Grounding Coomputational Lingustics in AI Planning Mark Steedman, University of Edinburgh There is a long tradition associating language and other serial cognitive behavior with an underlying motor planning mechanism (Piaget 1936, Lashley 1951, Miller et al. 1960, passim). The evidence is evolutionary, neurophysiological, and developmental. It suggests that language is much more closely related to embodied cognition than current linguistic theories of grammar suggest. The fact of child language acquisition seems to require this representation to be symbolic. However, attempts to induce such a cognitive representation from raw sensory-motor robot data have been disappointing. It does not seem possible to duplicate the evolution of vertebrate cognition by sheer force of machine learning. The talk argues that practically every aspect of language reflects this connection transparently. Building on planning formalisms developed in Robotics and AI, with some attention to applicable machine learning techniques, two basic operation corresponding to seriation and affordance will be shown to provide the basis for both plan-composition in animals, and long-range dependency in human language, of the kind found in constructions like relative clauses and coordination. A connection that is this direct raises a further obvious question: If language is so closely related to animal planning, why don't any other animals have language? The talk will further argue that the specific requirements of human collaborative planning, involving actions like helping and promising that depend on a type of interpersonal relationship that seems to be lacking in other animals, provides a distinctively semantic precursor for recursive aspects distinguishing human language from animal communication. It will show that the automaton that is minimally necessary to conduct search for collaborative plans, which is of only slightly greater generality than the push-down automaton, is exactly the automaton that also appears to characterize the parsing problem for natural languages, at a new "near context-free" level in the Chomsky hierarchy..