Translating Sentences from Written French into LSF Annotation
Résumé
The automatic translation of written language into sign language can be approached as a two-stage process. In the first stage, written sentences are transcribed into gloss-based annotations that represent the structure of the target sign language; in the second, these annotations are used to generate 3D animations via a signing avatar. This paper focuses on the first stage: the automatic generation of gloss-based representations for French Sign Language (LSF) from written French. Beyond the general methodological framework, our main contributions are as follows: (i) the design of a parameterized glossed-based specification language that accurately captures the grammatical properties of LSF; and (ii) the implementation of a method for transcribing written French into gloss-based LSF annotations, using large language models trained on a limited amount of data. Our approach is evaluated using a bilingual corpus collected within the SignToKids project [27]. The results are promising and demonstrate the method's potential for scaling to larger datasets.
| Origine | Fichiers éditeurs autorisés sur une archive ouverte |
|---|---|
| Licence |