Representation of semantic information is going to be a crucial issue as we are going to have large numbers of modules or agents, which have to communicate in a sufficiently flexible way. The complexity of the systems which can be put together, will be limited by our ability to interface and interconnect modules. If this is going to depend only upon standards specifications, this limit will be reached early.

In addition to standards with respect to "bits, bytes and formats", representations of information have to be organised in such a way that they allow an adaptive connection of information channels. We may view this as some form of a semantic "plug-and-play" capability. This requires that information implies a specification of a distance metric for objects and events (aspects of similarity - dissimilarity), in a general fashion.

A typical situation is that one module sends control signals to another module. From the response derived from that other module, the first one can infer what functionality, scaling, etc is involved.

In addition, semantic representations have to carry a component of certainty or confidence, which allow a flexible combination of data from several sources and relations to models. This allows an update of confidence of statements given different situations of data and model support. This is necessary for efficient handling, computation and search of semantic or cognitive information. It is also necessary for adaptive combination of information and fusion of information from several sources.

Today most handling of cognitive information content is done in language form using keywords and dictionaries. This will not prove sufficient for development of highly flexible and intelligent systems.

Goesta H. Granlund	Phone: +46 13 281303
Department of Electrical Engineering	Fax: +46 13 138526
Computer Vision Laboratory	email: gegran@isy.liu.se
Linkoeping University
S-581 83 Linkoeping, Sweden