Discussion and Conclusion
This chapter gives an overview and a brief discussion of the results obtained in the previous chapter and it will also present possibilities for future research in this area.
The first part of the discussion involves general results of both networks which do not depend on the different fixation frequencies. The second part of it will look in more detail at the differences and similarities between the control and the fixation net.
From the two simulations (i.e. the one with a full training corpus and the one
with a smaller training corpus), it can be concluded that both the control and the fixa- tion network have accomplished the general task of learning the grapheme-to-phoneme mappings of most of the words in the training corpus. Both networks get around 98.6% of all the words in the training corpus correct. The overwhelming majority of errors were made on low frequency words, which is as expected since the networks have seen these words only a very limited number of times. The tests with different word lists (i.e. regular, irregular and ambiguous words) showed that both networks have success-