发表机构
Software Engineering Institute, East China Normal University（华东师范大学软件工程学院）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
本文发现文档解析失败源于页面崩溃而非复杂性，提出基于失败条件的分解修复方法，在低令牌消耗下显著提升解析质量。
AI 中文摘要
端到端文档解析器日益为复杂页面提供可选的推理模式。在一个包含180页、按熵分层发现的样本上，使用一个冻结的4B检查点，复杂性是错误的决策变量。推理将平均质量降低2.21 Overall，同时消耗1.54倍的令牌；一个预注册的仅输入模型无法预测其符号化收益（留出法AUROC为0.47，与随机猜测无异）。收益集中在普通通过（ordinary pass）已经崩溃的页面上，而这些页面看起来并不复杂：共享崩溃的布局熵低于健康页面，但消耗的令牌却是后者的19倍，表现为退化性重复，即使将预算加倍也无法解决。切换模式很少能修复这些页面：83%的页面在推理下仍会复发。我们转而从普通通过的轨迹中检测崩溃，通过投影对页面进行分解，并重新解析每个区域。修复在1.13倍令牌消耗下获得1.40 Overall的提升（95%置信区间[0.68, 2.16]），在三个检查点上均可复现，并且在所有参数冻结的情况下，在剩余的1,175个基准页面上获得2.41的提升（置信区间[1.64, 3.46]）。
英文摘要
End-to-end document parsers increasingly offer an optional reasoning mode for complex pages. On a 180-page entropy-stratified discovery sample with one frozen 4B checkpoint, complexity is the wrong decision variable. Reasoning lowers mean quality by 2.21 Overall at 1.54x tokens; a preregistered input-only model cannot predict its signed benefit (held-out AUROC 0.47, indistinguishable from chance). The benefit concentrates on pages whose ordinary pass has already collapsed, and they do not look complex: shared collapses have lower layout entropy than healthy ones yet consume 19x the tokens as degenerate repetition that doubling the budget does not cure. Switching modes rarely repairs them: 83% recur under reasoning. We instead detect collapse from the ordinary-pass trace, decompose the page by projection, and re-parse each region. Repair gains 1.40 Overall (95% CI [0.68, 2.16]) at 1.13x tokens, replicates across three checkpoints, and, with all parameters frozen, gains 2.41 (CI [1.64, 3.46]) on the remaining 1,175 benchmark pages.
Comments5 pages, 3 figures, 2 tables. Submitted to ICASSP 2027
