2001 Conference Proceedings

Go to previous article 
Go to next article 
Return to 2001 Table of Contents


Speech Recognition Technology in the Writing Workshop: Advantages for Students with Learning Disabilities

Thomas Quinlan
Scott Beers
University of Washington

This study examines a Writing Workshop using speech recognition technology (SR), where students with learning disabilities (LD) participated with normal achieving (NA) students. By translating utterances into written words, SR may help mitigate the writing problems of students with LD (De la Paz, 1999), which are at least partly due to the mechanical aspects of getting words on the page (Graham, 1990). Inefficient handwriting processes can compound writing problems by interfering with other cognitive processes (Bourdin & Fayol, 1994), such as planning and revising. Motivation may also be adversely affected, since students may perceive handwriting difficulties as a sign of poor writing skills. According to self-efficacy theory, negative judgements of their own writing capabilities can lead students to avoid writing (Bandura, 1986). While SR may afford students with LD an easier mode of transcription, we also wondered whether it might offer particular advantages to students with LD in the context of a Writing Workshop.

Handwriting dysfluency is a general disadvantage in the classroom, where many activities require handwritten answers. But the Writing Workshop in particular presents formidable obstacles for students with LD. Since Atwell’s "In the Middle" (1987) described the Writing Workshop approach to writing instruction, teachers have increasingly employed it to emphasize the process aspect of writing. Accordingly, the writing classroom has been socially redefined as a community of writers. Ironically, while the Writing Workshop may empower normally-achieving students to assume ownership of their own writing, it may also effectively exclude students with LD. Participation in a classroom community of writers depends to a large extent upon the ability to produce a text. For many Writing Workshop activities, a completed draft is a "ticket" to participate. Because of their difficulties in producing a satisfactory draft, students with LD may be excluded from such activities. On the other hand, if they do complete a draft, they may be embarrassed by its appearance, whether due to brevity, illegibility, or misspelled words. The Writing Workshop can be a dreadful place for students with LD.

Is SR an optimal mode of transcription (Reece & Cumming, 1996)? Dictation studies indicate that speech-based transcription can help students write more (e.g. Scardamalia, Bereiter, & Goelman, 1982), and those with LD to compose better quality text (Graham, 1990). Yet, by displaying the emerging text, SR is superior to dictation. In a series of studies comparing writing modes, Reece and Cumming (1996) found that texts composed by poor younger writers using simulated SR were longer and rated superior in quality to handwritten texts. Higgins and Raskind (1995) found that a group of college students with LD composed higher quality texts using a discrete speech SR system, relative to those composed via a) unassisted handwriting or b) a human transcriber. The best predictor of holistic quality was words with "seven or more letters," suggesting that the SR system allowed students to use longer words that they could not spell. In Wetzel (1996), one sixth grade student with LD, who normally did not produce handwritten texts, was able to compose short narratives using a discrete speech SR system. Because of the high-frequency of recognition errors made by the SR systems employed in the two previous studies, these results may be distorted or at best underinflated.

With recent improvements, the accuracy of the technology has increased, making a SR Writing Workshop seem feasible. The following study focuses upon seven students using SR in a Writing Workshop, each writing their own chapter in a serial narrative project. Our research questions were two: 1) Is SR viable for the classroom and 2) would SR prove to be an effective writing tool for students with LD? We hypothesized that students would be able to compose substantial texts using SR. In addition, we expected that longer texts, and the relative ease of SR transcription, would contribute to substantial mean-level revising. Evidence of longer texts and adequate revising behavior would suggest that SR may enable students with LD to participate more effectively in a classroom Writing Workshop. Finally, we expected SR to make these students with LD to feel like more successful writers.

Method

Participants

Promoted as a Writing Workshop for students with LD, the summer session writing class attracted ten students between the ages of 10 and 14. At the beginning of the four-week session, students chose to learn and compose by either word processing (Microsoft Word 3.0) or SR (Dragon Naturally Speaking, Professional 4.0). Three students chose word processing and seven SR. Four of these SR students had been previously identified by their schools as LD, and they became the focus of this study. A pretest indicated that these four students demonstrated handwriting and/or spelling skills at least two grade levels below the norm. The other three SR students had not been so identified and did not demonstrate significant writing-related deficiencies.

Materials

The class was conducted in a computer lab with a complement of Pentium III-based PCs (450 Megahertz; 128 Megabytes RAM), each equipped with a Soundblaster sound card and a Plantronics (VX1) headset microphone. Before Naturally Speaking, Professional 4.0 can be used effectively, a voice profile must be created for each student. This was accomplished on the second day, with students receiving two hours of direct SR system instruction on the third day. The training included instruction in error-correction, the process whereby students fix recognition errors made. Subsequently, all SR skill instruction occurred in the context of writing activities.

Design and Procedure

Located on the campus of a Seattle area private school, the class took place five days a week, three hours a day, for four weeks. The authors of this proposal served as instructors. Our primary goal was to encourage students to compose a lot of text. Consequently, we attempted to plan writing activities that were fun and personally relevant to the students. Both instructors shared the philosophy that writing should serve an authentic communicative purpose, and that writers should take ownership of their own writing. Accordingly, students were informed that the finished products of the three major projects would be posted on a Web page, which then could be read by friends, their parents—and the rest of the world. Very little direct writing instruction was given; however, planning and revising activities were integral to virtually all writing activities.

The serial narrative project began at the beginning of the second week. The instructions were minimal. Students were asked to plan and write a story, with each student contributing one chapter. All students were to have a ‘voice’ in the planning sessions. The instructors guided the composing process by organizing a series of planning and revising activities. After composing the initial draft of their chapters, each student made three or four revisions. From the initial planning, through each successive revision, the students remained very enthusiastic about their story, which came to be titled "Automatic Eddy."

Instrumentation

Three sources of data informed our analysis: field notes, revised drafts, and end-of-session assessments. Field notes recorded our observations of the SR Writing Workshop. These were coded to describe student activities, particularly those pertaining to a) student-machine interaction and b) the writing process. All drafts were scored for text length, as the total number of written words. Revised drafts were coded according to type of revision change, based upon categories of syntactic level, type of operation, change of meaning, and quality change (Bridwell, 1980; Faigley & Witte, 1981). Two other assessments were administered. A handwritten narrative task was given at the beginning and end of the course, with a SR version being included in the end-of-session assessment. At posttest, a self-efficacy questionnaire (adapted from Shell, Murphy, & Bruning, 1989) assessed students’ self-perceptions of their capacity to complete various writing tasks using alternate modes, SR and handwriting.

Results

Observations indicated that SR was well suited for the Writing Workshop. Students were often engaged in their writing tasks, and even seemed to enjoy writing with SR. This may be due to many factors: The novelty of the technology, the exciting nature of their story "Automatic Eddy," or perhaps the radical departure from handwriting and spelling. Nevertheless, students seemed to appreciate SR as an alternative mode of writing. Students were generally not distracted by the sound of other dictating voices, with some exceptions (see below). While dictating their chapters, students paused often, which effectively dissipated the sound of the various voices. These observations affirmed the viability of SR in a Writing Workshop setting.

In writing their chapters for "Automatic Eddy," the four students with LD differed widely in SR dictating skill and chapter writing performance. This group as a whole tended to write less, and make fewer revision changes than the NA students. However, the high variability within both groups on most measures undermined the usefulness of descriptive statistics in reporting. Consequently, the nature of the data compelled us to focus our analysis on two students, John and George, who seemed to represent a range of performance in both writing and SR dictating.

George

George was consistently able to dictate effectively to the SR system. He quickly mastered the operation of the program, such that by the third and fourth week his error rate was quite low. This was attributed to his development of a dictation style that facilitated accurate recognition by the system. He judged himself to be much more capable of completing various writing tasks using SR than handwriting.

George’s favorite phrase was "I’m done." In writing his chapter, George composed a substantial (186 words) draft, then proceeded to revise very little in the subsequent weeks. However, the six revision changes he did make were mostly sentence-level and meaning changing, rather that surface level. He dictated very rapidly, producing a respectable draft very quickly. Run on sentences remained a problem, one which SR did not alleviate. In George’s case, SR appears to have increased the fluency of text production, but did little to improve revising behavior. His writing problem seems localized to spelling. While proud of being one of the best readers in his class, he admits to being a poor speller. Inspection of his handwritten narrative posttest, while substantial (103 words), indicates numerous misspelled words. By eliminating the need to spell words letter-by-letter, SR appears to have removed one main obstacle to text production for him.

John

John was less able to use SR effectively. He continued to be plagued by numerous recognition errors into the third and fourth weeks. This was curious, because he had already received SR training prior to the beginning of class, and even had his own system. He experienced similar difficulties whether working on his own or classroom PCs. He was often observed in off-task behavior, such as ferreting out non-utilitarian aspects of the SR program. In addition, he sometimes became impatient with his inability to cope with the errors. John remained relatively oblivious to how the tone and pitch of his voice affected recognition accuracy. For example, he sometimes drew complaints from his classmates, who claimed his loud voice was disrupting their SR system. In spite of being one of the less effective SR users, he still judged himself to be a more effective writer with SR than handwriting.

John’s initial draft of his chapter was relatively short (52 words), but he did make substantial revision changes (11). These were mostly "quality improving" changes made at the sentence and phrase level. The final revision, however, was not long (70 words). The narrative posttest indicated that John composed more words in SR mode (70) relative to handwriting (38).

Discussion

SR appears to be viable in the Writing Workshop, and an advantageous mode of transcription for this group of students with LD. In spite of divergent writing and SR dictating skill, both George and John a) experienced gains in writing fluency and b) judged themselves to be more effective writers using SR. These two findings are especially important, since the relationship between writing self-efficacy and writing performance are reciprocal: The experience of successful writing can increase self-efficacy, while positive self-efficacy can lead student writers to persevere during difficult writing tasks (Schunk, 1989). Improvement in these two basic aspects of writing could help students with LD to participate more fully in classroom writing activities such as Writing Workshops. In the long run, improvement in these two basic aspects of writing could serve as a basis for gains in overall writing skill.

References

Atwell, N. (1987). In the middle: Writing, reading, and learning with adolescents. Portsmouth, NH: Heinemann.

Bandura, A. (1986). Social Foundations of thought and action: A social cognitive theory. Englewood Cliffs, NJ: Prentice Hall.

Bernhardt, S. A., Wojahn, P. G., & Edwards, P. R. (1990). Teaching college composition with computers. Written Communication, 7 (3), 342-374.

Bourdin, B., & Fayol, M. (1994). Is written language production more difficult than oral language production? A working memory approach. International Journal of Psychology, 29 (5), 591-620.

Bridwell, L. S. (1980). Revising strategies in twelfth grade students’ transactional writing. Research in the Teaching of English, 14 (3), 197-222.

de la Paz, S. (1999). Composing via dictation and speech recognition systems: Compensatory technology for students with learning disabilities. Learning Disabilities Quarterly, 22 (3), 173-185.

Faigley, L., & Witte, S. (1981). Analyzing revision. College Composition and Communication, 32, 400-414.

Gould, J. D. (1980). Experiments on composing letters: Some facts, some myths, and some observations. In L. W. Gregg & E. R. Steinberg (Eds.), Cognitive processes in writing. Hillsdale, NJ: Lawrence Erlbaum Associates.

Graham, S. (1990). The role of production factors in learning disabled students’ compositions. Journal of Educational Psychology, 82 (4), 781-791.

Graham, S., & MacArthur, C. (1987). Written language of the handicapped. In C. Reynolds & L. Mann (Eds.), Encyclopedia of special education (pp. 1678-1681). New York: John Wiley & Sons.

Hidi, S. E., & Hildyard, A. (1983). The comparison of oral and written productions in two discourse types. Discourse Processes, 6, 91-105.

Higgins, E. L., & Raskind, M. H. (1995). Compensatory effectiveness of speech recognition on the written composition performance of postsecondary students with learning disabilities. Learning Disability Quarterly, 18, 159-174.

MacArthur, C. A. (1988). The impact of computers on the writing process. Exceptional Children, 54, 536-542.

McCutchen, D. (1987). Childrens’s discourse skill: Form and modality requirements of schooled writing. Discourse Processes, 10, 267-286.

Reece, J. E., & Cumming, G. (1996). Evaluating speech-based composition methods: Planning, dictation, and the listening word processor. In C. M. Levy & S. Ransdell (Eds.), The science of writing (pp. 361-380). Mahway, NJ: Lawrence Erlbaum Associates.

Scardamalia, M., Bereiter, C., & Goleman, H. (1982). The role of production factors in writing ability. In M. Nystrand (Ed.), What writers know: The language, process, and structure of written discourse (pp. 75-210). San Diego, CA: Academic Press, Inc.

Schunk, D. H. (1989). Social cognitive theory and self-regulated learning. In B. J. Zimmerman & D. H. Schunk, Self-regulated learning and academic achievement: Theory, research, and practice (pp. 83-110). New York: Springer-Verlag.

Shell, D. F., Colvin-Murphy, C., & Bruning, R. H. (1989). Self-efficacy and outcome expectancy mechanisms in reading and writing achievement. Journal of Educational Psychology, 81 (1), 91-100.

Wetzel, K. (1996). Speech-recognizing computers: A written-communication tool for stu9999dents with learning disabilities. Journal of Learning Disabilities, 29 (4), 371-380.


Go to previous article 
Go to next article 
Return to 2001 Table of Contents 
Return to Table of Proceedings


Reprinted with author(s) permission. Author(s) retain copyright.