In this paper, we study inferability of Prolog programs from positive examp
les alone. The importance of studying inductive inference From positive dat
a alone stems from the scarcity of negative information in practical applic
ations. Starting with Shinohara's result on bounded finite thickness and in
ductive inference from positive data, we investigate a broad spectrum of cl
asses of Prolog programs inferable from positive data and the relationships
between these classes. First we consider programs without local variables
and show that a few classes of programs (without any bound on the length of
clauses) are inferable from positive data. Then we consider programs with
local variables and identify a very useful class of programs inferable from
positive data, using mode annotations of predicates and linear predicate i
nequalities. (C) 2000 Elsevier Science B.V. All rights reserved.