Autonomous Neural Dynamics to Test Hypotheses in a Model of Spatial Language

Mathis Richter	Institut für Neuroinformatik, Ruhr-Universität Bochum, 44780 Bochum, Germany
Jonas Lins	Institut für Neuroinformatik, Ruhr-Universität Bochum, 44780 Bochum, Germany
Sebastian Schneegans	Institut für Neuroinformatik, Ruhr-Universität Bochum, 44780 Bochum, Germany
Yulia Sandamirskaya	Institut für Neuroinformatik, Ruhr-Universität Bochum, 44780 Bochum, Germany
Gregor Schöner	Institut für Neuroinformatik, Ruhr-Universität Bochum, 44780 Bochum, Germany

Abstract

Resolving relational spatial phrases requires that a coherent mapping emerges between a visual scene and a triad of two objects and a relational term. We present a theoretical account that solves this problem based on neural principles. A neural dynamic architecture represents perceptual information in activation fields that make detection and selection decisions through neural interaction. Activation nodes and their connectivity to the perceptual fields represent concepts. Dynamic instabilities enable the autonomous sequential organization of the processing steps needed to resolve relational spatial phrases. These include bringing visual objects into the attentional foreground, performing spatial transformations, and making matching decisions. We demonstrate how the neural architecture may autonomously test different hypotheses to resolve relational spatial phrases. We discuss how this neural process account relates to existing theoretical perspectives and how to move beyond the entry point sketched here.

Files

Autonomous Neural Dynamics to Test Hypotheses in a Model of Spatial Language (409 KB)