What is argumentation?

Formal argumentation is the study of how arguments attack and support each other, and which sets of arguments can stand together without contradicting themselves.

It sounds abstract, but it's the formal machinery behind auditable reasoning. When a courtroom jury decides which testimony to believe, or when a scientific review board weighs competing studies, they're doing argumentation informally. Formal argumentation gives you primitives — argument, attack, support, acceptance — that a computer can reason about.

The mental model

Picture a knowledge base as a graph. Nodes are arguments — conclusions backed by reasoning. Edges are attacks — one argument undermining another. Acceptance semantics tell you which subsets of arguments can coherently stand together, given the attacks.

The library treats arguments as abstract by default — opaque IDs in a graph, no internal structure. This is Dung (1995). When you need to look inside arguments (to attack a premise rather than a conclusion, say), the argumentation-schemes and ASPIC+ crates give you that structure.

What argumentation is not

A few common confusions worth heading off:

• Not natural-language understanding. This library doesn't parse sentences or extract arguments from text. You give it argument IDs; it tells you which sets are coherent.
• Not probabilistic reasoning. Acceptance is a yes/no question (per semantics). Probabilistic AF is a separate research line — see open areas.
• Not theorem proving. Argumentation handles defeasible reasoning — conclusions that can be defeated by counter-arguments, even when each step looks valid in isolation. Classical logic doesn't admit defeat.
• Not a chatbot. No language generation. The library is the reasoning scaffold underneath; a language model on top is a separate concern.

Why it matters for scene AI

Formal argumentation gives you a structural layer for scenes: a graph of arguments and attacks, a semantics for deciding which stand. In a scene, each beat carries a record of:

• The arguments asserted by whom.
• The attacks that bound under the current scene tension.
• The acceptance semantics that produced the beat outcome.

You can replay the scene deterministically, tune one parameter (β, scene intensity) and see the beats change in a transparent way, or combine this with other tools — language models for surface prose, handwritten branches for pivotal moments. Argumentation gives you a reasoning scaffold; what you build on top is up to you.

The smallest example

Two arguments that attack each other — the Nixon diamond:

Nixon diamond

Press enter or space to select a node.You can then use the arrow keys to move the node around. Press delete to remove it and escape to cancel.

Press enter or space to select an edge. You can then press delete to remove it or escape to cancel.

attacksdrag nodes · scroll to pan

Neither argument "wins" in isolation. Dung's acceptance semantics tell you which subsets of these arguments can coherently stand together — in this case, either-or, but not both. The Nixon diamond example walks through this in detail.

The four big ideas

If you read no other concept page, know these four:

1. Acceptance semantics — multiple ways to "accept" an argument (credulous, skeptical, grounded, preferred, stable). Pick one per query.
2. Walton schemes — ~60 named patterns of human argumentation (expert opinion, cause to effect, analogy, ...) with their critical questions baked in.
3. Weighted attacks and β — attacks have strengths; a single dial decides which strengths bind.
4. Value-based argumentation — characters' value priorities make the same framework reach different conclusions.

The encounter bridge ties these together for narrative scenes.

In our library

The workspace is layered: a Dung core, then extensions, then the bridge:

• argumentation — the Dung + ASPIC+ core.
• argumentation-bipolar — attacks + supports.
• argumentation-weighted — edge weights and β-budgets.
• argumentation-weighted-bipolar — composition of the above.
• argumentation-schemes — Walton's 60+ presumptive schemes.
• argumentation-values — value-based argumentation (Bench-Capon 2003 + multi-value).
• encounter-argumentation — the bridge into scene AI via the encounter crate.

See the reference overview for the curated entry-point types per crate.

Further reading

Start with Dung (1995), the paper that founded the field. Then Walton, Reed & Macagno (2008) for schemes. See the reading order for a full curriculum.

If you'd rather do than read, jump to your first scene.