Phrase-Level Grouping for Lexical Gap Resolution in Korean-Vietnamese SMT

Abstract

A lexical gap easily leads to word alignment errors, which impairs a translation quality. This paper proposes some simple ideas to resolve the difficulty of handling the lexical gap. In morphologically rich languages, a predicate has a complex structure consisting of many morphemes, so we mainly address the issue of how to group the component morphemes by employing morpho-syntactic filters and statistical information from the SMT phrase table. In addition, we abstract grouping results depending on a lexical choice of the target side to enhance translation probabilities. In the experiment, we not only investigate how each method has an effect on Korean-to-Vietnamese SMT, but also show a promising improvement of BLEU score.