GSolver: Artificial solver of word association game

Abstract

In this paper we propose an artificial solver for a word association game. The possibility of a player to solve associations counts on the richness and deepness of players language and cultural qualifications. In order to provide answer(s) a human participant must accomplish a multiple memory search tasks for meanings of huge number of concepts and their frame of references. Hence the knowledge background (KB) of the proposed artificial solver is based on a large information repository formed by utilizing machine reading techniques for fact extraction from the web. As a KB we indirectly use the Albanian world-wide-web and the Gjirafa as a search engine. Complementary, the central processing unit (CPU) of the artificial solver is designed as a spreading activating network. The CPU treats provided hints and finds associations between them and concepts within the KB in order to incrementally compute and update a list of potential answers. Furthermore the CPU module is enriched by proposing a schema for finding the most promising solutions to be provided as the final answers. Experiments show that the accuracy of the system is as good as the average human player performance.