The Rational Speech Act model
This week we are returning briefly to word learning, in order to introduce the Rational Speech Act (RSA) model, which is a way of applying Bayesian modelling to pragmatic inference and communicative behaviour. In week 7 we’ll combine insights from the RSA model with iterated learning to build a complete (but simple!) model of language transmission,where languages are used in communication and the language produced during communication form the data for learning in new individuals.
In the lecture I’ll go over the basics of the RSA model, and in the labs you’ll get a chance to play with it. I recommend you read Frank & Goodman (2014) as an intro to the RSA model (before or after the lecture is fine, take your pick) – the details of the model are laid out in section 2 of that paper and are quite brief, I’ll go through these more slowly in the lecture, but the paper provides some useful background on pragmatic inference in word learning and a nice series of experiments verifying that adults and children are indeed capable of making this kind of inference.
If you want an overview of other applications of RSA models and extensions beyond what we’ll cover on the course, you could (optionally) check out Goodman & Frank (2016) – it’s more technical but full of really interesting ideas about how to apply the same model to e.g. hyperbole.
Frank, M. C., & Goodman, N. D. (2014). Inferring word meanings by assuming that speakers are informative. Cognitive Psychology, 75, 80-96.
Goodman, N. D., Frank, M. C. (2016). Pragmatic language interpretation as probabilistic inference. Trends in Cognitive Sciences, 20, 818-829.
This page was written by Kenny Smith. All aspects of this work are licensed under a Creative Commons Attribution 4.0 International License.
Project maintained by centre-for-language-evolution Hosted on GitHub Pages — Theme by mattgraham