Scholarlink

<p>This study introduces a hybrid neuro-symbolic framework that achieves deterministic detection of statutory inconsistency in complex law. We use the U.S. Internal Revenue Code (IRC) for our investigation because its complexity makes it a fertile domain for identifying conflicts. Our research advances a solution for detecting inconsistent provisions by combining Large Language Models (LLMs) with a symbolic reasoner.</p><p>To evaluate this approach, we conducted experiments using GPT-4o, GPT-5, and Prolog. First, we used GPT-4o to translate IRC Section&#xa0; 121 into formal Prolog rules, which were then refined in the SWISH environment. Next, these rules were incorporated into GPT-4o prompts to test whether Prolog-augmention would improve the ability to detect legal inconsistencies.</p><p>The results show that Prolog augmentation did not improve the model’s success rate. Whether relying on natural language alone or using the Prolog-augmented prompts, GPT-4o only detected the inconsistency in one out of three testing strategies, resulting in a low 33% accuracy rate. Furthermore, the model’s underlying reasoning was actually worse when using the augmented prompts: the natural-language approach achieved 100% rule coverage, whereas the Prolog-augmented approach only reached 66%, indicating that the model missed critical steps in its statutory analysis.</p><p>In contrast to probabilistic prompting, the hybrid Prolog model produced deterministic and reproducible results. Guided by GPT-5 for refinement, the model formalized the IRC section’s competing interpretations and successfully detected an inconsistency zone. Validation tests confirm that the Prolog implementation is accurate, internally consistent, deterministic, and capable of autonomously identifying inconsistencies. These findings show that LLM-assisted formalization, anchored in symbolic logic, enables transparent and reliable statutory inconsistency detection.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

LLM-assisted formalization for deterministic detection of statutory inconsistency in tax law

  • 期刊论文

  • 发表日期：

    2026年7月10日

查看全文
  • Borchuluun Yadamsuren,
  • Steven Keith Platt,
  • Miguel Diaz

摘要

This study introduces a hybrid neuro-symbolic framework that achieves deterministic detection of statutory inconsistency in complex law. We use the U.S. Internal Revenue Code (IRC) for our investigation because its complexity makes it a fertile domain for identifying conflicts. Our research advances a solution for detecting inconsistent provisions by combining Large Language Models (LLMs) with a symbolic reasoner.

To evaluate this approach, we conducted experiments using GPT-4o, GPT-5, and Prolog. First, we used GPT-4o to translate IRC Section  121 into formal Prolog rules, which were then refined in the SWISH environment. Next, these rules were incorporated into GPT-4o prompts to test whether Prolog-augmention would improve the ability to detect legal inconsistencies.

The results show that Prolog augmentation did not improve the model’s success rate. Whether relying on natural language alone or using the Prolog-augmented prompts, GPT-4o only detected the inconsistency in one out of three testing strategies, resulting in a low 33% accuracy rate. Furthermore, the model’s underlying reasoning was actually worse when using the augmented prompts: the natural-language approach achieved 100% rule coverage, whereas the Prolog-augmented approach only reached 66%, indicating that the model missed critical steps in its statutory analysis.

In contrast to probabilistic prompting, the hybrid Prolog model produced deterministic and reproducible results. Guided by GPT-5 for refinement, the model formalized the IRC section’s competing interpretations and successfully detected an inconsistency zone. Validation tests confirm that the Prolog implementation is accurate, internally consistent, deterministic, and capable of autonomously identifying inconsistencies. These findings show that LLM-assisted formalization, anchored in symbolic logic, enables transparent and reliable statutory inconsistency detection.