• computational analysis
• language models
• polysemous patterns
• polysemy
• semantic representation

In natural languages, words frequently have multiple but related meanings. This phenomenon is called polysemy. The relations between words’ senses are often metaphorical or metonymic. This thesis focuses on polysemy motivated by metonymy as an universal language mechanism through the collection and the analysis of human data ratings, and the exploration of those polysemic words’ semantic disambiguation in Language Models, showing up the importance of considering the meanings’ possibility to colexify in a strict or loose way in different languages. The project born from the contrast between human ratings and computational driven data in the existing literature.