Grounded Language Learning in Robots-An Information-Theoretic Approach to Adjective Learning

Abstract

Grounded language learning refers to viewing the language learning process as forming mappings between words and concepts of entities in the physical world. It is important for embodied agents, like factory robots, that have to learn to interact in the real world and communicate with human coworkers. We are concerned with scenarios where robots learn the correct words in an online fashion by interacting with humans that talk about referents while using them.