Common Types of Errors in Generative Artificial Intelligence and Countermeasures: A Case Study of E-C Translation of Chemical Text

Abstract

With the qualitative analysis method, the paper uses three types of generative AI such as ChatGPT, Claude2, and ERNIE Bot to translate chemical text, aiming to conclude common types of errors committed by generative AI during E-C translation of chemical text and provide some countermeasures. The findings show that common types of errors include five aspects: terminological aspect, lexical aspect, syntactic aspect, discourse aspect, and format aspect. Based on these types of errors, the paper proposes the following countermeasures: (1) Translators can obtain accurate translation of chemical terminology by utilizing the “Method of Suspicion, Searching, Determination, and Expression” and multiple methods. In addition, they can also make use of the terminology database in computer-assisted translation tools like Trados to ensure consistency in the translation of terminology; (2) During the post-editing process, translators can adopt the Communicative Translation Principle to conduct their post-editing. This involves not only modifying content-related errors produced by generative AI but also addressing format issues, with the overarching aim of producing translations that are well-adapted to target readers; (3) For translations that have been modified in the post-editing process, translators can establish translation memory to promptly store corpus, thereby preventing the unnecessary re-translation or post-editing of identical or similar content.