There was a long tradition in philosophy according to which good reasoning had to be deductively valid. However, that tradition began to be questioned in the 1960’s, and is now thoroughly discredited. What caused its downfall was the recognition that many familiar kinds of reasoning are not deductively valid, but clearly confer justification on their conclusions. Here are some simple examples.

Relevance of premises to conclusion can be explicated through Toulmin’s notion of warrant, understood as an inference rule, albeit not necessarily formal. A normative notion of relevance requires the warrant to be reliable. To determine reliability, we propose a fourfold classification of warrants into a priori, empirical, institutional, and evaluative, with further subdivisions possible. This classification has its ancestry in classical rhetoric and recent epistemology. Distinctive to each type of warrant is the mode by which such connections are intuitively discovered (...) and the grounds on which we ultimately justify them. The classification of warrants is thus epistemic. We illustrate the difference by contrasting empirical physical with institutional intuition, and argue for the advantages of this approach over Toulmin’s conception of field dependence. (shrink)

We perceive relevance by virtue of inference habits, which may be expressed as Pierce's leading principles or as Toulmin's warrants. Hence relevance in a descriptive sense is a ternary relation between two statements and a set of inference rules. For a normative sense, the warrants must be properly backed. Different types of warrant to empirical generalizations, we introduce L.J. Cohen's notion of inductive support. A to empirical generalizations, we introduce L.J. Cohen's notion of inductive support. A generalization H is supported (...) by evidence E to degree i/in iff E indicates that H passes canonical test i, where there are n canonical tests. In a canonical test, one or more relevant variables, factors which may falsify H, are varied. H passes a test if it is not falsified. The tests are cumulative. Degree of support is relative to the canonical test, and may be modeled as relative to a point in a dialectical situation. A value of a variable at which H is falsified is a rebutting value. A is normatively relevant to B with respect to W iff sup[associated generalization(W), E] = i/n and for j > i, there is a presumption that the values of j are non-rebuting. (shrink)