2017/08/01 by Nafise Sadat Moosavi, Michael Strube, Moosavi, Nafise Sadat +1
Computer Science · #Computation and Language (cs.CL) #Explainable Artificial Intelligence (XAI) #FOS: Computer and information sciences #Natural Language Processing Techniques #Neural Networks and Applications #Topic Modeling
paper · pdf · doi:10.48550/arxiv.1708.00160
openalex publication_date 2017/08/01 · openalex created_date 2022/08/19 · openalex updated_date 2026/07/28
Coreference resolution is an intermediate step for text understanding. It is\nused in tasks and domains for which we do not necessarily have coreference\nannotated corpora. Therefore, generalization is of special importance for\ncoreference resolution. However, while recent coreference resolvers have\nnotable improvements on the CoNLL dataset, they struggle to generalize properly\nto new domains or datasets. In this paper, we investigate the role of\nlinguistic features in building more generalizable coreference resolvers. We\nshow that generalization improves only slightly by merely using a set of\nadditional linguistic features. However, employing features and subsets of\ntheir values that are informative for coreference resolution, considerably\nimproves generalization. Thanks to better generalization, our system achieves\nstate-of-the-art results in out-of-domain evaluations, e.g., on WikiCoref, our\nsystem, which is trained on CoNLL, achieves on-par performance with a system\ndesigned for this dataset.\n