20 February – Pia Sommerauer

Polysemy (and metaphorical mappings) in computational models of meaning representation

In this talk, I am going to introduce computational approaches to representing word meaning and, in particular, polysemy, which is one of the core aspects of my PhD project. Polysemy is a semantic phenomenon that affects a vast number of problems in Natural Language Processing (NLP), including the automatic identification and interpretation of metaphors. In this (probably interactive) talk, I aim at introducing how different computational approaches to representing word meaning deal with polysemy. In particular, I will present the way distributional models (i.e. models that are purely based on words as they appear in linguistic corpora) represent words and how they are affected by polysemy (and, possibly, metaphorical mappings between domains). If there is time, I might show a demo of automatic word sense disambiguation based on a distributional model of word meaning with some examples. Perhaps we can have a discussion about the strengths and weaknesses of the approaches.