Over the past thirty years there have been significant achievements in the field of auto-mated theorem proving with respect to the reasoning power of the inference engines.Although some effort has also been spent to facilitate more user friendliness of the de-duction systems, most of them failed to benefit from more recent developments in therelated fields of artificial intelligence (AI), such as natural language generation and usermodeling. In particular, no model is available which accounts both for human deductiveactivities and for human proof presentation. In this thesis, a reconstructive architecture issuggested which substantially abstracts, reorganizes and finally translates machine-foundproofs into natural language. Both the procedures and the intermediate representationsof our architecture find their basis in computational models for informal mathematicalreasoning and for proof presentation. User modeling is not incorporated into the currenttheory, although we plan to do so later.
This paper presents a new way to use planning in automated theorem provingby means of distribution. To overcome the problem that often subtasks fora proof problem can not be detected a priori (which prevents the use of theknown planning and distribution techniques) we use a team of experts that workindependently with different heuristics on the problem. After a certain amount oftime referees judge their results using the impact of the results on the behaviourof the expert and a supervisor combines the selected results to a new startingpoint.This supervisor also selects the experts that can work on the problem inthe next round. This selection is a reactive planning task. We outline whichinformation the supervisor can use to fulfill this task and how this informationis processed to result in a plan or to revise a plan. We also show that the useof planning for the assignment of experts to the team allows the system to solvemany different examples in an acceptable time with the same start configurationand without any consultation of the user.Plans are always subject to changeShin'a'in proverb
We present a method for learning heuristics employed by an automated proverto control its inference machine. The hub of the method is the adaptation of theparameters of a heuristic. Adaptation is accomplished by a genetic algorithm.The necessary guidance during the learning process is provided by a proof prob-lem and a proof of it found in the past. The objective of learning consists infinding a parameter configuration that avoids redundant effort w.r.t. this prob-lem and the particular proof of it. A heuristic learned (adapted) this way canthen be applied profitably when searching for a proof of a similar problem. So,our method can be used to train a proof heuristic for a class of similar problems.A number of experiments (with an automated prover for purely equationallogic) show that adapted heuristics are not only able to speed up enormously thesearch for the proof learned during adaptation. They also reduce redundancies inthe search for proofs of similar theorems. This not only results in finding proofsfaster, but also enables the prover to prove theorems it could not handle before.
We investigate one of the classical problems of the theory ofterm rewriting, namely termination. We present an ordering for compar-ing higher-order terms that can be utilized for testing termination anddecreasingness of higher-order conditional term rewriting systems. Theordering relies on a first-order interpretation of higher-order terms anda suitable extension of the RPO.
This paper introduces a multi-valued variant of higher-order resolution and provesit correct and complete with respect to a natural multi-valued variant of Henkin'sgeneral model semantics. This resolution method is parametric in the number of truthvalues as well as in the particular choice of the set of connectives (given by arbitrarytruth tables) and even substitutional quantifiers. In the course of the completenessproof we establish a model existence theorem for this logical system. The workreported in this paper provides a basis for developing higher-order mechanizationsfor many non-classical logics.
Coloring terms (rippling) is a technique developed for inductive theorem proving which uses syntactic differences of terms to guide the proof search. Annotations (colors) to terms are used to maintain this information. This technique has several advantages, e.g. it is highly goal oriented and involves little search. In this paper we give a general formalization of coloring terms in a higher-order setting. We introduce a simply-typed lambda calculus with color annotations and present an appropriate (pre-)unification algorithm. Our work is a formal basis to the implementation of rippling in a higher-order setting which is required e.g. in case of middle-out reasoning. Another application is in the construction of natural language semantics, where the color annotations rule out linguistically invalid readings that are possible using standard higher-order unification.
In this paper we will present a design model (in the sense of KADS) for the domain of technical diagnosis. Based on this we will describe the fully implemented expert system shell MOLTKE 3.0, which integrates common knowledge acquisition methods with techniques developed in the fields of Model-Based Diagnosis and Machine Learning, especially Case-Based Reasoning.
We first show that ground term-rewriting systems can be completed in apolynomial number of rewriting steps, if the appropriate data structure for termsis used. We then apply this result to study the lengths of critical pair proofs innon-ground systems, and obtain bounds on the lengths of critical pair proofsin the non-ground case. We show how these bounds depend on the types ofinference steps that are allowed in the proofs.