Abstract
Conditional Abstract Dialectical Frameworks (cADFs) represent a significant generalization of standard Abstract Dialectical Frameworks (ADFs). By allowing acceptance conditions to be assigned to complex propositional formulas rather than just atomic arguments, cADFs can model sophisticated scenarios such as attacks on sets of arguments and disjunctive logic programming. This paper establishes the formal foundations of cADFs, introducing a non-truth-functional four-valued logic and a generalized operator to handle indeterminism and complex dependencies.
Introduction
In formal argumentation, knowledge is typically represented as a graph where nodes are arguments and edges denote relations like attack or support. Abstract Dialectical Frameworks (ADFs) advanced this by introducing Boolean acceptance conditions for each argument, describing which combinations of neighbors justify its acceptance. However, ADFs are limited to assigning conditions to atomic arguments. This constraint makes it difficult to represent "attacks on sets" (e.g., "you cannot accept both $p$ and $q$ if $r$ is true") or disjunctive outcomes (e.g., "either $a$ or $b$ is true"). Conditional Abstract Dialectical Frameworks (cADFs) address these limitations by allowing rules of the form $\phi \leftarrow \psi$, where both the statement $\phi$ and the condition $\psi$ can be complex propositional formulas.
Methodology
The authors build the mathematical foundation for cADFs using several novel components designed to handle the increased expressiveness and potential indeterminism of the framework.
Syntax and 4CM Logic
A cADF is defined as a set of pairs $(\phi, \psi)$, interpreted as "$\phi$ is the case if and only if $\psi$ is the case." To handle the resulting complexity, the authors move from standard 3-valued logic to a four-valued logic (4CM) adding "Inconsistent" (I) to the set {True (T), False (F), Undecided (U)}. Crucially, this logic is non-truth-functional, meaning the truth value of a complex formula preserves classical tautologies (e.g., $p \lor \neg p$ is always True) even when components are Undecided.
The Generalized $\Gamma_\Pi$ Operator
The fundamental operator of ADFs is generalized in cADFs. Instead of mapping an interpretation to a single successor, the $\Gamma_\Pi$ function maps an interpretation to a set of interpretations that align with the given rules. This allows the framework to naturally represent "choice" or indeterminism in reasoning.
Semantics
Standard argumentation semantics—admissible, complete, preferred, and grounded—are redefined to work within this new, non-deterministic framework. Because cADFs allow indeterminism, they might not have a unique minimal model. The authors introduce the "grounded state"—a set of interpretations that collectively represent the minimal information entailed by the framework.
Key Insights and Results
- Indeterminism by Design: A rule like $(p \lor q) \leftarrow \top$ does not force $p$ or $q$ to be true individually, allowing cADFs to model disjunctive logic programming naturally.
- Handling Conflicts: The four-valued logic allows the framework to gracefully handle "overspecified" systems where rules might contradict each other, assigning them the value "Inconsistent" rather than causing logical failure.
- Generalization: The paper proves that cADFs are a proper generalization; applying cADF machinery to a standard ADF yields identical results to the original formalism.
- Implementation: These semantics have been implemented in the Tweety library, a popular Java framework for artificial intelligence and logic.
Conclusion and Future Work
cADFs provide a powerful extension to the argumentation landscape, enabling the representation of complex logical dependencies previously out of reach. While existence proofs for major semantics are established, future research is needed to define stable semantics, map out computational complexity, and explore the realizability of arbitrary model sets within the cADF framework.
Summary updated