arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2403.04400cs.CL
收藏

探索自然语言推理中组合泛化的持续学习

Exploring Continual Learning of Compositional Generalization in NLI

Heidelberg University(海德堡大学)

Xiyan Fu, Anette Frank

AI总结:

本文提出C2Gen NLI挑战,探讨持续学习对NLI组合泛化的影响,发现模型在持续场景下泛化失败,但通过合理排序子任务及利用依赖关系可增强该能力。

摘要:

组合自然语言推理已被探索用于评估神经模型执行NLI的真实能力。然而,当前的评估假设模型预先拥有对所有基本推理的完全访问权限,这与人类不断获取推理知识形成对比。在本文中,我们引入了推理中的持续组合泛化(C2Gen NLI)挑战,模型在该挑战中持续获取构成基本推理任务的知识,作为组合推理的基础。我们通过为组合NLI推理任务设计持续学习设置,探讨持续学习如何影响NLI中的组合泛化。我们的实验表明,模型在持续场景中无法进行组合泛化。为解决这一问题,我们首先对各种持续学习算法进行基准测试并验证其有效性。随后我们进一步分析C2Gen,重点关注如何排序基本推理和组合推理类型,并检查子任务之间的相关性。我们的分析表明,通过持续学习子任务并观察它们的依赖关系及不断增加的难度,持续学习能够增强组合泛化能力。

英文摘要:

Compositional Natural Language Inference has been explored to assess the true abilities of neural models to perform NLI. Yet, current evaluations assume models to have full access to all primitive inferences in advance, in contrast to humans that continuously acquire inference knowledge. In this paper, we introduce the Continual Compositional Generalization in Inference (C2Gen NLI) challenge, where a model continuously acquires knowledge of constituting primitive inference tasks as a basis for compositional inferences. We explore how continual learning affects compositional generalization in NLI, by designing a continual learning setup for compositional NLI inference tasks. Our experiments demonstrate that models fail to compositionally generalize in a continual scenario. To address this problem, we first benchmark various continual learning algorithms and verify their efficacy. We then further analyze C2Gen, focusing on how to order primitives and compositional inference types and examining correlations between subtasks. Our analyses show that by learning subtasks continuously while observing their dependencies and increasing degrees of difficulty, continual learning can enhance composition generalization ability.

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。

↑