14 February 2008
Applying Robust Semantics
Room 308 (Top Floor) Maths & Computing Dept
12:30pm - 13:45pm
Prof Ann Copestake - University of Cambridge
Semantic representations for natural language ideally provide an abstraction which allows generalisations to be made across very different surface forms and different syntactic formalisms. But deriving conventional semantic representations requires more information than can be extracted from morphology and syntax, even by deep grammars. The standard way of dealing with this is to use underspecified semantic representations, which can, in principle, be monotonically enriched to provide deeper semantics which inference engines can work with. However, researchers often work directly with the underspecified representations, using a variety of operations to process them. I will discuss such operations in the context of the (Robust) Minimal Recursion Semantics approach and give an overview of some of the uses to which they have been put. In particular, I will discuss the relationship between robust semantic matching and logical entailment in the context of question answering.
Save to your Calendar