A new study published on arXiv explores the trade-off between fluency and faithfulness in literary translation by both humans and machines. Researchers analyzed over 130,000 translated paragraphs from 106 novels, using a translationese classifier to measure fluency and the COMET-KIWI metric for faithfulness. The findings indicate a consistent negative correlation between fluency and faithfulness, suggesting that as translations become more fluent, they may deviate more from the source text's meaning. This pattern was observed in both human and Google Translate outputs, but was less pronounced in TranslateGemma. AI
IMPACT Highlights potential limitations in LLM-based literary translation, suggesting a need for models that better balance stylistic flair with semantic accuracy.
RANK_REASON Academic paper on translation quality metrics. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
