Holographic Reduced Representations for Oscillator Recall: A Model of Phonological Production

Abstract

This paper describes a new computational model of phonological production, Holographic Reduced Representations for Oscillator Recall, or HORROR. HORROR's architecture accounts for phonological speech error patterns by combining the hierarchical oscillating context signal of the OSCAR serial-order model~\cite{VousdenEtAl:2000,BrownEtAl:2000} with a holographic associative memory~\cite{Plate:1995}. The resulting model is novel in a number of ways. Most importantly, all of the noise needed to generate errors is intrinsic to the system, instead of being generated by an external process. The model features fully-distributed hierarchical phoneme representations and a single distributed associative memory. Using fewer parameters and a more parsimonious design than OSCAR, HORROR accounts for error type proportions, the syllable-position constraint, and other constraints seen in the human speech error data.

Available Versions of this Item

• Holographic Reduced Representations for Oscillator Recall: A Model of Phonological Production. (deposited 26 Jun 2002) [Currently Displayed]
  • Holographic Reduced Representations for Oscillator Recall: A Model of Phonological Production. (deposited 03 Jul 2002)