Semantics for finite state approximations.
- Speaker: Professor Stephen Pulman (Somerville College, Oxford)
- Host: Ian Pratt-Hartmann
- 9th November 2005 at 14:15 in 1.5
Various methods for producing finite state approximations for context-free or richer grammatical formalisms have been proposed in recent years. These are of interest for various reasons: efficient parsing; compilation into tight language models for commercial speech recognisers, etc. However, the resulting approximations lose the syntactic information needed to provide a compositional semantics for the original grammars. This paper presents a set of semantic transforms to parallel the left-corner transforms described in Mark Johnson's 1998 paper (http://acl.ldc.upenn.edu/P/P98/P98-1101.pdf) which enable a compositional semantics expressed in a typed higher order logic to be propagated through from the original to the transformed grammar, meaning that logical forms can be recovered directly from the finite state approximation.