Evidential-Based Higher-Order Set Argumentation Framework

Motivation

• Existing evidence-based formalisms cannot jointly represent evidential support, higher-order targets, collective sources, and unified logical or fuzzy semantics.Higher-order and set-based generalizations exist, but principled labelling analyses remain restricted to first-order bipolar frameworks.Source: §1 · pp. 2–4

EHSAF framework

• EHSAF integrates evidential support chains, higher-order attacks and supports, and collective sources, generalizing HSAF and extending REBAF.Interactions may target arguments, attacks, or supports, while their sources may be sets of arbitrary elements.Source: §1 · p. 4

Semantics

• Adjacent complete labelling semantics tolerates unsupported arguments in evidential support cycles, whereas extension-based complete semantics rejects arguments lacking current prima-facie support.The adjacent semantics enumerates true, false, and undecided possibilities when future evidence may emerge, while the strict semantics adopts an evidentialist stance.Source: §1 · pp. 4–5
• The proposed labelling and extension semantics conservatively generalize EAS, REBAF, and HSAF, with equivalence between extension and labelling semantics under acyclicity conditions.The paper also introduces adjacent complete, stable, preferred, and grounded labelling semantics for EHSAFs.Source: §1 · pp. 5–6

Propositional encoding

• EHSAF admits a normal propositional encoding whose three-valued Łukasiewicz models correspond exactly to adjacent complete labellings.This establishes a computational and logical foundation for reasoning over the unified framework.Source: §1 · pp. 5–6

Fuzzy encoding

• Continuous fuzzy normal encoded semantics satisfy continuity, monotonicity, boundary conditions, and solution existence across Gödel, Product, and Łukasiewicz t-norms.The instantiated fuzzy semantics correspond to the three-valued adjacent complete semantics under the paper’s stated conditions.Source: §1 · pp. 5–6

2.1.1 Syntax and Semantics of EASs

• EASs represent arguments with attacks and supports, requiring validity to trace through support chains to the environmental prima-facie argument η.The revised definition treats η as trivially e-supported, resolving deadlocks in support chains.Source: §2.1.1 · pp. 6–7

2.1.2 Syntax and Semantics of REBAFs

• REBAFs generalize evidential frameworks by allowing prima-facie arguments, attacks, and supports, collective sources, and targets that may be any such element.Their semantics preserve standard relationships and conservatively generalize Dung-style frameworks when supports are absent and all elements are prima-facie.Source: §2.1.2 · pp. 7–9

2.2 Propositional Logic Systems

• The logic preliminaries define three-valued Łukasiewicz logic and [0,1]-valued fuzzy logics based on standard negation, continuous t-norms, and residuated implications.The paper fixes Gödel, Product, and Łukasiewicz fuzzy systems and notes that zero-divisor-free t-norms preserve key implication properties.Source: §2.2 · pp. 10–12

3.1 Syntax of EHSAFs

• EHSAFs extend preparatory frameworks with higher-order attack and support relations over arguments, relation names, and source sets, while adding ⊥, ⊤, and prima-facie elements.The construction adds default attacks or supports for otherwise unconnected elements and treats relation names and supports as prima-facie where specified.Source: §3.1 · p. 12

3.2 Basic Semantics of EHSAFs

• Labellings assign values in [0, 1] while fixing ⊥ at 0, ⊤ at 1, and inherently prima-facie elements at 1; equational systems encode one equation for each universal-set element.Each equation depends on elements occurring in source sets of attacks or supports targeting the element, together with the corresponding relation names.Source: §3.2 · pp. 12–13

3.2.1 Labelling-Based Semantics of EHSAFs

• Three-valued equational semantics exactly coincides with adjacent complete labelling semantics for every EHSAF.The three possible values represent acceptance, rejection, or the otherwise case when neither the acceptance nor rejection condition holds.Source: §3.2.1 · pp. 14, 16

3.2.2 Extension-Based Semantics of EHSAFs

• Extension-based semantics derives elements through finite support trees rooted at the target, with non-root nodes in the extension and leaves among prima-facie elements.The support operator satisfies x ∈ Sup(S) iff a finite support tree for x exists, and it is monotone and continuous with a least fixed point.Source: §3.2.2 · pp. 17–18
• Extension-based semantics define complete extensions as conflict-free fixed points of acceptance, preferred extensions as maximal admissible structures, and grounded extensions as minimal complete structures.Support is computed by a monotone operator whose least fixed point exists and stabilizes after at most |U| iterations because U is finite.Source: §3.2.2 · pp. 19–20
• Every extension-based complete labelling is an adjacent complete labelling, while auxiliary arguments receive the same fixed labels in both semantics.Thus, extension-based semantics is at least as restrictive as adjacent semantics, with the theorem established by checking all three labelling cases.Source: §3.2.2 · pp. 20, 21, 23
• Support-acyclicity assigns strictly decreasing ranks along support chains, ensuring that every supported element ultimately derives from prima-facie elements without support cycles.Under this condition, positive adjacent labels imply evidential support and remain supported when the candidate set expands within U \ Def_a(E).Source: §3.2.2 · p. 24
• For support-acyclic EHSAFs, positively labelled elements form conflict-free sets, while generally any negatively labelled element is defeated through an attack or unsupported support condition.The conflict-freeness results distinguish full defeat, attack defeat, and the additional condition E ⊆ Sup(E).Source: §3.2.2 · p. 25
• Under support-acyclicity, every adjacent complete labelling is extension-based complete, and the two semantics are equivalent.The result follows by establishing that defeat implies negative labelling and that the accepted set equals the complete extension.Source: §3.2.2 · pp. 27–28
• The support-cycle example has exactly three adjacent complete labellings, denoted L1, L0, and L1/2; support names t1 and t2 are labelled 1 in every labelling.For L1, a and b are 1; for L0, a and b are 0.Source: §3.2.2 · p. 29
• For the same cycle, extension-based semantics has a unique labelling that assigns 0 to both cyclic arguments and 1 to both support names.This coincides with the pessimistic adjacent labelling because the cyclic arguments cannot be grounded in a complete extension.Source: §3.2.2 · p. 30
• Thus, extension-based semantics takes a strict evidentialist stance, while adjacent labelling semantics remains open to future confirmation, refutation, or undecidability.Consequently, the equivalence claim fails when the support relation contains cycles.Source: §3.2.2 · p. 30

• For an EHSAF, the propositional language uses constant set Con and variable set Var, with U ⊆ Con ∪ Var.Source: §4 · p. 31
• The constant set contains ⊥ and ⊤, interpreted as 0 and 1, respectively.Source: §4 · p. 31
• Elements in (Re ∪ Re) \ (Ra ∪ Re) are also included as constants and assigned truth value 1.Source: §4 · p. 31

Encoding and model semantics

• An EHSAF encoding is a propositional formula whose variables and constants coincide exactly with the framework’s universal set U.The encoding is a function from EHSAFs to formulas in a chosen propositional logic.Source: §4.1 · p. 31
• Encoded semantics treats models of the encoded formula as labellings of the EHSAF under the selected propositional logic.The induced semantics is defined through an encoding function and its models.Source: §4.1 · p. 31

Skeptical and credulous semantics

• Skeptical semantics assigns each element its minimum truth value across all basic-encoded models, yielding a unique intersection-based labelling.This captures the infimum of possible interpretations.Source: §4.1 · p. 31
• Credulous semantics assigns each element its maximum truth value across all basic-encoded models, yielding a unique union-based labelling.This captures the supremum of possible interpretations.Source: §4.1 · p. 31

Normal encoded semantics

• Normal encoding is defined as a specialized EHSAF encoding, with normal encoded semantics induced by the corresponding propositional-logic models.The section introduces both the normal encoding function and its associated semantics.Source: §4.1 · p. 32

Discrete Encoded Semantics

• Theorem 4 identifies 3 ecn(EHSAF) with LSac(EHSAF) for every EHSAF.This equality is the section’s central discrete-semantics correspondence.Source: §4.2 · p. 32
• Corollary 3 follows immediately from Theorem 1 and the preceding theorem.Source: §4.2 · p. 32
• The section also states Theorem 5 for every EHSAF, although its specific conclusion is not included in the supplied passage.Source: §4.2 · p. 32

Continuous Fuzzy Operator-Based Equational (CFOE) Semantics

• CFOE semantics assigns [0,1]-valued elements satisfying continuous t-norm, negation, and t-conorm equations built from attacks and supports.Solutions of the CFOE system induce the continuous fuzzy operator-based equational semantics.Source: Continuous Fuzzy Operator-Based… · pp. 32–33

Continuous Fuzzy Normal Encoded (CFNE) Semantics

• CFNE semantics is induced by the EHSAF normal encoding interpreted in a fuzzy propositional logic with continuous negation, t-norm, and residuated implication.It provides the normal-encoded counterpart to the CFOE construction.Source: Continuous Fuzzy Normal Encoded (CFNE)… · p. 33

Equivalence between CFOE and CFNE Semantics

• CFOE and CFNE semantics are equivalent for every EHSAF, so assignments satisfying the equational system are exactly models of the normal encoding.The equivalence follows by evaluating fuzzy connectives through the continuous negation, t-norm, and induced t-conorm.Source: Equivalence between CFOE and CFNE… · pp. 33–34

4.3.2 Core Properties of CFNE Semantics

• For every β, fβ is continuous, monotone non-increasing in attack-related elements and non-decreasing in support-related elements, satisfies extremal boundary conditions, and exhibits trifurcation when its t-norm is zero-divisor-free.Source: §4.3.2 · pp. 34–37

4.3.3 Relationships between CFNE Semantics and 3-Valued Semantics

• With a continuous zero-divisor-free t-norm, ternarizing any CFNE model produces an adjacent complete labelling.The ternarization maps fuzzy values 1, 0, and intermediate values to the corresponding three-valued labels.Source: §4.3.3 · pp. 38–39
• Every adjacent complete labelling satisfies the CFNE encoding F; for each β, ∥β∥ = Aβ⊙Bβ, and auxiliary elements and added attacks/supports satisfy the system.Source: §4.3.3 · pp. 40, 43
• Under a continuous 1/2-idempotent, zero-divisor-free t-norm with standard negation, ternarised CFNE models are exactly the adjacent complete labellings.The result states {∥·∥ | ∥·∥ |= LS_ac F} = {T3(∥·∥) | ∥·∥ |= CFNE F}.Source: §4.3.3 · p. 43

4.4 Key Instances of Fuzzy Encoded Semantics

• The three canonical instances use Gödel, Product, and Łukasiewicz t-norms to represent distinct aggregation patterns, with equational and encoded semantics equivalent by Theorem 6.The instances inherit the unified CFOE/CFNE framework and are paired with algebraic interpretations.Source: §4.4 · p. 43

4.4.1 G¨odel Fuzzy Encoded Semantics

• Gödel semantics applies a weakest-link rule: collective attacks use minimum source strength, supports use maximum successful support, and target acceptability combines these by minimum.It extends discrete max-min semantics to [0, 1]; the Gödel CFOE and CFNE semantics coincide, and every EHSAF has at least one solution.Source: §4.4.1 · pp. 44–45

4.4.2 Product Fuzzy Encoded Semantics

• Product semantics models mutually independent attacks and supports through multiplication, so multiple attacks cumulatively weaken acceptability while multiple supports cumulatively enhance it.Product CFOE and CFNE semantics coincide; continuity, commutativity, boundary conditions, monotonicity, and solution existence are guaranteed.Source: §4.4.2 · pp. 46–47

4.4.3 Lukasiewicz Fuzzy Encoded Semantics

• Łukasiewicz semantics provides threshold-based additive aggregation for competitive or mutually exclusive attacks and supports, with full effects emerging only after total strength exceeds a threshold.Its encoded form is obtained by substituting the Łukasiewicz t-norm and is equivalent by Theorem 6, but detailed derivation is omitted because the expression is cumbersome.Source: §4.4.3 · p. 47

5.1 Dung’s Abstract Argumentation and Basic Extensions

• EHSAF conservatively generalizes Dung’s framework, exactly recovering its extension-based semantics when support is empty, elements are prima-facie, sources are singleton arguments, and targets are arguments.Source: §5.1 · p. 48

5.2 Evidential Support Frameworks

• EHSAF extends evidential frameworks and REBAF by allowing arbitrary-element sources and higher-order targets, while recovering their semantics under the corresponding source and target restrictions.For REBAF, restricting sources to arguments makes EHSAF’s structure semantics coincide with REBAF’s; restricting EHSAF to argument targets and argument sources recovers EAS semantics.Source: §5.2 · p. 48

5.3 Higher-Order and Collective Attack Frameworks

• EHSAF extends HSAF [28] with evidential support: with empty support and all elements prima-facie, its semantics and logical encoding reduce exactly to HSAF.Source: §5.3 · p. 49

5.4 Logical Encodings of Argumentation Frameworks

• EHSAF’s propositional encoding provides a SAT-compatible counterpart to prior first-order encodings, conservatively recovering REBAF and HSAF encodings while uniformly handling arbitrary-element sources and higher-order attacks and supports.Its fuzzy encoding additionally supplies a numerical semantics without a direct counterpart among the cited first-order encodings.Source: §5.4 · p. 50

5.5 Fuzzy and Gradual Argumentation Semantics

• EHSAF’s fuzzy semantics extends prior gradual approaches with parameter-free, continuous and monotone aggregation derived directly from logical encodings, while recovering complete semantics on the three-valued domain under the Gödel t-norm.It also supports arbitrary-element sources, uniform higher-order attacks and supports, and systematic construction of new equational semantics through alternative fuzzy operations [27].Source: §5.5 · pp. 50–51

• EHSAF unifies evidential support, higher-order interactions, and collective sources while conservatively generalising EAS, REBAF, and HSAF.Evidence is rooted in prima-facie elements, interactions may target arbitrary elements, and sources are sets of arbitrary elements.Source: §6 · p. 51
• Adjacent and extension-based complete semantics diverge on support cycles but coincide under support-acyclicity.The adjacent semantics permits true, false, or undecided values, whereas the extension-based semantics requires well-founded support chains.Source: §6 · pp. 51–52
• Three-valued Łukasiewicz models of the normal propositional encoding correspond exactly to adjacent complete labellings.This encoding provides a computational foundation for EHSAFs.Source: §6 · p. 52
• Continuous fuzzy encodings over Gödel, Product, and Łukasiewicz logics satisfy continuity, monotonicity, boundary conditions, and solution existence.Under zero-divisor-free and 1 2-idempotent t-norms, ternarisation recovers adjacent complete labellings, and every adjacent labelling is a ternary fuzzy model.Source: §6 · p. 52
• Future work targets incremental computation, preference-based or weighted extensions, and applications to legal reasoning and scientific evidence evaluation.These applications are motivated by the epistemic distinction between current and prospective evidence.Source: §6 · p. 52