We often reach conclusions partially on the basis that we do not have evidence that the conclusion is false. A newspaper story warning that the local water supply has been contaminated would prevent a person from drinking water from the tap in her home. This suggests that the absence of such evidence contributes to her usual belief that her water is safe. On the other hand, if a reasonable person received a letter telling her that she had won a million dollars, she would consciously consider whether there was any evidence that the letter was a hoax or somehow misleading before making plans to spend the money. All to often we arrive at conclusions which we later retract when contrary evidence becomes available. The contrary evidence defeats our earlier reasoning. Much of our reasoning is defeasible in this way. Since around 1980, considerable research in AI has focused on how to model reasoning of this sort. In this paper, I describe one theoretical approach to this problem, discuss implementation of this approach as an extension of Prolog, and describe some application of this work to normative reasoning, learning, planning, and other types of automated reasoning.

this report, henceforth called DefDRT, employ a unification-based top-down parser. The reader is assumed to have a basic grasp of Prolog, parsing, the use of feature structures, and unification. DefDRT accepts as input multi-sentence discourse --- possibly over several different program executions --- and converts the information in the discourse into appropriate strict or defeasible facts and rules within a d-Prolog database. By asking a question the user can query the database or can investigate whether a certain inference can be proven from the database using defeasible reasoning. To achieve this final implementation, many modifications were made to what once was DRT.GLP. The modifications were necessary to deal with generic plurals and for distinguishing between sentences that express strict rules and those that express defeasible rules. 2 Background

This paper describes the construction of a general-purpose defeasible reasoner that is complete for first-order logic and provably adequate for the argument-based conception of defeasible reasoning that I have developed elsewhere. Because the set of warranted conclusions for a defeasible reasoner will not generally be recursively enumerable, a defeasible reasoner based upon a rich logic like the predicate calculus cannot function like a traditional theorem prover and simply enumerate the warranted conclusions. An alternative criterion of adequacy called i.d.e.-adequacy is formulated. This criterion takes seriously the idea that defeasible reasoning may involve indefinitely many cycles of retracting and reinstating conclusions. It is shown how to construct a reasoner that, subject to certain realistic assumptions, is provably i.d.e.-adequate. The most recent version of OSCAR implements this system, and examples are given of OSCAR's operation. 1. Introduction The aim of the OSCAR proj...