研究发现文学翻译中流畅与忠实存在权衡，长段落影响评估结果。
Fluency and Faithfulness in Human and Machine Literary Translation
- 用词性短语分析段落流畅度，以COMET-KIWI衡量翻译忠实度
- 13万段落数据显示流畅与忠实呈负相关，尤其在长文本中更明显
- 人类与谷歌翻译均现此现象，但Gemma模型表现较弱，适合关注评估偏差的研究者
文学翻译需兼顾目标语流畅性与源语文本忠实度。现有大语言模型虽常生成流畅译文，但其流畅性是否真正保留原意尚不明确。本研究基于106部小说、16种源语的130,486段译文，涵盖人工、Google Translate及TranslateGemma的翻译结果。流畅度通过针对词性n-gram训练的translationese分类器衡量，忠实度采用自动评估指标COMET-KIWI。控制段落长度后，发现流畅度与忠实度之间存在稳定负相关。该模式在人类与Google Translate中显著，但在TranslateGemma中较弱且多数不显著。结果表明段落长度影响自动评估效果，并提示文学翻译中存在流畅性与忠实度的权衡关系。
原文摘要 · Abstract (English)
Literary translation requires balancing target-language fluency with faithfulness to the source. Recent large language models (LLMs) often produce fluent translations, but it remains unclear whether fluency corresponds to semantic preservation in literary text. We examine this relationship using 130,486 translated paragraphs from 106 novels in 16 source languages, including human, Google Translate, and TranslateGemma translations. Fluency is measured as original-likeness with a translationese classifier trained on paragraph part-of-speech n-grams, and faithfulness with the automatic translation evaluation metric COMET-KIWI. We control for paragraph length and find a consistent negative correlation between fluency and faithfulness. The pattern appears for both human and Google Translate, but is weaker and often non-significant for TranslateGemma. These results show that segment length matters for automatic evaluation and suggest a tradeoff between fluency and faithfulness in literary translation.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。