Zero-shot Word Sense Disambiguation using Sense Definition Embeddings

Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 5670–5681 Florence, Italy, July 28 - August 2, 2019. c⃝2019 Association for Computational Linguistics 5670 Zero-shot Word Sense Disambiguation using Sense Definition Embeddings Sawan Kumar1 Sharmistha Jat1 Karan Saxena2,∗ Partha Talukdar1 1 Indian Institute of Science, Bangalore 2 Carnegie Mellon University, Pittsburgh {sawankumar,sharmisthaj,ppt}@iisc.ac.in, karansax@cs.cmu.edu Abstract Word Sense Disambiguation (WSD) is a long- standing but open problem in Natural Lan- guage Processing (NLP). WSD corpora are typically small in size, owing to an expensive annotation process. Current supervised WSD methods treat senses as discrete labels and also resort to predicting the Most-Frequent- Sense (MFS) for words unseen during train- ing. This leads to poor performance on rare and unseen senses. To overcome this chal- lenge, we propose Extended WSD Incorpo- rating Sense Embeddings (EWISE), a super- vised model to perform WSD by predicting over a continuous sense embedding space as opposed to a discrete label space. This allows EWISE to generalize over both seen and un- seen senses, thus achieving generalized zero- shot learning. To obtain target sense em- beddings, EWISE utilizes sense definitions. EWISE learns a novel sentence encoder for sense definitions by using WordNet relations and also ConvE, a recently proposed knowl- edge graph embedding method. We also com- pare EWISE against other sentence encoders pretrained on large corpora to generate defini- tion embeddings. EWISE achieves new state- of-the-art WSD performance. 1 Introduction Word Sense Disambiguation (WSD) is an impor- tant task in Natural Language Processing (NLP) (Navigli, 2009). The task is to associate a word in text to its correct sense, where the set of possi- ble senses for the word is assumed to be known a priori. Consider the noun “tie” and the following examples of its usage (Miller, 1995). • “he wore a vest and tie” • “their record was 3 wins, 6 losses and a tie” ∗ Work done as a Research Assistant at Indian Institute of Science, Bangalore. It is clear that the implied sense of the word “tie” is very different in the two cases. The word is associ- ated with “neckwear consisting of a long narrow piece of material” in the first example, and with “the finish of a contest in which the winner is un- decided” in the second. The goal of WSD is to predict the right sense, given a word and its con- text. WSD has been shown to be useful for popu- lar NLP tasks such as machine translation (Neale et al., 2016; Pu et al., 2018), information extrac- tion (Zhong and Ng, 2012; Delli Bovi et al., 2015) and question answering (Ramakrishnan et al., 2003). The task of WSD can also be viewed as an intrinsic evaluation benchmark for the semantics learned by sentence comprehension models. WSD remains an open problem despite a long history of research. In this work, we study the all-words WSD task, where the goal is to disambiguate all ambiguous words in a corpus. Supervised (Zhong and Ng, 2010; Iacobacci …

Content truncated to 500 words