With the event-based phonotactic description of German given in the previous section, a flexible knowledge component has been provided for the syllable parser in the ASL project which caters not only for actual syllables of the language but also generalises to all potential syllables. This is the type of linguistic information which is needed below the word level in order to be able to cater for new forms which do not belong to the scenario corpus. It may be useful to distinguish at this level already between attested forms (i.e. those in the corpus) and unattested forms which are permitted by the phonotactics. With this type of information, the syllable parser together with the morphophonological parser can optimise underspecified phonetic input below the word level thus reducing the search space required at the levels above. Furthermore, phonotactic knowledge can be used to guide event recognition at the acoustic-phonetic level, by supplying information as to which class of events is expected next. Optimisation of event data delivered by the acoustic-phonetic can be made by asking for further information on a particular event based on what is known at the phonological level. It seems that it is this type of knowledge component which goes some way to solving the projection problem in speech recognition as discussed in section 1.