Machine Learning applied to Rule-Based Machine Translation

Abstract

Lexical and morphological ambiguities present a serious challenge in rule-based machine translation (RBMT). This chapter describes an approach to resolve morphologically ambiguous verb forms if a rule-based decision is not possible due to parsing or tagging errors. The rule-based core system has a set of rules to decide, based on context information, which verb form should be generated in the target language. However, if the parse tree is not correct, part of the context information might be missing and the rules cannot make a safe decision. In this case, we use a classifier to assign a verb form. We tested the classifier on a set of four texts, increasing the correct verb forms in the translation from 78.68\%, with the purely rule-based disambiguation, to 95.11\% with the hybrid approach.