Speaker: Matthew Skala
Title: Words in space: applying geometry to language

Abstract:
Natural language processing, for applications like question
answering and machine translation, requires a computer to solve tricky
combinatorial problems in large quantities and as fast as possible.  In
this talk I describe my recent work, with others at the University of
Toronto, on applying high-dimensional geometry to problems arising from
computational linguistics.  Partial order embedding is the starting
point: mapping an arbitrary finite partial order into a subset of some
better-behaved space to speed up logical calculations within the partial
order.  The work on finite partial orders leads naturally to questions
involving more elaborate data structures representing semantic
information.  The talk will be aimed primarily at a computer science
audience, with linguistics background given where needed.