Detection, Disambiguation, Re-ranking: Autoregressive Entity Linking as a Multi-Task Problem

Findings of the Association for Computational Linguistics: ACL 2022, pages 1972 - 1983 May 22-27, 2022 c⃝2022 Association for Computational Linguistics Detection, Disambiguation, Re-ranking: Autoregressive Entity Linking as a Multi-Task Problem Khalil Mrini1, Shaoliang Nie2, Jiatao Gu2, Sinong Wang2, Maziar Sanjabi2, and Hamed Firooz2 1University of California, San Diego, La Jolla, CA 92093 khalil@ucsd.edu 2Meta AI, Menlo Park, CA 94025 {snie, jgu, swang, maziars, hfirooz}@fb.com Abstract We propose an autoregressive entity linking model, that is trained with two auxiliary tasks, and learns to re-rank generated samples at in- ference time. Our proposed novelties address two weaknesses in the literature. First, a re- cent method proposes to learn mention detec- tion and then entity candidate selection, but relies on predefined sets of candidates. We use encoder-decoder autoregressive entity link- ing in order to bypass this need, and propose to train mention detection as an auxiliary task instead. Second, previous work suggests that re-ranking could help correct prediction errors. We add a new, auxiliary task, match prediction, to learn re-ranking. Without the use of a knowl- edge base or candidate sets, our model sets a new state of the art in two benchmark datasets of entity linking: COMETA in the biomedical domain, and AIDA-CoNLL in the news do- main. We show through ablation studies that each of the two auxiliary tasks increases per- formance, and that re-ranking is an important factor to the increase. Finally, our low-resource experimental results suggest that performance on the main task benefits from the knowledge learned by the auxiliary tasks, and not just from the additional training data. 1 Introduction Entity linking (Zhang et al., 2010; Han et al., 2011) is the task of linking entity mentions in a text doc- ument to concepts in a knowledge base. It is a ba- sic building block used in many NLP applications, such as question answering (Yu et al., 2017; Dubey et al., 2018; Shah et al., 2019), word sense disam- biguation (Raganato et al., 2017; Uslu et al., 2018), text classification (Basile et al., 2015; Scharpf et al., 2021), and social media analysis (Liu et al., 2013; Yamada et al., 2015). Early definitions decompose the task of entity linking (EL) into two subtasks: Mention Detection (MD) and Entity Disambiguation (ED). Many sta- tistical and LSTM-based methods propose to cast SOCCER - Japan Get Lucky Win, China In Surprise Defeat. Japan began the defence of their Asian Cup title with a lucky 2-1 win against Syria in a Group C championship match on Friday. But China saw their luck desert them [...] Source Text SOCCER - Japan Get Lucky Win, China national football team In Surprise Defeat. Japan national football team began the defence of their AFC Asian Cup title with a lucky 2-1 win against Syria national footballer team in a Group C championship match on Friday. But China Chinese Super League [...] GENRE (De Cao et al., 2021) SOCCER - Japan national football team Get Lucky Win, China national football team In Surprise Defeat. Japan national footballer …

Content truncated to 500 words