Identifying Negative Exemplars in Grounded Language Data Sets

There has been substantial work in recent years on grounded language acquisition, in which paired language and sensor data are used to create a model of how linguistic constructs apply to the perceivable world. While powerful, this approach is hindered by the difficulty of obtaining unprompted negative examples of natural language annotations. In this paper, we describe an initial pilot of a system that uses natural language similarity metrics to automatically select negative examples from a paired corpus of perceptual and linguistic data.