Bridging the Defined and the Defining: Exploiting Implicit Lexical Semantic Relations in Definition Modeling

Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing, pages 3521–3527, Hong Kong, China, November 3–7, 2019. c⃝2019 Association for Computational Linguistics 3521 Bridging the Defined and the Defining: Exploiting Implicit Lexical Semantic Relations in Definition Modeling Koki Washio∗†1 Satoshi Sekine†2 ∗Department of Language and Information Sciences, The University of Tokyo †RIKEN Center for Advanced Intelligence Project {1kokiwashio@g.ecc, 3kato@boz.c}.u-tokyo.ac.jp 2satoshi.sekine@riken.jp Tsuneaki Kato∗3 Abstract Definition modeling includes acquiring word embeddings from dictionary definitions and generating definitions of words. While the meanings of defining words are important in dictionary definitions, it is crucial to capture the lexical semantic relations between defined words and defining words. However, thus far, the utilization of such relations has not been explored for definition modeling. In this pa- per, we propose definition modeling methods that use lexical semantic relations. To uti- lize implicit semantic relations in definitions, we use unsupervisedly obtained pattern-based word-pair embeddings that represent seman- tic relations of word pairs. Experimental re- sults indicate that our methods improve the performance in learning embeddings from def- initions, as well as definition generation. 1 Introduction Dictionary definitions are rich resources of seman- tic information for both humans and machines. Recent studies on definition modeling are primar- ily divided into two categories. One is definition generation, in which a definition is generated for a target word from its word representation. Defi- nition generation involves analyzing word embed- dings using generated definitions (Noraset et al., 2017) and machine explanations of word mean- ings for human readers (Ni and Wang, 2017; Ishi- watari et al., 2019). The other is learning word embeddings from definitions to obtain semantic- oriented word representations (Tissier et al., 2017; Bosc and Vincent, 2018). Although previous methods for encoding or de- coding definitions using recurrent neural networks (RNNs) yielded promising results in definition modeling (Noraset et al., 2017; Bosc and Vincent, 2018), they did not explicitly utilize lexical se- mantic relations between defined words and defin- Figure 1: Definition of knife from WordNet (Fellbaum, 1998) and lexical semantic relations between the de- fined word and the defining words. ing words. Various lexical relations exist in defi- nitions (Amsler, 1981), as displayed in Figure 1, where the defined word knife exhibits an Is-a rela- tion with the defining words, tool and instrument, Has-a relation with edge, and Used-for relation with cutting. Utilizing structures of definitions about lexical semantic relations facilitates the un- derstanding and generation of definitions. Based on this observation, we propose defini- tion modeling methods that exploit lexical seman- tic relations between defined and defining words. However, lexical semantic relations in definitions are not explicit. To solve this problem, we use un- supervisedly learned word-pair embeddings that represent semantic relations of word pairs based on co-occurring relational patterns in a corpus (Turney, 2005; Washio and Kato, 2018b). Exper- imental results show that our definition modeling methods improve previous models, with respect to both definition generation and the acquisition of word embeddings from …

Content truncated to 500 words