Coordination and Neutralization in HPSG Roger Levy and Carl Pollard Stanford University and Ohio State University 1 Introduction Ingria (1990) pointed out that neutralization (also called indeterminacy), including the special case of feature value syncretism, seemed to pose a problem for unification-based grammar. Few lin- guists adopted Ingria’s proposal, which involved replacing unification in the relevant cases with a nondistinctness check, but the paper was influential in the sense that it persuaded many people that unification-based grammars and the constraint-based frameworks that superseded them were a hopeless case. In the mid-90’s, when Bayer and Johnson (1995) showed how to solve the prob- lems in question using Lambek’s additive meet type constructor, it made a big impact and helped persuade many people to abandon constraint-based grammar for type-logical grammar. In recent years type-logical accounts of coordination of unlikes have been closely linked to those of neutralization. On the approaches of Bayer and Johnson, and Heylen (1996), which build on earlier work by Morrill (1990, 1994), coordination is treated in terms of Lambek’s additive join constructor, so that coordination and neutralization are lattice-theoretic dual operations (where the lattice in question is the Lindenbaum algebra of the Lambek calculus, namely the free residuated lattice generated by the basic type symbols). In the prosodic interpretation (“frame semantics”), these operations correspond, respectively, to intersection and union of sets of word strings. In this paper, we use lattice theory to present an account of neutralization and coordination of unlikes that can easily be incorporated into current HPSG, where linguistic objects are modelled as totally well-typed, sort resolved feature structures. The account, which is consistent with the HPSG analysis of neutralization proposed by Levine et al. (2001), was developed by the alphabetically first author during autumn 2000, and the lattice-theoretic presentation by the alphabetically second author during winter and spring 2001. Another HPSG-based account of the same phenomena, developed independently by Daniels (2001) at about the same time, is shown by lattice-theoretic methods to be essentially equivalent to ours. In both our approach and Daniels’, neutralization and coordination turn out to be dual lattice operations, just as in the type-logical account; the lattices in question are, however, quite different from the type-logical one. Unfortunately, space limitations necessitate postponing detailed critique of the type-logical account, which we intend to provide elsewhere. The same applies to the recent LFG-based account of Dalrymple and Kaplan (2000), which employs set values in several different ways to model neutrality, coordination, ambiguity, multiple selection (e.g. of adjuncts), and ‘principled resolution’ of agreement features. Proceedings of the 8th International HPSG Conference, Norwegian University of Science and Technology (3-5 August 2001) Frank van Eynde, Lars Hellan and Dorothee Beermann (editors) 2002.CSLI Publications. http://csli-publications.stanford.edu/ 221