AN LTAG PARSER FOR KERNEL ENGLISH SENTENCES BASED ON THE PHRASAL APPROACH

Abstract

This dissertation discusses the theory, design and implementation of a Parser, using Lexical Tree Adjoining Grammar (LTAG), for Kernel English sentences. Parsing, in general, for any natural language requires a coherent combination of many different faculties, viz., Grammar formalisms, algorithm and data structures, formalism for representing world knowledge, etc. A parser with LTAG's, requires various finite structures i.e. Trees. There are many trees, viz. sentential trees, phrasal trees and trees associated with each part of speech and recursive trees. There are one or more operations for composing the structures these are substitution and adjoining. The parser intelligently decides which trees to pick for substitution and adjoining from the context and features of words and phrases. The parser separates out the various phrases in sentences and then interconnects the phrases into the final sentential tree. (The various modules of a parser are Lexicon, morphological analyzer, phrase grouper and a core parser). The phrasal trees are constructed using top down approach, while sentential trees are constructed using bottom up approach. This Parser is written in C and runs on a UNIX platform, but with slight modifications it can be easily ported to a PC XT/AT MS DOS platform.