Next-generation Language Games: the Guessing Game Revisited

In this paper we show how a Guessing Game can be reimplemented combining two recently developed formalisms: Fluid Construction Grammar and Incremental Recruitment Language. The main difference from older experiments is that the meaning of a sentence is conceived as a semantic constraint program which the speaker wants the hearer to execute and the use of a simple grammar expressed by suffixes. In this paper we show which constraints might be involved and how a semantic constraint program can be encoded in and decoded from an utterance and show how the agents can learn how to express these semantic programs. We conclude by showing results of a grounded multi-agent experiment using a Talking Heads set-up.