A full account of how match values are calculated in the spatial coding model (based on a superposition match) is provided in Davis (2010). To investigate the model's match values we can plot the input to the word layer. Right click on that layer and select Plot net input. This input actually combines the superposition match and the match signal that comes from the initial and final letter banks. The latter two inputs reflect the match between the initial and final letters of the orthographic input and the word coded by given word node. This can be made clearer by considering an example. Try running the same form trial as before, and then change the plot settings …

The resulting plot should look like this:

There are at least three things that can be noted in this plot:

1. The match value changes over time
   This is because the match is sensitive not only to position overlap but also the strength of the input signal (the other dimension of the spatial code). Because the letter level activities take a little while to asymptote the match value increases during the initial part of the trial.
2. The maximum value in this plot is 6
   More generally, the maximum net input on this channel will be L+2, where L is the number of letters in the word coded by the node computing the match. The +2 reflects the fact that this net input signal includes not only the superposition match but also the initial and final letter match. To see this, try changing the stimulus to xformz. The maximum net input is still associated with the form node, but the value is now 4. This is because the addition of extra letters on either side disrupts the initial and final letter match for form, but not the superposition match (which still has a peak of four). Another way to test this would be to selectively lesion the feedforward connections to the word layer and examine the effect on the net input.
3. Not all neighbours are substitution neighbours
   An important consequence of spatial coding is that many words that are not orthographically similar according to slot coding are now similar. That is, the orthographic neighbourhood has changed for the spatial coding model is quite different than for the interactive activation model.

Next: Orthographic neighbours in the spatial coding model