Machine translated letters of credit of real-life trade exchange from English to Chinese: a qualitative approach to the evaluation of machine translation

Abstract

International trade documents are very important in trade transactions. Letters of credit in international trade are one of the most widely used transaction methods. Many letters of credit require translation from one language to another, which has contributed to a high demand for machine translation (MT) by many companies mainly due to the low cost and fast speed in comparison with human translation.

However, because of its special stylistic features, letter of credit are yet to be verified whether MT can perform the task well or well enough. This study attempts to examine the performance of neural network machine translation (NMT) and to identify which aspects it should be improved. More specifically, Meta Translate, a newly-developed neural network machine translation, was applied in this study to examine whether the NMT translation techniques were a good solution for the specialized type of translation task. All the data, which refers to letters of credit in this thesis, were collected from real-life trade transactions in trade companies.

After analysing and comparing two translated versions, an analysis based on skopos theory was conducted. The results show that though Meta Translate can perform well for most of the contents, it still has its shortcomings in terms of certain details. For example, it lacks technicality in terms of formal language and its inability to deal with date expressions is also a weak point. As part of the result of the study, the characteristics of language use in letters of credit were identified and summarized. It is believed that these results can help international trade practitioners to better understand the requirements for machines as well as for humans in the translation of letters of credit, with the aim to improve working efficiency in the businesses of international trade.