Regular Language Inference for Learning Rules of Simplified Boardgames

摘要

We deal with the problem of learning game rules by observing the play, the study initiated by Björnsson for the class of Simplified Boardgames, describing a rich family of chess-like games. In this paper we restate the problem in terms of regular language inference and improve Björnsson's algorithm combining applications of existing DFA learning algorithms with our domain-specific approach. We present the results of tests on a number of games, including human-made and artificially generated ones.