Abstract
This paper introduces the logical concept of "forgetting"—the principled removal of information—to Abstract Argumentation Frameworks (AFs). We define a set of desiderata for forgetting an argument and demonstrate that while simple node removal is often insufficient, sophisticated methods can preserve the logical integrity of the remaining debate. We provide a constructive algorithm for stable semantics that satisfies key properties by strategically introducing auxiliary arguments.
Introduction
In formal logic, "forgetting" a variable involves updating a knowledge base such that the variable no longer appears, while preserving the logical relationships between the remaining variables. While well-studied in propositional logic and AI planning, this concept has been largely unexplored in Abstract Argumentation Frameworks (AFs). An AF is a directed graph where nodes represent arguments and edges represent "attacks." The goal is to determine sets of acceptable arguments, known as extensions, under various semantics.
The Problem of Ad Hoc Removal
The primary challenge in AFs is the interconnectedness of arguments. A naive "ad hoc" approach—simply deleting an argument and its associated attacks—can lead to "collateral damage." For instance, if Argument A defended Argument B from an attack, removing A might cause B to be rejected, even if the reason for forgetting A was unrelated to B's validity. This research seeks a principled, axiomatic method to remove arguments while minimizing such effects on the remaining framework.
Axiomatic Framework and Desiderata
The authors propose several "desiderata" (desirable properties) for forgetting, categorized into three blocks:
- Semantical (Extensions): Rules (e1–e4) defining how the sets of winning arguments should evolve. For example, the "Reiter condition" (e1) requires that new extensions be the original ones minus the forgotten argument.
- Semantical (Acceptance): Rules (r1–r4) focusing on the continued acceptance status of specific arguments.
- Syntactical: Rules (s1–s3) concerning the graph structure, where s3 represents the ad hoc approach.
The analysis reveals a fundamental tension between maintaining a simple graph structure (syntactic) and preserving the debate's logical meaning (semantic). Notably, "Strong Persistence" (e2) is shown to be mathematically impossible in AFs.
Algorithm for Stable Semantics
The paper identifies stable semantics as unique in allowing the clean removal of "contaminated" extensions without introducing further logical issues. To operationalize this, the authors present Algorithm 1.
This algorithm does not merely delete nodes; it strategically introduces "fresh" auxiliary arguments to suppress unwanted extensions that might emerge post-removal. This ensures the resulting graph $G$ satisfies the desired properties, specifically the set {s1, e4, v3}.
Key Insights and Results
The research provides a comprehensive analysis of which properties are compatible under different semantics. Key findings include:
- Inadequacy of Ad Hoc Removal: Simple deletion (s3) is highly restrictive and often incompatible with sophisticated semantical requirements.
- Stable Semantics Advantage: It is the only semantics studied that facilitates the removal of specific extensions without collateral logical damage.
- Vacuity Principle: The proposed construction adheres to a "vacuity" principle, leaving the framework largely untouched if the forgotten argument was not pivotal to the debate.
Conclusion
This work establishes a formal foundation for forgetting in abstract argumentation. By defining clear desiderata and providing a constructive algorithm for stable semantics, the authors move beyond ad hoc methods toward principled information removal. Future work will extend these results to "batch" forgetting of multiple arguments and develop algorithms for other major semantics, such as preferred and grounded frameworks.
Summary updated