arXiv cs.CL· Maryia Zhyrko, Suzan Verberne, Lifeng Han·· 10 小时前AI 评分34
时间即结构：面向法律文档抽取与推理的时间骨架
Time as Structure: A Temporal Backbone for Extraction and Reasoning over Legal Documents
AI 导读
研究提出将带日期事实及其依赖关系抽取为时间依赖图，再用日历正确的引擎计算法律期限，而非让语言模型直接作答。在英国就业上诉法庭判决上，该引擎复现了 7 项时效裁定中的 6 项，日期与法官认定完全一致；四个语言模型中最强的一个虽算对算术却答错结论，21 次回答中有 6 次自相矛盾。
正文 · 原文
Abstract:Miss a filing deadline by one day and the claim is barred, however strong the case. Computing that deadline is rarely simple: the period runs from a triggering event, is counted by a statutory convention, and may be suspended by a mandatory conciliation window. We ask whether a language model should answer such questions directly, or read the document and leave the arithmetic to code. We extract dated facts and their dependencies into a temporal dependency graph and compute deadlines from it with a calendar-correct engine. On UK Employment Appeal Tribunal judgments the engine reproduces six of seven timeliness rulings, and matches the judges' own dates to the day. The strongest of four language models, asked the same cases, gets the arithmetic right and the answer wrong: in two of its seven responses its stated verdict contradicts its own working, and across the three GPT models six of twenty-one responses do, each time calling a late claim timely. To test the systems at scale we move the dismissal date across the statutory boundary, generating 427 cases whose answers are computed rather than annotated. On the cases both systems answer, the pipeline is right 90.2% of the time against 61.2% for direct answering. It answers two thirds of the cases, and counting its abstentions as errors it does not exceed direct answering (60.4% against 64.6%). The limit is extraction: on contracts the errors are almost never in the arithmetic, but in choosing which event the period starts from.
| Comments: | 17 pages, 2 figures, 12 tables. Accepted at the AACL-IJCNLP 2026 Student Research Workshop |
| Subjects: | Computation and Language (cs.CL) |
| ACM classes: | I.2.7; I.2.4; J.1 |
| Cite as: | arXiv:2608.15270 [cs.CL] |
| (or arXiv:2608.15270v3 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2608.15270 arXiv-issued DOI via DataCite |
Submission history
From: Maryia Zhyrko [view email]
[v1]
Sat, 15 Aug 2026 15:14:46 UTC (196 KB)
[v2]
Sat, 10 Oct 2026 16:37:08 UTC (45 KB)
[v3]
Tue, 10 Nov 2026 04:53:35 UTC (47 KB)
来源：arXiv cs.CL · arxiv.org