Word and Sentence Prediction: Using the best of the two worlds to assist AAC users

Issue title: Papers from the AAATE 2013 Conference

Guest editors: Pedro Encarnação

Article type: Research Article

Authors: Garcia, Luís^{a; *} | de Oliveira, Luis^b | de Matos, David^b

Affiliations: [a] Polytechnic Institute of Beja, Rua Pedro Soares, Beja, Portugal | [b] INESC ID Lisboa, Instituto Superior Técnico, Universidade de Lisboa, Lisboa, Portugal | Catholic University of Portugal, Portugal

Correspondence: [*] Corresponding author: Luís Garcia, Polytechnic Institute of Beja, Rua Pedro Soares, 7800-295 Beja, Portugal. Tel.: +351 284 311 540; Fax: +351 284 361 326; Mobile: +351 96 684 84 16; E-mail: [email protected].

Abstract: Background:Persons who rely on Augmentative and Alternative Communication (AAC) systems face tremendous difficulties to maintain a conversation, in part as a consequence of a poor output rate. Word Prediction is a popular AAC technique that can save up around 50% keystrokes. However, only modest communication rate improvements have been reported in the literature. Objective:Therefore, we studied the effect of using Sentence Prediction as a complementary, and faster technique, to Word Prediction, in a text-based AAC system. Methods:To evaluate this strategy we conducted user tests with a Word and Sentence Prediction prototype we have been developing for a client from a local rehabilitation center. Communication rate was measured with the system having full and partial knowledge of sentences to be composed. Results:With able bodied non-AAC users, mean rates of 18.8 WPM and 21.0 WPM were obtained, respectively, combining Sentence Prediction with Word Prediction, and using Sentence Prediction only. The Sentence Prediction with Word Prediction was the fastest configuration for the AAC user participant, with 7.2 WPM. These results were obtained with the system having knowledge about all the sentences the subjects had to produce (100% sentence knowledge). In a subsequent test, sentence knowledge conditions were degraded to measure performance under non-ideal conditions. The conditions with less sentence knowledge (25% and 0%) had results close to the Word Prediction only condition, around 8 WPM for the able bodied users and 1.3 WPM for the AAC user, which is an indicator that under low sentence reuse conditions Sentence Prediction does not compromise user performance. Conclusions:Since Sentence Prediction can potentially improve communication rate, we think this technique should be considered as a valuable complement to Word Prediction on text-prediction AAC solutions.

Keywords: Augmentative and alternative communication, rate enhancement techniques, vocabulary prediction, word prediction, sentence prediction

DOI: 10.3233/TAD-140406

Journal: Technology and Disability, vol. 26, no. 2-3, pp. 79-91, 2014