LNN-EL: A Neuro-Symbolic Approach to Short-text Entity Linking

Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing, pages 775–787 August 1–6, 2021. ©2021 Association for Computational Linguistics 775 LNN-EL: A Neuro-Symbolic Approach to Short-text Entity Linking Hang Jiang1∗, Sairam Gurajada2∗, Qiuhao Lu3, Sumit Neelam2, Lucian Popa2, Prithviraj Sen2, Yunyao Li2, Alexander Gray2 1MIT 2IBM Research 3University of Oregon hjian42@mit.edu, {sairam.gurajada, alexander.gray}@ibm.com, luqh@cs.uoregon.edu, sumit.neelam@in.ibm.com, {lpopa,senp,yunyaoli}@us.ibm.com Abstract Entity linking (EL), the task of disambiguat- ing mentions in text by linking them to enti- ties in a knowledge graph, is crucial for text understanding, question answering or conver- sational systems. Entity linking on short text (e.g., single sentence or question) poses partic- ular challenges due to limited context. While prior approaches use either heuristics or black- box neural methods, here we propose LNN- EL, a neuro-symbolic approach that combines the advantages of using interpretable rules based on first-order logic with the performance of neural learning. Even though constrained to using rules, LNN-EL performs competitively against SotA black-box neural approaches, with the added benefits of extensibility and transferability. In particular, we show that we can easily blend existing rule templates given by a human expert, with multiple types of fea- tures (priors, BERT encodings, box embed- dings, etc), and even scores resulting from pre- vious EL methods, thus improving on such methods. For instance, on the LC-QuAD-1.0 dataset, we show more than 4% increase in F1 score over previous SotA. Finally, we show that the inductive bias offered by using logic results in learned rules that transfer well across datasets, even without fine tuning, while main- taining high accuracy. 1 Introduction Entity Linking (EL) is the task of disambiguat- ing textual mentions by linking them to canoni- cal entities provided by a knowledge graph (KG) such as DBpedia, YAGO (Suchanek et al., 2007) or Wikidata (Vrandeˇci´c and Krötzsch, 2014). A large body of existing work deals with EL in the context of longer text (i.e., comprising of multiple sentences) (Bunescu and Pasca, 2006). The general ∗Equal contribution; Author Hang Jiang did this work while interning at IBM. approach is: 1) extract features measuring some degree of similarity between the textual mention and any one of several candidate entities (Mihalcea and Csomai, 2007; Cucerzan, 2007; Ratinov et al., 2011), followed by 2) the disambiguation step, ei- ther heuristics-based (non-learning) (Hoffart et al., 2011; Sakor et al., 2019; Ferragina and Scaiella, 2012) or learning-based (Mihalcea and Csomai, 2007; Cucerzan, 2007; Ratinov et al., 2011; Hof- fart et al., 2012; Ganea and Hofmann, 2017), to link the mention to an actual entity. A particular type of entity linking, focused on short text (i.e., a single sentence or question), has attracted recent attention due to its relevance for downstream applications such as question answer- ing (e.g., (Kapanipathi et al., 2021)) and conversa- tional systems. Short-text EL is particularly chal- lenging because the limited context surrounding mentions results in greater ambiguity (Sakor et al., 2019). To address this challenge, one needs to ex- ploit …

Content truncated to 500 words