We present a recurrent neural network which learns to suggest the next
move during the descent along the branches of a decision tree. More p
recisely, given a decision instance represented by a node in the decis
ion tree, the network provides the degree of membership of each possib
le move to the fuzzy set ''good move''. These fuzzy values constitute
the core of the probability of selecting the move out of the set of th
e children of the current node. This results in a natural way for driv
ing the sharp discrete-state process running along the decision tree b
y means of incremental methods on the continuous-valued parameters of
the neural network. The bulk of the learning problem consists in stati
ng useful links between the local decisions about the next move and th
e global decisions about the suitability of the final solution. The pe
culiarity of the learning task is that the network has to deal explici
tly with the twofold charge of lighting up the best solution and gener
ating the move sequence that leads to that solution. We tested various
options for the learning procedure on the problem of disambiguating n
atural language sentences. (C) 1998 Elsevier Science Ltd. All rights r
eserved.