Language & Learning for Robots

Robot technology will find wide-scale use only when a robotic device can be given commands and taught new tasks in a natural language. How could a robot understand instructions expressed in English? How could a robot learn from instructions? Crangle and Suppes begin to answer these questions through a theoretical approach to language and learning for robots and by experimental work with robots.

The authors develop the notion of an instructable robot—one which derives its intelligence in part from interaction with humans. Since verbal interaction with a robot requires a natural language semantics, the authors propose a natural-model semantics which they then apply to the interpretation of robot commands. Two experimental projects are described which provide natural-language interfaces to robotic aids for the physically disabled. The authors discuss the specific challenges posed by the interpretation of "stop" commands and the interpretation of spatial prepositions.

The authors also examine the use of explicit verbal instruction to teach a robot new procedures, propose ways a robot can learn from corrective commands containing qualitative spatial expressions, and discuss the machine-learning of a natural language use to instruct a robot in the performance of simple physical tasks. Two chapters focus on probabilistic techniques in learning.

Patrick Suppes (1922-2014) was the Lucie Stern Professor of Philosophy, Emeritus at Stanford University. He was the founder of the Computer Curriculum Corporation and the Suppes Brain Lab at Stanford, as well as the co-founder of the Institute for Mathematical Studies in Social Sciences.