3.1.1 The Input Layer

The network has been trained by showing it one word at a time. (Note 2) Each input word is represented by 85 units, 60 of which are used for the orthographic representation, while the other 25 units represent the meaning of the word concerned.

Note 2
It would also have been possible to just represent one letter at a time and let the network find out about the words for itself, but I felt that this extra task would have proven little: previous research has already shown that neural networks can learn to detect words from a continuous stream of letters (Elman 1990; Norris 1992).