arXiv cs.CL· Mahmoud Alyosify, Mirna Imbabi, Eman Elkhamisi·· 11 小时前AI 评分31
超越截断？LLM 解码中的蕴含许可省略与语义保留边界
Beyond Truncation? Entailment-Licensed Elision and the Limits of Semantic Preservation in LLM Decoding
AI 导读
研究提出 Entailment-Licensed In-Decoding Elision（ELIDE），在解码时仅当自然语言推理模型判定某段文本已被前文蕴含时才向读者省略该段，且不改变模型生成。ELIDE 在三个模型族上优于等长截断，但在 670 条公开提示的 936 次省略中 59% 损害答案，且无自动指标能检出；将许可范围收窄到话语单元后，两次新测试中有害比例降至约十分之一。
正文 · 原文
Abstract:Instruction-tuned large language models (LLMs) answer at greater length than a request needs. We ask whether a training-free controller inside the decoding loop can remove that surplus while preserving content, judged against the simplest possible baseline: truncating the model's own answer to the same length. An entropy-gated logits processor removes 25.5-34.9% of tokens on MT-Bench development questions and 19.9-34.2% on a sealed test split within pre-registered BERTScore non-inferiority margins, yet we prove and confirm by audit that its output is always a prefix of the unmodified answer: the saving is truncation. A survey of 23 training-free mechanism families finds none that selects content better than truncation without oracle labels. We therefore introduce Entailment-Licensed In-Decoding Elision (ELIDE), which, at each sentence boundary, withholds a span from the reader only when a natural-language inference model judges it entailed by the text already shown, while leaving the model's generation unchanged. ELIDE beats matched-length truncation on three model families, but reading 936 of its elisions on 670 public prompts shows that 59% harm the answer, a failure that no automatic measure detects. Scoping the licence to its discourse unit reduces the harmful share to about one elision in ten on two fresh tests.
| Comments: | 8 pages, 3 figures, 2 tables |
| Subjects: | Computation and Language (cs.CL) |
| Cite as: | arXiv:2611.08975 [cs.CL] |
| (or arXiv:2611.08975v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2611.08975 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Mahmoud Sayed Youssef Kotb [view email]
[v1]
Fri, 6 Nov 2026 04:39:25 UTC (103 KB)
来源：arXiv cs.CL · arxiv.org