Language Oriented Parsing Through Morphologically Closed Word Classes in Urdu

To find correctness of the grammatical structure of a natural language sentence, unambiguous parse is the basic requirement. Therefore, Parsing of the source language plays a key role for reliable machine translation. In this paper a language oriented parsing algorithm is presented for Urdu language sentences by initiating tagging only for morphologically closed classes of words like postpositions, conjunctions, verb morphemes, etc. By utilizing linguistics features of these closed classes neighbor words are collected into chunks. The chunks are formed by applying grammar rules in order through ordered context free grammar. Finally, full parsing on chunks is achieved which have much lesser search space. The functional structures are unified throughout the process of chunking and parsing to support the correctness of parsing. It is found that use of closed classes ahead of open classes and chunking process reduces the number of grammar rules and enhances the reliability of final parse.