Teaching a robot spatial expressions

Describing the semantics of natural language spatial expressions such as (1) ‘moving forward slowly’ and (2) ‘you’re in front of the desk’ is not a straightforward task. In this paper we describe a setting with a mobile robot where the meanings of such expressions are learned from a set of robot data and natural language descriptions made by a human commentator. We start with simple robot-centered spatial expressions like (1). These do not make reference to the environment external to the robot. We then extend the learning to prepositional expressions like (2) which denote relations between the objects in the environment.