An Analysis of Error Types in Chinese to English Translation by Google Neural Machine Translation
Yanqi Lu · 2023
Abstract: Due to the rapid development of globalization and digitalization, neural machine translation (NMT) systems have gradually developed into the mainstream technology in the field of machine translation (MT). Even so, MT output cannot meet the end-user's expectation in terms of translation quality. This has prompted many scholars to research on improving the quality of MT output. This study is based on Google Neural Machine Translation (GNMT) application and the Government Work Report in 2022 to compare the GNMT output with the official translation by means of manual error analysis. It summarizes the error types of GNMT in such texts, with 4 first-level error types and 15 second-level error types, and MT errors of such texts are decreasing according to the first-level error types: semantic, syntactic, lexical, and other errors. Thus, even though NMT systems lead the way in MT, it cannot replace human translation. This study also provides some case studies of these error types so as to provide practical reference for post-editing of NMT. Based on the error analysis, four implications for post-editing are summed up. So when doing translation, post-editors need to enhance awareness of context, focus on non-subject-predicate sentence structure, adhere to logical connection, and focus on four-character phrases.