Coping with Topological and Directional Relations Based on 3-Dimensional Geometric Representations

Peter Wazinski, Gerd Herzog,

SFB 314 - Project VITRA
Universität des Saarlandes
D-66041 Saarbrücken
Germany

Appeared in: Spatial Concepts: Connecting Cognitive Theories with Formal Representations, Workshop Notes, 10th ECAI, Vienna, Austria, 1992.

Introduction

In the VITRA (Visual Translator) project we are concerned with natural language access to visual data. Our previous work on integrating computer vision and natural language processing focused on 2-dimensional representations of the spatial environment. The aim now is to generalize our former work in order to deal with 3-dimensional geometric scene representations.

Two main applications are under investigation:

  1. The automatic generation of incremental natural language route descriptions taking into account the changing visual context while following the route
  2. Natural language dialog with an autonomous mobile robot about its spatial environment in order to control the positioning of the robot system and its effectors
For both scenarios, a detailed 3-dimensional geometric representation forms the starting point for the definition and representation of spatial relations that in turn are used for the analysis and synthesis of spatial reference expressions in natural language.

The Two-Level Approach for Topological and Directional Relations

The theoretical framework of our research is the idea of two-level semantics that originates in the work of Bierwisch (Bierwisch (1983)). The conceptual system functions as a representational mediator for various cognitive subsystems such as the visual, auditory and -- of course -- the linguistic system. Within the linguistic system the meaning of a lexical item is represented as a grammatically determined, item-specific formation of certain parameters, the values of which are available on the conceptual level. Thus, the conceptual and the linguistic system are independent, and conceptual knowledge is represented in a language-independent way. For directional prepositions and dimensional adjectives this approach was elaborated in Lang (1991) and Lang et al. (1991). Herweg (Herweg (1989)) used the same theoretical basis for topological prepositions.

Both, Lang's and Herweg's approaches to spatial expressions are purely qualitive. Quantitative aspects like the distance evaluation between the reference object and the object to be located or linguistic hedges (``X is fairly behind Y.'') are not taken into account. For these quantitative aspects of spatial descriptions we intend to adopt our own work on applicability degrees (Herzog et al. (1990), Wazinski (1992)).

The mathematical basis for Herweg's approach to topological prepositions is algebraic topoplogy. Within the context of geographic information systems Egenhofer has devoloped a more elaborate method that also relies on algebraic topology and provides the formal definition of (binary) topological relations as well as an inference mechanism for these relations (c.f. Egenhofer (1991)).

Spatial Reasoning

Although we assume that fairly complete 3-dimensional geometrical representations of the domain under discussion will be available the task of reasoning about spatial information can not be neglected:

As mentioned above, Egenhofer provides a framework for reasoning about topological relations. Work on qualitative reasoning for directions has been done by Frank (Frank (1991)). Hernandez (Hernández (1991)) gives an account for both, topological and directional reasoning. None of these approaches deals with possibly graded applicabilities of spatial relations or coverings of objects. Thus, in order to exploit applicability degrees on the level of base relations and to cope with coverings of objects, these approaches will have to be extended.

Conclusion

Our intention is to develop an integrated approach for directional and topological relations in order to provide natural language access to visual data on the basis of a 3-dimensional geometric representation.

In our talk we will show how to put together the various building stones of related and previous work for the directional and topological relations. In addition, we will present first results on the unified inference mechanism for spatial reasoning about both types of relations.

References

M. Bierwisch.

Semantische und konzeptuelle Repräsentation lexikalischer Einheiten. In: R. Ruzicka, W. Motsch, eds., Untersuchungen zur Semantik. Studia grammatica 22, pp. 61-99, Akademie-Verlag, Berlin, 1983.

M. J. Egenhofer.

Reasoning about Binary Topological Relations. In: O. Günther, H.-J. Schek, eds., Advances in Spatial Databases, pp. 144-160, Springer, Berlin, Heidelberg, 1991.

A. U. Frank.

Qualitative Spatial Reasoning with Cardinal Directions. In: H. Kaindl, ed., 7. Österreichische Artificial-Intelligence-Tagung, pp. 157-167, Springer, Berlin, Heidelberg, 1991.

D. Hernández.

Relative Representation of Spatial Knowledge: The 2-D Case. In: D. M. Mark, A. U. Frank, eds., Cognitive and Linguistic Aspects of Geographic Space, pp. 373-385, Kluwer, Dordrecht, 1991.

M. Herweg.

Ansätze zu einer semantischen Beschreibung topologischer Präpositionen. In: C. Habel, M. Herweg, K. Rehkämper, eds., Raumkonzepte in Verstehensprozessen: Interdisziplinäre Beiträge zu Sprache und Raum, pp. 99-127, Niemeyer, Tübingen, 1989.

G. Herzog, T. Rist, E. André.

Sprache und Raum: Natürlichsprachlicher Zugang zu visuellen Daten. In: C. Freksa, C. Habel, eds., Repräsentation und Verarbeitung räumlichen Wissens, pp. 207-220, Springer, Berlin, Heidelberg, 1990.

E. Lang.

A Two Level Approach to Projective Prepositions. In: G. Rauh, ed., Approaches to Prepositions, pp. 127-167, Narr, Tübingen, 1991.

E. Lang, K.-U. Carstensen, G. Simmons.

Modelling Spatial Knowledge on a Linguistic Basis: Theory--Prototype--Integration. Springer, Berlin, Heidelberg, 1991.

J. R. J. Schirra.

Ein erster Blick auf ANTLIMA: Visualisierung statischer räumlicher Relationen. In: D. Metzing, ed., GWAI-89: 13th German Workshop on Artificial Intelligence, pp. 301-311, Springer, Berlin, Heidelberg, 1989.

P. Wazinski.

Generating Spatial Descriptions for Cross-modal References. In: Proc. of the 3rd Conf. on Applied Natural Language Processing, pp. 56-63, Trento, Italy, 1992.




Gerd Herzog
Last update: Sun Nov 10 00:52:35 MET 1996


Send comments to herzog@acm.org