AI Translation Quality Evaluation of Attributive Clauses Based on Faithfulness, Expressiveness and Elegance Principle

Abstract

The widespread use of AI translation has made AI translation quality assessment an increasingly important topic, but related researches need deepening. There exist no sufficient studies comparing human translation and AI translation through statistical analysis. This paper compares human translation, Google translation and Youdao translation of attributive clauses. Data analysis of the online questionnaire shows that the quality of human translation is higher than that of AI translation. The quality of Google English-Chinese translation is higher than that of Youdao. Qualitative analysis is under the guidance of “Faithfulness, Expressiveness and Elegance” principle and the BCC corpus. The qualitative analysis results verifies the results above. Compared with Youdao, Google has better control over the translation methods of attributive clauses, even closer to the expression of human translation. Preposition translation, substitution of pronouns for antecedents, repeated antecedents, additional translation methods are used in line with the original text. However, Google translation is not perfect in some aspects and may neglect some Chinese expression habits. Occasional mistranslation of words or neglect of some Chinese idiomatic collocations are found. Sometimes nouns cannot be flexibly translated into verbs according to Chinese language rules in Google translation.