An Empirical Study towards Characterizing Neural Machine Translation Testing Methods

Abstract

Due to the rapid development of deep neural networks, in recent years, machine translation software has been widely adopted in people's daily lives, such as communicating with foreigners or understanding political news from the neighbouring countries, and it is embedded in daily applications such as Twitter and WeChat. The neural machine translation (NMT) model is the core of machine translation software, and it is very challenging to test it as a deep neural network model due to the Inexplicability of neural networks and the complexity of model output. In this paper, we introduce three latest machine translation testing methods and provide a preliminary analysis of their effects.