Error correction strategies of professional speech recognition users: Three profiles Mariëlle Leijten a, * , Daniel Janssen b , Luuk van Waes a a University of Antwerp, Belgium b Utrecht University/UiL-OTS, The Netherlands article info Article history: Available online 29 March 2010 Keywords: Cognitive processes Error correction Inputlog Keystroke logging On-line writing processes Pauses Pause analysis Research methods Speech recognition Writing modes Writing observation abstract One of the challenges in writing research in general is to explain the structural variation in writing pro- cesses within and between subjects. More or less recursivity has been attributed to writing experience, proficiency, task characteristics and the writing mode or medium. This study focuses on professional writers (n = 10) who use a modern writing instrument – speech recognition – as their primary tool for text production and revision. More specifically we are interested in the way this new technology affects the cognitive processes that underlie text production. In our study we have focused on error correction. We provide a description of the errors that professional speech recognition users need to deal with, how they deal with them and why they opt for various error correction strategies. Different converging research methods were used: (1) product, (2) process, and (3) protocol analysis. The results are described on two levels: the overall level and the subgroup level (three writer groups). The results show that the contrast between immediate and delayed error correction is quite decisive for the way in which writers structure their writing process. Next to this, the distinction between technical problems and revisions also plays an important role. Most writers prefer solving technical problems immediately. The same does not necessarily hold for other revisions. However, the revision behavior is not random: overall results show three distinct patterns or profiles of error correction. First, there are writers who prefer writing a first time final draft and solve technical problems immediately as well as revising the text produced so far immediately (handle profile). Second, writers who solve more than half of the deficiencies in the text produced so far immediately, but who also delay or postpone various tech- nical problems and revisions (postpone revisions profile). Finally, writers who prefer delaying error correc- tion and who delay technical problems to a second draft (postpone technical problems profile). Ó 2010 Elsevier Ltd. All rights reserved. 1. Introduction One of the challenges in writing research in general is to explain the structural variation in writing processes within and between subjects. Some writers seem to prefer a more or less linear path; they plan first, then formulate and finally revise with little or no recursion (Schilperoord, 1996; Selzer, 1983, 1984). Some writers go through a number of cycles of planning, translating and revising (Flower & Hayes, 1981; Flower, Hayes, Carey, Schriver, & Stratman, 1986; Hayes, Flower, Schriver, Statman, & Carey, 1987). Others go through a more or less chaotic process with some recursion and some linearity in which it is hard to distinguish a pattern at all. But even within subjects we see differences in the way they orga- nize their writing processes. Within the single writing process of a single writer we may find phases in which the writer works more linearly or recursively (Van den Bergh & Rijlaarsdam, 1996; Van Waes & Schellens, 2003). Or the same writer may approach specific writing tasks differently. Simple or well-rehearsed tasks, for in- stance, may be a matter of mere ‘knowledge telling’ and thus be executed in a more or less linear fashion; other tasks may require ‘knowledge transforming’, which often leads to more recursion (Bereiter & Scardamalia, 1987). Many researchers have come up with explanations for the ob- served differences. More or less recursion has been attributed to writing experience, monitor-configuration, proficiency, task char- acteristics and – of course – the writing mode or medium. In this article, we relate writing strategies to the use of new writing media, viz. speech recognition as a dictation device (for a historical review, see Honeycutt, 2003, 2004). More specifically, we are interested in the way this new writing technology affects the cognitive pro- cesses of writers on the job. As explained later in this paper, we will focus on a particular aspect of writing, namely error correc- tion. Previous studies on the influence of speech recognition (SR) on the writing process (Leijten, 2007a, 2007b; Leijten & Van Waes, 2005) have explored how novice speech recognition users adapted to the new writing medium. Professional writers were observed in 0747-5632/$ - see front matter Ó 2010 Elsevier Ltd. All rights reserved. doi:10.1016/j.chb.2010.02.010 * Corresponding author. Address: Department of Management, University of Antwerp, Prinsstraat 13, BE 2000 Antwerp, Belgium. Tel.: +32 265 5072. E-mail address: marielle.leijten@ua.ac.be (M. Leijten). Computers in Human Behavior 26 (2010) 964–975 Contents lists available at ScienceDirect Computers in Human Behavior journal homepage: www.elsevier.com/locate/comphumbeh