1.   This gives a rough idea of the number of choices that have to be carried forward in order to produce word graphs of this size.

2.   If a match to long words fails, then the short words are retrieved from the buffer and stored on the word graph at the appropriate point.

3.   Since there are no long words that begin in this way, the short word I,, is retrieved from the buffer and stored on the word graph.

4.   Subsequently, since there are no long words beginning with the sequence, the short word am,, is retrieved and added to the word graph.

5.   Paths could then be continued from all new short words that are stored on the word graph.

6.   We tested two strategies for reducing the word graph which used different kinds of information.

7.   The experiments in the previous chapter showed the effects that word boundary ambiguity can have on the size of the word graph given different phonemic inputs.

8.   We can think of the word graph as a search tree with each path through the tree corresponding to each of the word strings.