Source-linked AI summary
A General Account of Argumentation with Preferences
Sanjay Modgil, Henry Prakken
TL;DR
The paper addresses how a general structured argumentation framework can combine preferences with diverse logical instantiations while retaining rationality guarantees. It modifies ASPIC+, formalises Tarskian and classical logic instantiations, and argues that defeasible rules are necessary for representing forms of argumentation not captured by deductive inconsistency handling. The framework supports broader instantiations and rationality results, while its scope remains bounded by assumptions and identified limitations.
Problem
Dung’s abstract framework does not guide which instantiations satisfy rational properties, motivating a framework that supports structured arguments, preferences, and diverse logics.
Method
The paper modifies ASPIC+ by revising conflict-freeness, distinguishing attack types, broadening instantiating logics, and combining the framework with Tarskian logic approaches and preferences.
Results
The resulting framework satisfies key properties and rationality postulates under assumptions, including for Tarskian logic instantiations with and without preferences.
Takeaways & Limitations
ASPIC+ can serve as a general framework for specifying systems that use either defeasible inference or deductive approaches to model non-monotonicity as inconsistency handling.
Takeaways & Limitations
Abstract logic consequence operators cannot distinguish strict from defeasible inference, making their consistency requirement too strong for defeasible reasoning.
Abstract
from arXiv · showhide
This paper builds on the recent ASPIC+ formalism, to develop a general framework for argumentation with preferences. We motivate a revised definition of conflict free sets of arguments, adapt ASPIC+ to accommodate a broader range of instantiating logics, and show that under some assumptions, the resulting framework satisfies key properties and rationality postulates. We then show that the generalised framework accommodates Tarskian logic instantiations extended with preferences, and then study instantiations of the framework by classical logic approaches to argumentation. We conclude by arguing that ASPIC+'s modelling of defeasible inference rules further testifies to the generality of the framework, and then examine and counter recent critiques of Dung's framework and its extensions to accommodate preferences.
1 Introduction
The paper develops a more general ASPIC+ framework that combines structured argumentation with preferences while supporting broader logical instantiations. It revises conflict-freeness, establishes rationality properties under assumptions, extends the framework to Tarskian and classical logic approaches, and addresses critiques of preference-based argumentation.
- Argumentation makes reasons for conclusions and conflict resolution explicit, supporting reasoning with inconsistent or uncertain information and disagreement between agents.
- Dung’s abstract framework supports many logical instantiations but does not specify which instantiations satisfy intuitively rational properties.
- ASPIC+ introduces an intermediate abstraction with structured arguments, attacks, and preferences, enabling consistency and closure postulates beyond Dung’s abstract level.
- The paper revises ASPIC+ conflict-freeness, broadens its admissible instantiating logics, and shows the resulting framework satisfies key properties and rationality postulates under assumptions.
- It formalises Tarskian and classical logic instantiations with preferences, and examines and counters recent critiques of Dung’s framework and preference extensions.
2 Logic, Argumentation and Preferences
The section grounds preference-aware argumentation in Dung-style attacks, defeats, and extensions, while distinguishing logical incompatibility from the preference-dependent use of attacks. It motivates revised conflict-freeness and differentiates attacks on conclusions, premises, and inference steps.
- A Dung framework uses a binary attack relation, with extensions evaluating acceptable, defended, and conflict-free sets under complete, preferred, grounded, and stable semantics.
- Preference-based approaches derive defeats when an attack succeeds against an argument that is not stronger, then evaluate justified arguments using defeats instead of attacks.
- Arguments are assumed to combine strict and defeasible inference rules, with attacks restricted to fallible premises or defeasible inferences rather than deductive conclusions.
- 2.2 The Two Roles of Attacks: Attacks represent both mutual incompatibility of information and the dialectical use of one argument as a counter-argument to another.
- 2.3 Distinguishing Preference Dependent and Independent Attacks: Undercuts are preference-independent because they express a reason not to draw an inference, whereas conclusion attacks are resolved through preferences.
- 2.4 The Distinct Uses of Attacks and Defeats: The paper argues that conflict-free sets should exclude attacking arguments, not merely defeating arguments, because attacks encode logical incompatibility while defeats govern dialectical acceptability.
3 The ASPIC+ Framework
The paper modifies ASPIC+ by revising conflict-freeness and generalising the framework to support broader logical instantiations, including deductive approaches. ASPIC+ represents arguments as inference trees built from premises using strict and defeasible rules within a logical language.
- Framework revisions: ASPIC+ is revised by changing conflict-freeness and generalising the framework to capture deductive approaches to argumentation.The revised framework is intended to accommodate instantiating logics whose arguments have consistent premises.
- Arguments: Arguments are inference trees formed by applying strict or defeasible inference rules to premises that are well-formed formulae in the logical language.Premises, conclusions, sub-arguments, and applied rules are tracked for each argument.
- Argumentation systems: An ASPIC+ argumentation system consists of a logical language, a contrary relation, strict and defeasible rules, and a naming convention for defeasible rules.Strict rules have the form ϕ1, ..., ϕn →ϕ, while defeasible rules have the form ϕ1, ..., ϕn ⇒ϕ.
- Knowledge bases: ASPIC+ distinguishes axioms from ordinary premises: axioms are certain and cannot be attacked, whereas ordinary premises are uncertain and can be attacked.The knowledge base is partitioned into disjoint subsets Kn and Kp for axioms and ordinary premises.
- Argument properties: Arguments are classified as strict or defeasible according to whether they contain defeasible rules, and as firm or plausible according to whether all premises are axioms.Arguments with fallible premises or defeasible rules are fallible; finite arguments contain finitely many rules.
- Conflict-freeness: The revised framework defines attack conflict-freeness directly through the attack relation, requiring that no ordered attack exists between members of the set.For a (c-)SAF ∆=⟨A,C,⪯⟩, S is attack conflict free iff no X,Y∈S satisfy (X,Y)∈C.
4 Properties and Postulates
The paper compares attack- and defeat-based conflict-freeness in (c-)SAFs, showing that reasonable argument orderings recover key Dung properties and rationality postulates. It further establishes consistency, closure, extension equivalence, and a conceptual preference for attack-based conflict-freeness.
- Properties under the Attack Definition of Conflict Free: Under reasonable argument orderings, well-defined (c-)SAFs satisfy Dung’s fundamental lemma, so admissible extensions form a complete partial order and each is contained in a preferred extension.The fundamental lemma follows by showing that adding an acceptable argument preserves conflict-freeness and admissibility.
- Properties under the Attack Definition of Conflict Free: A reasonable argument ordering prioritizes strict and firm arguments over fallible ones and preserves preference relations when arguments are extended strictly.It also prevents every strict continuation of a set with one argument removed from being strictly less preferred than that removed argument.
- Rationality Postulates: The attack-based framework satisfies rationality postulates for complete, grounded, preferred, and stable semantics.These results apply to well-defined SAFs and c-SAFs under the attack definition of conflict free.
- Rationality Postulates: Complete extensions are closed under sub-arguments and strict inference, while admissible and complete extensions satisfy direct and indirect consistency.Sub-argument closure and strict-rule closure concern complete extensions; direct consistency concerns admissible extensions, and indirect consistency concerns strict closures of complete extensions.
- Comparison of Attack and Defeat Definitions: Admissible, complete, grounded, preferred, and stable extensions coincide under the attack and defeat definitions of conflict free.Consequently, rationality postulates established for SAFs under defeat also hold for c-SAFs under defeat.
- Comparison of Attack and Defeat Definitions: The attack definition is advocated because defeat-based conflict-free sets can contain mutually inconsistent arguments even when neither argument defeats the other.The paper therefore assumes attack conflict-freeness for subsequent extensions, while noting when results also hold under defeat conflict-freeness.
5 Instantiating Structured Argumentation Frameworks
The paper instantiates the revised ASPIC+ framework across preference orderings, abstract logics, and classical logic, establishing rationality results and equivalences under stated conditions.
- The framework adds c-SAFs with mutually consistent premises and an alternative attack-based definition of conflict-free sets.
- Preference orderings: Last-link and weakest-link argument orderings are reasonable when based on reasonable inducing set comparisons, and preserve strict-partial-order structure under corresponding assumptions.
- Abstract logic instantiations: ASPIC+ is formally instantiated by Tarskian abstract logics with preferences, assuming a symmetric contrary relation connected to Tarskian consistency.
- Abstract logic instantiations: For complete, grounded, preferred, and stable semantics, ASPIC+-based and AL-c-SAF extensions are equivalent under the defined construction.
- Abstract logic instantiations: The ASPIC+ and abstract-logic arguments themselves are not generally equivalent because abstract-logic arguments require subset-minimal premises.
- Results and rationality: The resulting c-SAFs preserve justified conclusions and satisfy closure under strict rules, direct consistency, indirect consistency, and sub-argument closure.
- Preference limitations: The non-strengthening assumption fails for some preference definitions, including democratic set comparison, but holds for elitist comparison under weakest- or last-link principles.
- Classical logic instantiations: Classical logic approaches are reconstructed as ASPIC+ special cases with preferences, and one classical instantiation is related to Brewka’s preferred subtheories.
6 A Discussion of Some Related Work
The discussion contrasts ASPIC+ with deductive and abstract-logic approaches, arguing that preferences must be applied to explicit argument structure and attacked sub-arguments. It uses examples to show how this avoids apparent inconsistency problems while retaining defeasible reasoning.
- 6.1 Comparison with General Frameworks for Argumentation: ASPIC+ is more complex than abstract logic because it combines preferences with explicit attacks on premises and defeasible inferences.Abstract logic treats inferences as certain and therefore cannot represent attacks on an argument’s internal inferential structure.
- 6.1 Comparison with General Frameworks for Argumentation: The Nixon example shows that an extension can remain indirectly consistent while the corresponding abstract logic is inconsistent because defeasible conclusions generate arguments outside the extension.The discrepancy arises when closure combines conflicting defeasible arguments even though those arguments are not jointly justified.
- 6.1 Comparison with General Frameworks for Argumentation: Abstract-logic consistency can be too strong because its consequence operator conflates strict and defeasible inference.Defeasible rules may be defeated even when their antecedents hold, so knowledge need not be closed under them.
- 6.1 Comparison with General Frameworks for Argumentation: The abstract-logic approach generalises classical argumentation but does not apply to mixed strict and defeasible argumentation.The paper nevertheless assigns deductive approaches a legitimate role while maintaining that a general account should also accommodate defeasible rules.
- 6.2 Comparison with other works on Preference-based Argumentation: In the reported examples, ASPIC+ changes the outcome by comparing preferences on attacked sub-arguments, preventing extensions that would otherwise contain contradictory conclusions.This changes the classical PAF outcome and makes the resulting extension satisfy consistency.
- 6.2 Comparison with other works on Preference-based Argumentation: The paper argues that PAF problems are repaired by making argument structure and attack type explicit, rather than by changing abstract definitions.Preference comparisons should target the attacked sub-argument, while undercutting and contrary attacks can remain preference independent.
7 Conclusions
The conclusion presents ASPIC+ as an intermediate framework combining Dung’s dialectical theory with structured arguments, attacks, and preferences. It argues that the framework supports deductive and defeasible instantiations and that argument structure is central to modelling preferences, while identifying several directions for further study.
- 7 Conclusions: ASPIC+ combines Dung’s theory with structured arguments, attacks, and preferences, allowing rationality postulates to be studied across concrete instantiating logics.The framework is intended to accommodate both deductive approaches and defeasible inference rules.
- 7 Conclusions: The paper reports that broader Tarskian instantiations and preference orderings satisfy assumptions needed for the framework’s stated properties and rationality postulates.It also addresses limitations in earlier definitions of argument orderings.
- 7 Conclusions: The conclusion’s central rhetorical claim is that properly modelling preferences requires taking argument structure into account.The authors support this claim through the paper’s results and its discussion of critiques of Dung-style and preference-based frameworks.
- 7 Conclusions: Future work includes specifying additional systems, structuring extended argumentation, studying non-interference and crash resistance, and evaluating alternative preference orderings.These directions extend the framework’s analysis beyond the systems and weakest- or last-link orderings considered here.
8 Appendix
The appendix establishes structural and rationality properties of preference-aware argumentation frameworks, then proves correspondences with abstract-logical and default-theoretic instantiations.
- Complete extensions are closed under sub-arguments and strict rules, while admissible extensions are directly consistent and complete extensions are indirectly consistent.
- Attack-based and defeat-based definitions yield the same admissible, complete, grounded, preferred, and stable extensions.
- Last-link and weakest-link preference principles preserve reasonableness, and their induced orderings are strict partial orders under the stated comparison conditions.
- Classical and abstract-logical instantiations preserve key properties and extension correspondences, including consistency, justified conclusions, and equivalence across premise-minimal and full frameworks.
- For default theories, preferred subtheories correspond to stable extensions: arguments from a preferred subtheory form a stable extension, and every stable extension induces a preferred subtheory.