Misinformation in Machine Translation: Error Categories and Levels of Recognition Difficulty

Abstract

Extensive research has been done on typologies of machine translation(MT) errors, but one subset of mistranslation—misinformation—is relatively understudied. Unlike other types of mistranslation, misinformation does not necessarily affect the readability or coherence of the translation, but will inhibit target readers from accessing the accurate information presented in the source text. It is unreasonable to expect post-editors(PE) to devote the equivalent levels of time and effort on the MT pre-translated text as in traditional translation projects, given that PE tasks have relatively lower pay but identical, if not tighter, deadlines. To gain more understanding on the concept of misinformation with the aim to improve the efficiency and accuracy of post-editors’ work, this study analyzed four English to Chinese MT texts to categorize the misinformation instances, observe distribution patterns of error types, and evaluate their levels of recognition difficulty. It was observed that the highest number of misinformation instances fell into the category of polysemy/named-entity errors, attributing to around half of all misinformation instances. The second most common misinformation category is the non-equivalent rhetoric/idiomatic expression. To help post-editors identify the hard-to-recognize misinformation, we propose three different approaches: (1) use interactive MT platforms or CAT tools that can provide alternative translation suggestions; (2) compare MT results generated by multiple MT tools, as the discrepancies in translations can alert post-editors of potential misinformation; (3) compare the original source text with the back translation result of MT to identify non-equivalent rhetorical/idiomatic expressions.