Causal Bayesian Networks, Signalling Games and Implicature of ‘More Than n’

Abstract

We use causal Bayesian networks for modelling experimental data on scalar implicatures of numerals modified by the comparative quantifier ‘more than’. The data are taken from the study of Cummins et al. (2012). They showed that subjects infer upper bounds for the true number n when ‘more than n’ has been used. They also showed that the difference between upper and lower bounds differs with the roundness of n. We show how the data can be explained by a production model of the speaker. We are interested in the architecture of such a model, and argue that causal Bayesian networks provide a natural and promising approach.