The project aims to investigate how language emergence in multi-agents environments. This will mainly be done by letting tabula rasa agents play different games where they have to communicate and coordinate in order to solve the game. In contrast to standard NLP, where one uses large datasets in order to find statistical relationships between words, we here try to take a more goal-oriented approach where the use of language is solely driven by a reward function. This is believed to be more closely related to how human language has developed.