Comparing Weak Admissibility Semantics to their Dung-style Counterparts (Extended Abstract)

Ringo Baumann, Gerhard Brewka, Markus UlbrichtIJCAI 2021 · DOI

TL;DR

This paper introduces and explores "weak admissibility" semantics for argumentation frameworks, a novel approach designed to handle self-defeating arguments more gracefully than existing methods. By leveraging a concept called the "reduct" and a new principle called "modularization," the authors demonstrate how this new semantics relates to and can even characterize classical semantics, offering a unified perspective on argumentation theory.

What Problem Does This Paper Address?

Argumentation frameworks (AFs) are a powerful tool for modeling reasoning and decision-making by representing arguments and their conflicts. However, a long-standing challenge in AFs, particularly for over 25 years, has been how to deal with "self-defeating" arguments – arguments that, directly or indirectly, attack themselves. Existing semantics often struggle with these scenarios, leading to counterintuitive results. This paper addresses this by proposing "weak admissibility," a new semantics that aims to provide a more robust way to handle self-defeat, particularly by limiting its impact.

Methods

The core of the paper revolves around two key concepts: the reduct and modularization.

  • The Reduct: For a given set of accepted arguments $E$ in an AF, the $E$-reduct ($F_E$) is a sub-framework containing only the arguments whose status is still undecided. This is achieved by removing arguments attacked by $E$ and their associated attacks. The authors show that this reduct is not just a tool for weak admissibility but is implicitly present in classical semantics as well, offering a unified way to understand them.
  • Modularization: This is a newly proposed abstract principle for semantics. It formalizes the idea that if you have a "solid" set of accepted arguments $E$ in an AF, and another "compatible" set $E'$ in the reduct $F_E$, then their union $E \cup E'$ should also be a "solid" set in the original AF. The authors demonstrate that many classical semantics (like preferred, grounded, and stable) and the new weak admissibility semantics satisfy this property.

Datasets

Not discussed in the paper.

Key Insights and Ideas

  • The Reduct Unifies Semantics: The paper reveals that the reduct, initially introduced for weak admissibility, is a fundamental concept that can be used to concisely characterize classical semantics like stable, admissible, preferred, and complete. This offers a new lens through which to view the relationships between different argumentation semantics.
  • Modularization as a Unifying Principle: The introduction of modularization as an abstract property for semantics is a significant contribution. It provides a powerful tool for understanding how extensions are built and how different semantics behave, showing that many standard semantics share this desirable characteristic.
  • Weak Admissibility's Place: The paper positions weak admissibility as a "liberal" semantics. While it satisfies conflict-freeness and modularization, it is less restrictive than classical admissibility, particularly in how it handles self-defeating arguments. This is formalized by showing that weak admissibility is the "least restrictive" conflict-free semantics satisfying modularization.
  • Characterizing Weakly Preferred Semantics: The paper provides a characterization of weakly preferred semantics that mirrors the characterization of classical preferred semantics, highlighting the parallels and differences between the two.

Key Results

  • Classical semantics (stable, admissible, preferred, complete) can be concisely characterized using the reduct. For instance, a set $E$ is stable if its reduct is empty, and admissible if no attacker of $E$ appears in its reduct.
  • Several classical semantics (admissible, complete, preferred, grounded, stable) satisfy the modularization property.
  • Weak admissibility semantics ($ad^w$) also satisfy modularization, and it's shown that $ad^w$ is the C-maximal semantics among all conflict-free semantics satisfying modularization.
  • Weakly preferred semantics ($pr^w$) satisfy "meaningless reduct," meaning that if $E$ is a weakly preferred extension, the reduct $F_E$ has no weakly admissible extensions. This implies $pr^w$ also satisfies modularization.

Limitations and Future Work

  • The paper is an extended abstract, so some proofs and detailed analyses are deferred to a full version.
  • The authors suggest further research into the relationship between modularization and other criteria from the literature, as well as exploring additional abstract criteria for characterizing semantics.
  • A deeper, more abstract investigation into why modularization is so prevalent across standard AF semantics is proposed, questioning if it's always linked to a specific notion of admissibility.
  • The paper mentions an analysis of strong equivalence and the odd cycle-free and acyclic fragments of AFs in the full version, indicating these are areas for further exploration.

Summary updated

BibTeX
@inproceedings{baumann2021comparing,
  author = {Ringo Baumann and Gerhard Brewka and Markus Ulbricht},
  title = {{Comparing Weak Admissibility Semantics to their Dung-style Counterparts (Extended Abstract)}},
  booktitle = {International Joint Conference on Artificial Intelligence},
  year = {2021},
  doi = {10.24963/ijcai.2021/642},
}
Figure