Incorporating Singletons and Mention-based Features in Coreference Resolution via Multi-task Learning for Better Generalization
This work addresses coreference resolution for natural language processing applications, offering incremental improvements in generalization and performance.
The paper tackles the problem of improving coreference resolution by incorporating singleton mentions and entity features via multi-task learning, achieving new state-of-the-art scores on the OntoGUM benchmark with a +2.7 point increase and enhanced robustness on out-of-domain datasets with an average +2.3 point gain.
Previous attempts to incorporate a mention detection step into end-to-end neural coreference resolution for English have been hampered by the lack of singleton mention span data as well as other entity information. This paper presents a coreference model that learns singletons as well as features such as entity type and information status via a multi-task learning-based approach. This approach achieves new state-of-the-art scores on the OntoGUM benchmark (+2.7 points) and increases robustness on multiple out-of-domain datasets (+2.3 points on average), likely due to greater generalizability for mention detection and utilization of more data from singletons when compared to only coreferent mention pair matching.