arXiv：cs.CL· Jianfei Ma, Zhaoxin Feng, Emmanuele Chersoni, Si Chen·· 6 小时前
只用语言学规则做提示词压缩：无需 LM 前向传播，推理成本最多可降
Every Time I Hire a Linguist, Inference Costs Go Down: On Linguistic Rules as Effective Prompt Compressors
AI 导读
EMNLP 2026 一篇论文提出仅靠语言学规则压缩提示词，压缩阶段完全不做 LM 前向传播，只用 CPU 处理；研究者通过离线进化搜索组合词汇、句法、语义和篇章规则，在短文本、多文档推理和对话记忆问答数据集上取得与近期先进提示词压缩策略相近的表现。该方法在轻中度压缩下效果最好，压缩率升高后性能下降，规则也从 token 裁剪转向句子抽取。
正文
Abstract:Prompt compression shortens LLM input to reduce inference cost, yet existing methods score token importance through LM forward passes. It remains questionable whether such nuanced, costly token selection is necessary. Compression requires identifying informative content, a problem that linguistic research has long addressed through cues that can be operationalized as deterministic rules. We therefore ask: can \textbf{linguistic rules alone} serve as effective prompt compressors, without LM-based scoring at compression time?
To address this, we conduct offline evolutionary search over lexical, syntactic, semantic, and discourse seeds to find competitive rule combinations. The resulting linguistic compressor requires no LM forward pass at deployment and uses only CPU-side processing for compression. We evaluate it with a dual-path protocol to balance compression quality and reconstruction fidelity.
Across short passages, multi-document reasoning, and dialogue-memory QA datasets, evolved compressors achieve performance similar to that of recent advanced prompt-compression strategies. Performance is strongest under light-to-moderate compression and degrades as compression becomes more aggressive, while the Direct and Reconstruction paths exhibit distinct patterns. Evolutionary analysis reveals that effective compression fuses signals across linguistic levels and, as the compression ratio increases, rules shift from token pruning to sentence extraction.
| Comments: | EMNLP 2026 |
| Subjects: | Computation and Language (cs.CL); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2607.25335 [cs.CL] |
| (or arXiv:2607.25335v3 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2607.25335 arXiv-issued DOI via DataCite |
Submission history
From: Zhaoxin Feng [view email]
[v1]
Tue, 28 Jul 2026 06:33:27 UTC (735 KB)
[v2]
Wed, 29 Jul 2026 12:56:44 UTC (735 KB)
[v3]
Wed, 14 Oct 2026 03:45:08 UTC (735 KB)
来源：arXiv：cs.CL · arxiv.org