Neural machine translation evaluation & error analysis in a Spanish-Korean translation
From RBMT to SMT and NMT, the MT field witnessed, first, a conceptual turn —from rule-based to data-base— and now, a technological turn —from MT algorithm to ML algorithm. Now that NMT became a new state of the art, this thesis quested for evaluating its performance in a Spanish-to-Korean translatio...
| Autor: | |
|---|---|
| Tipo de recurso: | tesis doctoral |
| Estado: | Versión publicada |
| Fecha de publicación: | 2019 |
| País: | España |
| Institución: | CBUC, CESCA |
| Repositorio: | TDR. Tesis Doctorales en Red |
| OAI Identifier: | oai:www.tdx.cat:10803/667853 |
| Acceso en línea: | http://hdl.handle.net/10803/667853 |
| Access Level: | acceso abierto |
| Palabra clave: | Neural machine translation MT evaluation Error analysis Spanish-Korean translation Traducció automàtica neuronal Avaluació de traducció automàtica Anàlisi d'errors Traducció de l'espanyol-coreà Traducción automática neuronal Evaluación de traducción automática Análisis de errores Traducción del español-coreano 81 |
| Sumario: | From RBMT to SMT and NMT, the MT field witnessed, first, a conceptual turn —from rule-based to data-base— and now, a technological turn —from MT algorithm to ML algorithm. Now that NMT became a new state of the art, this thesis quested for evaluating its performance in a Spanish-to-Korean translation, which, for the best of our knowledge, was the first attempt in this regard. The results reported that the NMT-based Google Translate (GNMT) had about 78% of reliability. In an experiment with post-editing, the post-editing was 37% more productive in GNMT than translation from scratch. An important finding was obtained from quantitative and qualitative error analysis. It reported that only 6% of the errors detected in the dataset were a syntactic error in such a distant pair like this. The results of this thesis served as a proof of a promising future of NMT in distant pairs. |
|---|