BaRDa:一个区分事实准确性与推理能力的信念与推理数据集
BaRDa: A Belief and Reasoning Dataset that Separates Factual Accuracy and Reasoning Ability
AI总结:
本文提出了 BaRDa 数据集,通过利用真假事实混合的人工标注蕴含树,清晰区分并量化了语言模型的事实准确性与推理能力。
摘要:
尽管有许多基准测试在比较现代语言模型(LMs)的性能,但最终任务评估常常将*事实准确性*(“真实性”)和*推理能力*(“理性”,或正确报告信念推论的“诚实性”)的概念混为一谈。我们的目标是构建一个能清晰区分这两个概念的数据集。我们的方法是利用并扩展一组人工标注的*蕴含树*,通过设计来表达良好和糟糕的推理链,并使用真假事实的混合,特别是包含反事实示例,以避免信念偏差(也称为“内容效应”)。生成的数据集名为 BaRDa,包含 3000 个蕴含关系(1787 个有效,1213 个无效),使用了 6681 条真实陈述和 2319 条虚假陈述。在四个 GPT 系列模型 GPT3(curie)/GPT3(davinici)/3.5/4 上进行测试,我们发现事实准确性(真实性)得分为 74.1/80.6/82.6/87.1,推理准确性得分为 63.1/78.0/71.8/79.2。这表明模型在事实准确性和蕴含推理方面有明显的提升,并且该数据集提供了一个新基准,能更清晰地分离并量化这两个概念。
英文摘要:
While there are numerous benchmarks comparing the performance of modern language models (LMs), end-task evaluations often conflate notions of *factual accuracy* ("truth") and *reasoning ability* ("rationality", or "honesty" in the sense of correctly reporting implications of beliefs). Our goal is a dataset that clearly distinguishes these two notions. Our approach is to leverage and extend a collection of human-annotated *entailment trees*, engineered to express both good and bad chains of reasoning, and using a mixture of true and false facts, in particular including counterfactual examples, to avoid belief bias (also known as the "content effect"). The resulting dataset, called BaRDa, contains 3000 entailments (1787 valid, 1213 invalid), using 6681 true and 2319 false statements. Testing on four GPT-series models, GPT3(curie)/GPT3(davinici)/3.5/4, we find factual accuracy (truth) scores of 74.1/80.6/82.6/87.1 and reasoning accuracy scores of 63.1/78.0/71.8/79.2. This shows the clear progression of models towards improved factual accuracy and entailment reasoning, and the dataset provides a new benchmark that more cleanly separates and quantifies these two notions.
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。