In this research project we are interested in the research question 'What is the function of the textual task environment ('text already produced') on the organization and the process of writing?' Previous experiments showed that writers have different strategies of dealing with the imperfectness of the text produced so far (TPSF).

In this experiment we explore the reasons why writers respond differently to deficiencies in the text they are writing. In that perspective, we have analyzed the writing behavior of ten expert writers that used speech recognition to write a business report. Because of its hybrid character, the speech recognition writing mode helps us to bring characteristics of writing processes to the surface that were previously less explicit. Keystroke logging, speech logging and prompted retrospective protocols were used to elucidate the writers’ cognitive processes. The data show us two diverse profiles: (a) a handle profile (solve problems in the text immediately) and (2) a postpone profile (delay error correction to a later stage in the writing process. In the presentation we explore different explanations why writers choose these different strategies.   

Context

Previous experiments showed that writers have different strategies in dealing with the imperfectness of the text produced so far (TPSF). Conducting experiments in this field with the use of speech recognition software helps us to bring characteristics of writing processes to the surface via the hybrid character of speech recognition that were previously less explicit. Theoretically, the interaction with the text on the screen could lead to a highly recursive organization of the writing process, but it could also lead to a less recursive process. Anyway, the correction of errors in the text produced so far in speech recognition could interrupt the linearity of the writing process in a different way than it does in keyboard based word processing. Error correction appears to be 'the key' for two reasons. On the one hand the correction of errors reveals an explicit awareness of the writing modes. This is shown by the difference in using speech input during error correction, and the related switches between writing modes (Leijten & Van Waes, 2005)). Writers need to compare their mental representation of the text with the visual representation. On the other hand the focus on error correction enables us to analyze the interaction with the text produced so far in a specific way. Writers who rely heavily on the text produced so far to prompt new ideas and guide planning ('reading to write'), can have difficulties because of the flaws in the text on the screen.  In other words, speech recognition can be seen as a perspective to gain better insight in the cognitive processes that are related to writing in general. In this experiment we would like to further explore the reasons why writers respond differently to these deficiencies.

Method

In this present experiment 10 professional writers (and expert users of speech recognition) wrote a business report using speech recognition software (Dragon Naturally Speaking 8.1, Dutch version). We combined converging research methods. The writing process data were logged with Inputlog (www.inputlog.net) and the writers were simultaneously observed via Morae.  Both programs did not constrain the subjects' behavior. In a retrospective interview the writers were asked to answer questions about their writing process during a retrospective interview - based on the concurrent observations and prompted by the selected fragments of their own writing process. The focus of the questions was on technical problems caused by the writing medium and on content based revisions.

Results

Based on the information of the experiment we can distinguish two diverse profiles:

1. Handle profile: (a) reading of the TPSF – (b) detection of the error - - - (c) diagnosis – (d) editing – (e) completion of the text.

Writers who do not delay the error correction follow this process linearly in the 'Handle profile'. Writers read the TPSF to evaluate and then choose a procedure to fix the detected text problem on a continuum from 'ill-defined' to 'well-defined' problems (Hayes et al. 1986). A small difference can be made in the initial text completion between writers who only detect the problem: they will prefer to 'say it again' (abd(ce) (rewrite) and writers who also make a thorough diagnoses of the problem and prefer to 'say it differently' (abcde) (revise).

2. Postpone profile: (a) reading of the TPSF – (b) detection of the error - (e) completion of the text - ((a) reading of the TPSF – (b) detection of the error))-  (c) diagnosis – (d) editing.

When following the 'Postpone profile' writers do not really 'read' the TPSF in order to comprehend, evaluate and define problems, but they are mainly focused on grasping the main gist of the text in order to produce new text. Detecting problems is of secondary importance. Only after completing (part of) the text they decide to diagnose possible errors and decide to correct them.

Both strategies consist of sub strategies. However, the main goal of this study is to describe why writers sometimes prefer strategies to solve problems (either technical problems or revisions) in their text immediately or sometimes prefer to postpone this process. Another issue related to the strategy writers use - and which might shed a light on the cognitive load of problems in the TPSF - is the use of writing modes during the execution of repairs. What strategies do writers use to make modifications to the TPSF? To answer this question we will describe the repairs in detail based on our categorization model (Leijten & Van Waes, 2005).