The integration of preference-based strategies into the new uniform model will be one of the main research direction in the near future.
We have mentioned several times the importance of preferences for natural language processing in the main chapters of this thesis (as well as for the EBL method) and we have been careful to avoid obstacles to this important future direction. The agenda mechanism of the uniform algorithm, for example, is already an important prerequisition for the incorporation of such strategies, since it allows processing of new items in any order. Also the architecture of the item sharing approach has been designed to support preference-based control.
The strategies described in [Uszkoreit1991] and [Barnett1994] seem to be suitable candidates for the new uniform environment. The work described in [Uszkoreit1991] is of importance since it focusses on the integration of preferences with the feature system of a constraint-based grammar as an appropriate means for obtaining plausible performance models. In [Barnett1994] a model is described that is able to handle specific preferences for parsing and generation, as well as shared preferences.
It is reasonable to assume that both strategies (even together) can be integrated into the new uniform model. If so, it would also be possible to realize a sort of preference-based monitoring strategy. We assume that the NLS in which the uniform model is integrated maintains different preference spaces for parsing and generation. Preference-based monitoring would then mean that the derivation of a produced utterance is directed so that it is consistent with respect to the assumed preferences of the interlocutor which have been used to direct parsing (clearly, this presumes that the NLS has as its disposal a user and discourse model). For example, if both prefer minimal-attachment of pp-modifiers then an utterance like ``Remove the folder with the system tools'' (with meaning ``Remove the folder by means of the system tools'') would cause no revision.