Section: New Results

Coreference resolution

Participants : Loïc Grobol, Éric Villemonte de La Clergerie.

In 2019 we have resumed our work on coreference resolution for French with the release in [25] of the first end-to-end automatic coreference resolution system for spoken French by adapting state-of-the art neural network system to the case of noisy non-standard inputs.

This first release uses no external knowledge beyond pretrained non-contextual word embeddings, making it suitable for applications to languages with less pre-existing resources. We also investigated the integration of further knowledge, both in the form of contextual embedding techniques such as CamemBERT and syntactic parsers developped at ALMAnaCH (works to be published in 2020).