Entity Linking via Explicit Mention-Mention Coreference Modeling

Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 4644 - 4658 July 10-15, 2022 ©2022 Association for Computational Linguistics Entity Linking via Explicit Mention-Mention Coreference Modeling Dhruv Agarwal, Rico Angell, Nicholas Monath∗, Andrew McCallum Manning College of Information and Computer Sciences University of Massachusetts Amherst {dagarwal,rangell,nmonath,mccallum}@cs.umass.edu Abstract Learning representations of entity mentions is a core component of modern entity linking sys- tems for both candidate generation and making linking predictions. In this paper1, we present and empirically analyze a novel training ap- proach for learning mention and entity repre- sentations that is based on building minimum spanning arborescences (i.e., directed spanning trees) over mentions and entities across docu- ments to explicitly model mention coreference relationships. We demonstrate the efficacy of our approach by showing significant improve- ments in both candidate generation recall and linking accuracy on the Zero-Shot Entity Link- ing dataset and MedMentions, the largest pub- licly available biomedical dataset. In addition, we show that our improvements in candidate generation yield higher quality re-ranking mod- els downstream, setting a new SOTA result in linking accuracy on MedMentions. Finally, we demonstrate that our improved mention repre- sentations are also effective for the discovery of new entities via cross-document coreference. 1 Introduction Natural language corpora, such as biomedical re- search papers (Leaman and Lu, 2016), news articles (Milne and Witten, 2008; Hoffart et al., 2011), and, more generally, web page text (Gabrilovich et al., 2013; Lazic et al., 2015a), often contain ambigu- ous mentions of entities. Resolving this ambiguity requires mentions to either be linked to a knowl- edge base (KB) of entities or discovered as a new KB concept if no suitable entry exists. Grounded entity mentions are beneficial for tasks such as question-answering (Das et al., 2019), semantic search (Leaman and Lu, 2016), recommendation ranking (Noia et al., 2016), and KB construction (Ling et al., 2015). The task is made particularly challenging in zero-shot settings, where not every ∗Now at Google. 1Code, datasets, and models are available at https:// github.com/dhdhagar/arboEL. entity has labeled training data (Lin et al., 2017; Logeswaran et al., 2019). In such settings, a com- mon approach is to make use of entity descriptions, types, and aliases to form entity representations, which can then be used for making predictions. Learned vector representations of entity men- tions are an integral part of modern linking systems (Gillick et al., 2019; Wu et al., 2020, inter alia). These representations are used for (a) retrieving a short-list of entity candidates for a mention to use with a re-ranker (Wu et al., 2020), (b) making linking predictions directly (Zhang et al., 2021; Liu et al., 2020; Sung et al., 2020), and (c) performing coreference by clustering mentions to form entities (Logan IV et al., 2020). In this work, we present a new objective and training procedure for learning mention and entity representations that explicitly model mention coref- erence relationships. Our proposed method uses a supervised clustering training objective based …

Content truncated to 500 words