This paper describes a working, computational model of word recognition that combines a letter classification component with a component that segments the string of classified letters into words and uses a dynamic programming method for matching the words against a lexicon of over 2,800 words. The letter classification component is a neural network trained to classify, in parallel, inputs corresponding to 20x188 pixel array images of letter sequences, 14 or more letters long. Consistent with human capabilities, the system can classify all 14 letters at a level above chance, and on average, classifies the first 7 or 8 letters in the sequence correctly. Dictionary lookup improves classification accuracy by 1 character per image. The model is robust, having been trained and tested on the entire text of the book The Wonderful Wizard of Oz, printed in multiple fonts and in both mixed and upper-case letters. It provides a computation-level understanding of word recognition capabihties, in which errors are attributable to the theoretically inevitable difficulties associated with learning to classify large input patterns. The model mimics human capabilities for circumventing some of these difficulties by imposing constraints on fixation positions that reduce image variability.

When people read, they classify a relatively long string of characters in parallel. Machine learning principles predict that classification learning with such high dimensional inputs and outputs will fail unless biases are imposed to reduce input and output variability and/or the number of candidate input/output mapping functions evaluated during learning. The present paper draws insight from observed reading behaviors to propose some potential sources of such biases, and demonstrates, through neural network simulations of letter-sequence classification learning that: (1) Increasing dimensionality does hinder letter classification learning and (2) the proposed sources of bias do reduce dimensionality problems. The result is a model that explains word superiority and word frequency effects, as well as consistencies in eye fixation positions during reading, solely in terms of letter classification learning.