Focus on what matters: Applying Discourse Coherence Theory to Cross Document Coreference

Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 1406–1417 November 7–11, 2021. c⃝2021 Association for Computational Linguistics 1406 Focus on what matters: Applying Discourse Coherence Theory to Cross Document Coreference William Held1, Dan Iter2, Dan Jurafsky2 1Sunshine Products 2Department of Computer Science, Stanford University will@sunshine.com, {daniter,jurafsky}@stanford.edu Abstract Performing event and entity coreference reso- lution across documents vastly increases the number of candidate mentions, making it in- tractable to do the full n2 pairwise compar- isons. Existing approaches simplify by consid- ering coreference only within document clus- ters, but this fails to handle inter-cluster coref- erence, common in many applications. As a result cross-document coreference algorithms are rarely applied to downstream tasks. We draw on an insight from discourse coherence theory: potential coreferences are constrained by the reader’s discourse focus. We model the entities/events in a reader’s focus as a neigh- borhood within a learned latent embedding space which minimizes the distance between mentions and the centroids of their gold coref- erence clusters. We then use these neighbor- hoods to sample only hard negatives to train a fine-grained classifier on mention pairs and their local discourse features. Our approach1 achieves state-of-the-art results for both events and entities on the ECB+, Gun Violence, Foot- ball Coreference, and Cross-Domain Cross- Document Coreference corpora. Furthermore, training on multiple corpora improves aver- age performance across all datasets by 17.2 F1 points, leading to a robust coreference resolu- tion model for use in downstream tasks where link distribution is unknown. 1 Introduction Cross-document coreference resolution of entities and events (CDCR) is an increasingly important problem, as downstream tasks that benefit from coreference annotations — such as question an- swering, information extraction, and summariza- tion — begin interpreting multiple documents si- multaneously. Yet the number of candidate men- tions across documents makes evaluating the full 1Code is available at https://github.com/ Helw150/event_entity_coref_ecb_plus n2 pairwise comparisons intractable (Cremisini and Finlayson, 2020). For single-document coref- erence, the search space is pruned with simple recency-based heuristics, but there is no natural corollary to recency with multiple documents. Most CDCR systems thus instead cluster the documents and perform the full n2 comparisons only within each cluster, disregarding inter-cluster coreference (Lee et al., 2012; Yang et al., 2015; Choubey and Huang, 2017; Barhom et al., 2019; Cattan et al., 2020; Yu et al., 2020; Caciularu et al., 2021). This was effective for the ECB+ dataset, on which most CDCR methods have been evaluated, because ECB+ has lexically distinct topics with almost no inter-cluster coreference. Such document clustering, however, keeps CDCR systems from being generally applicable. Bugert et al. (2020b) shows that inter-cluster coref- erence makes up the majority of coreference in many applications. Cremisini and Finlayson (2020) note that document clustering methods are also un- likely to generalize well to real data where doc- uments lack the significant lexical differences of ECB+ topics. These issues present a major barrier for the general applicability of CDCR. Human readers, by contrast, are able to perform coreference resolution with minimal pairwise …

Content truncated to 500 words