Relevance Logic

Relevance logic (also called relevant logic) is a family of non-classical logical systems that reject the validity of inferences in which the premises and conclusion have no meaningful connection. Where classical propositional logic and classical logic more broadly declare "if 2+2=5, then the Moon is made of cheese" a true conditional — the notorious paradox of material implication — relevance logic insists that for an implication A → B to hold, A must be relevantly connected to B. The connection is not merely a matter of truth values but of shared content.\n\nThe field was developed primarily by Alan Ross Anderson and Nuel Belnap at the University of Pittsburgh in the 1950s–1970s, with major contributions from Robert K. Meyer and Michael Dunn. Its motivations were simultaneously technical and philosophical: classical implication seemed to misrepresent the inferential practices of mathematics, science, and ordinary reasoning, all of which demand that conclusions follow from premises by virtue of content, not merely by logical form.\n\n== The Variable-Sharing Criterion ==\n\nThe central technical device of relevance logic is the variable-sharing criterion: in any valid implication A → B, A and B must share at least one propositional variable. This syntactic constraint blocks the paradoxes directly. The statement "if P then Q" cannot be a theorem when P and Q are completely unrelated atoms; the antecedent must contribute something to the consequent.\n\nThis criterion captures a topological intuition: valid inference is navigation within a connected graph of propositions, not teleportation between isolated nodes. The requirement that A and B share content mirrors the constraint in causal inference that a claimed cause must be connected to its effect by a mechanism — mere correlation, however strong, is not causation. The parallel is not decorative. Both fields are attempting to formalize the difference between genuine connection and accidental co-occurrence.\n\n== Systems and Semantics ==\n\nThe best-known relevance logics are R (the logic of relevant implication), E (the logic of entailment, which adds modal constraints on relevance), and T (ticket entailment). Each is strictly weaker than classical logic — they invalidate some classical theorems — but they are not merely fragments. They are alternative foundational frameworks with their own proof theories and algebraic semantics.\n\nThe Routley-Meyer semantics provides a model theory using a ternary accessibility relation: A → B holds at a world when, in all worlds where A is true and that are relevantly accessible from the current world, B is also true. This ternary relation generalizes the binary accessibility of modal logic and captures, in formal terms, the idea that implication is a three-place relation between premise situation, conclusion situation, and a context of relevance.\n\n== Relevance Beyond Logic ==\n\nRelevance logic's insistence on meaningful connection resonates far beyond proof theory. In artificial intelligence, the Frame Problem exposes the impossibility of tracking all unchanged facts after every action; relevance filtering — reasoning only about what is relevantly affected — is one of the principal strategies for making tractable the explosion of non-effects. The logic of relevance is, in this sense, the formal shadow of attention: not everything computed is computed about.\n\nThe connection to connectionist and neural network models is equally deep. A classical inference system treats all premises as equally active; a relevance-sensitive system modulates activation by relation strength. The variable-sharing criterion is, in essence, a topological constraint: conclusions must lie in the connected component of their premises. This topological intuition — that inference is navigation on a graph of meaningful relations — is the unifying thread linking relevance logic to graph-based learning and even semantic memory models in cognitive psychology.\n\nIn automated theorem proving, relevance guidance — selecting lemmas that share symbols or structural features with the goal — is a standard heuristic. Modern neural-guided provers use learned relevance scores to prune search spaces that would otherwise be intractable. The technical machinery has changed; the underlying principle has not.\n\nThe paradox of material implication is not a quirk of classical semantics but a symptom of a deeper representational failure: the assumption that logic can be reduced to truth tables without reference to the structure of what is being said. Relevance logic restores that structure — and in doing so, it reveals that logical consequence is not a matter of truth preservation alone but of meaningful connection. A logic that cannot distinguish between genuine inference and coincidental truth is not a logic of reasoning at all. It is a logic of bookkeeping.\n\n\n\n