This paper is devoted to the study of bare nominal arguments (i.e., determinerless NPs occurring in canonical argumental positions) from a crosslinguistic point of view. It is proposed that languages may vary in what they let their NPs denote. In some languages (like Chinese), NPs are argumental (names of kinds) and can thus occur freely without determiner in argument position; in others they are predicates (Romance), and this prevents NPs from occurring as arguments, unless the category D(eterminer) is projected. Finally, there are languages (like Germanic or Slavic) which allow both predicative and argumental NPs; these languages, being the ‘union’ of the previous two types, are expected to behave like Romance for certain aspects of their nominal system (the singular count portion) and like Chinese for others (the mass and plural portions). This hypothesis (the ‘Nominal Mapping Parameter’) is investigated not just through typological considerations, but also through a detailed contrastive analysis of bare arguments in Germanic (English) vs. Romance (Italian). Some general consequences of this view, which posits a limited variation in the mapping from syntax into semantics, for current theories of Universal Grammar and acquisition are considered. 1. G OALS AND M AIN D ATA The current view of the syntax-semantics interface is based on the following set of guiding principles: (1) a. Syntactic categories at the relevant level of representation, say LF, are mapped onto corresponding semantic types (thereby determining for each expression what its denotation is going to be). b. Logical Forms are compositionally interpreted using a small set of universal rules (like functional application and abstraction). GENNARO CHIERCHIA Natural Language Semantics 6: 339–405, 1998.  1998 Kluwer Academic Publishers. Printed in the Netherlands. REFERENCE TO KINDS ACROSS LANGUAGES* * A first draft of this paper has been circulating since November 1996. Versions of it were presented at MIT, Going Romance, Salt VII, the 1997 LSA Summer Institute, DIPSCO, the University of Florence, and the University of Pisa. All of those audiences have greatly contributed to shaping the present, final draft. Especially important has been the feedback I got from the Discussion Group on Semantic Variation, organized by M. Diesing, S. McConnell-Ginet, and myself (for which support through NSF Grant SBR 9710984 is grate- fully acknowledged). I am moreover grateful to M. Bittner, C. Cecchetto, D. Fox, T. Guasti, J. Higginbotham, J. Huang, F. Landman, A. Li, P. Longobardi, A. Moro, and D. Pesetsky. Thanks also to S. Vovk for his expert help with Russian. The greatest debt of all I owe to the particularly detailed and thoughtful comments provided by V. Dayal, I. Heim, and M. Krifka. Regrettably I cannot confidently say that any of the people mentioned agrees with all of the specifics of what is presented here.