LTAG BASED HINDI SENTENCE GENERATION FROM ENGLISH PARSE TREES

Abstract

Lexical tree adjoining grammar (LTAG) is a very effective approach for machine translation of sentences from one language to another where the input language sentence has been analyzed into a parse tree structure using LTAGs. This thesis presents a method for LTAG based hindi sentence generation from english parse trees. It involves the transformation of the words at the leaf level, and then the translation of the phrasal trees. This is followed by restructuring of the sentential parse tree, thus yielding the hindi parse tree. From this the hindi sentence can be extracted by the pre-order traversal of the leaf nodes of the parse tree. The information pertaining to the hindi words and their transformations is -stored in a transfer lexicon. The lexicon is implemented using a structural semantic approach based on paradigm (or model) tables to reduce the redundancy and the access time. The trie structure of the lexicon further reduces the access time. The program for the hindi sentence generator is written in C-language in the IRIX (UNIX) environment on the Tata Elxsi Power Series 3220 (dual RS6000 RISC -processors) system. With slight modifications, the program can be easily ported on to the MS-DOS platform.