alright — here’s the mega-compendium you asked for: a single, exhaustive, human-only definitions deck that consolidates every distinct way “reasoning” was defined or carved up in the sources we covered. it’s long on purpose, and it sticks to definitions (with brief clarifiers where needed).

master integrative definition (all-purpose)

Human reasoning is the family of cognitive processes by which people form, revise, and justify beliefs, intentions, and actions by drawing consequences from information they already accept (perception, memory, testimony, data) together with background knowledge, goals, and norms. It (i) constructs new representations from given ones, (ii) updates degrees of belief (and sometimes retracts them), (iii) tests conclusions by counterexample search or probabilistic calibration, and (iv) is often conducted socially through argument, where audience, source reliability, and utilities matter. Reasoning is defeasible (open to retraction), context-sensitive, and can aim at logical necessity (deduction), evidential support (induction), best explanation (abduction), or effective action (practical reasoning).

definitions by normative target (what “good” means)

deductive reasoning

Reasoning in which a conclusion is licensed by necessity from premises: in every interpretation where the premises are true, the conclusion is true. Standard: validity (form-based) and soundness (valid + true premises). • Practical gloss: decides what must follow if the premises hold. • Human wrinkle: people also apply content and world knowledge when judging “valid.”

inductive reasoning

Reasoning in which a conclusion is licensed by degree of support from premises—probable but not guaranteed. Standards: inductive strength (how much support) and calibration (match between confidence and truth frequency). • Practical gloss: project from known cases/data to new cases or generalizations.

abductive reasoning (inference to the best explanation)

Reasoning that selects a hypothesis that, if true, would best explain the data. Standards: explanatory virtues (simplicity/parsimony, coherence/integration, breadth/coverage, mechanistic adequacy) balanced against prior plausibility. • Practical gloss: hypothesis choice guided by explanation quality.

probabilistic/Bayesian reasoning

Reasoning that represents beliefs as degrees of belief (probabilities) and updates them by Bayes’ rule (for certain evidence) or Jeffrey conditionalization (for uncertain evidence). • Core equation for conditionals: Pr(if p then q) = Pr(q | p) (the “Equation”). • Standards: coherence (obey probability axioms), posterior rationality (correct update), information gain (useful data selection).

practical (instrumental) reasoning

Reasoning that arrives at an intention or action by weighing means against ends, expected outcomes, constraints, and risks. • Standards: instrumental rationality (effectively achieves goals), expected utility (if formalized), reasons-responsiveness (sensitivity to relevant reasons). • Note: practical reasoning can conclude in action, not just in a belief about what action maximizes utility.

desiderative reasoning

Reasoning about what to value or want (e.g., adopting/adjusting aims). • Standards: coherence among values, reflective endorsement, reasons-responsiveness (are putative reasons for valuing actually good reasons?).

definitions by representational format (what’s being computed over)

rule-based (mental-logic) reasoning

Reasoning by applying abstract inference rules (e.g., modus ponens) to symbolic representations with variables that can be bound to case content. • Signature: variable binding, form sensitivity, stepwise derivations. • Diagnostic criteria for genuine rule use: performance with abstract/unfamiliar content, priming of rules, overextension to exceptions early in learning, domain-independent transfer, etc.

mental-models (possibilistic) reasoning

Reasoning by constructing iconic models of possible situations consistent with the premises and knowledge, following the principle of truth (represent what is true in each possibility; often omit falsities). Conclusions hold if they hold in all models; violation found by counterexample search (a model where premises hold but conclusion fails). • Difficulty metric: number of models required. • Strength: explains nonmonotonicity (adding information removes models) and classic “illusory” inferences when omitted falsities matter.

similarity/exemplar-based reasoning

Reasoning by retrieving and comparing cases; conclusions track similarity and feature overlap (often weighted by context) rather than explicit rules. • Frequent in categorization, property induction, and quick judgments.

causal-model reasoning

Reasoning over causal Bayes nets (variables with directed edges), combining causal power estimates and graph structure to predict interventions, explaining-away, and suppression effects in conditionals. • Standard: causal coherence with data and background knowledge.

definitions by dynamics (how beliefs change)

belief revision (coherence-based)

Reasoning as revising a set of beliefs to improve overall coherence (especially explanatory coherence) while honoring conservatism (minimize change) and resource limits. • Rules are defeasible: even from p and (p→q) you might withhold q if it worsens coherence or you retract a premise. • Avoiding inconsistency is a pro tanto (defeasible) norm, not absolute.

nonmonotonic reasoning

Reasoning where adding premises can defeat prior conclusions (e.g., “birds fly” vs. “penguins are birds”). • Standard: defeasible warrant; conclusions are default and withdrawable.

dynamic (non-invariant) updating

Reasoning where new information changes not only a marginal probability (e.g., Pr(p)) but also a conditional (Pr(q|p)). • Formal repair: choose the closest revised distribution to the prior (e.g., minimize Kullback–Leibler distance) subject to the new constraints.

definitions by architecture (how it’s implemented in minds)

dual-process/systems definition

Reasoning arises from two interacting processes: Type 1 / associative — fast, automatic, similarity/contiguity-driven, parallel constraint satisfaction; produces intuitions. Type 2 / rule-based — slower, controlled, symbolically structured, serial/strategic; produces justified analyses. • Conflict is common; Type 2 can override Type 1, but intuitive pull often persists.

hybrid (rules + instances) definition

Reasoning is a cooperative/competitive mix: retrieved instances can access abstract rules (and vice-versa); outcomes depend on salience, task demands, and practice.

argumentation (reasoning as persuasion and coordination)

Reasoning as presenting, evaluating, and revising claims in dialogue to resolve disagreement, reduce ignorance, and coordinate action. • Standards (Bayesian/pragmatic): priors, likelihoods, source reliability, cost–benefit utilities, and audience goals. • “Fallacies” (ignorance, ad hominem, slippery slope, circularity) can be rational depending on test reliability, source credibility, categorical flexibility, and explanatory structure.

reasons-responsiveness

Reasoning as the capacity to recognize, weigh, and act for reasons as reasons, not merely to compute. • Standard: sensitivity to the right reasons in context (epistemic, practical, moral).

task/genre-level definitions (what counts as reasoning in specific domains)

conditional reasoning (everyday “if…then”)

Reasoning about if-sentences interpreted (often) via conditional probability (Pr(if p then q)=Pr(q|p)) or via possibility constraints (mental models). • Nonmonotonicity: strengthening the antecedent need not preserve truth for real-world conditionals. • Utility/deontic versions (promises, threats, warnings, permissions/obligations) guide action by linking goals and norms to conditionals.

quantified/syllogistic reasoning

Reasoning with quantifiers (all, some, none, most, few). • Probabilistic semantics: quantifiers map to conditional probabilities; conclusions predicted by dependency graphs and min/max heuristics (least informative premise sets quantifier; confidence by most informative premise). • Prototype sampling models: build a probabilistic prototype, sample a small set (≈ working-memory capacity), then test candidate conclusions against the sample.

property induction

Reasoning that generalizes a property from one or more categories to others. • Standards: similarity/typicality, diversity of premises, mechanism plausibility, and background theories.

data-selection (hypothesis testing)

Reasoning that chooses which data to gather to learn most efficiently. • Standard: expected information gain (not mere falsification); choices depend on priors, base rates, and utilities.

explanatory evaluation

Reasoning that ranks explanations for observed data. • Standards: simplicity, integration/coherence, coverage, mechanism, narrow latent scope (don’t over-explain unobserved facts), all tempered by prior plausibility.

micro-definitions of core normative/computational notions

Validity (logical): impossibility of true premises with false conclusion (form-based).
Soundness: valid argument with true premises.
Inductive strength: degree to which premises raise probability of conclusion.
Coherence (probability): conformity to probability axioms and constraints; coherence interval = range of permissible conclusion probabilities implied by uncertain premises.
p-validity: probabilistic analogue of validity—conclusion’s uncertainty cannot coherently exceed sum of premise uncertainties.
Bayesian conditionalization: update Pr(H) to Pr(H E) when E is certain.
Jeffrey conditionalization: update Pr(H) when evidence E is uncertain (update Pr(E) and mix conditionals).
Dynamic (non-invariant) update: revise both marginals and conditionals; often solved by minimum-disturbance (KL-minimization).
Likelihood ratio: Pr(E H)/Pr(E ¬H); multiplicative force of evidence or argument.
Causal power (W): strength of a cause independent of base rate of the effect; often ΔP/(1−base).
ΔP (delta-P): Pr(effect cause) − Pr(effect ¬cause); index of inferential relevance.
Explaining away: in common-effect structures, evidence for one cause lowers belief in alternatives.
Information gain: expected reduction in uncertainty (various measures; classically Shannon entropy reduction).
Kullback–Leibler distance: divergence between prior and posterior distributions; used to find closest update compatible with new constraints.
Default/defeasible rule: an inference licensed unless an exception is known (withdrawable).

definitional contrasts & caveats (to avoid category errors)

Reasoning vs. proof: Reasoning is belief/plan revision under constraints; proof is a formal product.
Rationality ≠ logical closure: It’s rational to limit derivations (clutter avoidance), to suspend consequences, or to retract premises.
Everyday “if” ≠ material conditional: People treat unmarked conditionals as rules/probabilistic constraints, not as “¬p or q.”
Deduction vs. induction are distinct targets: “Necessary consequence” and “evidential support” are different evaluation modes, though they can interact.
Group vs. solo: Much human reasoning is argumentative; standards include source reliability and audience-relative goals, not just private computation.

ultra-concise definitional map (pocket version)

Reasoning (umbrella): constructing, revising, and justifying beliefs/intentions by drawing consequences from information + knowledge + goals, individually or socially, under uncertainty and resource limits.
Targets: • Deductive (necessity) · Inductive (support) · Abductive (best explanation) · Practical (action) · Desiderative (what to value).
Formats: • Rules (symbolic) · Models (possibilities) · Instances (similarity) · Causal nets (structured causes).
Dynamics: • Belief revision (coherence + conservatism) · Nonmonotonic defaults · Bayesian/Jeffrey updates · KL-minimal revisions.
Social: • Argumentation (evidence, reliability, utilities) · Reasons-responsiveness.
Standards: • Validity/soundness · Strength/calibration · Coherence/p-validity · Explanatory virtues · Information gain · Goal satisfaction.

That’s the whole toolkit of definitions—stitched together and human-only. If you want, I can drop this into a clean PDF or a one-page “definitions wall” you can paste at the top of a doc.

AIT SAID Azzedine Idir