Proximity in Context: An Empirically Grounded Computational Model of Proximity for Processing Topological Spatial Expressions

Proceedings of the 21st International Conference on Computational Linguistics and 44th Annual Meeting of the ACL, pages 745–752, Sydney, July 2006. c⃝2006 Association for Computational Linguistics Proximity in Context: an empirically grounded computational model of proximity for processing topological spatial expressions∗ John D. Kelleher Dublin Institute of Technology Dublin, Ireland john.kelleher@comp.dit.ie Geert-Jan M. Kruijff DFKI GmbH Saarbru¨cken, Germany gj@dfki.de Fintan J. Costello University College Dublin Dublin, Ireland fintan.costello@ucd.ie Abstract The paper presents a new model for context- dependent interpretation of linguistic expressions about spatial proximity between objects in a nat- ural scene. The paper discusses novel psycholin- guistic experimental data that tests and verifies the model. The model has been implemented, and en- ables a conversational robot to identify objects in a scene through topological spatial relations (e.g. “X near Y”). The model can help motivate the choice between topological and projective prepositions. 1 Introduction Our long-term goal is to develop conversational robots with which we can have natural, fluent sit- uated dialog. An inherent aspect of such situated dialog is reference to aspects of the physical envi- ronment in which the agents are situated. In this paper, we present a computational model which provides a context-dependent analysis of the envi- ronment in terms of spatial proximity. We show how we can use this model to ground spatial lan- guage that uses topological prepositions (“the ball near the box”) to identify objects in a scene. Proximity is ubiquitous in situated dialog, but there are deeper “cognitive” reasons for why we need a context-dependent model of proximity to facilitate fluent dialog with a conversational robot. This has to do with the cognitive load that process- ing proximity expressions imposes. Consider the examples in (1). Psycholinguistic data indicates that a spatial proximity expression (1b) presents a heavier cognitive load than a referring expression identifying an object purely on physical features (1a) yet is easier to process than a projective ex- pression (1c) (van der Sluis and Krahmer, 2004). ∗The research reported here was supported by the CoSy project, EU FP6 IST ”Cognitive Systems” FP6-004250-IP. (1) a. the blue ball b. the ball near the box c. the ball to the right of the box One explanation for this preference is that feature-based descriptions are easier to resolve perceptually, with a further distinction among fea- tures as given in Figure 1, cf. (Dale and Reiter, 1995). On the other hand, the interpretation and realization of spatial expressions requires effort and attention (Logan, 1994; Logan, 1995). Figure 1: Cognitive load Similarly we can distinguish be- tween the cognitive loads of processing different forms of spatial relations. Focusing on static prepositions, topo- logical prepositions have a lower cognitive load than projective prepositions. Topological prepositions (e.g. “at”, “near”) describe proximity to an object. Projective prepositions (e.g. “above”) describe a region in a particular direction from the object. Projective prepositions impose a higher cognitive load because we need to consider different spatial frames of reference (Krahmer and Theune, 1999; Moratz and Tenbrink, 2006). Now, if we want …

Content truncated to 500 words