This is the first year funding of a three-year continuing award. Natural Language learning is a paradigm problem for Artificial Intelligence and Cognitive Science. The world's languages differ significantly in grammar and conceptual structure but all are learned easily by children. This research continues the exploration of language acquisition in the restricted domain of simple geometric scenes and sequences. The methodology is structured connectionist modelling with an extended version of back-propagation learning techniques. One critical task of this research is developing a system capable of learning individual spatial relation terms from any of a wide range of languages, as part of a broad inter-disciplinary effort on natural language acquisition. It includes the learning of grammatical rules and more complex sentences and adding inferential capabilities to the existing system. //
