In this paper, the machine translation models were applied to the Thai-English and English-Thai machine translation task. We investigated three models of machine translation on Thai and English sentence pairs. The translation performance of the transformer model is better than that of the recurrent neural network and the traditional statistical machine translation models. We found that the BLEU scores of the transformer model were the highest in both Thai-English (44.22%) and English-Thai (46.48%) translations. Besides, the results were also analysed linguistically. In comparison with the three models, the errors about detailed description and wrong word ordering were mostly found in the SMT model, whereas wrong word choice and missing words were mostly found in the RNNs model. Although the transformer model could perform much better than others, three error categories - under-translation, over-translation, and incorrect lexical choice-were also found.
展开▼
