The Intelligent Voice Prosthesis is a communication tool which reconstructs the meaning of an ill-structured sequence of icons or symbols, and expresses this meaning into sentences of a Natural Language (French). It has been developed for the use of people who cannot express themselves orally in natural language, and further, who are not able to comply to grammatical rules such as those of natural language. We describe how available corpora of iconic communication by children with Cerebral Palsy has led us to implement a simple and relevant semantic description of the symbol lexicon. We then show how a unification-based, bottom-up semantic analysis allows the system to uncover the meaning of the user's utterances by computing proper dependencies between the symbols. The result of the analysis is then passed to a lexicalization module which chooses the right words of natural language to use, and builds a linguistic semantic network. This semantic network is then generated into French sentences via hierarchization into trees, using a lexicalized Tree Adjoining Grammar. Finally we describe the modular, customizable interface which has been developed for this system.