Chapter 2. The Simulations
Table 2.11: Frequencies with which four-letter words are fixated at different fixation
positions. In the example, ’-’ indicates the fixation position (same as table 1.1).
beneficial for visual word recognition since it allows equal amounts of information about the word to go into each hemisphere.
Fixation Network For the fixation network, the situation is slightly different than for the control network. Compared to the control, it comits less errors at fixation positions 1 and 4, more errors at fixation positions 2 and 3 and the same number of errors at position 5. Similarly, the mean sum-square error given in table 2.10 for the fixation network has the lowest score at position 4 with 0.187 and the highest at position 2 with 0.212. Note that, overall, the average mse for the fixation postions is higher in this network (0.198) than in the control (0.174). This seems to suggest that it was harder for this network to learn when a condition on the fixation positions is imposed. However in total the fixation network has 3 errors less than the control (see table 2.1). Hence the added condition resulted in basically the same number of words being learned, however with less ’certainty’.
In order to be able to compare this data to the frequency used for each fixation position in this network, table 1.1 from section 2 is reproduced here as table 2.11. Looking at table 2.11, most of the changed statistics in the fixation network make sense. In this network, words have been presented with the highest frequency slightly right of centre at position 4. Consequently, this position has the lowest mse and less errors than in the