Lexical Resource Mapping via Translations

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 7147–7154 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 7147 Lexical Resource Mapping via Translations Hongchang Bao, Bradley Hauer, Grzegorz Kondrak Alberta Machine Intelligence Institute, Department of Computing Science University of Alberta, Edmonton, Canada {hongchan, bmhauer, gkondrak}@ualberta.ca Abstract Aligning lexical resources that associate words with concepts in multiple languages increases the total amount of semantic information that can be leveraged for various NLP tasks. We present a translation-based approach to mapping concepts across diverse resources. Our methods depend only on multilingual lexicalization information. When applied to align WordNet/BabelNet to CLICS and OmegaWiki, our methods achieve state-of-the-art accuracy, without any dependence on other sources of semantic knowledge. Since each word-concept pair corresponds to a unique sense of the word, we also demonstrate that the mapping task can be framed as word sense disambiguation. To facilitate future work, we release a set of high-precision WordNet-CLICS alignments, produced by combining three different mapping methods. Keywords: lexical resources, WordNet, concept mapping, lexical translations, multilingual resource 1. Introduction A lexical resource associates words in one or more lan- guages with concepts they can express. Each word- concept pair corresponds to a unique sense of the word. For example, the word plant in Figure 1 has distinct senses corresponding to the industrial and vegetation concepts it can lexicalize. Lexical resources vary in how they are constructed and how they represent con- cepts and senses, which makes it difficult to combine information from multiple resources. The task that we address in this paper is mapping (aligning) concepts or senses across lexical resources. Given a concept in one of the resources, such a mapping allows us to identify a matching concept in the other resource. Aligning concepts between lexical resources facilitates several tasks. First, combining information from mul- tiple resources increases coverage of words and lan- guages. For example, Open Multilingual WordNet (Bond and Foster, 2013) contains the translations of a large number of senses in all the languages covered by the aligned multilingual resources from which it was constructed. Second, in addition to word senses, lexical resources may contain other types of information, such as relations between concepts, glosses, and usage ex- amples. By mapping the senses in one resource to their equivalents in another, information about a given con- cept can be retrieved from both resources, increasing the total knowledge available about each concept. Fi- nally, lexical resources are important for various tasks in natural language processing (NLP). Inter-resource concept mapping has been shown to yield performance improvements compared to using resources in isolation (Ponzetto and Navigli, 2010). The task of concept mapping has been addressed in prior work with a wide variety of resources, techniques, and applications. Numerous works consider map- ping between Princeton WordNet and other resources, including OmegaWiki, Wiktionary, and Wikipedia. Figure 1: An overview of lexical resource mapping task Similarity-based methods (Meyer and Gurevych, 2011) define similarity measures on pairs of concepts across the …

Content truncated to 500 words