A Sentiment Preservation-based Framework for Evaluating Machine Translation Quality of Classical Chinese Literature

ACL ARR 2025 February Submission4107 Authors

Abstract: We present a novel framework for evaluating sentiment preservation in machine translation of classical Chinese literature, introducing two complementary metrics: the Sentiment Deviation Index (SDI) and Sentiment Preservation Score (SPS). Through a comprehensive parallel corpus of 19,999 classical Chinese-English sentence pairs annotated with fine-grained sentiment labels, we demonstrate that modern MT systems show promising yet varied capabilities across genres (mean SPS=0.841 for GPT-4o), with legal texts achieving exceptional preservation (mean SPS=0.954) compared to literary works (mean SPS=0.831). Our framework, supported by empirically validated weights for balancing polarity and intensity preservation, reveals fundamental challenges in preserving cultural and emotional nuances in classical literature translation, establishing a foundation for advancing cross-cultural sentiment analysis and emotionally intelligent translation systems.

Paper Type: Long

Research Area: Machine Translation

Research Area Keywords: Machine Translation Evaluation, Classical Chinese Literature Translation, Cross-lingual Sentiment Analysis

Contribution Types: Data resources

Languages Studied: Chinese, English

Submission Number: 4107