1.   One of the major areas of research in speech recognition concerns the ways in which general linguistic knowledge can compensate for such errorful or ambiguous acoustic input.

2.   The problem is not just one of poor acoustic input but also of ambiguous word boundaries.

3.   The representation must allow us to compute a link between descriptions of acoustic input and stored descriptions.

4.   Moreover it may sometimes be the case that the acoustic input is simply insufficient to discriminate between hypotheses.

5.   The hope is that they occur infrequently and that most would be eliminated after a word or two by mismatches to the acoustic input.