FII_CROSS at SemEval-2021 Task 2: Multilingual and Cross-lingual Word-in-Context Disambiguation

Proceedings of the 15th International Workshop on Semantic Evaluation (SemEval-2021), pages 787–792 Bangkok, Thailand (online), August 5–6, 2021. ©2021 Association for Computational Linguistics 787 Abstract This paper presents a word-in-context disambiguation system. The task focuses on capturing the polysemous nature of words in a multilingual and cross-lingual setting, without considering a strict inventory of word meanings. The system applies Natural Language Processing algorithms on datasets from SemEval 2021 Task 2, being able to identify the meaning of words for the languages Arabic, Chinese, English, French and Russian, without making use of any additional mono- or multilingual resources. 1 Introduction The computational task of disambiguating a word in the context of its sentence is still a very challenging topic facing natural language processing (NLP). In this study, we refer to word meaning that requires a multidisciplinary approach for its detection. From sense-based and contextualized embeddings, all tries are aimed at providing an understanding of words in context. We notice that evaluating such approaches is not easy. For instance, traditional Word Sense Disambiguation (WSD) fails to test latent representations unless these are linked to explicit sense inventories such as WordNet (Matusevych, 2016) or BabelNet (Navigli and Pozetto, 2012; Luan et al., 2020). To resolve the problem of disambiguation for both lingual dimensions, we tried to use a combination of well-known algorithms to provide an optimal system. The legitimate research questions this paper intend to answer: Is Word-in-Context Disambiguation a barrier for NLP techniques? The approach we propose in this paper investigates two models of cross-lingual word embeddings, comparing them to the shared- translation effect and the cross-lingual coactivation effects of false and true friends (cognates) found in human language. We find that the similarity structure of the cross-lingual word embeddings space yields the same effects as the human bilingual lexica (Merlo and Rodriguex, 2019). Research on bilingual lexica has uncovered fascinating interactions between the L1 (native language) and L2 (second language) lexica showing that both production and comprehension coactivate lexical items in both languages, indicating that bilinguals store lexical representations from their native and their second language in the same space. The rest of the paper is organized as follows: section 2 describes the literature related to sense disambiguation, section 3 presents the dataset and method of this study, section 4 resumes the results of the conducted experiments, followed by section 5 with the conclusions and discussions about how to increase the accuracy. 2 Background This topic has attracted significant attention in recent years, evidenced by increasing number of workshops (e.g., SemEval-2013 Task 10: Cross- lingual world Sense Disambiguation - CLWSD). Participating in such competitions is especially attractive since teams have thus access to labeled data. For binary tasks, as the case of this competition, there are many computational models to be used in detecting the right word sense. FII_CROSS at SemEval-2021 Task 2: Multilingual and Cross-lingual Word-in-Context Disambiguation Ciprian Bodnar1, Andrada Tapuc1, Cosmin Pintilie1, Daniela Gîfu1,2, Diana Trandabăț1,3 1Faculty of Computer Science, “Alexandru Ioan Cuza” University of Iasi 2Institute of Computer …

Content truncated to 500 words