We present the model of language implemented in the words prediction engine of the "Plateforme de Communication Aternative" (PCA), an assistive communication software designed for impaired persons. The model relies on a large coverage lexicon for french langage which provides for each entry its word frequency and its set of associated morphosyntactic categories. The engine includes a user model (personal lexicon for unknown words typed by the user, computation of personal word frequencies, storage of the generated sentences) and a morphosyntactic prediction system which weight the word frequencies of predicted words according to the syntactical context of the sentence. The evaluation of the model of language is quite encouraging : a keystrokes saving rate of 55% for 9 propositions. The major contribution comes from the raw prediction obtained by using the words frequencies of the general lexicon. MOTS-CLÉS : Communication assistée pour personnes handicapées, système de prédiction de mots, modèle utilisateur, prédiction morphosyntaxique.
Download Full PDF Version (Non-Commercial Use)