A Study on how Humans Describe Relative Positions of Image Objects

Information describing the layout of objects in space is commonly conveyed through the use of linguistic terms denoting spatial relations that hold between the objects. Though progress has been made in the understanding and modelling of many individual relations, a better understanding of how human subjects use spatial relations together in natural language to is required. This paper outlines the design and completion of an experiment resulting in the collection of 1920 spoken descriptions from 32 human subjects; they describe the relative positions of a variety of objects within an image space. We investigate the spatial relations that the subjects express in their descriptions, and the terms through which they do so, in an effort to determine variations and commonalities. Analysis of the descriptions determines that common elements of spatial perception do indeed exist between subjects, and that the subjects are quite consistent with each other in the use of spatial relations.