Rational Speech Acts (RSA) is a model of replicating cognitive theories of human communication based on the concepts from Bayesian decision theory. It's a form of cognitive AI in linguistics, where we try to replicate various pragmatic phenomena of the human mind, such as how individuals infer implied meanings, resolve ambiguity, and understand context-dependent communication. By modeling the recursive reasoning process between speakers and listeners, RSA allows us to capture the interplay of literal and pragmatic meaning, and it is a computational approach to understanding and simulating human-like communication.
This repo contains an implementations of the politeness RSA model in Gen probailistic programming language. Politeness RSA model examines how polite language affects cooperative communication, where polite utterances may seem misleading but serve to protect the listener's and speaker's face.