LLM能准确评估UGC机器翻译质量，但存在回复拒绝和输出不稳定问题。
Are Large Language Models State-of-the-art Quality Estimators for Machine Translation of User-generated Content?
- 用参数高效微调（PEFT）提升LLM对UGC翻译质量的预测能力
- 在无参考译文情况下，PEFT-LMM得分预测准确率优于基线模型
- 适合关注翻译质量评估且需可解释性的研究者使用
本文研究大型语言模型（LLMs）是否可作为无需参考译文的用户生成内容（UGC）机器翻译质量评估的最先进工具，尤其针对包含情感表达的文本。我们基于已有情感相关数据集，结合人工标注错误，采用多维质量度量（Multi-dimensional Quality Metrics）计算评估分数。在上下文学习与参数高效微调（PEFT）两种场景下，对比多种LLM与微调基线模型的表现。结果表明，对LLM进行PEFT后，在评分预测上表现更优，且能提供人类可理解的解释。然而，人工分析发现其在评估过程中仍存在拒绝响应提示、输出不稳定等问题。
原文摘要 · Abstract (English)
This paper investigates whether large language models (LLMs) are state-of-the-art quality estimators for machine translation of user-generated content (UGC) that contains emotional expressions, without the use of reference translations. To achieve this, we employ an existing emotion-related dataset with human-annotated errors and calculate quality evaluation scores based on the Multi-dimensional Quality Metrics. We compare the accuracy of several LLMs with that of our fine-tuned baseline models, under in-context learning and parameter-efficient fine-tuning (PEFT) scenarios. We find that PEFT of LLMs leads to better performance in score prediction with human interpretable explanations than fine-tuned models. However, a manual analysis of LLM outputs reveals that they still have problems such as refusal to reply to a prompt and unstable output while evaluating machine translation of UGC.
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。