Database semantics (dbs) was developed originally to model the basic mechanism of natural language communication (Hausser 2001). This important application is extended here to another crucial task of cognitive modelling, namely the task of pattern completion during recognition. It consists in efficiently reconstructing a complex concept from its basic parts.

It is shown that the data structure of dbs, called a word bank, supports recognition by providing for any elementary concept matching part of a complex external object all potential candidates connected to it. In this way recognition is provided with a limited set of concepts to actively try to match against other parts of the complex object. Once a second basic concept has been matched successfully and connected to the first, the database provides a much smaller list of potential candidates for a third basic concept, etc. This algorithm converges very quickly.

The basic concepts are defined as geons, in accordance with the RBC (recognition by components) or geon theory by (Biederman, 1987). Geons are structures of a complexity intermediate between features and templates, e.g., cubes, spheres, or cylinders. The approach works also for the hypothetical reconstruction of the unseen side of a known object, explained by (Barsalu, 1999) on the basis of frames.