Ranking paraphrases in context
Speaker: Georgiana Dinu
Abstract:
In this talk I will present a vector space model that supports the computation of context-sensitive vector representations for words. More precisely, this model allows us to combine the meaning representation of a verb with the meaning representations of its subject and object to obtain a vector representation for the verb in context; this vector representation helps distinguishing between paraphrases of the verb associated to the correct meaning and paraphrases associated to other meanings. An evaluation on the SemEval 2007 Lexical Substitution task shows promising results.
This is joint work with Stefan Thater and Manfred Pinkal.
