arXiv cs.CL· Adib Sakhawat, Syed Rifat Raiyan·· 12 小时前AI 评分31
Literal Lock-In：测试语言如何改变孟加拉语习语在类型化决策模型中的字面吸引
Literal Lock-In: How Test Language Shifts Literal Attraction in Typed Decision Models on Bengali Idioms
AI 导读
研究审计了编码器打分器、解码器打分器和闭源 API 三类非生成式决策模型，发现把习语自身的字面释义放入选项列表会吸引所有模型，占各模型剩余孟加拉语错误的 65% 至 96%。
正文 · 原文
Abstract:Decision models, non-generative scorers that report a probability over caller-supplied options and emit no text, are increasingly deployed as routers, classifiers and guardrails. Because they generate nothing, rationales and chain-of-thought do not apply. We audit three: an encoder scorer, a decoder scorer and a closed API, entirely through interventions on their input, using expert-annotated Bengali idioms carrying a gold figurative and a gold literal gloss. Dropping an idiom's own literal gloss into the option list draws every arm toward it, taking 65% to 96% of each arm's remaining Bengali errors. A model-by-language matrix over one shared item set shows how far this depends on the test language: every arm is 1.9 to 6.4 times more captured in Bengali than in English on the same paired entries, the commercial model's apparent resistance is a property of its test language, and capability does not order the failure. Calibration degrades in all three arms, and on the open arms confidence inverts, so gating on it is worse than gating at random. On a semantics-free option list, where accuracy is at chance, the best-calibrated arm still reports mean confidence 0.82. Code and data: this https URL.
| Comments: | Under review, 33 pages, 12 figures, 22 tables |
| Subjects: | Computation and Language (cs.CL) |
| Cite as: | arXiv:2610.22526 [cs.CL] |
| (or arXiv:2610.22526v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2610.22526 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Syed Rifat Raiyan [view email]
[v1]
Thu, 15 Oct 2026 18:19:02 UTC (1,631 KB)
来源：arXiv cs.CL · arxiv.org