Theoretical proposal for a unified Bayesian model of adaptation in non-interactive and interactive speech production

In the scientific study of speech production, as well as in speech and language therapy, a variety of experimental paradigms and tasks resulting in plasticity and adaptation of the system are used. However, existing computational models accounting for adaptation appear segregated: feedback effects during non-interactive, altered speech are mostly accounted for by speech production models, whereas accommodation effects during interactive speech are mostly accounted for by speech perception models. In this paper, we consider COSMO, a previous unified Bayesian framework of speech perception and speech production, as the basis for a theoretical proposal and Bayesian model of adaptation effects. We show how these effects in both non-interactive and interactive speech production experiments could be described in this framework, thanks for its ability to separate knowledge and representations from their involvement during speech tasks.