Encoder: a connectionist model of how learning to visually encode fixated text images improves reading fluency.

This article proposes that visual encoding learning improves reading fluency by widening the span over which letters are recognized from a fixated text image so that fewer fixations are needed to cover a text line. Encoder is a connectionist model that learns to convert images like the fixated text images human readers encode into the corresponding letter sequences. The computational theory of classification learning predicts that fixated text-image size makes this learning difficult but that reducing image variability and biasing learning should help. Encoder confirms these predictions. It fails to learn as image size increases but achieves humanlike visual encoding accuracy when image variability is reduced by regularities in fixation positions and letter sequences and when learning is biased to discover mapping functions based on the sequential, componential structure of text. After training, Encoder exhibits many humanlike text familiarity effects.

Recognizing Hand-Printed Letters and Digits Using Backpropagation Learning.

We report on results of training backpropagation nets with samples of hand-printed digits scanned off of bank checks and hand-printed letters interactively entered into a computer through a stylus digitizer. Generalization results are reported as a function of training set size and network capacity. Given a large training set, and a net with sufficient capacity to achieve high performance on the training set, nets typically achieved error rates of 4-5% at a 0% reject rate and 1-2% at a 10% reject rate. The topology and capacity of the system, as measured by the number of connections in the net, have surprisingly little effect on generalization. For those developing hand-printed character recognition systems, these results suggest that a large and representative training sample may be the single, most important factor in achieving high recognition accuracy. Benefits of reducing the number of net connections, other than improving generalization, are discussed.