Entity Disambiguation with Entity Definitions

Proceedings of the 17th Conference of the European Chapter of the Association for Computational Linguistics, pages 1297–1303 May 2-6, 2023 ©2023 Association for Computational Linguistics Entity Disambiguation with Entity Definitions Luigi Procopio1,2∗ Simone Conia1 Edoardo Barba1 Roberto Navigli1 1 Sapienza NLP Group, Sapienza University of Rome 2 Sunglasses.ai luigi.procopio@sunglasses.ai conia@di.uniroma1.it {barba,navigli}@diag.uniroma1.it Abstract Local models have recently attained astound- ing performances in Entity Disambiguation (ED), with generative and extractive formu- lations being the most promising research di- rections. However, previous works have so far limited their studies to using, as the tex- tual representation of each candidate, only its Wikipedia title. Although certainly effective, this strategy presents a few critical issues, es- pecially when titles are not sufficiently infor- mative or distinguishable from one another. In this paper, we address this limitation and in- vestigate the extent to which more expressive textual representations can mitigate it. We eval- uate our approach thoroughly against standard benchmarks in ED and find extractive formu- lations to be particularly well-suited to such representations. We report a new state of the art on 2 out of the 6 benchmarks we consider and strongly improve the generalization capa- bility over unseen patterns. We release our code, data and model checkpoints at https: //github.com/SapienzaNLP/extend. 1 Introduction Being able to pair a mention in a given text with its correct entity out of a set of candidates is a crucial problem in Natural Language Processing (NLP), referred to as Entity Disambiguation (Bunescu and Pa¸sca, 2006, ED). Indeed, since ED enables the identification of the actors involved in human lan- guage, it is often considered a necessary building block for a wide range of downstream applications, including Information Extraction (Ji and Grishman, 2011; Guo et al., 2013; Fatahi Bayat et al., 2022), Question Answering (Yin et al., 2016) and Seman- tic Parsing (Bevilacqua et al., 2021; Procopio et al., 2021). ED generally occurs as the last step in an Entity Linking pipeline (Broscheit, 2019), pre- ceded by Mention Detection and Candidate Gen- ∗ Work carried out while at the Sapienza University of Rome. eration, and its approaches have traditionally been divided into two groups, depending on whether co-occurring mentions are disambiguated indepen- dently (local methods; Shahbazi et al. (2019); Wu et al. (2020); Tedeschi et al. (2021)), or not (global methods; Hoffart et al. (2011); Moro et al. (2014); Yamada et al. (2016); Yang et al. (2018)). Despite the limiting operational hypothesis of in- dependence between co-occurring mentions, local methods have nowadays achieved performances that are either on par or above those attained by their global counterparts, mainly thanks to the ad- vent of large pre-trained language models. In par- ticular, among these methods, generative (De Cao et al., 2021) and extractive (Barba et al., 2022) formulations are arguably the most promising di- rections, having resulted in large performance im- provements across multiple benchmarks. Regard- less of their modeling differences, the key idea behind these methods is to part away from the pre- vious classification-based approaches and, instead, adopt formulations …

Content truncated to 500 words