硬提示压缩中的指称悬空：范式级失败模式
原标题：Relevant but Incomplete: Referential Dangling as a Paradigm-Level Failure Mode in Hard Prompt Compression
AI 摘要
该研究识别了硬提示压缩中的一种结构性失败模式——指称悬空，即独立选择可能拆分依赖证据对，保留答案但删除解释所需的实体。在0.30压缩比下，Beaver在三个多跳问答数据集上的悬空率为34-54%，所有六种测试压缩器在HotpotQA上均表现出高达60%的悬空率。重新插入缺失的支持段落可将准确率提高29-34个百分点，而自动恢复探针在压缩比仅增加0.01的情况下带来4.7个百分点的提升。研究结论是压缩应同时优化相关性和指称完整性。
正文节选
Relevant but Incomplete: Referential Dangling as a Paradigm-Level Failure Mode in Hard Prompt Compression Abstract Hard prompt compression reduces long-context inference cost by independently scoring tokens, sentences, or chunks and retaining the highest-scoring units under a budget. We identify a structural failure in this procedure: independent selection can split dependent evidence pairs, retaining one member while deleting the other. When retained text contains an answer but deleted text def