Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

The linearization of affixes : evidence from Nuu-chah-nulth Wojdak, Rachel 2005

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
831-ubc_2005-105887.pdf [ 11.6MB ]
Metadata
JSON: 831-1.0092361.json
JSON-LD: 831-1.0092361-ld.json
RDF/XML (Pretty): 831-1.0092361-rdf.xml
RDF/JSON: 831-1.0092361-rdf.json
Turtle: 831-1.0092361-turtle.txt
N-Triples: 831-1.0092361-rdf-ntriples.txt
Original Record: 831-1.0092361-source.json
Full Text
831-1.0092361-fulltext.txt
Citation
831-1.0092361.ris

Full Text

THE LINEARIZATION OF AFFIXES: EVIDENCE FROM NUU-CHAH-NULTH  by RACHEL WOJDAK B . A . (Hon), The University of British Columbia, 2000  A THESIS S U B M I T T E D IN P A R T I A L F U L F I L M E N T O F THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY in THE F A C U L T Y OF G R A D U A T E STUDIES (Linguistics)  T H E UNIVERSITY OF BRITISH C O L U M B I A October 2005  © Rachel Wojdak, 2005  ABSTRACT This dissertation addresses the linearization of affixes, and argues for a particular model of the way in which syntax maps to phonology. According to the proposal, syntax is spelled-out to phonology in minimal cycles equivalent to a single application of syntactic Merge (cf. Epstein et al. 1999). I term this proposal the local spell-out hypothesis. The empirical grounds on which this hypothesis is assessed is Nuu-chah-nulth (Nootka), a Southern Wakashan language spoken in British Columbia, Canada. Nuu-chah-nulth has a class of morphologically bound predicates termed affixal predicates which participate in a linearization strategy of suffixation. I claim that affixes in Nuu-chah-nulth are linearized at spell-out with respect to 'hosts' as a consequence of the PF requirement that utterances be sequentially ordered. Spell-out induces in Nuu-chah-nulth a relationship which I label PF Incorporation. The affixal predicate 'incorporates' its host in order to achieve a pronounceable form, that of a linearized affix. An affixal predicate in Nuu-chah-nulth consistently suffixes to a host chosen from its derivational sister, its complement.  This suffixation pattern is subject to a string adjacency  effect: an affixal predicate incorporates only the leftmost element from its complement, which happens to be contiguous with the affixal predicate. I present the argument that the local spellout hypothesis elegantly captures this dual sensitivity to derivational sisterhood and linear adjacency.  Although the spell-out mechanism which induces linearization of affixes is non-  syntactic, syntactic sisterhood conditions linearization opportunities at PF via the composition of local spell-out domains. This dissertation presents a variety of diagnostics for Nuu-chah-nulth clausal structure, with a particular focus on the argument structure of affixal predicates.  Amongst affixal  predicates which take nominal complements, predicates range from unaccusative to extended unaccusative, transitive to ditransitive. Unergatives, however, are systematically absent from the inventory of affixal predicates in Nuu-chah-nulth. Amongst affixal predicates which take verbal complements, both modal and main predicates are found. A consequence of the analysis is that syntax is 'phonologized' over the course of the derivation, in minimal stages induced by application of Merge. Linearization is thus established in increments.  This analysis has implications for the grammatical locus of head movement  operations: head movement is not strictly phonological (contra Chomsky 1995, 2001).  ii  T A B L E OF CONTENTS  Abstract Table of Contents Abbreviations Dedication Acknowledgements  ii iii vii viii ix  CHAPTER 1 Introduction  1  1.0 Introduction 1.1 Theoretical assumptions 1.1.1 Interface requirements 1.1.2 'Bottom-to-top' syntactic derivation 1.1.3 Syntax 'all the way down' 1.2 Linearization 1.2.1 Proposal: local spell-out 1.2.2 Alternative spell-out models 1.3 Affixal predicates in Nuu-chah-nulth 1.4 Research context 1.4.1 Previous literature 1.4.2 Methodology 1.5 Outline of the dissertation  1 4 4 5 6 7 7 10 13 18 18 20 21  CHAPTER 2 PF Incorporation  23  2.0 Introduction 2.1 Morpho-phonological dependency 2.2 Linearization is local 2.2.1 Iterativity 2.2.1 Modifier incorporation 2.2.1.1 Adjective incorporation 2.2.1.2 Adverbial incorporation 2.2.2 Coordinated objects.. 2.3 Insensitivity to syntactic category 2.4 Absence of L F effects 2.4.1 Discourse transparency 2.4.2 Absence of scopal effects 2.5 Edge effects 2.5.1 DP edges and the bare nominal requirement 2.5.2 C P edges and 'restructuring' effects 2.5.3 Comparison to 'phasal' spell-out 2.6 Conclusion  23 28 33 36 39 39 45 49 52 54 55 56 57 59 60 61 66  iii  CHAPTER 3 Clausal architecture of Nuu-chah-nulth  68  3.0 Introduction 3.1 Linearization of syntactic terminals 3.1.1 A universal linearization scheme? 3.1.1.1 Conceptual problem 3.1.1.2 Empirical problems 3.1.2 Directionality is determined at spell-out 3.1.2.1 Directionality conventions 3.1.2.2 Comparison to headedness 'parameters' 3.1.2.3 Affixation patterns 3.1.2.4 Directionality conventions of Nuu-chah-nulth 3.2 Configurationality 3.2.1 Partial head marking 3.2.2 A n asymmetry in possessor raising '• 3.2.3 A n asymmetry in incorporation 3.3 Predicate-initial word order 3.3.1 Proposal: 'Basic' predicate-initial order (VOS) 3.3.2 Evidence for right-branching specifiers 3.3.3 Evidence against verb-raising 3.3.4 Evidence against raising of verbal phrase 3.3.5 Implications for the linearization of affixal predicates 3.3.5.1 The 'complement' effect 3.3.5.2 Evidence against directionality of affixation 3.4 Cliticization domains 3.4.1 DP domain : 3.4.2 C P domain 3.5 Conclusion  68 69 70 72 73 73 74 76 76 77 79 79 81 83 84 87 89 90 94 95 96 97 99 99 102 104  v  '.  CHAPTER 4 Nominal complements of affixal predicates  105  4.0 Introduction 4.1 Predication configurations 4.1.1 Basic structures 4.1.2 Transitivization 4.1.3 Flexibility in theta-role mapping 4.1.4 Predicate inventory 4.1.5 Unergatives are necessarily non-affixal 4.2 Unaccusatives 4.2.1 Incorporation 4.2.2 Intransitivity in Nuu-chah-nulth 4.2.3 Absence of unergative affixal predicates 4.2.4 'Long' possessor raising as a diagnostic for unaccusativity 4.3 Transitives 4.3.1 Incorporation 4.3.2 Subject agreement 4.3.3 Possessor raising  105 108 108 109 111 112 113 115 115 117 118 120 128 129 131 132  iv  4.4 Extended unaccusatives 4.4.1 Locatum predicates 4.4.1.1 Incorporation 4.4.1.2 Subject agreement 4.4.1.3 Possessor raising 4.4.2 Location predicates 4.4.2.1 Incorporation 4.4.2.2 Subject agreement 4.4.2.3 Possessor raising 4.5 Ditransitives 4.5.1 Locatum-type ditransitives 4.5.2 Location-type ditransitives 4.6 Outstanding issues 4.7 Conclusion  135 139 140 142 143 144 145 146 147 148 148 150 152 155  CHAPTER 5 Verbal complements of affixal predicates  157  5.0 Introduction 5.1 Two classes of 'verb incorporating' affixal predicates 5.1.1 Affixal main predicates 5.1.2 Affixal auxiliary predicates 5.1.3 Affixal auxiliaries are non-thematic 5.1.3.1 Rigid vs. alternating complementation 5.1.3.2 Ability to license a 'mismatched' subject 5.1.3.3 Citation forms 5.1.4 Summary 5.2 PF Incorporation occurs only in infinitival environments 5.3 Absence of clausal morphology 5.3.1 No tense 5.3.2 No person/mood agreement 5.3.3 No complementizer 5.4 Lack of clause-boundedness effects 5.4.1 'Long' w/i-movement 5.4.2 'Long' possessor raising 5.4.2.1 'Long' possessor raising with affixal auxiliaries 5.4.2.1 No 'long' possessor raising with affixal main predicates 5.5 Conclusion  157 158 158 159 161 162 162 166 166 167 170 170 171 172 173 174 176 177 179 182  CHAPTER 6 Implications  184  6.0 Introduction 6.1 Summary 6.2 Theoretical implications 6.2.1 On the nature of PF operations 6.2.2 The head movement question 6.2.2.1 Linearization is purely phonological 6.2.2.2 Syntactic head movement is more complex 6.2.3 Local spell-out creates 'outside-in' dependencies 6.2.3.1 Tnside-out' dependencies are not possible via local spell-out 6.2.3.2 Evidence from Kwakw'ala (Northern Wakashan) 6.2.3.3 Affixation as primitive or derived  184 184 186 186 187 188 191 193 194 194 195  v  r  6.3 Typological implications 6.3.1 The typology of noun incorporation 6.3.2 The governing/restrictive hypothesis in Wakashan 6.3.3 Lexical suffixation as an areal feature of the Pacific Northwest 6.4 Conclusion  197 198 199 201 202  References  204  Appendix Appendix Appendix Appendix  215 220 221 228  A: Background on the Nuu-chah-nulth language B: Orthography C: Inflectional paradigms (Ahousaht dialect) D: Additional example sentences  vi  ABBREVIATIONS  ABS AGR ASP AUG AUX BEN CAUS COMP COND CONF CONJ CONT DEIC DEP DET DIM DIR DUR FUT GO HAB IMP IND INST INV IPOSS IREL ITER L LOC NEG NOM OBJ PAS PERF PL POSS PRE PS PST  Q QUOT R RD REL RES RL  absolutive mood agreement aspect augmentative auxiliary benefactive causative complementizer conditional mood confirmation mood conjunction continuous deictic dependent mood determiner diminutive directive durative future tense "go" imperative habitual imperative indicative mood instrumental invisible inalienable possessive "indefinite" relative mood iterative vowel lengthening locative negative nominaliser object passive perfective plural possessive prefix possessive mood past tense interrogative mood quotative mood reduplication round relative pronoun resultative (simple) relative mood  S SG SPOR SUB SUF TR TEMP TENSE  vii  vowel shortening singular sporadic subordinate mood suffix transitivizer temporal tense  This dissertation is dedicated to all of my hundred and twenty babies.  viii  ACKNOWLEDGEMENTS The path to this dissertation was in many ways a circuitous one, and so my gratitude spans a lot of ground. M y greatest thanks are due to the speakers of Nuu-chah-nulth who shared with me their time, their humour, and their love for their language: Mary Jane Dick, Katherine Fraser, Carrie Little, Archie Thompson, Barbara Touchie, Barney Williams Jr., Barney Williams Sr., Christine Nicolaye, Sarah Webster. Their patience and dedication was humbling, and I am indebted to them for the chance they gave me to learn so much. It was a true privilege to be welcomed into their lives and their homes, and I enjoyed the time I spent with them. Thanks in particular to Mary Jane for being a tireless consultant, a devoted researcher, and my determined teacher from the very start back in January 2000. I am also especially grateful to Katherine for her leadership over the years, and for letting me see the world beyond Ahousaht with her help in starting the dialect survey. It can be a challenge to bridge the gap between theoretical and practical linguistics, but I hope that the examples of the Af-verbs in this dissertation will be of some use as documentation of the language, and so a start to being able to give something back. This dissertation is written in a technical style, but really it is just about how to form sentences using those special /b-verbs. It would often delight my Nuu-chah-nulth consultants when they heard me speak a full sentence in their language, since - asthey pointed out - if a mama-hii could learn how to do this, then there is hope that anyone can. I know I have a long way to go before completely understanding the language, but it is my hope that parts of this dissertation will be able to be transferred to a practical use as a guide for how to speak in those "full sentences", which are often very long words. At the very least, the aim of this dissertation is to do justice to the elegance that nuucaanui- speakers have shown me characterises their language. The members of my committee - Henry Davis, Rose-Marie Dechaine, Martina Wiltschko - were instrumental in shaping the content and presentation of this dissertation. Henry and Rose-Marie often had opposite effects on the evolution of this project. I appreciate Henry for giving me great latitude, and equally appreciate Rose-Marie for hauling me back in. Martina, to her credit, can be thanked for doing a share of both. Consequently, the ideas of this dissertation were given the freedom and discipline they needed to grow. I'd like to thank all three of my committee members for letting me inhabit their minds from time to time, and for the effort they spent reading and reflecting on my many handouts and drafts for this project. As my thesis supervisor, I particularly owe Henry not only for ensuring that the U B C Nuu-chah-nulth project was relatively well-funded over the past few years, but also for influencing my approach to fieldwork. Most of all, I can thank Henry for seeing promise in me that I didn't really know that I had. Henry was the very first person to encourage me to do a Ph.D. - a possibility that might not have occurred to me if he hadn't. M y meetings with Rose-Marie convinced me that the topic of affixal predicates was a fruitful one, and her critiques made this a much stronger project. This project also benefited greatly from Martina's sharp questions and reliable feedback. Thanks to Florence Woo, Naomi Sawai, Eun-Sook Kim, and Christine Ravinski for their company in the universe, and for all those Nuu-chah-nulth "support group" sessions that always went longer than we planned. Fieldwork trips would never have been as much fun without the absurd sing-alongs and Christmas carols: thank you, Florence & Christine, and wiwisfaqkuksis p/ispis\ The origins of this project are in joint work with Florence when we started working on the language together; data and theory discussions with her continued to be very important to me over the years. I am also grateful to the Wakashanists outside U B C who've helped provide an exciting intellectual environment to work in. Thanks especially to Adam Werle for helping promote a sense of community amongst Wakashanists. Thanks to Matt Davidson, Henry Kammler, Terry Klokeid, Ryan Waldie and all the participants in thefirst-everWakashan Linguistics Conference for an invigorating exchange of ideas.  ix  A number of members of the UBC Linguistics department have provided thoughtful advice on various aspects of this project over the years, inside and outside of research seminars and formal presentations. I'd especially like to single out Felicia Lee, Gunnar Hansson, Doug Pulleyblank, and Lisa Matthewson. It was an honour to be a part of the U B C department. I am also grateful to Eric Reuland for giving me the opportunity to visit Utrecht University in early 2003, and Marisa Rivero for welcoming me to the University of Ottawa as a visiting student in 2004-2005. Each provided me with key feedback, and introduced me to opportunities I would have been without at U B C . Many other people were very kind to pass on helpful advice at different stages of this project, including Peter Ackema, David Adger, Wallace Chafe, Marianne Mithun, Henk van Riemsdijk and Susi Wurmbrand. The questions and comments I received from participants of the CLA, GLOW and ICSNL conferences were particularly encouraging to me. On the fieldwork side, I would like to thank the folks at UVic, particularly Leslie Saxon and Su Urbanczyk, for allowing the U B C troupe's access to UVic department facilities for Victoria elicitation sessions. Inside and outside of the U B C Linguistics department, I've spent the past five years in the company of some truly inspiring samples of humanity. I am grateful for everything I learned from you all. Carrie G(illon), Linda T(amburri Watt) and Leora B(ar-el) were the heart of the department during my formative years at U B C , and I thank them for their advice and perspectives, and many good times. Thanks to Carolyne Albert for lots of support and shared meals. Thanks to Jane Ellis for being a loyal friend from way back. Thanks to Joanna Sio for adventures in Vancouver and in Amsterdam, even though we never did get around to talking that much linguistics! Thanks to Dot Chow for her exuberance. Thanks to Andrea Gormley for good talks, rabbit-pancakes, and Pendulum peanut butter squares in the early years. Thanks to Gessiane Picango for her friendship. Thanks to Solveiga Armoskaite for her wisdom and for whatever that Ethiopian bread was called. Cheryl Wiramanaden and Brian de Alwis deserve credit for being my all-time favourite brunch hosts. Thanks to Siva Athreya for ameliorating my view of humanity and for his legendary dinner parties in my first year of grad school. I am grateful to all my fellow grad students for helping make the U B C Linguistics department the amazing place it is. I couldn't imagine these years without the personalities and work of Oladiipo Ajiboye, Leszek Barczak, Yunhee Chung, Clare Cook, Suzanne Gessner, Yoko Ikegami, Peter "Jacbos" Jacobs, Michele Kalmar, Masaru Kiyota, Jeff Muehlbauer, Sugunya (Add) Ruangaroon, Kayono Shiobara, Scott Shank, Olga Steriopolo, James Thompson, and Ian Wilson. Thanks to Edna Dharmaratne for being incredible and for keeping us all under control. In Ottawa, thanks especially to Dana Geber and Anousha Sedighi for providing such a warm welcome. Thanks to my family, for their support and for trying to understand what on earth it is that I do: my mom, dad, Becky, Graham, Hugh, Oma & Nana. Finally, I owe a great deal to my husband Antal Jdrai for the comforts and adventures he's added to my life over the past few years. I'm tremendously lucky to have had such a thoughtful and curious non-linguist audience for my dissertation "puzzles". Nothing forces j clarity of reasoning faster than having to explain one's linguistic analyses to a Hungarian mathematician, and I can directly credit my edes for some of the conceptual progress I made along the way. I gratefully acknowledge funding for my fieldwork from the Jacobs Research Fund (Whatcom Museum Society, Bellingham WA), Phillips Fund (American Philosophical Society), U B C Hampton Fund Research Grant in the Humanities and Social Sciences (to Henry Davis), SSHRC 410-1998-1597 (to Henry Davis), SSHRC 410-2002-1715 (to Lisa Matthewson) and SSHRC 410-2002-1078 (to Martina Wiltschko). M y post-graduate studies were financially supported by SSHRC doctoral fellowship 752-2002-1855, an Izaak Walton Killam pre-doctoral fellowship, a Patrick David Campbell Fellowship, and a University Graduate Fellowship.  x  CHAPTER 1  Introduction What we shall call the beginning is often the end And to make an end is to make a beginning. The end is where we start from. ~ T . S . Eliot Four Quartets no. 4  1.0 Introduction This dissertation proposes a new understanding of the means by which affixes in natural language come to be linearized. Affixal elements, whether prefixal or suffixal, share the property of requiring a 'host' with which they may form a word. An affix is not permitted to stand on its own. For example, the English morpheme un- must occur as an affix (specifically, a prefix), and not as an independent word. (1)  a. b.  I am unhurried. * I am un.  cf.  I am not rushed.  cf.  I am not.  It is often said that morphology governs affixation, as it imposes restrictions on the internal composition of words. It is this aspect of the grammar which distinguishes between 'bound' morphemes (such as un-), which form subparts of words, and 'free' morphemes (such as not), which are permitted as independent words. The system of syntax, on the other hand, can be understood to be the means by which words are grouped together to form larger phrases. This dissertation presents an argument that syntax plays a determining role in the combinatory properties not only of words, but of affixes (cf. Baker 1988, Lieber 1992, among others). According to the analysis, affixes have a syntax which is indistinct from that of nonaffixes. I propose that the local syntactic environment of a morpheme conditions the available patterns of affixation.  Under this view, affixes are subject to the same sorts of structural  relationships in the syntax as are non-affixes. However, morphemes with affixal status introduce a tension to the grammar in a way in which non-affixal elements do not. If one or the other of a or P in the syntactic construct in (2) is an affix, then a host for the affix must be found in order for the expression to be pronounced.  a What this study undertakes is an examination of the way that this phonological 'neediness' of affixes is resolved by the grammar.  I will argue that affixation derives a restricted set of  1  linearizations for syntactic constructs such as (2). linearization of oc-P results.  If, for example, a is a prefix, then a  If, however, a is a suffix, then an ordering of p-oc arises. The  consequence of this analysis is that phonological considerations induce linearization on inherently unordered syntactic entities (cf. Chomsky 1995, Fox and Pesetsky to appear). The language of investigation for this study is Nuu-chah-nulth (Nootka), a member of the Wakashan family spoken in British Columbia, Canada. Nuu-chah-nulth has a class of affixal predicates which participate in a linearization strategy of suffixation. In the following example, the affixal predicate -caas "bet" suffixes to the nominal kikuuk "dishes". The affixal predicate and its gloss are indicated by highlighting. (3)  ki^uukcasitnis?aa+ ki9ajuk-caas-mit-niis-?aaF  huu?ak?uyi huu?ak-?uyi  dishes-bet-PST-1 PL.IND-HAB  early-ago  We always used to bet dishes long ago. Affixal predicates in Nuu-chah-nulth are obligatorily bound, and may never occur as nonsuffixes.  This is illustrated by the example in (4), which shows that it is impossible for the  affixal predicate -caas "bet" to appear without a host which it may suffix to.  In this  ungrammatical example, the nominal kifcuuk "dishes" occurs separately from the affixal predicate. (4)  * caasitnis?aaF kifcuuk huu?ak?uyi caas-mit-niis-?aaF ki&uuk huu?ak-?uyi bet-PST-1 PL.IND-HAB dishes early-ago We always used to bet dishes long ago.  This dissertation develops a Minimalist approach to the linearization of affixal predicates in Nuuchah-nulth and argues that suffixation provides a means of satisfying the requirement that linguistic outputs be linearized.  According to the proposal, Nuu-chah-nulth is a 'proof-of-  concept' for the hypothesis that linearization is induced in stages corresponding to strictly minimal syntactic units (cf. Epstein et al. 1998).  This analysis employs only the smallest  necessary constructs for the syntax-to-phonology mapping. The linearization of affixal predicates in Nuu-chah-nulth may be observed to induce a 'displacement effect'. Displacement effects are ubiquitous in natural language, and can be found at various levels in the grammar.  When displacement applies at a syntactic level, words or  phrases are 'dislocated', such that they are pronounced in one position in a sentence although they are interpreted in another. For example, in the English sentence Who did Kyle see? the word who is felt to be the object of the verb see, even though it occurs in initial position within the clause and not in canonical object position following the verb (parallel to Sarah in the sentence Did Kyle see Sarah?). In such contexts, transformational linguists analyse the fronted word as  2  having undergone an abstract form of movement.  Considerable syntactic research lies in the  examination of the displacement properties of words and larger constituents (eg. Ross 1967, Chomsky 1977). Dislocation may also be examined at a sub-word level. In the case of Nuu-chah-nulth suffixation, the 'piece' of word which acts as a host for an affixal predicate is removed from the post-verbal position it would otherwise occupy. In (5a), the nominal ?aapinis "apple" serves as a host for the affixal predicate - He "consume". Non-nominal hosts are also available to affixal predicates.  In (5b), the modifier ha?um "tasty" acts as a host for the affixal predicate - 'He  "consume", while the nominal Paapinis follows the initial predicate complex. (5)  a.  ?aapiniyic?is?a+ ?aapinis-4ic-?iis-?a+ apple-consume-3.lND-PL They are eating apples.  b.  ha7um?ic?is?aF ?aapinis ha?um-'iic-?iis-?aF ?aapinis tasty-consume-3.lND-PL apple They are eating delicious apples.  I will argue that the post-verbal position of the nominal Paapinis "apple" in (5b) is the standard orientation for objects in Nuu-chah-nulth. Thus, the ordering of Paapinis "apple" as a pre-verbal host for the affixal predicate in (5a) represents a departure from the general linearization pattern. What drives the displacement properties of affixation? This question is to be addressed here from the perspective of the linearization of affixal predicates in Nuu-chah-nulth. In a sense to be made more explicit over the course of this chapter, suffixation in Nuu-chah-nulth is triggered by the need to pronounce an affix. Affixes, unlike free morphemes, are bound elements which cannot be pronounced in isolation. When an affixal predicate attaches to a host, its boundedness requirement is met, and a well-formed word results (cf. Lasnik 1981). It is a consequence of the affixal predicate's status as a suffix that the internal components (host and affix) of the resulting word are fixed in a particular (host-initial) order. The next sections of this introductory chapter are organized into four parts.  The  following section, §1.1, presents an overview of the Minimalist framework which is employed for the analysis of affixal predicates. The theoretical back-drop is further developed in §1.2, in which the linearization of syntactic constructs is discussed. This leads to the central claim of this dissertation, introduced in §1.3: that the affixation pattern of Nuu-chah-nulth is a reflex of the linearization of linguistic outputs.  In §1.4, I situate the discussion of Nuu-chah-nulth affixal  predicates within the broader context of Wakashan linguistics. Finally, §1.5 outlines the form which the remaining chapters of this dissertation takes.  3  1.1 Theoretical assumptions This section lays out the Minimalist theoretical framework which this dissertation adopts. Following Chomsky (1995, 2001), I pursue a strongly derivational approach to syntactic structure.  1.1.1 Interface requirements The grammar is charged with the task of delivering linguistic expressions which are serviceable to two external systems: the system of thought, and the sensorimotor system (Chomsky 1995, 2001). The grammar thereby allows spoken languages to map an abstract form to meaning and to sound.  According to the Minimalist approach, a linguistic expression exiting the generative  system is viable only if it meets the interface requirements imposed by these external levels. In Chomsky's terminology, linguistic expressions must be 'legible' to each interface level, Logical Form (LF) and Phonetic Form (PF). (6) Model of the grammar lexicon spell-out  LF  PF  meaning sound In this system, the lexicon acts as the source of the elements which enter the computation. The lexicon codes the semantic, syntactic, and phonological properties which are specific to each lexical item.  Lexical items enter the computation from the lexical array known as the  numeration. Syntactic structures are composed using the lexical building blocks provided by the numeration, and are interpreted at the L F and PF interfaces at the point of spell-out. (7) Spell-out: map syntax to the interfaces Spell-out to the PF portion feeds the outputs of syntactic structure-building (Merge and Move) to the phonological system, where they receive a pronounceable form. With the exception of the interfaces at L F and PF, no other levels exist in the Minimalist grammar. Representational levels such as 'deep structure' and 'surface structure', which existed in earlier principles-and-parameters models, are eschewed in favour of a more barebones model which contains only the conceptually necessary meaning/sound interfaces. This move away from representational levels corresponds to the minimalist ideal of paring down the grammar to those elements which are necessary design features of natural language. A l l constraints are abandoned save those which hold at the interfaces and "are motivated by the properties of the interface" 4  (Chomsky 1995: 171).  These interface requirements are known as bare output conditions:  external to the syntax, these interpretative conditions are mandated by the requirements of the conceptual and sensorimotor systems, and ensure that the requirements of lexical items are met over the course of the derivation.  1.1.2 'Bottom-to-top' syntactic derivation This dissertation adopts from Chomsky (1995, 2001) the notion that syntactic derivations are built up from 'bottom-to-top', through successive applications of two concatenative operations: Merge and Move. Merge operates on elements selected from the numeration, and conjoins pairs of items in a binary fashion: (8)  Merge: concatenate a with P, forming y  If X and Y are merged, the category label of one of these conjoined elements is projected. For example, in (9), the category of X is projected as X(P). (9)  Merge (X, Y) XP X  Y  Merge applies iteratively, building a syntactic structure by pairing the output of a prior instance of Merge with a lexical item freshly introduced from the numeration. In the following representation, Z is added to the structure of (9) via an additional application of Merge. (10)  Merge (Z, XP) ZP  A l l binary merger creates two sisters - a pairing which Epstein et al. (1998) label derivational sisterhood. In the trees above, [X, Y] are derivational sisters, as are [Z, XP]. The operation of Move (or 'remerge') parallels Merge in that it also pairs two syntactic objects and projects a single category label (Kitahara 1994, 1995; Epstein et al. 1998). Move differs from pure Merge, however, in that it re-inserts a syntactic object already introduced in the derivation, rather than selecting a new item from the numeration.  Like Merge, Move is an  instance of binary concatenation. In this dissertation, I will abstract away from the differences between Move and Merge, and assume simply that Move can be captured by a restatement of the simple Move operation, such as in (11).  5  (11)  (re)Merge:  concatenate a (where a is an existing terminal) with 3, forming y  By reducing all operations of the syntax to operations of binary concatenation, non-branching nodes are eliminated from the syntax. That is, there will be no instances in which an element does not have a derivational sister (Epstein et al. 1998).  ^  1.1.3 Syntax 'all the way down' Following Halle and Marantz (1993) and other work in the Distributed Morphology framework, I assume that word formation parallels sentence formation in that both occur outside of the lexicon. There is no independent module for word-forming operations: 'morphology' is a cover term for syntactic or postsyntactic processes. That is, concatenation of morphemes may apply through syntactic processes of head movement, or it may be conditioned by the post-syntactic interface with PF (Embick and Noyer 2001). Under this view, a complex string such as the Nuu-chah-nulth sentence in (12a) has a syntactic structure similar to a sentence in which the individual morphemes are expressed as separate words, as in the English sentence in (12b). (12)  a.  ?aapiniyicmahsa?is?a+ ?aapinis-'iic-mahsa-?iis-?a F apple-consume-want.to-3.lND-PL They want to eat apples. ,  b.  They want to eat apples.  Syntactic structures for the Nuu-chah-nulth and English sentences are illustrated below.' (13)  a.  consume  apple  In each case, the morphemes occupy syntactic terminals. The claim of this dissertation is that affixation requirements of bound morphemes are responsible for the distinct linearization patterns of the sentences in (12). In Nuu-chah-nulth, affixes are linearized attached to a host with which they may form a word. English, which lacks affixal predicates parallel to -?iic "eat" or -mahsa "want to", linearizes these morphemes as independent words. According to the maxim of 'syntax  The trees in (13) abstract away from several syntactically relevant properties. For one, I assume that Nuu-chah-nulth makes use of a covert pronominal (pro) in cases such as (13a) in which the 3 person argument is phonologically null. The third person plural ending -7is(?a+) is not to be taken to be equivalent to English "they". I refer the reader to Chapter 3 for discussion of Nuu-chah-nulth clausal structure. rd  6  all the way down', the affixal or non-affixal status of the morphemes has no consequence for syntactic representation.  1.2 Linearization In the theoretical framework I have adopted, there are two sets of requirements which must be met over the course of the derivation by elements entering the syntactic computation. In the bifurcated model of the Minimalist grammar, requirements may be necessitated by the interface to L F , or the interface to PF. With respect to the PF branch, how must elements be arranged so that the sensorimotor systems can make use of them? Crucially, a (spoken) linguistic expression must be sequentially ordered so that it may be represented as a speech stream (Kayne 1997). Linearization is a bare v  output condition on PF (Chomsky 1995). A linearization scheme is not provided by the inherent mechanics of the syntax. As described in §1.1.2, syntactic structure-building reduces to two concatenative operations, Merge and Move, which are not inherently oriented for directionality. When Merge unites two elements, a and P, there is no restriction whether a must precede P, or whether a follows p. A l l that binary concatenation requires is that a combine with P, joining an unordered set of {a, P). Given the unordered nature of binary concatenation, the representations in (14) are therefore to be interpreted as syntactically equivalent. (14)  a.  y a  b. P  y P  a  1.2.1 Proposal: local spell-out If one or the other of the nodes (a, P) are understood to be an affix, then a specific type of linearization requirement is imposed on the orientation of these elements. I hypothesize that this requirement applies at spell-out, the point at which syntactic structures such as (14) take on a phono-temporal form. Affixation induces a particular linearization of terminal elements: if an element a is an affix, it must realised as a suffix (-a) or as a prefix (a-). Suffixation occurs when the affix is pronounced to the right of a host with which it forms a word; prefixation takes place when the affix forms a word with a host by attaching to its left. I do not consider 'infix' to be a distinct boundedness requirement, as I assume that infixation is reducible to either prefixation or suffixation. In Nuu-chah-nulth, for example, the plural 'infix' -t- (eg. Ti-t-niifc "dogs") can be analysed as a suffix which is positioned prosodically following the first syllable of its host (Stonham 1999, Wojdak 2002).  7  I take the choice of prefixation or suffixation for a given bound element to be a spell-out convention, relatable to language- or morpheme- specific considerations.  2  The logical  possibilities for linearization of the syntactic terminals in (14) are listed in the following table: (15) Linearization of bound and free elements  -a p-a  -p  a ap pa a-p  p-  p-a  p-a  p  aa-P a-p a-pp-a-  In three cases (shaded in the above table), there is more than one option available for linearization. When neither a nor p is an affix, affixation cannot serve as an ordering mechanism. When both a and P are suffixes, their relative orientation is indeterminate; the same sort of indeterminacy applies when both a and P are prefixes. In these indeterminate cases, one or the other of the affixes does not receive its required type of host. In the case of  (-a-p), for example,  the element -a is not bound as a suffix. Indeterminacy in this sense therefore entails that the affixation requirement of the elements is not met: no appropriate host has been provided for the affixes. As such, there is no interpretable output for the linearization of the affixes. The remaining, fully specified, orientations of affixes have just two surface realisations: a-p or P-a. For each of these linearization patterns, there are three distinct types of underlying bound/free dependencies. These structural ambiguities with respect to the surface forms of a-P and p-a are listed below. (16) Structural ambiguities with respect to outputs output: a. a, -p b. a-, p c. a-, -P  a-P (a is free; p is a suffix) (a is a prefix; p is free) (a is a prefix; p is a suffix)  output: d. a, Pe. -a, P • f. -a, P-  P-a (a is free; P is a prefix) (a is a suffix; P is free) ( a i s a suffix; P is aprefix)  Thus, although the syntactic device of binary concatenation provides no instructions for linearization, a restricted set of linearizations arises when the merged element is an affix. This linearization is, by its very nature, non-syntactic: the syntax itself can be assumed to be  As described in Chapter 3, the status of an affix as a prefix or a suffix is determined for a language learner during the process of acquisition, from salient evidence in the input. This can be considered to be on par with syntactic headedness conventions, a topic I return to in Chapter 3.  8  unordered. I refer to the means by which the relative ordering of affixes is fixed as local spellout.  This mechanism provides the input to interpretation at PF (and LF) based on strictly  minimal syntactic domains. This linearization mechanism is defined in the following statement: (17)  Local spell-out: for Merge (a, (3), spell-out (oc, (3)  This mechanism is 'local' in that it is hypothesized to apply at each minimal step of the syntactic derivation, to derivational sisters conjoined by Merge (see also Epstein et al. 1998, Matushansky to appear). It is inherently a pairwise function, because each step of the derivation is an operation of binary concatenation. In the Minimalist framework, the necessity of orienting an affix with respect to a host is a consequence of spell-out to PF. Affixes require linearization so that the arrangement may be phono-temporally ordered. A n earlier formulation of this affixation requirement is the Stranded Affix Filter of Lasnik (1981, 2000).  Although this filter does not make reference to phono-  temporal sequencing, it does capture the notion that a derivation is not viable if an affix does not find a host. A mechanical apparatus for affixation is supplied by the Morphological Merger operation of Marantz (1988, 1989; see also Bobaljik 1994), and its more recent incarnations, Lowering and Local dislocation (Embick and Noyer 2001).  In each of these variations, an  affixation rule forces two elements to 'switch places', with a single word resulting. (18)  Morphological Merger: X .... Y —» [Y + X]  In the Minimalist program, bare output conditions are adopted instead of derivational filters such as the Stranded Affix Filter or rules such as Morphological Merger.  The local spell-out  hypothesis provides a Minimalist alternative to these filter- or rule-based mechanisms for affixation.  3  In Chapter 2, I return to discussion of how the local spell-out model captures  affixation possibilities not handled by Morphological Merger. In the PF branch, spell-out necessarily induces linearization: linguistic outputs must be linearized. Thus, the two elements treated by local spell-out inevitably undergo linearization with respect to each other. According to the local spell-out hypothesis, the PF branch receives directions to resolve affixation requirements each time the syntactic tree is expanded, as indicated in (19). We can take the diagram in (19) to be the output of three successive applications of Merge: the first uniting a and (3 (Merge a, (3); the second uniting 5 and y (Merge 8, y); the third uniting 8 and <j) (Merge 6, <j)).  An additional type of output 'filter' on the positioning of affixes are Optimality Theory constraints on alignment. In Optimality Theory, constraint rankings determine whether an affix is right-aligned to a host as a suffix by AlignR[Affix] or left-aligned to a host as a prefix by AlignL[Affix]. A minimalist solution for the positioning of affixes with respect to hosts is developed in Chapter 3. 3  9  (19) Iterative application of local spell-out local spell-out  e  local spell-out 8  1 a  local spell-out  <  P  For each application of Merge, local spell-out applies to the derivational sisters. The interface requirements of elements introduced in the derivation are thus subject  to aggressive  interpretation, in which PF and L F needs are assessed at each step of the derivation. This entails that syntactic outputs are richly 'phonologized' over the course of the derivation, via addition of directions for pair-wise interpretations at PF (Epstein et al. 1998). According to the proposal, spell-out enriches the derivation, due to iterative interpretations at the interfaces. This echoes the argument of Fox and Pesetsky (to appear), who hypothesize that the sole function of spell-out is to add information. By their Order Preservation hypothesis, information established in one cycle of spell-out is never deleted over the course of the derivation. This determines that linearization is established cumulatively: Each time a new Spell-out domain D' is constructed, Spell-out linearizes the material in D' and adds information about its linearization to the information cumulatively produced by previous applications of Spell-out. (Fox and Pesetsky to appear) Fox and Pesetsky argue for the Order Preservation hypothesis in the context of successive-cycle syntactic movement, assuming a model in which spell-out applies at certain designated syntactic nodes, rather than at each instance of Merge. This timing of spell-out contrasts with the present proposal, in which local spell-out applies incrementally, at each step of the syntactic derivation. The next section compares the local spell-out hypothesis to alternative models of the timing of spell-out.  1.2.2 Alternative spell-out models There are three logical possibilities for the point at which spell-out occurs. Mapping of the syntax to the interfaces may apply once, multiple times, or at every stage possible in the derivation. The third option corresponds to the local spell-out hypothesis.  10  (20) Timing of spell-out a.  SINGLE SPELL-OUT HYPOTHESIS:  root branch node  b.  MULTIPLE SPELL-OUT HYPOTHESIS:  multiple branch nodes  c.  LOCAL SPELL-OUT HYPOTHESIS:  every branch node  This section briefly compares the characteristics of each model, and discusses a unique prediction of the local spell-out hypothesis. If spell-out is restricted to applying to the root branch node, then spell-out applies once per utterance.  This represents the traditional view, in which mapping between syntax and  phonology takes place at a single point, after the completion of the syntactic derivation (Chomsky 1995). In the following diagram, spell-out applies at 7C, the root node. (21) Single spell-out hypothesis 7t  spell-out  a  P  The multiple spell-out hypothesis was introduced as an alternative to the single spell-out hypothesis (Uriagereka 1999, Chomsky 2001). In a multiple spell-out system, it is proposed that spell-out occurs once the derivation reaches certain designated syntactic nodes, such as CP, vP and possibly DP (Chomsky 2001, Fox and Pesetsky to appear). Spell-out thus applies cyclically over the course of a derivation. This is represented abstractly in the following diagram, in which spell-out applies at the node 7t, as well as at the mid-derivational point y. Spell-out does not apply at every branching node: for example, in (22), spell-out does not occur at the point <j). Instead, 8, the terminal node of <j), is not spelled-out until the higher spell-out node of 7t is reached. (22) Multiple spell-out hypothesis n  spell-out  spell-out a  p  In the terminology of Chomsky (2001), spell-out is induced at distinct phases of the derivation. A node such as y belongs to a phase separate from the mid-derivational point (|). A noted  11  challenge for the multiple spell-out hypothesis is the empirical task of determining which derivational points are equivalent to spell-out nodes, and which are not (Legate 2003, Matushansky 2005). Multiple spell-out entails that phonological operations have access to mid-derivational units formed by syntactic structure-building. By this view, PF and L F operations are limited by the same derivational units which constrain the syntax, because mid-derivational constructs created by the syntax are translated simultaneously to the PF and L F components (Chomsky 2001; Svenonius 2001, 2004). The local spell-out hypothesis takes this isomorphism one step further by imposing phonological-semantic integrity at each step of the derivation (Epstein et al. 1998). Where the local spell-out and multiple spell-out hypotheses diverge is the stages at which spell-out to the interfaces obtains. In a multiple spell-out system, only certain syntactic nodes such as the sisters to CP, vP (Chomsky 2001) and possibly DP (Fox and Pesetsky to appear) are spell-out domains. With the local spell-out hypothesis, there is no such stipulation. Instead, PF and L F requirements of lexical elements are assessed aggressively over the course of the derivation: interpretation at the interfaces applies after each step of the syntactic derivation, rather than in larger derivational chunks. A consequence of the local spell-out hypothesis is that derivational sisterhood is predicted to be a uniquely privileged relationship for relationships established at spell-out. Specifically, lexical requirements of elements must be met at the strictly minimal stage of the derivation in which only the element and its derivational sister are present. Let us take the linearization of affixes as an illustration of this prediction. By hypothesis, the linearization of affixes applies at spell-out, due to the PF requirement that linguistic outputs be temporally ordered. According to the local spell-out hypothesis, if an element a takes p as its derivational sister, then a is necessarily linearized at spell-out with respect to p. If a is lexically specified as an affix, then for the tree in (23), spell-out of y induces a linearization in which the affix a is linearized with P as its host. (23) Local spell-out of derivational sisters spell-out of K  71 <•  9  0  <  spell-out of 0  spell-out  5 a  12  of y  According to the local spell-out hypothesis, elements which are not the derivational sister of a are ineligible to act as the host of a. Thus, it is predicted that elements such as 8 or 9 should not serve as the host for a: these elements are not derivational sisters of a. Only derivational sisters of affixes are present at the point of local spell-out of the affix. In contrast, for the single spell-out and multiple spell-out models, derivational sisterhood is not predicted to be a uniquely privileged relationship. Both the single and multiple spell-out models allow for the possibility that certain nodes are not spell-out points. In (24), for example, spell-out applies at (j), but it does not apply at y. (24) Delayed spell-out: single and multiple spell-out <> f 8  <•spell-out y  a  p  If a in (24) is an affix, then it is predicted that there are multiple elements which are available to serve as the host for the affix. A l l else being equal, if spell-out does not apply until <|), then single and multiple spell-out models predict that 8 or p should be equally eligible as hosts for a. Thus, for these models, the derivational sisterhood which a shares with p does not guarantee that a will be uniquely linearized relative to P at spell-out. In the next section, I introduce the empirical grounds on which the local spell-out hypothesis will be tested.  Over the course of this dissertation, it will be demonstrated that  affixation in Nuu-chah-nulth is sensitive to derivational sisterhood. Affixal predicates in Nuuchah-nulth find a host at spell-out which is chosen from the phrase with which they were syntactically merged.  Moreover, there is critical evidence that this sensitivity to derivational  sisterhood is not reducible to an independent effect such as directionality: in Nuu-chah-nulth, P is not chosen as the host for a simply because it is right-adjacent to a. Instead, I will argue that the Nuu-chah-nulth facts fall out elegantly from a model in which linearization consistently occurs between derivational sisters.  1.3 Affixal predicates in Nuu-chah-nulth Affixal predicates in Nuu-chah-nulth participate in two types of bound/free dependencies. A n affixal predicate (-a) suffixes to a free (P) or bound (P-) host.  In either case, the surface  realisation is P-oc. The place of these Nuu-chah-nulth linearization patterns within the typology of  y 13  bound/free dependencies is indicated in (25).  The Nuu-chah-nulth behaviour corresponds to  (25e-f). (25) Typology of bound/free dependencies output: a-p a. a, -P (a is free; P is a suffix)  output: p-a d. a, P- (a is free; P is a prefix)  b. a-, P (a is a prefix; P is free) c. a-, -P (a is a prefix; P is a suffix)  eV.-atP' (a is a suffix; P is free) f. -a, P- (a is a suffix; p is a prefix)  An illustration of the dependency of (25e) is supplied by the following examples. In (26a), the affixal predicate -siik "make" suffixes to the host -hicPin "dress", a free noun. In (26b),the affixal predicate -mahsa "want to" suffixes to the host wa-f-sik "go home (PERF)", a verbal complex which is likewise a non-affixal host. (26)  a.  nupititsa nupit-mit-sa once-PST-lSG.DEP I made a dress once.  -hicrinsiik +uc?in-siik dress-make  b.  wa-rsifcmahsak wa-fr-sifc-mahsa-k go.home-PERF-want.to-2SG.Q Do you want to go home?  Descriptively, these suffixation patterns may be labeled as 'noun incorporation', and 'verb incorporation', respectively. In the first case, the affixal predicate suffixes to (or 'incorporates') a noun, while in the second case, it suffixes to (or 'incorporates') a verb. However, despite the difference in these descriptive labels of noun and verb incorporation, both types of incorporation share an identical phonological dependency: the affixal predicate suffixes to a free host. In contrast, an example of the dependency of (25f) is given in (27). Here, the affixal predicate -iic "consume" suffixes to a bound nominal host, suuh- "spring salmon". This pattern of suffixation to a noun may also be considered to be a type of noun incorporation. (27)  suuwiicsis suuh-'iic-siis salmon-consume-1 SG.IND I'm eating salmon.  Superficially, this pattern resembles (26a), since in each case, the affixal predicate has suffixed to a noun. However, (26a) and (27) differ in that in (26a), the nominal host ivc?in "dress" is free, while in (27), the nominal host suuh- "spring salmon" is bound. Bound hosts in Nuu-chah-nulth, such as suuh- "spring salmon" in (26), belong to a closed-class set of nominals which have been referred to in the Nuu-chah-nulth literature as 'combining forms' (Rose 1981: 287, Davidson 2002). These bound nominals are often truncated  14  versions of free-standing nominals in the language (Rose 1981).  For example, the bound  nominal capx- "man" is a truncated allomorph of the free-standing nominal cakup "man". Other 'combining forms' are equivalent to the free form minus its aspectual or 'absolutive' affix (Rose 1981: 286-287). The bound nominal suuh- "spring salmon", for example, is related to the free form suuh-aa "spring salmon", which is inflected for continuative aspect -(y)aa (CONT). Another common pattern for bound nominals, in particular vowel- or nasal-final forms, is the stem plus a final -q- (Rose 1981). A n example of this type of alternation is taanaltaanaq- "money", in which the bound form ends with a -q-. For the bound/free alternants of Nuu-chah-nulth, the bound form occurs if and only if it is suffixed to an affixal predicate. However, not all free nominals have a bound alternant. In fact, for the youngest generation of Nuu-chah-nulth speakers, free nominal forms are often preferred over bound variants in the case of alternations with truncated allomorphs. For the remainder of this dissertation, I set aside the issue of alternation of bound 4  and free nominals, and focus instead on the properties of affixal predicates. Affixal predicates in Nuu-chah-nulth do not show an allomorphic alternation with free predicates.  Instead, free predicates constitute a distinct class, unrelated in form to affixal  predicates. (28) Free and bound classes of predicates in Nuu-chah-nulth  I label this class of free elements independent predicates. While affixal predicates require suffixation to a host, independent predicates never occur as suffixes. For example, although an affixal predicate such as -siik "make" incorporates a host, this option is unavailable to a nonaffixal predicate such as rii-ci& "sew (PERF)".  In (29a), the affixal predicate -siik  "make"  suffixes to -fvcPin "dress". As shown in (29b), the independent predicate ni-cik "sew (PERF)" cannot suffix to this nominal. (29)  a.  "hic?insiikitsis -Fuc?in-siik-mit-siis dress-make-PST-1 SG.IND  I made a dress. As Nakayama (2000: 39) notes, this is testament to the productive nature of Nuu-chah-nulth incorporation. If affixal predicates and their hosts constituted lexicalized chunks, then such novel forms would be unanticipated.  4  15  b.  * +uc7innici&itsis 4aic?in-ni-cik-mit-siis dress-sew-PERF-PST-1 SG.IND  I sewed a dress. Affixal predicates require a host which they may suffix to. The ungrammaticality of (30a) arises because the affixal predicate -siik "make" fails to find an appropriate host.  Independent  predicates, in contrast, do not occur as suffixes. In (30b), the independent predicate ni-cifc "sew (PERF)"  appears separately from the nominal -hic?in?akqs "my dress".  (30)  a.  * siikitsis siik-mit-siis make-PST-1 SG.IND  +uc?in?akqs +uc?in-?ak-qs dress-POSS-lSG.PS  I made my dress. b.  nicifcitsis ni-cifc-mit-siis  -Kic?in?akqs Tuc?in-?ak-qs  sew-PERF-PST-1 SG.IND dress-POSS-1 SG.PS  I sewed my dress. Thus, affixal and independent predicates can be seen to have distinct behaviours. A similar distinction can be drawn between the affixal predicate -mahsa "want to" and the independent predicate Tapaak "willing to". As noted earlier, the affixal predicate -mahsa "want to" takes a verbal host, in a suffixation pattern I descriptively labeled 'verb incorporation'. This is shown in (31a), in which the affixal predicate suffixes to wa-f-si&"go home (PERF)". This example may be contrasted with the one in (31b), which shows the independent predicate iapaak "willing to".  As an independent predicate, Tapaak "willing to" is incompatible with being  linearized as a suffix. As shown in (31b), it is ungrammatical for the independent predicate to suffix to the verbal host wai--sik"go home (PERF)". (31)  a.  b.  wa+sifanahsak wa4--sifc-mahsa-k go.home-PERF-wantto-2SG.Q Do you want to go home? * wa+sMapaakk wa+-si9t-'Tapaak-k go.home-PERF-willing-2SG.Q Are you willing to go home?  The inverse pattern is shown in (32), in which the verb wa+-si9e"go home (PERF)" follows the predicate, as a separate word. A n affixal predicate is impossible in an environment such as (32), since this would entail that it would not be linearized as a suffix. The ungrammaticality of (32a) confirms that the affixal predicate -mahsa "want to" must appear as a suffix. In contrast, it is grammatical for the independent predicate iapaak "willing to" to occur in this environment. As (32b) shows, iapaak "willing to" is not a suffix.  16  (32)  * mahsak mahsa-k want.to-2SG.Q Do you want to b.  wa+si?c wa+-sik go.home-PERF go home?  Tapaakk Yapaak-k  wa+si& wa+-si?e  willing-2SG.Q  go.home-PERF  Are you willing to go home? I hypothesize that the classes of affixal and independent predicates are lexically differentiated. In particular, I propose that affixal predicates in Nuu-chah-nulth are marked in the lexicon with an affixation requirement, [affix]. (33)  [affix]:  a * co  This lexical specification states that the morpheme (a) is not equivalent to a phonological word (CO). The consequence of this requirement is that the affix requires a host with which it may form a phonological word. This lexical requirement must be met over the course of the derivation. Specifically, since this lexical requirement involves morpho-phonological instructions, this requirement must be met in the PF branch. Sample lexical entries for affixal and independent predicates in Nuu-chah-nulth are given in (34-35), which indicate the properties of the predicates which are idiosyncratic to the lexical items.  5  (34) lexical entries for affixal predicates a.  -siik "make":  I  MAKE  /siik/ affix b.  -mahsa "want to":  W A N T TO  /mahsa/ affix (35)  lexical entries for independent predicates a.  /w "sew":  b.  fapaak"willing  I SEW/ /ml to":  WILLING T O  Aapaak/ To foreshadow the content of the following chapters, we will see how the affixation requirement of Nuu-chah-nulth is met at spell-out, resulting in a type of suffixation I term PF Incorporation. This incorporation process is sensitive to linear adjacency.  That is, an affixal  This abstracts away from the families of features (formal, semantic, phonological) which lexical properties fall into.  17  predicate incorporates any element which abuts it, showing an insensitivity to syntactic constituency (eg. the Coordinate Structure Constraint), as well as syntactic category. Yet, this condition on string adjacency will be shown to exist hand-in-hand with a 'complement' effect with incorporation: an affixal predicate only incorporates an element from its syntactic complement, and not from projections which c-command the predicate. I will argue that the local spell-out hypothesis allows an elegant means of reconciling these dual sensitivities to string adjacency and syntactic configuration. This dissertation explicitly argues against a 'hybrid' treatment of PF as a module operating both on hierarchically- and linearly-arranged constructs (contra Embick and Noyer 2001). Rather, I illustrate that the seemingly syntactic complement restriction on incorporation is a by-product of the derivation. By adopting the notion that syntax-to-PF mapping occurs at each instance of syntactic merge (Epstein et al. 1998), local spell-out is predicted to apply only to derivational sisters. This derives the complement effect as an epiphenomenal consequence of the mechanism of spell-out. We now turn to a discussion of how this study of the linearization properties of affixal predicates in Nuu-chah-nulth fits in with existing research on the Wakashan language family.  1.4 Research context Nuu-chah-nulth is a dialectally-diverse language spoken along western Vancouver Island in British Columbia, Canada (see  APPENDIX  A). The data presented here come from original  fieldwork on Ahousaht, a central Nuu-chah-nulth dialect spoken on Flores Island, off the west coast of Vancouver Island. The next sections present an overview of existing research on the language, followed by a discussion of the methodology for the present study.  1.4.1 Previous literature The linguistic study of Nuu-chah-nulth has been spurred by three major waves of fieldwork activity, conducted over the course of the last century. The roots of Nuu-chah-nulth linguistics are in the work of Edward Sapir, who engaged in intensive fieldwork on the language in 1910 and 1913-14. Working together with Nuu-chah-nulth speaker Alex Thomas and Sapir's one-time student Morris Swadesh, Sapir oversaw the collection of extensive textual materials, which were published in part as Nootka Texts (Sapir and Swadesh 1939). This project yielded a concise dictionary of the southern Nuu-chah-nulth dialect Tseshaht and a brief but foundational grammatical description of this dialect, published together as an appendix to Nootka Texts. A  18  second volume of texts was published as Sapir and Swadesh (1955), while additional linguistic notes appeared in Sapir (1911b, 1924, 1929), and Swadesh (1939, 1948). The material collected during the Sapir-Thomas collaboration forms the empirical basis for much subsequent analytic work on the language, including Stonham (1999, 2004), and the Nuu-chah-nulth component of Davidson (2002). A dictionary of the Tseshaht dialect has also been published (Stonham 2005), based on Sapir's fieldnotes. Textual materials which were residual to the original two published volumes of the Sapir-Thomas texts (Sapir and Swadesh 1939, 1955) were recently brought to light with the appearance of Sapir et al. (2000, 2004). The final two installments of the SapirThomas series are due to be published in the near future (Eugene Arima and Terry Klokeid, p.c). After a lull in linguistic fieldwork on Southern Wakashan stretching from the 1920s into the 1960s, a second wave of activity ensued with research expanding to Ditidaht (eg. Haas 1969, 1972; Klokeid 1976, 1978a,b) and Makah (eg. Jacobsen 1969, 1979; Renker 1987), as well as a subset of Nuu-chah-nulth dialects (Paik 1968, Rose and Carlson 1984, Kess and Kess 1986, among others). This period saw the completion of the first grammar of Nuu-chah-nulth, based on the northern dialect Kyuquot (Rose 1981). Recent years have seen a renewed period of concerted fieldwork, largely in response to the advancing age of the last fluent speakers of the Southern Wakashan languages. This ongoing wave of activity has to date produced three Ph.D. dissertations on Southern Wakashan: one is a treatment of grammatical properties of the Ahousaht dialect of Nuu-chah-nulth (Nakayama 1997), another addresses this dialect's morpho-phonology (Kim 2003), and the third is a comparison of Makah grammar to that of the Tseshaht dialect of Nuu-chah-nulth (Davidson 2002). Other publications include Nakayama (1998, 2001), Davis and Sawai (2001) and Wojdak (2001). This renewal of research activity is marked with the forthcoming publication of a special edition of the Canadian Journal of Linguistics devoted to Wakashan linguistics (Davis and Wojdak (eds.) in prep). The topic of affixal predicates - or 'lexical suffixes', as they have frequently been labeled - is a recurrent theme in research on Nuu-chah-nulth. The role of these morphemes in Nuu-chah-nulth was initially highlighted in Sapir's (1921) discussion of Nuu-chah-nulth polysynthesis, as well as in Swadesh's (1939) article entitled Nootka Internal Syntax. Inventories of these predicates are found in Sapir and Swadesh (1939), Rose (1981) and Davidson (2002), while detailed descriptive treatments of their suffixation patterns are discussed in Rose (1981) and Nakayama (1997, 1998, 2001). To my knowledge, the first reference to the word formation properties of Southern Wakashan affixal predicates as a type of incorporation is found in Klokeid (1976). A noun incorporation analysis in terms of syntactic head movement is developed in Woo  19  (2000), Davis and Sawai (2001), Y i u (2001), and Stonham (2004). A n alternative to the noun incorporation analysis is presented by Waldie (2004), who analyses the suffixation as a type of denominal verb formation. While suffixation to nominals ('noun incorporation') has been the primary focus of this previous literature, suffixation to verbs ('verb incorporation') has been relatively under-investigated.  Cursory treatments are found in Rose (1981) and Nakayama  (1997, 1998). The topic of Nuu-chah-nulth verb incorporation plays a central role in the present study.  1.4.2 Methodology This study is based on fieldwork conducted during 2000 to 2005 with fluent speakers of Nuuchah-nulth. Sentences and short stretches of discourse were elicited in a series of person-toperson interviews. This elicitation method is crucial to syntactic research as it allows for the targeted study of phenomena which may be rare in texts. Moreover, syntactic elicitations permit access to native speaker intuitions about grammaticality, which are otherwise inaccessible in the format of textual analysis. The data obtained in elicitation sessions were transcribed and subsequently inputted to a computer database. Some recordings were also made to supplement the transcriptions. Transcriptions were proofread by Mary Jane Dick, a fluent speaker of the Ahousaht dialect of Nuu-chah-nulth who has formal training in linguistics. The majority of the data in this dissertation has its genesis in a project I began in 2002 with Mary Jane Dick to document Ahousaht usages of the 'lexical suffixes' catalogued in Sapir and Swadesh (1939), Rose (1981) and Davidson (2002). Mary Jane Dick worked independently and with her mother, Sarah Webster, to create illustrative example sentences for more than two hundred suffixes (see APPENDIX D for a sample). These example sentences were an invaluable resource, and provided a large data set which formed the basis for follow-up elicitation sessions during 2002 to 2005.  Many of the 'lexical suffix' example sentences presented in this  dissertation were used as input to an Ahousaht dictionary project currently underway at the University of British Columbia. The two primary language consultants for this project are Mary Jane Dick and her mother, Sarah Webster, who are each native speakers of the Ahousaht dialect. Both were born in Ahousat, British Columbia - Sarah in 1924, and Mary Jane in 1945. Nuu-chah-nulth was the dominant household language while both were young, influenced by the fact that Mary Jane's grandparents and great-grandparents,  who Mary Jane spent about half her time with while  growing up, were monolingual Nuu-chah-nulth speakers. Both Sarah and Mary Jane began to learn English when they started schooling. Sarah attended residential school in Ahousat, while  20  Mary Jane went to a school in Ahousat until the age of twelve, when she moved away from her family to attend residential school in Port Alberni. Currently, the two usually see each other daily and try to use their language as much as possible with each other. This study benefited from elicitations with several secondary language consultants, who are speakers of Ahousaht or other Nuu-chah-nulth dialects. Katherine Fraser is a speaker of the Ahousaht dialect and has formal training in linguistic analysis. She is in her 60s. Caroline Little, also a speaker of the Ahousaht dialect, is in her 80s. Barbara Touchie, who is in her 70s, is a speaker of the Ucluelet dialect, while her older brother, Archie Thompson self-identifies as a speaker of the Toquaht dialect. Josephine Thompson, a speaker of the Ahousaht dialect, is in her 70s. Barney Williams, Sr., who is in his 80s, identifies with both the Tla-o-qui-aht and Chickliset dialects. His son, Barney Williams, Jr., is in his 60s and speaks the Tla-o-qui-aht dialect. Each of these language consultants speaks Nuu-chah-nulth as their first language, and English as their second.  Sessions with secondary language consultants were used to confirm various patterns  indicated by primary language consultants.  1.5 Outline of the dissertation There are five chapters which follow this introduction. This section gives an overview of their contents. Chapter 2 argues that local spell-out gives rise in Nuu-chah-nulth to a specific type of dependency termed PF incorporation. I argue that mid-derivational units constructed in the syntax form minimal spell-out domains for the post-syntactic linearization mechanism for affixal predicates. An affixal predicate is linearized as a suffix to the first word in its derivational sister, a condition I label the string adjacency effect. The linearization shows an insensitivity to syntactic category and an absence of L F effects. However, opacity effects come into play in that an affixal predicate cannot incorporate a host across a DP or CP. Instead, when an affixal predicate has a DP or C P as its complement, an expletive host surfaces to rescue the potentially stranded affix. Chapter 3 presents in detail the Nuu-chah-nulth clausal architecture which underlies PF incorporation. I argue that Nuu-chah-nulth has a syntactically underived V S O word order, with arguments introduced within the verbal projections.  The phenomenon of possessor raising is  introduced as a diagnostic for a syntactic asymmetry between subject and object. With respect to the linear ordering, I present the argument that linearization of syntactic terminals is achieved exclusively at PF, as syntactic constructs  are not inherently ordered for directionality.  Furthermore, I argue that syntactic c-command does not unambiguously determine linear  21  precedence (contra Kayne 1994). Instead, directionality is imposed at the point of spell-out via choice of one of two logically possible linearizations. Regularities in directionality (i.e. left or right syntactic 'headedness') are proposed to be spell-out conventions which arise through the process of language acquisition. Chapter 4 gives an inventory of the argument structures of affixal predicates which take nominal complements.  The structure-building operation of Merge generates the nominal  arguments of affixal predicates in a range of orientations, from unaccusative to extended unaccusative, transitive to ditransitive. Unergatives, however, are absent from the inventory of affixal predicates in Nuu-chah-nulth - a systematic gap which I attribute to the need for an affixal predicate to have a phonologically distinct complement. For each type of affixal predicate, the predicate incorporates a host from its complement, often giving rise to noun incorporation. I present a range of independent diagnostics for syntactic structure, including possessor raising and subject agreement. Chapter 5 surveys types of affixal predicates which take verbal complements. I propose that these predicates fall into two general classes: affixal main predicates and affixal auxiliary predicates, both of which permit verb incorporation.  The PF incorporation pattern of these  predicates may result in incorporation in contexts in which the complement is a reduced verbal projection (vP), rather than an 'edged', fully inflected clause (CP).  Given that these PF  incorporation contexts are monoclausal, a range of 'restructuring' effects are predicted, including 'long' wh-movement and 'long' possessor raising. Chapter 6 contains concluding remarks. A brief summary of the dissertation is given, followed by discussion of theoretical and typological implications of the analysis.  On the  theoretical side, I present general implications for the characteristics of the PF branch. I also illustrate consequences which local spell-out has for the question of the grammatical locus of head movement operations.  A final theoretical implication is the prediction that 'inside-out'  dependencies, the antithesis of Nuu-chah-nulth affixation, should be possible only through post spell-out linearization, and should not be available through local spell-out.  The Northern  Wakashan language Kwakw'ala is presented as evidence for affixation which arises by post spell-out linearization, at a derivationally later stage than that at which Nuu-chah-nulth affixation occurs.  Additional typological factors are also considered.  I state that there is a range of  variation attested in noun incorporation dependencies cross-linguistically. Furthermore, the 'lexical suffixes' of the Pacific Northwest sprachbund do not, as has been previously assumed, reduce to a single type of 'bound nominal' pattern (Gerdts 1998). We now turn to the chapter on PF Incorporation.  22  CHAPTER 2  PF Incorporation So Ifindwords I never thought to speak In streets J never thought I should revisit ~ T . S . Eliot Four Quartets no. 4  2.0 Introduction In the previous chapter, the observation was made that a syntactic construct, formed derivationally by binary concatenation, is not inherently ordered. In the syntax, the elements joined by Merge are an unordered set {oc, p}. (1)  y a  P  Before such a derivational output can be pronounced, however, the elements occupying the terminal positions a and P must be linearized. I hypothesize that it is strictly non-syntactic principles in the grammar which are responsible for ordering them. In Chapter 1,1 proposed that affixation requirements constitute one means by which the linguistic elements a and P may be ordered. If one or the other of a and P is an affix, a restricted set of linearization patterns arises: either oc-P or p-a.  These orderings emerge in the syntax-to-PF mapping, via local spell-out.  Repeated from Chapter 1, this claim may be defined as the following: (2)  Local spell-out: for Merge (a, P), spell-out (a, P)  The need for an affix to be linearized with respect to its host is a bare output condition on PF the portion of the grammar sensitive to temporal sequencing. The Wakashan language Nuu-chah-nulth was introduced as the empirical grounds on which the local spell-out hypothesis will be assessed. This chapter develops the notion that spellout induces in Nuu-chah-nulth a particular arrangement which I refer to as PF incorporation. In this linearization, an affixal predicate -a suffixes to a host p(-), yielding an ordered pair of p-a. The affixal predicate 'incorporates' its host in order to achieve a pronounceable form, that of a linearized affix. In (3), this pattern is exemplified by the string lc aaq-caaqa "busy with spawned w  herring eggs", which is comprised of the affixal predicate -caaqa "busy with" and its nominal host k" aq "spawned herring eggs". w  23  (3)  k aaqcaaqa?is ?aahuus?ath k aq-caaqa[+L]-?iis faahuus-'ath s,h.eggs-busv.with-3.iND place.name-from The Ahousahts are busy with spawned herring eggs. w  w  According to my analysis, the string £ aaq-caaqa emerges as a reflex of the need to linearize the w  affixal predicate -caaqa "busy with". The host chosen for the affix is its derivational sister, the nominal k' aq "spawned herring eggs". w  (4)  ^ -caaqa  \ k aq w  busy with  s.h.eggs  The resulting dependency is a case of PF Incorporation. The term 'incorporation' has a lengthy history in Amerindian linguistics. In the context of what has been referred to as 'noun incorporation', this label applies to instances in which a noun and verb are combined into a single word. Over the past century, a series of high-profile debates have occurred regarding this phenomenon.  Kroeber (1909, 1911) and Sapir (1911a)  capitalize on the free-bound contrast to make a distinction between noun incorporation languages in which the verb is a free stem, and those 'verbalizing suffix' languages in which the verb is bound. This notion resurfaces in an exchange between Mithun (1984, 1986) and Sadock (1980, 1986).  M y contribution to this discussion is to specify a four-way range of affixation  relationships which exist in the typology of noun-verb dependencies. If the noun and verb are each either free or bound, four patterns of dependencies are available: a bound noun may affix to a bound verb or to a free verb; a free noun, in turn, may affix to a bound verb or to a free verb. Thus, the affixation dependencies of 'noun incorporation' are not a uniform phenomenon (see Chapter 6 for further discussion).' (5) Affixation dependencies of noun incorporation bound noun N-V btiwtd'verb''*'  • free noun N-V  (Nuu-chah-nulth) _ , (Nuu-chah-nulth) free verb N - V  N-V  Nuu-chah-nulth instantiates two of these four options.  In a Nuu-chah-nulth complex  denoted by P-a, the bound verb - a takes a bound or free nominal P(-) as its host. Because the affixal predicate is obligatorily bound, it can never go without a host.  The example in (6a)  ' Baker (1988: 143) reaches a similar conclusion, although he proposes a three-way morphological distinction: root noun and root verb, root noun and affixal verb, and affixal noun and root verb. Baker does not discuss the possibility of an affixal noun attaching to an affixal verb.  24  indicates a grammatical instance of the affixal predicate -?aap "buy" suffixing to the bound nominal mahia- "house". In (6b), even though the free form of "house", mahtii, is used, it is not possible for the noun and verb to be separated. (6)  a.  maht'a?amit?is cakup mahta-'aap-mit-?iis cakup house-buy-PST-3 .IND man A man bought a house.  b.  * ?aamit?is mahtii cakup 'aap-mit-?iis mahtii cakup buy-PST-3.lND house man A man bought a house.  The example in (6b) is ruled out because the affixal predicate -?aap"buy" must be linearized as a suffix. A purely syntactic account of incorporation does not capture the significance of the bound or free status of the verb in affixation contexts. In languages in which the 'incorporating 2  verb' is obligatorily bound, two logical possibilities emerge for the host for the verb, according to the local spell-out hypothesis. As previously noted, if cx in (7) is a bound verb, then (3 may take on the role of host for cc at spell-out. (7)  Y a  p  There is a logically possible alternative, however: an expletive host may be inserted at the point of spell-out to act as 'dummy' placeholder for the bound verb. This predicted pattern is attested in Nuu-chah-nulth, which utilizes an expletive host, Pu-, in contexts in which the affixal predicate has not incorporated a host.  3  A n example is given below, in which ?u- acts as a host for the  affixal predicate -?aa/>"buy". (8)  a.  7u?aarnit?is mahtii cakup 7u-'aap-mit-?iis mahtii cakup 0-bu_y.-PST-3.IND house man A man bought a house.  b.  maht'a?amit?is cakup mahta-'aap-mit-Tiis cakup house-buy-PST-3 .IND man A man bought a house.  Baker (1988: 72) argues that it is a morphological component of the grammar which determines whether (syntactic) incorporation is obligatory in some cases, or forbidden in others. This is ascribed to a filtering effect of the morphology. In the framework of Distributed Morphology, ?u- in Nuu-chah-nulth qualifies as a 'dissociated' morpheme - that is, one that is inserted at the point of spell-out (Embick 1997, Noyer and Embick 2001). 2  3  25  In (8a), '/^-support' occurs as an alternative to the noun incorporation of (8b). This expletive is also employed in cases in which the affixal predicate takes a sentential complement. In (9a), for example, the expletive Pu- appears as a host for the affixal predicate -cuk "necessary".  In (9a),  the predicate takes the conditional complement 6uk iPatquu qaawic?i "that the potatoes be w  washed".  For the predicate -cuk "necessary", the presence of  Pu- alternates with the  incorporation strategy. In (9b), the affixal predicate -cuk "necessary" suffixes to the verbal host, cu-k i&"wash (PERF)", rather than to the expletive Pu. w  (9)  a.  ?ucuk?is ?u-cuk-?iis  cuk i?atquu cu-k i?e-'at-quu  qaawic?i qaawic-?ii  0-need-3.lND  wash-PERF-PAS-3.COND potato-DET  w  w  It is best to wash the potatoes, (lit: "it is necessary that the potatoes be washed") b.  cuk i9ecuk?is cu-k ifc-cuk-?iis wash-PERF-need-3 .IND The potatoes need washing. w  w  qaawic?i qaawic-?ii potatoes-DET  The expletive Pu- in Nuu-chah-nulth is regularly used in the citation forms of affixal predicates, as in Pu-Paap "buy" and Pu-cuk "need".  From this point onwards, I adopt the convention of  writing affixal predicates in their citation forms when I mention them. Greenlandic (Eskimo) provides another example of this expletive pattern.  In the  Greenlandic language, 'incorporating verbs' are suffixes, just as in Nuu-chah-nulth (Waldie 2004). There is a 'placeholder' morpheme, pi-, in Greenlandic which surfaces in contexts in which no incorporation occurs (Sadock 1980). The following examples are adapted from Sadock (1980: 306, ex. 18a and 307, ex. 24). (10) Greenlandic a.  Qimme-qar-poq dog-have-3SG.lND He has a dog.  b.  Qimmi-mik dog-lNST  pe-qar-poq 0-have-3SG.IND  He has a dog. In (10a), incorporation unites the nominal qimme "dog" with the suffixal verb -qar "have". In (10b), in contrast, no incorporation of the nominal occurs, and instead, the bound verb is attached to the empty form pi- (surfacing as pe-). Unlike languages with bound verbs, languages with free verbs make no use of an expletive host for a verb in contexts in which incorporation fails to apply. A n example of a nounincorporating language with free verbs is Mohawk (Iroquoian). The incorporation pattern of Mohawk is indicated in (11a), in which the nominal -nuhs- "house" is incorporated into the  26  inflected verb ye-nuhwe'-s "like 3FS/3N". Note that in Mohawk the verb can surface detached from the noun, as in (1 lb). (11) Mohawk (examples from Postal 1962, as cited in Baker 1988: 81-82, ex. 14a-b) a.  Yao-wir-a'a  ye-nuhs-nuhwe'-s  PRE-baby-SUF 3FS/3N-house-like-ASP  The baby house-likes. b.  Yao-wir-a'a  ye-nuhwe'-s  ne  ka-nuhs-a'  PRE-baby-SUF  3FS/3N-like-ASP  DET  PRE-house-SUF  The baby likes the house. When incorporation of the noun into the verb does not occur, there is no process similar to Pusupport for the Mohawk verb. The inflected verb ye-nuhwe'-s "like" in (1 lb) does not receive an expletive host.  This difference symbolizes a key contrast between the Nuu-chah-nulth and  Mohawk patterns of incorporation. The contrasting patterns fall out from an analysis which takes into account the underlying bound/free status of the verb in affixation contexts. To summarize, we have seen two ways in which the affixation requirement of an affixal predicate may be met in Nuu-chah-nulth. On one hand, local spell-out may attach an affixal predicate to an incorporated host, yielding PF Incorporation. On the other hand, the expletive element ?u- may be introduced at spell-out as a host. Under this view, /fcnnsertion receives an analysis similar to that which Lasnik (1981, 2000) proposes for do-support in English: the 'dummy' do is inserted to meet the requirements of a potentially stranded affix.  Over the  following chapters, we will return to the discussion of /^-support as a spell-out solution which applies in cases in which an 'edge' separates the affixal predicate from a potential incorporate host. In Nuu-chah-nulth, the use of an expletive host for an affixal predicate is obligatory when the predicate takes a D P or CP complement: incorporation of a host cannot occur across these 'edged' domains. For example, the use of the expletive is necessary when the object of an affixal  -hic?in  predicate is marked with a determiner. The example in (12a) shows a grammatical instance of incorporation when the nominal host the determiner  "dress" is bare. Incorporation is im  appears, either in (12b) as the host for the affixal predicate, or in (12c) when  suffixed to the nominal. The example in (12d) shows the mandatory /i/-support which occurs when the nominal is marked with the determiner. (12)  a.  ruc?insiikitsis -hic?in-siik-mit-siis dress-make-PST-1 SG.IND I made a dress.  27  b.  * -hic?in?isiikitsis -fuc?in-?ii-siik-mit-siis dress-DET-make-PST-1 SG.IND  I made the dress. c.  d.  * ?iisiikmitsis ?ii-siik-mit-siis DET-make-PST-1 SG.IND I made the dress.  Fuc?in +uc?in dress  ?usiikitsis ?u-siik-mit-sis  +uc?in?i Fuc?in-?ii  0-make-PST-l SG.IND  dress-DET  I made the dress. We return to the topic of 'edge' effects in PF Incorporation in §2.5. For either expletive insertion or incorporation of a host, the satisfaction of the affixation requirement in Nuu-chah-nulth constitutes a bare output operation on PF.  A n inserted or  incorporated host allows the bound predicate to meet its linearization requirement. Bound status and linearization are not relevant to the syntax proper; instead, these are conditions on phonological representation. We now turn to discussion of the trademark properties of PF Incorporation in Nuu-chahnulth which serve to corroborate the claim that this phenomenon is derived post-syntactically, at the point of spell-out. In §2.1, independent evidence is presented in support of the claim that an affixal predicate and its host, whether incorporated or expletive, share an intimate phonological relationship. This is followed in §2.2 with discussion of the observation that incorporation in Nuu-chah-nulth is sensitive to string adjacency, rather than to syntactic c-command. In turn, §2.3 argues that incorporation in Nuu-chah-nulth displays an insensitivity to syntactic category, in the sense that the hosts of incorporation come from a range of lexical and functional categories. Evidence is presented in §2.4 that incorporation in Nuu-chah-nulth shows an, absence of L F effects. In §2.5, I illustrate that incorporation in Nuu-chah-nulth, although otherwise insensitive to syntactic and semantic properties, cannot cross a DP or CP. I attribute this 'edge' effect to the status of DPs and CPs as saturated domains (cf. the notion of 'phase' in Chomsky 2001). The chapter concludes with §2.6, which sums up evidence against a strictly syntactic view of Nuuchah-nulth affixation.  2.1 Morpho-phonological dependency The analysis which I am proposing states than an affixal predicate is united with a host at spellout so that it may be linearized. According to the local spell-out hypothesis, an affixal predicate -a is spelled-out with its host p(-), inducing a P-a linearization.  28  (13)  Local spell-out: for Merge (cx, (3), spell-out (cx, (3)  This section presents independent evidence for a phonological dependency between a and p. This evidence comes from the morpho-phonological 'subcategorization' of affixal predicates. In Nuu-chah-nulth, a striking property of bound morphemes is their ability to prosodically condition their hosts (Sapir and Swadesh 1939, Rose 1981, Davidson 2002, K i m and Wojdak 2002, K i m 2003)." For example, the repetitive iterative suffix -(y)a [+R+L] (ITER) causes vowel lengthening and reduplication of the first syllable of a monosyllabic root, as well as vowel lengthening of both the base and the reduplicant (Sapir and Swadesh 1939, Wojdak 2002, K i m and Wojdak 2002). (14)  a.  cuuscuusa cus-a[+R+L]  b.  kii&kiika ki&-a[+R+L]  dig-lTER  break-lTER  'digging continuously'  'breaking continuously'  Affixal predicates share this ability to prosodically condition their hosts, while independent predicates never induce prosodic conditioning. Each affixal predicate is associated with a characteristic pattern, although many affixal predicates are 'neutral' in that they do not impose changes on their host. A given affixal predicate may induce reduplication, a long vowel, a short vowel, or some combination of the three. (I refer the reader to Kim and Wojdak 2002 and Kim 2003 for a thorough description of which combinations are available.) (15) Prosodic conditioning a.  [+R]  b.  [+L]  long vowel  c.  [+S]  short vowel  1  reduplication  Both the expletive morpheme Pu- and incorporated hosts are affected by the prosodic requirements of affixal predicates. The examples below illustrate the behaviour of the affixal predicate Puu-hwaP "use", which triggers vowel lengthening of the first syllable of its host.  In (16a), the vowel of the  expletive morpheme Pu- is lengthened to Puu-, while in (16b) the first vowel of yaxyak "broom" is lengthened to yaaxyak (16) Long initial vowel (+L) a.  ?uuhwa+?i ?u-hwaTf+L1-'ii  yaxyak yaxyak  0-use-2SG.IMP>3OBJ  broom  Use a broom!  29  b.  yaaxyakhwa+?i yaxyak-hwa+[+L]-'ii broom-use-2SG.lMP>30BJ  Use a broom! The following example shows how reduplication is triggered by the affixal predicate ?u?u-q "travel with (in a vessel)". In (17a), the expletive morpheme Pu- surfaces as PuPu-, while in (17b), the morpheme Puus "someone" appears as PuuPuus. In this pattern, the first consonant and vowel of the host are reduplicated. The vowel length of the reduplicant is determined by the underlying vowel length of the morpheme which serves as the base. (17) Reduplication with neutral vowel length (+R) a.  ?u?uq?is Louis huupuuk asuk?i ?u-c[[+R]-?iis Louis huupuuk as-uk-?i 0-travel.with-3.IND Louis car-POSS-3.PS Louis is travelling in Robin's car. w  w  b.  TuuPuusqPis ?uus-(j[+R]-?iis someone-travel.with-3.IND  Robin Robin Robin  Louis Louis Louis  Louis is travelling with someone (eg. in his canoe). Some affixal predicates impose restrictions on vowel length as well as inducing reduplication. For example, the affixal predicate PuPuu-sapi "depend on" requires reduplication, as well as a short vowel in the reduplicant and a long vowel in the initial syllable of the base.  In the  examples below, the expletive surfaces as PuPuu- (18a), while Puus "someone" surfaces as PuPuus (18b). (18) Reduplication with short initial vowel & long second vowel (+R+S+L) a.  ?u?uusapi?is ?u-sapi[+R+S+Ll-?iis 0-depend.on-3.IND Louis is depending on  Louis Tin ?ayanak taana waa+ak Vancouver. Louis ?in ?aya-naak taana waa+ak Vancouver. Louis COMP many-have money go.to Vancouver having a lot of money in going to Vancouver.  b.  TuTuussapiTis ?uus-sapi[+R+S+L1-?iis someone-depend.on-3.IND S/he is depending on someone.  In contrast, the predicate PuPu-sum "want" triggers reduplication with a short vowel in both the reduplicant and the initial syllable of the base. In (19a), the expletive appears as PuPu-, while in (19b) the morpheme taanaq- "money" surfaces as tatanaq-. (19) Reduplication with short initial vowel & short second vowel (+R+S+S) a.  ?u?usum?is Louis ?u-sum[+R+S+S]-?iis Louis 0-want-3.lND Louis Louis wants his money.  taanaak?i taana-?ak-?ii money-POSS-DET  30  b.  tatanaqsum?is taanaq-sum[+R+S+S]-?iis monev-want-3.IND Louis wants money.  Louis Louis Louis  It is also possible for an affixal predicate to require reduplication of the first syllable of the base, as well as a long vowel in the initial syllable of base. In this pattern, the reduplicated syllable retains the underlying vowel length of the base. This is demonstrated in (20) with the predicate PuPuu-yuk "cry for". In (20a), the host is reduplicated to PuPuu-, while in (20b), the host is reduplicated to siisii- "you (pi)". (20) Reduplication with neutral initial vowel & long second vowel (+R+L) a.  ?u?uuyuk?anitwa?is?a+ nana?iiqsak ?uh?at Kyle ?u-y_uk[+R+L]-'at-mit-wa?is-?aT na?iiqsu-[+R]-?ak ?uh?at Kyle 0-cry.for-PAS-PST- 3. QUOT-PL aunt/uncle-PL-POSS by Kyle Kyle was crying for his uncles/aunts, (lit: "his uncles were cried for by Kyle")  b.  siisiiyuk?anitwa?icuus sii-yuk[+R+L]-'at-mit-wa?icuus vou-cry.for PAS-PST- 2PL. QUOT  Kyle Kyle Kyle  Kyle was crying for you (pi), (lit: "you (pi) were cried for (by) Kyle") Only in the 'neutral' pattern is the host prosodically unaffected by the affixal predicate. As the examples in (21) show, the phonemically contrastive vowel lengths of the host are unaltered by the affixal predicate -uPaa-f "find", and no reduplication is triggered. Accordingly, in (21a), resurfaces without reduplication or a change in vowel length, and in (21b), the same applies for taanaq- "money". (21) Neutral pattern a.  ruyuTaa+sis ?u-u?aa+-siis 0-find-l SG.IND I found money.  b.  taanaqu?aa+sis taanaq-u?aa+-siis money-find-1 SG.IND I found money.  taana taana money  Many affixal predicates in the language display this neutral pattern, and impose no prosodic conditioning (see Sapir and Swadesh 1939). Kim (2003) provides an analysis of the reduplicative patterns triggered by affixal predicates within the framework of Optimality Theory. I present here a brief sketch of the form which an analysis of prosodic conditioning may take if it is articulated in accordance with Minimalist assumptions. I assume that the prosodic 'subcategorization' of affixal predicates is  31  specified at the level of the lexical entry. A predicate such as Puu-hwal- "use" has the following lexical entry: (22)  -hwa+ "use":  USE  /hwaf/ affix [+L] Together with lexical semantics, phonemic representation, and information on the affixal status of this morpheme, this lexical entry contains a specification for a [+L] feature.  This vowel  lengthening requirement forms a condition on convergence at the PF interface. When the affixal predicate Tuu-hwai- "use" is spelled-out to PF, the lengthening requirement is obligatorily realised on its host. If yaxyak "broom" is the derivational sister of the affixal predicate, then as a host, it must undergo vowel lengthening. (23) -hwa+  yaxyak  This induces a form of yaaxyak-hwafbroom-use" at PF. I present this prosodic conditioning as independent evidence for the idea that local spellout links an affixal predicate with its host at spell-out to PF. The prosodic conditioning of a host by an affixal predicate indicates that the two must be interpreted together at PF. In effect, prosodic conditioning leaves a detectable 'footprint' of local spell-out.  The local spell-out  hypothesis offers a maximally restrictive prediction regarding which morphemes may be treated by prosodic conditioning, and which may not. According to this hypothesis, the only element which may be prosodically influenced is the derivational sister of the element which is specified for a prosodic requirement. For instance, for (24), the local spell-out hypothesis determines that the morpheme yaxyak "broom" is eligible for prosodic conditioning by the affixal predicate ?uuhwa+"use", but the imperative marker -'ii (2SG.IMP>30BJ) is ineligible. (24)  yaaxyakhwa+?i yaxyak-hwa+[+L]-'ii broom-use-2SG.IMP>3QBJ Use a broom!  According to the local spell-out hypothesis, a derivational sister to the affixal predicate is present at the point of spell-out of the affix, but the imperative marker is not. In the diagram in (25), the imperative marker is shown to occupy a higher position in the tree than the affixal predicate ?uuhwa+ "use", and its complement yaxyak "broom".  (See Chapter 3 for motivation for this  syntactic representation.)  Through local spell-out, the lengthening requirement of the affixal  predicate  with  is interpreted  yaxyak "broom", and not the imperative  32  marker  -ii  (2SG.IMP>30BJ).  Thus, only yaxyak "broom" is compatible with prosodic conditioning by the  affixal predicate. (25)  ^  \  <j - 'n IMP  spell-out -hwa-h\ yaxyak use broom  In contrast, the single spell-out and multiple spell-out hypotheses open the door for the possibility that morphemes from the derivation other than derivational sisters may be prosodically influenced by the affixal predicate.  With these delayed spell-out models, an additional  stipulation would be required to rule out why the imperative marker is not equally eligible for prosodic conditioning. Note that both the nominal yaxyak "broom" and the imperative marker 4  - y are wordmates of the affixal predicate. In contrast, the local spell-out hypothesis makes the correct generalisation as a direct prediction of the model.  2.2 Linearization is local When an element reaches spell-out, it must be linearized with respect to its neighbour. This is the essence of the local spell-out proposal. In the discussion up until this point, the locality constraint on this linearization process has been trivial in that only two syntactic terminals, a and p, were represented as the input to the spell-out rule: (26)  y  a  p  The syntactic configuration in (26) can be considered to be the basic step of the syntactic derivation, equivalent to a single application of Merge (a, P). A n example of this simple arrangement is when an affixal predicate selects a bare noun complement, as in taanaq-u?aa-r "find money" (from 21b). (27) -u?aaifind  taanaqmoney  The linearization forced at spell-out for taanaq-u?aa+ is a case of PF Incorporation. (We will return to the topic of the nominal complements of affixal predicates in Chapter 4.) In this  A possible stipulation could be directionality. That is, prosodic conditioning should only affect a morpheme to the left of the affixal predicate. The local spell-out hypothesis does not need to resort to directionality in determining the site of prosodic conditioning. Chapter 3 discusses problematic aspects of a directionality-sensitive mechanism for affixation.  4  33  section, we take a first step towards defining the linearization of more complex syntactic constructs. As we will see in this section, the linearization of affixal predicates in Nuu-chah-nulth is strictly local.  For an explanation of this locality constraint, consider the following syntactic  construct: (28)  y  a 8  7t  This configuration is derived via two separate applications of binary concatenation. In the first, 8 and TC are selected from the numeration [cx, 8, TC] and are joined through Merge (8,7t). The output of Merge (8, 7t) is (3, the abstract node label designating the contents of the pairing. For the second concatenation, a is introduced from the numeration. This concatenation unites a with P, through Merge (a, P). The syntactic output of this sequence of operations is y, the root node label. According to the local spell-out hypothesis, after the first round of Merge, 8 and 7t are spelled-out. Let us assume that a consequence of spell-out is that the two elements, 8 and TC, are ordered with respect to each other. (The means by which this ordering takes place is addressed in Chapter 3, but for now we can adopt this assumption.) Take this ordering to be specified first-tolast as <8,7t>. When the next element, cx, enters the computation and attains spell-out, a must be linearized with respect to p, just as it was in the simpler case of (26). (29)  Local spell-out: for Merge (a, P), spell-out (a, P)  With the derivation in (28), however, p is not a simplex construct; in (28), P is equal to the linearized object <8,7t>. Assume that a is an affixal predicate (-a).  At spell-out to PF, an ordering of a with  respect to p requires that the affixal predicate (-a) must be linearized relative to the ordered object <S, 7t>. The claim that this section makes is that spell-out of this arrangement consistently yields in Nuu-chah-nulth a linearization of <(8-a), TC>: an affixal predicate in Nuu-chah-nulth only ever suffixes to the element at the initial edge of its derivational sister. For the linearized object <8, 7t>, the host for an affixal predicate is identified as 8. The alternative of <8, (7C-a)> never arises in Nuu-chah-nulth.  34  Furthermore, when even larger derivational samples <0, 8, n> are considered, we will also see that linearization never 'skips' a potential host. Take <9, 8, 7t> to be the linearized object specified after two initial applications of Merge.  These two concatenations (and  corresponding local spell-outs) are following by a third application of Merge, introducing the affixal predicate oc. (30)  Y  a  8 At the spell-out point of y, the affixal predicate -a must be linearized with respect to the ordered object <0, 8, 7t>. Affixation is based on linear adjacency in that the resulting linearization is <(0-oc), 8,7i> and not <0, (8-oc), 7C>. It is the single leftmost element which can serve as the host for the affixal predicate. I label this constraint in Nuu-chah-nulth the string adjacency effect: (31) string adjacency effect: An affixal predicate must be linearized as a suffix to an immediately adjacent element. Which property of the grammar is responsible for this adjacency requirement? I consider this effect to be a reflex of the spell-out of the affix. It arises from local spell-out, in which the affixal predicate is evaluated relative to its derivational sister. The simplest view of the interpretative capacity of spell-out is that it operates without recourse to counting. Formally, linearization is therefore insensitive to ordinal position within a complex string <0, 8, 7i>.  Instead, the  linearization mechanism attends to the boundaries of the string (cf. Klavans 1985). The initial element of the derivational sister is chosen as host for an affixal predicate in Nuu-chah-nulth due to its peripheral position in the string. This derives the string adjacency effect. The string adjacency effect follows from the need to linearize an affix with respect to the boundary of its derivational sister. In Nuu-chah-nulth, affixes select as host the element at the left periphery of their derivational sister.  There also exists a logically possible alternative:  orientation to the right boundary of a derivational sister. Although this option does not arise in Nuu-chah-nulth, it may be observed in languages with 'phrasal' affixes which are positioned relative to the final element in a syntactic constituent (Klavans 1985). The two options for orientation of an affix at the boundaries of its derivational sister are illustrated in (32). In (32), the affix a has as its derivational sister the linearized object <0, 8,7C>.  35  (32) Peripheral positions within the derivational sister of an affix y a  e  A  5  initial boundary  71  final boundary  Orientation to the initial boundary will yield a string adjacency effect in which the element 8 will be selected as host for a, as in Nuu-chah-nulth. Positioning relative to the final boundary will determine that n is chosen as host for a. Thus, the local spell-out model predicts two alternative orientations for choice of hosts. linguistically (Klavans 1985).  These alternative realisations appear to be attested cross-  This prediction distinguishes the local spell-out model from  Marantz's (1988, 1989) Morphological Merger, which allows only string adjacent elements to be chosen as host for an affix. In the following sub-sections, we will see empirical evidence for the string adjacency effect in Nuu-chah-nulth affixation.  I start in §2.2.1, by showing that affixation in Nuu-chah-  nulth does not skip potential hosts, but instead feeds a potentially iterative affixation process. In §2.2.2, I present evidence that PF incorporation targets non-heads of a syntactic constituent, so long as these elements are positioned at spell-out contiguous to the affixal predicate. In §2.2.3, it is shown that this operation breaks up coordinated objects - disregarding the Coordinate Structure Constraint - by targeting the conjoined element which abuts the affixal predicate at spell-out.  2.2.1 Iterativity This section discusses the make-up of complex strings of dependencies. Consider (33), in which a sequence of affixes (including two affixal predicates, -'iih "try to" and -mahsa "want to") are suffixed to the verb huhtak "know". (33)  huuhtaksiihmahsa?is Lucy huhtak-si^-'iih[+L]-mahsa-?iis Lucy know-PERF-try.to-want, to-3 .IND Lucy Lucy wants to learn how to speak Nuu-chah-nulth.  quuquu?aca quu?ac-[+R]-(y)a person-speak-CONT  I have described the process of affixation in Nuu-chah-nulth as one applying to pairs of items: an affix and a host. If the linearization specified by local spell-out is a pairwise function, then how  36  can complex sequences such as (33) be formed? It seems on the surface that there are many affixes, and only a single host (the verb huhtak"know"). If the host for an affixal predicate must be linearly adjacent to the affixal predicate, then why is that -mahsa "want to" is attaching to another bound element (-'iih "trying to"), and not attaching directly to the free form huhtak "know"? Strictly speaking, the affixal predicates -mahsa and - iih cannot be serving as hosts for each other. Each of these affixal predicates are suffixes, so if they must find a host between them, then one will necessarily be left without. Recall from Chapter 1 that the combination of two suffixes, -oc and -(3, results in an indeterminate ordering of (-oc-P) or (-p-oc). If the ordering of (-CX-P) is selected, the result fails in that the element - a is not bound as a suffix. Conversely, if the ordering of (-p-cc) is picked, then -p fails to appear as a suffix.  Thus, the relationship  between the elements is inherently incompatible, and as such, no dependency obtains. The solution to this problem is iterative application of local spell-out, an idea first introduced in Chapter 1. In the framework I am assuming, spell-out applies for each occurrence of Merge. By the Order Preservation hypothesis (Fox and Pesetsky to appear), spell-out of later cycles adds information to previous cycles. A consequence of this proposal is that successive applications of local spell-out enforce a build-up of hosts, induced when one affixal predicate finds a host, and then this affix-host complex in turn serves as the host for another affix. We can take the data in (34) as an illustration of this process. (34)  carnaspa+cuqsi?in camas-paT-cuq-sifc-'in sweets-taste-in.mouth-PERF-1 PL.IMP Let us put something sweet in our mouths.  In (34), there are two affixal predicates: Tu-pa-r "taste o f and ?u-cuq "in mouth". The affixal predicates are followed by the perfective marker —sifc (PERF), and the imperative marking - 'in (1 PL.IMP).  Recall that syntactic derivations are built from bottom-to-top. We assume the first step of the syntactic derivation to be one in which the predicate -pa+ "taste o f joins with camas "sweet" via Merge (pa+, camas). (In Chapter 4, I show in detail how arguments of affixal predicates such as Pu-pai- taste o f are syntactically introduced.) u  (35) -pa+ taste  camas sweet  Because -pa+ is a suffix, this arrangement must be linearized at spell-out as <camas-pai> "sweet-tasting". Successive steps of Merge result in a longer string of morphemes. In the next stage, -cuq "in mouth" is merged into the derivation, via Merge (cuq, camas-pai).  37  -cuq in mouth  -pa-t taste  camas sweet  Because -cuq is a suffix, this string is linearized as <camaspai--cuq> "something sweet tasting in the mouth". Suffixes introduced later in the derivation follow a previously positioned suffix. "Tucking in" of suffixes is not permitted, ruling out a form such as * <camas-cuq-pa-f>, in which a later suffix (-cuq"m mouth") would adjoin directly to the original host of the derivation (camas "sweet"), rather than falling outside the last suffix (-pa-f "taste o f ) . The impermissibility of "tucking in" follows directly from the role that local spell-out plays in fixing phonological content. If suffixes were to adjoin cyclically to the original host, rather than the last suffix, then this would disrupt the placements of earlier cycles. Under the local spell-out model, linearization is fixed at each cycle. Returning to the derivation of (34), the non-predicative suffixes, -sik (lPL.IMP),  -sik  (PERF)  and - 'in  are also eligible for positioning through local spell-out. When the perfective suffix  (PERF)  is merged into the derivation, it is spelled-out to PF with its derivational sister [  camaspaivuq, inducing a linearization of <camaspa+6uq-sik> "put something sweet tasting in the mouth". (37) -sik PERF  -cuq  in mouth -pa+ taste Finally, the imperative suffix -'in  camas sweet (lPL.IMP)  is then positioned at spell-out after it has been  merged. (38) - m IMP  -sik PERF  -cuq  in mouth -pa+ taste As a suffix, the imperative marker -'in  camas sweet (lPL.IMP)  is spelled-out following the previously  linearized components. The resulting arrangement is <camaspa+cuqsi-?irt> "let us put something sweet in our mouths". The principles of iterative local spell-out therefore resemble the effects of the Mirror Principle of Baker (1988): the left-to-right arrangement of suffixes reflects the first-tolast steps of the syntactic derivation. Suffixes introduced later in the derivation will be linearized  38  towards the end of the word. In effect, each step of the syntax induces a 'phonologization' of the elements of the syntactic tree. According to my proposal, the 'phonologizing' effects of local spell-out are reflected in a sensitivity to string adjacency at PF.  Sensitivity to linear ordering is a property of the  phonological system, not the syntax (Chomsky 1995, Fox and Pesetsky to appear). As I have described, the syntax does not operate on the basis of linear arrangements: it is simply a device of binary concatenation. The next sections add weight to the argument that PF Incorporation is a non-syntactic phenomenon. The data which I will present demonstrates that PF Incorporation operates in Nuu-chah-nulth on linearly adjacent items, irrespective of their internal syntactic structure.  2.2.2 Modifier incorporation The analysis predicts that any element which is string adjacent to an affixal predicate should be chosen to act as its host; elements which are not linearly adjacent to the affix should not be eligible as hosts. Here, we examine the phenomenon of modifier incorporation as evidence for sensitivity to string adjacency in Nuu-chah-nulth incorporation. The prediction is that a modifier will serve as host for an affixal predicate if it occurs at the left-periphery of its derivational sister. Two types of modifier incorporation will be considered in detail: the first, targetting adjectives; the second, targetting adverbials.  2,2.2.1 Adjective incorporation Within nominal phrases in Nuu-chah-nulth, there is a strict ordering relationship between constituent elements, such that a modifier necessarily precedes the nominal . (In Chapter 3,1 will attribute this regularity to a spell-out 'convention'.) A modifier such as haPum "tasty" must obligatorily precede a nominal such as Paapinis "apples". (39)  a.  b.  ?u?iic?is?a+ 7u-'iic-?iis-?aF  ha?um ?aapinis ha?um Paapinis 0-consume-3 . I N D - P L tasty apples They are eating delicious apples.  * ?u?iic?is7a+ ?aapinis ?u-'iic-?iis-?aF ?aapinis 0-consume-3.IND-PL apples They are eating delicious apples.  ha?um ha?um tasty  Let us assume for the present discussion that in (39), the affixal predicate ?u-?iic "consume" takes the nominal phrase haPum Paapinis "tasty apples" as its complement. (Chapter 4 provides an analysis of the syntactic configuration of arguments of affixal predicates.) In this section, we will  39  see that, as predicted, the choice of host for an affixal predicate is determined by the linear ordering of elements with the nominal phrase that it takes as a complement. A n affixal predicate incorporates whatever host is "leftmost in the order Quantifier] > Q[uantity] > Afdjective] > N[oun]" (Rose 1981: 294).  In other words, an affixal predicate obligatorily attaches to the  element in the complement which is string adjacent to the affixal predicate. It is this sensitivity to linear ordering which determines that PF Incorporation is not 'noun incorporation' in a strict sense. Although affixal predicates may select a noun as host in a simplex nominal complement, this preference switches once a pre-nominal constituent enters the picture. For example, although the nominal Paapih/s "apples" is the host for ?u~?iic "consume" in (40a), it cannot act as a host when the prenominal modifier ha?um "tasty" appears, as in (40b). (40)  a.  b.  ?aapiniyic?is?a+ ?aapinis-Jic-?iis-?a+ apples-consume-3 .IND-PL They are eating apples. *  ?aapiniyic?is?a+ ha?um ?aapinis-'iic-?iis-?a4harum apples-consume-3 .IND-PL tasty They are eating delicious apples.  In this context, the adjective is necessarily chosen to be the host, rather than the modified noun. In (41), the affixal predicate ?u-?iic "consume" suffixes to the modifier ha?um "tasty". (41)  ha7um?ic?is?a"t?aapinis ha?um-'iic-?iis-?a-f ?aapinis tasty-consume-3. IND-PLapples They are eating delicious apples. Quantifiers, which like adjectives necessarily precede a nominal, show a parallel pattern  of being selected as host for the affixal predicate. The example in (42a) shows this quantifierinitial order. A n affixal predicate incorporates a quantifier (42b), rather than the quantified noun (42c). (42)  a.  ?u?is?is ?aya rhuks?i ?u-^is-?iis ?aya rftuks?i 0-on.beach-3 .IND many rocks There's lots of rocks on the beach.  b.  ?ayiis?is muks?i ?aya-%-?iis muks?i many-on.beach-3.IND rocks There's lots of rocks on the beach.  c.  * muks?i?is?is ?aya rhuks?i-^is-?iis ?aya rock-on.beach-3.IND many There's lots of rocks on the beach.  40  This sensitivity to linear adjacency extends to object w/i-questions formed by incorporation into an affixal predicate. In "which"-questions, the w/Vword hosts the predicate, while the restriction is stranded (Davis and Sawai 2001): (43)  waayaTamith Louis waayaq-'aap-mit-h Louis which-buv-PST-3.0 Louis Which sweater did Louis buy?  cupcupsunvr cupcupsunvr sweater  The restriction which Nuu-chah-nulth incorporation has on targetting the "leftmost" element (Rose 1981: 295) is not in the vocabulary of the syntax. This is because incorporation in Nuu-chah-nulth is not a syntactic process. In the Minimalist model I have adopted, syntactic processes operate on the basis of hierarchical relationships - created by binary concatenation while PF processes operate on the basis of linearly-defined relationships. In the terminology of Lasnik (2000), the two types of analyses may be teased apart in contexts in which linear adjacency does not correspond to hierarchical adjacency. Let us consider how these types of adjacency differ. Syntactic head movement is possible only in accordance with the Head Movement Constraint, a restriction which operates on the basis of hierarchical dominance relations within a syntactic tree: (44)  Head Movement Constraint (Travis 1984): A head Y may only move to X if Y is the sister of X .  Matushansky (to appear) terms this relation a 'head-of-the-complement'  locality.  Following  Matushansky, I will assume that it is fundamentally a syntactic restriction, as the same notion of locality is at play in c-selection, the means by which heads select the syntactic category of their complement.  5  Indeed, as Matushansky argues, head movement is possible only between  elements when one is c-selected by the other. For the tree in (45), Y is the sister of X , but Z is not. In other words, Y is the head of the complement of X . (45)  ^ X  \ Y(P) Z(P)  Y  Matushansky's (to appear) proposal counters the speculation of Chomksy (2001) that head movement is non-syntactic. See also Donati (2003) and Suranyi (2003), among others for related arguments that head movement is a syntactic phenomenon. I return to this topic in Chapter 6.  5  41  Thus, according to the Head Movement Constraint, only Y should be eligible to move to X .  6  Movement of Z to X should be impossible. The 'head-of-the-complement' locality of head movement is a hierarchically-defined restriction. This locality is sensitive to which element of a syntactic pairing is the one to project. In (45), Y projects after Merge (Z, Y). Therefore, Y is the head of the complement. Y may be said to be 'hierarchically' adjacent to X , because it is Y(P) which is united with X in the next round of structure-building, Merge (Y, X ) . This notion of hierarchical adjacency is not equivalent to linear adjacency.  If the syntactic object of (45) is  realised as the linearized object <X, Z, Y>, then X shares a linear adjacency to Z despite the fact that Z is not hierarchically adjacent to X. As Baker (2003: 152) describes, it is a consequence of this hierarchical adjacency restriction that adjective incorporation is explicity forbidden syntactically: "the head movement constraint implies that one can never incorporate the attributive modifier of a noun to form a A + k  V [NP t N] structure..." Thus, even if an adjective in complement position of a verb is linearly k  adjacent to the verb ('leftmost' of the complement), it is not 'hierarchically adjacent' to the verb in the syntax. In the tree in (46), the N is the head-of-the-complement of V ; A is not the head-ofthe-complement. (46)  ^  \  V  N(P) A  N  By the Head Movement Constraint, it is therefore predicted to be impossible for A to move to V . In contrast, operations which are processed on the basis of linear adjacency are not restricted by head-of-the-complement  locality.  Let us consider the case of adjectival  incorporation in Nuu-chah-nulth as evidence that incorporation operates on the basis of linear adjacency within derivational sisters, and not on the basis of hierarchical adjacency. Recall from the preceding discussion that an affixal predicate incorporates an adjective contained in its complement. I repeat here the example of adjective incorporation shown earlier in (41). (47)  ha?um?ic?is?a-l' ?aapinis ha?um-Mic-?iis-?a+ ?aapinis tasty-consume-3.IND-PL apples They are eating delicious apples.  I wait until Chapter 6 to discuss a particular technical implementation of syntactic head movement. The traditional model of head adjunction (Travis 1984, Baker 1988) is at odds with the Minimalist requirement that movement (remerge) only be effected at the root node (the Extension Condition). Suranyi (2003) and Matushansky (to appear) each offer analyses of syntactic head movement compatible with the Extension Condition. Either of these approaches would be compatible with my assumptions here. See Chapter 6 for further discussion of syntactic head movement. 6  42  It is impossible for an affixal predicate to incorporate the noun if an adjective is present. Crucially for our discussion, it was also noted that an adjective necessarily precedes a modified noun in Nuu-chah-nulth. Thus, where 8 is a modifer and 7t a noun, the affixal predicate - a is attached to the leftmost element in its complement <8,7i>, yielding a linearization of <(8-oc), %>. Adjective incorporation is an ideal test case for determining whether linearization of affixal predicates is a syntactic or PF phenomenon because for an adjective contained with the complement of a verb, linear adjacency between the adjective and verb does not correspond to hierarchical adjacency.  That is, while an adjective may directly follow a verb, it is the noun  which heads the complement that is hierarchically adjacent to the verb, not the adjective contained within the complement. The diagram in (48) shows the proposed syntactic structure for a modified noun phrase HaPum Paap/nis "tasty apples" which occurs as the complement of the affixal predicate ?u-?iic "consume". (48)  VP V - 'He  NP A haPum tasty  N Paap/nis apples  If incorporation in Nuu-chah-nulth were a case of syntactic head movement, then the movement should be in accordance with the Head Movement Constraint. This constraint determines that for (48) , the noun should incorporate, but the adjective should not. This is not what occurs in Nuuchah-nulth incorporation. In Nuu-chah-nulth, it is ungrammatical i f the noun incorporates into the affixal predicate when an adjective is present. This is shown in (49a), in which the affixal predicate - 'He "consume" incorrectly suffixes to the noun Paapinis "apples".  As indicated in  (49b), the affixal predicate must instead suffix to the adjective haPum "tasty". (49)  a.  b.  * ?aapiniyic?is?a+ ha?um ?aapinis-^iic-?iis-?aT harum appl es-consume-3. I N D - P L tasty They are eating delicious apples. haTumPicTisPa-l?aapinis ha7um-4ic-?iis-?a+ ?aapinis tasty-consume-3.iND-PLapples They are eating delicious apples.  This behaviour follows from the generalisation that an affixal predicate suffixes to whatever element is left-most in the noun phrase. The fact that incorporation in Nuu-chah-nulth targets whatever element is contiguous to the affixal predicate, irrespective of hierarchical adjacency, is  43  predicted by a local spell-out analysis in which incorporation operates on the basis of string adjacency. However, such behaviour is not predicted by a syntactic head movement analysis of incorporation. In  order  to  maintain  a  syntactic  head-movement  analysis  of Nuu-chah-nulth  incorporation, it would become necessary to abandon a noun-headed analysis of the complement phrase, in favour of a representation in which modifiers head the phrase containing the nominal (Stonham 2004). Such an analysis is represented in (50). (50)  VP V -'He consume  AP A HaPum tasty  N Paapinis apples  However, this putative case of syntactic incorporation is at odds with independent evidence for the nominal headedness of adjective-noun combinations. This independent evidence comes from categorial restrictions on modification in Nuu-chah-nulth. Modification in Nuu-chah-nulth is sensitive to a constraint on syntactic category (Wojdak 2000, 2001).  While adjectives may modify a noun (such as haak aak "girl"), an w  adjective (such as PaapM"friendly") is barred from serving as the modifier of another adjective. (51)  a.  ?u?uk inkitsis [q aca+aq?i ?u-k ink-[+R]-mit-siis [q aca+-aq[+S]-?ii 0-with-speak-PST-l SG.IND beautiful-AUG-DET I talked with the beautiful girl. w  w  w  b.  w  * ?u?uk inkitsis ?u-k ink-[+R]-mit-siis w  w  0-with-speak-PST-l SG.IND  haak aa?c] haak aak] girl w  w  [q aca+aq?i [q aca-r-aq[+S]-?ii  ?aaphii] ?aaphii  beautiful-AUG-DET  friendly  w  w  I talked with the beautiful friendly (one). The distinction between nominals and adjectives in such modification constructions is clearly not reducible to a non-categorial difference such as a contrast between stage-level and individuallevel predication (Kratzer 1995) or between transitives and intransitives. Lexical items like haak aafc "girl" and q acai- "beautiful" are each one-place predicates which denote individualw  w  level (non-transient) properties. Thus, I follow the argument which Demirdache and Matthewson (1995) make for Salish in proposing that the factor which distinguishes the class of these lexical items is inherent lexical category. The argument  for the headedness of adjective-noun  consideration of iteratively modified forms such as (52).  44  combinations comes from  (52)  ?u?uk inkitsis ?u-k ink-[+R]-mit-siis 0-with-speak-PST-l SG.IND w  w  [q aca+aq?i [q acar-aq[+S]-?ii w  w  beautiful-AUG-DET  ?aaphii haak aafc] ?aaphii haak aa9e] friendly girl w  w  I talked with the beautiful friendly girl. Given the category-sensitive restriction in (53a-b), we can deduce for an iteratively modified form that adjectival headedness of the phrase is ruled out (54b), in favour of nominal headedness (54a). In other words, since we know that adjective-adjective modification is impossible, it must be the case that the internal consituent denoted by bracketing in (54) must be behaving syntactically as a noun phrase, and not an adjective phrase.  If the internal constituent in (54)  were behaving as an adjective phrase, then it would be incorrectly predicted that combination with another adjective, as in (54b), should be impossible. (53)  a. b.  *  AP + NP  =>  AP+ AP  =>  (54)  a.  A P + [ A P + NP]  b.  * A P + [ A P + NP]  N P  A P  Thus, in an adjectivally-modified noun, it must be the noun that is the head of the phrase. This, in turn, implies that incorporation in Nuu-chah-nulth applies to morphemes (eg. adjectives) which are not the syntactic heads of the complement of an affixal predicate. This constitutes strong evidence against a syntactic account of Nuu-chah-nulth incorporation in which the head movement is subject to the Head Movement Constraint.  Adjectives are not  hierarchically adjacent to the affixal predicate; however, they do maintain a linear adjacency to the affixal predicate.  A linearization mechanism sensitive to linear adjacency, but not  hierarchical adjacency, predicts this incorporation pattern in Nuu-chah-nulth.  2.2.2.2 Adverbial incorporation This sensitivity to linear ordering may also be observed with affixal predicates which take verbal, rather than nominal, complements. Affixal predicates in this class include -qaath "claim" and Tii-fifa "come upon".  These predicates allow incorporation of a verb from their logical  complement. (55)  a.  wa+sifcqafhitsis wa+-sifc-qaath-mit-siis go.home-PERF-claim-PST-1 SG.IND  I claimed I went home. b.  wa?icvi9citsis Ken wa?ic-fifc-mit-siis Ken sleep-come.upon-PST-1 SG.IND Ken I came upon Ken sleeping.  This section considers the pattern of adverbials which respect to these 'verb-incorporating' affixal predicates.  45  Outside of incorporation contexts, adverbials in Nuu-chah-nulth split into two classes, depending on whether they are subject to flexible or rigid positioning relative to the main predicate. 'High' adverbials (Cinque 1999), such as subject-oriented or temporal adverbs, have a flexible order, and may either precede or follow a main predicate. This class includes subjectoriented q a?uuh "purposely" and temporals caani "first" and na?iik "immediately". The two w  patterns for these flexibly-positioned adverbs are shown in the examples below. In each of the (a) cases, the adverb appears before the main verb. In the (b) examples, the adverb follows the main verb. (Inflectional morphemes, as 'second position' enclitics, suffix to whatever word is first in the clause.) (56)  a.  q a?uuh?is Florence q a?uuh-?iis Florence purposely-3.lND Florence Florence is purposely singing.  nunuuk nunuuk sing  w  w  b.  nunuuk?is Florence q a?uuh nunuuk-?iis Florence q a?uuh sing-3.lND Florence purposely Florence is purposely singing. (context: Florence's neighbour kept her awake last night and now she wants to get even by being loud) w  w  (57)  (58)  a.  caani?aq?e?is wa?ic caani-?aq9e-?iis wa?ic first-FUT-3.IND sleep He will sleep first (i.e. before doing something else).  b.  wa?ic?aq^?is caani wa?ic-?aq&-?iis caani sleep-FUT-3.IND first He will sleep first (i.e. before doing something else).  a.  na?iik?aq&sis na?iik-?aq&-siis  wa+sik wa+-si&  immediately-FUT-1 SG.IND  go.home-PERF  hawii?a?equu hawii-'ak-quu finish-TEMP-3.COND  I will immediately go home when it's finished. b.  wa+si?aq^sis wa+-si?t-?aq9e-siis  na?iik hawii?afcquu na?iik hawii-'ak-quu go.home-PERF-FUT-1 SG.IND immediately finish-TEMP-3.COND I will immediately go home when it's finished.  Manner adverbials, in contrast, belong to a second class which must rigidly precede the main predicate.  These manner adverbials include wityax "slowly", camaqh "properly" and hacuk  "(sleep) deeply". As shown in the following (a) examples, it is grammatical for the manner adverbial to precede the predicate it modifies. In the (b) examples, in contrast, ungrammaticality arises when the manner adverbial follows the main predicate.  46  (59)  a.  b.  (60)  *  a.  b.  (61)  wityaxits waa+sifc wityax-mit-s wa+-[+L]-si?e Slow-PST-1 SCABS go. home-CONT-PERF I was going home slowly. waa+sifcits wa+-[+L]-si?e-mit-s gO.home-CONT-PERF-PST-lSG.ABS I was going home slowly.  wityax wityax slow  camaq9e?is titiqs Florence camaq9e-?iis titiqs Florence properly-3.lND dry Florence Florence is drying dishes properly. *  a.  b.  titiqs?is camaqfc Florence titiqs-?iis camaqfc Florence dry-3.lND properly Florence Florence is drying dishes properly. hacuk it?is wa?ic hacuk-mit-?iis wa?ic deeply-PST-3.IND sleep Ken was sleeping deeply. w  Ken Ken Ken  * wa?icif?is hacuk Ken wa?ic-mit-?iis hacuk Ken sleep-PST-3.lND deeply Ken Ken was sleeping deeply.  What is the suffixation pattern of affixal predicates which take adverbially-modified complements?  Rose (1981: 296) makes the following general statement about sentential  complements: "[p]arallel to N P incorporation, it is the left-most and highest constituent of the clause governed by the suffix which serves as base to the suffix". Rose's generalisation makes the correct predictions about the incorporation pattern of adverbials.  In the case of flexibly  positioned adverbials, an affixal predicate has the option of attaching to either the adverbial or the verb. In the following (a) examples, the affixal predicate suffixes to the adverbial; in the (b) examples, suffixation is to the verb. (62)  a.  q a?uuhqath?is q a?uuh-qaath-?iis purposely-claim-3.IND Ken is pretending to be  ta?i+ ta?i+ sick sick on  Ken Ken Ken purpose.  ta?i-kjath?is ta?i-j--qaath-?iis sick-claim-3.lND Ken is pretending to be  Ken Ken Ken sick on  q a?uuh q a?uuh purposely purpose.  w  w  b.  w  w  47  (63)  (64)  a.  caaniwitassis wa?ic caani-wit'as-siis wa?ic first-gonna-1 SG.IND sleep I'm gonna sleep first (ie. before doing something else)  b.  wa?icwit'assis caani wa?ic-witas-siis caani sleep-gpnna-1 SG.IND first I'm gonna sleep first (ie. before doing something else)  a.  naTiikqathitsis na?iik-qaath-mit-siis  wa+si?e wa+-si&  ?atquu ?atquu  wiktumsa wik-tum-sa  immediatelv-claim-PST-1 SG.IND go.home-PERF although NEG-PST-1 SG.DEP  I claimed I went home immediately, but I didn't. b.  waFsi&qathitsis na?iik ?atquu wiktumsa waF-sifc-qaath-mit-siis na?iik ?atquu wik-tum-sa go.home-PERF-claim-PST-1 SG.IND immediately although N E G - P S T - 1 SG.DEP I claimed I went home immediately, but I didn't.  For adverbials which rigidly precede a predicate, however, the choice of host for the affixal predicate is inflexible: the affix must attach to the adverbial, rather than to the verb.  As the  following (a) examples illustrate, it is grammatical for the affixal predicate to attach to the manner adverbial.  However, the (b) examples show that it is ungrammatical for the affixal  predicate to suffix to the verb, rather than the manner adverbial. (65)  a.  wityaxmahsasis wityax-mahsa-siis  waa-Fsifc wa+-[+L]-si&  slow-want.to-1 SG.IND go.home-CONT-PERF I want to go home slowly. b.  (66)  a.  b.  (67)  a.  *  waa-rsifanahsasis wityax wa+-si9e[+L]-mahsa-siis wityax go.home-CONT-PERF-want.to-1 SG.IND slow I want to go home slowly. camaq&qath?is titiqs Florence camaqfc-qaath-?iis titiqs Florence properl y-claim-3 .IND dry Florence Florence is pretending to dry dishes properly.  * titiqsqath?is camaqfc Florence titiqs-qaath-?iis camaqk Florence dry-claim-3 .IND properly Florence Florence is pretending to dry dishes properly. hacukfifotsis hacuk-fifc-mit-siis deepl v-come.upon-PST-1 SG.IND I came upon Ken in a deep sleep.  48  wa?ic wa?ic sleep  Ken Ken Ken  b.  *  wa?ic ri9eitsis hacuk Ken wa?ic-fifc-mit-siis hacuk Ken sleep-come.upon-PST-1 SG.IND hacuk Ken I came upon Ken in a deep sleep. (  This difference in the incorporation pattern of the two sets of adverbials follows from an analysis in which PF incorporation is sensitive to the linear ordering of elements. If an adverbial permits a postverbal syntactic positioning, then it is possible for the verb to serve as a host for the affixal predicate at local spell-out. However, if an adverbial is only ever linearized preceding the verb, then the verb will not be in the 'leftmost' position which allows it to be string adjacent to the affixal predicate at spell-out. The distinct linearization schemes of preverbal and postverbal adverbials are indicated in (68), in which the verbal phrase is the derivational sister to an affixal predicate - a . require (68a). (68)  Variably positioned adverbs allow either ordering, while manner adverbials  7  a. preverbal adverbials -a  b. postverbal adverbials  V ADV  -a V  V V  ADV  At spell-out, the affixal predicate - a requires a host. In the case of preverbal adverbials as in (68a), it is the adverbial which is contiguous to the affixal predicate; as such, the adverbial acts as the host for the affixal predicate. For postverbal adverbials such as (68b), however, it is the verb which is 'leftmost' in the phrase; consequently, the verb is selected as the host for the affixal predicate. Thus, adverbial incorporation in Nuu-chah-nulth shows a sensitivity to independently attested restrictions on linear ordering.  2.2.3 Coordinated objects According to the local spell-out hypothesis, an affix is linearized with respect to an immediately neighbouring element. The host for an affix is necessarily selected from the derivational sister of the affix. Since this linearization process operates on the basis of string adjacency, the analysis predicts that the internal syntactic structure of the derivational sister to the affix should be irrelevant to the linearization process. This section concerns the specific prediction of this model for coordinated structures.  The string adjacency effect determines that an affix should be  linearized relative to a linearly adjacent element in a coordinated object.  In the following  This raises the question of what determines this ordering. See Chapter 3 for discussion.  49  diagram, a conjunction  (CONJ)  joins two conjuncts, (3 and %. This coordinated phrase is the  derivational sister of the affix -a.  X  CONJ  The prediction of the analysis is that in Nuu-chah-nulth (3 should act as the host of the affix -a. This section illustrates that this prediction holds. The behaviour of coordinated objects provides evidence that PF incorporation operates on the basis of linear adjacency in Nuu-chah-nulth.  The conjunction PuAP/is  (CONJ)  is used  exclusively to conjoin nominals in Nuu-chah-nulth. Its use is shown in (70), in which it appears between the two conjuncts, haak aa9e "girl" and ma?ikqac "boy" in (70a) and Bill and Mary in w  (70b). (70)  a.  b.  naacpiihamitsis naac-piiha-mit-siis  haak aa?e haak aafc see-glimpse-PST-1 SG.IND girl I caught a glimpse of a girl and a boy. huu+huu+amit?is huu+-a[+R]-mit-?iis  Bill Bill dance-ITER-PST-3.IND Bill B i l l and Mary were dancing.  w  w  ?uh?iis ma?ifcqac Mrtiis ma?ifcqac CONJ boy  ?uh?iis Mary ?uh?iis Mary CONJ Mary  First, we must note that the Coordinate Structure Constraint (CSC) is operative in syntactic movement in Nuu-chah-nulth, as with the w/i-questions shown below. The examples in (71) are object w/z-questions, while the examples in (72) are subject w/i-questions. In the (a) examples, we have a grammatical case of w/z-movement which does not make use of conjunction. In the (b) and (c) examples, however, it is shown that it is ungrammatical for w/r-movement to target a single conjunct of the argument. These CSC-violating examples are ruled out in Nuu-chah-nulth, parallel to the English cases which are given as their literal translation. (71)  a.  ?aacaci+itk ?acaq-ci+[+L]-mit-k  naacpiiha naac-piiha who-AUX-PST-2SG.Q see-glimpse Who did you catch a glimpse of?  b.  * ?aacaci+itk ?acaq-ci+[+L]-mit-k  naacpiiha ?uh?iis ma?ifcqac naac-piiha ?uh?iis ma?ifcqac who-AUX-PST-2SG.Q see-glimpse CONJ boy (lit: "Who did you catch a glimpse of and a boy?")  50  c.  * ?aacaci+itk naacpiiha haak aa?e ?acaq-ci+[+L]-mit-k naac-piiha haak aa?c who-AUX-PST-2SG.Q see-glimpse girl (lit: "Who did you catch a glimpse of a girl and?") w  w  (72)  a.  ?acaqith ?acaq-mit-h  huuxhuu+a huur-a[+R]  who-PST-3.Q  dance-ITER  ?uh?iis TYjhriis CONJ  Who was dancing? b.  *  ?acaqith ?acaq-mit-h  huu+huu+a huu+-a[+R]  ?uh?iis Mary ?uh?iis Mary who-PST-3.Q dance-ITER CONJ Mary (lit: "Who and Mary was dancing?")  c.  *  ?acaqith huu-rhuuta Bill ?acaq-mit-h huuT-a[+R] Bill who-PST-3.Q dance-ITER Bill (lit: " B i l l and who was dancing?"  ?uh?iis ?uh?iis CONJ  In the case of PF incorporation, however, a different pattern emerges. In the Ahousaht dialect of Nuu-chah-nulth, the first word of a coordinated object is chosen to host an affixal predicate, in striking contrast to the pattern of CSC-obeying syntactic movement.  Examples of this  characteristic of PF Incorporation are shown below. For example, in (73b), the affixal predicate ?u-k isiap w  "take away" incorporates the nominal homuut "bones", leaving stranded the  remainder of the conjunction PuhP/is kuuna "and gold".  Additional cases of incorporation  targetting the first word of the conjunct are shown in (74b) and (75b). (73)  ?uk istamit?is mama+ni 7u-k ist-'ap-mit-?iis mama-hii 0-move.away-TR-PST-3 .IND white.people White people took away the bones and gold. w  w  hamuut ?uh?iis hamuut ?uh?iis bones  hamuutk ist'amit?is mama+ni hamuut-k ist-'ap-mit-?iis mama-i-ni bones-move.awav-TR-PST-3 .IND white.people White people took away the bones and gold. w  w  (74)  ?u?aamitsis ru-'aap-mit-siis 0-buv-PST-l SG.IND I bought a canoe and a  CONJ  ?uh?iis ?uh?iis CONJ  capac ?uh?iis cima capac ?uh?iis cima canoe C O N J net net.  c'apac?amitsis capac- 'aap-mit-siis  ?uh?iis cima ?uh?iis cima  canoe-buy-PST-1 SG.IND  CONJ  net  I bought a canoe and a net. (75)  ruhaayasci Tu-haa-'as-cii 0-buy-go-2SG.GO Go buy flour and sugar!  9ei?cick'uk ?ei?eickuk flour  51  ?uh?iis suuk aa ?uh?iis suuk aa w  w  CONJ  sugar  kuuna kuuna gold kuuna kuuna gold  b.  &9eickukhaayasci ^i^ickuk-haa-'as-cii  ?uh?iis suuk aa ?uh?iis suuk aa  flour-buy.-go-2SG.GO  CONJ  w  w  sugar  Go buy flour and sugar! Under a purely syntactic analysis of incorporation, examples such as (73b-75b) should be banned by the Coordinate Structure Constraint. However, since sensitivity to linear order is a property of phonology (Chomsky 1995), this behaviour is directly predicted under an account which derives these linearizations from spell-out to PF.  8  In this section, we saw that the linearization of affixal predicates in Nuu-chah-nulth shows an insensitivity to syntactic constituency. The next section discusses the observation that this process is similarly blind to syntactic category.  2.3 Insensitivity to syntactic category The local spell-out hypothesis states that an affix is linearized with whatever element is adjacent to the affix at the point of spell-out. This linearization is predicted to be insensitive to syntactic category, as the process is attuned to string adjacency.  As this section demonstrates, PF  Incorporation in Nuu-chah-nulth is unselective for syntactic category. Affixal predicates which select nominal complements may incorporate a noun, adjective, quantifier, w/i-pronoun or relative pronoun. This range of possible hosts is illustrated in (76) for the affixal predicate ?u-Paap"buy". (76)  a.  cupcupsum+?amitsis c upc upsumF- 'aap-mit-siis sweater-buy-PST-1 SG.IND  b.  I bought a sweater.  (noun)  9eih?aamitsis c'upc'upsumF siya 9tth-'aap-mit-siis c'upc'upsumF siya red-buy- PST-1 SG.IND sweater lSG / bought a red sweater.  (adjective)  Additional evidence comes from the Tseshaht dialect of Nuu-chah-nulth. In Tseshaht, the first component of a complex numeral is selected as the host for an affixal predicate, while the other elements are stranded (Stonham 1998, 2004). The following example is adapted from Stonham (1998: 391, ex. 21b) (i)  hayuu'lisuksi ?is hayu-iiq-'iis-uk-si ?is ten-score-consume-POSS-1 SG.IND C O N J I spent three hundred (dollars) on him.  suc'iiq suca-iiq five-score (Tseshaht dialect)  This constitutes a difference between the Tseshaht and Ahousaht dialects, since in the Ahousaht dialect, parallel examples with coordinated numerals are ungrammatical. In Ahousaht, coordinated numbers act as impenetrable chunks for incorporation, perhaps indicating lexicalization.  52  hiyaapatuk?is nuutinum(minh) his-'aap-'at-uk-?iis nuutinum(-minh) all-buy.-PASS-P0SS-3.IND necklace(-PL) A l l his/her necklaces were bought.  (quantifier)  ?aqi?amith Taqi-'aap-mit-h what-buy-PST-3.0 What did Louis buy?  (wh-pronoun)  Louis Louis Louis  k in?a+it?is John FucPin yaTaamifPitk k in?a+-mit-?iis John +uc?in yaq-'aap-mit-?iitk like-PST-3.IND John dress REL-buy.-PST-2SG.RL John liked the dress you bought. w  w  (relative pronoun)  In a strict sense, PF incorporation is therefore not equivalent to 'noun incorporation'. Although an affixal predicate (which takes a nominal complement) is capable of incorporating a noun, elements with a range of other syntactic categories can serve the role of host. The same is true for the phenomenon described as 'verb incorporation' in Nuu-chahnulth. As previously discussed, adverbial modifiers show the ability to incorporate, along with verbs. (77)  a.  kamatqukmahsa?is kamatq-uk-mahsa-?iis run-DUR-want.to-3.IND Florence wants to run.  Florence Florence Florence  b.  ^aTixmahsaTis kamatquk ^aTix-mahsa-Tiis kamatq-uk fast-want.to-3.IND run-DUR Florence wants to run fast.  Florence Florence Florence  The negative particle wik can also incorporate into an affixal predicate. However, this pattern is restricted by constraints on string adjacency, and is thus indirectly conditioned by syntactic structure.  In Nuu-chah-nulth, suffixation to the negative particle wik applies only with 'low'  scope negation, and not with 'high' scope negation. In the 'low' negation example of (78a), the affixal predicate -qaath "claim" suffixes to wik NEG; in the 'high' negation example of (78b), the affixal predicate instead suffixes to the verb Pu-uc"own". (78)  a.  wikqaath?is wik-qaath-?iis NEG-claim-3.IND Ken claims he doesn't  Ken Ken Ken own the  ?uuc suwis?i ?u-ic suwis-?ii 0-own shoes-DET shoes.  wik?iis Tuucqath Ken wik-?iis ?u-ic-qaath Ken 0-own-claim Ken NEG-3.IND Ken doesn't claim to own the shoes.  53  ('low' negation)  suwis?i suwis-?ii shoes-DET ('high' negation)  As with adverbial incorporation, the availability of wik N E G to incorporate relates to linear ordering restrictions on the incorporation process. As Rose (1981: 296) describes, "it is the leftmost and highest constituent of the clause governed by the suffix which serves as base to the suffix". With 'low' negation, wik N E G hosts the affixal predicate, while in 'high' negation, a verb takes over as host. This can be related to the spell-out position of the respective hosts. In 'low' negation, the negative particle wik appears in the derivational sister to the affixal predicate -qaath "claim", and is spelled-out adjacent to the affixal predicate. (Chapter 5 provides argumentation for the syntactic representation that I assume here.) In 'high' negation, however, the verb Pu-uc "own" is spelled-out adjacent to the affixal predicate. Only in 'low' negation does the negation particle wik (NEG) fall into a position where it is the left-most element in the derivational sister of -qaath "claim", as shown in (79a). In (79b), in contrast, it is the verb Pu-uc "own" which is the left-most element of the derivational sister of -qaath "claim". (79) a. 'low'negation  b.  vP  'high'negation NegP  suwisPi the shoes  suwisPi the shoes  Thus, due to the string adjacency effect on incorporation, for (79a), a linearization of <wik-qaath Puuc suwisPi Ken > is anticipated. In (79b), the ordering of <Puucqaath suwisPiKen> is predicted. This linearization mechanism is not sensitive to a difference in syntactic category between the negation host wik (NEG) and the verbal host Pu-uc "own".  2.4 Absence of LF effects Under a model in which L F effects are restricted to the narrow syntax (Chomsky 1995), spell-out to PF is predicted to have no semantic effects. This section considers two domains in which an observable L F effect is absent from Nuu-chah-nulth PF Incorporation: first, the referentiality of the incorporee (§2.4.1), and secondly, scopal effects (§2.4.2).  54  2.4.1 Discourse transparency In many languages in which verbs affix to nouns, the incorporated noun shows distinct referential properties from independent nouns in the language (Mithun 1984).  As Mithun reports, in  Huahtla Nahuatl, when a noun is first introduced in the discourse, it must not be incorporated. The following examples are from Mithun (1984: 860-861, ex. (58)). (80) Nahuatl A:  askeman ti-'-kwa never you-it-eat You never eat meat.  nakatl. meat  B:  na' ipanima ni-naka-kwa. I always I-meat-eat I eat it (meat) all the time.  In the first portion of this discourse, the noun nakatl "meat" occurs independently of the verb kwa "eat". In the following sentence, however, naka "meat" is able to be incorporated into the verb. The incorporability of the noun corresponds to its discourse role in Nahuatl. The restriction that incorporated nouns may not appear as the initial token of the argument suggests that incorporated nouns in Nahuatl cannot establish a discourse referent. If incorporation in Nuu-chah-nulth occurs at spell-out to PF, then it is predicted that incorporated nouns should be discourse transparent. That is, the spell-out properties of the noun should have no consequence for discourse effects.  Indeed, as this section demonstrates,  incorporation of an element into an affixal predicate in Nuu-chah-nulth does not affect this element's ability to occur as a discourse antecedent (Rose 1981). The referentiality of a Nuuchah-nulth incorporee is indicated in examples (81a-c). In (81a), the incorporated bound nominal capx- "man" serves as the discourse referent for the pronoun (pro) "he" in ?aphiqaqit?is"Yie. was very friendly". Similarly, the free nominal siicpax "cougar" which is incorporated into the affixal predicate  in (81b) acts as the antecedent for the object  qaaccanak'uuh/twaPis John "John saw three (of them)".  of the following  sentence,  A final example of the discourse  transparency of incorporated elements is given in (81c), in which ?a9daqum+ "two kinds" is incorporated into the affixal predicate, but still serves as the discourse referent for the following question waayaqhwaimahsak"Which of them do you want to use?". (81)  a.  capxcaasitsis hintsifc ?aphiqaqif?is capx-caas-mit-siis hint-sifc ?aphiiq-aq[+S]-mit-?iis man-beside-PST-1 SG.IND come-PERF friendly-AUG-PST-3. IND I was (sitting) beside a man coming this way. He was very friendly.  55  b.  sicpaxpa-FPis siicpax-pai-[+Sl-?iis cougar-around-3,lND There's cougars around.  qaaccanakuuhitwaTis qacca-nakuuh[+L1-mit-wa?is three-observe-PST-3 .QUOT John saw three (of them).  John John John  c.  ?a&taqum4naksis waayaqhwa+mahsak ?a^-taqum+-naak-siis waayaq-hwa+[+L]-rnahsa-k two-kinds-have-1 SG.IND which-use-want-2SG.Q I have two sets (eg. of lahal bones). Which (of them) do you want to use?  The referentiality of the incorporee follows if PF Incorporation has no L F consequences.  2.4.2 Absence of scopal effects Additional evidence for the lack of an L F effect comes from quantifier scope. As predicted by the local spell-out hypothesis, the choice of host for an affixal predicate should have no effect on scope at L F . As predicted, the surface order derived by PF incorporation appears to have no consequence for quantifier scope. Quantified subjects are ambiguous between wide and narrow scope over objects if the object is incorporated into the affixal predicate or if it is not. In (82), 9  the object muunaa "engine" of the affixal predicate Puu-taq "fix, work on" hosts the predicate, and linearly precedes the quantifier hisuk "all".  Both a wide scope and a narrow scope  interpretation of the quantifier are permiited. (82) a. muunaataqifPis hisuk muunaa-taqr+L] -mit-?iis hisuk motor-fix-PST-3.lND all-DUR All the men were working on an engine.  caakupiih. caawacna?aa+?a+ cakup-iih[+L] cawa-na'Taa+[+L]-0-?a+ man-PL one-handle-3.ABS-PL They were all working on the same one.  b. muunaataqifPis hisuk caakupiih muunaa-taq[+L]-mit-?iis his-uk cakup-iih[+L] motor-fix-PST-3 .IND all-DUR man-PL All the men were working on an engine. They each had one.  cacawanak cawa-nak-0-[+R] one-have-3.ABS-PL  In the following example, muunaa "engine" does not host the affix, and it instead appears below the quantifier. Here, the expletive Pu- hosts the affixal predicate Puu-taq "fix, work on". Still, both scope interpretations are possible. (83)  9  a.  ?uutaqit?is hisuk 7u-taq[+L]-mit-?iis his-uk 0-fix-PST-3.lND all-DUR All the men were working on  cakupiih muunaa caawacna?aaF?a+ cakup-iih[+L] muunaa cawa-naTaa+[+L]-0-?a+ man-PL motor one-handle-3.ABS-PL an engine. They were all working on the same one.  b.  TuutaqifPis hisuk cakupiih muunaa 7u-taq[+L1-mit-?iis his-uk cakup-iih[+L] muunaa 0-fix-PST-3.IND all-DUR man-PL motor All the men were working on an engine. They each had  The source of this ambiguity remains to be determined.  56  cacawanak cawa-nak-0-[+R] one-have-3.ABS-PL one.  The equivalence of (82a)-(83a) and (82b)-(83b), respectively, follows under an analysis in which spell-out to PF has no L F effect.  2.5 Edge effects As I have described, an affixal predicate in Nuu-chah-nulth incorporates a host chosen from its derivational sister, the complement of the verb. This linearization process has been stated to be insensitive to syntactic category.  A l l else being equal, we should therefore expect that  incorporation should be possible when any of NP, DP, vP or C P are the complements of the affixal predicate. (84)  a.  b. V  NP  c. V  DP  d. V  vP  ^ V  \ CP  This section presents evidence that not all else is equal. I hypothesize that D P and CP differ from other projections in that they are 'self-contained' units in the derivation (cf. the notion of 'phase' in Chomsky 2001). The borders of DPs and CPs therefore constitute 'edges'.  10  Intuitively, this  corresponds to the propositional independence of DPs and CPs. In (85), the D P and CP qualify equally as the propositional complement of recall. (85)  a.  I recalled [the city's destructionjrjp.  b.  I recalled [that the city was destroyedJcp.  Furthermore, a variety of syntactic evidence has been presented for an inherent symmetry between D and C, to the exclusion of other categories (Abney 1987, Szabolcsi 1994). According to these analyses, DPs and CPs possess parallel functional structures. According to the DP/CP edge hypothesis, when a D(P) or C(P) is spelled-out, the contents of these phrases become inaccessible to further derivational manipulations (cf. Chomsky 2001, Svenonius 2004, Fox and Pesetsky to appear). In effect, the details of the make-up of DPs and CPs are removed from the workspace of the computation.  A n attractive conceptual  consequence of this postulation is that it reduces the amount of information held in active memory (Chomsky 2001, Matushansky 2005). As this section demonstrates, Nuu-chah-nulth lacks any 'escape hatch' at the edge of CP or DP, which would allow a host for affixation to cross these opaque derivational domains. In Nuu-chah-nulth, there is independent prosodic evidence that DPs and CPs constitute phonologically 'impenetrable' domains. This evidence comes from the two distinct cliticization domains found in Nuu-chah-nulth. (See Chapter 3 for further discussion.) Clitic strings may be  Thanks to Rose-Marie Dechaine for suggesting this term. See Svenonius (2004) for a related usage.  57  built up within a DP, or at a clausal level which excludes the DP(s). In (86), there are two independent cliticization domains, indicated by bracketing. In the C P domain, the clitics -mit (PST)  and -siis(\SG.IND)  are found; in the D P domain are placed the clitics - u £ ( P O S S ) and -?iitk  (2SG.PS).  (86)  [hiixtaqci/n/te/il p hiixtaq-cip-zmif-.s/is C  d o m a  i  [huupuu£ asuA#'/£] domain huupuuk as-u£-/}}Vi: w  n  D P  w  have.accident-BEN-PST-1SG.IND car-POSS-2SG.PS  I had an accident with your car. As presented in more detail in Chapter 3, these DP-level and CP-level clitics may each be described as 'second position' morphemes: they occur (potentially in a string) at the left edge of the phrase, attached in 'second position' as a suffix to a host.  The DP/CP edge hypothesis  presents a straightforward means of explaining how two sets of items (DP-level and CP-level clitics) may each equally be described as 'second position' morphemes, when in fact they occur in distinct positions in the clause. According to the DP/CP edge hypothesis, there is no single 'second position' in a clause, because the derivation is considered in distinct chunks: CPs and DPs are independent domains, and each have their own 'second position'. The derivational independence of CP is also reflected in a ban on long-range syntactic movement in Nuu-chah-nulth.  Nuu-chah-nulth lacks cross-clausal w/i-movement (Davis and  Sawai 2001). The example in (87b) shows an ungrammatical attempt at w/i-questioning the subject of an embedded clause which occurs as the complement of the matrix predicate taaquk "believe". (87)  a.  t'aaquk i?a&?is t'aaquk-ci9e-'a?e-?iis  John ?in kuuwi+ifhuk Mary capac John ?in kuuwi+-mit-huuk Mary capac believe-PERF-TEMP-3.IND John C O M P steal-PST-3.DEP Mary canoe John believes that Mary stole the canoe, (adapted from Davis and Sawai 2001: 133) w  b. * ?acaqh ?acaq-h who-3.Q Who does  t'aaquk i?ak taaquk-ci?e-'ak  John ?in kuuwi+ithuk capac John ?in kuuwi+-mit-huuk capac believe-PERF-TEMP John COMP steal-PST-3.DEP canoe John believe stole the canoe? (adapted from Davis and Sawai 2001: 133) w  Instead, as I describe in Chapter 5, 'long' w/i-movement in Nuu-chah-nulth only occurs in monoclausal 'restructuring' environments. In §2.5.1 and §2.5.2, I show that PF Incorporation can never cross a DP or CP in Nuuchah-nulth. According to my proposal, the members of DPs and CPs are inaccessible as hosts for an affixal predicate when the contents of the DP or CP are removed from active memory. This 'edge' effect is schematized in (88).  58  (88)  When an affix -a taking a D P or CP complement is spelled out in Nuu-chah-nulth, the expletive Pu- must surface in order to provide a host for the affix and rescue the derivation. In 2.5.3, I discuss the source of this opacity effect under the local spell-out model.  2.5.1 DP edges and the bare nominal requirement In Nuu-chah-nulth, an affixal predicate may not suffix to an element of its complement when the phrase contains the determiner -PH. The consequence of this restriction is a bare nominal requirement, on incorporation. In (89a), incorporation targets the bare nominal -hicPin "dress". The example in (89b) indicates that a nominal marked with the determiner -PH cannot be incorporated. Furthermore, as shown in (89c), the determiner cannot itself act as a host for the affixal predicate  When the nominal complement of an affixal predicate is definite, /i/-support  must occur, as shown in (89d). (89)  a.  Fuc?insiikitsis •fuc?in-siik-mit-siis dress-make-PST-1 SG.IND  I made a dress. b.  * Fuc?in?isiikitsis Fuc?in-?ii-siik-mit-siis dress-DET-make-PST-1 SG.IND  I made the dress. c.  d.  * ?iisiikmitsis ?ii-siik-mit-siis DET-make-PST-1 SG.IND I made the dress.  FucPin +uc?in dress  Tusiikitsis Tu-siik-mit-siis  +uc?in?i -hic?in-?ii  0-make-PST-1S G.IND  dress-DET  I made the dress. This restriction may be attributed to the identity of DPs as 'edged' units in the syntax. The presence of a determiner, as in (89b-c) imposes an 'edge' between the affixal predicate and its complement. In such contexts, an expletive host must be inserted to satisfy the affixal predicate's affixation requirement, as in (89d). This restriction on incorporation in contexts with the determiner -PH can also be illustrated in the following stretch of discourse. The sentences in (90) establish the context for the test sentences in (91).  59  (90)  wa+aakitsis naquwas ?athiimit?i ?uk ink Tuushyumsukqs Lucy waraak-mit-siis naqu-was ?athii-mit-?ii ?u-k ink Tuushyums-uk-qs Lucy go.to-PST-1 SG.IND drink-building night-PST-DET 0-with friend-POSS-lSG.PS Lucy I went to the bar last night with my friend Lucy. w  w  qaccii+it?is caakupiih mama+ni huwayii?ath ?uh?iis qacca-ii-t-mit-Tiis cakup-iih[+L] mama+ni huwayii?ath ?uh?iis three-AUX-PST-3.IND man-PL white black and There were three guys there: a white guy, a black guy, and a First Nations  quu?as quu?as First.Nations guy.  This context establishes the nominal huwayiiPath "black person" as discourse-familiar. In this context, the preferred means of stating "Lucy liked the black man" is with an expletive host for the affixal predicate Puu-Pa-teumhi "to like", as in (91a). It is marked for the definite nominal huwayiiPath "black person" to be incorporated, as shown in (91b). Furthermore, it is not possible to incorporate the nominal if it is marked by the determiner -?//, as in (91c). (91)  a.  ?Aiu?a+sumhimit?is ?u-^a4^umhi[+L]-mit-?iis 0-like-PST-3.lND Lucy liked the black man.  Lucy Lucy ,Lucy  huwayii?ath?i huwayii?ath-?ii black.person-DET  b.  #  huwayii?ath?a+sumhimif?is huwayii?ath-'a4-sumhi-mit-?iis black.person-like- PST-3.IND Lucy liked the black man.  Lucy Lucy Lucy  c.  *  huwayii?ath?i?a+sumhimit?is huwayii?ath-?ii- a't sumhi-mit-?iis black.person-DET-like-PST-3.IND Lucy liked the black man.  Lucy Lucy Lucy  ,  ,  The restriction that a discourse-familiar nominal cannot serve as a host for the affixal predicate follows from the bare nominal requirement on Nuu-chah-nulth incorporation.  I attribute this  effect to the status of DP as an 'edged' constituent.  2.5.2 CP edges and 'restructuring' effects Parallel examples may be supplied for the ban on incorporation across CPs. Incorporation is possible only out of uninflected complements, not full CPs (see Chapter 5). In (92a), an example ^ is given showing a full CP complement for the affixal predicate Puu-nak'uuh "observe". This full complement contains the complementizer Pin ( C O M P ) , the past tense marker -mit (PST), and the dependent mood inflection -suuk ( 2 S G . D E P ) .  In contexts of verb incorporation, it is  ungrammatical for such clausal demarcations to appear, as indicated in (92b). The grammatical instance of verb incorporation in (92c) shows no complementizer, no tense marking, and no dependent mood inflection. In effect, the clausal content of (92c) is 'restructured' to exclude these clausal demarcations.  60  (92)  a.  tuuxtuux amitsuk tuux -a [+R]-mit-suuk jump-IT-PST-2SG.DEP 0-observe-PST-l SG.IND COMP I observed that you were jumping. w  w  b.  * muxmux amitnakuuhitsis tuux -a r+Rl-mit-nakuuh[+Ll -mit-siis jump-lT-PST-obserye-PST-1 SG.IND I observed you jumping. w  w  c.  (Pin) suwa (?in) suwa (COMP) you  tuuxtuux ana^uuhitsis suwa mux -a[+R]-nakuuh[+L]-mit-siis suwa jump-lT-obserye-PST-1 SG.IND you I observed you jumping. w  w  In Chapter 5, I will analyse the verb incorporation case in (92c) as having a vP complement, rather than the sort of CP complement in (92a). The inability of incorporation to occur across a CP complement may be attributed to the status of CP as an 'edged' domain. When no CP is present, there is no demarcation imposed between the affixal predicate and its complement. r  2.5.3 Comparison to 'phasal' spell-out According to the local spell-out hypothesis, the spell-out mechanism applies iteratively over the derivation, directing the outputs of syntactic concatenation to interpretation at the L F and PF interfaces. A characteristic of this model is that the mapping mechanism of spell-out does not remove the interpreted elements from the derivation. As first discussed in Chapter 1, the Order Preservation hypothesis (Fox and Pesetsky to appear) states that information established by spell-out is not deleted over the course of the derivation. As was shown in §2.2.1, the proposed iterative spell-out model allows a build-up of affixes, which occurs when one affix finds a host at spell-out, and this affix-host complex in turn serves as a host for another affixal at a subsequent application of spell-out. In (93), if - a and -8 are each suffixes, then a linearization of (|>-8 is induced in a first round of local spell-out. Next, when -oc is spelled-out, its host is this complex, yielding a linearization of <))-8-a. (93)  y <£.  local spell-out  B <.  -a  local spell-out  -8 If spell-out removed an affix and its host from the derivation, then such strings of affixes would be predicted to be impossible with iterative spell-out. That is, if spell-out occured at (3, and if this spell-out removed the contents of (3 from the derivation, then -cc and -8 would be predicted to  61  belong to distinct phonological domains. Instead, I assume that spell-out enriches the derivation by establishing cumulative linearization (see also Fox and Pesetsky to appear). A contrasting view is offered by Chomsky (2001), who hypothesizes that spell-out to the interfaces has the effect of removing spelled-out elements from the derivation. According to Chomsky (2001), spell-out domains form phonologically and semantically opaque phases. These spell-out domains are assumed to be the sisters to C(P) and v(P), although Chomsky leaves open the possibility that other categories, such as the sister of D(P) (Fox and Pesetsky to appear), belong to this set. According to this phase hypothesis, when a spell-out node (eg. C(P), v(P)) is reached in the derivation, the sister to the node is made inaccessible to further operations. This inaccessability is imposed by the Phase Impenetrability hypothesis (Chomsky 2001), as defined below. (94) Phase Impenetrability If X P is spelled-out, the sister to X may not be accessed later in the derivation. The diagram in (95) illustrates how C(P) imposes a boundary between two phases of the derivation. (95) V  CP  phasal boundary  In (95), the V and C remain active in the derivation, while the sister to C - TP - is made inaccessible to operations after spell-out. According to this analysis, the spell-out node of C(P) is an 'escape hatch' for movement in that it forms a potential link for movement across the phasal boundary: elements within the C P phase may move to C(P) to carry on in the derivation. D(P) likewise functions as an 'escape hatch', under the view that it is a designated spell-out node. Assuming that the complementizer ?in  (COMP)  in Nuu-chah-nulth occupies C, the Phase  Impenetrability hypothesis makes an incorrect prediction regarding the spell-out properties of affixal predicates. According to this hypothesis, the head of a spell-out domain (eg. C, v) should be available to the higher phase. For example, in (95), C will attain spell-out along with V when the next phase is reached. This predicted behaviour does not appear to occur in Nuu-chah-nulth. When an affixal predicate takes a C P complement, the complementizer is not available as a host to the affixal predicate. In (96a), it is shown to be ungrammatical for ?in ( C O M P ) to host the  62  affixal predicate -nalcuuli "observe". Instead, when the affixal predicate takes a C P complement, the expletive /i/-mandatorily hosts the affix, as in (96b). (96)  a.  * ?iinna£uuhitsis tuuxtuux amitsuk ?in-nakuuhr+L1-mit-siis tuux -a[+R]-mit-suuk COMP-obserye-PST-l SG.IND jump-lT-PST-2SG.DEP I observed that you were jumping. w  w  b.  Tuunakuuhitsis ?in tuuxfuux amitsuk ?u-nakuuh[+L]-mit-siis ?in tuux -a[+R]-mit-suuk 0-observe-PST-l SG.IND COMP jump-IT-PST-2SG.DEP I observed that you were jumping. w  w  This behaviour is unexpected if the affixal predicate and the complementizer were spelled-out together.  Thus, the technical implementation of the Phase Impenetrability hypothesis for CPs  appears to be problematic for Nuu-chah-nulth. This problem is avoided under the current DP/CP edge proposal, since, by this analysis, C P is an altogether distinct domain of the derivation. (97)  a. DP/CP edge hypothesis: V  b. Phase Impenetrabttity hypothesis  /'" CP C  V TP  CP C  /  phase boundary TP  Under the local spell-out model, what forces this special status of DP and CP? If spellout proceeds phase-by-phase (Chomsky 2001), the privileged behaviour of DP and CP falls out directly from their status as designated spell-out nodes. With local spell-out, in contrast, DP and CP are not special spell-out units: every instance of syntactic Merge induces spell-out. Tentatively, the opacity effects associated with DP and CP in Nuu-chah-nulth may reduce to parametric variation in choice of spell-out domains. According to this idea, choice of phasal or local spell-out is linked to whether D P and CP are 'active' derivational nodes in the language." If spell-out in a language proceeds cyclically at the level of D P and C P phases, then the active status of the D P and C P nodes corresponds to the existence of 'escape hatches' at the boundaries of each phase - allowing these units to form transparent domains for movement (Chomsky 2001). Languages with local spell-out show contrasting behaviour. In Nuu-chah-nulth, 'escape hatches' appear to be lacking in that D P and CP are opaque domains. Affixation cannot occur across a DP or CP in Nuu-chah-nulth; nor can syntactic movement cross a CP (Davis and Sawai 2001; see  11  Thanks to Henry Davis (p.c.) for bringing up this idea.  63  Chapter 5).  12  In effect, DP and CP are inert in Nuu-chah-nulth. Their impoverished role may be  related to their failure to act as designated spell-out nodes. The spell-out properties of vP are unlike that of DP and CP in Nuu-chah-nulth. The vP domain is transparent for affixation: no 'edge' effects occur. These transparency effects are inconsistent with a phasal spell-out analysis of Nuu-chah-nulth. According to Chomsky (2001), vP is a phase. Thus, assuming phasal spell-out, the sister to vP is predicted to be an impenetrable domain, separated from vP and higher categories by a phasal boundary. (98)  TP  A consequence of the.phasal boundary in (98) is that the spell-out of V and N P should be fully independent from the spell-out of categories higher than v.  By the Phase Impenetrability  hypothesis, a phasal boundary entails that lower portions of a derivation are inaccessible to higher ones. In Nuu-chah-nulth, however, the independence of the phasal domains in (98) does not occur. What is observed in Nuu-chah-nulth is a cumulative build-up of affixes and hosts (§2.2.1), which appears to be insensitive to a vP 'boundary'. A n affixal predicate and its host forms a single phonological domain (a word) together with a range of affixal morphemes, including tense and agreement clitics. In (99), for example, the affixal predicate - 'in+ "serve" and its host camas "sweets" forms a word with the tense morpheme -m/t (PST) and the agreement morpheme -siis(\  (99)  SG.IND).  camayin-Htsis camas-'urfr-mit-siis sweets-serve-PST-1 SG.IND  I served sweets. I will argue in Chapter 3 that tense and agreement morphemes in Nuu-chah-nulth are clitics whose position is not determined by raising of the verb. Thus, the single phonological domain which tense and agreement morphemes occupy along with the verb is not predicted to exist, according to an analysis in which tense and agreement have a spell-out domain fully independent from the verb and its complement.  Additional research is required into movement out of DPs. A prediction is that possessor raising should not be possible out of a DP. The determiner is obligatorily absent in possessor-raising contexts (Ravinski 2005).  64  A related problem for the notion that vP imposes a phasal boundary in Nuu-chah-nulth is the possibility of affixation across a vP phrase.  When an affixal predicate takes a verbal  complement, a host for the affixal predicate is chosen from the complement. In (100), the affixal predicate -f«tr"come upon" suffixes to its host hacuk "deeply". (100)  hacukfifotsis hacuk-fik-mit-siis deeply-come.upon-PST-1 SG.IND I came upon Ken in a deep sleep.  wa?ic Ken wa?ic Ken sleep Ken  Syntactically, this complex may be represented as in (101), in which the predicates Pu-Tik"come upon" belongs to a different verbal projection than hacuk "deeply". (The syntactic structure I assume in (101) will be motivated in Chapters 3 and 5.) (101)  vP  Ken phasal boundary ADV hacuk deeply  wa?ic sleep  The Phase Impenetrability hypothesis predicts that a boundary should be imposed between the matrix affixal predicate and its complement. By this view, an affixation relationship established by spell-out is not predicted between -ffit-"come upon" and its attested host, hacuk "deeply". On the assumption that verbs (vP) form distinct spell-out domains, the complex in (99) is thus problematic. In contrast, as I will describe in greater detail in Chapter 3, the local spell-out hypothesis entails that components of the clause above and below a vP are linked through iterative, monotone spell-out: there is no evidence for an 'edge' associated with a vP in Nuuchah-nulth. Assuming that the opacity effects of DP and CP in Nuu-chah-nulth relate to parametric variation in spell-out algorithms (i.e. phasal spell-out or local spell-out), a possible implication is that Universal Grammar does not accord the same privileged status for spell-out of vP as it does for DP and CP. In other words, in the local spell-out framework, the failure of vP to form a  65  special spell-out domain does not appear to render this unit derivationally inert. This matter requires further study.  2.6 Conclusion In this chapter, evidence was presented for the analysis that the positioning of affixal predicates in Nuu-chah-nulth is a reflex of the need to linearize bound morphemes.  According to the  proposal, the affixal predicate must find a host from within its derivational sister at the point of spell-out. This derivational sister is a linearized object, leading to the restriction that suffixation operates on the basis of linear adjacency.  I have argued that the notion of local spell-out  elegantly captures the dual sensitivities which affixal predicates show to derivational sisterhood and linear adjacency.  The syntax indirectly conditions the input to the linearization process  through its composition of local spell-out domains. Two conditioning effects of the syntax were emphasized: the first is the locality requirement induced by the binary concatenation of the syntax, which yields a complement restriction in Nuu-chah-nulth; the second, the creation of DP and CP domains which form impenetrable domains for affixation processes. A purely syntactic analysis of incorporation fails to predict the sensitivity which Nuuchah-nulth affixal predicates have to linear adjacency.  As described in §2.2.2, an affixal  predicate incorporates a modifier, rather than the element which syntactically heads the complement.  Such insensitivity to hierarchical adjacency is at odds with the syntactic  incorporation process described by Baker (1988), as it violates the syntactic Head Movement Constraint (Travis 1984). Moreover, in §2.2.3, it was shown that incorporation in Nuu-chahnulth displays an insensitivity to a restriction on extraction from a conjoined object, the Coordinate Structure Constraint.  This constraint was shown to hold in Nuu-chah-nulth in  contexts of true syntactic movement. Just as a purely syntactic analysis of Nuu-chah-nulth fails, so too does a strictly phonological one.  Note that the host for an affixal predicate cannot be defined in strictly  phonological terms, such as a syllable or a prosodic foot (cf. Halpern 1992). Instead, the host is equivalent to a unit of the syntactic derivation (one of the two elements treated by syntactic Merge). As the following examples illustrate, hosts in Nuu-chah-nulth can be mono- or polysyllabic. The host in (102a), Puus "someone", is monosyllabic. The host in (102b), quu?ac"(First Nations) person", is bisyllabic. Finally, the host in (102c), mama-hiiq- "white person" is trisyllabic.  66  (102)  a.  ?uusi+?a?c?is ?uus-'ii-'a?e-?iis someone-inside-TEMP-3.IND There's someone inside (the house) now.  b.  quu?aci-F?is quu?ac-'rt-?iis person-inside-3.IND There's a person inside (the house).  c.  mama+niqi-FPis mama+niq-'rf-Piis white.person-inside-3.IND There's white people inside (the house).  Feet in Nuu-chah-nulth are maximally bisyllabic (Werle 2002, Stonham 2004). Thus, hosts for an affixal predicate may be equal to, smaller than, or larger than a prosodic foot.  These  mismatches with prosodic definitions lend support to a syntactically-constrained definition of the host (i.e. a derivational sister). In the following chapters, it will be shown that the combinatory properties of affixal predicates are conditioned by their argument structure.  The syntactic positions in which  arguments are introduced indirectly constrains the post-syntactic linearization operation of affixation.  Argument structure determines derivational sisterhood, which defines the domains  which local spell-out applies to.  The next chapter serves as an introduction to the clausal  architecture of Nuu-chah-nulth.  67  CHAPTER 3  Clausal architecture of Nuu-chah-nulth Every phrase and every sentence is an end and a beginning... ~T.S. Eliot Four Quartets no. 4  3.0 Introduction Up until this point, there has been an implicit assumption that affixal predicates take an argument as their derivational sister. The following chapters will provide evidence for such a conclusion. In this chapter, I take a first step towards this analysis by arguing that nominal arguments of Nuuchah-nulth predicates are introduced within verbal projections of the affixal predicate.  This  analysis has consequences for the local spell-out hypothesis, since it determines that an argument (or part of an argument), as the derivational sister of an affixal predicate (V), will be selected as the host for an affixal predicate. (1)  VP V  <•  local spell-out  NP  This chapter adopts the analysis that the 'basic' word order for Nuu-chah-nulth is VOS. I propose that this word order is not derived via movement; instead this ordering results from a syntactically underived predicate-initial system in which subjects appear in right-linearized specifier positions. The consequence of this analysis is that when an affixal predicate (V|) takes a propositional complement, the embedded verb (V ) precedes the embedded arguments as the 2  leftmost element in the complement. This is represented in (2), in which the affixal predicate's complement (circled in the diagram) is assumed to be equivalent to vP. (2)  VP  V  2  < l o c a l spell-out  0 /  68  The v head is represented here as phonologically null (0).' As such, the embedded verb (V ) is 2  string adjacent to the affixal predicate, (V,). By the string adjacency effect, it therefore follows naturally that the verb (V,) should select the embedded verb (V ) as its host at spell-out. 2  In the next section, we turn to a discussion of how the PF component is responsible for the linearization of syntactic terminals.  According to the analysis, syntactic outputs are  linearized at the point of spell-out.  3.1 Linearization of syntactic terminals In the framework assumed for this dissertation, the syntax is equivalent to the structure-building operations of Merge or Move ("remerge").  The syntactic constructs formed by this binary  concatenation are unspecified for linear order. (3)  Merge: concatenate a with (3, forming y  Therefore, if Merge (X, Y ) applies, this does not dictate that X must precede Y , or vice versa. Thus, although it is possible to pictorially represent Merge (X,Y) as in (4a) or (4b), by default the two must be understood as syntactically equivalent representations. (4)  a.  XP X  b. Y  XP Y  X  Indeed, in fixing the right-left orientation of the elements X and Y , these two-dimensional diagrams unavoidably overspecify the unlinearized nature of the structure-building operations of the syntax. Although the binary concatenation of Merge or Move fails to impose a linearization of the joined elements, it is nonetheless an unmistakable characteristic of natural language that linear orderings do exist. In the following English sentence, for example, not must precede rushed, and not the reverse. (5)  a.  I am [not rushed],  b.  * I am [rushed not].  If linearization is not reducible to the structure-building properties of the syntax, then what is responsible for its effects? In the Minimalist model, the linear ordering of linguistic constructs constitutes a bare output condition at PF (Chomsky 1995). The temporal nature of speech determines that linguistic representations must be mapped to a linearized speech stream. Thus, the 'phonologization' of syntactic constructs entails that relative orderings must be imposed on the syntactic terminals {a, P). In the previous chapters, I introduced the idea that affixation is ' The v may be overt or covert in Nuu-chah-nulth. If overt, the v is realised as an aspectual suffix on the embedded verb.  69  one means by which ordering may be imposed on these elements. However, this linearization mechanism is only applicable if one or the other of a and P is an affix. Nothing has yet been said about linearization in non-affixal contexts. Linearization is itself necessary, but what particular linearization scheme of syntactic terminals is required?  Two hypotheses may be distinguished, one of which attributes  linearization to an invariant syntax-phonology mapping, and the other which opens the door to variations in linearization schemes for syntactic terminals: (6) PF Linearization mechanisms (i) Universal linearization scheme (ii) Directionality conventions In the following subsections, I will consider each of these possibilities in turn, and will make arguments for the latter.  3.1.1 A universal linearization scheme? According to the Universal Base Hypothesis (Lakoff 1970, Bach 1968), all languages share a common architecture. With respect to linearization schemes, Kayne (1994) presents the hypothesis that languages have an invariant mapping algorithm from syntax to linear ordering. According to Kayne's Linear Correspondence Axiom (LCA), the structural relationships of the syntax unambiguously determine ordering in that asymmetric c-command entails precedence. Originally formalised as a syntactic filter, this principle has been recast in various Minimalist approaches as a post-syntactic principle which applies at spell-out (eg. Chomsky 1995, Dobashi 2003). This linearization scheme may be defined by the following statement: (7) Linear correspondence axiom If X asymmetrically c-commands Y , then the terminals in X precede the terminals of Y . This axiom entails that the sole linearization option available for a head projecting a specifier (Spec) and taking a complement (Comp) is that shown in (8a).  70  (8) Linearization schemes (a) "Spec - Head - Comp"  (b) "Spec - Comp - Head"  a y  a a  y  p  a  (c) "Head - Comp - Spec"  a  p  a  (d) "Comp - Head - Spec"  a a a  a y  oc  p  p  y  a  C-command relationships of the above configurations can be evaluated according the following definition in (9), supplied by Reinhart (1979). (9)  2  X c-commands Y iff: (i) The first branching node dominating X dominates Y , and (ii) X does not dominate Y , and (iii) X is not equal to Y  Asymmetric c-command, in turn, may be determined through reference to the following additional statement: (10)  X asymmetrically c-commands Y iff: (i) X c-commands Y , and (ii) Y does not c-command X Let us now consider how the L C A applies to each of the linearization schemes in (8).  The "Spec - Head - Comp" configuration of (8a) satisfies the L C A , since the specifier (y) both asymmetrically c-commands and precedes the head (terminal a); the head (terminal a), in turn, asymmetrically c-commands and precedes the complement (P).  Assuming transitivity of  precedence relations, asymmetric c-command therefore exhaustively orders the terminals in (8a). The structure of (8b), in contrast, violates the precedence requirement with respect to sequencing Epstein (1999) proposes a derivational notion of c-command which follows from the properties of binary concatenation. According to derivational c-command, an element c-commands those elements with which it was paired in the course of a derivation. The choice of a representational or derivational view of ccommand does not affect the point that I make here regarding the stipulative nature of the precedence relation associated with c-command. 2  71  the head (a) following the complement (P). According to the L C A , since a asymmetrically ccommands P in (8b), a should precede P, not follow it. The configuration in (8c) shows a similar precedence violation involving the specifier (y). By the L C A , it is impermissable for y to follow the head and complement, since this node asymmetrically c-commands them. Finally, the option in (8d) is ruled out by the L C A , since its linear ordering is the reverse of that required by the L C A . In (8d), nodes follow, not precede, the terminals they asymmetrically c-command. If the L C A is adopted, directionality parameters are inapplicable as a means of accounting for different surface word orders. By the L C A , only a single linearization scheme is universally available, whether it applies to the base-generated syntax (as in the pre-Minimalist formulation originally proposed by Kayne 1994), or in the syntax-to-PF mapping (as in Chomsky 1995).  The consequence of this proposal is that word order variations between languages  (Greenberg 1966) must derive from different combinations of movement.  Before discussing  empirical challenges faced by this state of affairs, let us first consider a key conceptual problem facing the L C A hypothesis.  3.1.1.1 Conceptual problem According to the Linear Correspondence Axiom, asymmetric c-command induces a linear ordering between syntactic terminals. Kayne (1994: 36) argues that this hierarchical relationship consistently results in a precedence relation. However, this particular linearization requirement is a stipulation. There is an alternative ordering which would equally allow a consistent mapping between asymmetric c-command and linear sequencing: if c-commanding terminals follow rather than precede other elements of the tree, then syntactic relations will also strictly determine linear correspondence.  The two possible options for invariant mappings between c-command and  ordering are given below. (11) Linear correspondence axiom (reformulated) option (i): If X asymmetrically c-commands Y , then the terminals in X precede the terminals of Y . option (ii): If X asymmetrically c-commands Y , then the terminals in X follow the terminals of Y . If the first option is selected, then a "Spec - Head - Comp" order is expected (8a). However, if the second option is chosen, then a "Comp - Head - Spec" order arises (8d). Note that the two configurations are mirror images of each other and whatever asymmetric c-command relations hold for one will necessarily hold for the other (see Uriagereka 2002 for discussion).  72  In a crucial sense, asymmetric c-command therefore underspecifies linear ordering. Kayne observes this complication, and rules out the "Comp - Head - Spec" order through an independent stipulation. According to Kayne (1994: 36-38), this alternative order is inconsistent with the asymmetry of time. However, I contend that this is simply a restatement of the phonotemporal mapping problem, rather than a solution. A linearization of any of (8a-d) would be compatible with forward-moving time, in that they each specify a temporal sequence for the terminals. Either of (8a) or (8d) allows an invariant branching-to-linearization mechanism. I therefore conclude that the conceptual motivation is lacking for a single, optimally 'harmonic' linearization scheme induced by the c-command relation of the syntax (cf. Uriagereka 2002).  3.1.1.2 Empirical problems A range of empirical evidence has been put forth in argument against a universal "Spec - Head Comp". It goes beyond the scope of the present discussion to survey them fully, but I will point the reader in the appropriate directions. First, a universal linearization scheme has difficulty representing languages which appear to utilize a configuration which does not match "Spec Head - Comp". Japanese and Korean are two examples of languages which appear to have a general head-final word order (Whitman 1991). In a similar vein, an invariant mapping scheme is inadequate as an account of the variation which has been argued to exist between languages, such as the base V O - O V contrast proposed for Dutch and Basque, respectively (Vicente 2004, 2005). A uniform mapping scheme is also challenged by variation within a language, such as the 'mirror' effects in word order variation in Dutch prepositional phrases (Ackema and Neeleman 2002). Filially, apparent instances of right-ward movement to a specifier position (Beerman, LeBlanc and van Riemsdijk 1997) are problematic for a universal "Spec - Head - Comp" configuration.  3.1.2 Directionality is determined at spell-out As argued in the previous sections, syntactic structures are indeterminate for linear ordering. Yet, linearization is a bare output constraint at the PF interface. A n independent mechanism is therefore necessary for fixing the sequencing of non-affixal syntactic terminals. I present here the hypothesis that linearization schemes arise at the point of spell-out to PF (cf. Chomsky 1995). For Merge (a, P), the syntactic object may be spelled-out as the phonological objects <a, p> or  <p, a>. Regularities in linearization may be attributed to language-specific directionality conventions, as will be discussed in §3.1.2.1. In §3.1.2.2, these conventions are compared to the  73  notion of 'parameters'.  I propose that these conventions are not specific to the linearization of  non-affixal syntactic terminals, and argue in §3.1.2.3 that directionality conventions are also at play in the choice of a prefixation or suffixation pattern for affixal elements. This discussion concludes in §3.1.2.4 with a summary of the directionality conventions assumed for Nuu-chahnulth.  3.1.2.1 Directionality conventions In many languages, syntactic objects and phonological objects stand in an implicational relationship. That is, a syntactic object may be consistently mapped to a particular linearization. The examples in (12) may be understood to be distinct linearization schemes for a syntactic object in which a 'heads' the phrase (owing to the fact that it is the category which projects its label). In a language in which syntactic form has implications for ordering, the syntactic object may be realised regularly as an a-initial linearization; alternatively, it may be consistently ordered as an a-final pattern. (12)  a. a-initial  b. a-final a  a  a  p  P  a  These regularities in linearization may be attribed to language-specific ordering 'conventions' imposed at spell-out. The directionality convention hypothesis may be stated as follows: (13) Directionality convention: Order (a, P). I propose that ordering conventions are established through language acquisition. The decision for an a-initial or a-final ordering is conditioned by the language which the learner is exposed to. I assume that an a-initial or a-final ordering may hold uniformly across the language or it may be specific for certain syntactic categories, dependent on the regularities of the language. In either case, the convention establishes one out of the two logically possible orderings as a regularity at spell-out. Languages with free word order, which show no regularity in ordering, are assumed to lack directionality conventions for spell-out: a linearization is necessarily induced at spell-out, but one linearization scheme is not preferable over the other.  In other words,  ordering is stipulated phrase by phrase and language by language. Let us consider the contrast between English prepositions and Hungarian postpositions as an illustration of the proposed spell-out directionality conventions. Each of these language show regular patterns. In English, prepositions (eg. above, beside, under) reliably precede a nominal. In Hungarian, however, postpositions (eg. folott "above", mellett "beside", alatt "under")  74  consistently follow a nominal. This difference is illustrated in (14) with abovelfoldtt. In (14a), above precedes the table; in (\4b), folott "above" follows az asztal "the table". (14)  a.  The picture is above the table.  b.  A DET  (English)  kep az asztal folott van. picture DET table above BE-3SG  The picture is above the table.  (Hungarian)  It is ungrammatical for English prepositions to follow the nominal, as shown in (15a-b). Conversely, it is ungrammatical in Hungarian for postpositions to precede the nominal, as shown in (16a-b). (15) English a.  There are pictures everywhere: under, over and beside the table.  b.  * There are pictures everywhere: the table under, over and beside.  (16) Hungarian a. Kepek vannak mindeniitt: az asztal alatt, folott, mellett. picture-PL BE-3PL everywhere DET table under above beside There are pictures everywhere: under, over, and beside the table. b.  * Kepek vannak mindeniitt: alatt, folott, mellett az asztal. picture-PL BE-3PL every-where under above beside DET table There are pictures everywhere: under, over, and beside the table.  How are the distinct English and Hungarian patterns derived? Note that if a preposition (P) is concatenated with a nominal (DP) via Merge (P, DP), then there are two logically possible linearizations for this unordered syntactic object: the preposition may precede (17a) or follow (17b) the nominal. (17)  a.  b.  P P  DP  P DP  P  According to the directionality convention hypothesis, the difference between  English  prepositions and Hungarian postpositions reduces to a spell-out regularity in linearization. English learners adopt (17a) as a linearization convention, while Hungarian learners opt for (17b). When exposed consistently to preposition-initial examples in the input, language learners of English carry on the tradition through adoption of an oc-initial directionality convention; Hungarian learners, when faced with preposition-final input, establish an oc-final convention. In the case of English, the pattern for prepositions is consistent with the general head-initial nature of the language. In Hungarian, however, head-final pattern for prepositions may be viewed as a convention specific to this syntactic category: determiners, for example, are subject to a headinitial pattern, as in az asztal "the table".  75  3.1.2.2 Comparison to headedness 'parameters' How does the notion of directionality 'conventions' compare to the principles-and-parameters options for headedness 'parameters'? In the principles-and-parameters framework, learning a language entails selecting one of the possible settings made available by the parameters of Universal Grammar. By this view, language learners are 'hard-wired' (or innately endowed) with a variety of possible parameters. For example, Universal Grammar is postulated to allow either an oc-initial or a-final setting for the headedness parameter(s). (18)  Headedness parameters) (i)  SETTING A  (a-initial):  a precedes (3  (ii)  SETTING B  (a-final):  a follows (3  If a learner is exposed to English prepositions, then exposed to Hungarian postpositions, then  SETTING B  SETTING A  is switched on. If a learner is  is switched on.  What the directionality convention hypothesis shares with this notion of parametersetting is the aspect of acquisitional choices. However, the directionality convention hypothesis diverges from the parameter hypothesis in the grammatical source of these choices. According to the directionality convention hypothesis, a language learner simply selects one of only two logically possible linearizations for a syntactic object at local spell-out: the choices are not innately endowed in the language faculty. This follows from the reasoning that the choices of ainitial and a-final orderings need not be designated biologically, because they are inherently learnable (Newmeyer 2005; Henry Davis, p.c).  3.1.2.3 Affixation patterns The previous discussion was concerned with the linearization schemes of non-affixal elements. According to the directionality convention hypothesis, a language learner selects one of the two logically possible linearizations when it comes time to order a with respect to P at spell-out. This choice determines whether a language utilizes, for example, a head-initial or specifier-final pattern. How does this specification of directionality for syntactic terminals compare with wordinternal directionality? Within a word, a bound element may be designated as either prefixal or suffixal.  A prefixal ordering linearizes the bound ele'ment to the left of its host; a suffixal  ordering linearizes the bound element to the right of its host. I propose that the choice between a prefixation and a suffixation pattern is also a spell-out convention.  In Nuu-chah-nulth, for  example, affixal predicates are consistently linearized as suffixes. A language learner adopts this pattern over the logically possible alternative of ordering these affixal predicates as prefixes.  76  Note that the learner is exposed to abundant evidence for the language-specific choice of suffixation or prefixation in Nuu-chah-nulth. Thus, prefixation-suffixation patterns may be seen as a sub-type of directionality convention. If a or (3 is an affix, then Order (a, B) establishes a preference for a regular prefixation or suffixation linearization.  3.1.2.4 Directionality conventions of Nuu-chah-nulth In this dissertation, five linearization regularities for spell-out are proposed for Nuu-chah-nulth. These order a specifier relative to a head (19a), a head relative to a complement (19b), an adjective relative to a noun (19c), and a manner adverb relative to a modified verb (19d). A final convention establishes the suffixation pattern of affixal elements in Nuu-chah-nulth (19e).  3  (19) Directionality conventions of Nuu-chah-nulth: a.  SPECIFIER-FINAL:  b.  HEAD-INITIAL:  c.  ADJECTIVE-INITIAL:  d.  MANNER ADVERB-INITIAL:  e.  SUFFIXATION:  a head precedes a specifier  a head precedes a complement an adjective precedes a modified noun a manner adverb precedes a modified verb  a host precedes an affixal predicate  As determined by (19a), specifiers are realised in a 'right-branching' orientation in which they follow the head of the phrase. I will argue in §3.3 that this convention applies to subjects of clauses, inducing a subject-final ordering of VOS. According to the convention of (19b), Nuu-chah-nulth has a head-initial pattern in which complements follow heads (Davis and Sawai 2001, Stonham 2004, Ravinski 2005, among others). In Nuu-chah-nulth, heads of phrases are often affixes, which may obscure this headinitial linearization pattern.  However, non-affixal heads such as wik (NEG) show a consistent  head-initial pattern with respect to their complements.  4  NegP  (20) Neg wik  vP wa+sik go home  In the following sentence, wik must precede, not follow, the phrase which it takes scope over, wa-f-si&"go.\\ome  (PERF)".  Clearly this is a heterogenous set. Perhaps these regularities reduce to a more abstract characterization. For the sake of simplicity, I exclude tense (-mit) and subject agreement {-siis) from this diagram. I assume these morphemes are introduced in higher projections, and surface suffixed to wik in (11a) through local spell-out of these affixes. 3  4  77  (21)  a.  b.  wikitsis wik-mit-siis NEG-PST-1 SG.IND I didn't go home.  wa+sik wa+-si& gO.home-PERF  * wa-rsi?eitsis wa+-si9e-mit-siis gO.home-PERF-PST-1 SG.IND I didn't go home.  wik wik NEG  As was discussed in Chapter 2, a consistent ordering relation holds between an adjective and a modified noun in Nuu-chah-nulth (§2.2.2.1). This is captured by the convention of (19c). As shown in (22), the adjective ha?um "tasty" must obligatorily precede the nominal Paapinis "apples". (22)  a.  b.  ?u?iic?is?a+ ha?um ?aapinis ?u-4ic-?iis-?a+ haPum ?aapinis 0-consume-3.lND-PL tasty apples They are eating delicious apples. * ?u?iic?is?a+ ?aapinis Pu-'iic-Piis-Pa-r Paapinis 0-consume-3.IND-PL apples They are eating delicious apples.  ha?um haPum tasty  This initial ordering of the modifier also applies to manner adverbials, as determined by the convention of (19d).  As discussed in Chapter 2, a manner adverbial such as wityax "slow"  precedes a verb in Nuu-chah-nulth (§2.2.2.2). (23)  a.  b.  wityaxits waa-fsik wityax-mit-s wa+-[+L]-si?e slow-PST-lSG.ABS gO.home-CONT-PERF I was going home slowly. *  waa+si9eits wa+-[+L]-si&-mit-s gO.home-CONT-PERF-PST-lSG.ABS I was going home slowly.  wityax wityax slow  The suffixation convention of (19e) determines that affixal predicates are linearized as suffixes to their hosts, and not as prefixes. In (24), the affixal predicate - 'aap "buy" attaches to 5  the right of its host, maht'a- "house", and not to its left. (24)  a.  mahtaPamifPis cakup mahta-'aap-mit-Piis cakup house-buy.-PST-3.lND man A man bought a house.  This suffixation appears to hold for all bound forms in Nuu-chah-nulth, with the exception of bound nominals which host an affixal predicate (§1.3). 5  78  b.  * ?aamahtamit?is cakup 'aap-maht'a-mit-?iis cakup buy-house-PST-3 .IND man A man bought a house.  In the next section, we return to the topic of the linearization schemes for syntactic terminals, with a focus on the syntactic structure of Nuu-chah-nulth.  3.2 Configurationality Before discussing evidence for a VOS linearization scheme in Nuu-chah-nulth in §3.3, let us first cover background topics on the position of arguments in a Nuu-chah-nulth clause.  In this  section, I argue that Nuu-chah-nulth is a 'configurational' language (cf. Hale 1983, Baker 1996) in which DP arguments are introduced with the verbal projections, as in (25). (25)  vP  v  VP V  DP j ob  ect  This section addresses the basic structural properties of the Nuu-chah-nulth language, and argues for a syntactic asymmetry between subjects and objects.  3.2.1 Partial head marking In 'partial head marking' languages, agreement morphology associated with a predicate obligatorily registers some, but not all, arguments of the predicate. Nuu-chah-nulth is a partial head marking language (Davis and Wojdak to appear): portmanteau mood/agreement enclitics agree with subjects (Rose 1981, Davidson 2002). Objects are not registered via agreement, with the exception of a limited form of object marking which appears in imperative environments (see Davidson 2002). The subject agreement paradigm for the Ahousaht dialect is indicated in the following table (see  APPENDIX  C). This table is closely based on Nakayama (1997, 2001),  although I distinguish between "absolutive" and "dependent" moods, as well as propose a "confirmation" mood. For third persons, plurality is optionally indicated by the plural enclitic -?af(Nakayama 1997: 30).  79  (26) Subject agreement in Ahousaht Mood/Person INDICATIVE INTERROGATIVE CONFIRMATION QUOTATIVE ABSOLUTIVE SUBORDINATE DEPENDENT SIMPLE RELATIVE INDEFINITE RELATIVE CONDITIONAL  f' singular -siis -hs -haas -waa?ic'as -s -qs -sa -qs -(y)iis -quus  2 singular -?iick -k nd  -  -waa?ick ?  -k -suuk -?iitk -(y)iik -quuk  3 -?iis -h -haa(c) -waa?is 0 -q -huuk -?iitq rd  -(y)ii -quu  I ' plural -niis -hin s  -waaricin -na -qin -na -qin -(y)in -q in w  2 plural -?iicuus -hsuu nd  -  -waa?icuus -suu -qsuu -suu -?iitqsuu -(y)iisuu -quusuu  The contrast between subjects and objects with respect to registering agreement correlates with the ability to license pro-drop. Subject DPs are freely omissible in Nuu-chahnulth, given appropriate discourse contexts. The sentences in (27) may equally be uttered, for example, in the following specified scenerio: (27) context: You've been out with Kyle on the beach and he saw a sea anemone for the first time. You want to report what happened. a.  naatsiicifot?is kin-kimc naatsii-si9e-mit-?iis kin+umc see-PERF-PST-3.IND sea.anemone Kyle saw a sea anemone.  b.  naatsiicifcit?is naatsii-si&-mit-?iis see-PERF-PST-3.lND  Kyle Kyle Kyle  kin+umc kin+umc sea.anemone  He saw a sea anemone.  -  For objects, however, pro-drop is not freely available, even with appropriate context. An overt object (28a) is permitted in the context below, but a covert object (28b) is not. (28) context: You know that Ken is planning to buy a sweater for his mother. You also know that Ken and Kay were just out shopping. You see a new sweater on the couch, and want to know if Ken bought it. a.  b.  maakuk ith cupcupsum+?i maakuk-mit-h c'upc'upsum+-?ii buy-PST-3.Q sweater-DET . Did he buy the sweater? w  * maakuk ith maakuk-mit-h buy-PST-3.Q Did he buy it? w  80  Following Rizzi (1986), I adopt the analysis that 'rich' subject inflection formally licenses a null pronominal (pro) as subject in cases where overt subject DPs are absent (Davis and Wojdak to appear). In effect, the properties of the subject can be recovered by the inflection.  Objects,  which lack rich inflectional agreement, do not formally license a null pronominal argument.  6  The asymmetry between subjects and objects with respect to agreement and the licensing of pro has key implications for the the structural representations of DPs in the language. Specifically, Nuu-chah-nulth does not meet the definition of a 'polysynthetic' language in the technical sense of Baker (1996). Baker (1996) draws a link between the rich inflection exhibited 7  by certain polysynthetic languages, and the lack of structural asymmetries between lexical DPs. Under Baker's analysis, lexical DPs in these non-configurational 'polysynthetic' languages occupy adjunct positions, while pronominal inflection occur directly as arguments. Pronominal inflection is licensed for theta role assignment by the verb via the Morphological Visibility Condition. This condition states that theta roles of the verb are to be found within the same word as the verbal head, either as (i) inflection or (ii) an incorporated noun. (29) Morphological Visibility Condition (Baker 1996: 17) A phrase X is visible for 9-role assignment from a head Y only if it is coindexed with a morpheme in the word containing Y via: (i) an agreement relationship, or (ii) a movement relationship In Nuu-chah-nulth, however, objects violate Baker's Morphological Visibility Condition on the licensing of pronominal arguments because they are not registered morphologically. Thus, by this line of reasoning, there is evidence that at least object DPs must occupy argument positions in Nuu-chah-nulth. In the next section, I turn to additional evidence for a syntactic asymmetry between the structural representation of subjects and objects in Nuu-chah-nulth.  3.2.2 An asymmetry in possessor raising Evidence for an asymmetry between subjects and objects in Nuu-chah-nulth comes from restrictions on a construction known as 'possessor raising'. In this construction, the possessive marker -uk/-(P)ak  (POSS)  is suffixed to a predicate, instead of (or in addition to) its base position  Nuu-chah-nulth does allow limited pro-drop of objects under certain discourse conditions; this is discourse-licensed and not agreement-licensed. Note that Sapir's (1921) discussion of 'polysynthesis' was partly in response to observations about Nuuchah-nulth. In the original sense of the term, a 'polysynthetic' language is one which exhibits a high morpheme-to-word ratio. 6  7  81  suffixed to the possessum (Davidson 2002, Ravinski 2005).  8  The basic pattern of possessor  raising is illustrated below with the intransitive predicate wiwisfaqk "lazy".  Note that either  possessor raising or possessor doubling behaviours are generally possible. In (30a), the nominal tana "child" is suffixed by the possessive marker and is interpreted as the possessum. In (30b), the predicate wiwisTaqft "lazy" is suffixed by -uk (POSS), while the nominal tana "child" receives no such marking but is still interpreted as the possessum. The example in (30c) shows both the predicate and the nominal suffixed by possessive markers. (30)  a.  wiwis?aq^;?is t'anaakri wiwis'Taqk-riis tana-?ak-?ii lazy-3.IND  ?uustaqyu?i Tuustaqyu-Tii  child-POSS-3.PS healer-DET  The doctor's child is lazy. wiwis^aq?cuk?is tana wiwis Taqfc-uk-?iis tana lazy-POSS-3.INI) child The doctor's child is lazy. <  c.  (unraised) ?uustaqyu?i ?uustaqyu-?ii healer-DET (possessor raising)  wiwisvTaqfcukPis wiwis Taqfc-uk-?iis'  tanaak?i tana-?ak-?ii  lazy-POSS-3.IND  child-POSS-3.PS healer-DET  (  ?uustaqyu?i ?uustaqyu-?ii  The doctor's child is lazy.  (possessor doubling)  In Chapter 4, I discuss the analysis in which the possessive marker -uk (POSS) on the predicate licenses a position for a raised DP possessor (Ravinski 2005). For our present purposes, we can set aside the mechanics of this raising operation and focus instead on the use of possessor raising as a diagnostic for distinguishing between subjects and objects. The  behaviour of transitives with respect to possessor raising gives evidence for a  distinction between subjects and objects in Nuu-chah-nulth. For transitives, there is a restriction that a possessive marker on the predicate is only ever associated with a possessum subject, and not a possessum object (Davidson 2002, Ravinski 2005). As indicated by (31a-b), a subject ("cat") and an object ("bird") are equally compatible with possessive marking in 'unraised' contexts. Crucially, however, it is shown by the interpretation of the possessor-raised (31c) that objects face a restriction which subjects do not. (31)  a.  hink a?iihit?is hin-k a?iih-mit-?iis w  w  piispis maamaatakqs piispis maamaati-?ak-qs  LOC-go.after-PST-3.IND cat  bird-POSS-lSG.PS  A cat was after my bird.  (unraised)  The choice between -uk (POSS) and -(Fjak (POSS) is phonologically constrained. As Ravinski (2005: 25) notes, -uk follows consonants, while -(?)akfollows vowels. Both -uk (POSS) and -(PjaJc(POSS) indicate alienable possession. A different suffix, -/^r(lPOSS) occurs with inalienable possession (see Rose 1981, Davidson 2002, Ravinski 2005).  82  b.  hink a?iihit?is hin-k a?iih-mit-?iis w  w  piispisukqs piispis-uk-qs  LOC-go.after-PST-3.IND cat-P0SS-lSG.PS  maamaati maamaati bird  M y cat was after a bird. c.  (unraised)  hink a?iihuksis hin-k a?iih-uk-siis  piispis piispis LOC-go.after-PST-POSS-1 SG.IND cat = (i) My cat was after a bird. ± (ii) A cat was after my bird. w  w  maamaati maamaati bird (possessor raising)  In (31c), the raised possessor controls clausal subject agreement, as is standard for Nuu-chahnulth possessor raising. The sole available interpretation for (31c) is one in which the nominal "cat" is the possessum; an interpretation in which the nominal "bird" is the possessum is unavailable. This difference in the behaviour of the two nominals reflects a systematic difference between subjects and objects in Nuu-chah-nulth (Ravinski 2005).  3.2.3 An asymmetry in incorporation Further evidence for the configurationality of arguments in Nuu-chah-nulth is the restriction that only complements of affixal predicates are able to act as hosts (Woo 2000; Davis and Sawai 2001; Yiu 2001; Wojdak 2003a,b, 2004). In the realm of 'noun incorporation', arguments which occur as objects of an affixal predicate can be suffixed by the affixal predicate, but, as is indicated by (32b), subjects do not undergo a similar type of suffixation. (32)  a.  haa?um?i9eas?is nuwiiq ha?um-?j&[+L]-'as-?iis nuwiiq food-take-go-3. IND father Father went to get food.  b.  \ nuwiic?i9cas?is ha?um nuwiic-?ifc-'as-?iis ha?um father-take-go-3 .IND food Father went to get food.  In (32a), the affixal predicate ?u-?ik "take" suffixes to the object nominal haPum "food". The example in (32b) shows that it is illicit for the subject nuwiic- "father" to be suffixed by the affixal predicate. Such asymmetries are found across all affixal predicates.in Nuu-chah-nulth, a topic that we will return to in Chapter 4. Possessor raising and incorporation thus pick out complementary sets of arguments: incorporation applies to objects, while possessor-raising applies to subjects. This can be taken 9  Note that the two constructions do not pick out completely complementary sets (Davis and Wojdak to appear). Incorporation operates on underlying objects, including the subjects of unaccusative verbs; in contrast, possessor raising is linked to surface subjects, also including the subjects of unaccusative verbs. 9  83  as strong evidence that Nuu-chah-nulth must distinguish hierarchically between subjects and objects, and therefore that clausal structure'cannot be 'flat' (Davis and Wojdak to appear). In the following section, I provide evidence for a particular asymmetrical representation of subjects and objects in Nuu-chah-nulth. M y analysis of Nuu-chah-nulth clausal structure proposes that canonical predicate-initial word order arises from an 'basic' VOS linearization.  3.3 Predicate-initial word order Word order in Nuu-chah-nulth is predicate-initial. When both subject and object are overtly expressed, either V S O or VOS word orders are generally available (Rose 1981, Whistler 1985, Jacobsen 1993). (33)  a.  b.  kuuwi+it?is cakup kuuwi+-mit-riis cakup steal-PST-3.IND man A man stole a car. (VSO)  huupuukwas huupuuk as car w  kuuwi-Ht?is huupuukwas kuuwi+-mit-?iis huupuuk as steal-PST-3.lND car A man stole a car. (VOS) w  cakup cakup man  There is evidence that post-predicative word order is not strictly free, and that it is linked to constraints on animacy and discourse (Rose 1981, Woo 2004). As Rose (1981: 179) maintains, alternate word orders in Nuu-chah-nulth are "a function of the communicative salience of the constituents, eg. newness of information, definiteness, particularity, contrastiveness, and role in the discourse." Woo (2004) generalises that VOS in Nuu-chah-nulth most readily obtains when the object is 'unmarked' - inanimate and/or indefinite. Consonant with Woo's generalisation, the primary Ahousaht consultants for this dissertation show a strong preference for the subject to precede the object when the object is animate. These speakers generally disprefer VOS with an animate object, as indicated in the sentence below, in which word order serves to restrict the possible interpretations. (34)  10  ?u?uuyuk?is Ken Kay ?u-Vukr+Rl-?iis Ken Kay 0-cry.for-3.lND Ken Kay = (i) Ken is crying for Kay. (VSO) ^ (ii) Kay is crying for Ken. (VOS)  As Davis and Wojdak (to appear) note, however, this should make no difference to the argument against non-configurationality. A topic worthy of future research is inter-speaker (or inter-dialectal) variation in word order. As a whole, factors influencing word order in this language have not been adequately investigated to date. 1 0  84  In (34), the nominal Ken must precede Kay in order for Ken to interpreted as a subject. This may be contrasted with the case in (33), in which the inanimate object huupuuk' as "car" has the w  option of preceding or following the subject. Rose (1981: 112) proposes that an object may precede a subject when the object is new and salient, as she illustrates with the following Kyuquot examples. (35)  a.  ?uk irHs qa?uuc Mary ?u-cii+-?iis qa?uuc Mary 0-make-3.IND basket Mary Mary is making a B A S K E T . (VOS)  (Kyuquot; Rose 1981: 113, ex. 373)  b.  hisimyuup maacq in his-qim+-'uup-0 maacq in all-around-CAUS-3.ABS fly  +uucma 4-uucma woman  The lady collected H O U S E F L I E S . (VOS)  (Kyuquot; Rose 1981: 113, ex. 375)  w  w  w  However, a flavour of optionality must be recognized, as she notes that an order of V S O is also possible in this discourse context. The presence of the determiner -?ii is correlated with word order permutations in the Ahousaht dialect of Nuu-chah-nulth. If an object is marked with this determiner, then the object preferentially appears after the subject. This is illustrated in (36-37). In (36), the determinermarked nominal maamaati-?i "the bird" must follow piispis "cat" in order for maamaati-?i "the bird" to be interpretable as the object. Similarly, in (37), the determiner-marked nominal finiik-Ti "the dog" must follow cakup "man" if it is to be interpreted as an object. (36)  a.  b.  (37)  a.  b.  ?u?iicit?is piispis maamaati?i Tu-'iic-mit-Tiis piispis maamaati-?ii 0-consume-PST-3.lND cat bird-DET A cat is eating the bird. (VSO) # ?u?iicit?is maamaati?i piispis ?u-'iic-mit-?iis maamaati-Tii piispis 0 - consume-PST-3.IND bird-DET cat = (i) The bird is eating a cat. (VSO) 1- (ii) A cat is eating the bird. (VOS) hisqin9e?is cakup finiifcPi Tuuhwa-r his-qin5c-?iis cakup Tiniifc-Tii ?u-hwa+[+L] hit-on.head-3.IND man dog-DET 0-use A man hit the dog on the head using a hatchet. (VSO- O B L )  hisyak hisyak hatchet  ! hisqin^Tis finiifcri cakup Tuuhwa-r his-qin?e-?iis Tiniifc-Pii cakup ?u-hwa+[+L] hit-on.head-3.IND dog-DET man 0-use = (i) The dog hit a man on the head using a hatchet. (VSO- OBL) ^ (ii) A man hit the dog on the head using a hatchet. (VOS- OBL)  hisyak hisyak hatchet  A variety of analyses has been proposed to account for the derivation of the world's predicate-initial systems. Although these analyses are constrained by theory-internal restrictions  85  on admissable clause structure (see, for example, Anderson 1984, Lee 2000), there is mounting cross-linguistic evidence that there are multiple 'routes' available to achieving predicate-initial word order (Chung 1998, Davis 2005). That is, predicate-initial status amounts to a superficial characteristic which masks potentially distinct underlying syntactic systems.  On one hand,  predicate-initial systems may be directly linearized as VOS," under a right-linearized specifier analysis (Chung 1991). This is represented schematically below, abstracting away from node labelling. (38) 'Basic' predicate-initial word order  S V  O  With this line of analysis, V S O is also possible as a derived word order, surfacing after movement of one or the other arguments, as has been argued for mixed V O S / V S O systems including Mayan (England 1991), Austronesian (Chung 1991, 1998, 2004) and Salish (Davis 2005). On the other hand, predicate-initial systems may be derived from an underlying SVO configuration.  Within this option, two general possibilities are to be distinguished: V-raising  targeting the predicate head, as in McCloskey's (1991) treatment of V S O in tensed clauses in Irish; or VP-raising which targets the predicate at the level of the phrase, as has been argued for Malagasy (Pearson 1998), Niuean (Massam 2000), and Zapotec (Lee 2000). In the latter case, VOS straightforwardly obtains. However, V S O may be derived in conditions of V P remnantraising (Lee 2000), in which the object evacuates the V P before the V P 'remnant' is raised. The diagrams in (33) illustrate head-raising and phrase-raising, respectively. (39) Derived predicate-initial word orders a. verb-raising  b. raising of verbal phrase  I adopt an analysis of Nuu-chah-nulth predicate-initial word order as a 'basic' VOS system. In what follows, I will describe this proposal in more detail, followed by a discussion of evidence against alternative raising approaches. In §3.3.1, a right-branching specifier analysis of subjects is proposed, while §3.3.2 extends this right-branching specifier analysis to possessive phrases. Evidence against a verb-raising analysis of Nuu-chah-nulth predicate-initial word is  86  presented in §3.3.3, and discussion of problems for a verb phrase- raising treatment is given in §3.3.4. The analysis of predicate-initial word order concludes in §3.3.5 with a summary of the implications of variable word order for the linearization of affixal predicates.  3.3.1 Proposal: 'Basic' predicate-initial order (VOS) According to the analysis proposed here, predicate-initial word-order in Nuu-chah-nulth is not derived by syntactic movement. That is, no syntactic movement operation is necessary in order for the verb to precede the arguments of the clause. This is achieved via implementation of a right-branching specifier system, as in (40a), rather than a left-branching one, as in (40b). These systems represent two distinct linearizations schemes for identical syntactic objects. (40)  a. right-branching specifier  b. left-branching specifier  XP  XP  Comp Right-branching specifiers are unavailable under the model of syntax proposed by Kayne (1994). However, according to the directionality convention hypothesis, right- and left- branching systems are equally available across languages as distinct post-syntactic linearization schemes for universally unlinearized syntactic objects. By hypothesis, a directionality convention at spell-out in Nuu-chah-nulth determines that specifiers follow, rather than precede, the inner layers of the projection (the head and complement). Based on this right-branching specifier analysis, I propose the following representation of Nuu-chah-nulth predicates relative to their arguments.  In (41), I adopt the analysis that a  verbal head v introduces a subject, and takes a V P as complement.  According to this  arrangement, arguments appear on the right-periphery of morphemes which head the verbal projection(s) of the clause: (41)  vP DPsubject  DPobject  This yields an underived V O S word order. This linearization is consistent with the following generalisation of Sapir (1924: 83): "verb, object, subject- this is the most common Nootka order".  87  As indicated at the outset, however, VSO word orders compete with VOS in Nuu-chahnulth, generating controversy as to which, if either, of these word orders can be construed as more 'basic' (Woo 2004). Indeed, Jacobsen (1993) counters Sapir's assertion by showing that V S O is in fact statistically more common than VOS in the text examined by Sapir (1924), and furthermore, that sentences with two overt arguments are rare in text contexts. I follow England (1991) in maintaining that frequency of natural occurrence is not a determining criterion in the analysis of underlying word order (see also Woo 2004), as syntactic and non-syntactic factors conspire on a language-by-language basis to determine the specific word order permutations which arise. Definiteness effects, animacy restrictions, and topic/focus are all observed to affect the available word orders of the daughter languages of Proto-Mayan (England 1991). Strictly 'stylistic' factors such as prosodic heaviness have also been argued to play a role in the argument order in predicate-initial Salish systems (Davis 2005). There is inadequate evidence at present to allow for a conclusive characterisation of which mix of factors are at play in restricting the word order permutations of Nuu-chah-nulth.  However, existing research is consistent with the  hypothesis that argument order is influenced by information structure. Adopting England's (1991) analysis of the mixed V O S / V S O systems of Mayan, I therefore propose that V S O in Nuu-chah-nulth may plausibly be derived from a 'basic' VOS order when a 'reordering' rule moves a marked object to the right-periphery. (42)  [ V _ S]  'reordered'O  (England 1991:480)  This falls in line with claims that 'marked' animate and/or definite objects in Nuu-chah-nulth often follow a subject (Rose 1981, Woo 2004). The 'reordering' of the object can be represented by movement of the object to a specifier position above the subject, as in (43). (43)  XP  The X P projection can be understood to be a privileged position for animate and/or definite objects. For example, in the sentence in (44a), the determiner-marked object maamaati-Ti "birdD E T " necessarily raises above the subject piispis "cat". indicated by (44b).  88  A V O S ordering is unavailable, as  (44)  a.  b.  ?u?iicit?is Tu-'iic-mit-Tiis 0-consume-pst-3 ,ind A cat is eating the bird.  piispis piispis cat (VSO)  maamaati?i maamaati-?ii bird-DET  # ?u?iicit?is maamaati?i ?u-'iic-mit-?iis maamaati-?ii 0-consume-pst-3.ind bird-DET = (i) The bird is eating a cat. (VSO) ± (ii) A cat is eating the bird. (VOS)  piispis piispis cat  3.3.2 Evidence for right-branching specifiers A right-branching specifier system elegantly captures the canonical word order of Nuu-chahnulth possessed phrases. As Ravinski (2005) describes, the unmarked word order for possessed noun phrases when the possessum is adjectivally modified is adjective- possessum-possessor. (45)  naatsiici&itwaTis naatsii-si?e-mit-wa?is see-PERF-PST-3.QUOT Christine saw [Rachel's  Christine Christine Christine new house].  [cusukuk?i maht'ii Rachel] [cus-uk-uk-?i maht'ii Rachel] new-DUR-POSS-3.PS house Rachel (Ravinski 2005: 41, ex 114c)  The availability of this word order is straight-forwardly predicted by an analysis in which the possessor (Rachel) occupies a specifier position to the right of the possessed nominal. Following Ravinski (2005), this may be represented by a structure in which a possessor occurs as an internal argument of the NP, as in (46)." (46) Right-branching possessor AgrP Agr -7i  PossP  Rachel  Following Ravinski's (2005) analysis, the possessive morpheme -uk (POSS) is shown to head the Possessive Phrase (PossP). A n Agreement Phrase (AgrP) is postulated, which is headed by an agreement marker registering the possessor. In (46), the possessor agreement is third person -?i (3.PS), to match the third person possessor Rachel.  This diagram abstracts away from syntactic locus of the aspectual marking -uk ( D U R ) on the adjective. The aspectual properties of Nuu-chah-nulth adjectives requires additional research.  89  If a left-branching analysis of possessors is assumed, then the possessor is predicted to precede, not follow, the noun possessum (see Braithwaite 2003 for discussion).  This is  illustrated in the tree in (47). (47) Left-branching possessor AgrP Agr  PossP Poss -uk  NP AP NP cusuk new Rachel  N mahtii house  Multiple derivations would be required to derive a possessor-final word order, because otherwise the order adjective-possessor-possessum would be anticipated (Ravinski 2005). These multiple derivations are complicated by the placement of the inflectional clitics within the phrase (Braithwaite 2003).  12  As Ravinski (2005) notes, a right-branching specifier analysis is a  straightforward alternative to these complicated derivations.  3.3.3 Evidence against verb-raising Previous treatments of Nuu-chah-nulth word order have analysed predicate-initial word order as arising from head-movement of the verb (Davis and Sawai 2001, Stonham 2004, among others). These analyses assume that head movement applies to an underlying SVO structure, raising the head of the predicate to a position past the subject, where it syntactically adjoins to tense and/or agreement occupying higher functional projection (eg. TP, Mood). This may be represented by the following:  I assume that the 'second position' behaviour of inflectional clitics is determined through local spell-out (§3.4.1).  90  (48)  In this section, I illustrate empirical problems with this analysis as applied to Nuu-chah-nulth. The first hurdle confronting a head-raising analysis is the question of evidence for verb movement to a functional projection above the vP. Since inflectional morphemes such as tense and subject agreement are 'second position' enclitics in Southern Wakashan (Klokeid 1978, Davidson 2002), their appearance suffixed to the verb is phonologically conditioned, and as such, does not necessarily entail a syntactic mode of placement (see §3.4.2 for a local spell-out analysis of this cliticization pattern). The examples below illustrate the 'second position' effect in which tense and subject agreement encliticize to the first word of the sentence, whether it is the predicate itself (49a), a preceding adverbial (49b), or a negation particle (49c). (49)  a.  waa+si?e/£s wa+-[+L]-s\ k-inJt-s !  gO.home-CONT-PERF-PST-1 SG. ABS  I was in the process of going home. b.  wityax/fc wityax-/22/f-.s  waa-rsifc wa+-[+L]-sifc  slowly-PST-1 SG. A B S  gO.home-CONT-PERF  I was slow in going home. c.  wik/te mk-mit-s  wityax wityax  waa-rsifc wa+-[+L]-si&  NEG-PST-lSG. ABS  slowly  gO.home-CONT-PERF  I wasn't slow in going home. The fact that a verb in Nuu-chah-nulth may bear tense and/or agreement morphology does not therefore constitute evidence for syntactic raising of the verb to these functional projections. If a verb is the first word in the predicate phrase, it will bear tense and/or agreement morphology; if it is not first, then it will not. Moreover, there is the broader issue of a trigger for the putative verb-raising. Based on the observation that V S O word order obtains in Modern Irish in tensed clauses, while S V O occurs in infinitival ones, McCloskey (1991) argues that Irish [-(-finite] verbs are attracted to an  91  inflectional projection. In Nuu-chah-nulth, however, no parallel argument can be constructed.  13  In Nuu-chah-nulth, predicate-initial word order is possible in small clause environments. On the assumption that small clauses lack functional projections above the vP which could house a raised verb, predicate-initial word order in this environment should be impossible. This verbinitial pattern is illustrated for the non-finite complements of the perception verb Puu-nakuuh "observe". (In Chapter 5,1 provide evidence that the complement here is equivalent to a vP.) (50)  Tuuna^uuhitsis [wa?ic'as haa ?u-nakuuh[+L]-mit-siis [wa?ic-'as haa 0-observe-PST-l SG.IND sleep-on.groundDEIC I observed that man sleeping on the ground.  cakupri] cakup-?ii] man-DET  The same generalization may be made of non-finite complements of negation (51a), as well as in contexts with auxiliaries (51b) and non-affixal modals (51c). Each of these environments allow the predicate to precede the subject, in the absence of an overt syntactic trigger (such as tense or finiteness).'  4  In the negation context in (51a), the predicate waPic "sleep" precedes the subject,  Ken. The example in (51b) shows a relative clause which is formed when the relative pronoun yaq (REL) is suffixed by the auxiliary Puu-k ii- (AUX): what follows this auxiliary is the verb w  maakuk "buy", crucially preceding the subject of the relative clause, cakup-Pi "the man". In (51c), the verb wa-J-sik "go home (PERF)" is sandwiched between the modal Tapaak "willing" and the subject Kay. (51)  a.  wikifPis hacuk wa?ic Ken wik-mit-riis hacuk wa?ic Ken NEG-PST-3.IND deeply sleep Ken Ken wasn't in a deep sleep.  (negation)  Davis and Sawai (2001: 125) argue based on the behaviour of the perception verb naatsiicik"\a see" that SVO word order occurs in non-finite complements. However, given that naatsiicik "to see" is compatible with nominal complements (eg. naatsiicikit?is Mary John "Mary saw John"), combined with the availability of null pronouns and null "absolutive" third person subject agreement in Nuu-chah-nulth, it is unclear whether their test sentence involves nominal complementation (as indicated by the bracketing below) or clausal complementation. Note that the third person inflection (-hu/d0) in the second clause is optionally overt. (i) [naatsiicikifPis Mary John] k ik ixasi&(huk) pro Wanda [naatsii-si?c-mit-?is Mary John] k ik ixas-si&(-huk) pro Wanda see-PERF-PST-3.lND Mary John kiss.on.cheek-PERF-(3.DEP) pro Wanda Mary saw John, (he was) kissing Wanda on the cheek, (cf. Davis and Sawai 2001: 125, ex. 5) In order to control for ambiguities in clausal composition (see Jacobsen 1993 for discussion), it is necessary to perform this test with a matrix perception verb (such as Puu-jiakuuA "observe") which is incompatible with nominal complementation (see example (50)). An alternative word order is also possible where the verb phrase follows the subject. The availability of this word order requires further research, as do other cases of word order variability in the language (see Rose 1981, Davidson 2002). What is crucial for the present argument against verb-raising is that the predicate-initial word should not be possible in this environment. 13  14  92  w  w  w  w  b.  9eiihum+?is &iih-um+-riis  suwis [yaaq i-Htii suwis [yaq-ci-i-[+L]-mit-ii red-RD-3SG.IND shoes R E L - A U X - P S T - 3 S G . I R L The shoes the man bought are red.  c.  TapaakPis wa+si?e Kay Yapaak-?iis wa+-sifc Kay willing-3.IND go.home-PERF Kay Kay is willing to go home.  w  maakuk cakup?i] maakuk cakup-?ii] buy man-DET (auxiliary environment)  (modal environment)  What these environments share is a word order in which the verb precedes the subject. However, in the negation context in (51a), the predicate phrase is irrealis, and so the trigger for the pututive raising cannot therefore be finiteness. Similar arguments can be constructed for (51b-c): why would the presence of the preverbal auxiliaries not preclude the need for verb itself to raise? Complex nominal predicates also constitute a considerable challenge to a head-raising analysis of predicate-initial word order.  15  Since head movement can apply to only a single head  in the predicate phrase, this operation is predicted to 'break up' a complex predicate composed of a predicate and its modifier, by forcing the subject to intervene between the raised element and unraised residue left in the V P . In fact, contrary to expectations, the standard pattern is for the subject to follow a complex nominal in Nuu-chah-nulth (Davidson 2002: 128).  16  In such  contexts, the 'fronted' element resembles a maximal projection, not a head. In (52), the nominal predicate kui-hiucma "good woman" precedes the subject Kay. (52) complex nominal predicate (subject-peripheral) [9cu+?iis +uucma] [9cu+-?iis -hiucma] good-3.lND woman Kay is a good woman.  Kay Kay Kay  In contrast, for an underived predicate-initial system, this subject-peripheral word order follows naturally, as the nominal occupies initial position in the clause, together with its modifer. This word order is represented in the following structure, adopted from Ravinski's (2005) analysis of nominal predicates.  17  In (53), the nominal head n introduces the subject (Kay) of the nominal  predicate.  Nuu-chah-nulth allows any of the lexical categories (A, V, N) to occur directly as predicates in clauseinitial position; there is no copula in the language (Wojdak 2000, 2001). Davidson (2002: 128) notes that "the words in a multi-word nominal predicate... are usually strictly ordered: quantifier/number > property > noun." Subject-intervening patterns, though marked, are attested in my own fieldwork, however. Additional research is required into this pattern, as with other instances of word order variability in Nuu-chah-nulth. 1 assume that encliticization of the subject agreement marker -?iis to the adjective is achieved through local spell-out (§3.4.1) 16  17  93  (53) complex nominal predicates in right-branching specifier system !  Kay n  NP A ku-fgood  N ivucma woman  In conclusion, I suggest that a plausible solution to the problems posed by a head-raising account - lack of evidence for raising, lack of a syntactic trigger for raising, unpredicted word orders - is to assume that Nuu-chah-nulth clause structure is V O S , with rightward movement of the object deriving the V S O variant.  3.3.4 Evidence against raising of verbal phrase The issues noted above with respect to head-raising also create complications for a predicateraising analysis of Nuu-chah-nulth. On the topic of word order variability, any analysis of Nuuchah-nulth clausal structure will need to employ special machinery to account for the VOS/VSO alternations and other word order permutations in the language. A potential advantage of a V P raising analysis over a head-raising approach, however, is that it has been demonstrated independently that it can successfully deal with VOS/VSO alternations (Massam 2000, 2001). In the predicate-initial system of Niuean, post-predicative word order is not strictly flexible, as it is tied to a definiteness effect on the object. In Massam's VP-fronting analysis of Niuean, V S O word order is obtained when a definite object vacates the V P before the remnant of the phrase raises. Indefinite objects which are 'pseudo-incorporated' into the V remain as part of the V P complex, generating VOS order when the V P raises. The distinct derivations of VOS and V S O orders is represented below. (54)  a. VP-raising (= VOS) TP  T  b. VP-remnant raising (= VSO) TP  T  vP  94  vP  The VP-remnant raising derivation is essentially a 'two-step' process in which the object raises on its own first, before the rest of the V P undergoes movement. It is apparent, however, that the analysis which Massam applies to Niuean cannot be directly translated to Nuu-chah-nulth. As Woo (2004) observes, this analysis makes the crucial prediction that existential clauses must have VOS order, since the indefinite object has no motivated escape route out of the VP.  Under this view, the indefinite object of an existential  clause should be 'pseudo-incorporated' into the V , and should never raise outside of the V P . This runs counter to the observation that V S O word orders are the preferred pattern for locative existentials in Nuu-chah-nulth (see Wojdak and Woo 2004). In (55a), the object haPum "food" follows the subject niisyak-Pi"the pot"; in (55b), the indefinite object ciixsac "frying pan" comes after the subject camaqkyak-?i'"the oven". (55)  a.  ?ucuu?is 7u-cuu-?iis 0-in.container-3.IND There's food in the pot.  b.  TuuqWiis ?u-'aqfc-?iis 0-inside-3.IND There's a frying  18  niisyak?i niisyak-?ii pot-DET (VSO)  ha?um ha?um food  camaqfcyak?i ciixsac c'amaqfcyak-?ii ciixsac oven-DET frying.pan pan in the oven. (VSO)  An indefinite object is incorrectly predicted to follow the verb for these existentials.  19  In sum, given the challenges faced by derivational mechanisms for generating predicateinitial word order in Nuu-chah-nulth, I present a 'basic' V O S configuration as a plausible alternative for this language. The factors which govern post-predicative word order variability in the language await clarification by future research. In the next section, I examine the implications of post-predicative word order variability for the linearization of affixal predicates.  3.3.5 Implications for the linearization of affixal predicates According to the local spell-out hypothesis, an affixal predicate finds a host chosen from its derivational sister.  In §3.3.5.1, I illustrate how this has the implication of creating a  'complement' effect in incorporation, whereby only elements from the complement of an affixal predicate are eligible as hosts.  I then argue that the local spell-out analysis is superior to  The indefinite locatum argument (haPum "food", ciixsac "frying pan") in these locative existentials can be clearly shown to be objects of the predicate (§4.4.1). A topic for future research is the surface position of the definite subjects in (55). As proposed in §3.3.1, determiner-marked objects are moved rightward in Nuu-chah-nulth. It is not known whether determinermarked subjects ever show a similar preference in Nuu-chah-nulth. 1 9  95  alternative models of linearization which select a host for the affixal predicate via constraints on directionality. In §3.3.5.2, it is shown that direction-sensitive mechanisms for affixations have difficulty coping with the variable post-predicative word order of Nuu-chah-nulth.  3.3.5.1 The 'complement' effect  /  According to the analysis I have proposed, affixal predicates are spelled-out in a minimal domain containing only the affix and its derivational sister. A consequence of spell-out is linearization. Thus, an affixal predicate is linearized with respect to its derivational sister, inducing affixation of the predicate to a host from its derivational sister.  Elements which are introduced at later  stages of the derivation, in higher projections, are not available as potential hosts for the affix because they are not present at the time that the affix is spelled-out. Thus, under the assumption that objects, but not subjects, occur as complements to a predicate, the local spell-out hypothesis predicts that only objects of an affixal predicate are eligible as hosts. This is illustrated in (56), in which an object acts as the derivational sister to the verb, while the subject is introduced in a higher projection. The verb is spelled-out with the object, its complement. (56) The 'complement' effect in Nuu-chah-nulth affixation local spell-out  For verbs which take nominal complements, there is robust evidence for an affixation asymmetry between the subjects and objects of affixal predicates (Woo 2000, Davis and Sawai 2001, Stonham 2004, among others).  (This is argued in detail in Chapter 4.) This effect is  illustrated in (57) for the affixal predicate ?u-?aap "buy". The object of this verb, mahta- "house" may serve as host for the affix, but the subject capx- "man" cannot. (57)  a.  b.  mahta?amif?is cakup mahta-'aap-mit-Tiis cakup house-buy-PST-3 .IND man A man bought a house. *  capx?aamit?is maht'ii capx-'aap-mit-Tiis maht'ii man-buy-PST- 3. IND house A man bought a house.  This complement effect is also observed with affixal predicates which take verbal complements, such as Puu-nalcuuh "observe".  The syntactic structure for an affixal predicate  which takes a verbal (sentential) complement may be represented abstractly as in (58). (A detailed analysis of the structure of such predicates is presented in Chapter 5.)  96  (58)  local spell-out  The affixal predicate occupies matrix position as V , , and takes the sentential complement (VOS) as its complement (circled). According to the local spell-out hypothesis, V i is linearized with respect to this complement, since it is its derivational sister.  The subject of V , (i.e. SO falls  outside of this local spell-out domain, predicting an asymmetry between S| and the complement. Evidence for such an asymmetry is given in (59). In (59a), the embedded verb tuuxtuux a "jump w  (ITER)"  hosts the affixal predicate Tuu-nak'uuh "observe". As shown in (59b), the subject of the  affixal predicate, capx- "man", is ineligible as host. (59)  a.  tuuxtuux anakuuhit?is cakup t'aatna?is tux -a[+R]-nakuuh-mit-?iis cakup taatna?is jump-ITER-observe-PST-3.lND man children A man observed the children jumping. w  w  b.  * capxnalcuuhifPis tuuxtuux a capx-nakuuh-mit-?iis tux -a[+R] man-observe-PST-3.IND jump-lTER A man observed the children jumping. w  w  taatna?is taatna?is children  This asymmetry follows if capx- "man" is not a derivational sister of the affix.  By the  'complement' effect in affixation, only derivational sisters of affixal predicates are able to act as hosts. The following section presents an argument against an alternative analysis of the observed affixation asymmetry.  3.3.5.2 Evidence against directionality of affixation This section argues that sensitivity to derivational sisterhood is not reducible to an independent effect of directionality. Let us label this alternative analysis the RlGHT-directionality hypothesis. This hypothesis is defined by the following statement: (60) RlGHT-directionality hypothesis: An affixal predicate attaches to whatever host is found to its right Given a syntactic structure as in (61), this rule would determine that an affix - a choses P as its host because P is right-adjacent to -oc. The element 8, in contrast, would be ineligible as a host for - a because it occurs to the left of -a.  97  (61)  -a  ^  P  R-adjacent  1  Under such an analysis, it is irrelevant that p is the derivational sister of -ex. A l l that matters according to the  RIGHT-directionality  Evidence against the  hypothesis is that |3 follows -a.  RIGHT-directionality  hypothesis comes from the post-predicative  word order possibilities of Nuu-chah-nulth. As noted in the earlier discussion, in Nuu-chahnulth, there is often variability in the word order of an object with respect to an overt subject. This variability is illustrated in (62) for the affixal predicate Pu-Paap "buy". In (62a), the object mahtii "house" precedes the subject cakup "man". In (62b), the ordering of the arguments is reversed. (62)  a.  ?u?aamit?is mahtii cakup 7u-'aap-mit-?iis mahtii cakup 0-buy.-PST-3.IND house man A man bought a house.  b.  ?u?aamit?is cakup mahtii 7u-'aap-mit-?iis cakup mahtii 0-buy-PST-3.lND man house A man bought a house.  Thus, either an object or a subject may follow the affixal predicate. Despite this flexibility in word order of subject and object, the affixation mechanism is invariant: an affixal predicate may only attach to mahta- "house" and not to capx-"man". (63)  a.  mahta?amit?is cakup mahta- 'aap-mit-?iis cakup house-buy-PST-3 .IND man A man bought a house.  b.  * capx?aamif?is mahtii . capx-'aap-mit-?iis mahtii man-buy.-PST-3.IND house A man bought a house.  This finding is at odds with the predictions of the  RIGHT-directionality  hypothesis. According to  this hypothesis, if an element can occur right-adjacent to the affix, it should be eligible as a host. Thus, a subject which precedes an object, such as cakup "man" in (62b), is incorrectly predicted to act as a host. Unlike with the  RIGHT-directionality  hypothesis, for the local spell-out hypothesis, the  host for an affixal predicate is determined by derivational sisterhood. Local spell-out matches an  98  affixal predicate and a host at an initial stage of the derivation: the point at which an affixal predicate is merged into the derivation. According to this analysis, the surface position of the arguments of the affixal predicate is irrelevant: affixation is determined before the relative ordering of the arguments is manipulated. Thus, the local spell-out hypothesis elegantly captures the generalisation that the affixation pattern is invariant, despite variability in post-predicative word order. In the next section, I demonstrate how the local spell-out hypothesis can be extended from affixal predicates to other affixes found in Nuu-chah-nulth.  3.4 Cliticization domains There are two distinct cliticization domains in Nuu-chah-nulth. Clitic strings may be built up within a DP, or at a clausal level which excludes the DP(s). The bracketing in (64) illustrates these two zones of cliticization. (64)  hiixtaqci[/J7/to5] p C  d  o  m  a  i  huupuul^as[&iA#/l] p domain  n  D  hiixtaq-cip-22J7]f-5H»  huupuul^as-uk-Pi'/tk  have.accident-BEN-PST-1 SG.IND car-POSS-2SG.PS  I had an accident with your car. In §3.4.1 and §3.4.2,1 consider each of these cliticization domains in turn.  3.4.1 DP domain Within a Nuu-chah-nulth DP, functional morphemes appear in a strictly ordered clitic string suffixed to a root (Davidson 2002, Werle 2002, Ravinski 2005). (65) Organization of the DP clitic sequence in Nuu-chah-nulth = POS S = T E N S E = A G R / D E T  This clitic string includes the following enclitics: possessive markers, tense, possessive agreement and the determiner -Pn.  20  These morphemes appear in the following examples,  suffixed to a nominal (in brackets). (66)  a.  naatsiici&itsis naatsii-sik-mit-siis see-PERF-PST-1 SG.IND  [maht'iimit?i] [maht'ii-mit-?ii] house-PST-DET  I saw the former house (that burnt to the ground).  (Ravinski 2005: 16, ex.29)  Nuu-chah-nulth permits tense markings in the nominal domain, as well as in the clausal domain. This phenomenon is not uncommon in the Pacific Northwest sprachbund (see Burton 1996 for discussion of the Salish language Halkomelem).  99  b.  &ihum+it?is &ih-um+-mit-?iis  [huupuuk asuk itqs] [huupuuk as-uk-mit-qs]  red-RD-PST-3.IND  car-POSS-PST-lSG.PS  w  w  w  M y former car was red. These inflectional morphemes encliticize to the leftmost root with a DP, illustrated below with the 'second position' placement of the enclitic determiner to haak aak w  In (67a), the determiner suffixes  "girl", while in (67b) it appears instead on the modifier q acai{aq) "(very) w  beautiful". In (68a), the determiner suffixes to pismis "problem(s)", and in (68b) it attaches to ?aya "many". (67)  a.  ?u?uk inkitsis ?u-k ink[+R]-mit-siis  [haak aa?c?i] [haak aa9e-?ii]  0-talk.with-PST-l SG.IND  girl-DET  w  w  w  w  I talked with the girl. b.  ?u?uk inkitsis [q aca+aq?i ?u-k ink[+R]-mit-siis [q acaF-aq[+S]-?ii 0-talk.with-PST-l SG.IND beautiful-AUG-DET I talked with the beautiful girl. w  w  w  (68)  a.  w  ?uucwa?is ?u-ic-wa?is  Kay Kay  0-own-3.QUOT Kay  q acii+ q a-cii+  [pismis?i] [pis-mis-?ii]  like-make  bad-NOM-DET  w  w  haak aafc] haak aak] girl w  w  Kay's the instigator of the problems. b.  ?uucwa?is Kay 7u-ic-wa?is Kay 0-own-3.QUOT Kay Kay's the instigator of  q acii+ [?aya?i q a-cii+ [?aya-?ii like-make many-DET the many problems. w  w  pismis] pis-mis] bad-NOM  As can be seen in these examples, when the DP contains only a nominal, the determiner encliticizes to this word; however, when a modifier or quantifier takes on leftmost position in the DP, the placement consistently shifts to this leftmost element (Davidson 2002). This 'affix hopping' behaviour is ably handled by the local spell-out analysis. Let us take the positioning of the determiner in the examples in (67) as illustration of the spell-out properties of DP-level clitics. When the determiner is syntactically merged with a noun, such as haak aak "girl" in (67a), local spell-out determines that the determiner and the noun must be w  linearized with respect to each other at spell-out to PF. (69)  DP D -?ii  N haak aak girl w  Because the determiner is a suffix, a linearization of <haak aa9c-?b> is induced. w  100  The placement of the determiner suffixed to the modifier in (67b) proceeds much the same way, although there is additional round of spell-out when the noun is first merged with the modifier q^ca+faq) "(very) beautiful". Before the determiner may be merged with the noun phrase, the following syntactic operation takes place: Merge (q aca+aq, haak aak). This builds w  w  a modified noun. (The category label of the noun is projected, as argued in Chapter 2). local spell-out A q aca+(aq) beautiful w  N haak aa& girl w  As derivational sisters, the adjective and noun are linearized with respect to each other at spellout.  A directionality convention establishes the modifier-initial pattern of Nuu-chah-nulth,  setting the stage for a spell-out ordering of <q acai{aq), haak aak>. w  w  In the next step of the syntactic derivation for (67b), the determiner is introduced by Merge. The determiner thus takes the NP as its derivational sister. (71)  DP  <  local spell-out local spell-out  A q aca+(aq) beautiful w  N haak aa& girl w  At spell-out, the determiner will need to be linearized with respect to its derivational sister. In particular, because the determiner is an affix, it must find a host from within its derivational sister. The earlier round of local spell-out established an ordering of <q aca+(aq), haak aak>. w  w  By the string adjacency effect, the host for the determiner -?iiis selected as the left-most element of the NP: q acai-(aq) "(very) beautiful". w  Thus, an ordering of  <q aca+(aq)-?i, haak aak> w  w  results at spell-out. The 'affix-hopping' behaviour of the enclitic determiner can therefore be seen to be an interaction between the string adjacency effect and the syntactic composition of the derivational sister of the determiner.  If the derivational sister is simplex, as in (53a), then the locality  restriction on affixation is trivial: the determiner must be spelled-out with the single element in its derivational sister. If the derivational sister of the determiner is complex, as in (53b), then the locality restriction determines that the single leftmost element in the derivational sister is selected as a host.  Note that the determiner has not actually 'hopped': in both types of cases, the  determiner consistently selects as a host the single element which is linearly adjacent to it at spell-out.  101  3.4.2 CP domain Functional morphemes outside of the DP are also subject to encliticization (see Klokeid 1978 for discussion of the Southern Wakashan language Ditidaht). Davidson (2002) identifies a range of inflectional morphemes which occur in a strictly ordered clitic sequence (simplified from Davidson 2002: 321): (72) Organization of the CP clitic sequence in Nuu-chah-nulth =TR=TEMP=PAS=POSS=TENSE=AGR/MOOD=PL=AGAIN=HAB  Parallel to cases of DP-level cliticization, the members of this sequence occur standardly in 'second position' relative to a host morpheme at the left edge of the clause.  This 'second  position' effect is exemplified in (73) with the positioning of the past tense morpheme - / n / f (PST) and the third person indicative subject agreement -?iis (3.IND). In (73a), these morphemes suffix to the verb kamatq-uk "run ( D U R ) " . In (73b), however, their position 'shifts' to the preverbal modifier (73)  "fast". kamatquk it?is kamatq-uk-mit-?iis w  a.  run-DUR-PST-3.IND  Florence Florence Florence  Florence was running. b.  &a?ixit?is &a?ix-mit-?iis  kamatquk kamatq-uk  fast-PST-3.IND  run-DUR  Florence Florence Florence  Florence was running fast. This suffixation pattern follows from the local spell-out hypothesis. We can consider the simpler case in (73a) first. Successive applications of Merge build the tree shown in (74), in which the tense and subject agreement morphemes occupy functional projections (Tense Phrase and Mood Phrase) above the lexical projections of the verb kamatq(uk) "run". Note that the subject, Florence, is represented as a right-linearized specifier. (74)  AgrP  v 0  Florence V kamatq(uk) run  Each application of Merge is subject to local spell-out for the derivational sisters conjoined by Merge. Early rounds of local spell-out determine that the subject Florence is linearized to the  102  right because of a right-branching specifier convention. When the past tense morpheme is added to the tree by Merge (T, vP), this has the result that -mit (PST) must be linearized with respect to its derivational sister <kamatq(uk), Florence> at spell-out. Because -mit (PST) is a affix, it must find a host from within its derivational sister.  The string adjacency effect determines that  kamatq(uk) is selected as this morpheme's host, inducing a linearization of <kamatq(uk)-mit, Florence>. The next morpheme to be linearized in accordance with the string adjacency effect is the subject agreement morpheme -?iis (3.IND). As a suffix, it is tagged on at spell-out to the end of the leftmost element in its derivational sister. This yields the ordering <kamatq(uk)mit-?iis, Florence>. The tree in (75) is a representation of the derivation when the verbal predicate is modified by an adverbial, as in (73b). Here, the adverbial /4krfar"fast" combines with kamatq(uk) "run" to form a complex verbal predicate.  Florence  kamatquk run Once again, successive applications of Merge determine that spell-out relationships are formed incrementally between derivational sisters.  Starting with the lower portions of the tree, a  directionality convention requires that the modifier ka?ix "fast" linearly precede kamatq(uk) "run" at spell-out.  Similarly, the right-branching specifier convention entails that the subject  Florence will be spelled-out the right of the rest of the vP. When it comes time for the past tense morpheme -mit (PST) to find a host at spell-out, the item selected as its host is the leftmost element in its derivational sister.  Because the modifier linearly precedes the verb, it is the  modifier which is determined to be the host for -mit (PST).  This yields a linearization of  <fca?ix-mit, kamatq(uk) Florence>. A final act of linearization suffixes the subject agreement morpheme -?iis (3.IND) to the tail end of the initial complex, resulting in an ordering of <ka?ixmit-?iis, kamatq(uk) Florence>.  103  3.5 Conclusion This chapter has touched upon areas of Nuu-chah-nulth grammar which bear on the present analysis of affixal predicates. I have presented evidence for the configurationality of Nuu-chahnulth syntax, ^and have represented this clausal structure within a right-linearized specifier system. By this analysis, Nuu-chah-nulth predicate-initial word order originates with a 'basic' VOS system. I have identified two domains of cliticization in the language, linked to DP-level and clause-level inflection. Inflectional clitics in the language find their positions via the same spell-out principles responsible for the linearization of affixal predicates. The  empirical coverage for the remainder of this dissertation corresponds to the  'polysynthetic' realm canonically situated at the left-edge of a Nuu-chah-nulth clause. It is in this morphologically complex sequence that affixal predicates may commonly be found united with their hosts. (76)  a.  b.  YicpaT?in+?anitnis , fic-pa-fr-'m-fr-'at-mit-niis rotten-taste-serve-PAS-PST-1 PL.IND We were served rotten-tasting spawned  !< aqmis k aq-mis s.h.eggs-NOM herring eggs by w  w  huuhtaksiihmahsaris Lucy huhtak-sifc-'iih[+L1-mahsa-?iis Lucy know-PERF-try.to-want.to-3.lND Lucy Lucy wants to learn how to speak Nuu-chah-nulth.  Mary Mary Mary Mary. quuquu?aca quu?ac-[+R]-(y)a person-speak-CONT  In the following chapters, I will present additional evidence for the syntactic structures underlying these morphologically complex sequences. In Chapter 4, the argument structure of affixal predicates which take nominal complements will be discussed.  In Chapter 5, the  argument structure of affixal predicates which take verbal complements will be discussed. Residual to my analysis are the factor(s) governing the word order variations found outside of the clause-initial polysynthetic complex.  104  CHAPTER FOUR  Nominal complements of affixal predicates ...where every word is at home, taking its place to support the others... ~ T.S. Eliot Four Quartets no. 4  4.0 Introduction The predicates in Nuu-chah-nulth which permit incorporation are a lexically specified set of affixal predicates. As discussed in earlier chapters, 'incorporating' predicates in Nuu-chah-nulth are invariably bound.  In this, Nuu-chah-nulth differs from incorporation languages such as  Mohawk which do not have a lexically defined subclass of incorporating predicates. In Mohawk, a single predicate can show an alternation between an incorporating and a non-incorporating option (Baker 1988). In (la), the inflected predicate ye-nuhwe'-s "like" incorporates its object -nuhs "house", while in (lb) it does not. (1) Mohawk (examples from Postal 1962, as cited in Baker 1988: 81-82, ex. 14a-b) a.  Yao-wir-a'a  ye-nuhs-nuhwe'-s  PRE-baby-SUF 3FS/3N-house-like-ASP  The baby house-likes. b.  Yao-wir-a'a  ye-nuhwe'-s  ne  ka-nuhs-a'  PRE-baby-SUF  3FS/3N-like-ASP  DET  PRE-house-SUF  The baby likes the house. For predicates in Nuu-chah-nulth, however, such an alternation is banned outright. Independent (non-affixal) predicates such as maakuk "buy" never permit incorporation. The example in (2a) illustrates the impossibility of incorporating the nominal mahta- "house" into the independent predicate maakuk "buy".  The nominal must always occur separately from the independent  predicate, as in (2b). (2)  a.  * maht'amaakuk it?is cakup mahta-maalaik-mit-Tiis cakup w  house-buy-PST-3.IND  man  A man bought a house. b.  maakuk it?is maakuk-mit-?iis  cakup cakup  buy-PST-3.lND  man  w  mahtii mahtii house  A man bought a house.  105  Incorporation is an option exclusively reserved for affixal predicates in Nuu-chah-nulth, such as ?u-?aap "buy". In (3a), ?u-?aap "buy" incorporates a nominal host, mahta- "house". As shown in (3b), it is impossible for an affixal predicate such as ?u-?aap "buy" to occur without a host. (3)  a.  b.  maht'a?amif?is cakup mahta-'aap-mit-?iis cakup house-buy-PST-3 .IND man A man bought a house. * ?aamit?is cakup mahtii 'aap-rnit-?iis cakup mahtii buy.-PST-3.IND man house A man bought a house.  I argued in Chapter One that affixal and independent predicates in Nuu-chah-nulth are distinguished via specification of an [affix] requirement in the lexical entry of an affixal predicate. That is, affixal predicates constitute a lexically designated subclass of predicates. We turn now to the syntactic characteristics of affixal predicates. What is the syntactic make-up of this lexically defined subclass? In this chapter, the syntactic structure of affixal predicates which take nominal complements is investigated.  (Affixal predicates which take  verbal complements, such as -qaath "claim", are discussed in Chapter 5.) (4) Classes of predicates in Nuu-chah-nulth A . affixal predicates (i) nominal complements (ii) verbal complements B. independent predicates (i) nominal complements (ii) verbal complements The aim of this chapter is to develop an inventory of the argument structures which are available to affixal predicates which take nominal complements. Following Hale and Keyser (1993, 2 0 0 2 ) , I pursue a syntactic approach to argument structure in which positions for arguments are projected syntactically in accordance with the lexical properties of the head. Argument structure is what limits the number of arguments which exist for a given predicate.  For example, a  ditransitive verb such as ?u-yii "give" has a lexically licensed position for a benefactive argument, although a transitive predicate such as ?u-?aap "buy" does not. (5)  taanaqayimifPis cakup ?um?iiqsak taanaq-ayii-mit-?iis cakup ?um?iiqsu-?ak money-giye-PST-3.IND man mother-POSS A man gave money to his mother.  106  (6)  a.  * mahta?amit?is cakup ?um?iiqsak mahta-'aap-mit-?iis cakup ?um?iiqsu-?ak house-buy-PST-3 .IND man mother-POSS A man bought a house for his mother.  b.  mahta?apcip?is mahta-'aap-cip-?iis  cakup ?um?iiqsak cakup ?um?iiqsu-?ak  house-buy-BEN-3.IND  man  mother-POSS  A man bought a house for his mother. In (5), PumPiiqsak "his mother" acts as the recipient of the predicate Pu-yii "give". The example in (6a) shows that such an argument is not directly licensed by the predicate Pu-Paap "buy". Instead, in order for a benefactive argument to appear with Pu-Paap, the predicate must be supplemented by the addition of the benefactive suffix -dip ( B E N ) , as in (6b). This chapter will demonstrate that the syntactic configuration of arguments of an affixal predicate plays a deterministic role in the pattern of incorporation in Nuu-chah-nulth. Specifically, the syntax conditions local spell-out operations by determining which elements will act as the derivational sister to the affixal predicate. As first discussed in Chapter 3, a host for an affixal predicate (-a) is chosen from the complement (p) of an affixal predicate. In contrast, an element from the affixal predicate's specifier (7t) will be ineligible as a host because it is not a derivational sister of the affixal predicate.  As illustrated in the diagram below, the affixal  predicate (-00 is spelled-out with its derivational sister, the complement (p).  The host for the  affixal predicate is determined by this local spell-out domain: the specifier (7t) falls outside of this domain. This derives the 'complement' effect in affixation: (7) The 'complement' effect in affixation y > 8  local spell-out -a  7C p  The syntactic limitations which are imposed on Nuu-chah-nulth incorporation follow directly from the PF Incorporation hypothesis in which incorporation occurs post-syntactically. (8) syntax PF Incorporation (linearization)  LF  PF  107  According to the hypothesis, the linearization operation responsible for resolving the affixation requirement of an affixal predicate applies to the outputs of syntactic structure-building, at the point of spell-out to PF. Only an element which is generated in the syntax in a position where it may be spelledout with the affixal predicate is compatible with incorporation.  Complements occupy a  privileged position in that their syntactic sisterhood to the affixal predicate guarantees that the affixal predicate will be oriented with respect to the complement by local spell-out.  4.1 Predication configurations Before turning to my diagnostics for Nuu-chah-nulth argument structure,.! present an overview of the predication configurations which will be discussed over the course of this chapter.  4.1.1 Basic structures Argument structures of predicates are built by syntactic concatenation. The simplest possible argument structure for a predicate results from a single application of Merge. This one-place relation maps an argument to the complement position of the predicate. This is an unaccusative relation. (9) Unaccusative syntax VP V  DP  A second application of Merge introduces another DP. This second argument is realised as a specifier of the predicate. This two-place relation may be referred to as an extended unaccusative. (10) Extended unaccusative syntax VP  A central proposal of this chapter is that the argument structures available to affixal predicates reduce to these two basic configurations of arguments.  Both basic types map an internal  argument to complement position, although the types differ lexically in the possibility of projecting a specifier (Hale and Keyser 2002). A n unaccusative predicate is a monadic relation which maps its single internal argument to its complement position. A n extended unaccusative, in contrast, is an inherently dyadic relation which allows two internal arguments to occupy complement and specifier positions, respectively.  108  The affixal predicate Tu-nii "arrive" is an example of a Nuu-chah-nulth unaccusative predicate. In the following example, this monadic predicate takes a single argument, paastin?ath "American(s)". (lh)  v  ?unii?is ru-nii-?iis 0-arrive-3.IND Americans came.  paastin?ath paastin?ath American  unaccusative  In contrast, a locative predicate such as Pu-k i"\n" is inherently 'birelational' in the sense of Hale w  and Keyser (2002). The function mapped by the locative predicate is necessarily saturated by two arguments, a locatum (the element which is located) and a location. In the example below, c'a?ak "water" corresponds to the locatum of ?u-k i "in", while c'ax ac "bucket" specifies the w  w  location of the water. (12)  ?uk i?is ca?ak cax ac ?u-ci-?iis c'a?ak cax ac 0-in-3.lND water bucket The water is in the bucket. w  extended unaccusative  w  w  In the following sections, I discuss two factors which allow a six-way classification of predicates to be derived from these two basic unaccusative and extended  unaccusative  configurations. In §4.1.2, the topic of transitivization is introduced. This is followed in §4.1.3 by the proposal that inherently birelational predicates show flexibility in their theta-role mapping.  4.1.2 Transitivization Transitive and ditransitive syntax is also available to Nuu-chah-nulth affixal predicates; however, following Hale and Keyser (2002), I assume that these are derived structures. Embedded in the transitive and ditransitive configurations are the underlying syntax of unaccusatives and extended unaccusatives, respectively. A transitive predicate is created when a 'light' verbal head, denoted as v, is merged with the basic unaccusative relation (Hale and Keyser 1993, 2002; Chomsky 1995).  The verbal head v introduces an additional ('external') argument, yielding a derived  dyadic verb. (13)  Transitive syntax  V  DP  109  Transitivization is associated with an agentive or causative interpretation which is not present with the basic predication configuration of an unaccusative (Hale and Keyser 1993, 2002; Kratzer 1994). By this analysis, transitive predicates such as ?u-naah "look for" achieve their dyadicity in a manner distinct from locatives such as ?u-k i "in". In the example in (14), the transitive w  affixal predicate Pu-naah "look for" takes a first person argument as the subject, and a second argument, cupcupsum-hikqs "my sweater" as an object. (14)  Tunaahsis c'upcupsum-rukqs Tu-naah-siis cupcupsum-r-uk-qs 0-look.for-l SG.IND sweater-POSS-lSG.PS I'm looking for my sweater.  transitive  Unlike locative predicates, which are inherent two-place relations, transitives are composite predicates which are formed by implanting an unaccusative predicate configuration within the additional syntactic structure projected by the 'light' verbal head v.  By this view, only  cupcupsuwhikqs "my sweater" in (14) is an internal argument. The first person subject is an external argument, made available by transitivization. A parallel transitivization process is hypothesized to occur with ditransitive predicates. While transitives build on basic unaccusative syntax, ditransitives are formed from a basic extended unaccusative relation. As indicated in the diagram in (15), the ditransitive is formed when the verbal head v is merged with an extended unaccusative. This, in turn, is merged with a DP projected as the specifier of the v projection. (15) Ditransitive syntax vP  DP V  DP  By this view, a ditransitive is conceptualized as a locative predicate which has an external argument added to the basic extended unaccusative relation (Freeze 1992, Hale and Keyser 2002). A n example of a Nuu-chah-nulth ditransitive affixal predicate is ?u-yii "give". In (16), ?u-yii'"give" takes three arguments: the second person subject, the recipient ?um?i"mother", and the theme taana "money.  110  (16)  ?uyii?i ?um?i taana Pu-ayji-'ii ?um?i taana 0-give-2SG.IMP>3OBJ mother money Give mom money!  4.1.3 Flexibility in theta-role mapping Transitivization represents one means by which the basic predication configurations are used to build a wider array of argument structures.  Another source of diversity is flexibility in the  mapping between theta-roles and underlying structure. In particular, I follow Hale and Keyser (2002)  in hypothesizing that when a locative lexical item projects two internal arguments, the  arguments of the extended unaccusative may be realised in one of two possible orientations. Following Hale and Keyser predicates.  (2002),  these two types may be termed iocatum' and 'location'  (The topic of locative predicates is discussed in detail in §4.4.) For the class of  locatum predicates, a locatum argument (the located element) is merged with the predicate as the basic step of the derivation; this is followed by another application of Merge in which the location argument is added to the structure.  For location predicates, the basic step is uniting the  location argument with the predicate by Merge; this precedes a secondary step in which the locatum is introduced. (17) Flexibility of theta-role mapping (a) 'locatum' predicate  V  locatum  (b) 'location' predicate  V  location  The mechanisms of syntactic concatenation thus allow inherently dyadic predicates to show variability in their theta-role assignment to complement and specifier positions. By this means, extended unaccusative predicates come in two distinct 'flavours', with inverse configurations of the internal arguments. In one, the locatum acts as the complement, and the location acts as the specifier; in the other, the order of Merge operations are reversed, and the relative positions are consequently the opposite. The availability of inverse argument structures necessitates abandonment of a strict oneto-one mapping between syntactic structure and theta-role assignment (Baker 1988, Hale and Keyser 1993). According to Baker's Uniformity of theta assignment hypothesis, theta-roles have a unique structural realisation.  Ill  (18) Uniformity of theta assignment hypothesis (Baker 1988: 46) Identical thematic relationships between items are represented by identical structural relationships between those items at the level of D-structure. A variety of evidence in Nuu-chah-nulth points to the need for a more flexible mapping mechanism (see §4.4, §4.5). For example, subject agreement in locative predicates shows two distinct patterns: for one class of predicates (locatum predicates), subject agreement is linked to the location argument, as in (19a); for the other class (location predicates), subject agreement is determined by the locatum, as in (19b): (19)  a.  Tukuxssis Tu-uxs-siis  0-on.head-1 SG.IND  ^iTicumF ^iTicumF straw.hat  I'm wearing a straw hat. (lit: "I'm heading a straw hat") b.  ?uk isis Tu-ci-siis  cucsac cucsac  0-in-l SG.IND  tub  w  (locatum predicate)  I'm in the tub.  (location predicate)  This pattern of subject agreement corresponds to a split between those locatives (locatum predicates) which suffix to a locatum argument, and those locatives (location predicates) which suffix to the location argument. (20)  a.  b.  ha?umcu?is qa?uuc?i ha7um-cuu-?iis qa?uuc-?ii food-contain-3.lND burden.basket-DET There's food in the burden basket. (lit: "The burden basket contains food")  (locatum predicate)  qa?uucci?is yama qa?uuc-ci-?iis yama burden.basket-in-3.IND salal.berries The salal berries are in a burden basket.  (location predicate)  I will argue in §4.4 that this contrast in incorporation behaviour follows naturally if locatum predicates such as Pu-cuu "contain" map a locatum argument to complement position, while location predicates such as ?u-k i "in" map a location to their complement. Such an analysis is w  incompatible with the rigid mapping mechanism of Baker's Uniformity of theta assignment hypothesis.  4.1.4 Predicate inventory  ^  In sum, there exists a six-way classification of affixal predicates which take nominal predicates, which are built from the two basic predication configurations.  Unaccusative and transitive  predicate classes are composed from the basic unaccusative relation, while locatum, location,  112  locatum-type ditransitive, and location-type ditransitive arise from the basic extended accusative relation. (21) Classification of affixal predicates which take nominal complement basic configuration (a) unaccusative  transitivization orientation of arguments underived n/a transitivized  predicate type  unaccusative  v  eg. Pu-ni'i "arrive"  transitive eg. ?u-naah "look for"  (b) extended unaccusative  underived  locatum predicate  locatum-type  eg. ?u-cuu "contain"  location predicate  location-type  eg. ?u-k i "in" w  transitivized  locatum-type  locatum-type ditransitive  location-type  location-type ditransitive  eg. Pu-yii"give" eg. ?u-?iip "give to"  4.1.5 Unergatives are necessarily non-affixal Absent from this inventory of affixal predicates are unergative predicates (Perlmutter 1978, Burzio 1981). That is, the Nuu-chah-nulth equivalents of intransitive verbs such as cry, run or dance occur only as independent predicates in the language. examples below, fiih-sik"cry-PERF",  As may be observed in the  kamatq-sik"xvLX\-Y^R¥" and huu+-"dance" are not suffixed  to any element. (22) Unergative predicates: exclusively non-affixal a.  fiihsi?aqkk Tiih-si9e-?aqfc-k cry-PERF-FUT-2SG.Q  Are you going to cry? b.  saya?ii?is saya-?ii-?iis  kamatqsifc kamatq-sifc  much-go-3.IND run-PERF  S/he ran far. c.  huu+huu+amitk huu+-a[+R]-mit-k  ?athiimit?i ?athii-mit-?ii  dance-ITER-PST-2SG.Q  night-PST-DET  Did you dance last night? I relate this systematic absence of affixal unergative predicates to the requirement that affixal predicates must have an independent internal argument. internal argument.  Following Hale and Keyser (1993,  analysis of unergatives.  113  Unergatives lack an independent  2002),  I adopt a concealed transitive  (23) Unergative syntax  V < ^ > NP  According to this concealed transitive analysis, unergatives are composite predicates which contain an inner V P layer embedded within a vP, as in (23). Unlike in transitives, however, the 'phonological signature' (or 'p-signature') of V is defective in unergatives. This defectiveness results in the internal argument undergoing conflation with the V . Hale and Keyser (2002: 63) describe this process in the following way: (24)  Conflation consists in the process of copying the p-signature of the complement into the p-signature of the head, where the latter is "defective".  Thus, the internal argument of an unergative is not independently realised under this analysis. This lack of independence presents a problem for affixation. By hypothesis, an affixal predicate requires a host chosen from its derivational sister. Two elements are required for this arrangement: an affix, -a, and a host, P(-), resulting in p-a. The problem with an unergative affix is that the local spell-out domain of the predicate would contain only one phonologically contentful element, that of the N P complement. In (25), for example, V lacks a phonological form independent of the N(P). Thus, no host-affix dependency may be established between the V and N(P) at local spell-out. (25)  vP  v  V P <•  local spell-out  V < ^ > NP  Since affixation is inherently a binary relationship, affixal predicates are not predicted to be possible as the V of an unergative.' Affixal predicates require an independent internal argument so that they may find a host within the local spell-out domain.  ' I take this to be a diachronically relevant fact: I assume that the restriction was at play when the class of affixal predicates developed. The origins of a distinct class of affixal predicates may possibly be traced as far back as Proto-Wakashan (see APPENDIX A ) . Perhaps the expletive host ?u- was not available as a host at the time that affixal predicates developed.  114  The remainder of this chapter is organised as follows. The first two sections deal with predicates formed from the basic unaccusative configuration: unaccusatives and transitives. I begin in §4.2 with a discussion of intransitive predicates in Nuu-chah-nulth, and argue for the existence of a class of unaccusative affixal predicates.  In §4.3, I discuss transitive affixal  predicates, and illustrate syntactic diagnostics for their structure, including subject agreement and possessor raising.  Next, I turn to the structure of predicates formed from the extended  unaccusative configuration: locatives and ditransitives. In §4.4,1 show how syntactic diagnostics motivate a distinction between two classes of locative predicates with inverse argument structures, which I label location and locatum predicates.  The behaviour of ditransitives is  discussed in §4.5, where I analyse these predicates as (di)transitivized locative verbs, which, like locative predicates, may be divided into two classes based on their asymmetrical argument structures.  In §4.6, outstanding issues are discussed. Finally, §4.7 gives a summary of the  findings.  4.2 Unaccusatives This section presents evidence for the existence of an unaccusative-unergative distinction in Nuuchah-nulth, and proposes that there are no unergative affixal predicates in Nuu-chah-nulth. By my analysis, intransitive affixal predicates in Nuu-chah-nulth are exclusively unaccusative. (26)  VP V  DP  4.2.1 Incorporation Predicates in the unaccusative class include Tu-nii "arrive", Tu-pai- "be present", ?u-?aa?atu "move down". As indicated in the following (a) examples, these unaccusative predicates allow suffixation to their argument. The examples in (b) show affixation to the expletive host ?u-. (27)  (28)  a.  paastin?athni?is paastin?ath-nii-?iis American-arrive-3.IND Americans came.  b.  Tuniiris paastin?ath?i ?u-nii-?iis paastin?ath-?ii 0-arrive-3.IND American-DET The Americans came.  a.  tapa-hvaris ta-pa+-wa?is sick-present-3 .QUOT  pisaqaqwa?is pis-aq[+S]-aq[+S]-wa?is bad-AUG-AUG-3.QUOT  There's sickness around, (and) it is really bad.  115  J  (29)  b.  7upa+wa?is tamis pisaqaqwa?is ?u-pa+-wa?is ta-mis pis-aq[+S]-aq[+S]-wa?is 0-present-3.QUOT sick-NOM bad-AUG-AUG-3.QUOT There's sickness around, (and) it is really bad.  a.  tatuus?a?arumit?is tatuus-Saa?atu-mit-?iis star-move.down-PST-3 .IND A star fell.  b.  ?u?aa?atumit?is ?u-Taa?ata-mit-?iis 0-move.down-PST-3.lND A star fell.  tatuus tatuus star  Incorporation of the argument of an unaccusative is predicted by the PF Incorporation analysis. The internal argument is a derivational sister to the affixal predicate, and thus forms a local spell-out domain with the affixal predicate. When the affixal predicate reaches spell-out, it looks to this derivational sister for its host. For example, if the affixal predicate Pu-iiii "arrive" takes paastinPath "American(s)" as its argument, then the reflex of spell-out will be a linearization of paastinPath-nii, as inpaastinPathniPis"Americans came" in (30a). (30) -nii arrive  paastinPath American  The affixal predicate suffixes to its derivational sister. The alternative to suffixation to a derivational sister is suffixation to a host inserted at spell-out, the expletive morpheme Pu-. In (31b), the presence of the determiner -/fr imposes an 'edge' between the affixal predicate and the members of its complement. (31)  paastinPath American ,  As noted in Chapter 3, DPs in Nuu-chah-nulth act as independent spell-out domains in that they form 'islands' for affixation. Inflectional clitics, for example, are built up within a DP and do not cross it. In a context such as (31), the expletive Pu- 'rescues' the stranded affixal predicate by acting as its host. PF Incorporation is not possible in this context.  2  A topic for further research is why the expletive ?u- is employed in cases such as (28b), which lack a determiner. Incorporation is apparently optional in such cases. Whether or not there is a detectable pause between the predicate and complement in cases such as (28b) remains to be tested, although I am not aware 2  116  In the next section, I discuss evidence for an unergative-unaccusative distinction in Nuuchah-nulth.  4.2.2 Intransitivity in Nuu-chah-nulth According to the Unaccusativity Hypothesis (Perlmutter 1978, Burzio 1981), there are two subclasses of monadic predicates - unaccusatives and unergatives - which are associated with different underlying syntactic configurations. While the argument of an unaccusative verb such as "arrive" or "die" is an internal argument, the single (overt) argument of an unergative verb such as "cry" or "dance" is an external argument. In the framework which I have adopted, the difference between these two intransitive types may be represented according to an asymmetry as to which syntactic head introduces the argument. The argument of an unaccusative is generated as the complement of V , in the same position as the object of a transitive predicate. For unergatives, however, the argument is introduced by a v head, in the same position as the subject of a transitive predicate (Hale and Keyser 1994, Kratzer 1994, Chomsky 1995). (32)  VP  (33)  unaccusative  vP  unergative  Recall that I adopt the analysis that unergatives are 'concealed' transitives which have a conflated internal argument. Thus, only the top argument (circled) of the unergative is overtly realised as an independent argument. In Nuu-chah-nulth, contrasts between unaccusatives and (non-affixal) unergatives are occluded by the fact that both types of predicates receive the same type of subject agreement. A predicate such fiih "cry" takes the same subject agreement as does a predicate such as hinin "arrive". In (34), both are inflected for the third person indicative subject agreement -?/>s(3.IND). (34)  a.  TiKiihamitTis ?iih-a[+R]-mit-?iis  na?iiqsakqs na?iiqsu-?ak-qs  cry-IT-PST-3.IND  aunt/uncle-POSS-1 SG.PS  M y auntie was crying.  of any (impressionistic) differences between (28a) and (28b). whether use of ?u- correlates with specific stylistic effects.  117  Textual analysis may also shed light on  b.  hinin?a?e?is hinin-?a?e-?iis arrive-TEMP-3.IND  na?iiqsakqs na?iiqsu-?ak-qs aunt/uncle-POSS-lSG.PS  M y auntie has arrived now. Despite this superficial similarity of unergatives and unaccusatives in Nuu-chah-nulth, I will advance two separate pieces of evidence for a distinction between unaccusative and unergative intransitives in the language. First, I propose that there is distributional evidence for unaccusativity, based on the class membership of affixal predicates (§4.2.3). Secondly, I propose that 'long' possessor raising constitutes a reliable syntactic diagnostic for unaccusativity in Nuuchah-nulth (§4.2.4).  4.2.3 Absence of unergative affixal predicates A systematic gap in the composition of the Nuu-chah-nulth lexicon supports a contrast in this language between the two monadic classes of unergatives and unaccusatives. While one-place predicates with the semantics of typical unaccusative verbs (eg. die, arrive) are found amongst both the affixal and non-affixal classes in Nuu-chah-nulth, to the best of my knowledge, oneplace predicates with the semantics of typical unergative predicates (eg. work, cry, dance) exist only as non-affixal predicates. The generalisation that unergative predicates are absent from the affixal predicate inventory is supported by the suffix lists in Sapir and Swadesh (1939), and by the grammars of Rose (1981) and Davidson (2002). For example, consider Rose's (1981) description of the two classes of 'verbal affixes' in Kyuquot, a northern dialect of Nuu-chah-nulth. Rose labels the two verbal classes 'governing' or 'restrictive'. Amongst the 'governing' category of verbal affixes that take an ' N P object base' (rather than a sentential one), we find predicates with transitive and ditransitive syntax, according to the present classification. Unaccusatives are found amongst Rose's class of 'restrictive verbal affixes'. Importantly, there is no other category of 'verbal affix' in Rose's list which plausibly resembles unergatives. The subclasses of 'verbal affixes' discussed in Rose (1981) are illustrated in the following table with their correspondences to the present classification system.  118  (35) Types of' Verbal affixes' Classification Ahousaht examples  Kyuquotcognates  unaccusative  -ni "come home, arrive" restrictive verbal -Fata "sink, go down" affix -suwi(k) (~suu(9c)) "die, get destroyed" "come into house"  transitive  Tu-nii "arrive" ?uu-?atu "sink, go down" ?u-suuk"die, get destroyed" ?u-yh?i&"come into house" ?uu-taq"fix, work on" 7u-k iH- "make" ?u-?iic "consume" Tu-ciii-h "use as fuel" ?u-?iip "give" Pu-y/i" give" Puu-ks "ask for" -s/nhi[+L] "try to continue" -mafiiq&"need to" (bodily functions) w  ditransitive  auxiliary  Label in Rose (1981)  governing verbal -taq[+L] "work on.." -(c)iii- "make.." affix -'iic "eat." -NP object base -ci+h "use., as fuel" -?iip "give.." governing verbal -aayi"g\ve." affix -^>>Ll"askfor.." -NP object base -sinh [+L] "try to (be).." governing verbal -ma'iaq&"'want to (be).." affix -sentential base  Note that a 'governing verbal affix' such as Tuu-taq "work on" is strictly transitive, with a meaning similar to "fix", rather than with a usage parallel to English intransitive work. (36)  a.  Tuutaqitsis Tuu-taq-mit-siis 0-work.on-PST-1 SG.IND I was working on an engine.  b.  muunaa muunaa engine  * Tuutaqitsis ?uu-taq-mit-siis 0-work.on-PST-1 SG.IND  I was working. (consultant's comment: "you have to tell what you were fixing or working on") In a similar vein, the affixal predicate Pu-Piic "consume" has only a transitive usage. In the example below, the object sapnii'"bread" is mandatorily expressed. (37)  a.  b.  *  ?u?iicamitsis ?u- 'iic- 'ap-mit-siis 0-consume-TR-PST-1 SG.IND I made Ken eat bread. ruPiicamitsis ?u- 'iic-'ap-mit-siis  Ken Ken Ken Ken Ken  0-consume-TR-PST-1 SG.IND  Ken  sapnii sapnii bread  I made Ken eat. This contrasts with the behaviour of the independent predicate haPuk "eat", which, like English eat, allows for both an intransitive and transitive usage. In (38a), the object sapnii "bread" is expressed; in (38b), it is not.  119  (38)  ha?uk amitsis ha?uk-'ap-mit-siis eat-TR-PST-1 SG.IND I made Ken eat bread.  Ken Ken Ken  ha?uk amitsis ha?uk-'ap-mit-siis eat-TR-PST-1 SG.IND I made Ken eat.  Ken Ken Ken  w  w  sapnii sapnii bread  In the next section, we see further evidence that intransitive affixal predicates in Nuuchah-nulth pattern as unaccusatives and not as unergatives.  4.2.4 'Long' possessor raising as a diagnostic for unaccusativity As first described in Chapter 2, Nuu-chah-nulth has a process of possessor raising in which the possessive marker -uk/-(PJak (POSS) suffixes to a main predicate, instead of (or in addition to) suffixing to the possessum (Davidson 2002, Ravinski 2005). In (39a), the possessum k aaPuuc w  "grandchild" is suffixed by -uk (POSS). In the possessor-raised (39b), -uk (POSS) suffixes to the predicate taPi-f "sick". (39) ,  a.  ta?i+?is k aa?uucukqs ta?i+-?iis k aa?uuc-uk-qs sick-3.lND grandchild-POSS-lSG.PS M y grandchild is sick.  (unraised)  taTiFuksis k aa?uuc taTiF-uk-siis k aa?uuc sick-POSS-1 SG.IND grandchild M y grandchild is sick.  (possessor raising)  w  w  b.  w w  In possessor raising, the possessor ends up determining subject agreement for the clause: in (39b), the subject agreement is —siis (1 SG.IND) because it matches the first person possessor of k aa?uuc "grandchild". w  Standardly, possessor raising targets the surface subject of the main  predicate in Nuu-chah-nulth, including the derived subjects of passives and unaccusatives (Ravinski 2005). For main predicates, possessor raising is insensitive to the difference between the arguments of unergatives and unaccusatives. As shown in (40), possessor raising is permitted with the argument of an 'unergative' intransitive such as fiih "cry", or the argument of an 'unaccusative' intransitive such as hinin "arrive". (40)  a.  fihSiihakitsis ?iih-a[+R]-?ak-mit-siis cry-IT-POSS-PST-1 SG.IND M y auntie was crying.  na?iiqsu na?iiqsu aunt/uncle  120  b.  hinin?akitsis hinin-?ak-mit-siis arrive-POSS-PST-1 SG.IND M y auntie came  na?iiq na?iiq aunt/uncle  However, when possessor raising applies in contexts of affixal auxiliaries such as -qaath "claim" or -mahsa "want to", a distinction emerges between unaccusatives and other types of predicates. When an unaccusative predicate combines with an affixal auxiliary, two possible interpretations are available in contexts of possessor raising. The example in (41) shows these two interpretations for the predicate  "sick", which is suffixed by the affixal auxiliary -qaath  "claim". (In Nuu-chah-nulth, ia/Stf-''sick" patterns with the class of unaccusative verbs.) In the 3  first interpretation in (41), the one who doing the claiming is the same one who is sick ('my grandchild'). A second interpretation is also available in which the claimer and the one who is sick are disjoint: the 'claimer' is a first person argument, and the one who is sick is 'my grandchild'.  (For the second type of interpretation, the person and number features of the  'claimer' necessarily matches the person and number of the possessor of the one who is sick.) (41)  ta?i+qathuk itsis k aa?uuc taTiF-qaatn-uk-mit-siis k aa?uuc sick-cIaim-POSS-PST-1 SG.IND grandchild = (i) M y grandchild claimed to be sick. = (ii) I claimed my grandchild was sick. w  w  w  ('short' possessor raising) ('long' possessor raising)  I refer to the first interpretation as a case of 'short' possessor raising, and the second as an instance of 'long' possessor raising, for reasons which will soon become apparent. For unergative predicates, only a 'short' possessor raising interpretation is available; 'long' possessor raising is impossible. This is indicated in (42) with the unergative predicate tiaq-sifc "shout (PERF)".  The 'claimer' and the 'shouter' must be the same person in this  possessor raising context. (42)  ?aaqsi?eqathuk itsis naniiqsu 9aaq-sifc-qaath-uk-mit-siis naniiqsu shout-PERF-claim-POSS-PST-1 SG.IND grandparent = (i) M y grandparent claimed to shout. ± (ii) I claimed my grandparent shouted. w  ('short' possessor raising) ('long' possessor raising)  Thus, the availability of a 'long' possessor raising interpretation distinguishes between the class of unaccusatives and the class of unergatives: unaccusatives such as ta?i+ "sick" are compatible with 'long' possessor raising, while unergatives such as ?aaq-sik"shout (PERF)" are not. The interpretations are disambiguated by context. For example, in a sentence such as ta?i+qathuk itsis k aa?uuc c'inuqk kiisMisa^as "My grandchild claimed she was sick because she didn't want to go to w w school", the meaning is clearly the first interpretation. For ta?i+qathuk itsisk aa?uuc cinuq%ap naPuuk nuwiiqsak"\ claimed my grandchild was sick because I didn't want her to go with her father", the second interpretation arises. w  w  121  A sketch of my analysis of the two types of possessor raising follows. As I will argue in Chapter 5, affixal auxiliaries such as -qaath "claim" in (43) are raising verbs which do not project a subject of their own, but which license raising of an embedded argument. (43)  nunuukqath?is Florence nunuuk-qaath-?iis Florence sing-claim-3.lND Florence Florence is pretending to sing.  The diagram in (44) represents how Florence takes on the role of 'shared' subject of the auxiliary -qaath "claim" and the main predicate nunuuk "sing". In (44), Florence originates as the subject of the main predicate, and raises to specifier position of the auxiliary. (44) Affixal auxiliaries as raising verbs FP  nunuuk sing Following Ravinski (2005), I assume that the possessive morpheme -uk  (POSS)  licenses a  position for raised possessors. This analysis is illustrated in (45) for the non-auxiliary case of possessor raising taPiFuksis k aa?uuc "My grandchild is sick", from (39b). Here, the first person w  possessor of k aaPuuc "grandchild" raises to specifier position of PossP. w  (45) 'Simple' possessor raising PossP  -uk ta?iF sick k aa?uuc grandchild w  t  —  1  DP  In 'short' and 'long' possessor raising with auxiliaries, the syntax of the auxiliary interacts with the syntax of possessor raising.  122  (46)  ta?iFqathuk itsis k aa?uuc ta?iF-qaath-uk-mit-siis k aa?uuc sick-claim-POSS-PST-1 SG.IND grandchild = (i) M y grandchild claimed to be sick. = (ii) I claimed my grandchild was sick. w  w  w  ('short' possessor raising) ('long' possessor raising)  In the case of 'short' possessor raising with auxiliaries, I propose that the entire possessed nominal ("my grandchild") raises to a 'subject' position of the auxiliary, in specifier position of FP. This is followed by possessor-extraction, which raises the possessor to specifier of PossP. In 'long' possessor raising, in contrast, the possessor is raised twice, on its own. A first move raises the possessor to 'subject' position of the auxiliary, while the second move takes the possessor to specifier of PossP. The difference between the 'short' and 'long' possessor raising is illustrated below. (47) Possessor raising with unaccusative main predicate a.  'short'possessor raising *  b. 'long'possessor raising  PossP  ,  PossP  grandchild Thus, in the 'short' case of (47a), "my grandchild" acts as the (derived) subject of -qaath "claim", while in the 'long' case of (47b), -qaath "claim" has a first person subject (equivalent to the first person possessor). In both cases, the first person possessor occupies specifier position of PossP, and ultimately ends up determining the first person subject agreement of the clause, -siis (1 SG.IND).  As noted in the previous discussion, only unaccusative predicates such as taTi-t "sick" permit 'long' possessor raising. As illustrated in (48), the argument of an unergative such as Taaq-sik"shout ( P E R F ) " is not compatible with 'long' possessor raising.  123  (48) Illicit 'long' possessor raising with unergative main predicate PossP  -sik PERF  The  / \  naniiqsu  tp D  Taaq grandparent" shout  cause for this restriction is in need of further investigation. Whatever the  grammatical motivations for this contrast, it serves as a reliable diagnostic for unaccusatives. The unaccusativity restriction is an empirically robust distinction, and holds for Nuu-chah-nulth speakers across a range of predicates. Intransitive predicates such as mamuuk "work", hita?ap "win", and yaac "walk" all disallow a 'long' possessor raising interpretation.  The possible  interpretations are shown below with the affixal auxiliary -mahsa "want to". (49)  (50)  (51)  mamuukmahsaksis na?iiq mamuuk-mahsa-?ak-siis na?iiq work-want-POSS-1 SG.IND aunt/uncle = (i) M y aunt/uncle wants to work. ^ (ii) I want my aunt/uncle to work.  ('short' possessor raising) ('long' possessor raising)  hita?apmahsaksis naniiq hita?ap-mahsa-?ak-siis naniiq win-want-POSS-1 SG.IND grandparent = (i) M y grandparent wants to win. ^ (ii) I want my grandparent to win.  ('short' possessor raising) ('long' possessor raising)  yaacsifrmahsaksis na?iiqsu yaac-sifc-mahsa-?ak-siis nariiqsu walk-PERF-want-POSS-1 SG.IND aunt/uncle = (i) M y aunt/uncle wants to go for a walk. ^ (ii) I want my aunt/uncle to go for a walk.  ('short' possessor raising) ('long' possessor raising)  These unergative intransitives pattern together with transitive predicates such as Tu-k^if 'make" his-sifc "hit ( P E R F ) " , and ?u-?aap "buy", which also disallow 'long' possessor raising. In (52), only a 'short' possessor raising interpretation is permitted with the transitive predicate ?u-k iiP w  "make". Likewise, in (53), a 'long' possessor raising interpretation is shown to be impossible for  124  the transitive predicate his-sik "hit ( P E R F ) " . Finally, the example in (54) shows this restriction against 'long' possessor raising for the transitive predicate ?u-?aap "buy". (52)  (53)  saapniqi-fmahsaksis naniiqsu saapniq-cii-F-mahsa-?ak-siis naniiqsu bread-make-want-POSS-1 SG.IND grandparent = (i) M y grandparent wants to make bread. ^ (ii) I want my grandparent to make bread. hissikqathuk itsis his-si?e-qaath-uk-mit-sis  yuk iiqsu yuk iiqsu  hit-PERF-claim-POSS-PST-1 SG.IND  y .sibling  w  w  w  = (i) M y younger sibling claimed s/he hit Ray. ^ (ii) I claimed my younger sibling hit Ray (54)  huupuk as?apqathuk itsis huupuk as-?aap-qaath-uk-mit-siis  yuk iiqsu yuk iiqsu  car-buy-claim-POSS-PST-1 SG.IND  y.sibling  w  w  w  ('short' possessor raising) ('long' possessor raising) Ray Ray Ray ('short' possessor raising) ('long' possessor raising)  w  w  = (i) M y younger sibling claimed s/he bought a car. ^ (ii) I claimed my younger sibling bought a car.  ('short' possessor raising) ('long' possessor raising)  Conversely, unaccusative predicates consistently allow the 'long' possessor raising interpretation. In the appropriate context, either a 'short' possessor raising or a 'long' possessor raising interpretation is allowed for the sentences below with the unaccusative predicate sahyut "healthy". The sentence in (55a) shows both readings in the context of the auxiliary -mahsa "want", while (55b) illustrates parallel readings with the auxiliary -qaath "claim". (55) a. sahyutmahsaksis naniiqsu sahyut-mahsa-?ak-siis naniiqsu healthy-want-POSS-1 SG.IND grandparent = (i) M y grandparent wants to be well. = (ii) I want my grandparent to be well. sahyutqathuk itsis naniiqsu sahyut-qaath-uk-mit-siis naniiqsu healthy-claim-POSS-PST-1 SG.IND grandparent = (i) M y grandparent claimed to be well. = (ii) I claimed my grandparent was well.  ('short' possessor raising) ('long' possessor raising)  w  ('short' possessor raising) ('long' possessor raising)  For some, but not all, of consultants, the 'long' possessor raising interpretation is highly salient for unaccusative examples such as (55). However, for all consultants there are some contexts in which only a 'long' possessor raising interpretation arises. When an unaccusative main predicate takes an inanimate argument, a 'short' possessor raising interpretation is ruled out -  perhaps due to the pragmatic restriction that the inanimate cannot control a desiderative  auxiliary such as -mahsa "want". This is illustrated with the unaccusative main predicate tuq-sik "melt ( P E R F ) " , which in the sentence below takes the inanimate argument pata "butter".  125  (56)  tuqsitanahsaksis pata mq-sifc-mahsa-?ak-siis pata melt-PERF-want-POSS-1 SG.IND butter ^ (i) ! M y butter wants to melt. = (ii) I want my butter to melt.  ('short' possessor raising) ('long' possessor raising)  In this example, the only pragmatically available reading is a 'long' possessor raising interpretation in which the controller of the auxiliary predicate -mahsa "want" is the same as the first person possessor of the main predicate's argument, pata "butter".  A 'short' possessor  raising interpretation is impossible, since this entails an absurd reading in which the controller of the main and auxiliary predicates is the possessive nominal itself ('my butter').  Additional  examples of 'long' possessor raising with unaccusatives are shown in (57). Predicates which are compatible with 'long' possessor raising include puux-sik "rise "shine (57)  (CONT)"  and cah-aa "leak  (PERF)",  ntiPatu "sink", path-aa  (CONT)".  a.  puuxsi&mahsaksis puux-sifc-mahsa-?ak-siis rise-PERF-want-POSS-1 SG.IND I want my bread to rise.  sapnii sapnii bread  b.  nii?atuqathuk it?ick nii?atu-qaath-uk-mit-?iick sink-claim-POSS-PST-2SG.iND You claimed your boat sank.  muut muut boat  c.  pathaamahsaksis tirfa path-aa-mahsa-ak-siis tii+a shine-CONT-want-POSS-1 SG.IND fishing.spoon I want my fishing spoon to shine.  d.  cahaaqathuk it?ick muut ?ata?ick Tuyiiq cah-aa-qaath-uk-mit-riick muut ?ata-riick Tu-yiiq leak-CONT-cIaim-POSS-PST-2SG.IND boat but-2SG.IND 0-travel.in You claimed your boat is leaking, and yet you're travelling in it.  w  w  In sum, 'long' possessor raising can be used as a diagnostic to separate unaccusative predicates from transitive and unergative classes, since only unaccusative main predicates ever allow 'long' possessor raising. The results of this diagnostic are summarized in (58).  126  (58) 'Long' possessor raising as a diagnostic for unaccusativity Main predicate A. Transitive MW'make" hissi&"h\t" Pu-Paap"buy" B. Unergative mamuuk "work" hitaPap "win" Taaqsifc "shout" yaacs/&"wa\k" C. Unaccusative sahyut "healthy" tuqsj'&"me\t" puuxsifc"r\se" nh'Patu "sink" pathaa "shining" cahaa "leaking" Pu-suu&"die" Pu-riti "arrive"  'long' possessor raising X X X  X X X X  •/ •/  s  If we apply the 'long' possessor raising diagnostic to intransitive affixal predicates, we see that this test confirms that these affixal predicates behave similarly to non-affixal predicates such as sahyut "healthy" and tuq-sifr "melt ( P E R F ) " in that they allow a 'long' possessor raising interpretation.  In other words, these intransitive affixal predicates pattern as unaccusatives. In  the sentences below, the intransitive affixal predicate Pu-suufc "die" appears in complex predicates with the affixal auxiliary -qaath "claim". Both 'short' and 'long' possessor raising is permitted, dependent on the discourse context. (59)  a.  ?usuukqathuk?ick nani ?atquu tiic ?u-suufc-qaath-uk-?iick nani ?at-quu tiic 0-die-claim-POSS-2SG.IND grandparent but-3.COND alive Your grandparent claimed she died, but she is alive, ('short' possessor raising) (context: life insurance fraud scenario)  b.  TusuukqathukPick nani ?ata?is tiic ?u-suufc-qaath-uk-?iick nani ?ata-?iis tiic 0-die-claim-POSS-2SG.IND grandparent but-3.IND alive You claim your grandparent died, but she is alive. ('long' possessor raising) (context: bereavement leave scenario)  'Long' possessor raising is possible with other intransitive affixal predicates, such as Pu-nii "arrive". (60)  ?uniiqathuk?ick ?u-nii-qaath-uk-?iick  naniiqsu naniiqsu  ?atquu ?at-quu  wi£iit wi£iit  0-arrive-claim-POSS-2.IND  grandparent  but-3.COND  NEG-?  You pretend your grandparent came, although she isn't here. ('long' possessor raising)  127  In conclusion, the 'long' possessor raising diagnostic supports a classification in which affixal predicates like ?u-suu&"die" and Tu-nii"arrive" are unaccusative.  4.3 Transitives  This section considers the behaviour of transitive affixal predicates. These predicates have the following syntactic structure, derived via abstract transitivization of an underlying unaccusative: (61)  vP  Representative examples are shown below, which illustrate suffixation of the predicate to a nominal host. The predicates ?u-?in+ "serve", ?u?u-?iih "gather" and ?u-k iiP w  "make" are each  proposed to be transitive. (62)  a.  camayhvHtsis camas-'m-fr-rnit-siis sweets-serve-PST-1 SG.IND  I served sweets. b.  t'utuciih?is nani tucup-'iihf+RJ-Tiis nani sea.urchin-gather-3.lND grandparent Grandparent is gathering sea urchin.  c.  saapniqirWis ?um?i saapniq-ciH--?iis ?um?i bread-make-3 .IND mom Mom is making bread.  In the absence of incorporation, these predicates suffix to the the expletive pronoun ?u(63)  a.  ?u?in-Htsis Pu-'in-f-mit-siis 0-serye-psT-1 SG.IND I served sweets.  b.  ?u?u?iih?is Mh[+R]-?iis  c.  ?uk irf?is ?um?i sapnii ?u-cii+-?iis ?um?i sapnii 0-make-3.IND mom bread Mom is making bread.  camas camas sweets  tucup nam tucup nani sea.urchin 0-gather-3.lNn grandparent Your grandparent is gathering sea urchin. w  I present evidence in §4.3.1 that only complements may act as the source of incorporation for transitive affixal predicates. Two tests are used to independently affirm the  128  existence of a distinction between syntactic positions of the arguments of a transitive predicate: subject agreement (§4.3.2) registers an argument in subject position; possessor raising (§4.3.3) is possible only out of a subject, and not out of an object. Thus, in active contexts, incorporation is exclusively reserved for an argument which is incompatible with subject agreement or possessor raising.  4.3.1 Incorporation There is an incorporation asymmetry in Nuu-chah-nulth between the two arguments of a transitive affixal predicate.  Recent work has shown that Nuu-chah-nulth affixal predicates  incorporate their object, and not their subject (Woo 2000, Davis and Sawai 2001, Stonham 2004). In Chapter 2, I termed this restriction on affixation the 'complement' effect. The asymmetry is illustrated in the examples below, which show that the two arguments of a transitive predicate do not have equal ability to serve as the host for the affixal predicate. While ?u-?aap "buy" can suffix to the nominal mahta- "house", it cannot suffix to the nominal capx-'man". (64)  a.  b.  mahta?amit?is cakup mahta-'aap-mit-?iis cakup house-buy.-PST-3.IND man A man bought a house. *  capx?aamif?is mahtii capx-'aap-rnit-?iis mahtii man-buy.-PST-3.IND house A man bought a house.  This incorporation asymmetry is robust. For example, in w/z-questions (Davis and Sawai 2001) and relative clauses formed with an affixal predicate, incorporation of the wh- or relativepronoun into the affixal predicate is obligatory for object-questions and object-relativizations, but is impermissible for subject-oriented ones. As Davis and Sawai (2001) describe, incorporation is mandatory for w/z-pronouns that occur as the object of an affixal predicate. In (65), the w/z-object Paqf-"what" must incorporate into the affixal predicate ?u-?iic "consume". (65)  a.  b.  ?aqiicith John ?aqi-'iic-mit-h John what-consume- PST-3.Q John What did John eat? * ?aaqici+h ?u?iic ?aqi-ci+[+L]-h ?u-2ic what-AUX-3.Q 0-consume  (Davis and Sawai 2001:127; ex.11) John John John  What did John eat?  (Davis and Sawai 2001:127; ex.16)  In contrast, incorporation into the affixal prediate is ruled out for w/z-subjects. The example in (66a) shows that it is ungrammatical for the w/z-subject ?acaq- "who" to incorporate into the  129  affixal predicate 7u-?iic "consume". As shown in (66b), a w/i-subject must occur independently of the affixal predicate. (66)  a.  b.  * TacaTiicith suuhaa ?acaq-'iic-mit-h suuhaa who-consume-PST-3.0 salmon Who ate the salmon?  (Davis and Sawai 2001: 129; ex. 19)  ?acaqith ?acaq-mit-h  ?u?iic ?u-4ic  hiiFas suuhaa?i hii+as suuhaa-Tii  who-PST-3.Q  0-consume  DEIC  Who ate this salmon?  salmon-DET  (adapted from Davis and Sawai 2001: 130; ex. 22a)  The same subject/object asymmetry is found with relative clauses, as illustrated below with the affixal predicate ?uu-?inhi "wait for". The relative pronoun yaq "who" incorporates in an object relativization (67a), but not in a subject relativization (67b). (67)  a.  b.  hacumsiqsaksis haa cakup?i yaaTinhi?itq hacumsiqsu-?ak-siis haa cakup-?ii yaq-?inhi[+L]-?iitq brother-POSS-1 SG.IND DEIC man-DET REL-wait.for-3.RL That man who Mary is waiting for is my brother. hacumsiqsaksis haa cakupTi yaq?itq ?uu?inhi hacumsiqsu-?ak-siis haa cakup-?ii yaq-?iitq ?u-?inhi[+L] brother-POSS-1 SG.IND DEIC man-DET R E L - 3 . R L 0-wait.for That man who is waiting for Mary is my brother.  Mary Mary Mary Mary Mary Mary  Under a PF Incorporation analysis, the distinct patterns of subjects and objects are anticipated. An object of a transitive predicate, but not its subject, is generated in complement position. As the derivational sister of an affixal predicate, an object nominal undergoes local spell-out with the affixal predicate. This derives the 'complement' effect. (68)  vP  local spell-out V -?aap buy  NP mahtahouse  At spell-out, a linearization of mahta-?aap "house-buy" is induced. In the next section, we turn to our first of two syntactic diagnostics which corroborate the analysis that it is the syntactic complement of a transitive affixal predicate which serves as the source of incorporation:  130  4.3.2 Subject agreement Agreement in Nuu-chah-nulth corresponds to the subject of a transitive predicate, not to an object. The table below lists the Ahousaht agreement paradigm for the indicative mood, one of several mood inflections in the language (see Chapter 3, APPENDIX C; Nakayama 1997, 2001). (69) Indicative subject agreement  1 2 3  singular -siis -?iick  plural -niis -?iicuus -?iis  Subject agreement registers the subjects of affixal and non-affixal predicates alike. For example, in (70a), the first person singular marker -siis references the (pro) subject of the affixal predicate Puu-Acii "cook"; in (70b) the same marker applies to the non-affixal predicate kith-sifr "phone (PERF)". (70)  a.  Tuuhciisis 7u-hcii[+L]-siis 0-cook-l SG.IND I am cooking meat.  cisqmis cisqmis meat  kithsi?aq9esis suwa ?athii wikquus haana?a?as kith-si ?e-?a q^-s i is suwa ?athii wik-quus haana?aq-'as ring-PERF-FUT-1 SG.IND you tonight NEG-lSG.COND lahal-ASP I'll phone you tonight if I don't go to the lahal game. I assume that subject agreement in Nuu-chah-nulth is licensed by the highest argument of a predicate (Ravinski 2005).  Higher arguments are chosen over lower arguments due to the  Minimal Link Condition (Chomsky 1995), in which short-distance relationships are preferred over long-distance ones.  This is represented in (71), in which the external argument of a  transitive is the closer to the agreement projection than the internal argument of the transitive. Subject agreement can be assumed to be licensed through an AGREE relation between a head (eg. Agr) and the most local D P (Chomsky 1995). (71)  AgrP  DP  131  Given the analysis that complements of an affixal predicate act as 'incorporated' hosts, and given that complements of transitives are not the closest nominal to Agr, we expect that an incorporated nominal should not determine the subject agreement of a transitive predicate. This indeed holds. In (72) below, the nominal cisqmis "meat" incorporates into Puu-hcii "cook", while it is the first person argument which determines subject agreement. (72)  ciisqmishcisis cisqmis-hcii|+L]-siis meat-cook-1 SG.IND I am cooking meat.  In the next section, we see further evidence that complements of the transitive affixal predicate are the source of incorporation.  4.3.3 Possessor raising An additional diagnostic for the syntactic structure of transitive affixal predicates is supplied by the possessor raising construction. Recall from Chapter 2 that possessor raising only ever targets subjects in Nuu-chah-nulth (Davidson 2002, Ravinski 2005). This is illustrated in the sentence below with the predicate hin-k a?iih "chase". w  (73)  hink a?iihuksis hin-k a?iih-uk-siis  piispis maamaati piispis maamaati  LOC-after-POSS-3.IND  cat  w  w  bird  = (i) M y cat was after a bird. # (ii) A cat was after my bird. In this example of possessor raising, the possessive morpheme -uk suffixes to the predicate, rather than to the possessum, and the possessor argument controls subject agreement (Davidson 2002, Ravinski 2005). What is noteworthy for our purposes is that the subject piispis "cat" is eligible to receive an interpretation as the possessum, but the object maamaati "bird" is not. Thus, we infer that only the subject is able to act as the source of possessor raising. The subject restriction on possessor raising is illustrated in (74).  132  (74) Subject restriction on possessor raising PossP  -uk  DP _ l  A  lSG maamaati bird  / \ lSG  Ravinski (2005) attributes this subject restriction on possessor raising to the Minimal Link Condition (Chomsky 1995). According, to this condition, shorter moves are preferred over longer ones. As the highest of the two arguments, the subject has the shortest move to Spec, DP. Thus, possessor raising from objects is predicted to be ruled out. Focusing now on affixal predicates, we find that the subject restriction on possessor raising distinguishes between the two arguments of these transitives. The sentence below is an instance of possessor raising with the transitive affixal predicate Tu-yuTaai- "find".  As in the  previous example, -uk (POSS) is suffixed to the predicate and it is the first person possessor which determines subject agreement. (75)  TuyuPaa+uksis ?u-u?aa4--uk-siis  hupkumF hupk-um+  "Tiniik "Tiniifc  0-find-POSS-l SG.IND  ball-RD  dog  = (i) M y dog found the ball. ^ (ii) The dog found my ball. In this example, the only argument which may be interpreted as the possessum is finiifc "dog"; hupkum-f"ba\\" is not interpretable as the possessum. This diagnoses finiifr "dog" as the subject of the predicate 7u-yu?aa+, since only subjects in Nuu-chah-nulth may act as the source of possessor raising. Note that this restriction on interpretation holds regardless of the relative word orders of subject and object. (Recall from Chapter 3 that surface word order of arguments in Nuu-chahnulth is often flexible.) In (76), the arguments are in reverse word order relative to (75).  (  133  (76)  TuyuTaa+uksis liniifc Tu-uTaai-uk-siis riniifc 0-find-POSS-1 SG.IND dog = (i) M y dog found the ball. ± (ii) The dog found my ball.  hupkum+ hupk-um+ ball-RD  Here again, the only argument which is interpreted as the possessum is Tiwik"dog". Also note that the restriction on interpretation in cases of possessor raising holds in spite of the fact that in cases with no possessor raising, either the subject or the object of the affixal predicate is eligible as a possessum. The examples below are instances of unraised possessives with the same predicate, Pu-uPaaJ-'Tind". (77)  a.  b.  ?uyu?aa+?is viniitoikqs ru-u?aa-fr-?iis viniifc-uk-qs 0-find-3.IND dog-POSS-lSG.PS M y dog found the ball.  hupkum+ hupk-um+ ball-RD ( unraised)  ?uyu?aa+?is iinii& hupkum+ukqs ?u-u?aa4'-?iis viniik hupk-unrr-uk-qs 0-find-3.lND dog ball-RD-POSS-lSG.PS The dog found my ball.  (unraised)  In (77a), the subject Tiniik is a possessum, and in (77b) it is the object hupkumi- which is a possessum. The possessive marker -uk is suffixed to the possessum in each case. The results of this possessor raising diagnostic can be used as support of the analysis that complements of a transitive affixal predicate are the source of incorporation. What we anticipate for a transitive affixal predicate is that the argument which tests as a non-subject by the possessor .raising diagnostic should be the same argument which permits incorporation. This predicted behaviour is shown to occur in (78). (78)  ?uyu?aa+uksis rVj-uTaa-fr-uk-siis 0-find-POSS-1 SG.IND M y dog found a bone.  viniifo viniik dog  hamuutu?aaTuksis  limine viniik dog  hamuut-u?aa4--uk-siis bone-find-POSS-1 SG.IND My dog found a bone.  hamuut hamuut bone  Both examples illustrate possessor raising. In (78a), - u?aa+ "find" suffixes to the expletive ?u-, while in (78b), incorporation of the nominal hamuut "bone" occurs. In both cases, the argument hamuut "bone" tests as a non-subject since it fails to receive a possessum interpretation under possessor raising. Since complements of a transitive predicate are not subjects, this behaviour is predicted.  Thus, the results of the possessor raising diagnostic coincide with the proposed  complement effect on incorporation.  134  4.4 Extended unaccusatives This section examines the properties of locative affixal predicates in Nuu-chah-nulth, which I analyse as having the syntax of extended unaccusatives.  These predicates project two internal  arguments, realised in complement and specifier position, respectively. (79)  VP  Locative suffixes are abundant in the Wakashan languages (Sapir and Swadesh 1939, Boas 1947, Anderson 1985). In Nuu-chah-nulth, locative affixal predicates make up a substantial percentage of the affixal predicate inventory. For the Kyuquot dialect of Nuu-chah-nulth, Rose (1981: 293) estimates that there are 127 locative suffixes out of a set of 406 noninflectional affixes in the language, amounting to approximately 31% of the total set.  Locative affixal  predicates denote a range of spatial relationships in Nuu-chah-nulth, including reference to physical locations (eg. ?u-?is "on the beach", ?uu-tsit ox\ the surface of a liquid") and body parts u  (eg. ?uu-wik "on the head", Pufij-qhta "on the foot/feet"), as well as more abstract relationships (eg. Pu-k i"iri\  Pu-kcaas "beside") (Davidson 2002). Representative examples are given below.  (80)  a.  ?u?is?is ?aya muks?i haaF ca?ak?i ?u-^is-?iis ?aya rhuks?i haa+ ca?ak-?ii 0-on.beach-3.lND many rocks DEIC island-DET There's lots of rocks on the beach of that island. (lit: "That island beaches/contains a lot of rocks")  b.  ?uutsit?is cisxmis ?uu-tsit-?iis cisx-mis 0-in.water-3.IND dirt-NOM There's dirt (in the water) in the pot. (lit: "The pot waters/contains dirt")  c.  ?uuwik?is ciisiicum?i ?u-wik[+L]-?iis ciisiicum-?ii 0-on.head-3.IND headband-DET S/he's wearing a headband. (lit: "She's heading a headband")  w  d. .  ?u?uqhtinu?i ?u-qhta[+R]-inufc-'ii  niisyak?i niisyak-?ii pot-DET  suwisuk?itk suwis-uk-?iitk  0-on.feet-PERF-2SG.lMP>3.OBJ shoes-POSS-2SG.PS  Put your shoes on! (lit: "Feet your shoes!") e.  ?uk i?is qa?uuc yama ?u-ci-?iis qa?uuc yama 0-in-3.lND burden.basket salal.berries The salal berries are in the basket. w  135  f.  ?ukcaasuksis ru-caas-uk-siis 0-beside-POSS-1 SG.IND My house is beside a church.  maht'ii saantiquwas maht'ii saanti-quvvas house Sunday-building  Since the first study of Southern Wakashan languages in the early twentieth century, researchers have observed that locative suffixes show contrasts in the type of relationship that holds between the suffix and its host (Sapir & Swadesh 1939, Swadesh 1939). Davidson (2002: 180-181) aptly notes that the locative suffixes -cuu "contain" and -ci "in" show opposite patterns with respect to the nominal they suffix to. In the examples below, the locative suffix ?u-cuu "contain" cannot suffix to the nominal gaPuuc 'burden basket' (81b), while the Pu-k i "in" does w  have the ability to do so (82b). (81)  (82)  a.  ha?umcu?is qa?uuc?i ha7um-cuu-?iis qa?uuc-?ii food-contain-3. IND burden.basket-DET There's food in the burden basket. (lit: "The burden basket contains food") (cf. Davidson 2002: 181, ex. 277)  b.  * qa?uuccu?is ha?um qaTuuc-cuu-Tiis ha?um burden, basket-contain-3. IND food There's food in a burden basket.  a.  * yarftaci?is qa?uuc yarha-ci-?iis qa?uuc salal .berries-in-3. IND burden.basket There's salal berries in a burden basket.  b.  qa?uucci?is yarha?i qa?uuc-ci-?iis yama-Tii burden.basket-in-3.IND salal.berries-DET The salal berries are in a burden basket, (cf. Davidson 2002: 181, ex. 276)  The claim that I develop in this section is that the suffixation patterns of locative suffixes in Nuuchah-nulth fall out from conditioning effects of their argument structure. Under my analysis, the predicate Pu-cufu) "contain" is classified as a locatum predicate, while ?u-k i"'m" is classified as w  location predicate. This terminology is borrowed from treatments of English denominal 'location' and 'locatum' verbs (Clark and Clark 1979, Hale and Keyser 2002). As Clark and Clark (1979) describe, English locatum verbs (such as clothe or saddle) are derivationally related to nominals which specify an object which is located (the "locatum"). In (83a), the locatum verb saddle references the object (the saddle) which is placed onto the horse. On the other hand, location verbs in English (such as shelve or bottle) are derived from nouns that specify the location of the object, and not the located object itself. references the position (the shelf) where the books end up.  136  In (83b), the location verb shelve  (83)  a.  She saddled the horse,  (locatum verb; Hale and Keyser 2002: 19, ex. 35b)  b.  I shelved the books.  (location verb; Hale and Keyser 2002: 19, ex. 35a)  Hale and Keyser (2002) present an analysis of English locatum and location verbs in which the two locative predicate types show inverse argument structures. Locatum predicates have a locatum argument in complement position of a covert preposition, and a location argument in specifier position of the preposition. Location predicates, in contrast, have a location argument in their complement, and locatum in their specifier. (84) a. locatum predicate  b. location predicate  V  V  V  V DP  DP  A  A  N saddle (locatum)  the horse (location)  the books P (locatum)  N shelf (location)  In Hale and Keyser's (2002) analysis, the predicative function of these denominals is represented by a synthetic structure composed of stacked prepositional and verbal predicates headed by covert elements. The nominal head in each configuration ("saddle", "shelf) comes to be realised as a verb via a process of conflation which ties together the nominal with the verbal head. Following Hale and Keyser's (2002) analysis of English locative denominals, I propose that the two locative classes in Nuu-chah-nulth differ in that locatum predicates such as Pu-cuu "contain" take a locatum argument as their complement, while location predicates such as ?u-k i w  "in" take a location argument as their complement. Assuming that the second argument of these dyadic verbs is introduced in a right-branching specifier position in Nuu-chah-nulth, this yields the following syntactic representation: (85)  a. locatum predicate  b. location predicate  VP  VP location  V -cuu contain  locatum V -ci in  locatum  location  This analysis of Nuu-chah-nulth locatives differs from Hale and Keyser's (2002) treatment of English denominal locatives with respect to the process of conflation. As noted earlier in the chapter, conflation copies the 'p-signature' of a complement onto a defective head. Conflation accounts for the 'denominal' characteristics of the English locatives. In Hale and  137  Keyser's analysis of English locatives, the lexical head islntroduced as a nominal complement of a phonologically 'defective' (ie. covert) preposition. This prepositional phrase is embedded as the complement of a covert verb. 'signature'  By conflation (represented by an arrow), the phonological  of the nominal comes to be associated with a phonologically defective  prepositional head, and in turn, a defective verbal head (0).  (0)  The ' V thus takes on the  phonological characteristics of the N , saddle. (86) denominal-forming conflation V  P  s  S N  0  saddle  In this way, the process of conflation derives the verbal behaviour of the nominal head. In Nuu-chah-nulth, a different set of empirical facts holds. Unlike the English denominal predicates ("shelve", "clothe") which show an overt correspondence to nominals ("shelf, "cloth(es)"), there is no evidence for a nominal alternation with Nuu-chah-nulth locatives. In other words, Nuu-chah-nulth locative predicates are not related to any free-standing nominals. Accordingly, I represent the lexical material of the Nuu-chah-nulth locative predicates as the predicate heads themselves, rather than as nominal heads embedded in additional structure. In other words, I propose that no abstract conflation occurs in Nuu-chah-nulth to derive the predicative function of locatives. In the analysis I have presented, the locative predicates Pu-cuu "contain" and Pu-k i "in" directly occupy a verbal head position, V . There is no mediating w  prepositional component to the predication. That is, I adopt a direct predication analysis over a prepositional conflation analysis. (87) a. prepositional conflation  b. direct predication  PP  VP  P <r=> NP locative head  locative head  A further difference between Nuu-chah-nulth locatives and English locative denominals is that Nuu-chah-nulth locatives lack agentive force.  In the Hale and Keyser analysis, the  agentivity of English denominals is represented in the synthetic predicate structure by a tacit  138  verbal element equivalent to "put". No such covert verbal head is warranted in the Nuu-chahnulth cases; in the vP shell framework which I have adopted, the absence of agentive force of Nuu-chah-nulth locatives may be denoted by the absence of a vP projection. This allows for a contrast between stative locatives in Nuu-chah-nulth, and agentive locatives formed by overt causativization. The following example expresses a complex predicate equivalent to English "put", derived by causativization of the location predicate ?u-k i"\x\". w  (88)  ?uk inup?aq?esis ?u-ci-nup-?aq?c-siis  &ahiqs?i &ahiqs-?ii  0-in-CAUS-FUT-1 SG.IND  box-DET  w  I'll put it in the box. I assume that agentive locatives such as that in (88) have a vP projection, but stative locatives do not. This is illustrated in the diagrams in (89). (89)  a. stative locative  b. agentive locative  VP  vP  -ci in We now to turn to evidence which supports the syntactic structures of locatives proposed in (85). The diagnostics which I first introduced for transitives in §4.3 (subject agreement and possessor raising) will be shown to support an analysis in which locative predicates in Nuu-chahnulth belong to two distinct classes which have inverse argument structures. In §4.4.1,1 consider locatum predicates, and in §4.4.2,1 turn to location predicates.  4.4.1 Locatum predicates This section presents diagnotics for the syntactic structure of locatum predicates such as ?u-cuu "contain" or Pu-hahu-f"'on front". (90)  a.  rucuu?is ca?ak ?u-cuu-?iis c'a?ak 0-contain-3 .IND water There's water in this. (lit: "This contains water")  139  b.  c'a?akcu?is ca?ak-cuu-?iis water-contain-3 .IND There's water in this. (lit: "This contains water") ?uhahu+?is ?imtii?ak?i haa Tu-hahu-r-Tiis ?imtii-?ak-?i haa 0-on.front-3.IND name-P0SS-3.PS DEIC That woman's got her name written on her front. (lit: "That woman is fronting her name")  (91)  b.  Fuucma?i -fuucma-Tii woman-DET  fimtiihahu-FPis haa -ruucma?i fimtii-hahu-fr-Tiis haa -fuucma-?ii name-on.front-3.IND DEIC woman-DET X That woman's got a'name written on her front. (lit: "That woman is fronting her name")  I analyse these predicates as taking a locatum argument as a complement, and a location as a specifier. (92) location locatum  4.4.1.1 Incorporation According to the PF Incorporation hypothesis, affixal predicates in Nuu-chah-nulth incorporate a host chosen from their derivational sister. Assuming the structure in (93), we predict that a locatum predicate such as Pu-cuu "contain" should be oriented at spell-out with respect to its locatum argument, and not to its location argument. It is the locatum argument of a locatum predicate which is a derivational sister of the affixal predicate. (93)  ^ V P ^ local spell-out  ^^^^^ V  location locatum  This prediction holds, as indicated in the sentences in (94), repeated from (81). In (94a), the affixal predicate Tu-cuu "contain" suffixes to the locatum argument, haPum "food". As shown in (94b), it is ungrammatical for this predicate to suffix to the location argument, gaPuuc "burden basket". (94)  a.  ha?umcu?is qa?uuc?i ha?um-cuu-?iis qa?uuc-?ii food-contain-3. IND burden.basket-DET There's food in the burden basket. (lit: "The burden basket contains food")  140  b.  *  qa?uuccu?is ha?um qaruuc-cuu-riis ha?um burden.basket-contain-3.IND food There's food in a burden basket.  Other locatum predicates show the same restriction. This is demonstrated below with Pu-kuxs "on the head", and Puu-tsit"on the surface of a liquid". In (95), the affixal predicate suffixes to the locatum argument &?icum+ "straw hat", but not the location argument ivc- "woman". In (96), the affixal predicate suffixes to cisx- "dirt", the locatum, and not to niisyak "pot", the location. (95)  a.  b.  (96)  a.  b.  ?ci?icumTUXs?is -fuucma &?icum+-uxs-?iis +uucma straw.hat-on.head-3.lND woman A woman is wearing a straw hat. (lit: " A woman is heading a straw hat") * +ucuxs?is 9c-ir.curn+ -hic-uxs-Piis 9ei?icum+ woman-on.head-3.IND straw.hat A woman is wearing a straw hat. ciisxsit?is ?uh niisyak?i cisx-si|[+L]-?iis ?uh niisyak-?ii dirt-on.surface.of.liquid-3.lND DEIC pot-DET There's dirt in the pot. Go dump it out! (lit: "The pot surfaces/contains dirt. Go dump it out!") * niisyaksit?is niisyak-sit-Tiis pot-in. water-3.lND There's dirt in a pot.  huqsaapci huq-saap-cii spill-CAUS-2SG.G0  cisxmis cisxmis dirt  The incorporation asymmetry derives from the properties of pairwise spell-out to PF. A n affixal predicate finds a host from its derivational sister.  As the derivational sister of a locatum  predicate, the locatum argument acts as a host for the affix. Location arguments are ineligible as a host because they are not derivational sisters to the affixal predicate. The incorporation asymmetry between the arguments of a locatum predicate extends predictably to the formation of w/z-questions and relative clauses. For locatum verbs such as Tu-uqfc "inside" or Pu-cuu "in a container", a w/z-word which corresponds to the locatum argument incorporates into the affixal predicate. In (97a), the locatum ?aqi- "what" incorporates into the affixal predicate Pu-uqk"inside". In (97b), the same locatum incorporates into the affixal predicate Pu-cuu "in a container". (97)  a.  ?aqiqkh ?aqi-aqfc-h what-inside-3.0 What's in the oven?  camaq?cyak?i camaq^yak-?ii oven-DET  141  b.  ?aqicuh ?ahkuu ?aqi-cuu-h ?ahkuu what-contain-3.Q DEIC What's in this? (lit: "What does this contain?")  Incorporation of a w/i-location is disallowed by a locatum verb.  As shown in (98a), it is  impossible for the location waayaq "which" to incoporate into the affixal predicate Pu-cuu "in a container". Instead, the location must remain unsuffixed to the locatum predicate, as in (98b). (98)  i.  * waayaqcuh waayaq-cuu-h which-in.container-3.0 Which one is my food in?  b.  ha?um?akqs ha?um-?ak-qs food-POSS-lSG.PS  waayaqh ?ucuu ha?um?akqs ?u-cuu waayaq-h ha?um-?ak-qs which-3.Q 0-in.container food-POSS-lSG.PS Which one is my food in? (lit: "Which contains my food?")  A parallel pattern is found with relative clauses formed with locatum predicates.  A  relative pronoun (yaq) which corresponds to the locatum argument incorporates into the affixal predicate Pu-kuxs"on the head" in (99a). As shown in (99b), a relative pronoun corresponding to the location argument does not incorporate into the locatum predicate. (99)  a.  ^u+cuu?ak?is ?ei?icum+ yaquxs?itq &uf-cuu-?ak-?iis 9ei?icum+ yaq-uxs-Tiitq good-RES-POSS-3.lND straw.hat REL-on.head-3.RL The staw hat that lady is wearing is very nice.  b.  yuk iiqsaksis +uucma?i yuk iiqsu-?ak-siis -hiucma-?ii y.sibling-POSS-1 SG.IND woman-DET The woman who is wearing a necklace w  w  haa +uucma?i haa -ruucma-Tii DEIC woman-DET (locatum- relative ) -  yaq?iitq nuutin+ yaq-?iitq nuut-win-t REL-3.RL necklace-on.neck is my younger sibling, (location-relative)  4.4.1.2 Subject agreement The subject agreement for a locatum predicate corroborates an analysis in which the location argument occurs higher than the locatum. As the highest argument, the location is predicted to trigger the subject agreement in AgrP. (100)  AgrP  location X V  locatum  '  142  This prediction holds: for locatum predicates, the person agreement corresponds to the location argument, not the locatum argument. A first person location argument is registered by the first person indicative inflection -siis, as illustrated in (101). (101)  a.  ?ucutumsis sackahs ?u-cutum-siis sackahs 0-on.side.of.head-1 SG.IND comb I've got a comb on the side of my head.  b.  Tuuwin-rsis nuutinum ?u?iihas Tuu-wip-f-siis nuutinum ?u-?iiha-s 0-on.neck-l SG.IND necklace 0-due.to-lSG.ABS Because I'm wearing a necklace I'm itchy.  c.  JeiTicum+uxssis &i?icum+-uxs-siis straw.hat-on.head-1 SG.IND I'm wearing a straw hat.  puuma+Tiha puuma+-?iha itchy-feeling  Here, the locatum argument is not registered by subject agreement.  4.4.1.3 Possessor raising In contexts with no possessor raising, either argument of a locatum predicate may be marked with possessive morphology and receive an interpretation in which the possessive-marked nominal is a possessum. In (102a), the location argument cupcupsumj'"sv/eater" is possessivemarked as a possessum, while in (102b) it is the locatum Pimtii "name" which acts as a possessum. (There is no restriction here as to whether the possessor Lucy must precede or follow the possessum: following (Ravinski 2005), I assume this is a case of 'scrambling'.) (102)  a.  ?u?a+?is fimtii 7u-?a4--?iis Timtii 0-on.flat.surface-3.lND name There is a name is on Lucy's sweater.  b.  ?u?a+?is ?u-?a+-?iis 0-on.flat.surface-3.lND Lucy's name is on a sweater. 1  Lucy Lucy Lucy  cupc'upsum-fuk?i c'upcupsum+-uk-?i sweater-POSS-3.PS (possessum= location)  ?imtii?ak?i Lucy cupcupsunvr imtii-?ak-?i Lucy cupcupsuirrr name-POSS-3.PS Lucy sweater (possessum= locatum) c  In possessor raising contexts, however, there is an asymmetry between the two arguments of the locatum predicate. Only the location argument of a locatum predicate may act as the source of possessor raising. This is reflected in the possessor raising sentences below, in which a possessive marker suffixes to the locatum predicate Pu-Pa-f-"on a flat surface". Here, the location cupcupsunrh "sweater" is obligatorily interpreted as the possessum. The sentence in (103a) shows the pattern with the locatum predicate suffixed to Pu-, and (103b) shows the same restriction on interpretation when the affixal predicate has suffixed to the locatum argument.  143  (103)  a.  7u?a-ruk?is virntii Lucy ?u-?a-fr-uk-?iis vimtii Lucy 0-on.flat.surface-POSS-3 .IND name Lucy = (i) There is a name on Lucy's sweater. (ii) Lucy's name is on a sweater.  cupcupsurrrf c'upc'upsum+ sweater (possessum = location) (possessum £ locatum)  b.  Yimtiqa-hik?is Lucy ?imtii-?a+-uk-?iis Lucy name-on.flat.surface-POSS-3.IND Lucy = (i) There is a name on Lucy's sweater. ^ (ii) Lucy's name is on a sweater.  cupc'upsum+ c'upcupsum+ sweater (possessum = location) (possessum ± locatum)  In (103), an interpretation of "Lucy's name is on a sweater" is unavailable. Thus, the locatum (fimtii'"name") proves to be ineligible as the source of possessor raising. It is predicted by the analysis that only a location argument of a locatum predicate should act as a source of possessor raising. Possessor raising in Nuu-chah-nulth is possible only out of subjects, and never objects. As the higher argument of a locatum predicate, the location takes on the role of a subject, and with this, the ability to serve as the source of possessor raising. In the next section, we turn to the discussion of location predicates, which I argue have an orientation of arguments which is the inverse of that of locatum predicates.  4.4.2 Location predicates This section presents diagnotics for the syntactic structure of location predicates such as Pu-k i w  "in" or Pu-kcaas "beside". Examples of each of these affixal predicates are given below. (104)  a.  ?uk i?is c'a?ak cax ac ?u-ci-?iis c'a?ak c'ax ac 0-in-3.IND water bucket The water is in the bucket. w  w  w  b.  c'ax acci?is ca?ak cax ac-ci-?iis ca?ak bucket-in-3.IND water The water is in the bucket. w  w  (105)  a.  ?ukcaas?is Kay ?u-caas-?iis Kay 0-beside-3.IND Kay man Kay's sitting beside a man.  b.  capxcaas?is Kay capx-caas-?iis Kay man-beside-3.iND Kay Kay's sitting beside a man.  cakup cakup  I analyse these predicates as taking a location argument as a complement, and a locatum as a specifier, as in (106).  144  (106) locatum location  4.4.2.1 Incorporation Only the location argument of a location predicate may incorporate; locata may not. This is illustrated in (107) for the predicate ?u-k i "in". In (107a), the predicate suffixes to the location w  c'ax ac "bucket". It is not possible for the locatum, c'a?ak "water", to serve as a host for the w  affixal location predicate, as indicated in (107b). (107)  a.  c'ax acci?is c'a?ak cax ac-ci-?iis c'a?ak bucket-in-3.IND water The water is in the bucket. w  w  b.  * ca?akci?is cax ac c'a?ak-ci-?iis cax ac water-in-3.IND bucket The water is in the bucket. w  w  This incorporation asymmetry is readily observed in w/i-questions, in which an incorporated whword references only a location, and not the locatum. In the following example, ?aqi "what" incorporates in the location predicate Pu-k i "in". The w/j-word corresponds only to the location w  argument, while the argument qaPuuc "basket" refers to the locatum. (108)  ?aqicih qa?uuc ?aqi-ci-h qa?uuc what-in-3.Q burden.basket = (i) What's the burden basket in? ^ (ii) What's in the burden basket?  ^ (wh = location) (wh± locatum)  When the locatum argument of a location predicate is w/i-questioned, the locatum is not incorporated. Compare the locatum-question in (109a) to the location-question in (109b): only in the latter case is the w/i-word waayaq "which" incorporated into the location predicate Pu-k i w  (109)  a.  b.  waayaqh qa?uucci waayaq-h qa?uuc-ci which-3.Q burden.basket-hi Which one is in the burden basket?  (wh = locatum)  waayaqcih waayaq-ci-h which-in-3.Q Which one has  (wh = location)  ha?um?akqs ha?um-?ak-qs food-POSS-lSG.PS my food in it?  145  Given the proposed analysis of the syntactic orientation of arguments of a location predicate, this incorporation asymmetry follows. A location argument is predicted to act as the host for an affixal location predicate because this argument is the derivational sister of the predicate. This is illustrated in the diagram in (110), which represents the argument structure of the sentence qaPuucciP/syarha?i "The salal berries are in a burden basket" (from (82b)). The location argument qa?uuc "burden basket" acts as host for the affixal predicate at local spell-out. (110)  VP local spell-out  DP V  -ci in  NP  yarha  qaPuuc salal burden basket  Spell-out induces a linearization of qaPuuc-ci "in a burden basket".  4.4.2.2 Subject agreement For location predicates, subject agreement matches the locatum argument. This is illustrated in the question-answer pair in (111), in which the respondent specifies his/her location with the location predicate ?u-k i "in". w  The predicate is inflected with the first person subject agreement  - . m s ( l S G . i N D ) , corresponding to the locatum argument of the predicate. (111)  a.  waasik waasi-k where-2SG.Q Where are you?  b.  ?uk isis cucsac ?u-ci-siis cucsac 0-in-l SG.IND tub I'm in the tub. w  This pattern of subject agreement follows if the locatum is the highest argument of the location predicate. In the diagram below, the locatum argument establishes a 'minimal link' to Agr. (112)  AgrP  146  4.4.2.3 Possessor raising For location predicates, locata act as the source of possessor raising. In the examples below, the possessum corresponds to the locatum, and not to the location. In (113), the locatum haPum "food" is treated as the possessum; in (114), the locatum maht'ii "house" is the possessum. (113)  (114)  waayaqcakhs ha?um waayaq-ci-?ak-hs ha?um which-in-POSS-lSG.Q food = (i) Which one has my food in it? 4- (ii) Which of mine has food in it?  (possessum = locatum) (possessum ^ location)  saantiquwascasuksis maht'ii saantiquwas-cas-uk-siis mahtii church-beside-POSS-1 SG.IND house = (i) My house is beside a church. 4- (ii) The house is beside my church.  (possessum = locatum) (possessum / location)  This restriction on possessor raising holds in spite of the fact that either a locatum or a location may act as a possessum in contexts with no possessor raising. The example in (115a) is a case of possessor raising in which only the locatum yama "salal berries" may be interpreted as the possessum. The example in (115b) is an unraised example showing yama "salal berries" as a possessive-marked possessum. The example in (115c) shows that there is no restriction on the location qaPuuc "burden basket" acting as possessum when no possessor raising occurs. In (115c), the location is marked with the possessive morpheme -uk (POSS), and no possessive marking appears on the predicate. (115)  a.  b.  c.  qa?uuccaksis yama qa?uuc-ci-?ak-siis yama basket-in-POSS-1 SG.IND salal.berries = (i) M y salal berries are in a burden basket. ^ (ii) The salal berries are in my burden basket.  (possessum = locatum) (possessum ^ location)  ?uk i?is qa?uuc ?u-ci-?iis qa?uuc 0-in-3.IND burden.basket M y salal berries are in a burden  (possessum = locatum)  w  yama?akqs yarna-?ak-qs salal-POSS-lSG.PS basket.  ?uk i?is qa?uucukqs ?u-ci-?iis qa?uuc-uk-qs 0-in-3 .IND burden.basket-POSS-1 SG.PS The salal berries are in my burden basket. w  yama yama salal  (possessum = location)  Thus, it is only in possessor raising contexts that the location argument of a location predicate is barred from acting as a possessum. This pattern supports an analysis in which the locatum argument is the subject of a location predicate, and not the object. extraction is possible only out of subjects in Nuu-chah-nulth.  147  Recall that possessor-  In the next section, we turn to ditransitive affixal predicates, which I analyse as transitivized extended unaccusatives.  4.5 Ditransitives There is evidence for two distinct classes of ditransitive affixal predicates in Nuu-chah-nulth. The first class, exemplified by Pu-yii "give", suffixes to a theme argument. In (116a), the affixal predicate Pu-yii "give" incorporates the theme taanaq- "money". The second type, exemplified by Pu-Piip "give to", suffixes to a goal argument. This pattern is illustrated in (116b), in which the affixal predicate Pu-Piip "give to" incorporates sut- (2SG). (116)  a.  taanaqayi?i taanaq-yii-'ii money-give-2SG.IMP>3 .OBJ Give mom money!  ?um?i ?um?i mother  b.  sufPiimitsis sut-fup-mit-siis 2SG-give.to-PST-l SG.IND I gave you money.  taana taana money  According to the analysis I propose, ditransitives are treated as extended unaccusatives which are abstractly transitivized when they are embedded within a vP 'layer'. (117)  vP  I propose that the two classes of ditransitives in Nuu-chah-nulth are distinguished according to the orientation of the arguments which belong to the extended unaccusative layer. Predicates such as Pu-yii "give" belong to a class I label 'locatum-type' ditransitives, while predicates such as Pu-Piip"g\ve to" belong to a 'location-type' class. The incorporation patterns of each of these types of ditransitives are illustrated in the following sections.  4.5.1 Locatum-type ditransitives This section considers the characteristics of locatum-type ditransitives, which I analyse as having a configuration of arguments as in (118).  i  148  (118)  theme According to the analysis, these predicates take a theme in their complement position, while a goal argument occupies specifier position of the embedded extended unaccusative relation. Predicates belonging to this class are ?u-yii"g\ve.", Puu-ks"ask for" and ?uu-pwin"to owe". (119)  (120)  (121)  a.  ?uyii?i ?u-ayii-'ii 0-give-2SG.IMP>3.OBJ Give mom money!  ?um?i taana ?um?i taana mother money  b.  taanaqayi?i taanaq-ayii-'ii money-give-2SG.IMP>3 .OBJ Give mom money!  ?um?i ?um?i mother  a.  ruukscii ?um?i ca?ak ?um?i ca?ak ?u-ks[+L]-cii mother water 0-ask.for-2SG.GO Go ask mother for water!  b.  c'aa?akscii ?um?i c'a?ak-ks[+L]-cii ?um?i water-ask.for-2SG.GO mother Go ask mother for water!  a.  ?uupwin?ick ?vju-pwin[+L]-?iick 0-OWe-2SG.IND You owe me money.  siya siya lSG.PRO  b.  taanaqapwin?ick taanaq-pwin[+L]-?iick money-owe-2SG.lND You owe me money.  siya siya lSG.PRO  taana taana money  For locatum-type ditransitive, incorporation of a goal is not possible. Only a theme argument may act as a host for the affixal predicate. This is illustrated by the sentences in (122). In (122a), the theme nacaaiyak "book" is the host for the affixal predicate Pu-yii "give". The recipient capx- "man" may not serve as the host, as shown in (122b). (122)  a.  nacaa+yakayimit?is Robin cakupri nacaa+yak-ayii-mit-?iis Robin cakup-?ii book-give-PST-3.lND Robin man-DET Robin gave a book to the man.  149  b.  * capxayimit?is Robin nacaa+yak capx-ayii-mit-Tiis Robin nacaa+yak man-give-PST-3.IND Robin book Robin gave a man a book.  This incorporation restriction is reflected in the formation of relative clauses. A relative pronoun corresponding to a goal argument may not incorporate into a locatum-type ditransitive. As shown in (123a), for goal relativizations, the relative pronoun yaq "who, which" occurs independently  of the  affixal  predicate  Pu-yii "give". The ungrammaticality  of (123b)  demonstrates that the relative pronoun cannot serve as the host for the affixal predicate ?u-yii "give" in a goal relativization. (123)  a.  yuk iiqsaksis haa yuk iiqsu-?ak-siis haa younger.sibling-POSS-1 SG.IND DEIC That is my younger sibling, who Robin  gave money to.  * yuk iiqsaksis haa yuk iiqsu-?ak-siis haa younger.sibling-POSS-1 SG.IND DEIC That is my younger sibling, who Robin  yaqayii?itq yaq-yii-Tiitq REL-give-3.RL gave money to.  w  w  b.  w  w  The  incorporation  asymmetry  for  yaq?iitq yaq-?iitq REL-3.RL  locatum-type  ?uyii ?u-yii  taana Robin taana Robin 0-give money Robin  ditransitives  Robin taana Robin taana Robin money follows  from  the  complement effect in PF incorporation. If a theme occupies complement position of a locatumtype ditransitive, then it is the derivational sister of the predicate, and is directed to act as the host for an affixal predicate at spell-out. Goal arguments, in contrast, do not have the privilege to form a local spell-out domain directly with the affixal predicate. This is illustrated for c'aaPakscii PumPi "Go ask mother for water!", from (120b): (124)  vP  local spell-out  DP V -ks ask for  N P PumPi c'aPak mother water  The reflex of this local spell-out is a linearization of c'aPak-ks.  4.5.2 Location-type ditransitive The second type of ditransitives has a configuration of internal arguments which is the inverse of that of locatum-type ditransitives.  150  (125)  theme  By this analysis, the goal argument acts as the complement of the ditransitive, while the theme is in specifier position. The predicate ?u-?iip "give to" is a location-type ditransitive. (126)  ?u?iimitsis ?u-?iip-mit-siis  suwa suwa you  0-give.to-PST-l SG.IND  I gave you money. b.  sufPiimitsis sut-Txip-mit-siis  taana taana money  taana taana money  you-give.to-PST-1 SG.IND  I gave you money. Location-type ditransitives incorporate a goal argument.  For example, the goals si- "me" and  sut- "you" incorporate into Pu-Piip "give to" in the examples below. (127)  a.  si?iipis siya-?iip-'iis me-give.to-2SG.IMP> 1 SG Give it to me! He owes me lots  ?ayapwin?a&ats taana ?aya-pwin-'afc-'at-s taana many-owe-TEMP-PAS-lSG.ABS money of money.  b.  sut?iimif?is Robin sut-?iip-mit-?iis Robin you-give-PST-3.lND Robin Robin gave you money.  taana taana money  In relative clauses formed with Pu-Piip "give to", a relative pronoun (yaq) corresponding to the goal incorporates into the affixal predicate. This is illustrated in (128). (128)  yuk iiqsaksis haa yuk iiqsu-?ak-siis haa y.sibling-POSS-lSG.IND DEIC That is my younger sibling, who Robin w  w  yaviipTitq Robin yaq-?iip-?iitq Robin REL-give-3.RL Robin gave money to.  taana taana money  For this location-type ditransitive, incorporation of a theme is not possible. As shown in (129), it is ungrammatical for the theme taanaq- "money" to act as the host for the predicate Pu-Piip "give to". (129) * taana?iimit?is taanaq-?iip-mit-?iis money-give.to-PST-3.IND Robin gave you money.  Robin suwa Robin suwa Robin you  151  The restriction that themes may not serve as the host for a location-type ditransitive follows from the analysis I have presented. PF Incorporation is sensitive to the argument structure of the affixal predicate. I have analysed location-type ditransitives as having a goal as a complement. The theme, in contrast, appears as an internal argument in specifier position. At local spell-out, the affixal predicate is linearized with respect to the goal argument, its derivational sister. This spell-out domain excludes the theme, and thus the possibility of the affixal predicate taking the theme as its host.  This is illustrated in (130) for the sentence  sutTiimitsis taana "I gave you money". Here, the goal sut- "you" is the derivational sister of the affixal predicate Pu-?iip"give to". (130)  vP  local spell-out V -?iip "give"  N P taana sut- "money" "you"  The reflex of spell-out for (130) is sut-?iip.  4.6 Outstanding issues This section discusses a usage of the affixal predicates discussed in this chapter which is in need of further research. A subset of affixal predicates show the ability to participate in a complex predicate strategy which I will refer to as 'serial verb affixation'. This predication construction 4  is illustrated in (131), in which the unaccusative affixal predicate Tu-wahswr "go out (of opening)" combines with the verbal host sa "crawl", forming the complex predicate sa-wahsui"crawl out of an opening". (131) sawahsuT?is histaqsik sa-wahsu-r-?iis histaq-sik crawl-go.out-3.IND from-PERF S/he crawled out from the cave.  "TavikTi ?avi?e-?ii cave-DET  These complex predicates disallow 'decomposition': the affixal predicate may not be separated from the verbal host. The example in (132a) shows serial verb affixation involving the affixal predicate Tu-TaaPatu"move down". In (132b), it is shown to be ungrammatical for Pu-to be used as the host for -?aa?atu "move down" when it is combined with the predicate mat- "fly". This is assumed to be a characteristic feature of the class of so-called 'restrictive' suffixes (Sapir and Swadesh 1939, Rose 1981, Davidson 2002; see Chapter 6).  4  152  (132)  a.  ,  b.  mariaa?atumit?is mat-9aa?atu-mit-?iis fly-move.down-PST-3 .IND A bird flew down. * ?u?aa?atumit?is ?u-yaa?atu-mit-?iis 0-move.down-PST-3.IND A bird flew down.  maamaati maamaati bird (cf. Davidson 2002: 198, ex. 292c) mataa mat-aa fly-CONT  maamaati maamaati bird  This lack of decompositionality stands in contrast to cases in which the affixal predicate takes a nominal complement. As shown in the 'noun incorporation' examples in (133), the predicate -TaaPatu "move down" can be separated from the nominal host t'at'uus"star" if the expletive host ?u- appears. (133)  a.  t'atuusWatumitPis t'at'uus-^aa?atu-mit-?iis star-move.down-PST-3.IND A star fell.  b.  7u Taa?atumif?is ?u-^aa?atu-mit-?iis 0-move.down-PST-3.lND A star fell. <  t'at'uus t'at'uus star  Unaccusative verbs of motion frequently occur in this serial verb construction, as do locatum predicates.  The examples in (134) show affixation of -ii?ik"%o  inside house" and  -TaaPatu "move down". The examples in (135) illustrate affixation of -cit'um "on side of the head" and-win+"ox\ the neck". (134) Serial verb affixation with unaccusatives a.  kamatqii?i?e?is kamatq-iffifc-Tiis run-go.inside.house-3.IND Tom ran inside.  Tom Tom Tom  b.  t'ftaa?ammit?is t'i-^aa?atu-mit-?iis fall-move.down-PST-3.IND A plane crashed.  matuk matuk plane  (135) Serial verb affixation with locatum predicates a.  9eikcit'um&sis Ken &ik-citum-u9e-siis Ken punch-side.of.head-PERF-1 SG.IND Ken I punched Ken on the side of the head.  b.  taaq in-r?anitsis taaq-win-r-Tat-mit-siis squeeze-on.neck-PAS-PST-1 SG.IND I was being choked by Ken. w  153  Ken Ken Ken  The morpheme -yaq- often intervenes between an initial predicate and the locative affixes - 7 - f "in the house", -'as "on ground", -'is "on beach".  In such contexts, the initial  predicate may be marked for aspect. This pattern is illustrated in (136) with the locative affix - ' / f " i n the house". The example in (136a) shows serial verb affixation without the use of the morpheme -yaq-. In (136b), the morpheme -yaq- appears and the initial predicate is marked with continuative aspect. (136)  a.  haptix?is Ken hapt-W-?iis Ken hide-in.house- 3 .IND Ken Ken is hiding in the house.  b.  haptaayaqi+?is hapt-aa-yaq-'i-r-?iis  Ken Ken Ken  hide-CONT- ?-in.house-3 .IND  Ken's been hiding in the house. The syntactic constraints on this process of serial verb affixation require additional research. The question should be tackled from two angles: (i) what restriction are there on which predicate may act as the initial element in the complex predicate?; (ii) what restrictions are there on which affixal predicates may act as the second element? This may shed light on the syntactic roles which arguments in these constructions play. Anderson (1985) discusses the role of arguments in the Northern Wakashan language Kwakw'ala, and argues that locative suffixes describe the position of subjects for an initial intransitive predicate, and the position of objects for an initial transitive predicate.  For the  Kwakw'ala complex predicate k 3 'l-i-r "lie down in the house", for example, Anderson (1985: w  31) argues that the locative suffix -iJ-"(on the floor) in the house" describes the position of the subject of the intransitive predicate k^'l "lie down". The restrictions on examples with the cognate affixal predicate - 7 f "(on the floor) in the house" in Nuu-chah-nulth are unclear at present, however. In the examples in (137), - 7 f " ( o n the floor) in the house" references the position of the subject of the initial predicate. (ITER)"  5  In (137a), it is the subject ofteaqteaqa"shout  which is described by - ' / f " ( o n the floor) in the house". In (137b), -'ii-"(on the floor)  in the house" references the position of the subject of hukqaa "stare ( C O N T ) " . (137)  5  a.  TaaqYaaqayaqi+Tis t'ariaak 7aaq-a[+R]-yaq-1+-?iis t'aria-?ak shout-lTER-?-in.house-3.lND child-POSS She is shouting at her child in the house. = (i) mother is inside, shouting at her child outside through an open window. * (ii) mother is outside, shouting at her child inside.  The morpheme -yaq- is obligatory in these examples. This matter requires future research.  154  b.  hufcqaayaqiF?is nuwiiqsak hu9eq-aa-yaq-'iF-?iis nuwiiqsu-?ak stare-CONT- ?-in.house-3. IND father-POSS She's staring at her father inside. = (i) she is inside and her dad is outside. * (ii) she is outside and her dad is inside.  In superficially similar examples, however, a different effect is found. In the examples in (138), - ' / f " ( o n the floor) in the house" does not reference the position of the subject of the initial predicate. In (138a), the position of the child who is spitting is irrelevant, so long as the spit ends up on the floor. A similar effect is shown in (138b), in which -'/¥• "(on the floor) in the house" references the aim of the bullets, and not the position of the shooter. (138)  a.  taaxtaaxayaqi-fuk?ick taax-a-yaq-'rt-uk-?iick  t'arWis tana-?is  spit-ITER-?-in.house-POSS-2SG.lND  child-DlM  Your child keeps spitting on the floor. = (i) child is on the floor inside = (ii) child is elsewhere b.  &9tumyi-ritwa?is Ken hiqaacip ?um?iiqsak &&im-'ii;-rnit-wa?is Ken hiqaa-cip ?um?iiqsu-?ak shoot-in.house-PST-3.QUOT Ken wreck-BEN mother-POSS Ken was shooting onto the floor. He wrecked his mother's house. = (i) Ken is inside. = (ii) Ken is elsewhere.  maht'ii maht'ii house  The restrictions on interpretation may ultimately be clarified by further research on the argument structure of the initial independent predicates involved (eg. faaqlaaqa "shout ( I T E R ) " and taaxtaaxa "spit ( I T E R ) " ) .  4.7 Conclusion In this chapter, the argument structure of affixal predicates was shown to be linked to a restricted set of attested incorporation patterns. I attributed this restricted pattern to a conditioning effect of the syntax at local spell-out, whereby only an argument introduced as a derivational sister to the affixal predicate is able to be spelled-out as the host for the affix. This induces the 'complement' effect of PF incorporation in Nuu-chah-nulth. This effect was demonstrated to hold across a variety of affixal predicates, which have a range of argument structures - from unaccusative to transitive, extended unaccusative to ditransitive. For all these predicates, only derivational sisters to the affix may act as hosts. The local spell-out hypothesis predicts the absence of unergative affixal predicates in the language. According to the analysis, affixal morphemes require linearization with respect to a host at the point of spell-out. Unergative predicates, which lack a pairing of a phonologically  155  contentful verb and a complement, fail to meet the binary requirement for affixation. affixation, two distinct elements are required at spell-out: a host, and an affix.  156  For  CHAPTER 5  Verbal complements of affixal predicates Through the unknown, unremembered gate... Is that which was the beginning... ~ T . S . Eliot Four Quartets no. 4  5.0 Introduction In the previous chapter, an analysis was presented of the argument structure of affixal predicates which take nominal complements. The suffixation pattern of these affixal predicates often results in 'noun incorporation' - although the process is not limited to targeting a single syntactic category. Rather, the choice of host is sensitive to string adjacency to the affixal predicate within its local spell-out domain. We now turn to the suffixation pattern of affixal predicates which take verbal complements, with particular consideration given to how the syntactic structure of these predicates interacts with restrictions on PF Incorporation. As we will see in this chapter, when an affixal predicate which takes a verbal complement suffixes to a host, the outcome may resemble 'verb incorporation'. In (1), for example, the affixal predicate -atah "try to" incorporates the verb suk- "reach" as its host. (1)  susuk atah?is kifcuuk suk-atah[+R]-?iis kiJoiuk reach-try-3.IND glass S/he's trying to grab for a glass. w  As in cases with 'noun incorporating' affixal predicates, however, the process of 'verb incorporation' does not strictly select for syntactic category: non-verbal elements may also serve as the host for these affixal predicate, as first discussed in Chapter 2.  By the string adjacency  effect, an affixal predicate which takes a verbal complement consistently incorporates whichever host is positioned at spell-out adjacent to it. When the complement of -mahsa "want to" contains just the verb wa+-sik "want to ( P E R F ) " , it is this verb which serves as the host for the affix, as in (2a). If a preverbal modifier, such as wityax "slowly" is present in the complement, however, this modifier takes on the role of host, as in (2b).  In contexts with preverbal modifiers,  incorporation of the verb is ruled out, as shown in (2c). Here, the affixal predicate consistently suffixes to the leftmost element of the complement, regardless of its syntactic structure.  157  (2)  a.  wa+si^mahsasis wafr-sifc-mahsa-siis go. home-PERF-want.to-1 SG.IND I want to go home.  b.  wityaxmahsasis waa-tsik wityax-mahsa-siis wa+-[+L]-si9e slow-want.to-1 SG.IND go.home-CONT-PERF I want to go home slowly.  c.  *  waa+sitanahsasis wityax wa-f-sifc[+L]-mahsa-siis wityax go.home-CONT-PERF-want.to-1 SG.IND slow I want to go home slowly.  The aim of this chapter is to investigate the syntactic structures which underlie the suffixation pattern of affixal predicates which take verbal complements. By 'verbal' complement, I refer to the syntactic frames projected by verbal heads, including any modifiers and functional projections associated with these verbs. The remainder of this chapter is organised in the following way. First, I argue in §5.1 that the class of 'verb incorporating' affixal predicates must be subdivided into categories of main predicates and auxiliary predicates, which show contrastive thematic properties. In §5.2, I present an analysis in which PF Incorporation is consistently barred across full clausal (CP) complements of affixal predicates. I relate this prohibition to an 'edge' effect. In §5.3,1 discuss a variety of morphological evidence that the complement in incorporation configurations is smaller than a Tense Phrase (TP). Syntactic evidence for a lack of clause-boundedness effects in these constructions is given in §5.4. The chapter concludes with §5.5.  5.1 Two classes of 'verb incorporating' affixal predicates This section presents evidence that affixal predicates which allow 'verb incorporation' are not a homogenous group, and must be divided into categories of main and auxiliary predicates. These classes show distinct behaviours with respect to complementation alternations and a same-subject requirement.  5.1.1 Affixal main predicates 'Verb-incorporating' affixal predicates vary in their ability to select a fully inflected complement. The class of verbs which I will term affixal main predicates alternate between an incorporation strategy, and a strategy in which the affixal predicate takes a full complement which is inflected for subject/mood agreement. In the latter case, the affixal predicate suffixes to the expletive morpheme ?u-, rather than to the embedded verb. The  158  (a)  examples  below  illustrate  the  incorporation strategy, while the (b) examples show /^-support in the case of an inflected complement. (3)  a.  tuuxtuux aS"i?eitsis t'aatrWis tux -a[+R]- ti^-mit-siis taatna-?is jump-ITER-find-PST-1 SG.IND children-DIM I came upon the children jumping.  b.  Tufifotsis ?in tuuxtuux ahuk . taatna?is Tu-fifc-mJt-siis ?in tux -a[+R]-huuk taatna-?is 0-find-PST-l SG.IND COMP jump-ITER-3.DEP children-DIM I came upon the children jumping. ( Pu-support)  w  w  <  (incorporation)  w  w  qaqahatu+itsis naniiq qah-aturt[+R]-mit-siis naniiq die-dream.of-PST-1 SG.IND grandparent I dreamt that grandparent passed away.  (4)  (incorporation)  Turuufu-Htsis ?in qahsiMiuk naniiq ?u-atot[+R]-mit-siis ?in qah-sik-huuk naniiq 0-dream.of-PST-l SG.IND COMP die-PERF-3.DEP grandparent I dreamt that grandparent passed away. (Pu-support) cuk ifccuk?is cu-k ifc-cuk-?iis wash-PERF-need-3.lND The potatoes need washing.  (5)  w  w  b.  qaawic?i qaawic-?ii potato-DET (incorporation)  ?ucuk?is cuk i?atquu qaawic-?i ?u-cuk-?iis cu-k i9e-?at-quu qaawic-?ii 0-need-3.IND wash-PERF-PAS-3.COND potato-DET It is best to wash the potatoes. (lit: "It is necessary that the potatoes be washed") w  w  ( Pu-support)  The inflected complement of matrix affixal predicates such as Pu-fi&"f'md" in (3) and PuPu-utui"dream o f in (4) is headed by the complementizer Pin (COMP) and is marked with dependent mood morphology.  For the the affixal predicate Pu-cuk "need" in (5), there is a conditional  complement which is inflected with conditional morphology -quu (3.COND).  5.1.2 Affixal auxiliary predicates The class of verbs which I label affixal auxiliary predicates are incompatible with an inflected complement, and rigidly select only the incorporation strategy.  The examples in (a) below  illustrate the incorporation strategy, while the (b-c) examples show the impermissibility of the auxiliary taking a fully inflected complement and being 'rescued' by /^/-support. Auxiliary predicates may never select an inflected complement, whether it is a dependent clause (6b-8b) or a conditional clause (6c-8c).  159  (6)  a.  c.  (7)  a.  (8)  wa+aakmahsasis wa+aak-mahsa-siis go.to-want.to-1 SG.IND I want to go to Victoria.  mituuni mituuni Victoria (incorporation)  Tumahsasis wa+aaksa Tin mituuni Tu-mahsa-siis Tin wa+aak-sa mituuni 0-want.to-l SG.IND COMP gO.tO-lSG.DEP Victoria I want to go to Victoria.  ( Pu-support)  Tumahsasis wa+aakquus Tu-mahsa-siis wa+aak-quus 0-want.to-l SG.IND gO.tO-lSG.COND I want to go to Victoria.  ( Pu-support)  mituuni mituuni Victoria  waaTicsinhiTis John waTic-sinhi[+L]-Tiis John sleep-try.to.stay-3.IND John John is trying to stay sleeping.  (incorporation)  TusinhiTis John Tin waTichuk Tu-sinhi[+L]-Tiis John Tin waTic-huuk 0-try.to.stay-3.lND John COMP sleep-3.DEP John is trying to stay sleeping.  (Pu-support)  TusinhiTis John waTicquu Tu-sinhi[+L]-Tiis John waTic-quu 0-try.to.stay-3.IND John sleep-3.C0ND John is trying to stay sleeping.  (Pu-support)  Tucaci9twitasitsis Tu-ca-ci^-witas-mit-siis 0-go.to-PERF-gonna-PST-1 SG.IND I was gonna go to Victoria.  (incorporation)  mituuni mituuni Victoria  b.  Tuwit'asmitsis Tin Tucaci&sa mituuni Tu-witas-mit-siis Tin Tu-ca-ci?c-sa mituuni 0-gonna-PST-l SG.IND COMP 0-go.to-PERF-lSG.DEP Victoria I was gonna go to Victoria. ( Pu-support)  c.  Tuwit'asmitsis Tucacifcquus Tu-witas-mit-siis Tu-ca-ci?e-quus 0-gonna-PST-l SG.IND 0-gO.tO-PERF-lSG.COND I was gonna go to Victoria.  mituuni mituuni Victoria (Pu-support)  Unlike main predicates, these auxiliary verbs have a same-subject restriction which requires that the notional subjects of the two predicates in the construction match. This is illustrated in the following examples with the affixal auxiliary predicate -qaath "claim", which incorporates the predicate Pu-uc"ovm"}  The ?u- in these examples occurs as a host to the affixal predicate A - » c " o w n " .  160  (9) Same subject requirement a.  ?uucqath?is Florence 7u-ic-qaath-?iis Florence 0-own-claim-3.lND Florence Florence claims she owns the shoes.  suwis suwis shoes  7  b.  * Tuucqathitsis Florence ?u-ic-qaath-mit-siis Florence 0-own-claim-PST-l SG.IND Florence I claimed Florence owns the shoes.  suwis suwis shoes  Here, the same-subject requirement determines that the "claimer" and the "owner" must corefer. In (9a) a 'match' occurs, while the ungrammatical (9b) illustrates a 'mismatch'.  5.1.3 Affixal auxiliaries are non-thematic I propose that the difference between main and auxiliary predicates in Nuu-chah-nulth is linked to their thematic properties.  Specifically, their behaviours fall out from an analysis in which  auxiliaries are functional elements, while main verbs are lexical (Cinque 2001, Wurmbrand 2004). Under this view, auxiliaries such as -qaath "claim" are non-thematic raising verbs which do not project a subject of their own. (10)  nunuukqath?is Florence nunuuk-qaath-?iis Florence sing-claim-3.lND Florence Florence is pretending to sing.  The 'same subject' effect of these auxiliaries arises when the subject of the main verb raises to specifier position of the auxiliary. In effect, the subject is 'shared' between the auxiliary and the main verb. The diagram in (11) shows that the subject of the main predicate, Florence, raises to the specifier of the auxiliary. (11) Auxiliaries as raising verbs FP  F  vP  Florence |  nunuuk sing In what follows, I will introduce evidence in favour of this analysis of auxiliaries as nonthematic verbs. In §5.1.3.1,1 show that auxiliaries have rigid complementation. This is followed  161  in §5.1.3.2 with an illustration of how transitivization allows an auxiliary to license an 'external' argument. In §5.1.3.2,1 discuss the lack of citation forms for these functional affixes.  5.1.3.1 Rigid vs. alternating complementation As functional morphemes, auxiliaries are predicted to show rigid complementation, in which they select only an infinitival complement (Cinque 2001). In (12), the infinitival complement of the auxiliary is represented as a vP. This vP lacks higher functional projections for tense (TP), agreement (AgrP) and complementizers (CP). (12) auxiliary predicate: strict selection of infinitival complement FP F  vP  I  A  auxiliary Main predicates, in contrast, are lexical verbs which permit lexically-specified alternations in complementation. For Nuu-chah-nulth, I propose that these main predicates may take either an infinitival or clausal complement. In (13a), the main verb selects an infinitival (vP) complement, while in (13b), it selects a 'full' CP complement. (13) main predicate: alternations in complement selection a.  ...  VP V  I main verb  b.  ...  vP  A  VP V  CP  I  A  main verb  5.1.3.2 Ability to license a 'mismatched' subject A distinction between the lexical and functional characteristics of affixal predicates has implications for these predicates' argument-taking abilities. As functional elements, auxiliary predicates are anticipated to be non-thematic (Cinque 2001, Wurmbrand 2004). This meshes with the observation that the valency-increasing morpheme - 'ap (TR) is employed in Nuu-chahnulth in contexts in which the same-subject requirement of auxiliaries is overridden (see also Rose 1981).  2  In the following (a) examples, - 'ap (TR) is used to permit a 'different subject'  reading. The examples in (b) show the illicit outcome without use of - 'ap (TR).  2  Rose (1981: 306) notes that - 'ap is used with Kyuquot desideratives "to indicate that the subject (the desirer) is not coreferential to the complement subject...". Evidence from Ahousaht indicates that - 'ap has a broad usage in which is it employed with the full range of auxiliary predicates (not simply desideratives). 162  (14)  ?uucqath?apsis Florence ?u-ic-qaath-'ap-siis Florence 0-own-claim-TR-3 .IND Florence I'm pretending Florence owns the shoes.  suwis suwis shoes  * ?uucqathsis Florence suwis Tu-ic-qaath-siis Florence suwis 0-own-claim-1 SG.IND Florence shoes I'm pretending Florence owns the shoes. (15)  miksitariahsapsis mifc-sifc-mahsa-'ap-siis rain-PERF-want.to-TR-1 SG.IND  I want it to rain. * miksi&mahsasis mifc-sifc-mahsa-siis rain-PERF-want.to-1 SG.IND  I want it to rain. (16)  a.  toiu+sinhapi'is John fcuF-sinhi- 'ap-?iis John good-try.to.stay-TR-3 .IND John John is trying to keep his shoes nice.  b.  9euu+sinhi?is John suwisuk?i 9eu+-sinhi-?iis John suwis-uk-?i shoes-POSS-3.PS good-try.to.stay-3.IND John John is trying to keep his shoes nice.  suwisuk?i suwis-uk-?i shoes-POSS-3.PS  Note that this behaviour contrasts with that of main predicates, which do not employ - 'ap (TR) in cases of subject mismatches. In (17a), the 'dreamer' and the 'one who dies' are disjoint, without use of -  'ap(TR).  (17)  a.  The example in (17b) shows that -  'ap(TR)  qaqahatu+itsis qah-atu+[+R]-mit-siis  suwa suwa  die-dream.of-PST-1 SG.IND  2SG  is impossible in this environment.  I dreamt that you died! b.  * qaqahatu+Pamitsis qah-atu+[+R]-'ap-mit-siis die-dream.of-TR-PST-1 SG.IND  suwa suwa 2SG  I dreamt that you died! Although - 'apis standardly labelled a "causative" (Sapir and Swadesh 1939), I contend that it has a more general transitivizing function than this term suggests (see also Rose 1981). With certain psych predicates (such as yaaPak "feeling", or Piihmis "be treasured"), usage of this morpheme converts an intransitive to a transitive. The examples in (18) show an intransitive usage of yaaPak "feeling" and Piihmis "be treasured", while the examples in (19) illustrate a transitive usage employing - 'ap. In (18a), the predicate is translated as "sore" and it takes a single argument (MsMnPatqs "my foot/feet"); in the transitivized (19a), it is translated as "love"  163  and takes two arguments. In (18b), the predicate is translated as "be treasured" and it takes a single propositional argument; in the transitivized case of (19b), the predicate is translated roughly as "won't part with" and it takes two arguments (?ux aapi "paddle" and naniiq w  "grandparent"). (18)  intransitive a.  r  yaa?ak?is yaa?ak-?iis  &is?ein?atqs &isfcin-?at-qs  feeling-3.lND  lower.leg-iPOSS-lSG.PS  M y feet are sore. b.  ?iihmis?is ?iihmis-?iis  k akuucna£at k akuuc-naak-'at  treasured-3.IND  grandchildren-have-PAS how-AUX-3.ABS let.go  w  w  q aaqh q aa-qh-0 w  w  +acyuu ruh Kay +acyuu ruh Kay DEIC Kay  Having grandchildren is something to be treasured, and yet Kay lets them go. (19) transitivized a.  yaa?a£apsis yaa?ak-'ap-siis feeling-TR-1 SG.IND  suwa suwa 2SG  I love you; I care for you. b.  ?iihmis?ap?is ?iihmis-'ap-?iis treasured-TR-3.IND Grandparent won't part  ?ux aapi naniiq ?ux aapi naniiq paddle grandparent with a paddle (ie. he's possessive or stingy about it). w  w  The transitivizing function of - 'ap (TR) is represented in the diagrams below. In (20a), the predicate yaa?ak "feel" is shown as an unaccusative, projecting a single, internal argument. In (20b), - 'ap(TR) transitivizes the predicate by introducing an external argument. (20)  a. intransitive  b. transitivized  VP V  DP  yaa?ak feel  vP  X\  9ds9dn?atqs my feet  For affixal auxiliary predicates, I propose that the morpheme -'ap (TR) serves a similar transitivizing function when it licenses a subject 'mismatch'.  Recall that affixal auxiliary  predicates show a same-subject restriction (as in 21a), and disallow mismatched subjects (as in 21b). The morpheme -ap must be used in cases in which the subjects do not match (as in 21c).  164  (21)  wa+aakmahsa?is Kay wa+aak-mahsa-?iis Kay Kay go.to-want.to-3.lND Kay wants to go to Victoria.  a.  mituuni mituuni Victoria (same subject)  /  Kay * wa+aakmahsasis Kay wa+aak-mahsa-siis go.to-want.to-1 S G . I N D Kay I want Kay to go to Victoria. wa+aakmahsapsis wa+aak-mahsa-'ap-siis g0.tO-want.to-TR-1 SG.IND  mituuni mituuni Victoria (different subjects) Kay Kay Kay  mituuni mituuni Victoria  I want Kay to go to Victoria.  (transitivization)  As noted in the earlier discussion, I propose that affixal auxiliaries are raising verbs which do not introduce an external argument (see Cinque 2001). The auxiliary 'shares' the thematic subject projected by the embedded verb. With the introduction of the transitivizer - 'ap, however, the auxiliary predicate inherits transitive syntax, and thus the same-subject restriction is overcome. In the same-subject case in (22a), the argument Kay is shared by the auxiliary -mahsa "want to" and the main predicate wa+aak"'go to". In (22b), in contrast, the affixal predicate -mahsa "want to" inherits a first person singular argument by virtue of the transitivizer. Therefore, (22b) allows a different subject than the one (Kay) used by the main predicate wa-fiaak"go to". (22)  a. same-subject  b. transitivization  FP  vP  wa+aak mituuni  want  go to Victoria wa+aak mituuni go to Victoria No derived transitivization occurs for affixal main predicates, because the status of these predicates as (thematic) lexical verbs allows them to introduce an external argument directly. Recall that there is no same-subject requirement for a main predicate such as ?u-fi&"come upon". (23)  wa?ic Tifotsis Ken wa?ic-9ik-mit-siis Ken sleep-come.upon-PST-1 SG.IND Ken I found Ken sleeping. <  165  In the following diagram, the affixal main predicate Pu-fifc "come upon" is represented as a thematic verb with inherent transitivity.  The transitivizer -'ap does not appear in this  construction. (24)  vP  v  VP  A waPic sleep The predicate Pii-fifa "find" takes a first person singular argument which is distinct from the one contained in its complement (Ken).  5.1.3.3 Citation forms Additional support for a split between functional and lexical characteristics of affixal predicates may be adduced from native speakers' judgements about the citation forms of these verbs. Affixal auxiliary predicates, which I analyse as functional elements, are unrecognizable to Nuuchah-nulth speakers without an incorporated verb, and therefore lack an independent citation form. Affixal main predicates, on the other hand, are pronounced in isolation as forms affixed to the expletive morpheme ?u-. Thus, while speakers of Nuu-chah-nulth reject 'words' such as *Puwit'as "gonna" or *Pu-qaath "claim", they freely accept PuPu-utui- "dream o f or Pri-fife "come upon". I suggest that this difference in isolatability is linked to the lexical status of these affixal 3  predicates.  5.1.4 Summary To recap, evidence has been presented for two distinct varieties of verb-incorporating affixal predicates. I have proposed that the behaviours of affixal main and auxiliary predicates are tied to the thematic properties which derive from their functional or lexical status. In the next section,  3  ^  An apparent exception is the affixal predicate -mahsa "want to", which I analyse as a auxiliary. Speakers accept ?u-mahsa as a 'word'. However, in this case it appears that the auxiliary -mahsa is homophonous with the non-auxiliary form Pu-mahsa "want", which takes a nominal complement, as in Pumahsasis haa c'apacPi "I want that canoe". The form Pu-mahsa is not accepted in contexts of verbal complementation (ex. 5). 166  I turn to an analysis of the suffixation patterns of main and affixal auxiliary predicates, which states that PF incorporation is only permitted in infinitival environments.  5.2 PF Incorporation occurs only in infinitival environments For affixal predicates that take verbal complements, incorporation arises only in monoclausal configurations in which the complement of an affixal predicate is an infinitival smaller than a Tense Phrase (TP).  PF incorporation is ruled out across biclausal structures in which the  embedded clause is demarcated by CP (cf. L i 1990). As first discussed in Chapter 2, this may be attributed to an 'edge' effect, in which CPs constitute self-contained units of the derivation. (25)  a. incorporation configuration  affixal predicate  vP  b. full C P complementation  affixal predicate/" C P AgrP AgrP  TP T  vP  In contexts in which an affixal predicate selects a full complement, a stranded-affix 'repair' strategy of suffixation to the expletive morpheme ?u- is implemented, rather than suffixation via incorporation. As was discussed in Chapter 2, a host may not be chosen from within the CP, because the CP is removed from the active workspace of the derivation. Let us sketch the analysis of the linearization stategy which occurs when the complement of an affixal main or auxiliary predicate is infinitival. When a main predicate such as ?u-fi& "come upon" incorporates a host, the choice of host is determined by string adjacency to the affixal predicate at spell-out. In (26a), the affixal predicate suffixes to the verb wa?ic "sleep", while in (26b), it suffixes to the modifier hacuk "(sleep) deeply", stranding the verb wa?ic "sleep". (26)  wa?icii9citsis Ken wa?ic-fi&-rnit-siis Ken sleep-come.upon-PST-1 SG.IND Ken I found Ken sleeping. hacukvifcitsis hacuk-fifc-mit-siis deeply-come.upon-PST-1 SG.IND I came upon Ken in a deep sleep.  167  wa?ic wa?ic sleep  Ken Ken Ken  The syntactic positions of the verb and the preverbal modifier are illustrated in (27). In (27a), -Ti& "come upon" takes a vP complement with an unmodified verb, waPic "sleep", as the embedded predicate. In (27b), this embedded predicate is modified by hacuk "(sleep) deeply". (27)  a. verb incorporation  b. adverbial incorporation  vP  vP  Ken  In each case, the embedded subject, Ken, appears as a right-linearized specifier of the embedded vP. In (27a), the embedded vP is linearized as <waPic, Ken>, while in (27b), the embedded vP is linearized as <hacuk waPic, Ken>. According to the local spell-out proposal, the linearization of the affixal predicate for (26/27a) and (26/27b) follows from the orderings of these embedded phrases. For (27a), it is the verb waPic "sleep" which is string adjacent to the affixal predicate -fik "come upon" when it attains spell-out. For (27b), in contrast, it is the modifier hacuk "(sleep) deeply" which has this privileged status of string adjacency to the affix.  Thus, verb  incorporation is induced for (27a), while adverbial incorporation is induced for (27b). A parallel process of linearization is proposed to occur for affixal auxiliary predicates, such as -wit'as "gonna". Just like affixal main predicates, these auxiliaries allow suffixation to either a verbal or adverbial host. The example in (28a) shows - wit'as "gonna" incorporating the verb waPic "sleep", while (28b) shows incorporation of the temporal adverbial caani "first". (28)  wa?icwitassis wa?ic-wit'as-siis sleep-gonna-1 SG.IND I'm gonna sleep.  As noted in Chapter 2, the temporal adverbal caani "first" is 'flexibly positioned': outside of incorporation contexts, it allows either a preverbal or postverbal positioning. Accordingly, verb incorporation is an alternative to the adverbial incorporation pattern of (28b).  4  168  4  b.  caaniwitassis caani-witas-siis first-gonna-1 SG.IND  wa?ic wa?ic sleep  I'm gonna sleep first (i.e. before doing something else) The  syntactic representations I assume for these examples are given in (29).  The vP  complements of the auxiliary -wit'as "gonna" contain a null first person singular pronominal as subject. In (29a), the verb wa?ic "sleep" is unmodified, while in (29b) the modifier caani"first" occupies a preverbal adjunct position. (29)  5  a. verb incorporation  b. adverbial incorporation  wa?ic sleep  ADVP A caani first  VP A wa?ic sleep  The preverbal position of the adverbial caani has consequences for the linearization of the affixal auxiliary -wit'as "gonna".  In (29a), the derivational sister of the auxiliary is equivalent to  <wa?i6>, while for (29b), the derivational sister is the linearized object <caani, wa?id>. At local spell-out of -wit'as "gonna", wa?ic "sleep" is selected as the host for (29a), while caani "first" takes on the role of host in (29b) due to its string adjacency to the affix. The proposed analysis states that while main and auxiliary predicates differ in their lexical properties, they share the ability to take an infinitival (vP) complement. It is in these infinitival environments that incorporation occurs. As will be shown in the following sections, a range of evidence exists for the analysis that the syntactic prerequisite for PF incorporation is an infinitival complement smaller than a Tense Phrase (TP). In the following sections, I consider three sets of arguments in favour of this analysis: the absence of embedded clausal morphology; 'restructuring' effects which indicate a lack of clause-boundedness; and finally, evidence for affixal main and auxiliary predicates being raising verbs.  I adopt an analysis in which the temporal adverbial caani "first" occupies the same preverbal adjunct position to the verb as a manner adverbial. This is not a necessary assumption, so long as each occur preverbally (see Chapter 2 for discussion). The syntactic position of adverbials in Nuu-chah-nulth is an understudied area in need of future research. 5  169  5.3 Absence of clausal morphology This section examines morphological evidence for the analysis that PF incorporation occurs in infinitival contexts. If the 'clauselet' in incorporation contexts is smaller than a Tense Phrase (TP), then this complement is predicted to lack specifications for tense, person/mood agreement and complementizers. If this reduced clause is equivalent to vP, then only aspectual morphology (occupying v) is predicted to be present. This section demonstrates that this prediction regarding the absence of higher-level clausal morphology holds. In incorporation contexts, the verbal host of an affixal auxiliary may only be inflected for aspect: independent specifications for tense, subject/mood agreement and complementation are systematically absent.  For example, the  verbal host wa-t--si&"go home ( P E R F ) " in (30) contains the perfective marking - s / ; t ( P E R F ) , but no tense marker -?aq&(F\n), subject/mood agreement -sa ( l S G . D E P ) or complementizer / f a ( C O M P ) . (30)  wa+si?c(*?aq^)(*sa)mahsasis waF-sifc-("?aqfc)-(*sa)-mahsa-siis  (*?in) (*?in)  gQ.home-PERF-(FUT)-(lSG.DEP)-want.to-lSG.IND  (COMP)  I want to go home. This follows from an analysis in which the complement of -mahsa "want to" is a reduced clause equivalent to vP, with perfective aspect shown to occupy v. (31)  6  FP  In the following sub-sections, I will present in detail the evidence for a lack of tense (§5.3.1), person/mood agreement (§5.3.2) and complementizers (§5.3.3) in the reduced clause.  5.3.1 No tense An incorporated verb is systematically free of tense morphology. Neither the past tense marker -mit (PST) nor the future tense marker -?aqfr(¥\]T) can occur with an incorporated verb. This is true for verbs hosting either an auxiliary or main predicate. In (32), with the affixal auxiliary predicate -qaath "claim", it is shown to be ungrammatical for the verbal host nunuuk "sing" to be  It is not a crucial assumption for aspect to occupy v. I propose that the reduced clause is smaller than a TP, but it is possible that projections intermediate to T P and vP exist which could house aspectual morphology. The syntactic representation of aspectual morphology in Nuu-chah-nulth is in need of further research, as this will shed light onto how much smaller than TP the 'clauselet' is.  170  inflected for the future tense -?aq&(FlTT). In (33), with the main verb Tuu-nalcuuh "observe", it is illicit for the host tuuxtuux a "jump (ITER)" to be marked for past tense -mit (PST). w  (32) Incorporation with auxiliaries: Absence of embedded tense a.  b.  nunuukqath?is Florence nunuuk-qaath-?i is Florence sing-dajm-3.lND Florence Florence is pretending to sing. *  ^  nunuuk?aq?eqath?is Florence nunuuk-Paqfo-qaath-riis Florence sing-FUT-claim-3.IND Florence Florence is pretending she's going to sing. cf.  nunuukqath?aq9t?is nunuuk-qaath-?aq9e-riis sing-claim-FUT-3.lND Florence will pretend to sing.  Florence Florence Florence  (33) Incorporation with main verbs: Absence of embedded tense a.  tuuxtuux anal«uuhitsis tuux -a[+R1-nalaiuh[+L]-mit-siis jump-ITER-observe-PST-1 SG.IND I observed you jumping.  suwa suwa you  * tuuxtaux amitnakuuhitsis tuux -a[+R]-mit-nalcuuh[+L]-mit-siis jump-lTER-PST-observe-PST-1 SG.IND I observed you jumping.  suwa suwa you  w  w  b.  w  w  This ban on embedded tense morphology stands in contrast to the availability of tense specification in full complements. In (34), the full complement Tin tuuxtuuxwamitsuk"that you were jumping" contains the past tense marker -mit (PST). (34)  ?uunal<uuhitsis ?u-nalcuuh[+L]-mit-siis 0-obserye-PST-l SG.IND I observed you jumping, (lit: "I  ?in tuuxtuux amitsuk ?in tuux*-a[+R]-mit-suuk COMP jump-ITER-PST-2SG.DEP observed that you were jumping") w  The lack of independent embedded tense morphology in incorporation contexts is predicted by an analysis in which the embedded clause does not project a Tense Phrase.  5.3.2 No person/mood agreement The complex predicate formed by incorporation is marked with only a single set of portmanteau person/mood inflection: the incorporated verb is not independendy inflected. In (35), it is shown for the auxiliary predicate -mahsa "want to" that the host is not inflected for the person/mood agreement -suuAr(2SG.DEP). In (36), this same restriction is demonstrated for the main verb ?u?ii-cik"hear, find out (PERF)".  171  (35) Incorporation with auxiliaries: Absence of embedded person/mood inflection a.  b.  ?acsi?emahsak ?ac-sifc-mahsa-k go .fi shi n g-PERF- want.to-2SG .0 Do you want to go fishing? * ?acsi&suukmahsak ?ac-sifc-suuk-mahsa-k goiishing-PERF-2SG.DEP-w^ntto-2SG.Q Do you want to go fishing?  (36) Incorporation with main verbs: Absence of embedded person/mood inflection a.  a.  taTi-FPiicifcitsis ta?i+-?ii-ci9t-mit-siis sick-hear-PERF-PST-1 SG.IND I heard you were sick. *  suwa suwa you  ta?i+suuk?iici?eitsis ta?i+-suuk-?n-ci9e-mit-siis sick-2SG,DEP-hear-PERF-PST-l SG.IND I heard you were sick.  (suwa) (suwa) (you)  An embedded verb is inflected within a full complement, however. In (37), the main verb ?u-?iicik "hear, find out (PERF)" takes a full complement, Tin taPi-huuk "that you were sick", which contains the dependent person/mood agreement -suu£(2SG.DEP). (37)  ?u?iici9eitsis ?u-?ji-ci9c-mit-siis 0-hear-PERF-PST-l SG.IND I heard that you were sick.  ?in taTi-rsuuk ?in ta?i+-suuk COMP sick-2SG.DEP  The lack of inflectional morphology on a verbal host falls out directly from an analysis in which the 'clauselet' containing the embedded verb does not project up to an Agr Phrase.  5.3.3 No complementizer There is no complementizer Pin in incorporation contexts.  In (38), it is shown that the  complementizer ?in (COMP) is obligatorily absent in contexts with the auxiliary -mahsa "want to".  In (39), the complementizer is shown to be illicit when the affixal main predicate ?uu-  nakuuh "observe" suffixes to the verbal host muu?ak a-ci9e"burn (PERF)". w  (38)  Incorporation with auxiliaries: Absence of complementizer a.  saapniqirrmahsah Kay saapniq-cii-r-mahsa-h Kay bread-make-want.to-3.0 Kay Does Kay want to make bread?  172  b.  (39)  * saapniqii+mahsah ?in Kay saapniq-cii-f-mahsa-h ?in Kay bread-make-want.to-3.0 C O M P Kay Does Kay want to make bread?  Incorporation with main verbs: Absence of complementizer a.  mu?ak aci?cfiakWhitsis w  mu?ak a-ci^-nakuuh[+Ll-mit-siis  +ayiipt -fayiipt  burn-PERF-observe-PST-1 SG.IND  leaves  w  I was watching leaves burn. b. J  * mu?ak aci^na]cuuhitsis mu?ak a-ci^-nakuuh[+L]-mit-siis w  w  burn-PERF-observe-PST-1 SG.IND  ?in ?in  +ayiipt +ayiipt  COMP  leaves  I was watching leaves burn. In contrast, within a full complement, complementizers are present. In (40), the affixal main predicate Tuu-mlcuuh "observe" is hosted by the expletive ?u. Here, the complementizer appears with the full complement. (40)  Tuunalcuuhitsis Pu-nalamhr+LI-mit-siis  ?in ?in  mu?ak aci9ehuk mu?ak a-ci?e-huuk  +ayiipt -tayiipt  0-observe-PST-1 SG.IND  COMP  burn-PERF-3.DEP  leaves  w  w  I was watching leaves burn. The absence of a complementizer in incorporation contexts follows if there is no Complement Phrase demarcating the boundary between the affixal predicate and its complement. According to the proposal, incorporation of a host is only possible when a C P 'edge' does not \ intervene between the affixal predicate and its potential host.  5.4 Lack of clause-boundedness effects A diagnostic property of infinitival constructions are transparency effects in which matrix and embedded constituents form a unitary domain for otherwise clause-bound processes (Cinque 2001). These so-called 'restructuring' phenomena (also known as 'clause union' or 'reanalysis') have been widely documented in languages, including Romance (see eg. Aissen and Perlmutter 1983, Rizzi 1982, Roberts 1997, Rooryck 2000, Cinque 2002) and Germanic (see eg. Evers 1975, Wurmbrand 2001). To illustrate an example, consider the optional process of clitic-climbing in Romance, which occurs in environments in which the matrix predicate belongs to a restricted class of 'restructuring' auxiliaries. In (41), the 'restructuring' verb is voudrais "would like". (41) Clitic-climbing (Cinque 2002: 620, ex. 4b & 5b) a.  Je voudrais y aller. I would-like there go I would like to go there.  (French)  173  b.  J'y  voudrais  aller.  l-there  would-like  go  I would like to go there.  (French)  With clitic-climbing, the pronominal clitic associated thematically with an infinitival verb is promoted to a position within the matrix clause. In (41b), the pronominal y "there" abandons its position within the embedded clause (y aller "go there") and surfaces instead as part of the pronominal proclitic string attached to voudrais "would like". In this case of 'restructuring', the infinitival does not constitute an independent cliticization domain distinct from the matrix verb. In this section, I provide evidence that cases of incorporation in Nuu-chah-nulth display a similar lack of clause-boundedness.  Thus, affixal  main and auxiliary predicates  are  'restructuring' verbs. The two cases I consider here are 'long' w/z-questions and 'long' possessor raising constructions which are formed with affixal predicates.  Following Cinque (2001) and  Wurmbrand (2001), I argue that these types of 'restructuring' effects in Nuu-chah-nulth are indicative of the reduced clausal status of the embedded complement: no clause-boundedness effects are found with incorporation because no clause boundary (i.e. CP) is projected between the matrix predicate and its infinitival complement.  x  5.4.1 'Long' w/i-movement As first noted by Davis and Sawai (2001), w/z-movement in Nuu-chah-nulth is strictly clausebound. This is shown in the example below, in which w/j-extraction out of the complement of the non-affixal predicate taaquk "believe" is disallowed.  The example in (42a) shows a  grammatical case with no w/i-extraction. In (42b), it is shown to be ungrammatical to question the subject of the embedded verb phrase kuuwii- c'apac "steal the canoe". (42)  a.  t'aaquk i?afc?is t'aaquk-ci&-'a&-?iis  John ?in kuuwi+ithuk Mary c'apac John ?in lcuuwi+-mit-huuk Mary c'apac believe-PERF-TEMP-3.IND John C O M P steal-PST-3.DEP Mary canoe John believes that Mary stole the canoe, (adapted from Davis and Sawai 2001: 133) w  b. * ?acaqh t'aaquk i?a?e John ?in kuuwi+ithuk c'apac ?acaq-h t'aaquk-ci?e-'ak John ?in kuuwi+-mit-huuk c'apac who-3.Q believe-PERF-TEMP John COMP steal-PST-3.DEP canoe Who does John believe stole the canoe? (adapted from Davis and Sawai 2001: 133) w  A different pattern is displayed by affixal predicates, however.  'Long' w/z-movement is  permitted out of the complement of an affixal auxiliary. In (43a), it is shown to be grammatical to question the subject of the verb phrase kuuwii- c'apac "steal the canoe", when the auxiliary -qaath "claim" is used. In (43b), a similarly grammatical example is shown with the auxiliary -mahsa "want to": here, the object of the predicate maakuk "buy" is questioned.  174  (43)  a.  ?acaqqafh?aph John kuuwi+ capac ?acaq-gaath-'ap-h John kuuwi+ capac who-claim-TR-3,Q John steal canoe Who does John claim stole the canoe?  b.  ?aaqici+mahsak ?aqi-ci+[+L]-mahsa-k what-AUX-want.to-2SG.O What do you wanna buy?  maakuk maakuk buy  The contrast between ungrammatical interclausal w/i-movement and grammatical 'long' whmovement across an infinitival can also be observed with affixal main predicates. W/i-movement is barred across a full clausal complement of an affixal main predicate, but 'long' wh-movement is allowed when the complement is infinitival. (44)  a. full complement: no long-range wh-movement *  ?aacina£uuhith ?acaq-ci+[+L]-na^uuh[+L]-mit-h  Florence Florence  ?in ?in  who-AUX-observe-PST-3.0  Florence  COMP  tuuxtuux amithuk tuux -a[+R]-mit-huuk w  w  jump-lTER-PST-3.DEP  Who was Florence watching jumping? b. infinitival complement: 'long' wh-movement ?aacinak\iuhith Florence ?acaq-ci+[+L]-nakuuh[+L]-mit-h Florence who-AUX-observe-PST-3.0 Florence Who was Florence watching jumping? This type of 'long'  tuuxtuux a tuux -a[+R] jump-lTER w  w  w/Vmovement with affixal predicates avoids true long-range  movement (crossing a CP), as indicated in the diagrams below. (45)  a. ungrammatrical long-range wh-extraction t'aaquk . [ p ?in . believe COMP C  [  A g r  -  *  p  -huuk  [ p T  /  -mit [  /  -  vP  ?acaq kuuwi+ capac] who steal canoe J  b. grammatical 'long' wh-extraction across an infinitival complement -qath claim +.  [  vP  ?acaq kuuwi-iwho steal  capac] canoe  J  W/i-movement in Nuu-chah-nulth does not cross a CP (Davis and Sawai 2001). However, in contexts in which an affixal predicate takes what I analyse to be an infinitival complement, movement out of the complement is unrestricted.  This transparency effect is predicted if the  infinitival complement of an affixal predicate lacks higher clausal projections such as CP.  175  5.4.2 'Long' possessor raising This section argues that 'long' possessor raising in Nuu-chah-nulth is indicative of the reduced clausal status of complements in incorporation contexts. As first described in §2.2.2.1, Nuuchah-nulth exhibits a process of possessor raising in which the possessive morpheme -uk/-?ak (POSS) appears on the predicate rather than the possessum (Davidson 2001, Ravinski 2005). In a possessor raising construction, the subject agreement of the clause matches the possessor. The example in (46a) shows no possessor raising: the possessive morpheme -u^r(POSS) suffixes to the possessum k aaPuuc "grandchild". In the possessor raising example (46b), the predicate taPi-P w  "sick" is suffixed by -uk (POSS), and the subject agreement is first person singular -siis (1 SG.IND), to match the features of the possessor. (46)  a.  ta?i-r?is k aa?uucukqs ta?i+-?iis k aa?uuc-uk-qs sick-3.lND grandchild-POSS-1 SG.PS M y grandchild is sick. w  w  ( unraised)  k aa?uuc taPrruksis ta?i+-uk-siis k aa?tiuc sick-POSS-1 SG.IND grandchild M y grandchild is sick. w  w  (possessor raising)  As discussed in Chapter 4, this pattern can be accounted for under an analysis in which the possessive morpheme -uk (POSS) licenses a position for a raised possessor (Ravinski 2005). In the unraised example (46a/47a), the possessor remains within the possessive NP. In the raised example (46b/47b), the possessor raises to Spec, PossP where, as the highest DP, it takes on the role of subject, and determines subject agreement. (47)  a. Unraised  b. Possessor raising  VP ta?i+  PossP NP  sick k aaPuuc ^ DP w  grandchild  / \  taPj'-f  lSG  sick kwaaPuuc grandchild  l-DP  There is evidence that possessor raising is only permitted within a clause. Possessor raising cannot cross a clause boundary marked by the complementizer Pin (COMP). The example in (48a) shows an unraised example for a sentence in which the matrix predicate iaqaak"believe" takes a complement containing the embedded nominal k aaPuucukqs "my grandchild". In (48b), w  intraclausal possessor raising occurs, in which the possessive morpheme -uir(POSS) is suffixed to  176  the embedded verb taPt'-f-''sick". This case can be contrasted with the ungrammatical example in (48c), in which possessor raising crosses the CP boundary of the embedded clause, to suffix the possessive morpheme -iMr(POSS) to the matrix predicate iaqaak"believe". (48)  a.  taqaaksis ?in taqaak-siis ?in believe-1 SG.IND COMP I believe my grandchild is  ta?i+huk ta?i+-huuk sick-3.DEP sick.  k aa?uucukqs k aa?uuc-uk-qs grandchild-POSS-1 SG.PS w  w  taqaaksis ?in ta?i+uksa taqaak-siis ?in ta?i+-uk-sa believe-1 SG.IND COMP sick-POSS-lSG.DEP I believe my grandchild is sick. c.  (unraised) k aa?uuc k aa?uuc grandchild (intraclausal possessor raising) w  w  taqaakuksis ?in ta?i+(uk)sa k aa?uuc t'aqaak-uk-siis ?in tari+-(uk)-sa k aa?uuc believe-POSS-1 SG.IND COMP sick-(POSS)-lSG.DEP grandchild I believe my grandchild is sick. (interclausal possessor raising) w  w  From the ungrammaticality of (48c), it is evident that possessor raising must be contained within a CP.  5.4.2.1 'Long' possessor raising with affixal auxiliaries Despite this restriction on interclausal possessor raising, 'long' possessor raising can be observed in environments in which an affixal predicate incorporates a host. As was initially described in Chapter 4, for affixal auxiliaries suffixed by -uk (POSS), either a 'short' possessor raising or 'long' possessor raising interpretation is generally possible with unaccusative hosts. In (49a), the affixal auxiliary -qaath "claim" is hosted by the unaccusative predicate sahyut "healthy". The possessive morpheme -uk (POSS) is suffixed to the predicate complex. In the 'short' possessor raising interpretation, "my grandparent" is construed as the notional subject of both -qaath "claim" and sahyut "healthy." In the 'long' possessor-raising interpretation, "my grandparent" is construed as the subject of sahyut "healthy", while the first person possessor of "grandparent" corresponds to the first person 'subject' of -qaath "claim". The example in (49b) indicates that when the argument of the unaccusative is inanimate ("bread"), only a 'long' possessor-raising interpretation is felicitious. (49)  a. sahyutqathuk itsis naniiqsu sahyut-qaath-uk-mit-siis naniiqsu healthy-claim-POSS-PST-1 SG.IND grandparent = (i) M y grandparent claimed to be well. = (ii) I claimed my grandparent was well. w  111  ('short' possessor raising) ('long' possessor raising)  b. pux aaqathuk it?ick sapnii ?ata?is wikyuu pux -aa-qaath-uk-mit-?iick sapnii ?ata-?iis wik-yuu rise-CONT-claim-POSS-PST-2SG.iND bread still-3.IND NEG-RES ^ (i) ! Your bread claimed it was rising, but it still hasn't, {'short' possessor raising) = (ii) You claimed your bread was rising, but it still hasn't. ('long' possessor raising) w  w  w  The syntactic prerequisite for 'long' possessor raising is feature matching between the notional subject of the auxiliary and of the possessor. For example, the sentence in (50) is ungrammatical if the 'claimer' and the possessor show a feature mismatch between a first person singular 'claimer' and and a second person singular 'possessor'. (50) * sahyutqathuk itsis sahyut-qaath-uk-mit-siis healthy-claim-POSS-PST-1 SG.IND I claimed your grandparent was well. w  naniiqs(ak?itk) naniiqsu(-?ak-?iitk) grandparent(-POSS-2SG.PS)  In contexts of feature mismatches, possessor raising is impossible: instead, - 'ap (TR) is employed to license the 'mismatch'. (51)  7  sahyutqath?amitsis sahyut-qaath-'ap-mit-siis healthy-claim-TR-PST-1 SG.IND I claimed your grandparent was  naniiqsak?itk naniiqsu-?ak-?iitk grandparent-POSS-2SG.PS well.  There is a contrast in Nuu-chah-nulth between illicit interclausal possessor raising and 'long' possessor raising in contexts with affixal auxiliaries. This contrast is represented in (52). Raising of a DP possessor across a C P boundary is ruled out in Nuu-chah-nulth, as indicated in (52a). According to the analysis that the 'clauselets' of affixal auxiliaries lack CPs, movement to a 'raised' position is predicted to be possible. In (52b), 'long' possessor raising does not cross a CP, because no CP is present.  This transitivization strategy is consistently available as an alternative to possessor raising with auxiliaries, whether or not possessor raising would be illicit in that context. In (i), transitivization expresses a meaning parallel to the ' l o n g ' possessor raising interpretation o f (ii). 7  (i)  ?uniiqath?ap?ick ?u-nii-q aath- 'ap-?iick  naniiqsak?itk naniiqsu-?ak-?iitk  ?atquu ?at-quu  wilciit wikiit  0-arrive-claim-TR-2SG.lND  grandparent-POSS-2SG.PS but-3.COND NEG  Y o u claimed your grandparent came, although she isn't here. (ii)  ?uniiqathuk?ick ?u-nii-q aath-uk-?i ick  naniiqsu naniiqsu-?ak-?iitk  ?atquu ?at-quu  wilciit wikiit  0-arrive-claim-POSS-2SG.lND grandparent-POSS-2SG.PS but-3.COND NEG = (i) Y o u r grandparent claimed she came, although she isn't here, ('short' possessor raising) - (ii) Y o u claimed your grandparent came, although she isn't here, ('long' possessor raising) ' L o n g ' possessor raising is only possible when a auxiliary takes an unaccusative verb as its complement. W h e n the embedded verb is transitive or unergative, use of - 'ap (TR) is mandatory for disjoint readings of the subject of the auxiliary and o f the embedded verb.  178  (52) Interclausal vs. 'long' possessor raising a. Ungrammatical interclausal movement  b. Grammatical 'long' movement  PossP  PossP  ta?i+ sick  NP  k aa?uuc grandchild  t p  w  D  Thus, the possibility of 'long' possessor raising in auxiliary environments can be viewed as evidence for the lack of clause-boundedness between the possessive argument of the embedded verb and the auxiliary.  5.4.2.1 No 'long' possessor raising with affixal main predicates Consider now the case of 'long' possessor raising with affixal main predicates.  A l l else being  equal, we would anticipate that 'long' possessor raising should be possible out of an infinitival complement of an affixal main predicate, as it is out of the complement of an auxiliary afffixal predicate. However, what is instead found is that possessor raising can cross neither the full nor infinitival complement of an affixal main predicate. The examples below show this restriction for the main predicate ?u-?ii-cifc "hear, find out ( P E R F ) " .  The examples in (53) show that  interclausal possessor raising (crossing a full complement clause) is impossible, as it is in all contexts in the language.  In (54), 'long' possessor-raising is shown to also be illicit.  The  example in (54a) shows a grammatical unraised example, while (54b) shows the ungrammatical outcome when the -uk (POSS) suffixes to the predicate complex. (53)  a. 1  ?u?iici?titwa?is ?u-?ii-ci9t;-mit-wa?is  Lucy ?in ta?i+ k aa?uucuk?i Lucy ?in ta?i+ k aa?uuc-uk-?i 0-hear-PERF-PST-3.QUOT Lucy C O M P sick grandchild-POSS-3.PS Lucy heard that her grandchild is sick. (unraised) w  w  179  b.  * Turiici^itukwaris ?u-?ii-ci?e-mit-uk-wa?is 0-hear-PERF-PST-POS S -3, Q U O T Lucy heard that her grandchild is sick.  (54)  a.  b.  ta?i+?iici9eit?is ta?i+-?ii-ci9e-mit-?iis sick-hear-PERF-PST-3 .IND Lucy heard that her grandchild  Lucy Lucy Lucy  Tin Tin  ta?i+ k aa?uuc ta?i+ k aa?uuc COMP sick grandchild (interclausal PR) w  w  Lucy k aa?uucuk?i Lucy k aa?uuc-uk-?i Lucy grandchild-POSS-3.PS is sick. (unraised) w  w  * ta?i+?iici9eituk7is ta?i+-?ii-ci9t-mit-uk-?iis sick-hear- PERF-PST-POSS-3.IND Lucy heard that her grandchild is sick.  Lucy Lucy Lucy  k aa?uu k aa?uuc grandchild ('long' possessor raising) w  w  A priori, the impermissibility of 'long' possessor raising out of an infinitival complement of an affixal main predicate is surprising given the transparency effects found with affixal auxiliary predicates. However, I propose that this behaviour is a direct consequence of the absence of a TP node in the complement of an affixal main predicate. Based on the proposal that nominative case is assigned by T (Chomsky 1995), the absence of TP predicts that an embedded subject is not case-licensed within an infinitival clause (Wurmbrand 2001). According to this proposal, main verbs such as 7u-?ii-ci9e (PERF) "hear, find out", Tu-'iik "come upon"and Puu-rialcuuh "observe" are analysable as verbs which govern promotion of an infinitival subject to a matrix object position. Under this analysis, an argument generated within the infinitival is unlicensed for nominative case, and so must raise to the matrix clause to receive abstract accusative case via the matrix v (Chomsky 1995). For a sentence such as in (55), this entails that k aa?uucukqs "my grandchild" is realised as an object of the matrix w  verb, although its genesis is in the subject position of the embedded 'clauselet'. (55)  tuuxtuux anal<uuhitsis tuux -a [+R]-nalaiuh-mit-siis  k aa?uucukqs k aa?uuc-uk-qs  jump-ITER-observe- P S T - l S G . l N D  grandchild-POSS-lSG.PS  w  w  w  w  I observed my grandchild jumping. This analysis is represented in (56). Here, the nominal phrase k aa?uucukqs "my grandchild" w  moves from its base position as an argument of the embedded verb tuuxtuux a "jump ( I T E R ) " to w  the matrix vP projection of the affixal main predicate Puu-nak'uuh "observe".  180  (56) Affixal main verb governing raising-to-object vP  k aa?uucukqs vP my grandchild w  The analysis treats these Nuu-chah-nulth main verbs as parallel to "Exceptional Case Marking" (ECM) verbs such as English "believe", which license accusative case assignment for the argument of an embedded infinitival. (57)  a.  I believe him to be sleeping.  b.  I believe that he is sleeping.  (ECM)  Once we assume the raising-to-object analysis for this main verb, the 'long' possessor raising facts directly fall out. Recall that 'long' possessor raising is not possible with affixal main predicates, as indicated by (58b). (58)  a.  8  tuuxmux anak\juhitsis tuux -a[+R1-na^uuh-mit-siis jump-lTER-obserye-PST-1 SG.IND I observed my grandchild jumping. w  w  b.  k aa?uucukqs k aa?uuc-uk-qs grandchild-POSS-1 SG.PS w  w  ( unraised)  tuux tuux anakuuhuk itsis k aa?uuc tuux -a r+RTnakuuh-uk-mit-siis k aa?uuc jump-lTER-observe- POSS-PST-1 SG.IND grandchild w  w  w  w  w  w  I observed my grandchild jumping.  ('long' possessor raising)  Also recall from Chapter 3 and 4 that possessor raising is possible only out of subjects in Nuuchah-nulth, not out of objects (Ravinski 2005). This is reflected in the interpretation of the possessor raised example of (59). Note that only the subject piispis "cat" is amenable to an interpretation as the possessum.  This restriction holds whether the embedded predicate is unaccusative (as with ta?i+ "sick" in 54b), or , unergative (as with tuux" -"]ump" in 58b).  181  (59)  hink a?iihuksis hin-k a?iih-uk-siis  piispis maamaati piispis maamaati  L0C-after-P0SS-3.IND  cat  w  w  bird  = (i) M y cat was after a bird. ± (ii) A cat was after my bird. The unavailability of 'long' possessor raising in the main verb environment is directly predicted by an analysis which states that the argument generated within the infinitival is behaving as an object, rather than a subject.  Under a raising-to-object analysis, the promoted argument is  predicted to be incompatible with possessor raising, since it fails to occupy a subject position. Possessor raising is inapplicable for objects in Nuu-chah-nulth. This analysis is represented in the diagram below. Here, only the subject of the matrix predicate is shown to allow possessor raising. (60) Raising-to-object verbs are incompatible with 'long' possessor raising PossP  -a  A  ITER  tuux  jump  5.5 Conclusion In summary, this chapter has argued that affixal predicates which take verbal complements fall into two general classes: affixal main predicates, and affixal auxiliary predicates.  The two  classes differ empirically in a variety of ways, including their ability to select a C P complement and to license a subject 'mismatch'. Moreover, only auxiliary affixal predicates permit 'long' possessor raising.  These behaviours were accounted for under an analysis in which main  182  predicates are lexical verbs which govern raising-to-object, while auxiliary predicates are nonthematic functional verbs which govern raising-to-subject.  9  I presented evidence for the infinitival nature of the complement of an affixal predicate in incorporation environments. The 'clauselet' behaves as a monoclausal configuration which lacks clausal demarcation between the affixal predicate and its complement. According to the analysis, the complement of an affixal predicate which governs incorporation is a phrase smaller than a TP. Thus, in these reduced clause contexts, no CP 'edge' is imposed between an affixal predicate and its host. This study expands the cross-linguistic typology of 'restructuring' effects.  The  manifestation of these effects differ considerably across languages, although they share the common property of a lack of clausal demarcation between matrix and embedded constituents (Wurmbrand 2001, Cinque 2001). For Romance languages, primary examples of 'restructuring' effects are clitic-climbing, long NP-movement, and auxiliary selection (see Rizzi 1982), while Germanic 'restructuring' behaviours include long object movement, long-distance scrambling, and verb raising (see Wurmbrand 2001). This study demonstrates that in Nuu-chah-nulth, 'restructuring' effects include incorporation, 'long' wh-movement and 'long' possessor-raising. These 'restructuring' effects were previously unrecognized in the Nuu-chah-nulth literature.  Additional research is necessary to determine if a wider range of raising configurations exist. For instance, preliminary research suggests that the affixal predicate ?u-cuk "need" may govern subject-tosubject raising.  183  CHAPTER 6  Implications We shall not cease from exploration and the end of all our exploring will be to arrive where we started and know the place for the first time. ~ T . S . Eliot Four Quartets no. 4  6.0 Introduction This concluding chapter first summarizes the effects of local spell-out in Nuu-chah-nulth, and then moves on to discussion of key theoretical and typological implications of the analysis. The chapter closes with concluding remarks.  6.1 Summary This dissertation has proposed that the positioning of affixal predicates in Nuu-chah-nulth is a reflex of the need to linearize these bound morphemes. I have advanced an argument for the way in which syntactic objects are mapped to phonological linearization. According to the proposal, the linearization of syntactic constructs is established incrementally over the course of the derivation, in minimal domains equivalent to a single cycle of Merge. (1) Local spell-out hypothesis 7t <  local spell-out local spell-out  8  y a  <£.  local spell-out  p  The local spell-out hypothesis determines that an affixal predicate finds a host from within its derivational sister.  This derivational sister is a linearized object at spell-out, leading to the  restriction that suffixation operates on the basis of string adjacency. The consequence of local spell-out of affixal predicates in Nuu-chah-nulth is a process termed PF Incorporation. Through PF Incorporation, an affixal predicate suffixes to a host, in a process unselective for syntactic category, and insensitive to the syntactic structure of the derivational sister.  The syntax and the linearization mechanism are not fully independent,  however. I have proposed that the syntax plays a conditioning role in the linearization process,  184  through the composition of local spell-out domains. For affixal predicates which take nominal complements, I presented evidence that the argument structure of the affixal predicate restricts the choice of host for the affix. Whether an affixal predicate is unaccusative, extended unaccusative, transitive or ditransitive, the position in which arguments are introduced has an influence on the pattern of incorporation. In each case, it is the argument which is introduced as the derivational sister to the affixal predicate which may serve as host. Higher arguments are ineligible as hosts. (2) Nominal complements of affixal predicates a. unaccusative local spell-out  ^  b. extended unaccusative VP  V  local spell-out d. ditransitive  c. transitive  local spell-out local spell-out  Moreover, the syntax was shown to condition spell-out of affixal predicates through the creation of 'self-contained' derivational units. According to the DP/CP 'edge' hypothesis, hosts for an affixal predicate are never selected from within a DP, because a DP is an independent phonological domain. Instead, in these contexts, the expletive Pu-is inserted to host the affix. For affixal predicates which take verbal complements, a similar analysis for the conditioning effect of the syntax was given. I argued that the inventory of affixal predicates in Nuu-chah-nulth includes both main and auxiliary verbs, which permit incorporation of a host chosen from its complement. The incorporation pattern occurs with infinitival (vP) complements, and not with C P complements.  By the DP/CP 'edge' hypothesis, affixation is barred from  crossing a CP. (3) Verbal complements of affixal predicates a. auxiliary predicate local spell-out  b. main predicate local spell-out  FP vP  A 185  .j> V P  6.2 Theoretical implications This section presents three theoretical implications of the local spell-out hypothesis. In §6.2.1, I distinguish the characteristics of PF operations from syntactic ones. Next, §6.2.2 discusses how PF Incorporation in Nuu-chah-nulth fits in to the broader question of the locus of head movement operations in the grammar.  Finally, §6.2.3 illustrates how the linearization effected by local  spell-out necessarily induces an 'outside-in' dependency.  6.2.1 On the nature of PF operations Under the analysis I have proposed, spell-out to PF manipulates the ordering of syntatic elements when linearization is induced.  This section argues explicitly against the notion that this  characterization of the derivational capacity of PF amounts to a 'second syntax' or to a 'hybrid' branch sharing both syntactic and phonological sensitivies (cf. Embick and Noyer 2001, Rivero 1999). The properties I have attributed to syntactic and PF operations are distinct. The syntax, under my analysis, plays a structure-building role in which linearization is irrelevant: syntactic constructs are inherently unordered. In contrast, at spell-out to PF, the structures of the syntax are inconsequential, while relationships based on string adjacency take on import. When the syntax is spelled-out to PF, the syntax indirectly conditions the input to the PF branch. The Nuu-chah-nulth facts indicate two ways in which syntax has a conditioning effect: the structure-building component is responsible for the creation of local spell-out domains (derivational sisters), as well as the creation of computationally independent 'edged' constituents. (4) Conditioning effects of the syntax (i) determining derivational sisterhood (ii) creating constituents (DP, CP) which correspond to computationally independent domains The first effect is responsible for the 'complement' restriction on Nuu-chah-nulth incorporation. In Nuu-chah-nulth, the host for an affixal predicate is necessarily selected from the derivational sister of the affix. Thus, only elements from the complement of the affixal predicate will qualify as hosts. The second effect determines that incorporation is impossible in Nuu-chah-nulth across a DP or CP. Thus, when an affixal predicate takes a DP or C P complement, the host for the affixal predicate must be the expletive ?u- In Chapter 2, I suggested that the opacity effect associated with DP and C P in Nuu-chah-nulth may arise from the inert status of these nodes as designated spell-out domains. Unlike in languages with phasal spell-out, D P and C P in Nuuchah-nulth fail to act as 'active' spell-out triggers, rendering them derivationally impermeable.  186  6.2.2 The head movement question In the wake of Chomsky's (1995, 2001) suggestion that head movement may be best analysed as a PF phenomenon, much attention has been paid to theoretical motivations for eliminating this operation from the narrow syntax. A well-noted objection to a standard syntactic treatment of head movement (eg. Baker 1988, Chomsky 1993) is its countercyclic violation of the Extension Condition, in which head movement builds the tree at a non-root node (Chomsky 2001, among others). According to a traditional syntactic treatment of head movement, a head X is moved to adjoin to an immediately c-commanding node Y , as in (5a).  In (5a), the Y node at which  structure-building is effected is not the root node. In cyclic movement, structure-building occurs at the root, as in (5b). (5)  a. countercyclic movement ^J(P)^ Y X  t  x  Y  Y  t  x  Countercyclicity entails that the moved element does not c-command its 'trace', under standard definitions of c-command. An additional problematic aspect of a traditional analysis of syntactic head movement is the stipulative nature of the Uniformity Condition on the phrasal status of chains (Suranyi 2003, Matushansky to appear). (6) Uniformity Condition (Chomsky 1995: 253) A chain is uniform with regard to phrase structure status. However, a variety of analyses have indicated that these problems posed by head movement within a Minimalist syntax may be reconciled through an adjustment in the syntactic machinery of head movement (Donati 2003, Suranyi 2003, Matushansky to appear). For example, Matushansky (to appear) argues that phrasal movement and head movement have identical landing sites, in which both extend the projection at the root.  By this analysis, the issue of  apparent countercyclicity evaporates. The residue of this theoretical debate on the admissibility of head movement as a syntactic operation is the question of empirical support for the notion that head movement occurs in PF (Boeckx and Stjepanovic 2001).  In this light, the empirical basis for diagnosing the  symptoms of a PF operation takes on considerable theoretical import (see also Rivero 1999, Embick and Noyer 2001). Specifically, if head movement is a PF operation, then what features should it be expected to have? This examination of incorporation data from Nuu-chah-nulth expands the empirical coverage of the 'head movement question'.  187  As I have shown,  incorporation in Nuu-chah-nulth has the hallmark properties of a post-syntactic, PF process. The phenomenon of PF Incorporation in Nuu-chah-nulth demonstrates adherence to string adjacency, and shows an insensitivity to syntactic category and constituency, and as well as an absence of L F effects. Moreover, the 'edge' effects associated with DP and C P constituents in Nuu-chah-nulth provide novel support for the computational independence of these phrases at a phonological level (cf. Chomsky 2001, Matushansky 2005). In what follows, I make explicit the claim that while the 'head movement' operation of affixal predicates in Nuu-chah-nulth is strictly phonological, other types of head movement show contrasting properties. In §6.2.2.1, I summarize the argument that the linearization mechanism for Nuu-chah-nulth affixes is achieved at spell-out to PF.  In §6.2.2.2, this linearization  mechanism is contrasted with syntactic head movement.  6.2.2.1 Linearization is purely phonological In a sense, the placement of affixal predicates in Nuu-chah-nulth is purely phonological. Although syntactic composition conditions the input to the linearization mechanism at local spellout, the syntax plays no role in the means by which an affix is attached with its host. As I have argued, an affixal predicate suffixes to its host as a reflex of the spell-out mechanism. For a syntactic construct such as (7), the suffix - a is the linearized object P-a. (7)  ^ -a  \ p  A characteristic of the linearization of Nuu-chah-nulth affixal predicates is that it is conditioned by the 'base' position of affixes and their arguments, via aggressive interpretation. That is, as soon as an affixal predicate is merged into the derivation, its linearization is determined: the affixal predicate is hosted by a string adjacent element from its derivational sister. Under the simplest assumptions, it follows from this analysis that subsequent movement in the derivation, of either the affix or its derivational sister, should not affect this relationship between the affix and its host.  This prediction is upheld in Nuu-chah-nulth. For example, when the  nominal object of an affixal predicate is passivized, the 'complement' effect in affixation remains. Whether an affixal predicate is active or passive, the affix suffixes to a host from its derivational sister, its complement.' The examples in (8) show this pattern for the transitive affixal predicate ?u-?nc "consume".  The active sentence in (8a) shows the affixal predicate suffixing to the  nominal Jc ag "spawned herring eggs", which the predicate takes as its syntactic object (as argued w  This claim directly contradicts Waldie's (2004) assertion that affixal predicates with nominal hosts cannot be passivized in the Ahousaht dialect of Nuu-chah-nulth. The Ahousaht speakers who I have worked with consistently allow passivization in appropriate discourse contexts, with semantically appropriate predicates. 1  188  in Chapter 4). In (8b), the predicate has been passivized: this is indicated morphologically by the addition of the suffix - 'at (PAS). What is relevant is that passivization has no consequence for the choice of host for the affixal predicate.  In (8b), the host for the affixal predicate is k' ag w  "spawned herring eggs", just as it is in the active case of (8a). (8)  a.  b.  k a9iic?is k aa?uuc k aq-'iic-?iis k aa?uuc s.h.eggs-consume-3.lND grandchild Grandchild is eating spawned herring eggs.  (active)  k^aTiicck at?is ?uush?at ^ aq- iic-ck i-'at-?iis Tuus-qh-'at s.h.eggs-consume-EVID-PAS-3.IND some-do.by-PAS Spawned herring eggs must have been eaten by someone.  (passive)  w  w  w  w  w  w  ,  w  According to a syntactic treatment of Nuu-chah-nulth passivization (Kim 2001), the object of the passivized predicate is promoted to a surface subject position. In (9), the passivized object k aq w  "spawned herring eggs" moves to specifier position of the vP projection which - 'a^(PAS) heads. (9)  vP  EvidP  -'at ck i w  VP 'He consume  k aq s.h.eggs w  Crucially, the syntactic movement of lc aqc\ots not affect its spell-out position hosting the affixal w  predicate. Furthermore, the relationship between an affixal predicate and its host is not affected by other types of word order manipulations which disrupt the surface adjacency between the predicate and its syntactic complement. This effect may be observed in contexts in which the complement of the affixal predicate contains multiple words, such as when the object contains a modifier.  In the following examples, the modified nominal cus(uk) suwis "new shoes" is the  complement of the affixal predicate PuPu-ciqa "lose". As shown in the examples in (10), in which the affixal predicate is hosted by the expletive ?u-, the subject Louis and the object cusuk suwis "new shoes" show a variable word order. (As described in Chapter 3, subjects and objects often show variable word order when the object is inanimate and indefinite.) The crucial examples are given in (11). In (11), the affixal predicate PuPu-ciqa "lose" is hosted by cus- "new", which originates syntactically in object position as the modifier of the nominal suwis "shoes". In (11a),  189  the subject Louis precedes the object 'remnant' suwis "shoes". In ( l i b ) , Louis follows suwis "shoes". (10)  (11)  L.  ?u?uciqa?is ?u-ciga[+R]-?iis 0-lose-3.IND Louis lost a brand new  Louis cusuk Louis cus-uk Louis new-DUR pair of shoes.  suwis suwis shoes  i.  ?u?uciqa?is ?u-ciqa[+R]-?iis 0-lose-3.IND Louis lost a brand new  cusuk cus-uk new-DUR pair of shoes.  Louis Louis Louis  a.  cucusciqa?is cus-ciga[+R]-?iis new-lose-3.lND Louis lost a brand new  Louis suwis Louis suwis Louis shoes pair of shoes.  cucusciqa?is cus-ciga[+R]-?iis new-lose-3.IND Louis lost a brand new  suwis Louis suwis Louis shoes Louis pair of shoes.  b.  suwis suwis shoes  What the example in (11a) suggests is that surface position of the predicate and its object is irrelevant for the affixation mechanism. Instead, affixation is determined at the earliest possible stage of the derivation, with the object adjacent to the predicate in complement position. For the syntactic structure in (12), local spell-out determines that the affixal predicate -ciqa "lose" selects cus "new" as its host because of the string adjacency the two share. (12)  local spell-out  N suwis shoes When movement derives an object-final order, this affixation relationship is unaffected.  In the  diagram in (12), the object-final word order of (11a) is shown to be derived by movement of the object to a right-linearized specifier position. Under this analysis, I assume that it is the bottom 'copy' of the moved element cus "new" which is pronounced in Nuu-chah-nulth. Note that it is otherwise impossible to separate the modifier cus(uk) "new ( D U R ) " from the nominal suwis "shoes", as shown in (13).  190  (13) * TuTuciqaPis ?u-ciqa[+R]-?iis 0-lose-3.IND Louis lost a brand new  cusuk cus-uk new-DUR pair of shoes.  Louis Louis Louis  suwis suwis shoes  The ungrammatically of such an example suggests that cus(uk) suwis "new shoes" moves only as a unitary constituent in Nuu-chah-nulth, lending support to the analysis in (12) that it is the NP which moves rightward, and not simply the nominal head suwis "shoes". In summary, syntactic movement appears to have no consequence for the linearization mechanism of affixal predicates in Nuu-chah-nulth.  The host for an affixal predicate is  determined by local spell-out of the affix at the stage in the derivation in which it is first introduced.  Linearization of the affix and its host is not disturbed by subsequent syntactic  movement of the syntactic category containing the host. Thus, syntactic manipulations play no role in the linearization of Nuu-chah-nulth affixal predicates, beyond the irreducible conditioning effect of the syntax in forming the primary local spell-out domain of the affixal predicate. In the next section, I argue that the linearization mechanism of Nuu-chah-nulth affixal predicates may be contrasted with syntactic head movement, which is achieved by an interplay between syntactic movement and local spell-out.  6.2.2.2 Syntactic head movement is more complex As described in Chapter 2, PF Incorporation in Nuu-chah-nulth is sensitive to linear adjacency: an affixal predicate suffixes to a string adjacent element from its derivational sister. In this, Nuuchah-nulth differs from those languages in which incorporation operates on the basis of hierarchical adjacency.  As noted in Chapter 2, the contrast between sensitivity to linear  adjacency and hierarchical adjacency is readily observed in environments in which the complement of the 'incorporating' verb is modified. With PF incorporation, it is a linearly adjacent modifier which is targeted for suffixation by an affixal predicate. In (14), the affixal predicate Pu-Piic "consume" suffixes to the modifier AaPum "tasty", stranding the nominal head Paapinis "apple". (14)  ha?um?ic?is?a+ ?aapinis ha?um- iic-?iis-?aF ?aapinis tasty-consume-3 . I N D - P L apples They are eating delicious apples. ,  In contrast, in syntactic incorporation, a modifier is necessarily inaccessible for incorporation (Baker 2003). As argued in Chapter 2, the prohibition on syntactic adjective incorporation is a consequence of the Head Movement Constraint, which determines that movement targets the head of a complement.  191  With respect to adjective incorporation, Nuu-chah-nulth may be contrasted with Mohawk (Iroquoian). As Mithun (1984: 870) describes, when a verb in Mohawk suffixes to a noun, it may strand a modifier.  2  In (15), the verb u:ni "make" suffixes to -akya'tawi'tsher- "dress", stranding  the modifier kanekwarunyu "polka-dotted". (15) Mohawk (Mithun 1984: 870, ex. 106) kanekwarunyu wa' -k-akya' tawi' tsher-u:ni it.dotted.DIST PST-I-dress-make I made a polka-dotted dress. In what follows, I suggest that the difference between Nuu-chah-nulth and Mohawk amounts to whether affixation is achieved purely through local spell-out, or via a more complex derivation. This section hypothesizes that syntactic head movement arises through an interaction between syntax and local spell-out. According to this notion, syntactic head movement reflects the local spell-out of a head not in its base position, but in the position it has attained via a local form of syntactic movement. This falls in line with other work which proposes that syntactic head movement is a multi-step process with individual components of syntactic movement and affixation (Donati 2003, Matushansky to appear). I adopt Matushansky's (to appear) proposal that head movement is a cyclic operation which remerges a head at the root node. For Matushansky (to appear), the prerequisite for this local movement is an A G R E E relation equivalent to c-selection. In (16a), the head X selects the category Y as its complement. In (16b), Y is (re)Merged at the root node X(P). (16)  a. c-selection  b. movement (remerge)  WP AGREE  t  Y  WP  As Matushansky argues, in syntactic head movement, the movement in (16b) is followed up by an affixation process which attaches the moved head X to the head Y . A syntactic analysis of Mohawk incorporation (Baker 1988) may thus be implemented within the Minimalist framework in which remerge is effected at the root node. This analysis translates into the following movement operation for noun incorporation which strands a modifier: Mithun (1984) explicitly argues against a syntactic treatment of noun incorporation. Baker (1988, 2003) in assuming that Mohawk incorporation is syntactic. 2  192  However, I follow  (17) Syntactic incorporation in Mohawk  akya 'tawi 'tsher- V dress u:ni  NP  make kanekwarunyu polka-dotted Here, the noun akya'tawi'tsher- "dress" is (re)Merged at the root node, as a projection of V(P). How does the movement operation in (17) result in suffixation of the verb to the noun? I suggest that the affixation reflex for this head movement operation relates to the bound status of Iroquoian nominals. With few exceptions, nominals in Iroquoian languages are obligatorily morphologically bound, and cannot occur independently (Bryan Gick, Marianne Mithun, p.c.).  3  Let us therefore assume that akya'tawi'tsher- "dress" is subject to an affixation requirement. On the assumption that a host for this affixal noun must be a linearly adjacent element from its derivational sister, then the verb u:ni "make" will be selected as host due to its position as leftmost element of the derivational sister. The choice of u:ni "make" as host of akya'tawi'tsher"dress" would therefore immediately follow. Thus, the linearization of akya'tawi'tsher-u:ni "dress-make" results.  6.2.3 Local spell-out creates 'outside-in' dependencies According to the local spell-out hypothesis, the relationship between a Nuu-chah-nulth affixal predicate and its host is necessarily 'outside-in'.  4  A suffix (-oc) is always 'outside', or higher  than, the syntactic terminals contained within its derivational sister ((3): (18)  y  In Nuu-chah-nulth, an affixal predicate (-a) must be linearized with respect to a host from its derivational sister (3. If its derivational sister is linearized as <S, 7t>, then the host for an affixal predicate is determined to be 8, due to its string adjacency to this morpheme (<(8-oc), TC>). I return to this topic in §6.3.1. Outside of incorporation contexts, noun stems in Mohawk are most commonly suffixed by -a', a 'noun suffix' which contributes no discernible meaning (Marianne Mithun, p.c). Thanks to Gunnar Hansson for suggesting this term to me. 3  4  193  What about 'inside-out' dependencies? In this configuration, an affix 'climbs up' to find a host higher in the tree. In the illustration below, let us take - a to be a suffix, and 8 to be its host. In an 'inside-out' dependency, an affix - a attaches to the host 8, even though 8 is not contained within its derivational sister. In the tree below, n is the derivational sister of a, not 8: (19)  y  6.2.3.1 'Inside-out' dependencies are not possible via local spell-out I hypothesize that the effect of an 'inside-out' dependency is possible in natural language. However, it crucially will never arise from local spell-out.  Instead, I suggest that this  arrangement is necessarily achieved later in the derivation than what is possible by the 'aggressive' linearization algorithm of local spell-out. This type of affixation is not local in the same sense as local spell-out, since it is done on full assemblies of derivations, rather than in the incremental steps taken by local spell-out. In the following section, I present the notion that 'inside-out' linearizations apply in later stages of the derivation than local spell-out.  6.2.3.2 Evidence from Kwakw'ala (Northern Wakashan) Evidence for the existence of 'inside-out' dependencies comes from the Northern Wakashan language Kwakw'ala (Anderson 1984, Klavans 1985). As Anderson (1984) notes, in Kwakw'ala, determiners are enclitics, and they attach to the right edge of a preceding word. The following example is slightly modified from Anderson (1984: 21, ex. 1) to match the morpheme gloss conventions of this dissertation: (20) Kwakwala 'inside-out' enclitic determiner k ix?id-ida  bsg an9ma-x-a qasa-s-is  clubbed-DET  man-OBJ-DET  w  t3lwag ayu  w  w  otter-INST-POSS club  The man clubbed the sea-otter with his club.  bag andma  In this example, the determiner -ida (DET) is semantically construed as the marker for the w nominal  -a  "man". Yet instead of attaching to this nominal, it attaches  preceding word, the verb k ix?id "clubbed". The same leftward pattern is also exhibited by the w  other determiner in the sentence,  (DET).  Although this determiner should be semantically  bag amma  construed with the object nominal qasa "otter", it nonetheless attaches to the right edge of a  w  different nominal, the subject nominal  194  "man", which happens to precede th  The Kwakw'ala example can be contrasted with the following example from Nuu-chahnulth, which shows an 'outside-in' dependency.  In Nuu-chah-nulth, the determiner  reliably  suffixes to the right edge of the first word in its complement. (This distribution is identical to that of a Nuu-chah-nulth affixal predicate.)  Here,  suffixes to the nominal huupuukwas "car".  (21) Nuu-chah-nulth 'outside-in' enclitic determiner kuuwi+it?is cakup kuuwi+-mit-?iis cakup  huupuulc as?i huupuuk as-?ii  steal-PST-3.lND man  car-DET  w  w  A man stole the car. The cases in (20) and (21) constitute a minimal pair for the 'inside-out'/'outside-in' distinction. The difference between these two types of dependencies is represented syntactically below: (22)  a. Kwakw'ala'inside-out'affixation b. Nuu-chah-nulth'outside-in'affixation  In (22a), the Kwakw'ala determiner orients itself leftward to attach to the verb. In (22b), the Nuu-chah-nulth determiner positions itself rightward to attach to the nominal under its semantic scope. In each case, the determiner takes a single step to attach to a neighbouring word. Why, then, would only (22b) qualify as a 'local' linearization? Recall that local spell-out applies only to derivational sisters. Only in (22b) does the affix attach to a host within its derivational sister: in (22a), the derivational sister of the enclitic determiner D is N , not the V which it takes as a host. Therefore, (22a) does not satisfy this strict definition of derivational locality. In the following section, I propose that linearization induced by local spell-out applies only in cases in which the affix is lexically specified for an affixation requirement.  6.2.3.2 Affixation as primitive or derived What may condition the choice between linearization which is effected by local spell-out and linearization which occurs after spell-out? I suggest that it may relate to whether affixal status is primitive or derived in the language (cf. Anderson's (1992) 'special' or 'simple' clitics). Recall that in Nuu-chah-nulth affixal predicates are lexically specified as affixes. This lexical distinction  195  serves to differentiate them from independent predicates in the language, which never occur as suffixes. (23) Free and bound classes of predicates in Nuu-chah-nulth  independent predicates  affixal predicates  The difference between affixal and independent predicates is not independently reducible to a factor distinct from affixhood, such as prosodic weight. With respect to the prosodic heaviness, affixal predicates come in a range of types - from the non-syllabic to the polysyllabic - and, as such, overlap with the syllabic and polysyllabic forms of independent (non-affixal) predicates. Examples of the different weights of affixal and independent roots are given below. (24) Polysyllabic affixal predicates  (25) Polysyllabic independent predicates  a.  -naJcuuh [+L] "observe"  a.  kuuwH- "steal"  b.  -hahu+\+L] "on front"  b.  pawa-f "to lose something"  (26) Monosyllabic affixal predicates  (27) Monosyllabic independent predicates  a.  -naah "trying to locate"  a.  rhaa  "to bite"  b.  -pa+  b.  Jc ix  "to kiss"  "be present"  w  (28) Non-syllabic affixal predicates a.  -q [+R] "travelling with in vessel"  b.  -s[+L] "asking for"  Affix, in Nuu-chah-nulth, must therefore be a lexically-specified primitive. It is not the case that affixation takes place because of a deficiency in prosodic weight: affixal predicates may be weighty or weightless.  5  If affixal status is a lexical property of an affix in Nuu-chah-nulth, rather than a prosodically conditioned characteristic, then it constitutes a bare output requirement for that morpheme. It is a tenet of the Minimalist grammar that the lexically-specified properties of an element must be satisfied by the point of spell-out, so that the features of the lexical item may receive an appropriate interface interpretation. Under this view, it is lexical specification that forces affixation in Nuu-chah-nulth to emerge via local spell-out.  1  In Anderson's (1992) terminology, Nuu-chah-nulth affixal predicates thus cannot be 'simple' clitics.  196  According to this line of reasoning, the lexical entry for the Nuu-chah-nulth determiner must come equipped with an affix specification, in the same way that affixal predicates' do. In other words, affixal  status in Nuu-chah-nulth is primitive, for affixal  predicates or for  determiners. Under a local spell-out analysis of Kwakw'ala, the implication of this analysis is that, conversely, the affixal status of determiners in Kwakw'ala must be derived, not inherent. Only for derived affixes will 'inside-out' dependencies be tolerated. Only for derived affixes will affixation not be a requirement at the stage of local spell-out.  Instead, if affixal status is  prosodically determined, and established at a derivationally later stage, then prosodic manipulations will apply to find a host for the derived affix. There is independent evidence that affixation of the Kwakw'ala determiner applies later in the derivation than affixation in Nuu-chah-nulth. The key to this idea is the observation that a Kwakw'ala determiner construed with the subject attaches leftward to the verb. That is, in the VSO word order of Kwakw'ala, the determiner of the S attaches to the V , as we saw in example (20), repeated here below. (29)  k ix?id-ida  b3g anama-x-a qasa-s-is  clubbed-DET  man-OBJ-DET  w  w  t9lwag ayu w  otter-INST-POSS club  The man clubbed the sea-otter with his club. Crucially, as I noted in Chapter 3, V S O is not possible as an underived word order for a language, since the verb and object form a constituent. Instead, V S O order must always be derived. The fact that affixation takes place in Kwakw'ala only after V S O order has been established entails that it must be occuring at a derivationally later stage than in Nuu-chah-nulth. In Nuu-chah-nulth, an affixal predicate can only ever find a host from the object, never the subject, even though either VOS or V S O surface orders are permitted. The analysis of the affixation patterns of Kwakw'ala rests on empirical confirmation of the prosodic characteristics of Kwakw'ala determiners. Future research is required to assess the accuracy of the speculation that the affixal status of Kwakw'ala determiners is derived from prosodic factors, rather than being a lexical designation. The contrast between the 'inside-out' linearization of Kwakw'ala determiners and the 'outside-in' pattern of Nuu-chah-nulth affixes implies a typological split between the Northern and Southern branches of the Wakashan family. In the following section, I discuss additional typological implications for my proposal.  6.3 Typological implications This section presents three typological implications for my analysis of PF Incorporation.  I  address implications for the typology of noun incorporation in §6.3.1, followed in §6.3.2 by  197  discussion of the traditionally-assumed 'governing/restrictive' distinction in Wakashan.  In  §6.3.3, I argue that it is inaccurate to portray 'lexical suffixation' as an uniform areal feature of the Pacific Northwest.  6.3.1 The typology of noun incorporation A claim of this dissertation is that the affixation phenomenon descriptively labelled 'noun incorporation' is not a uniform construction (cf. Mithun 1984, 1986; Sadock 1986; Baker 1988). According to my analysis, a noun may come to be suffixed by a verb via four distinct 'routes' of bound/free dependencies. As first discussed in Chapter 2, the noun may be free or bound, and the verb may itself be free or bound. This four-way typology is given in (30). (30) Typology of noun-verb dependencies bound noun  free noun  bound verb Nuu-chah-nulth  Nuu-chah-nulth  free verb  English  Mohawk  This section reviews the affixation patterns for each of these four underlying noun-verb dependencies. Nuu-chah-nulth exhibits two of these four relationships. In Nuu-chah-nulth, bound predicates mandatorily require suffixation to a host, and never occur independently. The inability of an affixal predicate to separate from its host is illustrated in (31) for the predicate Pu-sh'k "make". In (31a), the affixal predicate suffixes to -hic?in "dress". In (31b), suffixation fails to occur and the result is ungrammaticality. (31)  a.  nupititsa nupit-mit-sa once-PST-1 SCABS I made a dress once.  -rucPinsiik +uc?in-siik dress-make  b.  nupititsa nupit-mit-sa once-PST-lSG.ABS I made a dress once.  siik +uc?in siik -ruc?in make dress  This lack of isolatability of the predicate sets the suffixation strategy in Nuu-chah-nulth apart from noun incorporation languages such as Mohawk. In Mohawk, a verb can be grammatically separated from an incorporated noun.  In (32a), the inflected predicate ye-nuhwe'-s "like"  incorporates the nominal nuhs- "house", while in (32b) it does not.  198  (32) Mohawk (examples from Postal 1962, as cited in Baker 1988: 81-82, ex. 14a-b) a.  Yao-wir-a'a  ye-nuhs-nuhwe'-s  PRE-baby-SUF 3FS/3N-house-like-ASP  The baby house-likes. b.  Yao-wir-a'a  ye-nuhwe'-s  ne  ka-nuhs-a'  PRE-baby-SUF  3FS/3N-like-ASP  DET  PRE-house-SUF  The baby likes the house. From this, I deduce that incorporation in Mohawk satisfies no boundedness requirement on the verb. An opposite effect is found with Mohawk nominals, however.  In Mohawk, a noun  separated from an incorporating verb typically cannot stand on its own. In Mohawk, the majority of nominals are obligatorily bound (Marianne Mithun, p.c). Outside of incorporation contexts, the regular pattern is that a noun stem is suffixed by a neuter prefix and a 'noun suffix'. (33)  6  NEUTER-noun.stem-NOUN.SUFFIX  An example of this pattern is given below for the Mohawk noun ohnenna:ta' "potato". (34)  ohnenna:ta' o-hnenna't-a' NEUT-potatO-NOUN.SUFFIX  'potato'  (Marianne Mithun, p.c.)  This example employs the most common noun suffix, -a', which contributes no discernible meaning (Marianne Mithun, p.c). This noun suffix appears to bear a formal resemblance to the expletive morpheme ?u- in Nuu-chah-nulth. In Nuu-chah-nulth, Pu- acts as a 'placeholder' for an affixal predicate requiring a host which it may suffix to. In Mohawk, a plausible hypothesis is that -a' acts as a 'placeholder' for an affixal nominal requiring a host which it may prefix to. Note that this morpheme does not surface when the noun has undergone incorporation, as indicated in (32) for the nominal nuhs-(a') "house". The final dependency is one in which a free noun suffixes to a free verb. This option corresponds to compounding in English. In (35), the noun file is free, as is the verb share. (35)  a.  They are sharing files,  b.  They are file-sharing.  6.3.2 The governing/restrictive hypothesis in Wakashan In arguing that the combinatory properties of Nuu-chah-nulth affixal predicates are conditioned by their argument structure, my analysis differs markedly from traditional treatments of these  There are some odd words used as lexical nouns which do not show this structure, but they are quite rare, and reportedly are never incorporated (Marianne Mithun, p.c).  6  199  suffixes (eg. Sapir and Swadesh 1939; Swadesh 1939; Rose 1981; Nakayama 1997, 1998; Davidson 2002).  In existing literature on the language, differences in these morphemes'  suffixation patterns have been noted, but it has not previously been recognised that these 'lexical suffixes' serve as predicates which show a range of distinctive argument structures.  Under  traditional classifications, affixal predicates are subject to a Wakashan-specific dichotomy between 'governing' and 'restrictive' suffixation. This distinction amounts to whether the suffix has a predicative or modificational role, respectively. Suffixes which I have analysed as transitive and location predicates are treated as predicative 'governing' lexical suffixes which take their morphological base as their object (Rose 1981). On the other hand, predicates which I classify as unaccusative and locatum predicates are grouped together with an assortment of other suffixes (such as plural markers) as modificational 'restrictive' lexical suffixes. According to Rose (1981: 313), these 'restrictive' suffixes "have an appositive relation to their base and do not influence the semantic class of the stem in which they occur." Syntactically, these suffixes are not considered to take complements. It is claimed that when a 'restrictive' suffix attaches to a nominal, the nominal does not serve as the object of the suffix, but rather as the main predicate.  For example, Rose (1981) analyses the locative  construction in (36) as being syntactically headed by the the nominal predicate siicpax "cougar". Crucially, by traditional analyses, the nominal siicpax "cougar" does not serve as the object of the locative suffix. (36)  siicpaxnaq siicpax-naq-0 cougar-pji_top.-3.ABS There's a cougar on top.  (Kyuquot dialect; adapted from Rose 1981: 314, ex. 447)  In contrast, the proposed analysis states that the locative -naq "on top" is a locatum predicate which takes siicpax "cougar" as its syntactic object.  (The subject is a null third person  pronominal, registered by absolutive agreement.) This analysis is represented in (37). (37)  VP  -naq on.top  siicpax cougar  Thus, by my analysis, it is not the case that the nominal siicpax "cougar" serves  as  the  main  predicate. At the heart of the governing/restrictive hypothesis is the notion that 'restrictive' suffixes, unlike 'governing' suffixes, do not syntactically head a predicate phrase. The proposal presented in this dissertation offers an alternative to the language-specific governing/restrictive hypothesis,  200  and derives the behaviours of affixal predicate sub-classes through universal tenets of argument structure (Wojdak 2004). According to this analysis, the syntax of affixal predicates is built from two basic underlying syntactic configurations of unaccusative or extended  unaccusative  structures. The host for affixal predicates is chosen from their complement.  6.3.3 Lexical suffixation as an areal feature of the Pacific Northwest There is a lengthy tradition of viewing the phenomenon of lexical suffixation as an areal feature of languages of the Pacific Northwest (see, for example, Sapir 1911, Swadesh 1948, Gerdts and Hinkson 1996, Kinkade 1998, Mithun 1999). Kinkade (1963) coined the term 'lexical suffix' due to "the semantic similarity between these suffixes and usual lexical items" (Kinkade 1998: 266). These bound morphemes bear a lexical load parallel to that of free morphemes. Under a strict definition, lexical suffixes are thus simply suffixal morphemes which convey lexical content (see Swadesh 1939). Within the Salishanist literature, however, the term 'lexical suffix' has come to be synonymous with the noun-like bound elements found in these languages (see Saunders and Davis 1975, Gerdts and Hinkson 1996). For Salishan languages, Gerdts (1998; citing Carlson 1989) notes that there is support for the notion that "lexical suffixes can be regarded as incorporated nouns that have lost their status as free-standing nominals" (Gerdts 1998: 97). For Wakashan