1.   We can view the space of possibilities at the lexical access level as a horizontal slice through the search space, representing a graph of word hypotheses.

2.   The syntactic and semantic components must select from word hypotheses that are constantly changing their status or score as more acoustic information becomes available.

3.   However, word hypotheses compete on the Chart, not in the lexicon, so this similarity of status should not matter from a processing point of view.

4.   Such a strategy would have to be very carefully designed, however, if it were not to result in a huge explosion of word hypotheses.

5.   LA passes several hundred word hypotheses to the higher level components, most of which are false positives.

6.   However, the grammar must be able to correctly distinguish word hypotheses or the number of paths will grow exponentially.

7.   Since partial interpretations did not have to be kept distinct over stretches longer than two words the combinatorial explosion of word string hypotheses should have been considerably reduced.