Joint Coreference Resolution and Named-Entity Linking with Multi-Pass Sieves

Proceedings of the 2013 Conference on Empirical Methods in Natural Language Processing, pages 289–299, Seattle, Washington, USA, 18-21 October 2013. c⃝2013 Association for Computational Linguistics Joint Coreference Resolution and Named-Entity Linking with Multi-pass Sieves Hannaneh Hajishirzi Leila Zilles Daniel S. Weld Luke Zettlemoyer Department of Computer Science and Electrical Engineering University of Washington {hannaneh,lzilles,lsz,weld}@cs.washington.edu Abstract Many errors in coreference resolution come from semantic mismatches due to inadequate world knowledge. Errors in named-entity linking (NEL), on the other hand, are of- ten caused by superficial modeling of entity context. This paper demonstrates that these two tasks are complementary. We introduce NECO, a new model for named entity linking and coreference resolution, which solves both problems jointly, reducing the errors made on each. NECO extends the Stanford determinis- tic coreference system by automatically link- ing mentions to Wikipedia and introducing new NEL-informed mention-merging sieves. Linking improves mention-detection and en- ables new semantic attributes to be incorpo- rated from Freebase, while coreference pro- vides better context modeling by propagat- ing named-entity links within mention clus- ters. Experiments show consistent improve- ments across a number of datasets and ex- perimental conditions, including over 11% re- duction in MUC coreference error and nearly 21% reduction in F1 NEL error on ACE 2004 newswire data. 1 Introduction Coreference resolution and named-entity linking are closely related problems, but have been largely stud- ied in isolation. This paper demonstrates that they are complementary by introducing a simple joint model that improves performance on both tasks. Coreference resolution is the task of determining when two textual mentions name the same individ- [Michael Eisner]1 and [Donald Tsang]2 announced the grand opening of [[Hong Kong]3 Disneyland]4 yester- day. [Eisner]1 thanked [the President]2 and welcomed [fans]5 to [the park]4. Figure 1: A text passage illustrating interactions between coreference resolution and NEL. ual. The biggest challenge in coreference resolu- tion — accounting for 42% of errors in the state- of-the-art Stanford system — is the inability to rea- son effectively about background semantic knowl- edge (Lee et al., 2013). For example, consider the sentence in Figure 1. “President” refers to “Donald Tsang” and “the park” refers to “Hong Kong Dis- neyland,” but automated algorithms typically lack the background knowledge to draw such inferences. Incorporating knowledge is challenging, and many efforts to do so have actually hurt performance, e.g. (Lee et al., 2011; Durrett and Klein, 2013). Named-entity linking (NEL) is the task of match- ing textual mentions to corresponding entities in a knowledge base, such as Wikipedia or Freebase. Such links provide rich sources of semantic knowl- edge about entity attributes — Freebase includes president as Tsang’s title and Disneyland as hav- ing the attribute park. But NEL is itself a chal- lenging problem, and finding the correct link re- quires disambiguating based on the mention string and often non-local contextual features. For exam- ple, “Michael Eisner” is relatively unambiguous but the isolated mention “Eisner” is more challenging. However, these mentions could be clustered with a coreference model, allowing for improved NEL through …

Content truncated to 500 words