The Predict4All open-source word prediction and word correction engine in French language: a feasibility study.

Abstract

Purpose: Information and communication technologies are crucial for social and professional integration, but access to technology can be difficult for people with physical impairments. Text entry can be slow and tiring. We developed a free and open-source module called Predict4All for use with AAC (augmentative/alternative communication) software in French language. It offers spelling error-tolerant word prediction. We aimed to evaluate the feasibility of using the system in people with learning or physical impairments and to perform a preliminary evaluation of its efficacy on spelling correctness and text input speed.

Materials and methods: We conducted a prospective interventional, two-center feasibility study between February 2020 and March 2022. We included people with learning difficulties or neurological diseases impacting verbal or written communication. We evaluated fatigue, cognitive load and satisfaction, spelling correctness and text input speed and technical issues. We used 3 experimental setups: WITHOUT (neither prediction nor correction), PRED (prediction without correction), PREDCORR (prediction with correction).

Results: Twelve participants completed the study. Fatigue (all conditions), cognitive load, satisfaction (PRED and PREDCORR) did not differ between Day 0, Day 30 and Day 120, for either center. No technical issues were reported. The percentage of errors decreased significantly between Day 30 and Day 120 with word prediction software (p = 0.026), with no difference between PRED and PREDCORR (p < 0.447).

Conclusion: Our study has shed light on the potential of the Predict4All engine with successful results, as the system performed effectively without technical or methodological obstacles. Furthermore, the engine has been adopted in clinical practice at both centers by both healthcare professionals and users.