Construction and Application of Automatic Evaluation Model for Machine Translation Quality Based on Fuzzy Logic Algorithm

With the widespread application of machine translation in cross-language communication, it is crucial to accurately evaluate its translation quality. Current machine translation evaluation faces problems such as lack of macro-level grasp of micro-similarity evaluation and difficulty for computers to perceive context and understand deep meaning. Based on fuzzy logic algorithm, this paper constructs an automatic evaluation model for machine translation quality. A three-tier hierarchical evaluation framework is established, comprising the target level, criterion level, and indicator level. The Analytic Hierarchy Process (AHP) is employed to calculate the weight vector, while a trapezoidal membership function is utilized to define the membership degrees. The final evaluation vector is derived through a fuzzy synthesis operation. The experiment selects four softwares: Google Translate, Baidu Translate, DeepL, and Youdao Translate. English news is used as the test text, and professional translators and MT researchers use a 5-level Likert scale to score. The results show that the grammatical error rate memberships of Google (A), DeepL (C) and Youdao (D) are 0.12, 0.07 and 0.15, respectively. The automatic evaluation model of machine translation quality can clearly present the performance differences of each software in terms of vocabulary, discourse, grammar and other dimensions, providing a scientific basis for selecting high-quality translation software and improving software quality.