ABSTRACT
To enable a robot to interpret natural language instructions with regard to a specific object in a configuration, the robot needs to be equipped with a repertory of representations that human users might employ for reference. In this paper, we focus on the ways in which objects are referred to in relation to other similar objects, by analysing the results of an experiment involving a robot and human users unfamiliar with it. The robot is designed to understand reference via basic intrinsic and relative reference systems. Results showed that our implemented computational model could handle standard instances of spatial reference, but needed refinement to be able to interpret specific variations that were employed by users based on contextual factors analysed in this paper.
