Mining Legal Arguments in U.S. Corporate Case Law

ACL ARR 2026 January Submission10660 Authors

Keywords: Legal argumentation, argument mining, information retrieval

Abstract:

The application of argument mining in the legal domain has the potential to influence various real-world tasks, including dense passage retrieval, argument completion, generation, and review. To address these applications, we introduce a simple, syllogism-centric scheme that constructs full arguments by chaining "atomic'' syllogisms. In contrast to taxonomies that distinguish between "claim'' and "premise,'' we draw inspiration from the IRAC annotation method by consolidating local (intermediate) conclusions into analysis or rule units, and reserving the “conclusion” label exclusively for the document’s final outcome. We formalize this scheme with concise guidelines for marking atomic links and evaluate it by manually annotating 42 U.S. corporate reorganization cases under I.R.C. S368.
Schema-based argument mining improves the predictive and retrieval performance of language models. Experimental results demonstrate that our scheme provides strong semantic separation (0.81 Macro-F1 for 4-class classification) and a structure that supports logical completion (60\% Recall@20) in a computationally efficient and straightforward setting. The dataset of cases with extracted arguments is released as a resource for future research.

Paper Type: Short

Research Area: Sentiment Analysis, Stylistic Analysis, and Argument Mining

Research Area Keywords: Information Retrieval and Text Mining, Resources and Evaluation, Sentiment Analysis, Stylistic Analysis, and Argument Mining

Contribution Types: NLP engineering experiment, Data resources, Data analysis

Languages Studied: English

Submission Number: 10660