Semantic variation and translation issues in positively evaluative Korean homographs

Abstract

In order to develop one for a morphologically difficult language like Korean, the paper investigates the problem of homograph disambiguation in NMT. Homographs – words with a common form but several meanings and context differences – cause significant issues for NMT systems because they often assign meanings at random, leading to translation errors. This work introduces UTagger, a tool that uses the Standard Korean Language Dictionary (SKLD) to provide homographs distinct sense-codes in order to resolve homograph ambiguity. These sense-codes are then integrated into the parallel corpus, where each indexed homograph is treated as an individual word by the NMT system, minimizing ambiguity-related translation errors. The addition of UTagger improves the quality performances of the translation systems with BLEU, TER, and DLRATIO measures, according to experiments conducted on Korean-English and Korean-Vietnamese translation pairings. The outcome shows that for languages with complex morphology, UTagger enhances translation accuracy and dependability. The study concludes with a discussion of extending the application of UTagger to a wider range of homographs, particularly favourable evaluations, and outlines future work to further enhance UTagger, including enhancing its capacity to adapt to various languages and dialects.