TL;DR
This paper introduces a unified theory of argumentation frameworks, defining argument acceptability precisely and demonstrating that various nonmonotonic reasoning and logic programming approaches are, at their core, forms of argumentation. This work provides a general method for building argumentation system metainterpreters, paving the way for more robust and integrated AI reasoning systems.
What Problem Does This Paper Address?
Humans naturally engage in argumentation to understand, express, and defend their opinions. This ability is fundamental to intelligence. However, in Artificial Intelligence, representing and reasoning with arguments, especially in the face of incomplete or conflicting information (nonmonotonic reasoning), has been approached through diverse, often disparate, formalisms. The paper highlights a gap in understanding the fundamental mechanism of argument acceptability and its relationship to existing AI reasoning techniques like logic programming. The authors aim to bridge this gap by developing a unified theory that explains how arguments are accepted and how this concept underpins various reasoning systems.
Methods
The core of the paper is the introduction of a formal argumentation framework, defined as a pair <AR, attacks>, where AR is a set of arguments and attacks represents the relationship between them (one argument attacking another). Building on this, the authors define:
- Conflict-free sets: Sets of arguments where no argument attacks another.
- Admissible sets: Conflict-free sets where every argument attacking an argument within the set is itself attacked by an argument in the set.
- Preferred extensions: Maximal admissible sets. These represent the "credulous" view of what a rational agent might accept.
- Grounded extension: The least fixed point of a characteristic function, representing the "skeptical" view.
- Complete extensions: Admissible sets that contain all arguments acceptable with respect to them.
The paper then demonstrates how existing formalisms like Reiter's default logic and logic programming can be mapped onto these argumentation frameworks, showing they are essentially different instantiations of this general concept. A key methodological contribution is the development of a general method for generating metainterpreters for argumentation systems by viewing argumentation as a form of logic programming with negation as failure.
Datasets
Not discussed in the paper.
Key Insights and Ideas
- Argumentation as a Unifying Principle: The most significant insight is that many seemingly different nonmonotonic reasoning and logic programming approaches are, in fact, different forms of argumentation. This provides a powerful unifying perspective.
- Formalizing Acceptability: The paper offers a precise, formal definition of argument acceptability, moving beyond intuitive notions and enabling rigorous analysis.
- Argumentation as Logic Programming: The insight that argumentation can be viewed as logic programming with negation as failure is crucial. It allows for the generation of metainterpreters, essentially creating a "compiler for argumentation systems."
- Skeptical vs. Credulous Reasoning: The distinction between grounded (skeptical) and preferred (credulous) extensions clearly maps to different reasoning stances, providing a formal basis for understanding these variations.
- The "Last Word" Principle: The paper implicitly formalizes the intuitive idea that the person who can successfully defend their argument against all counterarguments wins, capturing the essence of "the one who has the last word laughs best."
Key Results
- Theory of Acceptability: A formal theory defining argument acceptability is presented.
- Unification of Reasoning Formalisms: The paper demonstrates that logic programming and major nonmonotonic reasoning formalisms are specific types of argumentation systems.
- Metainterpreter Generation: A general method for generating metainterpreters for argumentation systems is introduced, leveraging the connection to logic programming.
- Semantics Correspondence: Theorems establish correspondences between argumentation semantics (stable, preferred, grounded) and the semantics of default logic and logic programming (R-extensions, stable models, well-founded models). For instance, stable models of logic programs correspond to stable extensions of their associated argumentation frameworks.
Limitations and Future Work
- Single Conflict Type: The current theory primarily handles one type of conflict (attack). The authors acknowledge that real-world argumentation often involves multiple conflict types (e.g., reductio ad absurdum, specificity). Future work should generalize the theory to handle these richer conflict scenarios.
- Efficiency: While the paper provides a theoretical foundation, practical implementation and efficiency improvements for argumentation systems are mentioned as areas for future work, suggesting the application of techniques like partial evaluation and program transformation.
- Deeper Insight into Conflicts: Further research is needed to gain deeper insights into the nature of various conflicts between arguments.
The paper concludes by emphasizing the importance of this unified framework for developing knowledge representation formalisms capable of inter-system communication and for constructing large, complex knowledge bases.
Summary updated