[Paper] Constrained Assumption-Based Argumentation Frameworks

Source: arXiv - 2602.13135v1

Overview

The paper introduces Constrained Assumption‑Based Argumentation (CABA), a new extension of the classic Assumption‑Based Argumentation (ABA) framework that lets arguments contain variables and constraints over potentially infinite domains. By moving beyond the traditional “ground‑only” (variable‑free) setting, CABA opens the door to richer, more expressive models that can capture real‑world reasoning tasks such as scheduling, configuration, and policy compliance.

Key Contributions

• Variable‑enabled ABA: Formal definition of ABA components (assumptions, rules, attacks) that may contain constrained variables.
• Non‑ground semantics: Introduction of several notions of attack (e.g., universal, existential, mixed) that work with variable‑bearing arguments.
• Conservative generalisation: Proof that when all variables are instantiated (i.e., the framework is ground), CABA collapses to standard ABA semantics, guaranteeing backward compatibility.
• Theoretical analysis: Formal properties (e.g., monotonicity, consistency) of the new semantics and relationships among the different attack notions.
• Illustrative examples: Demonstrations of how CABA can model problems that are cumbersome or impossible to express in traditional ABA (e.g., resource allocation with numeric constraints).

Methodology

1. Language Extension: The authors start from the usual ABA language (atoms, rules, assumptions) and augment it with constrained variables—variables that range over a domain together with logical constraints (e.g., x > 0, x ∈ ℕ).
2. Argument Construction: An argument is a tree of rule applications that may leave some variables uninstantiated. The leaf nodes are assumptions, possibly containing variables.
3. Attack Definitions: Three families of attacks are defined:
   • Universal attack: An argument A attacks B if every grounding of A’s variables defeats some grounding of B.
   • Existential attack: A attacks B if there exists some grounding of A that defeats some grounding of B.
   • Mixed attack: Combines universal and existential conditions to capture more nuanced interactions.
4. Semantics Transfer: Classic ABA semantics (e.g., admissible, preferred, stable extensions) are lifted to the non‑ground setting by interpreting them over the new attack relations.
5. Conservativity Proof: By restricting the variable domains to a singleton set, the authors show that CABA semantics coincide exactly with the original ABA semantics.

Results & Findings

• Expressiveness Boost: CABA can encode problems involving arithmetic, ordering, and other constraints without flattening them into propositional atoms, dramatically reducing the size of the underlying knowledge base.
• Semantic Preservation: The new attack notions respect the core properties of ABA (e.g., conflict‑freeness, defense) while allowing reasoning over infinite families of arguments.
• Computational Trade‑offs: While the framework is more expressive, the authors note that checking attacks may require solving constraint satisfaction problems (CSPs), which can be NP‑hard or worse depending on the constraint language.
• Case Studies: Sample encodings of a simple job‑shop scheduling scenario and a role‑based access control policy demonstrate that CABA yields more natural models and can be evaluated using existing CSP/SMT solvers.

Practical Implications

• AI Planning & Scheduling: Developers can model planning domains with numeric resources directly in an argumentation engine, delegating the heavy lifting to off‑the‑shelf solvers.
• Policy & Compliance Engines: Complex regulatory rules often involve thresholds and ranges; CABA lets you capture these as constraints rather than enumerating every possible case.
• Explainable AI (XAI): Argumentation‑based explanations can now reference variable‑rich premises (e.g., “Because the load x exceeds 80 %”) making them more intuitive for end‑users.
• Integration Path: Existing ABA tools can be extended with a preprocessing layer that translates constrained rules into ground instances on demand, enabling a gradual migration to CABA without rewriting the whole system.

Limitations & Future Work

• Scalability: The reliance on external CSP/SMT solvers means that performance hinges on the difficulty of the underlying constraints; large or highly non‑linear domains may become intractable.
• Tooling Gap: No dedicated CABA implementation exists yet; the paper provides a theoretical foundation but leaves engineering of efficient solvers to future work.
• User Guidance: Selecting the appropriate attack notion (universal vs. existential) can be non‑trivial for practitioners; guidelines or automated heuristics are needed.
• Future Directions: The authors suggest exploring hybrid semantics that combine ground and non‑ground reasoning, optimizing constraint handling (e.g., incremental solving), and applying CABA to domains like security analysis, legal reasoning, and multi‑agent negotiation.

Authors

• Emanuele De Angelis
• Fabio Fioravanti
• Maria Chiara Meo
• Alberto Pettorossi
• Maurizio Proietti
• Francesca Toni

Paper Information

• arXiv ID: 2602.13135v1
• Categories: cs.AI, cs.LO
• Published: February 13, 2026
• PDF: Download PDF