PDF《Ebner, Marc and Levine, John and》

2 Previous work Several attempts have been made in the past to model aspects of both simple arcade games and board games. Most of them had the purpose of being able to generate complete games (through evolutionary computation or constraint satisfaction) but there have also been attempts at modelling games for the purposes of game design assistance. The Stanford General Game Playing Competition de?nes its games in a special-purpose language based on ?rst-order logic (4). This language is capable of modelling a very large space of games, as long as they are perfect information, turn-taking games;

those games that can practically be modelled with the Stanford GDL tend to be board games or games that share similar qualities. The language is very verbose: an example de?nition of Tic- Tac-Toe runs to three pages of code. We do not know of any automatically generated game descriptions using this GDL, but even if they exist, their space is unlikely to be suitable for automatic search, as any change to a game is likely to result in an unplayable and perhaps even semantically inconsistent game. One noteworthy attempt at describing and evolving board games is Browne'

s Ludi language and the system powered by this language that evolves complete board games (1).........