One game card describes one fallacy at a time; the repository’s ontology describes them all together, with their kinships and their links. Its notation follows the reference AIF format, and the gesture AIF performs for only a handful of fallacies, Argumentum extends to its entire tree.
Illustration: docs/publication/articles/illustrations/2a-liens-transverses.png, Argumentum repository, commit 981e5233 — 2× crop of the cross-links map (Cards/Fallacies/Mindmaps/fr/Fallacies_fr.links.svg): the arrows cut across the branches instead of following the tree.
A machine-readable description
An ontology is a formal description of the concepts of a domain and their relations, written to be read by a program rather than by the human eye. Argumentum publishes two of them, in the semantic web’s OWL language: one for the 1,408 nodes of the fallacy taxonomy, the other for the 223 nodes of the virtues. Each node carries a stable identifier, a label in French and English, and its place in the hierarchy — written in SKOS, the standard vocabulary for publishing thesauri. The game’s eight languages live in the cards and the mind maps; the ontology itself deliberately remains bilingual.
Both files are published online, and the publishing channel re-serves them every time the repository is updated.
The fallacy as an exception to a scheme
AIF — Argument Interchange Format — is the reference format that came out of artificial argumentation research for writing down an argument, the inference that produces it, and the conflicts that attack it. It relies on the argumentation schemes drawn up by Douglas Walton: a scheme is a legitimate form of argument — expertise, analogy, consequences — accompanied by its critical questions, those whose answer separates honest use from misuse.
A fallacy, in this framework, is an exception to a scheme: it imitates a valid form and skips the question that would have brought it down. AIF and Walton do this for a handful of fallacies; Argumentum applies it to its whole tree, node by node, describing for each one the legitimate scheme it imitates, what its attack targets, and the critical question it skips.
Measured on the repository’s corpus: 1,064 nodes out of 1,408 have their imitated scheme, 1,400 their attack target, 1,212 their critical question. Most often, the fallacy does not contradict a conclusion: it goes after what leads to it, the inference rule that produces it.
The question asked at the highest node
The taxonomy descends up to ten levels, and the mechanics of a node often hold for all of its descendants. The critical question is therefore asked at the highest node where it is true, and the descendants inherit it. The twelve questions written so far thus cover entire groups: “Biased reasoning,” 257 nodes, is covered by ten questions asked at its head, at its three children, and at six of its grandchildren; “Rhetorical device” is covered by two, including this one, asked at the “Humor” node:
Has the laughter established anything on the substance, or has it merely avoided answering it?
Each question lives in the corpus, and the ontology generator knows how to write it next to its node. The work progresses group by group.
Cross-links, drawn on the map
The corpus also says how fallacies answer one another, through eight relations: relying on another, denouncing it, enabling it, opposing it, inverting it, mirroring it, preceding it, being related to it. 487 fallacies out of 1,408 carry at least one of these links.
Each language has its mind map of the whole tree; its “links” variant draws each relation in color, one hue per verb. The French map carries exactly its 581 arrows, counted one by one. Making each arrow clickable — seeing where it leads, and why it exists — is the next project.
The virtues, and an export for programs
The virtues pack received the same treatment, with a difference: a virtue diverts nothing, it holds up. The corpus therefore links it to the good argument it embodies: 222 “holds up” links to 14 legitimate schemes.
All of it is also exported as simple CSV files that are easy for a program to consume: the 145 attack edges of the fallacies, the virtues files, the taxonomy tree, and the graph of relations between fallacies, each symmetric link being written in both directions. The public teaching project CoursIA uses them for its argument analysis series.
Where to find them
Sources: Argumentum repository, commit 32408fdd. Counts established on November 8, 2026 from Cards/Fallacies/Argumentum Fallacies - Taxonomy.csv and the published files of the docs/ontology/ folder (ontologies and exports, links verified). AIF modeling and attack graph export: docs/ontology/aif-export/README.md. Critical question quoted verbatim (“Humor” node, path 2.1.2). Consumer cited: CoursIA repository.