arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2510.07178cs.CL
收藏

无偏语言模型以不自然的方式学习:LLM如何无法区分可能与不可能

Biasless Language Models Learn Unnaturally: How LLMs Fail to Distinguish the Possible from the Impossible

Tel Aviv University(特拉维夫大学);Ecole Normale Supérieure(巴黎高等师范学院);Earth Species Project(地球物种项目);EHESS, PSL University, CNRS(法国社会科学高等研究院、巴黎文理研究大学、法国国家科学研究中心)

Imry Ziv, Nur Lan, Emmanuel Chemla

AI总结:

研究发现GPT-2在学习自然语言和不可能语言时表现相似,反驳了LLM具备区分能力的假设,表明其学习过程无系统性区分。

摘要:

大型语言模型(LLMs)是否对人类可能与不可能的语言有区分敏感性?此问题最近被用于讨论LLM与人类是否共享相同的固有学习偏见。先前研究通过比较现有语言数据集和通过扰动函数生成的“不可能”数据集上的LLM学习曲线得出肯定回答。使用相同方法,我们对更广泛的语言和不可能扰动进行了检验。我们发现,在大多数情况下,GPT-2在学习每种语言及其不可能对应物时同样容易,与先前发现相反。我们还通过测试GPT-2是否在跨语言变异性指标上对自然语言与不可能语言提供任何分离来施加更宽松的条件。综合来看,这些视角表明GPT-2在可能与不可能之间并无系统性区分。

英文摘要:

Are large language models (LLMs) sensitive to the distinction between humanly possible and impossible languages? This question was recently used in a broader debate on whether LLMs and humans share the same innate learning biases. Previous work has answered it in the positive by comparing LLM learning curves on existing language datasets and on "impossible" datasets derived from them via various perturbation functions. Using the same methodology, we examine this claim on a wider set of languages and impossible perturbations. We find that in most cases, GPT-2 learns each language and its impossible counterpart equally easily, in contrast to previous findings. We also apply a more lenient condition by testing whether GPT-2 provides any kind of separation between the whole sets of natural vs. impossible languages, based on cross-linguistic variance in metrics derived from the learning curves. Taken together, these perspectives show that GPT-2 provides no systematic separation between the possible and the impossible.

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。

↑