Chapter 1. Introduction and Background
Figure 1.8: The split-fovea-model. Here the word clam is presented with a fixation point
in-between the letters ’l’ and ’a’. Activation spreads into the hidden layers and from
there to the output layer which produces the correct output /klaem/.
network in different fixation points. For example Figure 1.9 shows the different inputs for the word clam. The network has to learn to map all the different possible inputs for a word to the same correct output target. To be able to do this task correctly for all the words and non-words presented to it, the network has to learn a shift-invariant mapping from the input to the output, so that shifting the input across the viewing positions doesn’t make a difference to the final computed output.
The Viewing Position Effect
The fact that words can be fixated at many different positions (once or more than once) during reading is an important aspect of visual word recognition. There is evidence that the probability of recognising a word depends on where it is fixated (O’Reagan and Javobs, 1992; Montant et al., 1998; Nazir, 2000). Words were presented to people while controlling the exact position at which the word was fixated by the person. This is done for words with different lengths and the words that were recognised correctly