Towards More Linguistically Constrained Coreference Resolution
TALN (Communication orale)
- Auteurs :
Ailloud Étienne (University of Zurich, Switzerland)
Klenner Manfred (University of Zurich, Switzerland)
- Résumé : We propose a filter model of coreference resolution that is based on the notions of transitivity and linguistic exclusivity. Starting from the general assumption that coreference sets remain coherent throughout a text, our model enforces the checking of some compatibility criteria (filters) between coreference candidates, improving resolution performance. This filtering is achieved at different stages of the workflow of machine-learning-based coreference resolution, including at the standard learning and testing steps, where it may help reduce the computational load and better distribute the actual occurrences to be learned.
- Format PDF