Exploring the Representation of Word Meanings in Context: A Case Study on Homonymy and Synonymy

This paper presents a multilingual study of word meaning representations in\ncontext. We assess the ability of both static and contextualized models to\nadequately represent different lexical-semantic relations, such as homonymy and\nsynonymy. To do so, we created a new multilingual dataset that allows us to\nperform a controlled evaluation of several factors such as the impact of the\nsurrounding context or the overlap between words, conveying the same or\ndifferent senses. A systematic assessment on four scenarios shows that the best\nmonolingual models based on Transformers can adequately disambiguate homonyms\nin context. However, as they rely heavily on context, these models fail at\nrepresenting words with different senses when occurring in similar sentences.\nExperiments are performed in Galician, Portuguese, English, and Spanish, and\nboth the dataset (with more than 3,000 evaluation items) and new models are\nfreely released with this study.\n