SynET: Synonym Expansion using Transitivity

Findings of the Association for Computational Linguistics: EMNLP 2020, pages 1961–1970 November 16 - 20, 2020. c⃝2020 Association for Computational Linguistics 1961 SYNET: Synonym Expansion using Transitivity Jiale Yu, Yongliang Shen, Xinyin Ma, Chenghao Jia, Chen Chen, Weiming Lu∗ College of Computer Science and Technology, Zhejiang University, China {27121115, syl, maxinyin, chjia, chen double, luwm}@zju.edu.cn Abstract In this paper, we study a new task of syn- onym expansion using transitivity, and pro- pose a novel approach named SYNET, which considers both the contexts of two given syn- onym pairs. It introduces an auxiliary task to reduce the impact of noisy sentences, and proposes a Multi-Perspective Entity Matching Network to match entities from multiple per- spectives. Extensive experiments on a real- world dataset show the effectiveness of our ap- proach. 1 Introduction Synonym discovery has become an important task, which can benefit many downstream applications, such as web search (Cheng et al., 2012), question answering (Zhou et al., 2013), knowledge graph construction (Boteanu et al., 2019), clinical text analysis (Wang et al., 2019b), and etc. One straightforward approach to obtain syn- onyms is from public knowledge bases, such as WordNet (Fellbaum, 2000) and DBpedia (Lehman- n et al., 2015). For example, WordNet groups terms into synsets, and DBpedia uses Redirects to URIs to indicate synonyms. However, these synonyms are constructed manually, which makes the coverage rather limited. Two types of approaches are widely exploited to discover synonyms automatically from text cor- pora, including the distributional based approach- es (Wang et al., 2019a,b; Fei et al., 2019) and the pattern based approaches (Nguyen et al., 2017). The distributional based approaches assume that if two terms appear in similar contexts, they are like- ly to be synonyms. For example, “USA” and “the United States” are often mentioned in similar con- texts, so they both refer to the same country. The pattern based approaches lay emphasis on the local ∗Corresponding author contexts, such as “commonly known as”. However, they both have some limitations. The distributional based approaches suffer from low precision, while the pattern based approaches suffer from low recal- l. In order to address these limitations, DPE (Qu et al., 2017) integrated these two approaches for synonym discovery. Intuitively, people believe that synonyms possess transitivity, that is (mi, synonym, mb) ∧ (mb, syn- onym, mj)→(mi, synonym, mj), where mi, mb and mj are three different mentions, and mb is the bridge mention of two synonym pairs (mi, mb) and (mb, mj). This transitivity can be used for syn- onym discovery directly from existing synonyms. For example, the United States of America and the United States are synonyms, the United States and U.S. are synonyms, so the United States of America and U.S. should also be synonyms. Distiller (Ali et al., 2019) even designed loss functions based on the synonym transitivity properties. Baidu Baike1 and Wikidata2 both use “Also known as” to indicate synonyms, as shown in Fig- ure 1(a) and Figure 1(b). Therefore, we can ex- tract synonym pairs such as ( 荷花 …

Content truncated to 500 words