A Domain Adaptation Method for Neural Machine Translation

Xiaohu Tian, Jin Liu, Jiachen Pu, Jin Wang

Abstract: With the globalization and the rapid development of the Internet, machine translation is becoming more widely used in real world applications. However, existing methods are not good enough for domain adaptation translation. Consequently, we may understand the cutting-edge techniques in a field better and faster with the aid of our machine translation. This paper proposes a method of calculating the balance factor based on model fusion algorithm and logarithmic linear interpolation. A neural machine translation technique is used to train a domain adaptation translation model. In our experiments, the BLEU score of the in-domain corpus reaches 43.55, which shows a certain increase when comparing to existing methods.