I am Yves Peirsman, a post-doctoral scholar at Stanford University and the University of Leuven. My research in lexical semantics brings together my interest in theoretical linguistics and that in computational linguistics. I am a member of the research group QLVL at Leuven and the NLP group at Stanford.
My research focuses on lexical semantics. I mainly work with vector space models, which identify the semantic relatedness between two words on the basis of the contexts in which those words occur. I have applied these vector space models to several tasks, including the automatic translation of words between different languages, and the recognition of semantic variation between two varieties of the same language. It is tasks like those, which require the combination of computational modelling with linguistic knowledge, that spark my interest most.
These are some of the issued I've tackled so far:
Automatic word translation. When two languages share a fair number of words (or even word stems), we can use these shared words in the construction of a bilingual vector space that aids the translation of individual words from one language to the other. I've successfully applied this technique to the language pairs English-Dutch, English-German, Dutch-German, and English-Spanish. (Publications c14 and in preparation)
The recognition of lexical variation. Because vector space models identify relatedness in meaning, they can help us find words that have different meanings in two language varieties. Moreover, they can automatically retrieve the alternative to a word in another language variety. (Publications a3, c11)
Linguistic interpretation and applications of vector space models. Despite the popularity of vector space models in computational linguistics, their semantic characteristics are not yet well-understood. Therefore I've investigated a wide range of vector spaces and their performance in a large number of tasks, like the recognition of synonyms or the prediction of associations. (Publications c4-10, c12-13)
Metonymy recognition. My MSc dissertation at the University of Edinburgh targeted the automatic recognition of metonymical words like Washington says, where the city stands for the politicians that work in that city (probably). (Publications c1-2, g2)
Metonymy in Cognitive Linguistics. My MA dissertation at the University of Leuven focused on metonymy from a cognitive-linguistic perspective. It suggested a new classification of metonymical patterns, based on the notions of contiguity and prototypicality. (Publications a1-2, b1, g1)