Computational linguistics has undergone something of a revolution in recent years, as the use of very large corpora and machine learning methods have become part of our standard toolbox. Corpora have both obvious advantages (authenticity, size, coverage of many dialects, domains, genres, registers) and disadvantages (lack of extra-linguistic context, pervasive ambiguity). In this context neuroimaging data can be seen as complementary, as it is typically much more limited in scale (due to the labour and cost of collecting it), but gives a well-controlled and direct, if noisy, snapshot of the mind in action. If recent rapid methodological progress continues, neural data could soon become a second principle empirical pillar for computational work that explores linguistic cognition. In this talk I will try to give an overview of recent work that decodes linguistic representations and processes from recordings of brain activity, with a particular emphasis on the topic we specialise in: lexical semantics. At the end I will also speculate on the broader potential of such methods, and explain why I think computational linguists are singularly well placed to contribute in this area.

If you would like to meet with the speaker, please contact Matthew Crocker.