In this paper, we report about some preliminary experiments in which we tried to improve the performance of a state-of-the-art Predictive Typing system for the German language by adding a collocation-based prediction component. This component tries to exploit the fact that texts have a topic and are semantically coherent. Thus, the appearance in a text of a certain word can be a cue that other, semantically related words are likely to appear soon. The collocation-based module exploits this kind of topical/semantic relatedness by relying on statistics about the co-occurrence of words within a large window of text in the training corpus. Our current experimental results indicate that using the collocationbased prediction module has a small but consistent positive effect on the performance of the system. 1

Word prediction systems (such as those embedded in most current augmentative and alternative communication systems) aim to predict what a user wants to type next on the basis of corpus-extracted n-gram counts. Good performance of such a system depends crucially on the size and quality of the underlying lexicon.

The domain of Augmentative and Alternative Communication (AAC) studies appropriate techniques and systems that enhance or accomplish the retaining or non-existing abilities for interpersonal communication. Some AAC users apply telegraphic language, as they attempt to speed up the interactive communication or because they are language impaired. In many AAC aids, a "sentence" is formulated by combining symbols of an icon-based communication system. To be accepted by the communication partner, the output should be a correct oral sentence of a natural language. In this paper we present our effort to develop a novel technique for expanding spontaneous telegraphic input to well-formed Greek sentences, by adopting a feature-based surface realization for Natural Language generation. We first describe the general architecture of the system that accepts compressed, incomplete, grammatically and syntactically ill-formed text and produces a correct full sentence. The NLP techniques of the two main modules, named preprocessor and translator/ generator, are then analyzed. A prototype system has been developed using Component Based Technology (CBT) which is under field evaluation by a number of speech-disabled users. Currently it supports fully the BLISS and MAKATON icon based communication systems. Some limitations of the module are also discussed along with possibilities for future expansions.

I would like to thank my mother, Mary G. Mair, for being a surrogate mother to my daughter, Rebecca, while I was working on my graduate studies. I would also like to thank both her and my father, Robert D. Mair, for being supportive of my work over the past few years. To my advisors, Kathleen F. McCoy and Patrick W. Demasco, I am indebted for their guidance, patience, and faith in me. When I started working with them in the Natural Language Processing Lab at the Center for Applied Science and Engineering, located at

istique n-gramme, clavier simul dynamique. 1 Introduction Cet article prsente un systme d'aide la communication pour personne handicape. Le handicap tudi concerne les personnes dont l'usage de la parole est fortement ou compltement altr, et dont les facults motrices sont trs rduites (hmiplgie, ttraplgie). C'est le cas de patients IMC (Infirmes Moteurs Crbraux), SLA (Sclrose Latrale Amyotrophique) ou Locked-In Syndrom. Quelle que soit la maladie, le handicap physique est svre, la communication est prive de son support oral habituel et les modalits d'interaction sont limites. 2 Un moyen de communication alternatif est le recours aux systmes dits de communication assiste ou AAC (Alternative and Augmentative Communication). On appelle systme de communication assist tout systme de supplance dont le rle est d'augmenter ou de restaurer, ne serait-ce que partiellement, la fonction de communication. Cette aide rend une certaine autonomie la personne handicape et amliore sa qualit de vie. P