16/04/24 - Emmanuella Martinod (LISN, University Paris-Saclay & CNRS)

Title: "Sign language modeling through a corpus-based approach: AZee"

AZee is an approach to formally represent French sign language, LSF. It has initially been designed for SL modeling (synthesis, automatic recognition, or computer-assisted translation). However, this approach shares common goals with SL linguistics. Indeed, while respecting the constraints associated with a modeling process, it aims to represent LSF by accounting for its linguistic properties and specificities. AZee is based on the concept of production rule, i.e. a form-meaning association. Each production rule can contain information traditionally considered semantic, syntactic, or prosodic. LSF utterances can be described by combining the set of existing production rules. If a phenomenon observed in LSF discourse cannot be represented using this set, a methodology exists to enrich the system and create new production rules. In this presentation, I will show how two corpora of different genres were described with AZee. The limits of this process will lead me to introduce the methodology for identifying new production rules in SL data. This methodology will be illustrated by cases of recently extracted production rules for plurality and facial expressions in LSF. Finally, I will discuss how AZee can enrich our understanding of SL in general, for instance, concerning the links between lexicon and illustrative structures and, more broadly, the potentialities to establish a grammar of LSF or even comparison between SLs.