The translation of Japanese literature requires not only linguistic accuracy but also a deep understanding of cultural and contextual nuances, making translation ability a complex task. Traditional evaluation methods often rely on manual scoring, which is subjective, time-consuming, and prone to inconsistency. Existing automated approaches, while improving efficiency, often struggle with semantic alignment, contextual interpretation, and the nuanced literary style inherent in Japanese texts. To address these challenges, this research proposes the Japanese Literature Translation Intelligent Evaluation System (JLT-IES), a deep learning-based framework designed to assess translation quality comprehensively. The paper utilizes Recurrent Neural Networks (RNNs) to capture semantic and contextual relationships in translated texts. Additionally, attention mechanisms are employed to enhance the detection of stylistic fidelity and cultural nuance. The proposed JLT-IES enables automated, consistent, and high-accuracy Evaluation of translation submissions, providing real-time feedback to learners and educators. Experimental results, evaluated on a Kaggle English–Japanese parallel corpus with human-annotated benchmarks, demonstrate that the system achieves superior performance in semantic alignment, contextual understanding, and stylistic assessment compared to existing methods, significantly reducing subjectivity and evaluation time. Performance metrics are reported as mean ± standard deviation across multiple test runs to ensure reliability: semantic alignment: 90 ± 1.3%; contextual understanding: 85 ± 2%; stylistic fidelity: 93 ± 1.8%; cultural nuance detection: 95 ± 1.2%; and translation consistency: 88.9 ± 1.5%. These findings highlight the potential of deep learning frameworks in advancing intelligent assessment systems for literary translation education.
Haiyan Wei (Sun,) studied this question.