An Influence Model for Reference Object Selection in Spatially Locative Phrases

Michael Barclay and Antony Galton

In Christian Freksa, Nora S. Newcombe, Peter Gärdenfors, and Stefan Wölfl, Spatial Cognition VI: Learning, Reasoning, and Talking about Space, Proceedings of International Conference Spatial Cognition 2008, Freiburg, Germany, September 15-19, 2008, Springer, pages 216-232
ISBN 3-542-87600-6

Abstract

A comprehensive influence model for choosing a reference object in a spatially locative phrase is developed. The model is appropriate for a Bayesian network implementation and intended as a step toward machine learning of spatial language. It takes its structure from the necessary steps a listener must take in utilising spatial communication and contains as variables parameters derived from the literature concerning characteristics of landmarks for wayfinding as well as reference objects in general. Practical limitations on the implementation and training of the model are discussed.

Online version


Antony Galton
Last modified: Mon Mar 29 13:06:51 BST 2010