arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2307.12976cs.CL
收藏

评估语言模型中知识编辑的涟漪效应

Evaluating the Ripple Effects of Knowledge Editing in Language Models

Tel Aviv University(特拉维夫大学);Google Research(谷歌研究院)

Roi Cohen, Eden Biran, Ori Yoran, Amir Globerson, Mor Geva

AI总结:

本文提出评估语言模型知识编辑的涟漪效应,构建RippleEdits基准,发现现有编辑方法难以一致更新相关事实,而简单的上下文内编辑基线表现最佳。

摘要:

现代语言模型捕获了大量的事实知识。然而,一些事实可能被错误地诱导或随时间变得过时,导致生成的事实不正确。这促使了各种编辑方法的发展,这些方法允许更新模型编码的事实。对这些方法的评估主要集中在测试单个事实是否被成功注入,以及针对其他主体的类似预测是否未改变。在此,我们认为这种评估是有限的,因为注入一个事实(例如,“Jack Depp 是 Johnny Depp 的儿子”)会引入一种“涟漪效应”,即模型需要更新的额外事实(例如,“Jack Depp 是 Lily-Rose Depp 的兄弟姐妹”)。为了解决这个问题,我们提出了一套新颖的评估标准,考虑编辑对相关事实的影响。利用这些标准,我们构建了 RippleEdits,一个包含 5K 个事实编辑的诊断基准,捕捉了各种类型的涟漪效应。我们在 RippleEdits 上评估了突出的编辑方法,表明当前方法未能引入模型知识的一致变化。此外,我们发现一个简单的上下文内编辑基线在我们的基准上获得了最佳分数,这为模型编辑提出了一个有前景的研究方向。

英文摘要:

Modern language models capture a large body of factual knowledge. However, some facts can be incorrectly induced or become obsolete over time, resulting in factually incorrect generations. This has led to the development of various editing methods that allow updating facts encoded by the model. Evaluation of these methods has primarily focused on testing whether an individual fact has been successfully injected, and if similar predictions for other subjects have not changed. Here we argue that such evaluation is limited, since injecting one fact (e.g. ``Jack Depp is the son of Johnny Depp'') introduces a ``ripple effect'' in the form of additional facts that the model needs to update (e.g.``Jack Depp is the sibling of Lily-Rose Depp''). To address this issue, we propose a novel set of evaluation criteria that consider the implications of an edit on related facts. Using these criteria, we then construct RippleEdits, a diagnostic benchmark of 5K factual edits, capturing a variety of types of ripple effects. We evaluate prominent editing methods on RippleEdits, showing that current methods fail to introduce consistent changes in the model's knowledge. In addition, we find that a simple in-context editing baseline obtains the best scores on our benchmark, suggesting a promising research direction for model editing.

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。

↑