It's About Time: Temporal References in Emergent Communication

The paper presents findings on the emergence of temporal references in agent communication, highlighting that architectural changes, specifically a modified batching method, are crucial for their development, as opposed to merely altering the loss function. In experiments, over 95% of agents using this new architecture successfully developed temporal references in a temporal referential games environment. This advancement is significant for AI practitioners as it suggests that incorporating temporal structures can enhance communication efficiency in emergent language systems, paving the way for more optimal coding strategies.