Improving Adequacy in a Rule-Based English-to-Igala Automatic Translation System through Word Sense Disambiguation

Abstract

Here, we describe a rule-based machine translation system that translates English sentences to the Igala language. This work was evaluated with respect to adequacy, a measure of the quantity of information existent in the original text that the translated text contains. It indicates whether the output is a correct translation of the original sentence in a sense that the right meaning is communicated. An analysis revealed that adequacy was poor for sentences that contain words that exhibit ambiguity. The errors originated from the fact that correct meanings of ambiguous words were not selected based on the context. A further study of the architecture of the translation system showed it has no built-in module for word sense disambiguation. This accounts for the poor adequacy. A word-sense model was developed and incorporated into the MT system to disambiguate sentences before passing it to the translator.  The model was implemented using JSP, a technology for building and deploying web applications in Java as the front end and MySQL as a backend. The model was tested on 100 previously translated sentences that contain ambiguous words. The level of adequacy increased from 32% to 68.2%.