Avatar signer - Synthesis of French sign language from text

In this paper, we present a multimodal synthesis system that translates text-to-LSF (French sign language) by means of a 3D virtual character, also called virtual signer. Our SignCom system is based on a multichannel composition mechanism, each channel being associated to linguistic information (from the phonological level to the lexical, syntactic or semantic levels), or to articulatory information. The composition is based on a spatio-temporal arrangement of these elements that are parallelized, and is able to edit and generate utterances in LSF. The new synthesis modules that enrich the initial system are described, including body movement synthesis, facial and hand movement synthesis. They implement grammatical dynamics specific to sign language, based on the fundamental concepts of spatiality and iconicity. Finally, we present the main technological challenges that remain before concluding.