TL;DR
This paper bridges the gap between theoretical non-monotonic reasoning (Preferred Subtheories) and practical argumentation systems. It proves that preferred and stable semantics for argumentation frameworks coincide, meaning existing proof theories for admissible extensions can now be used for credulous inference in Preferred Subtheories. Furthermore, a new dialectical approach to argumentation offers a more nuanced, less "sceptical" inference for agents with limited reasoning capabilities.
What Problem Does This Paper Address?
Non-monotonic reasoning deals with drawing conclusions that can be retracted in light of new information. "Preferred Subtheories" is a formal framework for this, often dealing with inconsistent knowledge bases by selecting consistent subsets of beliefs. Argumentation frameworks (AFs) provide a way to model this by representing knowledge as arguments that attack each other. However, directly applying standard argumentation frameworks to Preferred Subtheories has limitations: it assumes agents can construct all possible arguments (often infinite) and requires checking for premise consistency, which is computationally infeasible for real-world agents. This paper addresses how to make these characterizations practical and rational, especially for agents with bounded reasoning capabilities.
Methods
The paper builds upon two main approaches to argumentation:
- Standard Classical Logic Argumentation (Cl-Arg): This approach uses Dung's abstract argumentation framework where arguments are defined by premises and conclusions, and attacks are based on logical negation. Preferences between arguments are introduced to define "defeats." The paper proves that for this standard approach, the preferred extensions (sets of arguments that are maximal consistent and acceptable) of an AF derived from a default theory are equivalent to the stable extensions (conflict-free sets where every non-member is attacked by a member).
- Dialectical Cl-Arg: This novel approach, introduced by the authors in prior work, distinguishes between an argument's premises (assumed true) and suppositions (assumed true "for the sake of argument"). This distinction allows for more flexible reasoning and avoids the "foreign commitment problem" in dialogues where agents don't have to commit to the truth of their interlocutor's premises. The paper shows that for this dialectical approach, preferred and stable extensions also coincide.
Datasets
Not discussed in the paper. The paper uses theoretical examples to illustrate its concepts rather than empirical datasets.
Key Insights and Ideas
- Preferred = Stable: The core insight is proving that for both standard and dialectical argumentation frameworks derived from Preferred Subtheories, the set of preferred extensions is identical to the set of stable extensions. This is a significant theoretical result.
- Practical Implications for Proof Theories: Because preferred extensions are now shown to be equivalent to stable extensions, existing proof theories and algorithms designed for admissible extensions (a weaker notion than preferred/stable) can be directly applied to credulous inference in Preferred Subtheories. This opens the door for more efficient reasoning.
- Dialectical Approach for Bounded Agents: The dialectical approach, by separating premises from suppositions, offers a more rational and less "sceptical" inference for agents with limited computational resources. It avoids forcing agents to commit to premises they don't necessarily believe.
- Grounding the Dialectical Approach: The paper demonstrates that the grounded semantics (which aims for the least sceptical conclusion) applied to the dialectical approach more closely approximates sceptical inference in Preferred Subtheories compared to the standard approach. This is achieved by allowing arguments to be defeated based on suppositions, rather than solely on premises.
Key Results
- Theorem 3 & Proposition 7: Show that for standard Cl-Arg, preferred extensions coincide with stable extensions.
- Proposition 9: Shows that for dialectical Cl-Arg, preferred extensions coincide with stable extensions.
- Correspondence: These results establish that credulous Preferred Subtheories inference corresponds to conclusions found in at least one stable (and now preferred) extension of the instantiated AF.
- Grounding Advantage: The dialectical approach, under grounded semantics, provides a less sceptical inference that better aligns with the desired outcome for practical applications, avoiding issues like the "foreign commitment problem."
Limitations and Future Work
- Argument Game Adaptation: While proof theories for admissible extensions can now be used for credulous inference, adapting argument games (which typically rely on admissible extensions) to the dialectical framework requires future work. Specifically, how to parameterize defeats to reference the set of arguments already committed to by the proponent.
- Sceptical Inference and Grounded Semantics: While the dialectical approach improves the alignment of grounded semantics with sceptical inference, a full correspondence doesn't always hold due to syntactic sensitivities in non-monotonic reasoning. The authors conjecture this failure is limited to "pathological" cases involving conjunctions of syntactically unrelated formulas.
- Generalizing to Partial Orderings: The current work focuses on total orderings for Preferred Subtheories. Future work could extend these characterizations to partial orderings.
Summary updated