On Structured Argumentation with Conditional Preferences

Phan Minh Dung, Phan Minh Thang, Tran Cao Son

Abstract

Structured argumentation provides a formal framework for reasoning with conflicting information. However, traditional systems often assume static preferences between rules. This paper extends these frameworks to handle conditional preferences—where the priority of rules depends on the specific context. By defining a set of "regular properties" for attack relations and proving the existence of a canonical conflict resolution method, the authors provide a logically sound framework for defeasible reasoning with situation-dependent priorities.

Introduction: The Challenge of Contextual Priority

In classical logic, statements are strictly true or false. Defeasible reasoning allows for "defaults" (e.g., "birds fly") that can be overturned by specific evidence ("Tweety is a penguin"). Modern systems like ASPIC+ use preferences to decide which argument prevails in a conflict. Yet, in real-world scenarios, preferences are rarely static. For instance, one might prefer taking the bus over driving, unless it is raining.

When preferences themselves are conditional and subject to debate, defining "attacks" between arguments becomes complex. Without a formal framework, different systems may yield inconsistent or counter-intuitive results. This research addresses this gap by providing an axiomatic foundation for meta-level argumentation, where arguments can target the validity of other attack relations based on context.

Formalizing Conditional Preferences

The authors propose an axiomatic approach to define what constitutes a "good" preference-based argumentation system.

  1. Defeasible Knowledge Bases (DKB): A DKB is expanded to include strict rules, defeasible rules, and preference rules (p-rules).
  2. Preference-Based Argumentation Framework (PAF): This framework allows "preference arguments" to attack the attack relation itself. In this meta-level view, an argument $P$ does not just support a conclusion; it can attack the fact that Argument $A$ successfully defeats Argument $B$.
  3. Regular Properties: The paper identifies five "regular properties" that any preference-based attack relation should satisfy:
  4. Inconsistency-Resolving: Conflicts must stem from logical contradictions.
  5. Effective Rebuts: Preferences must correctly identify and eliminate the "weaker" side of a conflict.
  6. Context-Independence: The validity of an attack should rely on the structure of the arguments rather than the specific knowledge base.
  7. Attack Monotonicity: If a preference eliminates a strong attack, it must also eliminate weaker variants.
  8. Link-Orientation: Attacks should be directed at the specific "culprit" links within an argument's structure.

Illustrative Case: The Sherlock Holmes Mystery

To demonstrate the framework, the authors use an extended murder mystery example involving multiple suspects ($P_1, P_2, S$) and conflicting evidence regarding motives and innocence.

Figure 6: A structured argumentation graph representing a complex murder mystery with conditional preferences and motives.

In this scenario, arguments for innocence ($A_1, A_2, A_s$) are challenged by counter-arguments ($N_1, N_2, N_s$). Preferences ($X_1$) based on the strength of evidence ($d_1 < d_2$) determine which arguments prevail. If a motive ($M_1$) is established, it may defeat a preference for innocence, showing how meta-level arguments dynamically reshape the conflict graph.

Key Theoretical Insights

The biggest contribution is treating preferences as defeasible arguments. If a preference rule is based on an assumption that is later proven false, the preference itself is defeated.

The authors prove that the set of all "regular" attack assignments forms a complete lower semilattice. This mathematical structure ensures the existence of a unique "least" element: the Canonical Preference Attack Relation. This canonical assignment follows the Minimal Removal Principle, removing only the absolute minimum number of attacks necessary to satisfy the logical axioms, thereby preserving the maximum amount of original conflict structure.

Results and Validation

The proposed framework is a conservative generalization of existing systems; if complex preferences are removed, it behaves identically to standard, well-accepted argumentation models. Furthermore, the authors define a "Normal" assignment that is computationally simpler and prove it is identical to the Canonical assignment in "p-stratified" systems.

Rigorous proofs confirm that any system satisfying the five regular properties will adhere to fundamental argumentation postulates, such as consistency (avoiding $A$ and $\neg A$ simultaneously) and closure (concluding everything that logically follows from accepted arguments).

Conclusion and Future Work

While providing a robust theoretical foundation, the paper notes that the computational complexity of finding the Canonical assignment in very large knowledge bases remains an open challenge. Future research may focus on developing efficient algorithms for these calculations and integrating this framework with probabilistic or fuzzy logic to handle non-binary preferences.

Summary updated

Abstract

We study defeasible knowledge bases with conditional preferences (DKB). A DKB consists of a set of undisputed facts and a rule-based system that contains different types of rules: strict, defeasible, and preference. A major challenge in defining the semantics of DKB lies in determining how conditional preferences interact with the attack relations represented by rebuts and undercuts, between arguments. We introduce the notions of preference attack relations as sets of attacks between preference arguments and the rebuts or undercuts among arguments as well as of preference attack relation assignments which map knowledge bases to preference attack relations. We present five rational properties (referred to as regular properties), the inconsistency-resolving, effective rebuts, context-independence, attack monotonicity and link-orientation properties generalizing the properties of the same names for the case of unconditional preferences. Preference attack relation assignment are defined as regular if they satisfy all regular properties. We show that the set of regular assignments forms a complete lower semilattice whose least element is referred to as the canonical preference attack relation assignment. Canonical attack relation assignment represents the semantics of preferences in defeasible knowledge bases as intuitively, it could be viewed as being uniquely identified by the regular properties together with the principle of minimal removal of undesired attacks. We also present the normal preference attack relation assignment as an approximation of the canonical attack relation assignment.
BibTeX
@inproceedings{dung2019structured,
  author = {Phan Minh Dung and Phan Minh Thang and Tran Cao Son},
  title = {{On Structured Argumentation with Conditional Preferences}},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {2019},
  doi = {10.1609/aaai.v33i01.33012792},
}
Figure