The evolution of efficient compression in signaling games

Abstract

Converging evidence suggests that natural language meaning systems are subject to an efficiency constraint: maximizing simplicity and informativeness (Kemp et al., 2018). It re- mains less clear how languages optimize for efficiency over time. Towards illuminating these dynamic aspects of efficient communication, we provide a model grounded in evolutionary game theory. We consider a similarity-maximizing signaling game in which a Sender and Receiver evolve to communicate about a richly structured meaning space, and per- form simulations of adaptive dynamics that require minimal rationality assumptions about agents. We show that emergent languages converge near the achievable bounds of efficient compression given by Rate-Distortion Theory. This unifies static, information-theoretic accounts of communicative efficiency with dynamic, evolutionary accounts of how meaning can emerge ex nihilo in biological systems.

Seeing through the mess: evolutionary dynamics of lexical polysemy

“…Intuitively, if discriminability is high then individuals are less likely to view two different uses of a word as belonging to the same sense (for instance, if individuals can discriminate well, they would suggest the word mess to have two different senses in the utterances I see through the mess and I have a burger in the mess). There is much less research on the effect of semantic discriminability on language change than for the other two factors discussed above, but modeling and simulation results suggest that high discriminability entails more complex signaling inventories (Chaabouni et al, 2021;Imel, 2023). From a cognitive point of view, interdependencies among word utterances are weaker if they are (perceived as being) different from each other, i.e., in the case of high discriminability (den Heyer and Briand, 1986).…”