Preliminary study for Conversational Korean-Vietnam Neural Machine Translation.

In this paper, we aim to build a conversational Korean-Vietnamese translator. We aim to develop a conversational translator that can be applied to interpreting using conversational data that encompasses features such as ill-formed sentences, anaphora, omissions, and contextual information commonly employed by real individuals in conversations. To this end, we utilized subtitle data to create a large-scale parallel corpus that reflects the characteristics of conversational data and overcome the problem of lack of data between languages, which is a problem in machine translation. We used the built data as training data for a neural network-based automatic translation model to create a conversational translator, which improved the BLEU score by 3.67 compared to the initial experiment.