Abstract: This paper presents an automated system for identifying "Dialectical Examples"—complex configurations of legal cases used in pedagogical settings to teach students the nuances of legal argumentation. By formalizing rhetorical moves into logical queries within a structured inheritance system, the researchers demonstrated that complex pedagogical examples can be generated in seconds, significantly outperforming manual search methods and improving student performance in preliminary tests.
Introduction
In legal education, mastering the art of argument requires more than just finding a single relevant case. Students must learn how multiple cases interact: how one case can "trump" another, how a "vanilla" case illustrates a single factor, or how a set of cases can collectively cover all defensive positions. Historically, finding these specific configurations, termed "Argument Contexts," has been an incredibly labor-intensive task for instructors, often requiring days of manual comparison across digital databases.
This research addresses this bottleneck by automating the generation of Dialectical Examples using symbolic Artificial Intelligence. By shifting the focus from simple case retrieval to complex case configuration, the system provides a powerful tool for legal pedagogy.
Methodology
The system was built using Loom, a structured inheritance system. The researchers modeled 26 legal cases from the domain of trade secret misappropriation, breaking each case down into 20 symbolic "factors" (e.g., "Security-Measures" or "Bribe-Employee") that favor either the plaintiff ($\pi$) or the defendant ($\partial$).
Formalizing Rhetoric
The core innovation of the study was the formalization of five distinct types of Dialectical Examples into first-order logic queries:
- Representative Examples: Cases that clearly demonstrate a factor's effect (Vanilla, Simple Conflict, or Packed cases).
- Conflict Resolution: Cases showing one set of factors outweighing another.
- Refutational Comparisons: Counterexamples that "trump" an opponent's cited case by being more "on point."
- Ceteris Paribus: Pairs of cases identical except for one factor, proving that factor's significance.
- Coherence Examples: Sets of cases that collectively show that none of an opponent's strengths are fatal to the overall argument.
Key Insights: Argumentation as Configuration
A central insight of this work is that legal argumentation is best viewed as a graph-like configuration rather than a linear search. The relationship between cases is often more pedagogically significant than the individual cases themselves. By defining rhetorical moves like "trumping" declaratively, the system can search for cases that prove specific logical relationships—for instance, finding cases where a bribe is legally more significant than a non-disclosure agreement.
Results and Impact
The system demonstrated remarkable efficiency, generating complex Argument Contexts in 3.5 to 22.9 seconds, a task estimated to take human instructors hours or even days.
Beyond efficiency, the pedagogical impact was significant. In a preliminary experiment with first-year law students, those taught using the program-generated examples scored 23–24 out of 25 on a complex argument task. In contrast, the control group, which did not receive the specialized tutorial, scored between 6 and 13. This suggests that presenting students with logical "Argument Contexts" that force difficult choices is a highly effective teaching method.
Limitations and Future Outlook
While successful, the study was limited by its small database of 26 cases. Scaling the system to thousands of cases would require substantial manual knowledge engineering to "factorize" the legal texts. Additionally, while the examples were generated automatically, the tutoring itself was conducted by a human. Future work aims to integrate these logical queries into a fully automated Intelligent Tutoring System (ITS) and to utilize independent evaluators to ensure objective grading of student performance.
Summary updated