A SANSKRIT BASED KNOWLEDGE REPRESENTATION BUILDER FOR MACHINE TRANSLATION

Abstract

The goal of Natural Language Processing (NLP) is to build computer models of natural language understanding and generation. Machine Translation has been the first area to be pursued with this goal. After the failures of the early machine translation systems, a lot of attention has been given on designing unambiguous representation schemes for the meaning of a natural language text. For machine translation an intermediate language or interlingua is used for the same. Shastric sanskrit has been found to be a natural language that allows paraphrasing a natural language sentence so as to encode the semantics explicitly and without any ambiguity. The present work was aimed at building a Knowledge Representation system based on Sanskrit. The key issue of the system was to determine the feasibility of Sanskrit as the interlingua when one language is a non-Indian language like English. The dissertation describes such a system that builds Knowledge Representation structures similar to Conceptual Graphs. The system has been implemented with Turbo C for the PC - AT compatibles. The concept of Local Word Grouping has been used to determine the Sanskrit relations from English sentences. Karak Charts have been used for providing the mapping from these word groups to karak relations, with the help of prepositions. The system successfully builds Knowledge Representation structures for a wide variety of declarative English sentences. The success of the system indicates that this Sanskrit approach may be extended to other languages too.