This paper describes in detail the design of the specialized text predictor for patients with Huntington's disease. The main aim of the specialized text predictor is to improve the text input rate by limiting the phrases that the user can type in. We show that such specialized predictor can significantly improve text input rate compared to a standard general purpose text predictor. Specialized text predictor, however, makes it more difficult for the user to express his own ideas. We further improved the text predictor by using the sematic database to extract synonym, hypernym, and hyponym terms for the words that are not present in the training data of the specialized text predictor. This data can then be used to compute reasonable predictions for words that are originally not known to the text predictor.