Seminar: Donders Institute for Brain, Cognition, and Behaviour Nijmegen, The Netherlands
Mr. Mark Blokpoel
Radboud UniversityDonders Institute for Brain, Cognition, and Behaviour Nijmegen, The Netherlands
Scaling Models of Communication
Department of Computer Science
Thursday, November 22, 2012, 1:00 p.m., Room EN-2022
It is common practice within cognitive psychology that computational explanations of cognitivephenomena should fit observations made in the lab. However, for these explanations to besatisfactory they should also scale to situations of real-world complexity outside the lab. This scalingis often taken for granted but, as it turns out, from a computational-complexity perspective, manyexplanations in cognitive science do not yet scale in an obvious way.Take, for instance, inferential models of communication. In such models, understanding thecommunicative intentions of others based on their behavior is seen as an ‘inference to the bestexplanation’, a.k.a. abduction. It is known that computations of this type can be computationallyintractable (NP-hard or worse) and hence can require on the order of years or centuries to computefor anything but trivially simple inputs. Models based on such computations thus cannot yet explainhow humans can efficiently communicate in complex real-world situations.In the first part of this talk I will highlight the importance of facing the challenge of scaling head on.I will present a constructive methodology based on complexity theory from theoretical computerscience to deal with this challenge and discuss its benefits for cognitive science. Then in the secondpart I will illustrate how one can apply this methodology to inferential Bayesian-based models ofcommunication.