Three different formalizations of concept-learning in logic (as well a
s some variants) are analyzed and related. It is shown that learning f
rom interpretations reduces to learning from entailment, which in rum
reduces to learning from satisfiability. The implications of this resu
lt for inductive logic programming and computational learning theory a
re then discussed, and guidelines for choosing a problem-setting are f
ormulated. (C) 1997 Elsevier Science B.V.