Learning Grammatical Constructions from Narrated Video Events for Human-Robot Interaction

The ability to provide natural language interaction with humanoids will likely require adaptive systems that learn the mappings from speech to meaning in a context dependant manner. The objective of this research is to develop a system for miniature language learning based on a minimum of pre-wired language-specific functionality, that is compatible with observations of perceptual and language capabilities in human development. In the proposed system, meaning is extracted from video images based on detection of physical contact and its parameters. Mapping of sentence form to meaning is performed by learning grammatical constructions that are retrieved from a construction inventory based on the constellation of closed class items uniquely identifying the target sentence structure. The resulting system displays robust acquisition behavior that reproduces certain observations from developmental studies, with very modest “innate” language specificity.