Term Generalization

Appeared in Volume 9/4, November 1996

Keywords: terms.

aramante@six.clo.tour.ist
Arch-image
11th June 1996

I am looking for references on term generalization. Term generalization means that, for example:

Given a set of ground examples:

E1(a, [c], [a, c])
E2(b, [a, b], [b, a])
E3(c, [a, b], [c, a, b])

A general form (by term generalization) is:
E(X, [Y|Ys], [X, Y|Xs])

Next, and more particularly, I am also interested in works on "active generalization" or "relation learning"; for example:

Given a set of ground examples:

E1(1, 2)
E2(2, 4)
E3(3, 6)

A constraint general form is:
E(X, Y) :- Y=2*X

js10@doc.ic.ac.uk
Joachim Schimpf
18th June 1996

You might be interested in Generalised Propagation, which infers (possibly approximate) generalisations from disjunctions. There is an implementation called Propia, which comes as a library in the ECLiPSe system. A small example:

[eclipse 2]: [user].
e(1, 2).
e(2, 4).
e(3, 6).
yes.

[eclipse 3]: e(X, Y) infers most.
X = X{[1..3]}
Y = Y{[2, 4, 6]}
yes.

The generalisations are represented in terms of the constraint system available, in this case the finite domains.