Competitive Dynamics in a Dual-route Connectionist Model of Print-to-sound Transformation
James A. Reggia
Department of Neurology, University of Maryland Hospital,
22 South Green Street, Baltimore, MD 21201, USA
and
Department of Computer Science,
University of Maryland Institute for Advanced Computer Studies,
College Park, MD 20742, USA
Patricia M. Marsland
Rita Sloan Berndt
Department of Neurology, University of Maryland Hospital,
22 South Green Street, Baltimore, MD 21201, USA
Abstract
This paper describes a connectionist model of print-to-sound transformation ("word naming" or "reading aloud"). The associative network it uses is based on published studies of oral reading, and simulation results are compared to experimental data in the psychological literature. The results obtained are of interest for two separate reasons. First, the print-to-sound connectionist model is based on an indirectly interactive dual-route hypothesis of reading aloud. The model confirms that this hypothesis, when implemented as a detailed and sizeable computer simulation, can account qualitatively for a number of behavioral phenomena such as regularity and word frequency effects. The model thus provides support for a modified dual-route hypothesis involving indirectly interactive routes and verifies the hypothesis' consistency with a set of replicable psychological data. The second reason the print-to-sound connectionist model is of interest is that it uses a new approach to implementing competitive dynamics in connectionist models. Focused spread of network activation and avoidance of network saturation are produced by using a competitive activation mechanism rather than explicit inhibitory links between competing modes. The print-to-sound model demonstrates for the first time that competitive activation mechanisms can function usefully in relatively large, complex situations of interest in cognitive psychology and artificial intelligence.
