATION 2: DYNAMICALLY CHANGING PATTERNS As the text grows, the task situation changes. Writers adapt to these changes by carrying out different processes or cognitive activities. This is shown in Figure 1. In this figure the (mean) occurrence of two activities, reading the assignment and generating information, is plotted against the moment in the writing process. As time goes on, or more precisely as the text grows, the probability of occurrence for ‘reading the assignment’ decreases. This activity is most likely to be observed at the beginning of the writing process, becoming increasingly unlikely towards the end. As is to be expected, at no time does the average writer engage in this activity as in the beginning of the writing process. 4 VAN DEN BERGH & RIJLAARSDAM -- INSERT FIGURE 1 ABOUT HERE -- Generating, however, shows a completely different course over the writing process. The mean probability of occurrence gradually increases from a low initial value to reach a peak (minute 20) from which is decreases to a very low level. Hence, in the beginning of the writing process, writers refrain from generating information and engage in other activities like ‘reading the assignment’. These general patterns, observed with fifteen-year-olds, also appear with younger children. Van der Hoeven (1997) showed the same patterns for both activities in eleven-year-olds. Most cognitive activities studied thus far show a distinct pattern of occurrence during the writing process. The essential discovery is that each cognitive activity has a higher or lower probability of occurrence depending on the moment in the writing process: at different points in the writing, different cognitive activities dominate. 3. OBSERVATION 3: DYNAMICALLY CHANGING RELATIONS BETWEEN PROCESSES AND TEXT QUALITY In Figure 2, we present the correlations between the frequencies of two cognitive activities (‘reading the assignment’ and ‘generating’) and text quality at various moments during the writing process. ‘Reading the assignment’ is positively related to text quality only during the initial stages of the writing process. During later stages, the correlation between this activity and text quality decreases and soon becomes negative. Hence, students who (relatively) often consulted the assignment in the beginning, other things being equal, produced text of high quality. WRITING PROCESS THEORY: A DYNAMIC & FUNCTIONAL APPROACH 5 Conversely, writers who hardly consulted the assignment in the beginning wrote poor(er) texts. Writers who consulted the assignment frequently at the end of the writing process wrote poor texts, whereas writers who refrained from reading the assignment at the end of their writing process produced better texts. The correlation between ‘generating’ activities and text quality also changes during the writing process. It increases during the initial phases and reaches a maximum in the middle of the writing process to decrease during later stages. -- INSERT FIGURE 2 ABOUT HERE -- 4. OBSERVATION 4: FUNCTIONAL ALLIANCES BETWEEN COGNITIVE ACTIVITIES Form and function of cognitive activities need to be distinguished. The recursive nature of writing does not imply that an activity that occurs at occasion t has the same function in the writing process as the same activity at occasion tx. It means, for instance, that revisions should be distinguished depending in the moment they are carried out. For instance, a writer with start-up troubles, beginning over and over again, might revise just as frequently as a writer who fluently produces a text and revises afterwards. These writers clearly write in different ways. We claim that the two types of revision in this example function differently. To explore functional relations between cognitive activities, we connected cognitive activities to preceding activities. For this reason, we went looking for functional relations between several other cognitive activities and generating (Van den Bergh & Rijlaarsdam, 1999). Here, we distinguish a several types of ‘generating’ (Table 1). We based our typology of gen- 6 VAN DEN BERGH & RIJLAARSDAM erating ideas on the cognitive activity that preceded the generating activity itself, if the preceding activity triggered the generation activity. -- INSERT TABLE 1 ABOUT HERE -Figure 3 shows that five combinations of generating activities described in table 1 have different patterns of distribution. This implies that the different combinations behave differently, which indicates a functional relationship. If the combinations were just random adjacent pairs, the distributions over the process would have overlapped. -- INSERT FIGURE 3 ABOUT HERE -- Therefore, making a distinction between the contexts that precede generating activities illuminates different processes in two respects. The mean the mean number of generating activities linked to each type of generating activity is different, and the distribution of these combinations varies during the process. This observation teaches us that in general we may predict that some combinations have a higher probability to occur at a certain moment in the process than others. This points to different functions of these pairs during the process. Note that we showed in Figure 3 the general pattern: individual differences are large (Van den Bergh & Rijlaarsdam, 1999, p. 109112). Individual differences are correlated with text quality: some functional pairs are more effective in the beginning of the process, others in a later phase of the process (Figure 4). For instance, from Figure 4 it appears that the correlation between ‘Translation-drivengeneration’ (TRDG) and text quality is time dependent. The correlation is significant in the first quarter (negative) and after (about) 25 minutes until (about) 75 minutes, where it is WRITING PROCESS THEORY: A DYNAMIC & FUNCTIONAL APPROACH 7 positive. After (about) 85 minutes, it reaches significance again, but now it is negative: the more this activity occurs during the end of the process time, the weaker the text. Please note, that these changes in correlation between the occurrence of cognitive activities and text quality do not only hold for ‘Translation-driven-generation’, but for all other combinations as well; in all cases the correlation with text-quality is dependent on the moment in the writing process. -- INSERT FIGURE 4 ABOUT HERE -- This observation leads one to reconsider the unit of analysis for theory building. When combinations of cognitive activities behave as functional relations, implying that the function of each activity varies according to the context – the preceding activity – then combinations rather than single activities should be considered as the unit of analysis. 5. OBSERVATION 5: FUNCTIONAL COMPENSATORY RELATIONS BETWEEN COGNITIVE ACTIVITIES Observation 4 leads to the assumption that cognitive activities can fulfil different functions, depending on the context in the process. The function of a cognitive activity may vary, depending of the context. The activity of ‘Re-reading already written text’ is a good example to demonstrate this point. Re-reading already written text functions within the reviewing component in the Hayes & Flower Writing Process Model. In Hayes’s revised model (Hayes, 1996), reading has a more central role in the writing process. Hayes distinguished several functions of reading during writing (Hayes, 1996, p. 18-20): ‘In addition to reading to evaluate, two other kinds of reading play an important role in writing: reading source texts and 8 VAN DEN BERGH & RIJLAARSDAM reading to define tasks’ (Hayes, 1996, p. 18). However, observation 4 pointed to the possible relation between ‘re-reading already written text’ and ‘generating ideas’: a rather rare combination over the writing process (see Figure 3) but an effective combination when the writing process proceeds (see Figure 4). At least some writers seem to use re-reading parts of their written text as input for new thoughts. This implies that re-reading already written text and generating must correlate. In this section, we will explore this relation by re-analysing the data reported in a former study (Breetvelt, Van den Bergh, & Rijlaarsdam, 1996). ‘Generating’ has a rather low probability overall. It reaches a maximum of about .12 around twenty minutes after the start of the writing process, and decreases slowly afterwards. That is, at around twenty minutes of writing 12 out of 100 cognitive activities are ‘generating’ activities (see Figure 1). The pattern for ‘Re-reading’ is clearly different. The probability increases during the first half hour, stays more or less constant for the next hour, and then rises to a probability of about .30. When we relate the distribution of these processes to the quality of the written text, two clear pictures emerge (Breetvelt, Van den Bergh & Rijlaarsdam, 1996, p. 17). Writers who generated relatively often at the beginning wrote weak texts; writers who gradually increased the number of generating activities wrote the best texts1. For re-reading the already-written text, the correlation changes during the course of writing, but is always positive. Note that these are the findings as long as we assume that these two processes are unrelated. However, it is tempting to hypothesize that re-reading already-written text can, in some instances, fulfil a supportive function for generating texts. Take, for example, the sequence that the writer writes something down, then re-reads what she wrote, and then generates another chunk of information. And indeed, if we account for re-reading, the correlation between generating and text quality changes (Figure 5). WRITING PROCESS THEORY: A DYNAMIC & FUNCTIONAL APPROACH 9 -- INSERT FIGURE 5 ABOUT HERE -- The correlation between generating and text quality changes dramatically if re-reading is taken into account. It thus seems that the occurrence of re-reading is related to the occurrence of generating. On the other hand (not in the figure), the correlation between re-reading and text quality changes only marginally if generating is accounted for. This implies that generating does not have the same effect on re-reading that re-reading has on generating. Therefore, it is plausible that re-reading serves generating in some writers, and that re-reading has other functions as well as serving as a tool for generating. In any case, the analysis shows that it is reasonable to map functional relations between cognitive activities. This observation is a rather crucial indication that a univariate view on cognitive activities that play a role in a writing process limits the interpretation of the data and the building of a writing process theory. Such view neglects the context in which cognitive activities do their work, or neglects the interdependency between activities, which can change during the process. From Figure 5, we learn that the effect of generating is larger when we correct for rereading. However, the correction has a different effect in both halves of the process. We might interpret this observation as follows. Much generating in the beginning of the process is ineffective; those writers who generate much in the beginning while leaning on re-reading already written text as input for generating appear to have written a poor text. In the second half of the process, generating contributes strongly to text quality. But those writers who use re-reading already written text in the second half as input for generating profit more from this effect compared to writers who do not use re-reading as input for writing. So the combination 1 Note that if time were left out in our analyses, the net result would have been a correlation of about zero. 10 VAN DEN BERGH & RIJLAARSDAM is inhibiting in the first half, but facilitating in the second half: functional relations appear to change over the process. The observation points to individual differences in the way activities are functionally combined: some writers make (relatively) more combinations of rereading already written text and generating of new information, other writers don’t. It is difficult to interpret these different strategies. One possible interpretation is that writers with weak generation skills need the input of the already written text as a springboard for generating a new idea to write about. This interpretation could be tested if data about the writers’ generation skills were available, which is not the case. However, from another study we got some indication that differences in processes are related to the quality of the skills involved in these processes. Van der Hoeven (1997), in her study on writing processes of 11-year-olds measured revision skills with special, independent tasks. She observed that revision skill was positively related to most of the cognitive activities appearing in the writing-aloud protocols: structuring, writing (production of written text), re-reading, evaluating and transforming already-written text. The higher the student’s competence in evaluating already-written text, the more instances of re-reading, evaluating and transforming were observed, and the better the resulting text. Interestingly, the competence of evaluating already-written text was negatively related with the quality of text. Only by employing writing process activities of revision was this negative relation changed into a positive relation. This implies that the skill itself is not sufficient; writers have to apply the skill when they write. Another important conclusion was that revision skill was not only related to the number of cognitive activities, but more importantly, to their temporal distribution over the writing process as a whole. Participants with low revision skills generated fewer ideas in the begin- WRITING PROCESS THEORY: A DYNAMIC & FUNCTIONAL APPROACH 11 ning of the writing process compared to participants with relatively high revision skills scores. While the number of ideas gradually decreased in the high revision skills group, this number increased in the low revision skills group. Students with high revision skills re-read, evaluated and revised relatively little in the beginning and more towards the end of the writing process. These findings suggest that the quality of revision – revision skill – is related to the way the writer organizes the writing process, and thus affects the quality of the resulting text. 6. OBSERVATION 6: INDIVIDUAL DIFFERENCES Until now, we showed general patterns of some cognitive activities, and the relation of these patterns to the resulting text quality. These patterns emerged from data from individual writers. We now turn to these individuals, to see in what respect they differ, and how we could qualify these differences. Hence, Figure 5 in fact portrayed the temporal distribution of two cognitive activities of nonexistent writers. In order to understand the writing process we need to turn to individual writers and study the cognitive activities they carry out. Statistics allow for the estimation of individual patterns of occurrence of a cognitive activity (Van den Bergh & Rijlaarsdam, 1996). In Figure 6, the differences in reading the assignment and generating between writers (N = 36) are presented. For each writer, the (estimated probability of) occurrence is plotted against the moment in the writing process. -- INSERT FIGURE 6 ABOUT HERE – Each line in Figure 6 represents one writer for either ‘reading the assignment’ (left) or generating (right). The differences between writers appear to be relatively large. Some writers con- 12 VAN DEN BERGH & RIJLAARSDAM sult the assignment (relatively) frequently during the beginning and considerably less during later phases in the writing process. For some writers, an increase in the probability of occurrence of ‘reading the assignment’ shows at the end of the writing. For a third group of writers the probability of observing this cognitive activity remains constant during the writing process. The temporal distribution for generating shows that for most students there is an increase in generating activities during the beginning of the writing process followed by a decrease later on. In effect, most writers appear to follow the mean pattern. Nevertheless, there are clear exceptions, like a student who generates much in the beginning and in the end but less in the middle stages, and students for whom the probability of generating continuously decreases. The relation between both activities also differs between writers. For some writers there is a positive relation between ‘reading the assignment’ and ‘generating’, whereas for other writers the relation between these two activities is negative. In fact, the correlation between the temporal order of both activities varies between writers from -0.90 and .90. For some writers we observed a negative correlation between these two activities, while for other writers there is a strong positive correlation (see also Figure 3). Hence, some writers use the information in the assignment to generate new information to write about, whereas others do not need the information in the assignment to generate content information. Perhaps this difference in functional relation between these activities is mediated by topic knowledge; if one knows enough on a topic, one does not need to consult the assignment to come up with ideas. However, if one does not know what to write, it seems a plausible strategy to see if the assignment contains unused information. WRITING PROCESS THEORY: A DYNAMIC & FUNCTIONAL APPROACH 13 To illustrate that the relation between cognitive activities can vary from writer to writer, we selected data from two writers. Figure 7 shows the (probability of) occurrence for ‘reading the assignment’ and ‘generating’. For Figure 7, two writers are singled out. -- INSERT FIGURE 7 ABOUT HERE – Remind that we observed and discussed two essential differences between writers in temporal organization of cognitive activities thus far. First, these differences in temporal organization are related to differences in text quality (e.g. see Figure 2). Second, the differences in temporal organization of one activity are related to the temporal organization of other activities. Figure 7 plots the change in (probabilities of) occurrence for re-reading and generating for two writers. It becomes clear from this figure that there is a huge difference in the temporal organization of both writers. This figure shows that the relation between the two activities differs between both writers. For the first writer, both activities are negatively correlated during the writing process (‘reading the assignment’ decreases as ‘generating’ increases), whereas for the second writer there is a positive relation between ‘reading the assignment’ and ‘generating’ (at least in the beginning of the writing process). These differences might be related to general procedural knowledge (compare, Alamargot & Chanquoy, 2001; Van der Hoeven, 1997), but might also be a consequence of the specific task execution, as both determine the ongoing writing process. 7. ISSUES RELATED TO AN EMPIRICALLY BASED WRITING PROCESS MODEL In this chapter, we illustrated that with a limited set of cognitive activities and the factor time, much of the variance in text quality can be predicted. These activities are the building blocks 14 VAN DEN BERGH & RIJLAARSDAM for a writing process model. This model should predict how these building blocks are organized during a writing process: how individual activities have distinct temporal distributions, and the change of functional relations over time. This model should also predict how differences in cognitive and linguistic skills cause different processes. In a number of studies we have tried to show empirical evidence for parts of the model presented above (e.g. Braaksma, Rijlaarsdam, Van den Bergh & Hout-Wolters, 2004; Breetvelt et al , 1994, 1996; Rijlaarsdam & Van den Bergh, 1996, 1997; Van den Bergh, Rijlaarsdam & Breetvelt, 1992, 1993; Van den Bergh & Rijlaarsdam, 2001) or showed the statistical model behind the analysis (e.g. Van den Bergh & Rijlaarsdam, 1996; or more in general Goldstein, 1995, especially chapter 7). Nevertheless there remain several unresolved issues that touch upon the heart of the model. One important issue to tackle is the controlling system: how does a cognitive system ‘know’ which activity should follow a particular activity, and how does a system ‘know’ that some combinations of activities must be frequent in a certain phase of the process? How do we specify the guide and control mechanisms, and the mechanisms that determine the employment of specific cognitive activities (Bruce, Collins, Rubin & Genter, 1979; Fayol, 1994)? Traditionally this task is reserved for the monitor. In the early Hayes and Flower model, a monitor was introduced, as a check and balance system, and fed with blueprints of processes (‘monitor configurations’ or writer style; Hayes & Flower, 1980, p. 20). In his revised model, Hayes implemented ‘task schemas’ that guide and control the process (Hayes, 1996, p. 4 and p. 17). Also, the co-ordination of cognitive activities is an area of debate (Alamargot and Chanquoy, 2001). During the writing process, writers must continuously decide which activity to employ next. Will a writer generate new information, structure the information already available, or reread the text already written? Sometimes it is assumed that a next activity starts when the products previously processed are in accordance with the activi- WRITING PROCESS THEORY: A DYNAMIC & FUNCTIONAL APPROACH 15 ties ensured by the component situated just after (compare, Hayes, 1996). Hence, this controlling mechanism would hardly be costly in terms of cognitive resources. However, in thinking-aloud protocols writers appear to instruct themselves from time to time (Breetvelt et al, 1994). Apparently, this monitoring activity, or the choice of activities to be employed, is at least partly under conscious control, and therefore requires attention. This allows for the intrusion of general knowledge (topic knowledge or procedural knowledge) in the writing process. It is assumed that the sequence of activities is based partly on routine, and therefore automatic and does not require conscious attention, whereas it allows for an intrusion of other activities based on some evaluation of the product thus far. How do these routines work? An interesting perspective on building a theory about processing routines is the parallel distributed processing model proposed and tested by Rummelhart and McClelland (1999). They propose a probabilistic system. Through experience, during processing, connections between units are established. This probabilistic system of connection strengths between units develops over time. This system is not a stored task schema that must be retrieved: the patterns of activation or connection strengths are part of the cognitive units themselves. It seems likely that if the text grows (changes) the activated cognitive nodes change as well. By these changes, processes like automatic spreading activation will activate other nodes (compare: Hinton & Anderson, 1981; Anderson, 1983; Rummelhart & McClelland, 1999; Schilperoord, 1996). Hence, the internal representation of the text changes with changes in the task situation. Writers react to these changes in terms of cognitive activities they employ. A second issue is the unit of observation. In our illustrations, we started with cognitive activities as units, and demonstrated that units can form functional pairs, where it is not the single cognitive activity that contributes to the quality of the resulting text, but a certain combina- 16 VAN DEN BERGH & RIJLAARSDAM tion, at a certain moment in the process. We need to identify larger functional units, consisting of at least two related cognitive activities. Where does a functional unit start, and where does it end? And how does it relate (co-ordinated, superordinated-cause, sub-ordinatedconsequence) to the former and the following functional unit? In Table 1 we presented adjacent pairs, with some context. In some cases, it is apparent that the adjacent pair is part of a larger unit. The Translation-Driven-Generation example for instance, seems to be driven by another fragment, where the writers re-reads already written text. With cognitive activities as the building blocks of the writing process theory, we must now try to identify the generative syntax that accounts for the hierarchically related patterns of building blocks. Besides, the unit of analysis may be different for different writers. It is generally known that the size of units depends on the skill of an executor. Therefore, highly skilled writers may process larger units then lower skilled writers. Alternatively, skilled writers might show patterns of activities in their writing process, e.g. they have a pattern of generating, structuring and formulating (compare Van den Bergh, Klein Gunnewiek & Herrlitz, 2000; Dansac & Alamargot, 1999). Hence, relatively skilled writers might not operate solely on the level of individual activities, but for these writers specific combinations of activities are to be considered as one unit (compare, Kunst, 1978). A third issue is that of the presumed causal relation between process and quality of the product. Above, a causal relation between distributional elements of an activity and quality of the resulting text has been suggested (at least). This conclusion seems obvious, and it is a crucial element in the theory thus far. It functions as an explanation of differences in processing, and pinpoints effective and less effective temporal patterns of a cognitive activity. However, from a methodological point of view one could dispute conclusion. In fact, a coincidence in time is WRITING PROCESS THEORY: A DYNAMIC & FUNCTIONAL APPROACH 17 observed. And although the temporal patterns are theoretically well founded, and the relations with text quality appear rather easy to interpret (it seems obvious that an activity like ‘reading the assignment’ is especially correlated positively in the beginning of the writing process; see figure 2), without proper experimental manipulation causality cannot be proven. A fourth issue concerns the linearization of the writing process. In thinking-aloud conditions, writers can only formulate one activity at a time. Therefore, according to the observations, writing is a linear process, in which activities are ordered in time. In thinking on writing, and in many writing theories, this linearization is posed without questioning. Above, we did the same, however, there is hardly any reason why the writing process should be linear. In principle, not only aspects of a monitor are allowed to function next to other activities. Think, for instance, on the manifold short editing activities that appear to intrude into almost every string of activities. But also, activities seem to be strongly related. For instance, structuring can trigger the generation of new ideas; that is, during structuring a new idea may pop up. In fact, this is a kind of parallel processing that we cannot do without, but which has received very little attention in writing research, That is, the higher-level units that appear to operate are, in fact, a kind of parallel processing, in which several sub-processes are fused in to one. We did not present a complete theory on writing and writing processes. In fact, we do not have such a theory. Nevertheless, we have tried to sketch our way of thinking, and made the skeleton of such a theory visible. It’s clear, however, that we do not have definite answers, and in some respects, we have just started to ask the right questions. 18 VAN DEN BERGH & RIJLAARSDAM LITERATURE Alamargot, D. & Chanquoy, L. (2001). Through the models of writing. Dordrecht: Kluwer Academic Publisher. Anderson, J.R. (1983). A spreading activation theory of memory. Journal of Verbal Learning and Verbal Behavior, 22, 261-295. Braaksma, M. Rijlaarsdam, G., Bergh, H. van den & Hout Wolters, B.A.M. (2004). Observational learning and its effects on the orchestration of the writing process.. Cognition & Instruction, 22, 1-36. 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. Breetvelt, I., Van den Bergh, H., & Rijlaarsdam, G.(1996). Rereading and generating and their relation to text quality: An application of multilevel analysis on writing process data. In G. Rijlaarsdam, H. Van den Bergh, & M. Couzijn (Eds.), Studies in writing. Volume 1. Theories, models and methodology in writing research (pp. 10-21). Amsterdam: Amsterdam University Press. 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, NJ, L.E.A. Dansac, C. & Alamargot, D. (1999). Accessing referential information during text composition: When and why? In: M. Torrance & D. Galbraith (Eds.), Knowing what to write: Conceptual processes in text production (79-98). Amsterdam: Amsterdam University Press. Fayol, M. (1994). From declarative and procedural knowledge to the management of declarative and procedural knowledge. European Journal of Psychology of Education, 6, 99-117. WRITING PROCESS THEORY: A DYNAMIC & FUNCTIONAL APPROACH 19 Goldstein, H. (1995). Mulitlevel Statistical Models (second edition). London: Arnold. 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 Erbaum Associates. Hayes, J. R., & Flower, L. (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. Hinton, G.E.. & Anderson, J.A. (Eds.). (1981). Parallel models of associative memory. Hillsdale, NJ.: Erlbaum. Kunst, H. (1978). Cognitie van semantische systemen: Strategieën en interpretaties bij een structure of intellect vaardigheid. [Cognition of semantic systems: Strategies and interpretation of a structure of intellect skill]. Dissertation. Amsterdam: University of Amsterdam. Rijlaarsdam, G. & Van den Bergh, H. (1996). Essentials for writing process studies: many questions and some answers. In, M.C. Levy & S. Ransdell (Eds.), The science of writing. New York: Lawrence Erlbaum ass. (p. 107-126). Rijlaarsdam, G. & Van den Bergh, H. (1997). Cognitieve activiteiten tijdens het schrijfproces en hun relaties met tekstkwaliteit: Plannen en formuleren. [Cognitive activities during the writing process and their relations with text quality: Planning and formulating]. In, H. van den Bergh, D. Janssen, N. Bertens,& M. Damen (Eds.), Taalgebruik ontrafeld. Dordrecht: Foris. (p. 275-284). Rumelhart, D.E. & McClelland, J.L., and the PDP Research Group (1999). Parallel distributed processing. Explorations in the microstructure of cognition. Vol I. Cambridge, MA., MIT-Press. 20 VAN DEN BERGH & RIJLAARSDAM Torrance, M. (1996). Is writing expertise like other kinds of expertise?. In G. Rijlaarsdam, H. Van den Bergh & M. Couzijn (Eds.), Studies in writing. Volume 1. Theories, Models and Methodology in Writing Research, (pp. 3-9). Amsterdam: Amsterdam University Press. Van den Bergh, H., & Rijlaarsdam, G. (1996). The dynamics of composing. Modeling Writing Process Data. C. Michael Levy and Sarah Ransdell. The Science of Writing. New York (N.J.): Lawrence Erlbaum Ass. p. 207-232. Van den Bergh, H., & Rijlaarsdam, G. (1999). The dynamics of idea generation during writing: An online study. In M. Torrance & D. Galbraith (Eds.), Studies in writing. Volume 4. Knowing what to write: Cognitive perspectives on conceptual processes in text production (pp. 99-120). Amsterdam: Amsterdam University Press. Van den Bergh, H. & Rijlaarsdam, G. (2001). Changes in cognitive activities during the writing process, and relations with text quality. Educational Psychology, 21, 373-385. Van den Bergh, H., Rijlaarsdam, G. & Breetvelt, I. (1992). Het stellen van doelen; relaties met productkwaliteit [Formulating aims; Relations with text quality]. Tijdschrift voor Taalbeheersing, 14, 221-233. Van den Bergh, H., Rijlaarsdam, G. & Breetvelt, I. (1993). Revision processes and text quality: An empirical study. In, G. Eigler & Th. Jechle (Eds.), Writing: Current trends in European research. Freiburg: Hochschul Verlag. p. 133-147. Van den Bergh, H., Herrlitz, W. & Klein Gunnewiek, L. (1999). Entwicklung von Fähigkeiten in einer Fremdsprache - vorsichtige Hypothesen. In G. Kischel & E. Gotsch (Eds.), Wege zur Mehrsprachigkeit im Fernstudium (p.115-138). Hagen: Fernuniversität. Van der Hoeven, J. (1997). Children’s composing. A study into the relationships between writing processes, text quality, and cognitive and linguistic skills. Utrecht Studies in Language and Communication, vol. 12. Amsterdam: Rodopi. WRITING PROCESS THEORY: A DYNAMIC & FUNCTIONAL APPROACH Figure 1. 21 Mean changes in occurrence of ‘reading the assignment’ and ‘generating’ during the writing process. Data from Breetvelt, Van den Bergh & Rijlaarsdam (1996). Reading assign. Probability of occurrence 0.3 0.2 Generating 0.1 Generating Reading assign. 0.0 0 20 40 Reading assign. Generating 60 Time (in minutes) 80 100 22 VAN DEN BERGH & RIJLAARSDAM Figure 2. Correlations between the temporal distribution of two cognitive activities and text quality: ‘Reading the assignment’ and text quality (R_Assignment) and between ‘Generating’ and text quality. Data from Breetvelt, Van den Bergh & Rijlaarsdam (1996). R_Assignment 0.7 Correlation Generating 0.3 -0.1 Generating R_Assignment -0.5 0 20 40 60 Time (in minutes) 80 100 WRITING PROCESS THEORY: A DYNAMIC & FUNCTIONAL APPROACH Figure 3 23 Changes in the mean probability of occurrence (y-axis) of Assignment-Driven- Generation (ASDG), Rereading-Text-Driven-Generation (REDG), Translation-DrivenGeneration (TRDG), Generation driven Generation (GEDG), and Pause-Related-Generation (PARG). Data from Van den Bergh & Rijlaarsdam (1999). PARG ASDG Probabillity of occurrence Probabillity of occurrence 0.15 0.10 0.05 0.00 TRDG TRDG REDG GEDG ASDG REDG TRDG GEDG REDG ASDG GEDG 0 20 TRDG 40 60 Time (in minutes) PARG 0.15 0.10 PARG 0.05 PARG PARG REDG GEDG ASDG ASDG 80 100 0.00 0 20 40 60 Time (in minutes) 80 100 24 VAN DEN BERGH & RIJLAARSDAM Figure 4. Correlation (y-axis) between occurrence of Assignment-Driven-Generation (ASDG), Rereading-Text-Driven-Generation (REDG), Translation-DrivenGeneration (TRDG), Pause-Related-Generation (PARG), Generation-Driven- Generation (GEDG) and text quality during writing (x-axis). Data from Van den Bergh & Rijlaarsdam (1999). 0.5 0.3 GEDG TRDG GEDG ASDG PARG TRDG 0.1 REDG REDG -0.1 PARG ASDG ASDG Correlation GEDG ASDG PARG GEDG REDG TRDG GEDG PARG REDG TRDG -0.3 -0.5 0 20 40 60 Time (in minutes) 80 100 WRITING PROCESS THEORY: A DYNAMIC & FUNCTIONAL APPROACH 25 Figure 5. The changes in correlation (vertical axis) between generating and text quality in two conditions: raw correlation (solid line) and purified for rereading (dashed line). Reanalysed data taken from Breetvelt, Van den Bergh & Rijlaarsdam (1996). 1.0 Correlation 0.6 Generating 0.2 Generating -0.2 -0.6 0 20 40 60 Time (in minutes) 80 100 26 VAN DEN BERGH & RIJLAARSDAM Figure 6. Changes in the probability of occurrence during the writing process for indi- vidual writers (left handed panel Generating; right handed panel Reading the assignment and documentation). Data from Breetvelt, Van den Bergh & Rijlaarsdam (1996). Generating Reading the assignment 0.8 Probability of occurrence Probability of occurrence 1.0 0.8 0.6 0.4 0.2 0.0 0.6 0.4 0.2 0.0 0 20 40 60 Time (in minutes) 80 100 0 20 40 60 Time (in minutes) 80 100 WRITING PROCESS THEORY: A DYNAMIC & FUNCTIONAL APPROACH Figure 7. 27 The temporal distribution over the writing process for two cognitive activities of two writers (Reading the assignment and Generating). Data from Breetvelt, Van den Bergh & Rijlaarsdam (1996). Probability of occurrence 1.0 0.8 Generating _2 Generating _2 0.6 Reading A_1 Generating _2 0.4 Generating_1 0.2 Reading A_2 0.0 Reading A_2 ReadingReading A_2 A_1 Generating_1 10 30 50 Time (in minutes) 70 90 28 Table 1 VAN DEN BERGH & RIJLAARSDAM Some examples of the distinguished types of generating activities defined by adjacent pairs. (Slightly revised adapted from Van den Bergh & Rijlaarsdam, 1999, p. 105) Type of generating Protocol Fragment (some context; adjacent pairs in italics) activity (Reading Assignment:) ‘....As adult who starts living alone, you can AssignmentDriven-Generation get a hard time. You must pass the time yourself. Social contacts were left at the parental home’ (Thinking:) ‘no friends any more’ (Writing down:) ‘no/less friends’ (Thinking): ‘eehm..’ (Thinking:) ‘... ehm.../this is a citation or something./let me have a Rereading-TextDriven-Generation look/ citation of prof. X, and then...something written down/ (Reads already written text): ‘I state that /’ (Thinking:) ‘what then? Am I against the citation? Yes/.....’ (Reading own text:) ‘It is not good that human beings stay alone’ Translation(Writing): ‘I ..am’ Driven-Generation (Thinking:) ‘I am disagreeing with this statement, I think/ no, I don’t agree/it’s not good....I have to think of something else’ (Thinking:) ‘...should I continue that line of thinking / eeehhhmm/ GenerationDriven-Generation something like../or did I write first something...about..eeh.. how people live nowadays /I think/ eeehh/ nowadays../ yes, nowadays it is very normal that ...people, eehh / no, nowadays they don’t find it that strange’ (Writing:) ‘Nowadays, they don’t find it that strange/...’ Pause-RelatedGeneration (Silence, for at least 0.5 seconds) Yes, I do agree with that point, I think nowadays people are too selfish. WRITING PROCESS THEORY: A DYNAMIC & FUNCTIONAL APPROACH 29 Appendix: Summary of regression weights per writing episode: the relation between occurrences of cognitive activities and the quality of the resulting text.(+ positive relation; - negative relation; blank (no significant) relation). Adapted with correction from Breetvelt, Van den Bergh & Rijlaarsdam, 1994. Cognitive Activity 1 Writing Process Episode 2 Reading writing task and documentation Self instructions Goal Setting Generating ideas Structuring Meta comments Pausing Writing/Text production Rereading text Evaluating text Revising text + – – – + + + – – 3 + + + + + – – –
© Copyright 2026 Paperzz