Intelligent System for Playing Tarok

Abstract

We present an advanced intelligent system for playing three-player tarok card game. The system is based on alpha-beta search with several enhancements such as fuzzy transposition table, which clusters strategically similar positions into generalised game states. Unknown distribution of other players’ cards is addressed by Monte Carlo sampling. Experimental results show an additional reduction in size of expanded 9-ply game-tree by a factor of 184. Human players judge the resulting program to play tarok reasonably well.