The Multiple Word Guessing Game

In this paper I bring together a number of insights in the field of semiotic dynamics. Inspired by recent advances in the understanding of the naming game and by previous attempts to define stages in the formation of lexical languages, we define two classes of language game problems called the single and multiple word guessing games. By being more complex than naming games, but not as complex as language games involving grammar, these problems form a next and tractable step in the understanding of the emergence of language. I show that a variety of previously proposed setups can be casted as instances of the guessing game, particularly those that investigate crosssituational learning from single word utterances. It is also shown how the problem of solving a multiple word guessing game can effectively be reduced to solving a single word guessing game. The effectiveness of the proposed technique is verified on a large number of problems of varying complexity. Insights gained from the conducted experiments are then used for defining a learning algorithm that outperforms several other algorithms as proposed in literature.