CogniGron Seminar - Hartmut Fitz

Title: Causal models of language processing

Abstract:
The language system is a physical system implemented in neurobiological
tissue. Despite enormous efforts in psycholinguistics and cognitive
neuroscience, an integrated understanding of this system remains a
formidable challenge. What is still missing from many theoretical
proposals is a specification of the underlying neural mechanisms that
support language processing. I will outline a causal modeling approach
which offers a novel perspective on how to bridge this gap. It combines
methods from systems neuroscience with core insights from the
neurobiology of language.
     I will focus on two functional subcomponents of language that
require memory on different timescales. First, how are words stored and
maintained in long-term memory, and how are they retrieved from partial
cues? I will present simulation work using spiking networks with
multiple interacting plasticity rules that sheds light on how this
system could be implemented in a neurobiological substrate. And
secondly, what is the nature of short-term processing memory needed
during sentence comprehension? I will show that intrinsic neuronal
adaptation provides a novel memory mechanism that is suitable for the
online interpretation of utterances and allows the resolution of binding
relations between words and semantic roles.
     This work provides a first step towards a computational
neurobiology of language.