Word prediction: what the research says
Silió, M. C., & Barbetta, P. M. (2010). The effects of word prediction and text-to-speech technologies on the narrative writing skills of Hispanic students with specific learning disabilities. Journal of Special Education Technology, 25(4), 17–32.
This quasi-experimental research study, published in a peer-reviewed academic journal, investigated whether, with writing time held constant, Hispanic students with specific learning disabilities (SLD) under:
- word processing,
- word processing with word prediction,
- word processing with text-to-speech, or
- word processing with word prediction and text-to-speech combined
- perform more effectively in narrative composition writing as defined by writing fluency, syntax, and spelling accuracy,
- improve their overall organization of narrative composition writing, and
- maintain skills in narrative composition writing and organization on maintenance tests given two, four, and six weeks after writing ended.
Participants were six fifth-grade Hispanic boys with SLD who were English Language Learners but no longer warranted ELL services based on English proficiency. A multiple-baseline design across subjects was used in the two separate but related studies. Participants were divided randomly into two equal groups, Cohorts A and B. Before the intervention each participant was provided private 30-minute individualized training on the use of WordQ’s word prediction and text-to-speech. Cohort A and B participants followed all general procedures including the use of Word to respond to the randomly assigned prompt. During the first intervention phase, Cohort A used the word prediction feature of WordQ in conjunction with Word. When the group moved to the second intervention phase, text-to-speech was added. Cohort B participants began the first intervention phase using Word with the text-to-speech feature of WordQ, with word prediction added during the second phase. Treatment fidelity and inter-rater reliability measures were collected by trained observers/raters on 12 of 38 (32%) of baseline, intervention, and maintenance sessions. A holistic rubric was used to evaluate writing samples. The mean inter-rater reliability scores for writing fluency, syntax, spelling accuracy, and overall organization were 99.7%, 91.3%, 99.6%, and 86.6%.
RESULTS: Although outcomes varied for individual participants, the overall results demonstrated that word prediction alone and in combination with text-to-speech had a positive impact on the participants’ writing. With word prediction alone and with text-to-speech, participants in both cohorts wrote longer, more syntactically mature compositions that were better organized and had fewer spelling errors. The use of text-to-speech alone, however, resulted in little or no improvement. With few exceptions, participants maintained a high percentage of the composition skills they developed.
Cullen, J., Richards, S. B., & Frank, C. L. (2008). Using software to enhance the writing skills of students with special needs. Journal of Special Education Technology, 23(2), 33–43.
This research article, published in a peer-reviewed academic journal, investigated the effects of computer software on the writing performance of students with mild disabilities. In particular, it focuses on the effects of a talking word processor with spell checker alone (Write: Outloud) and in conjunction with word prediction software (Co:Writer) on journal entries. The study addressed the specific research question: “What are the effects on the performance of seven students with special needs when a talking word processor with spell checker software is used independent of and in conjunction with word prediction software as accommodations in daily writing exercises?” Participants were seven fifth grade students with mild disabilities, diagnosed with a mild cognitive delay or a learning disability. The researchers used a case study approach with a modified multiple baseline. The study contained three phases: baseline, intervention using a talking word processor, and intervention using word prediction software in conjunction with a talking word processor. The modification of the multiple-baseline design involved simultaneous implementation of the intervention phases with each participant rather than in a staggered fashion as in the traditional multiple-baseline design. Baseline writing samples were converted from handwritten. However, they were converted to word processed text to eliminate the possibility of scoring bias. The two intervention phases lasted three weeks each, with a maximum of nine writing samples per participant in each phase. All writing samples were quantitatively assessed on:
- mean number of words,
- mean number of misspellings,
- accuracy percentage, and
- total rubric score.
RESULTS: In general, both supports positively impacted the students’ writing, Co:Writer more so than Write: Outloud. As a whole group, the participants improved on each dependent variable during both intervention phases. However, these effects did not yield uniform outcomes across students. Collins, Richards and Lawless Frank conclude that while computer software that provides writing supports such as text-to-speech, spell checker and word prediction benefits students with disabilities, it is necessary to consider each student’s individual strengths and weaknesses when choosing software.
Mirenda, P., Turoldo, K., & McAvoy, C. (2006). The impact of word prediction software on the written output of students with physical disabilities. Journal of Special Education Technology, 21(3), 5–12.
This quasi-experimental research study, published in a peer–reviewed academic journal, examined the impact of word prediction software on written output. The research questions were:
- What are the perceptions of students with physical disabilities and their adult supporters about the benefits of using a word prediction program such as Co:Writer?
- Are there significant differences in the rates of text entry; the proportions of legible words, correctly spelled words, and correct word sequences; and/or the mean lengths of consecutive correct word sequences produced by students with physical disabilities using handwriting, word processing, and word processing + word prediction software?
The participants included 24 students, 16 males and 8 females, with physical disabilities that affected their ability to write by hand. Fifteen attended elementary school, one attended middle school, and eight attended high school. Twenty of the participants were enrolled in general education classrooms with six of them receiving part-time support in resource rooms. The remaining four participants attended congregated classrooms for students with special needs. Surveys and three writing samples (10 minutes about “something you like to do”) using:
- a computer with word processing software only, and
- a computer with both word processing software and Co:Writer
The order of writing mode was counterbalanced across students to control for an order effect. Two-thirds or more of the students believed that Co:Writer helped them to spell better; use a wider variety of words; write faster; produce neater, easier-to-read work; and write more correct sentences. More than half of the teachers/adult supporters also identified all but one of these benefits: write more correct sentences. Two-thirds or more of the teachers/adult supporters believed that Co:Writer helped their students to write more without tiring, experience less frustration when writing, and read what they had written. More than half of the students noted the same benefits. One-way repeated-measures analyses of variance (ANOVAs) were used to determine whether there were significant differences between the 10-minute writing samples produced with Co:Writer, word processing, and handwriting with regard to five variables:
- total number of words written,
- percent of legible words,
- percent of words spelled correctly,
- percent of correct word sequences, and
- mean length of consecutive correct word sequences.
RESULTS: Results indicated that the word processor with word prediction supported students to achieve higher percentages of legible words, correctly spelled words, and correct word sequences than handwriting. However, use of this assistive technology did not significantly impact students’ rate of text production. Interestingly, almost three-quarters of student-adult pairs believed that Co:Writer helped students to write faster and write more without tiring (i.e., produce more words per minute), neither of which was supported by the writing sample data.
Tam, C., Archer, J., Mays, J., & Skidmore, G. (2005). Measuring the outcomes of word cueing technology. Canadian Journal of Occupational Therapy, 72(5), 301–308.
This research article, published in a peer-reviewed scientific journal, reports the findings of a one-year program evaluation project in which the Canadian Occupational Performance Measure (COPM) was used as an outcome measure to evaluate the effectiveness of WordQ. The study occurred in a clinical setting at a writing aids clinic in a pediatric rehabilitation centre in Toronto, Ontario. The clinic’s clients are children aged 19 and under who have a physical disability that affects their ability to write. All children seen at the writing aids clinic using WordQ (Version 1) were included in the 12-month data collection period between 2003 and 2004. These children may have received handwriting training at some point in their life, but they had come to the clinic because they had been identified as non-functional writers and were in need of technological intervention. The COPM was successfully conducted with 42 children and their families, but only 29 families responded to the request for a follow-up interview. Of the 29 children, 15 were female and 14 were male. The average age of the children was 11.1 (SD = 3.7, range = 3.9–19.1).
RESULTS: Generally, families and children found WordQ helpful. They reported increased productivity, increased motivation to write, and use of a broader variety of words in writing. Children gained enhanced independence, as they did not need their parents to be available to help them with spelling. Parents reported that their children were more willing to experiment with words and were therefore using a richer variety of words in their writing. The findings of this project indicate that the COPM is an effective tool for measuring children’s perceived outcome of word cueing technology. The use of the COPM supports evidence-based, client-centred assistive technology practice. The findings also support the effectiveness of WordQ to enhance written productivity.
Handley-More, D., Deitz, J., Billingsley, F. F., & Coggins, T. E. (2003). Facilitating written work using computer word processing and word prediction. The American Journal of Occupational Therapy, 57(2), 139–151.
This quasi-experimental research study, published in a peer-reviewed academic journal, investigated whether occupational therapy intervention that focused on teaching children to use word processing, either alone or with word prediction, was effective in improving the written communication skills of children with learning disabilities and handwriting problems. A single-subject alternating treatments design was replicated across three children in grades 4 and 5. During the baseline phase the children wrote stories by hand; during the intervention phase, the children wrote stories, alternating among handwriting, word processing, and word processing with word prediction. Each student was asked to look through a packet of pictures and select 36 that he or she could write about. These pictures were randomly assigned to each session. During the six baseline sessions, the students produced handwritten stories about their preselected pictures. Each student was prepared for the intervention phase by being individually trained in keyboarding, word processing, and word prediction. During the intervention sessions, the students were asked to write stories about their preselected pictures using the three text production methods (handwriting, word processing, and word processing with word prediction). A random numbers table was used to assign handwriting, word processing, or word processing with word prediction to each session in order to control for sequential confounding. Dependent variables focused on percentages of legible words, percentages of correctly spelled words, total amount written, and rate of writing. Prior to the intervention phase of the study, a second rater was trained in evaluating student writing until the level of inter-rater agreement was at least 80% on all parameters. The second rater evaluated randomly selected stories for each student. Agreement was checked for two stories from the baseline phase and one-third of the stories from the intervention phase. Average inter-rater agreements for the student stories written during the baseline (BL) and intervention (I) phases were as follows: legible words (BL = 95%, I = 96%); correctly spelled words (BL = 94%, I = 97%); and total words (BL = 100%, I = 99%). To ensure the use of standard procedures within and across conditions, procedural reliability checklists that delineated equipment set-up, the amount and type of reinforcement, and when cues could be provided were developed for each text generation method. The checklists were used at each session. An observer checked procedural reliability at 18% of the sessions. Overall procedural reliability scores remained at 100% during the baseline and ranged from 96% to 100% (average = 99.6%) during the intervention.
RESULTS: Results were variable. Two children had clear improvements in legibility when using either word processing alone or with word prediction. These same children demonstrated clear improvements in spelling when using word prediction. Though rate of writing was best for two children when using handwriting, relative to total amount produced, one method was not clearly preferable to another. This suggests that occupational therapy intervention involving word processing and word processing with word prediction can facilitate written work by improving legibility and spelling for some students with learning disabilities. The success of the technology varied depending on each child’s unique needs, talents, and environmental supports. This suggests that it is important to evaluate each child individually and provide training and ongoing support for technology use.