A bit less magic! How to Recognize the Relevant Intentions

Abstract

Humans have a remarkable ability for recognizing the intentions of others. From verbal and non-verbal actions they are capable to derive a plausible intention of a communicator. Computational modeling of intention recognition, as it occurs in human communication, can contribute to a better understanding of the cognitive mechanisms underlying this everyday capacity. Blokpoel et al. (2011) have recently proposed a computational model of intention recognition using a Bayesian formalism. The model assumed, however, that the relevant set of intentions that the observer selects from to interpret a communicator is always prespecified. This assumption seems to significantly limit the psychological plausibility of the model. In this project I will extend the model proposed by Blokpoel et al. by incorporating a computational subprocess for selecting the set of relevant intentions. Implications of the extension of the model for its computational tractability will also be discussed.