Paper: Prediction Of Lexicalized Tree Fragments In Text

ACL ID H93-1048
Title Prediction Of Lexicalized Tree Fragments In Text
Venue Human Language Technologies
Session Main Conference
Year 1993

There is a mismatch between the distribution of information in text, and a variety of grammatical formalisms for describing it, including ngrams, context-free grammars, and dependency grammars. Rather than adding probabilities to existing grammars, it is proposed to collect the distributions of flexibly sized partial trees. These can be used to enhance an ngram model, and in analogical parsing. 1. THE PROBLEM WITH PROBABILIZED GRAMMARS For a variety of language processing tasks, it is useful to have a predictive language model, a fact which has recently led to the development probabilistic versions of diverse grammars, including ngram models, context free grammars, various de- pendency grammars, and lexicalized tree grammars. These enterprises share a common problem: there is a mismatch be...