Acceptance semantics
An acceptance semantics is a rule for deciding which subsets of arguments in a framework can coherently "stand together." Dung's original paper defined four main ones; we'll walk through them with examples.
The building blocks
- Conflict-free set — no argument in the set attacks another in the set.
- Admissible set — conflict-free AND defends itself: every attacker of a member is itself attacked by some member.
- Complete extension — an admissible set that contains every argument it defends. (The minimum complete extension is the grounded extension; maximum complete extensions are preferred.)
- Stable extension — a conflict-free set that attacks every argument outside it.
Credulous vs skeptical
- Credulously accepted — the argument is in at least one preferred extension.
- Skeptically accepted — the argument is in every preferred extension.
Skeptical is strictly stronger than credulous. Our library exposes both via EncounterArgumentationState::is_credulously_accepted and is_skeptically_accepted.
When they diverge
Most of the time, all four semantics (grounded / preferred / stable / complete) agree on which arguments are accepted. The Nixon diamond is the canonical case where they diverge:
- Grounded:
{}(no argument is "unconditionally acceptable"). - Preferred:
{A}and{B}(two maximal admissible sets). - Stable:
{A}and{B}(both are stable). - Complete:
{},{A},{B}.
Credulous acceptance holds for both A and B (each is in some preferred extension). Skeptical acceptance holds for neither (neither is in every preferred extension).
Why we care in scene AI
Credulous acceptance is a natural fit for "is this claim live right now?" A beat can proceed as long as the proposer's argument is credulously accepted at the current β. Skeptical acceptance is too strict — it would stall scenes the moment any counter-argument exists.
Our bridge's StateActionScorer boosts affordances whose argument is credulously accepted; StateAcceptanceEval rejects only when the responder has put forward a credulously-accepted counter.
Further reading
- Dung (1995) — the foundational paper.
- Baroni, Caminada, Giacomin (2011) — a survey of semantics beyond the original four.