arXivDaily arXiv每日学术速递 周一至周五更新
北美退伍军人节(Veterans Day),12 日无更新
arXiv 2306.08042cs.CLcs.AI
收藏

FLamE: 基于自然语言解释的少样本学习

FLamE: Few-shot Learning from Natural Language Explanations

University of Chicago(芝加哥大学)

Yangqiaoyu Zhou, Yiming Zhang, Chenhao Tan

AI总结:

针对自然语言解释提升分类效果有限的问题,提出两阶段少样本学习框架FLamE,先通过GPT-3生成解释再微调小模型,在e-SNLI上准确率显著提升,并揭示生成解释中标签特定线索的重要作用。

摘要:

自然语言解释有潜力提供丰富信息,原则上可指导模型推理。然而,Lampinen等人(2022)的近期研究显示,自然语言解释在提升分类效果方面的作用有限。为有效从解释中学习,我们提出FLamE——一个两阶段少样本学习框架,该框架首先使用GPT-3生成解释,随后用生成的解释微调一个更小的模型(如RoBERTa)。我们在自然语言推理任务上的实验证明了该方法优于强基线,在e-SNLI数据集上,准确率较GPT-3 Babbage提升17.6%,较GPT-3 Davinci提升5.7%。尽管分类性能有所提升,但人工评估出人意料地显示,大多数生成的解释无法充分证明分类决策的合理性。额外分析指出了生成的解释中标签特定线索(如中性标签对应的“not know”)的重要作用。

英文摘要:

Natural language explanations have the potential to provide rich information that in principle guides model reasoning. Yet, recent work by Lampinen et al. (2022) has shown limited utility of natural language explanations in improving classification. To effectively learn from explanations, we present FLamE, a two-stage few-shot learning framework that first generates explanations using GPT-3, and then finetunes a smaller model (e.g., RoBERTa) with generated explanations. Our experiments on natural language inference demonstrate effectiveness over strong baselines, increasing accuracy by 17.6% over GPT-3 Babbage and 5.7% over GPT-3 Davinci in e-SNLI. Despite improving classification performance, human evaluation surprisingly reveals that the majority of generated explanations does not adequately justify classification decisions. Additional analyses point to the important role of label-specific cues (e.g., "not know" for the neutral label) in generated explanations.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑