Transformational Reanalysis of Underspecified Parse Trees
Erik Oltmans
Department of Computer Science
University of Twente, The Netherlands
oltmans@cs.utwente.nl
In this talk I will discuss the second module of a two-stage
NLP-system that is part of the indexing process of a large-scale
information retrieval project. The first stage of the NLP-system
consists of a parser that produces shallow analyses in cases of
extra-grammatical input. One of the objectives of the second
stage is to reanalyze these underspecified parse trees in order
to generate appropriate trees that can be used in the remainder
of the indexing process. I will discuss the transformational
rules of the second stage and I will demonstrate their contribution
to indexing.