Exploiting structural similarities of Philippine languages for a multilingual machine translation system
Abstract/Summary
A multilingual machine translation system is designed for Tagalog, Cebuano and English. It exploits structural similarities of the Philippine languages, Tagalog and Cebuano, and handles the free word order languages. It has two modules: the analyzer and synthesizer. Analyzer parses the input sentence and converts it to its feature structure representation based on the rules and lexicon. The synthesizer then checks for structural similarities. If no structural similarities are found as reflected in the representation produced by the analyzer, the synthesizer converts the source representation to a target representation otherwise, it takes the source representation as input and generates target strings as a means of translation short cut.
The system was inputted with test sentences and evaluated for translation accuracy. Free word order sentences of Tagalog were inputted and checked if these sentences had been accepted syntactically and translated correctly to Cebuano and English. Results showed that the system accepted and translated 53 out of the 55 (or 96%) sentences.
