Modeling Inferences in Mental Models
 

Klaus Eyferth, Robin Hörnig, Ute Schmid, Sylvia Wiebrock, Fritz Wysotzki

TU Berlin, Department for Informatics

Duration: 1996 - 2000

Project homepage: http://ki.cs.tu-berlin.de/~sppraum/spacecog.html
 

This project pursues three purposes: First, an experimental psychological inquiry into the representation of spatial relations given in texts. Second, the development of a computer based model representing spatial relations that are given by or inferred from information in a propositional format, as a contribution of artificial intelligence (AI). Third, the comparison and mutual evaluation of the presuppositions as well in the psychological as in the AI model. The psychological model draws upon the theory of mental models, proposing that a recipient reconstructs spatial textual relations in an analogous mental representation. The experiment takes up and revises suggestions of N. Franklin, B. Tversky and colleagues (c.f. 1990). The AI model defines and explores inference algorithms promising to be more efficient than approaches based on classical logic. This model uses geometric transformations, as used e.g. in computer graphics and robotics (Ambler & Popplestone 1975), in combination with constraint satisfaction. Spatial scenes are represented on an intermediate level by labelled graphs. The nodes of the graph are labelled by coordinate systems, representing objects introduced in the text. The arcs are labelled by transformations of coordinate systems, coding spatial prepositions given in the text. Spatial relations not available from the text can then be computed (inferred) by finding appropriate paths in the graph, multiplying the coordinate transformations along the path and matching the product with the definitions of spatial prepositions.
The project as a whole is set to develop a model of spatial inferences that is consistent with psychological evidence and allows for computer implementation according to clear and parsimonious principles.

Approaches: empirical investigation; modeling; implementation
Areas of Research: artificial intelligence; cognitive psychology; linguistics
Topics: mental models; language; inference
 

Project publications