ROLF NOYERCLITIC SEQUENCES IN NUNGGUBUYU AND PFCONVERGENCEABSTRACT. Although X0movement places clitics within sentences, ordering withinclitic clusters must be readjusted at PF to conform to nonsyntactic ordering requirements.This paper addresses exactly how the conflicting demands of syntactic ordering and purelymorphological ordering are mediated, with data from Nunggubuyu (Australia, documentedby Heath (1980, 1982, 1984)). Two mechanisms are proposed. First, a pre-Spell-Out dele-tion of features by means of Impoverishment rules (Bonet 1991, 1995) bleeds the insertionof certain clitics. Impoverishment rules also feed and bleed one another, giving rise tocomplex surface patterns. Second, clitics which survive Impoverishment must move inMorphology from their abstract (syntactic) position to conform with surface ordering re-strictions. Heath formalized these movements as Affix Hopping transformations, but thispaper shows that Hopping is always local, and can thus be modelled as an instance of Mor-phological Merger (Marantz 1988). Where clitic sequences cannot be properly reorderedby local movements, clitic deletion applies as a Last Resort to allow PF convergence. Theproposed locality restriction explains a battery of seemingly unrelated clitic deletions in aprincipled way.Perlmutter (1971) drew attention to surface constraints in the ordering ofRomance clitic sequences, first posing the question of how output con-straints on clitic orderings are imposed in a generative grammar. Sincethat time two lines of thought have emerged. According to one view,clitic sequences are base-generated in the lexicon, and their ordering isa non-syntactic, morphological peculiarity imposed either by lexical sub-categorization (Inkelas 1993) or templates (Simpson and Withgott 1986).Other accounts similarly propose that clitic clusters merge together ab-stractly at PF and are then arrayed post-syntactically onto a spell-outtemplate (Bonet 1991, 1995). The alternative view, espoused in manysyntactic treatments such as Kayne (1989 et seq.), Haverkort (1993) orUriagereka (1995) contends that the answer is to be found in the syn-tactic component: clitic order reflects directly some aspect of derivationalhistory, such as the order that clitics head-adjoin to their host functionalI would like to thank Morris Halle, Alec Marantz, David Embick, and three anonym-ous reviewers for many helpful comments. I also extend a special thanks to Ken Hale forhis knowledge, guidance and encouragement. All errors remain my own.Natural Language & Linguistic Theory 19: 751–826, 2001.© 2001 Kluwer Academic Publishers. Printed in the Netherlands.752 ROLF NOYERprojection. The syntactic view has met with a certain degree of success,but as Uriagereka (1995) admits, some additional mechanism of PF re-ordering appears inescapable. While some reordering is driven by prosodicconsiderations, as argued by Halpern (1995), Bonet (1991) concluded in adetailed study of Catalan clitic ordering that there remains a residue ofpure idiosyncrasy to clitic ordering. The same conclusion was drawn byCummins and Roberge (1994) in a survey of clitic ordering in Romance.A natural question to pose at this juncture is whether some combinationof syntactic conditions and PF-well-formedness constraints together giverise to complex ordering phenomena, and more important still, how suchconflicting conditions are mediated.In this article I address this question directly through an analysis ofclitic sequences in Nunggubuyu, a language of Northern Territory, Aus-tralia, meticulously documented by Heath (1980, 1982, 1984). With 22different categories of pronominal arguments, Nunggubuyu exemplifiescomplexities of clitic ordering and neutralization easily more complex thanthat of Romance languages such as Catalan, but in principle no different.Drawing on many of Heath’s original insights, I show that clitic orderingmust conform both to syntactic conditions as well as to specific morpho-logical conditions on ordering. Abstract clitic morphemes are positionedsyntactically, but the phonological actualizations of these morphemes mustmove in the mapping from their Spell-Out positions to their ultimate sur-face positions to conform to PF ordering restrictions. These movementswere formalized by Heath (1984) as Affix Hopping transformations muchas in Chomsky (1957). The primary advance over Heath’s original analysisthat I propose here is to show that Hopping obeys important conditionsof locality, specifically, the conditions of MORPHOLOGICAL MERGERas proposed in Marantz (1988). Where clitics cannot move locally to aposition licensed at PF, the derivation crashes. To permit a convergent de-rivation in the sense of Chomsky (1995), Last Resort clitic deletion mustapply. The locality condition adduced plays a role in a battery of seem-ingly unrelated clitic deletion phenomena, and accounts for the absence ofsemantically expected clitics in roughly 40% of the logically possible argu-ment combinations in Nunggubuyu. Alternative descriptions of the sameset of facts yields a heterogeneous set of rules of little interest. Instead,the strict locality of reordering provides evidence that the mapping fromSpell-Out to PF is constrained by universal principles of some interest.These principles reside in a post-syntactic component of AutonomousMorphology such as envisioned in the theory of DISTRIBUTED MORPHO-LOGY (Halle and Marantz 1993, 1994), the essential premises of which areadopted here.CLITIC SEQUENCES IN NUNGGUBUYU AND PF CONVERGENCE 753Not all cases of clitic disappearance can be explained by the localityrestriction on clitic movement. Instead, some clitic sequences are unreal-izable owing to rather specific rules of neutralization or clitic antipathy.Syntax appears to have no role in explaining these neutralizations, which Itake to be part of the periphery with respect to syntax proper. On Heath’soriginal analysis some of these neutralizations were accomplished by rulesdeleting clitics, but the present account differs in situating these neutraliz-ations at the featural, pre-Spell-Out level. Following Bonet (1991, 1995)I propose that such neutralizations are accomplished by IMPOVERISH-MENT rules, strictly morphological rules deleting certain values before thephonological realization of functional projections. Although these rules arelargely particular to Nunggubuyu, they nevertheless interact in an intricateway, giving rise to a complex pattern of surface