This paper presents Sibyl, a new AAC (Augmentative and Alternative Communication) computer system, that aims at improving text typing for persons with sever speech and motor impairments. Typical AAC systems display virtual keyboard on screen which enables key selection via a few switches device. However, text typing is cumbersome. Sibyl aims at allowing faster typing by means of two predictive modules SibyLetter and SibyWord. SibyLetter facilitates key selection through a dynamic keyboard by predicting next letter. It takes advantage of a n-gram statistical model applied on letters. SibyWord allows keystroke saving by word completion. It predicts next word using natural language processing techniques. A robust chunk parsing (non-recursive constituent) is achieved and prediction is based on the last chunk heads occurrences. Best predictions are displayed in a word list. This paper presents the two predictive modules and the Sibyl software used in the Kerpape French rehabilitation center.