Belief Revision in a Discrete Temporal Probability-Logic

We describe a discrete time probability logic for use as the representation language of a temporal knowledge base In addition to the usual expressive power of a discrete temporal logic our language allows for the speci cation of non universal generaliza tions in the form of statistical assertions This is similar to the probability logic of Bacchus but di ers in the inference mech anisms In particular we discuss two in teresting and related forms of inductive in ference interpolation and extrapolation Interpolation involves inferences about a time interval or point contained within an interval for which we have relevant statis tical information Extrapolation extends statistical knowledge beyond the interval to which it pertains These inferences can be studied within a static temporal knowl edge base but the further complexity of dynamically accounting for new observa tions makes matters even more interesting This problem can be viewed as one of be lief revision in that new observations may con ict with current beliefs which require updating As a rst step toward a full edged temporal belief revision system we consider the tools of inductive logic We suggest that Carnap s method of con rma tion may serve as a simple mechanism for belief revision