Textual inference rules for recognizing textual entailment
Speaker: Georgina Dinu
Institution: Saarland University
Abstract:
My talk concerns collections of semantically related phrases acquired from large monolingual corpora based on the Distributional Hypothesis. The Distributional Hypothesis states that words occurring in similar contexts have similar meaning. This idea has been the core of several algorithms developed for building collections of inference rules (an example of such a rule is: X /prevents/ Y entails X /decreases the risk of/ Y).
Much of the recent work focuses on methods of making these rule collections more accurate. Ways of refining them are, for example, determining the context in which the rules hold, the direction of the inference relations or identifying incorrect rules. In my talk I will present preliminary results which are concerned with a complementary aspect of this: using inference rules acquired through these methods for recognizing textual entailment.