多语言 LLM 否定表征的因果分析：「否定」在翻译中并未丢失
"Not" is Not Lost in Translation: A Causal Analysis of Negation Representations in Multilingual LLMs
AI 导读
一项研究对六个开源大语言模型在英语、法语、越南语和中文上的否定表征进行因果干预分析，基于 FLORES-200 衍生的否定翻译数据集与合成对比数据集，发现否定表征可被因果控制且跨语言基本共享。从一种语言提取的对比激活方向可诱导或消除其他语言的否定，且每个模型中单个稀疏自编码器（SAE）特征即可跨语言控制极性。密集与稀疏干预还能泛化到另外四种语言。
正文 · 原文
Abstract:Negation occurs in every language, yet its surface realization differs drastically. Despite this, multilingual large language models (LLMs) translate negation robustly, raising the question of whether they share an underlying representation of negation across languages. We study this question in six open models across English, French, Vietnamese, and Chinese, using causal interventions on a synthetic contrastive dataset and a negation translation dataset derived from FLORES-200. Across models, we find strong evidence that negation representations are causally controllable and largely shared across languages. In particular, we show that a contrastive activation direction extracted from one language can induce or remove negation in the others, and directions extracted from different languages are substantially aligned. We further find that negation is represented in sparse features: a single sparse autoencoder (SAE) feature per model suffices to control polarity across languages, while responding selectively to negated inputs. Finally, both dense and sparse interventions generalize to four additional languages and FLORES-derived sentences. We release data and code to support future work: this https URL.
| Comments: | 29 pages, 20 figures |
| Subjects: | Computation and Language (cs.CL); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.32281 [cs.CL] |
| (or arXiv:2610.32281v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2610.32281 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Giang Son Nguyen [view email]
[v1]
Fri, 23 Oct 2026 07:38:41 UTC (1,034 KB)
来源：arXiv cs.AI · arxiv.org