Paragraph Boundaries Are Not White Space:Compression Depth as the Signature of Hierarchical Structure
- 类型:arxiv
- 标识:2609.23551
- 链接:https://arxiv.org/abs/2609.23551
- 主分类:engineering
- 形态:position
- TLDR:Standard positional encodings represent position as a one-dimensional reading-order coordinate, but reading order alone does not determine hierarchical textual structure. We use a hierarchical rotary positional encoding (hRoPE) that represents paragraph, sentence, and token indices as separate channels, hold the token sequence fixed, intervene on the paragraph coordinate p1, and measure cross-paragraph attention with a token-distance-exact estimator. Attention is compressed relative to a token-distance-matched baseline in every corpus, but compression alone is not diagnostic of true structure:
- 待LLM分类:是
- 标题中文:[标题中文] 段落边界并非空白:压缩深度作为层级结构的特征
- TLDR中文:[TLDR中文] 标准位置编码将位置表示为一维的阅读顺序坐标,但仅凭阅读顺序无法确定层级化的文本结构。我们使用层级旋转位置编码(hRoPE),将段落、句子和 token 索引表示为独立的通道,在固定 token 序列不变的情况下,对段落坐标 p1 进行干预,并使用与 token 距离精确的估计器衡量跨段落注意力。在所有语料库中,相对于按 token 距离匹配的基线,注意力均呈现压缩现象,但仅凭压缩不足以诊断真正的结构:
- 来源文件:
- /inbox/tom/_candidates/2026-09-28-agent-rag-longcontext-candidates.json