Minimal Indirect Reference

0 downloads 0 Views 746KB Size Report
Kaisse (1985) notes that this palatalization does not apply to all words. For ex- ...... Art`u arthur. [p]apa pope. 'They have elected Arthur pope' b. Credo. I-believe che ..... Native speakers I have consulted also confirm the acceptability of (37a). x '.

Minimal Indirect Reference: A Theory of the Syntax-Phonology Interface Amanda Seidl

Contents Preface

ix

List of Abbreviations

xii

1 Introduction 1.1 The problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2 Two accounts of the interface . . . . . . . . . . . . . . . . . . . . 1.2.1 Indirect reference . . . . . . . . . . . . . . . . . . . . . . 1.2.2 Direct reference . . . . . . . . . . . . . . . . . . . . . . . 1.3 The expressiveness of the phonological parser . . . . . . . . . . . 1.3.1 Indirect reference is too constrained . . . . . . . . . . . . 1.3.2 Traditional indirect reference is not constrained enough . . 1.3.3 Evidence in support of indirect reference: mismatches . . 1.3.4 The advantages of a syntax-only account . . . . . . . . . 1.3.5 Empty categories and the interface . . . . . . . . . . . . . 1.4 Domain Paradoxes . . . . . . . . . . . . . . . . . . . . . . . . . 1.5 A new proposal for the interface: Minimal Indirect Reference (MIR) 1.6 Outline for the book . . . . . . . . . . . . . . . . . . . . . . . . . 1.7 Summary of proposals made in the book . . . . . . . . . . . . . .

3 3 5 5 8 9 9 10 10 11 12 12 14 15 16

2 Domain Paradoxes 2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 The structure of the argument . . . . . . . . . . . . . . . . . . . . 2.3 Domain Paradoxes and violations of domain clustering . . . . . . 2.4 Domain clustering violations in Kpa Mende . . . . . . . . . . . . 2.4.1 The domains for tone sandhi and mutations . . . . . . . . 2.4.2 The solution to the domain clustering problems in Mende . 2.4.3 Supporting evidence from English . . . . . . . . . . . . . 2.4.4 Supporting evidence from Korean . . . . . . . . . . . . . 2.5 Kimatuumbi: another violation of domain clustering . . . . . . .

19 19 20 21 22 23 27 33 35 38

v

Contents

vi

2.6 2.7 2.8

2.5.1 A solution to the domain problem in Kimatuumbi 2.5.2 In support of Odden’s account . . . . . . . . . . L AYEREDNESS violations in Luganda . . . . . . . . . . 2.6.1 A solution to the Luganda domain paradoxes . . Yoruba Domain Paradoxes . . . . . . . . . . . . . . . . 2.7.1 The solution for Yoruba . . . . . . . . . . . . . Summary of the chapter . . . . . . . . . . . . . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

41 45 46 48 51 54 55

3

Contrasting various recent Phonological Domain Generators 57 3.1 Four accounts of Chimwi:ni and Chichewa ˆ . . . . . . . . . . . . . 57 3.1.1 A relational account of Chimwi:ni . . . . . . . . . . . . . 58 3.1.1.1 Problems with the restructuring relational account 59 ˆ 60 3.1.1.2 A relational account cannot account for Chichewa 3.1.2 An End-Based account of Chimwi:ni . . . . . . . . . . . 60 3.1.2.1 Problems with the End-Based theory . . . . . . 61 3.1.2.2 The End-Based theory cannot account for Chichewa ˆ 62 3.1.3 The Optimality Theoretic account . . . . . . . . . . . . . 62 3.1.3.1 Problems with the Optimality Theoretic account 65 3.1.4 The Null Theory: a direct reference account of the interface 66 3.1.4.1 The Null Theory’s account of German stress . . 67 3.1.4.2 The Null Theory when applied to Chichewa ˆ . . 69 3.2 The inadequacy of a theory which assumes a unique P . . . . . . . 71

4

The Minimal Indirect Reference approach 4.1 A thumbnail sketch of MIR . . . . . . . . . . . . . . . 4.1.1 Other aspects of MIR . . . . . . . . . . . . . . 4.2 Supporting evidence for theta-domains from Kinyambo 4.3 Supporting evidence for theta-domains from K nni . . 4.4 Summary of the chapter . . . . . . . . . . . . . . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

73 74 76 78 79 84

MIR applied to the Bantu data 5.1 Two distinct patterns of phonological domains 5.1.1 The single domain pattern . . . . . . 5.1.2 The split domain pattern . . . . . . . 5.2 An important correlation . . . . . . . . . . . 5.3 How phonological domains are constructed . 5.4 A syntactic account of Bantu . . . . . . . . . 5.4.1 The symmetrical passive . . . . . . . 5.4.2 The asymmetrical passive . . . . . . 5.5 Verb movement and object shift . . . . . . . 5.5.1 Asymmetric Bantu languages . . . . 5.5.2 Symmetric Bantu languages . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

85 86 86 88 89 89 91 91 91 92 92 94

5

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

vii

5.5.2.1 Less movement in Asymmetric languages . 5.5.3 Sentences with OMs and passives . . . . . . . . . . . 5.5.4 Some variation in symmetrical languages: Chaga . . . ˆ 5.5.5 Some variation in asymmetrical languages: Chichewa British English and phonological domains . . . . . . . . . . . Summary and further directions . . . . . . . . . . . . . . . .

. . . . . .

. . . . . .

95 97 100 102 103 106

Revisiting the visibility conditions on rules 6.1 What is a functional projection? . . . . . . . . . . . . . . . . 6.1.1 The Government Binding view of lexical categories . . 6.1.2 Baker’s view of the categories . . . . . . . . . . . . . 6.1.3 Adjectives and the lexical/functional distinction . . . . 6.1.4 Is vP relevant to the phonology? . . . . . . . . . . . . 6.1.4.1 The lexical/functional distinction and Bantu 6.2 Stress in Spanish . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . .

. . . . . . .

107 107 108 109 112 116 117 118

Kaisse (1985) and MIR 7.1 The Connected Speech proposal for the interface . . . . 7.1.1 The data CS succeeds in explaining . . . . . . . 7.1.1.1 A P2 rule: Flapping . . . . . . . . . . 7.1.1.2 A P1 rule: Raddoppiamento Sintattico 7.2 Problems with CS and differences between CS and MIR 7.2.1 P2 rules are not just rules of fast speech . . . . . 7.2.1.1 P2 late rules are not entirely prosodic . 7.2.1.2 There is a Prosodic Hierarchy . . . . . 7.2.2 C-command . . . . . . . . . . . . . . . . . . . . 7.3 P2 rules have lexical exceptions, not P1 rules . . . . . . 7.4 Summary of the chapter . . . . . . . . . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

121 121 122 122 123 125 125 126 131 131 131 132

8 Conclusion 8.1 Revisiting the mismatch data . . . . . . . . . . . . . . . . . . 8.1.1 An alternative account of mismatches . . . . . . . . . 8.1.2 The MIR account of mismatches . . . . . . . . . . . . 8.1.2.1 The MIR explanation of English mismatches 8.1.2.2 MIR in Tohono O’odham . . . . . . . . . . 8.2 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . .

. . . . . .

133 133 134 136 136 137 138

5.6 5.7 6

7

. . . . . . . . . . .

. . . . . . . . . . .

Notes

139

Bibliography

151

Index

159

ix

Preface This book is a slight revision of my doctoral thesis (Seidl 2000), written at the University of Pennsylvania. The book investigates the hypothesis that phonology can refer to two levels of structure, one which has domains corresponding to domains of the Phase (in the sense of Chomsky 1999) and another which consists of phonological domains constructed by a simple mechanism referring to the domains in which theta-roles are assigned. The book also makes a methodological point: It is proposed that the locus of variation in the size of the domains in which phrasal phonological rules apply should first be assumed to be syntactic. It is argued that although there may be some variation in the phonology of languages in the construction of phonological domains, the primary locus of variation which causes differences in the phonological domains of languages is located in their syntax. It is proposed that aspects of the Prosodic Hierarchy which make reference to constituents larger than the word are unnecessary. In standard Prosodic Hierarchy Theory deriving from Nespor and Vogel (1986), Selkirk (1984,1996), inter alia, it is claimed that UG contains a finite set of constituent types (such as ‘Phonological Phrase’). Since each of these constituent types is a representational prime within UG, then UG may also contain specific constraints referring to these constituents (as in Optimality Theory (Selkirk 1996, Truckenbrodt 1995,1998)). This book rejects the premise that UG contains such a set. It follows from this that UG also cannot contain constraints referring to members of this set, since the set does not exist. A theory of minimal indirect reference (MIR) is proposed for the interface between syntax and phonology. In MIR, all variation across languages in phonological domain size is attributed to syntactic differences, along with a single domain parameter specific to phonology. MIR is shown to be more constrained than previous theories of direct reference (Cinque 1993, Odden 1995, Kaisse 1985, inter alia) since in MIR phonological domains are sensitive only to the output of a post-syntactic morphology component and not to the workings of syntax proper. Finally, MIR is shown to have greater coverage of the data than previous theories of indirect reference (Selkirk 1996, Truckenbrodt 1998, Hayes 1989, inter alia).

Acknowledgements Thanks to the members of my thesis committee; Gene Buckley, Tony Kroch, Mark Liberman and Martha McGinnis, for their contributions to this book; and thanks especially to Rolf Noyer for his tireless efforts. In addition I thank Rolf Noyer, my

x

Preface

committee members, my fellow graduate students and postdocs, and Ellen Prince for aiding my development as a linguistic researcher. Their help was instrumental in informing my ability to think like a linguist. I thank my friends, family and mentors for helping me through all the challenges of the past five years, for making it possible for me to achieve here as well as elsewhere, and above all for being my friends and family. Finally, thanks to Akin Akinlabi, Vincent Ochwo, Eon-Suk Ko, Moses Pessima, Alwiya Omar, Monde Makiwane, and Andrew Kulemeka for data from ˆ respecYoruba, Luganda, Korean, Kpa Mende, Swahili, Isixhosa and Chichewa tively. Thanks are also due to Brian McHugh for the phrasing data on Chaga and to Michael Kenstowicz, Charles Kisseberth, Lisa Selkirk, Bob Ladd, Geraldine Legendre, Hubert Truckenbrodt and the audiences at WCCFL 18, NAPHC 1, and at Johns Hopkins for many helpful comments and suggestions. This thesis is dedicated to the memory of my father Robert Friedman (1935– 1990) for inspiring me to do any sort of academic work and to my best friend Cynthia Elbaum (1966–1994) for thinking I could do anything.

xii

List of Abbreviations 

3p 3s 1p ACC AP Appl Art Aux CAUS C Comp DO FOC FUT FV Indic INF Inter

Phonological Phrase Third person plural Third person singular First person plural Accusative case Applicative Marker Applicative Article Auxiliary Causative Clitic Group Complmentizer Direct Object Focus Marker Future tense Final Vowel Indicative Infinitive Marker Interrogative

Impf IO IV LOC NC NEG NOM OM PAS Perf Pros PRS PST Quan RecP SM Stat T

Imperfect Indirect Object Initial Vowel Locative Noun Class Marker Negation Nominative case Object Marker Passive Marker Perfect aspect Prospective aspect Present tense Past tense Quantifier Recent Past tense Subject Marker Stative aspect Tense

C HAPTER 1

Introduction 1.1 The problem The problem that this book addresses is an interface problem: certain phonological rules apply to domains larger than the word when words occur in particular syntactic relations, and they are blocked when these relations do not obtain. An example of such a process is intrusive-r in Received Pronunciation of English (RP). In RP [r] is inserted before a vowel in certain syntactic contexts (1a), but not in others (1b). (1.1)

Vogel (1986) a. Amanda [r]always lies. . . b. Let’s go to Canada. (*[r]) It’s Chip. . .

Another such rule occurs in Hausa. In Hausa, vowel shortening applies when a verb directly precedes a full NP direct object (2b), but not when it precedes a pronoun direct object (2a) or an indirect object (2c). 

(1.2)

Kaisse (1985) a. n´a: k´a:m`a: sˇ´ı I have-caught it ‘I have caught it’ b. n´a: k´a:m`a k´ı:f´ı: I have-caught fish ‘I have caught a fish’ c. n´a: k´a:m`a: w`a mu:s´a: k´ı:f´ı: I have-caught for musa fish ‘I have caught Musa a fish’ 3

Minimal Indirect Reference

4

Thus certain rules must refer to some hierarchical structure — either syntax or some phonological constituent structure based indirectly on syntax. Chomsky and Halle (1968) propose an algorithm for rules which refer to syntactic structure. They suggest that boundaries are inserted at the beginning and end of lexical categories and the beginning and end of phrases, and that the phonology sees these boundaries: The boundary # is automatically inserted at the beginning and end of every string dominated by a major category, i.e., by one of the lexical categories “noun,” “verb,” “adjective,” or by a category such as “sentence,” “noun phrase,” “verb phrase,” which dominates a lexical category. (p.366) They provide the example in (1.3): (1.3)

Chomsky and Halle (1968) a. The book was in an unlikely place. b.

#



#







an



 

# the #book# #un #likely# # 













#





# # # #was# #place# # # # #















































in





Chomsky and Halle’s algorithm also includes an automatic readjustment rule that deletes the innermost of any three consecutive #’s. This yields the structure in (1.4). (1.4)



an



#





#un



# 







the #book# # #likely# # #place#



 

















 



#



 





#



 

 





#



#was# # 







in #





This structure is then converted into what Chomsky and Halle call Phonological Phrases, units larger than the word to which phonological rules apply. Phonological Phrases are constructed in a way which is not explicitly specified, but Chomsky and Halle suggest a system where increasing numbers of #’s represent higher prosodic boundaries. The system suggested in Chomsky and Halle (1968) is spelled out explicitly by Clements (1978), who proposes that two #’s represent a Phonological Phrase boundary in Ewe (Kwa). An account such as Chomsky and Halle’s (and the subsequent elaboration of it in Clements (1978)), which suggests that the rules which refer to domains larger than the word can always be described by reference to the number of boundaries separating words, has difficulty accounting for the data in (1.2) because presumably the same number of syntactic boundaries exist whether a constituent is a head of NP or non-head. Compare (2b) and (2a): the only difference between the two sentences is that in one the direct object is a pronoun and in the other it is a full

Introduction

5

NP. In addition, Chomsky and Halle’s algorithm requires the phonology to count boundaries, which results in a less restrictive theory. In general, in natural language, grammars that count to specified values are unattested because they are too powerful.

1.2 Two accounts of the interface There are essentially two ways in which phonological rules that refer to syntactic relations have been handled: direct reference and indirect reference. Let us assume that there is parse P which has domains which phonological rules may refer to. Both indirect reference and direct reference theories assume that this parse P exists and that it is unique. The difference between indirect reference accounts and direct reference accounts is found in the character of P. On indirect accounts of the interface between syntax and phonology, P is composed of purely phonological primes, while on direct accounts of the interface P is composed of purely syntactic primes. In addition, the indirect reference hypothesis assumes that there is a single point in a derivation where a constituent structure unique to phonological domain construction (P) is assigned. The direct reference hypothesis, on the other hand, assumes that there is no point in a derivation where a constituent structure unique to phonological domain construction is assigned. Understanding the nature of the interaction between phonology and syntax is not only crucial for linguistic theory (with the aim towards a unified theory and use of phonological information to evaluate our syntactic theories), but is also crucial for understanding the nature of language in the mind. For example, if the construction of phonological domains is not directly dependent on the syntax but rather is dependent on some sort of independent prosodic structure, then it would be possible (for e.g., an agrammatic aphasic) to lose prosodic phrasing and have intact syntax.

1.2.1

Indirect reference

Indirect reference theories (Hale and Selkirk (1987), Selkirk (1984, 1996), Nespor and Vogel (1986), Truckenbrodt (1995)) suggest that this P is a level of prosodic representation whose constituent types are ranked and strictly layered. P is generated using what I call a Phonological Domain Generator (PDG), a set of rules or constraints which creates P from a syntactic representation. A schematic diagram of indirect reference is shown in figure (1.1).

Minimal Indirect Reference

6

Morphosyntactic Representation (M ) 





Phonological Domain Generator

Prosodic Representation (P)





Phonological Rules

Surface Phonological Representation Figure 1.1 Indirect Reference

Indirect reference theories propose that phonological rules cannot refer directly to syntactic structure, but refer to prosodic constituents which exist as an independent level in a derivation (Hale and Selkirk (1987), Truckenbrodt (1995), Selkirk (1984), Nespor and Vogel (1986), inter alia). Prosodic domains in this kind of theory are constructed from the syntax, but are independently manipulable. To illustrate this theory, first suppose a derivation consists of stages S , S , . . . S . At each stage S , let M be the morphosyntactic representation of an expression E at S , and let P be the phonological parse of E at S . An indirect reference rule applying to an expression E after stage S , but before stage S refers only to information present in the phonological parse P , but not to information in morphosyntax at S . In addition, indirect reference theories support a hypothesis of specialized parsing primes. These representational primes are crucially not the same as those which compose a morphosyntactic parse. Such theories also propose that the set of constituent types unique to the phonological parse is finite and fixed by UG. These make up the Prosodic Hierarchy above the Prosodic Word level. Nespor and Vogel (1986), like Selkirk (1981), propose that there is Prosodic Hierarchy for P both above and below the word level. Their hierarchy is illustrated in figure (1.2). 

   

















 







Introduction

7

Utterance Intonational Phrase Phonological Phrase Clitic Group Word Foot Syllable Figure 1.2 The Prosodic Hierarchy (Nespor and Vogel 1986)

In addition, Nespor and Vogel’s account of indirect reference and the interface includes a proposal that there are four basic principles of prosodic phonology concerning the unique P. These principles are listed in (1.5); (Principle 1 is inspired by the Strict Layer Hypothesis proposed in Selkirk (1981)), i.e., it is a statement that the constituent types in P must be strictly layered). (1.5) Nespor and Vogel (1986) Principle 1: A given non-terminal unit of the prosodic hierarchy, X is composed of one or more units of the immediately lower category X . Principle 2: A unit of a given level of the hierarchy is exhaustively contained in the superordinate unit of which it is a part. Principle 3: The hierarchical structures of prosodic phonology are n-ary branching. Principle 4: The relative prominence relation defined for sister nodes is such that one node is assigned the strong value (s) and the other nodes are assigned the weak value (w). 







Selkirk’s (1996) End-Based account reformulates a few of the principles proposed by Nespor and Vogel (1986) and Selkirk (1981) as ranked and violable constraints. These constraints limit the kinds of end-based accounts that are possible in a language, so that units such as recursive structures and non-strictLayeredness are penalized. These constraints are shown in (1.6). Just as a caveat, the formal definition of indirect reference that I have provided above does not properly highlight the attractiveness of indirect reference and Prosodic Morphology at the date of its inception. Indirect reference is an attractive idea for several reasons: (1) Prosody clearly contains units of differing 

Minimal Indirect Reference

8

(1.6) Selkirk (1996) N ONRECURSIVITY: No C dominates C , j =i Phrases are not recursive E XHAUSTIVITY: No C immediately dominates a constituent C , j i-1 There is no skipping of levels within the hierarchy L AYEREDNESS : No C dominates a C , j i Lower levels cannot dominate higher levels H EADEDNESS : Any C must dominate a C All parts of the prosodic hierarchy except the syllable must dominate something 























Figure 1.3 Constraints for the Strict Layer Hypothesis (Selkirk 1996)

lengths and strength which indicate a prominence relation and Prosodic Morphology gives structure to these units. (2) Syntax and semantics have structure, thus for reasons of symmetry with these other systems it is attractive for phonology to have analogous units. (3) The assumption that phonology has its own structure serves to explain certain mismatch data (perhaps too easily), i.e., because phonology has its own hierarchical structures and syntax has its own hierarchical structure it is only natural that there should be times when these two structures do not match up.

1.2.2

Direct reference

In contrast with indirect reference theories described in 1.2.1, direct reference theories suggest that the parse P is the output of syntax (not a separate level of phonological structure). Thus the only P which can be referred to in direct reference theories is a level of syntax (Cinque (1993), Odden (1990)). A diagram of direct reference is shown schematically in (1.4). 

Morphosyntactic representation (M ) 





Phonological Rules

Surface Phonological Representation Figure 1.4 Direct Reference

As illustrated above, in direct reference theories, phonological rules refer directly to syntactic constituents (Cinque (1993), Odden (1990)). A direct reference

Introduction

9

rule applies to expression E at stage S and refers to information which is present in the morphosyntactic structure of E at S. Specific direct reference theories are discussed more extensively in chapter 3. A limiting of information or loss of information is common to both indirect and direct reference accounts, but indirect reference accounts typically also add information (usually in the form of additional structure) at the unique level P. However, as mentioned earlier the crucial point is that traditional indirect and direct reference accounts assume that there is only one parse P to which phonological rules can refer.

1.3 The expressiveness of the phonological parser in classic accounts In this section, I show that because they adhere to the assumption that all rules refer to the same parse, classic indirect reference and direct reference accounts are at once too expressive and not expressive enough.

1.3.1

Indirect reference is too constrained

This book presents evidence that not only does the phonology make reference to syntactic bracketing, but phonological rules may also depend directly on the syntax without having to be filtered through Phonological Phrases. I discuss some data supporting this claim in chapter 2. If reference by phonological rules to syntactic structure is unconstrained, then the phonology ought to be able to refer to any piece of syntactic information it requires. This clearly does not occur, e.g., a phonological rule never refers to whether or not Tense is +/– past . According to theories of indirect reference, the limitations on syntactic information available to the phonology are proof that the syntax is mediated by an additional level of structure, namely Phonological Phrases (Selkirk 1984, Nespor and Vogel 1986). I argue that this is not sufficient evidence that Phonological Phrases exist as primitives. In chapter 4, I discuss the ways in which reference by the phonology to syntax is limited. In addition, I propose that because the phonology can, and in fact, as we will see, must be able to refer to a degree of syntactic structure, a theory of indirect reference is too constrained. On the theory proposed here, the degree of access is constrained, but not so constrained as to not refer to syntax at all. I show that phonology is constrained in that it refers only to certain pieces of syntactic information, namely Phases (units of a derivation) and parts of syntax which remain in order to give a sentence its interpretation —theta-roles. Specifically, after non-interpretable features have been checked (presuming that syntax precedes morphology in the formal grammar in a derivational framework) the parse contains only information about theta-roles; all other features have been 



Minimal Indirect Reference

10

checked and are thus not accessible to phonology. Theta information remains because it is needed by another component as well, namely the semantic component. The level of syntax accessible to the phonology is what limits the interface to the desired small window that is theoretically preferred.

1.3.2

Traditional indirect reference is not constrained enough

A theory of indirect reference, unless constrained in certain ways, predicts the occurrence of Phonological Phrase types which are not attested in human language. Such a theory needs additional constraints in order to restrict it from generating unattested Phonological Phrasing structures (see chapter 3). An account based on syntax needs no such additional constraints because the syntax carries its own constraints, e.g., constraints on the degree of branching and other properties of syntactic well-formedness. Thus, the theory of indirect reference is at once too restrictive ( 1.3.1) and not restrictive enough (as mentioned in this section, 1.3.2). 

1.3.3



Evidence in support of indirect reference: mismatches

Proponents of indirect reference (Selkirk (1984, 1986, 1996), Nespor and Vogel (1982, 1986), Truckenbrodt (1995), and Hannahs (1996)) argue that the relation between syntactic structures and phonological representations is mediated by a level of phonological structure which includes prosodic constituent boundaries. Thus phonological rules cannot refer directly to syntactic structure, but refer only to prosodic constituents. The advocates of indirect reference suggest that because Intonational Phrasing and Phonological Phrasing are not always isomorphic with the syntactic parse, the direct reference account is inadequate. This is illustrated in the classic example in (1.7), where the mismatch between Intonational Phrasing and syntactic phrasing is clearly exemplified for English (Chomsky and Halle (1968)). (1.7)

Chomsky and Halle (1968) a. This is the cat that caught the rat that stole the cheese syntax b. (this is the cat) (that caught the rat) (that stole the cheese) phonology   













In (1.7) the locations of syntactic boundaries before the cat, the rat and the cheese are not the locations of Intonational Phrase boundaries, i.e., intonation diagnoses a parse which is not isomorphic with the syntactic parse. This nonisomorphy between the syntax and intonation is called a mismatch. This mismatch may be exemplified for Phonological Phrases as well as Intonational Phrases, as Hale and Selkirk (1987) show for Tohono O’odham. In (1.8) the location of major syntactic boundaries once again does not correspond to the location of the domains of tone assignment (LHL tonal sequences are assigned within these domains). For example, although a major syntactic boundary occurs after

Introduction

11

’at, nonetheless ’at is phrased with the NP that follows it w´akial, thus resulting in a seeming mismatch. (1.8)

Hale and Selkirk 1987 c´epos ’at w´akial w´ısilo brand did t t t cowboy calf (brand) (did cowboy) calf) ‘The cowboy branded the calf’  

























  

syntax phonology

In this book I take the point of view that such mismatches can be accounted for easily without invoking phonological constraints (Truckenbrodt 1995, 1999) which increase the expressiveness of the Phonological Domain Generator, but rather by relying primarily on a more elaborated version of the syntax of the languages in which mismatches have been attested. By examining in detail the typological similarity of a range of mismatches, I show that mismatches do not require a theory of indirect reference and restructuring constraints, with their associated increase in the expressiveness of the Phonological Domain Generator. I suggest that the Phonological Domain Generator is greatly simplified and that this piece of the system can be less expressive than previous theories have suggested. In addition, I discuss arguments made by Phillips (1996) concerning Tohono O’odham to the effect that mismatches between the domains of phonological rule application and syntactic structure are not really mismatches in the conventional sense. Specifically, according to Phillips, mismatches between phonological rule domains and the syntax result from the parsing of sentences from left-to-right, such that Intonational Phrases are constituents at the time of parsing and seeming mismatches are the result of a sort of ‘garden path effect’ (Trueswell, Tanenhaus and Garnsey 1994). !

1.3.4

The advantages of a primarily syntax-only account: explanatory value

This book will discuss the advantages of a primarily syntactic account of the interface between syntax and phonology. For example, a syntax-only account is simpler because no additional language-specific constraints are required; the domains of phrasal phonological rules fall out directly from syntactic structure mediated only by some parameters of visibility in the phonology. In addition, if differences in phonological domains follow directly from differences in syntactic domains, the child learner can really learn something about the syntax directly from the phonology, which eases the task of acquisition. This book takes a position in support of prosodic bootstrapping (Gleitman and Wanner 1982), the idea that acoustic information can provide infants with the relevant domain bracketings to lead them to syntax, i.e., that infants use prosodic cues for bottom-up segmen-

Minimal Indirect Reference

12

tation of speech.

1.3.5

Empty categories and the interface

Both indirect reference and direct reference accounts must deal with the problem of empty categories. Specifically: are empty categories visible to the phonology or not? It has been alternately proposed that empty categories are visible to the phonology (Cowper and Rice (1987)) and that they are invisible to the phonology (Nespor and Vogel (1986)). For example, Cowper and Rice (1987) suggest that consonant mutation in Mende is sensitive to empty categories. In the example in (9a), they claim that mutation is blocked from occurring on the verb when there is an empty object. In contrast, the same verb’s initial consonant mutates when the verb is used intransitively (9b). (1.9)

Cowper and Rice (1987) &

&

a. nd`opo´ı e f` mb´ ng`a child it swung ‘the child swung it’  " # $

&

% #

&

b. nd`opo´ı v` mb´ ng`a child swung ‘the child swung’ $

% #

I show in chapter 2 that although Cowper and Rice’s analysis of these examples is not quite correct, Mende does provide some evidence in support of the visibility of empty categories because there are cases where empty categories are relevant to the phonology. (In chapter 2 I also discuss additional such cases from Kimatuumbi where empty categories seem to be relevant to phonological rule application.) Nevertheless, Cowper and Rice’s analysis of Mende is incorrect because they assume that there is just one parse (P) which the phonology can refer to. In chapter 3 of this book I discuss why the assumption that there is just one P cannot be correct. I briefly outline this argument in the next section.

1.4

Domain Paradoxes: A problem for theories where P is unique

Domain Paradoxes are cases where two rules make reference to distinct domains which are not nested, but are instead overlapping. In chapter 2, I show that both direct reference accounts and indirect reference accounts fail to account for all of the data of natural language because both theories assume that P is unique; but P is not unique. An example of one such Domain Paradox in Luganda is illustrated in (1.10). In (1.10) the domains for a tonal plateauing rule (TP) and a vowel

Introduction

13

shortening rule (VS) can be identical (10a), the TP domain can be larger than the VS domain (10b), or the TP domains can be smaller than the VS domains (10c). These data are also a problem for an indirect reference account which assumes the Prosodic Hierarchy Theory, since these Domain Paradoxes violate the L AYEREDNESS constraint of Prosodic Hierarchy Theory. These data are discussed in greater detail in chapter 2. (1.10)

Hyman et al. (1987) a. ( t´u-ly-´aa´ =kˆo ) we-eat-indic=of ‘We eat a little’ 





'



mbw´a b. ( t´u-ly-´a we-eat-indic dog ‘We eat dog’ 

c.









(te-t´u-ly-`aa` ) (=kˆo) Neg-we-eat-indic=of ‘We don’t eat any’ '





'





)

'



Vincent Ochwo, p.c.





By examining data from Mende, Kimatuumbi, Korean, and Yoruba I show that it is necessary for there to be a P and a P (i j), i.e., two parses which phonological rules may apply to. In addition, I argue that some rules, early rules, apply to P and other rules, late rules, apply to P , and that late rules are sensitive to a different parse than early rules are. In addition, I show that where it seems we “need” two such parses P and P , then P is always a “trivial image” of M (the first morphosyntactic parse). If this is the case, then we can simply require that P be equal to M and conclude that only one parse aside from M , namely P , is needed for proper phonological rule application. I then show that if a rule R refers to M and a rule R refers to P, then it is necessary that R precede R . These two parses and the rules which apply to them are illustrated in figure (1.5).  (



































P /M









P







R



R



Figure 1.5 Different rules refer to different parses As evident from the supposition that both P /M and P are necessary, I show 





Minimal Indirect Reference

14



that R and R do not apply to the same P , that is not all rules apply to P /M . Based on these findings, in chapter 4 I make a proposal about the architecture of the postsyntactic grammar which makes a distinction between rules which apply to M (early rules) and rules which apply to P (late rules). However, before I make this proposal, in chapter 3, using cases for which it has been claimed that the phonology needs indirect access to syntactic information, I will attempt to show that direct reference is, in some cases, necessary and that indirect reference in some cases makes the wrong predictions. The theory which I adopt allows for both direct reference rules and a more limited set of indirect reference rules. I examine data from certain Bantu languages and show that if a unique P is assumed it is not possible to account for these Bantu data. 









1.5

A new proposal for the interface: Minimal Indirect Reference (MIR)

I propose a theory of Minimal Indirect Reference (MIR). This MIR theory has three parts. The first part is a proposal about the architecture of the postsyntactic grammar. Namely, I propose that P is not unique, and that there are rules which refer to M and rules which refer to P ; this part of my proposal is a hybrid of the indirect and direct reference theories in that it allows reference to both P and M . I show that there exists a class of rules which refer to M (early rules) and a class of rules which refer to P (late rules). The second part of MIR is a proposal about the visibility conditions on early rules: It limits the kinds of syntactic information available to both early and late rules. I propose that for early rules the phonology cannot see any information not in the syntax, and the phonology is unable to see certain aspects of syntax, so that there are a limited number of ways in which phonological domains can vary. I suggest that the only information early rules may refer to is the edges of the domains of a Phase (in the sense of Chomsky 1999). According to Chomsky (1999) a derivation proceeds by Phase, thus a Phase is a unit in a derivation. Specifically, a Phase consists of a clause unit involving tense. I suggest that the phonologies of languages vary only in that the phonology of a given language can be insensitive to only certain aspects of the syntax, or certain edges of the Phase. The idea that the phonology is selectively sensitive is not a new one. Calabrese (1995) proposes something similar for the visibility of certain distinctive features. Namely, he proposes that phonological rules be parameterized to “see” either all contrastive feature values, or just marked feature values. I propose that, in the same way that certain degrees of feature specification are hypothesized to be visible only to certain phonological rules, certain aspects of syntax are visible only to certain phonological rules. The third part of MIR is a theory of a Prosodic Domain Generator. I propose that 









Introduction

15

phonological domains are generated using Halle and Idsardi’s (1995) “Simplified Bracketed Grid Theory” of the metrical parse. I also show, contra previous indirect reference accounts (Selkirk (1981, 1996), Nespor and Vogel (1986), inter alia), that P does not need to have constituent types. This more restricted version of indirect reference allows for direct reference rules as well as some very limited indirect reference rules. I show that if we assume MIR we can greatly simplify the Phonological Domain Generator (PDG). A simpler PDG serves to ease the task of acquisition. The architecture (inspired by Kaisse (1985)) which I adopt for the architecture of the post-syntactic grammar for MIR is shown in figure (1.6). )

Morphosyntactic Representation (M )











Early Rules

Phonological Domain Generator

Prosodic Representation (P ) 





Late Rules

Surface Phonological Representation Figure 1.6 Suggested architecture for the syntax-phonology interface The theory which I propose has two levels: One level has rules which are conditioned by information in the morphosyntax (that is, they are “direct reference” in the sense defined above) and a second level has rules which apply to domains constructed from the syntax. The second level, which I call the Prosodic Representation (P), represents an extremely constrained version of indirect reference. I propose that its prosodic domains are based on the syntax and vary only on one parameter: whether the phonology makes reference to a right or left bracket of a syntactic domain where theta-roles are assigned. The account I propose makes no reference to the specialized parsing primes (or the Prosodic Hierarchy) that previous accounts found necessary (Selkirk (1984, 1995), Nespor and Vogel (1986), inter alia).

1.6 Outline for the book The outline of this book is as follows: In chapter 2 I discuss Domain Paradoxes. In chapter 3 I argue that previous indirect reference accounts that add information at a unique level make predictions which are not borne out. I show that the theories which use Phonological Phrases as primitives make false predictions; thus

Minimal Indirect Reference

16

there is no motivation for the adoption of Phonological Phrases as primitives. I also show that additional constraints which refer to phonological phrasing (for example, those proposed by Truckenbrodt (1995)) are unnecessary, and suggest that there is no need for a Prosodic Hierarchy above the word level. I also argue that previous direct-reference-only accounts are not expressive enough. In chapter 4 I develop the theory of MIR. In chapter 5 I refute arguments suggested in the recent literature that require greater complexity in the PDG (e.g., Truckenbrodt 1995, 1999). The indirect reference theory has become extremely complex in order to account for certain data from Bantu languages, but this complexity is unnecessary if we assume MIR. I show that there are several reasons why Phonological Domain Generators have become unnecessarily complex in recent proposals. The primary reason for the complication of PDGs is that morphosyntactic structure has been misunderstood: I show by examining cases of Bantu phonological domains that a better understanding of morphosyntactic structure nullifies arguments for a more expressive or more complex PDG. PDGs have also been complicated because early rules were presumed to be late rules, that is, rules which apply at M have been presumed to apply at P. In chapter 2, I show that in Kimatuumbi an assumption that the phonological rules under consideration are rules which refer to M and not to P leads to a simpler PDG. In chapter 6 I revisit the visibility conditions on rules, and I propose that the distinction made in the literature between lexical projections and functional projections is misleading (Selkirk (1986), Truckenbrodt (1995), Chomsky and Halle (1968) Hale and Selkirk (1987), inter alia). I show that the theta-domain approach introduced in chapter 4 makes this distinction unnecessary. In chapter 6 I argue, on the basis of recent work (e.g., Baker 1999), that the linking of the lexical/functional distinction with the phonological visibility/invisibility distinction is also unmotivated for independent reasons. I also argue, on the basis of recent work on the assignment of stress in Spanish (Oltra-Massuet and Arregi 2000), that an account which makes reference only to lexical categories will make the wrong predictions for stress assignment in Spanish. Next, in chapter 7, I show the ways in which MIR is different from the system proposed in Kaisse (1985). Namely, I argue that late rules can apply to prosodic domains which are not just domains of fast speech, and that the domains to which early rules apply are not simply c-command domains, but rather domains of the Phase. 



1.7

Summary of proposals made in the book

To summarize this chapter and the goals of the book, I will argue: (1.11) (i) There is more than one parse which phonological rules can refer to: (a) M (the first morphosyntactic parse) and (b) P (phonological domains constructed from the syntax). 



Introduction

17

(ii) There are early rules and late rules. (iii) Early rules refer to syntactic structure directly, but the only types of syntactic information available to early rules are Phase boundaries or the edges of Phases. (iv) Late rules refer to a specialized phonological parse P constructed from domains in which theta-roles are assigned; these are the only domains left after morphosyntactic information is erased by feature checking. (v) P is composed of constituent types which are not the same as those advanced in Prosodic Hierarchy Theory. (vi) P is unique, and all late rules refer to P and only P . (vii) Since M is not isomorphic with P , the problem of Domain Paradoxes is resolved. (viii) A clearer view of the syntax of languages results in a less complex theory of the phonology of languages: variation in the application of phrasal phonological rules is primarily a reflection of syntactic differences, and only secondarily a reflection of variations within phonology proper. 













18

Minimal Indirect Reference

C HAPTER 2

Domain Paradoxes In this chapter I show that the problem of Domain Paradoxes makes it impossible for there to be only one parse (P) to which the phonology can refer. Rather, an account which assumes that phonological rules apply to a morphosyntactic parse (M ) as well as to a phonological parse (P ) can account for Domain Paradoxes. In chapter 1 I suggested that there is no real evidence that the Prosodic Hierarchy is necessary for the explanation of differences in domains in languages, but I have yet to provide specific evidence against Prosodic Hierarchy Theory. In this chapter I show that even if there is a Prosodic Hierarchy above the word level it cannot be the case that it is crucial to the syntax-phonology interface, because it would make false predictions about the phonological domains of certain languages. 



2.1 Introduction According to classic indirect reference accounts of the syntax-phonology interface (Hale and Selkirk (1987), Selkirk (1984, 1996), Nespor and Vogel (1986), inter alia), phonological rules cannot refer directly to syntactic structure, and there is a unique phonological phrasing structure for each language. Phonological rules can refer only to these phrases (i.e., phrases in the Prosodic Hierarchy (Nespor and Vogel 1986)). Inkelas and Zec (1995) describe the predictions of the indirect reference theory in the following sentence (p.548): No matter how many rules exist in a system, Prosodic Hierarchy theory predicts them to utilize a maximum of four domains (“domain clustering”); moreover, those domains must enter into a hierarchical relationship (the “Strict Layer Hypothesis”). The Strict Layer Hypothesis (Selkirk 1984) referred to above states that within the Prosodic Hierarchy, each level must consist only of instances of the next lower 19

Minimal Indirect Reference

20

level of the hierarchy. The Strict Layer Hypothesis is formulated as a series of constraints in Selkirk’s recent work. These constraints are shown in (2.1). Notably, Selkirk (1996) suggests that both L AYEREDNESS and H EADEDNESS are inviolable, but E XHAUSTIVITY and N ONRECURSIVITY are violable. (2.1)

Selkirk (1996) N ONRECURSIVITY: Phrases are not recursive (violable) E XHAUSTIVITY: There is no skipping of levels in the hierarchy (violable) L AYEREDNESS: Lower levels cannot dominate higher levels (undominated) H EADEDNESS: All parts of the prosodic hierarchy except the syllable must dominate something (undominated)

In this chapter, I provide evidence against “domain clustering”. That is, I provide evidence against the idea that there are only a maximum of four domains above the Prosodic Word, i.e., only the domains of the Prosodic Hierarchy. In addition, I present evidence against the L AYEREDNESS constraint of the Strict Layer Hypothesis. This evidence shows that different domains are required for different phrasal phonological rules, i.e., contra the Strict Layer Hypothesis there is not one unique phrasing structure to which all rules can refer. This combined evidence shows that the predictions of Prosodic Hierarchy theory do not hold of all phrasal rules (Prosodic Hierarchy Theory could still hold of P, but P is not unique). In addition, it shows that a theory which allows reference to only one parse cannot be maintained.

2.2

The structure of the argument

The evidence that I discuss in this chapter comes from five languages. In three of these languages, Yoruba, Luganda and Korean, I show that the domains of phrasal rule application are overlapping; thus there is a violation of L AYEREDNESS and a Domain Paradox. In two other languages, Kimatuumbi and Mende, I show that more domains are needed than a classic indirect reference theory (such as the one proposed in Nespor and Vogel 1986) will allow, and that these domains cannot be domains within the Prosodic Hierarchy. Thus, these languages violate the “domain clustering” aspect of Prosodic Hierarchy Theory. Based on the evidence from these languages I propose that there are phonological rules which apply to a morphosyntactic representation (M ), and are sensitive to phrases based on XPs, or what we can think of as a Phase (Chomsky 1999). There are other phonological rules which apply to a phonological parse (P ) which is derived from syntactic domains. These rules apply to a post-Morphological Merger structure (Marantz 1988, et. seqq.). The suggested architecture for the grammar that I proposed in chapter 1 is repeated in (2.1). 



Domain Paradoxes

21

Output of Morphosyntax (M )











Early Rules

Phonological Domain Generator

Prosodic Representation (P )







Late Rules

Surface Phonological Representation Figure 2.1 The architecture of the grammar of MIR Crucially, I do not argue that there are no prosodic considerations to phrasing, but rather that there is no universal inviolable Prosodic Hierarchy above the Prosodic Word level, and that prosodic rules apply at a different level than direct syntax rules. I suggest that prosodic considerations are relevant only to a later parse, and that prosodic constituent types are not involved in the phonology-syntax mapping. For certain phonological rules, the phonetic environment is crucial. I propose that such rules are essentially different from pure syntax rules (rules which refer to M ) in two ways: (1) the syntax-phonology mapping at M consists of rules which make reference to syntactic domains and show no consideration of phonetic factors such as speech rate and phonological weight. (2) Rules which make reference to prosodic bracketing occur after the direct syntax rules, as can be demonstrated by the fact that these rules can be dependent upon speech rate and other prosodic factors. 



2.3 Domain Paradoxes and violations of domain clustering Prosodic Hierarchy theory cannot account for phenomena which I term Domain Paradoxes (e.g., Hyman (1987) calls these Layeredness violations). Domain Paradoxes are most convincingly illustrated by overlapping domains with different phonological rules making reference to different bracketing of domains. I illustrate these with examples from Yoruba, Luganda and Korean. Another problem for Prosodic Hierarchy Theory arises in violations of “domain clustering”, i.e., situations where it does not seem that there are enough levels in the Prosodic Hierarchy to cover the domains in which prosodic rules apply, or

Minimal Indirect Reference

22

where the levels of the hierarchy do not seem to fit the domains of rule application. In the next section I illustrate domain clustering violations with examples from Mende and Kimatuumbi. The solution that I provide for the Mende and Kimatuumbi data is then applied to Domain Paradoxes as well.

2.4

Domain clustering violations in Kpa Mende

In Mende (a Western Mande language closely related to Vai, Kpelle and Loma) there are two phrasal phonological rules which rely on different domains: a tone sandhi rule and a consonant mutation rule (Seidl 1998). The tone sandhi rule is an OCP-driven lowering rule which changes a H-tone to a L-tone when it is preceded by another H-tone within the same phonological domain. For example, in (2.2) the tone on the first vowel in f´aj`ı changes from a Htone to a L-tone because ny´ and f´aj`ı are within the same phonological domain. &

&

+

&

(2.2) ny´ f´aj`ı w´ -`ıt`a six fish buckets ‘six fish buckets’

*

&

ny´ -

$

% #

&

v`aj`ı w´ -`ıt`a

The consonant mutation rule lenites certain consonants when they are preceded by a terminal element within a domain which is potentially different from the domain for tone sandhi. As an example of mutation, consider that when /k/ occurs at the left edge of a word it mutates into [g] (3a) and when [ng] is at the left edge of a word it mutates into [w] (3b). ,

(2.3)

Consonant Mutation +

a. ngˇı k`an´aa´ his case ‘his case’ &

b. b´ı ng´ul´´ı your oil ‘your oil’

ngˇı

+

b´ı

$

$

% #

g`an´aa´

% #

&

w`ul´´ı

The consonant changes which occur in Mende in certain contexts are shown in (2.4). (2.4) Consonant Mutations: f v kp gb s j mb b p w +

+

+

+

+

Domain Paradoxes

23

+

nd l t l ny y k g ng y/w (depending on harmony with the following vowel) +

+

+

+

There are also certain consonants which never mutate regardless of the context they are found in. These consonants are shown in (2.5). (2.5) Immutable consonants: h, m, n, , l, y, w, g, b, d, gb, j, v 

2.4.1

-



The domains for tone sandhi and mutations

The previous literature has accounted either for consonant mutation (Cowper and Rice 1987) or tone sandhi (Dwyer 1969), but has not discussed the difference in the domains for tone sandhi and consonant mutation. This difference in domains will be the focus of this section. The object and verb are in separate domains for tone sandhi in (6a) and (6b). If the object shared a domain for tone sandhi with the verb, the H-tone on the verb would be changed to a L-tone. Likewise in (6b) the possessor would cause the H-tone on the inalienable noun ear to lower if the possessor and the possessed noun were in the same domain. But in both examples, those same elements are in the same domain for the purposes of consonant mutation. (2.6)

mutation and sandhi domains w´ok´ l` l´ (*`ı ny´a p´ok´ l` l´ ) a. `ı ny´a mutation domain (Subj+Pst) (me imitated Cert) tone sandhi domain (Subj+Pst) (me) (imitated Cert) ‘he imitated me’ $

b. ny´a w´oli (my ear) (my) (ear) ‘my own ear’ $

% #

% #

&

&

(*ny´a nd´oli) mutation domain tone sandhi domain

The different domains for tone sandhi and consonant mutation are illustrated in (2.6). Observe that the domains for consonant mutation are larger than the domains for tone sandhi. (Note that because example (6b) is an inalienable, (6b) does not have tone sandhi, although the alienable (3b) does. The syntactic differences and subsequent prosodic differences between alienables and inalienables is .

Minimal Indirect Reference

24

discussed in detail in Seidl 1998.) Several alternatives are available to explain these domains using the Prosodic Hierarchy. Alternative 1: Because the domains for tone sandhi are often smaller than the domains in which consonant mutation occurs, suppose we associate the domain for tone sandhi with the Prosodic Word. If the domain for tone sandhi is the Prosodic Word, then the domain for consonant mutation is the Phonological Phrase, as proposed by Cowper and Rice (1987), who argue that the initial consonant of a word mutates when it is preceded by another word that is in the same Phonological Phrase. In their system, Phonological Phrase boundaries correspond to the left edges of branching XPs (X ). In other words, provided that an XP is branching, a word adjacent to it will mutate if it is preceded by another word im. Thus knife mutates in (2.7) mediately to its left that does not constitute an X because it is preceded by a word to its left (child) that does not constitute an X . $

/ 0

$

/ 0

$

/ 0

&

(2.7) nd`op´o`ı b`ow`´ı child-the knife ‘the child’s knife’ $

% #

However, their proposal about the domain of mutation being the Phonological Phrase is not defensible. Recall that if the domain of mutation is the Phonological Phrase then the domain of tone sandhi must be the Prosodic Word. I conclude that the Phonological Phrase cannot possibly be the domain for mutation because tone sandhi occurs between Prosodic Words, not just within them, as in (2.2), repeated here as (2.8). Hence the domain of tone sandhi cannot be identified with the Prosodic Word. &

&

+

(2.8) ny´ f´aj`ı w´ -`ıt`a six fish buckets ‘six fish buckets’

&

ny´

$

% #

&

v`aj`ı w´ -`ıt`a

In addition, Cowper and Rice’s system is insufficient because it depends on the assumption that Phonological Phrases correspond to branching XPs. But there is independent evidence for another kind of phrasing (perhaps Phonological Phrasing), namely the domains in which tone sandhi occurs, and these tone sandhi domains show that Phonological Phrases are not where Cowper and Rice’s mutation rule would place them. In (2.9), for example, tone sandhi is blocked, but mutation does occur. Therefore, Cowper and Rice’s generalization about mutation being dependent upon Phonological Phrase boundaries cannot be correct.

Domain Paradoxes

25

(2.9) (ny´a) (k´ w` ) my foot ‘my foot’ 1

+ 1

ny´a g´ w`

mutation, no tone sandhi

In addition to this type of exception, there are constituents which mutate even though they are not preceded by a terminal element in their Phonological Phrase as diagnosed by tone sandhi or by Cowper and Rice’s algorithm. In particular, inalienably possessed nouns always mutate, but do not form Phonological Phrases with their possessors. (In earlier work (Seidl 1998), I showed that an account based on direct reference can account for these problems (partially by being sensitive to empty categories).) An inalienable noun preceded by nothing overt is shown in (2.10). +

w´oli (2.10) nd´oli ear ‘my own ear’ 2

$

% #

Alternative 2: Because the tone sandhi domain is larger than the Prosodic Word, as demonstrated by (2.8), but smaller than the consonant mutation domain as shown by (2.6), suppose that the domain in which tone sandhi operates is the Phonological Phrase. Therefore, the domain that mutation operates in must be an Intonational Phrase. There are several reasons why analyzing the consonant mutation domains as Intonational Phrases cannot be correct. (1) Nespor and Vogel (1986) suggest that the Intonational Phrase usually consists of parenthetical expressions, non-restrictive relative clauses, and lists; mutation domains are none of these. (2) It is not always true that the consonant mutation domains are larger than the tone sandhi domains; in fact, most of the time they are the same size (3b). (3) It would be odd if the subject or object in (6a) was itself an Intonational Phrase, i.e., an Intonational phrase rarely consists of one syllable and usually is sensitive to pause insertion, while consonant mutation is not sensitive to these things. Alternative 3: consonant mutation domains and tone sandhi domains are recursive Phonological Phrases. Selkirk (1996) argues that in order to account for the special behavior of clitics it is not necessary to have an additional prosodic constituent (the Clitic Group (Nespor and Vogel 1986)). Instead, rather than introduce another level, Selkirk proposes recursion of the Prosodic Word. Specifically, she suggests that both (11a) and (11b) are possible structures. (2.11) a.

1

clitic

3

lex

Minimal Indirect Reference

26

b.

1

3

clitic

3

lex If this is true, N ONRECURSIVITY (Selkirk 1996) is violated multiple times because in example (6a) (repeated below as (2.12)) the object pronoun must occupy two different phrases. (2.12) &

`ı ny´a w´ok´ l` l´ ((Subj+Pst) ((me) (imitated Asp) ) ) ‘he imitated me’ $



% #









It is also unclear how a rule/constraint could target one level of recursive phrasing and not another. Therefore I conclude that it is not possible to explain the Mende consonant mutation and tone sandhi domains using Prosodic Hierarchy Theory and present a fourth alternative: Alternative 4: A non-segmental morphological case marker or clitic on the object in (6a) and the possessor in (6b) is rebracketed as a single M with its host after the tone sandhi rule has applied. This rebracketing is illustrated in (2.13). (2.13) (cl) (host)

2

+

(cl host)

Thus, consonant mutation is not conditioned by purely phonological factors; the environments where mutation occurs in Mende appear to be syntactic. As evidence for this, consider the fact that most languages with mutation seem to have this phonological property due to the decay of some morphological affix. For example, Fula has been analyzed as having mutation triggered by the remnant of a noun class marker (Dwyer (1969)); Celtic has been analyzed as having mutation triggered by the remnant of a case marker (Borsley and Tallerman (1996)). Such rebracketing is a problem for the Prosodic Hierarchy Theory, because rebracketing entails that there is not one unique phrasing to which all rules refer, since some rules refer to rebracketed domains and other rules refer to an earlier parse. In support of Alternative 4, note that the two rules of tone sandhi and consonant mutation have a different character in that borrowed words and kinship terms are 



Domain Paradoxes

27

lexical exceptions to the consonant mutation rule but not to the tone sandhi rule. In (14a) we see that tone sandhi applies but consonant mutation does not. Thus I conclude that consonant mutation and tone sandhi apply at different levels. (2.14) a. ny´a k´enya my uncle ‘my uncle’

2

&

b. b´ı ng´ul´´ı your oil ‘your oil’

+

ny´a k`enya

+

b´ı

$

% #

loan words and native vocabulary loan word: TS applies, CM does not

&

w`ul´´ı

native word: TS and CM apply

Because kinship terms which are borrowed from mutable nouns do mutate in all contexts where they are being used as kinship terms, we know that kinship terms such as the one in (14a) are lexical exceptions. For example the word nj´e (‘sheep’) is mutated when it is used to mean ‘mother’ in (2.15). (2.15) ny´a nj´e my mother ‘my mother’

2

+

ny´a

$

% #

y`e

sheep used as kinship term: TS and CM

That there are lexical exceptions for consonant mutation and not to tone sandhi provides additional evidence in support of Alternative 4, because it suggests that the two rules apply at different stages in the derivation.

2.4.2

The solution to the domain clustering problems in Mende

I adopt an analysis similar to Kaisse (1985) to account for the domains of tone sandhi and consonant mutation in Mende. On Kaisse’s account, P1 rules, which operate on syntactic domains only, precede another set of rules, P2 rules. Kaisse describes the P2 rules as rules of fast speech. I, however, propose that P2 rules are not only rules of fast speech, but rather that they are rules which apply after Morphological Merger (see chapter 7 for further discussion). (This architecture was illustrated in (2.1).) Because grouping into larger domains causes information about structure to disappear, an individuated parse cannot be reconstructed (if there is no look-back). It follows from this that the tone sandhi rule must apply before the consonant mutation rule. (I suggest that the tone sandhi rule applies within the Phase (Chomsky 1999).) The problem with having larger domains precede smaller ones is illustrated schematically in (2.16), which shows that it is never clear what parse the grammar

Minimal Indirect Reference

28

would return to if the consonant mutation rules precede the tone sandhi rule. 



(2.16) X X X 





+

(X X) (X)



+

?(X) (X) (X)/?(X) (X X)

I propose that, in Mende, consonant mutation applies if a word is preceded by a morphological non-segmental case marker within the smallest YP containing the case-marker and the word (2.17). Specifically, a word is mutated when preceded by a case-marked X which does not end a maximal projection.  



(2.17) Word

+

 4



+mutated / X case+ 

X and Y need not be distinct.



Rebracketing is governed by rule (2.18), which brackets a clitic with a word in Morphological Merger (Marantz 1988). Rebracketing creates the environment for the rule in (2.17) to apply. 



(2.18) . . . X +case) (X . . .

2

+





. . . X ) (case+X . . .

Because mutation follows rebracketing, it has a different character (with respect to lexical exceptions, as discussed in 2.4.1) than the tone sandhi rule (as illustrated by (14a), where consonant mutation does not apply to certain forms, but tone sandhi does). In (2.19), because both the verb and the object are in a functional projection and there is no lexical XP boundary intervening between them the clitic is able to be rebracketed with the verb. Thus the mutated form appears because the clitic is bracketed with the verb. 

(2.19) &

(`ı) (ny´a+case) (p´ok´ l` l´ ) Sbj+Pst me imitate Asp ‘he imitated me’

+

(`ı) (ny´a) (case+

$

% #

&

w´ok´ l` l´ )

The subject clitic cannot cause mutation on the object because it is in a separate XP. This is illustrated by (2.20). (2.20) &

&

ng´ı [s´ l ´ı] kulˇı n`ı I banana drop Perf ‘I dropped the banana’ 



+

ng´ı

 " $

% #

&

&

s´ l ´ı

$

% #

gulˇı n`ı (*ng´ı

5

6 7

j´8 l8 ´ı 5

6 7

gulˇı n`ı)

The object in (2.20) is in a separate XP; therefore, the subject cannot be rebracketed with it. In (2.20), because the smallest XP containing the verb at Spell-Out is

Domain Paradoxes

29

AspectP (which also contains the object), the object triggers mutation on the verb. This is shown in (2.21). (2.21) Mende clauses: MoodP DP

Mood

SUBJ



9

Mood

AspectP

Mood DP

Aspect

OBJ

9



Aspect Verb +Aspect

vP DP t

VP DP

V

t

t

 

If we assume that Mende is head-final, then the fact that the verb appears before a PP complement (2.23), but after an NP complement in cases where the object must move up to get case (2.22) shows that verb movement to a functional head occurs before Spell-Out. With no PP:

 &

(2.22) ta l nik isia waa ma he Cert cattle kill Imperf ‘He is killing the cattle’ With a PP: (2.23) ta l li ma suku he Cert go at school ‘he is going to school’ The fact that A-bar movement blocks mutation falls out from this view of mutation, because if, for example, a case-marked object NP is fronted, then it has

Minimal Indirect Reference

30

been moved out of an XP (24a)-(2.25). Thus the object NP is no longer in the appropriate local relationship to cause mutation on a verb. The subject NP cannot trigger mutation on the following verb because there is always an intervening XP (24b). Therefore A-bar movement bleeds mutation. (2.24)

A-bar movement and mutation &

a. ng´ul ´ı m`ıa nd`op´o`ı kp`and`ıa´ oil it-is child-the heated ‘it’s oil that the child heated’ 

b. [ [ oil ] 





" $

% #

it-is [ [child ]

&





[[t ]







Verb ]



: ;  < # 

]

=

" "  

]

>



&

(2.25) nj`alˇa gul ˆı mia `ı kp`atˆ m`a gara cloth it-is Mood make Prog ‘It was gara cloth she was making’  " $

% #

When there is A-bar movement of the subject, the object is still present to trigger mutation on the verb. Thus mutation still occurs in (2.26): &

&

(2.26) t`ı a´ mia mˇu nd´o ng` mˇu y pˇ 3pl mood it-is 1pl desire Stat 1pl talk.. ‘It is they who want us to be talking...’

+

t`ıa´ mia mˇu

$

% #

&

&

l´o ng` mˇu y pˇ

If mutation were just a reflex of abstract Case assignment we would not necessarily expect to find that A-bar movement bleeds mutation, since A-bar traces get abstract Case. However A-bar movement of the object does block mutation (see (24a)), so morphological case must be involved. The Mutation rule in (2.17) also explains why mutation occurs between an intransitive verb and its subject (2.27). NP(Subj)-V &

&

(2.27) m´u v` mb´ `ıl´ ng´ul´ı´ı h`u we swung tree on ‘we swung on the tree’ $

% #

According to the mutation rule in (2.17), the subject ought not to trigger mutation on the verb in (2.27) because the verb is in a different XP. Thus it is still necessary to explain how the case morpheme “gets around” the Mood morpheme to cliticize with the verb. Another proposal for mutation in discussed by Tateishi (1990). He argues that all intransitive verbs in Mende are unaccusatives and it is unaccusative syntax which

Domain Paradoxes

31

causes them to mutate. This is clearly not the case given that the list of intransitives includes verbs such as fight, work, swim, and climb. It is implausible that these unergative intransitives would share the same syntax as the unaccusatives he mentions. Here I observe that intransitive verbs mutate not because they are unaccusatives, as Tateishi (1990) argues, but because they are in the appropriate relationship to the preceding NP to trigger mutation. For intransitives which are unaccusatives it is crucial to my argument that traces of A-movement do not block mutation the way traces of A-bar movement do. (2.28)

IP Subj +case

I



Mood

9

AspectP Aspect

9



V +Aspect

VP 

NP t

t



I suggest that Mood cliticizes with case and then the entire structure triggers mutation on the verb, i.e., it is a complex structure which triggers the mutation. This structure is shown in (2.29). (2.29)

Case Case Subj

Case

Asp Mood

V

Asp

Minimal Indirect Reference

32

The third person singular pronoun is a low tone proclitic. This clitic pronoun blocks mutation, for phonological reasons. In (2.30) the low tone appears between the case morpheme and the verb and there is no mutation on the following verb. 

?

&

&

(2.30) nd`op´o-`ı f` mb´ ng`a child-the it swung Prospective ‘the child swung it’  " $

% #

?

&

&

(*nd`op´o´ı v` mb´ ng`a)

The lack of mutation in (2.30) is unexpected because with full NPs and other pronouns there is mutation when the object appears before a verb. In addition, because mutation may be triggered allomorphy and not an actual phonological process this clitic pronoun should not block mutation if morphological case is syntactic. So how do we explain the fact that this proclitic blocks mutation in examples such as (2.30) and (2.31)? Object of a ditransitive verb is just low tone: (2.31)

?

&

t` b´ı w`umb´a it lift your top ‘put it on top of your head!’ 

" $

% #

Whereas when the object is overt there is mutation: &

&

l´ b´ı w`umb´a (2.32) haˆak ˆı load lift your top ‘put the load on top of your head!’ $

% #

So far we have been assuming that case morphemes are inserted syntactically as soon as they can be. However, the data above show that the insertion of case morphemes must follow syntactic cliticization so that the case morpheme does not trigger insertion of the mutated allomorph when there is a 3rd person clitic pronoun. (Recall from the previous section that left leaning clitics, such as Mood, do not block the insertion of a mutated allomorph.) This follows from Distributed Morphology (Halle and Marantz (1993)), because in Distributed Morphology there is morpheme insertion after syntax (Embick 1997). The fact that a case morpheme’s realization is dependent upon the surface realization of a pronoun shows that case morphemes must be inserted after syntactic lowering of the pronoun. When a pronoun is realized as a low tone clitic the case morpheme is not rebracketed because, as the mutation rule states, rebracketing only occurs with bare X s (see (2.18)), not when there is a syntactic clitic attached to an X by lowering (see (33c)). 



Domain Paradoxes

33

(2.33) a. b.

3sg clitic A @ XB A A C D E @ @ +case A @ 3sg clitic+X B A insertion c. @ @ +case A @ 3sg clitic+XB A @

@

@

@

@

A C

D

A C

D

The low tone proclitic and rebracketing 3Sg clitic+XB A A C D syntax case E @ @ +case A @ 3sg clitic+X B A A C D E

@

@

+case A @ 3sg clitic+XB

A

A C

D

PF

Because the low tone proclitic is a syntactic clitic which cliticizes to the verb or preposition before the case morpheme can be rebracketed with it, it prevents mutation. On this account, phonological operations apply as soon as they can. The information relevant to the tone sandhi rule is syntax: tone sandhi applies within each XP (e.g., (2.2) is an XP and so are the subject, object and verb in (2.19)), thus tone sandhi applies early before this syntactic information is lost. Consonant mutation applies as soon as its structural description is met. Because its structural description is not met till rebracketing, it applies later (that is, it may apply earlier but it would not be possible to see it if it did because the environment is not met for the rule early on) (see architecture in (2.1)). The tone sandhi rule applies to the morphosyntactic parse (M ) and the consonant mutation rule applies to the phonological parse (P ). On this analysis no domain clustering violation arises, because there are enough syntactic and morphological categories/domains to account for the domains of phonological rule application. 



2.4.3

Supporting evidence from English

Some supporting evidence that the late phonological parse (P ) can access rules which apply to lexically restricted forms comes from palatalization in English. Because palatalization in English is a fast speech rule (Kaisse 1985), it must occur late, after speech rate information is incorporated. Some examples of palatalization in English are shown in (2.34). 

(2.34)

Kaisse (1985) a. Could you pass the peas? cou[ˇj]ou pass the peas b. Can’t you leave now? can[ˇc]ou leave now c. As you pass the tower turn left. a[ˇz]ou pass the tower turn left 2

2

+

+

2

+

Kaisse (1985) notes that this palatalization does not apply to all words. For example, palatalization is not found between dreaded and yearly in (35a). Some further examples of a lack of palatalization are shown in (2.36).

Minimal Indirect Reference

34

(2.35)

Kaisse (1985) a. . . . my dreaded yearly checkup *. . . my dreade[ˇj] yearly checkup b. Just use your head *Jus[ˇc] use your head c. He’s my third urologist in as many months *He’s my thir[ˇj] urologist in as many months +

+

+

(2.36) a. b. c. d.

Some questionable cases It was a bad yearly checkup *It was a ba[ˇj] yearly checkup You shouldn’t use the vacuum ?You shouldn[ˇc]use the vacuum ?Shouldn’[ˇc]anni be here by now Shouldn’t Yanni be here by now? Where’s urology ?Where[ˇz]urology +

+

+

+

Palatalization seems to occur most often before an unstressed pronoun. Therefore it might be advantageous to assume Selkirk’s (1996) analysis in which unstressed pronouns are rebracketed with the Prosodic Word. However, this analysis cannot explain why palatalization in (36b) is not particularly bad, and why you can be focalized and nonetheless still undergoes palatalization in (37a).  

(2.37) a. it sure isn’t YOU

+

it sure isn’[ˇc]OU

b. I’d eat ivy too. Wouldn’t you? song

+

from the sitcom Phyllis

wouldn’[ˇc]OU

from the children’s

In addition, interestingly, there is only one reading possible for the palatalized version of (2.38), which may mean that palatalization of you may be allomorphic or at least a rule which only applies to the lexically specified form you. (2.38) where[ˇz ] anus can only mean the anatomical region Where’s Uranus/your anus F G

My preliminary studies of don’t use, don’t you and don’t choose in online spoken corpora from the Linguistic Data Consortium and in quickly read speech reveal that the palatalized consonant in don’t you has the same duration as don’t choose and far longer duration than don’t use (when there is palatalization at all). Given these data, it seems that the palatalized don’t you is an example of a rule that only applies to lexically specified forms as well as being a fast speech rule. Based on the fact that palatalization is a fast speech rule and thus applies to P and not to M , I conclude that rules which only apply to lexically specified forms may apply late in a derivation, and so must apply to P and not to M . 







Domain Paradoxes

2.4.4

35

Supporting evidence from Korean

Certain prosodic rules from Korean which are dependent on speech rate can also be shown to apply late. In Korean there is a formally similar paradox to the one in Mende: /n/-insertion in Standard (Seoul) Korean occurs between a Prosodic Word ending in a consonant when it is followed by a Prosodic Word beginning with a high front vowel, viz., the rule operates at the prosodic word juncture (Han 1994) (39b). Consider the examples in (2.39), here /n/-insertion occurs in different types of phrases.  !

(2.39) a. (c am yep’ -ta) very be pretty-DECL ‘(It) is very pretty’ H

I

2

Han (1994)



+ 

c am nyep’ da H

I

b. (k sacin- l y hi-ka c’ik- s’-ta) k sajin ll y higa c’ig tt’a that picture-ACC Younghee-NOM take-PST-DECL ‘Younghee took that picture’ I



I

I

J -

I

J

J -

+



J

In (39b), /n/-insertion applies between that picture and Younghee. Notice that /n/-insertion does not occur in (2.40), which reveals that there are two domains for the rule. If there were not two domains for /n/-insertion we would expect an [n] to be inserted between t´o sae and yep’ ta. -

-

I

(2.40) (t´o sae ) (yep’ ta) Brother is handsome ‘brother is handsome’ -

-

Eon-Suk Ko, p.c.

I

In Korean, there is also an OCP-motivated rule which deletes a H-tone on a word when it is in the same domain as another word with a H. Domains for the OCP rule are shown in (2.41). (2.41)

Eon-Suk Ko, p.c. a. (t´o sae yep’ ta) Brother is handsome ‘brother is handsome’ -

-

I

(*t´o sae y´ep’ ta) -

b. (k s´acin- l) (y h´ı-ka c’ik- s’-ta) that picture Younghee took ‘Younghee took that picture’ I

I

J -

J

-

I

Minimal Indirect Reference

36

But the OCP does not hold between that picture and Younghee in (41b). Thus, the domains for the OCP tonal rule in Korean are both larger and smaller than the domains for /n/-insertion. (2.42)

The overlapping domains a.

(k s´acin- l)



I

I

b. ( t´o sae 

-

K

L

-   



(ly h´ı-ka c’ik- s’-ta) J -

 :  M # 

"  

J



yep’ ta I

  

K

L

  

 :  M # 

 :  M # 



"  

)

K

L

"  



In Korean the /n/-insertion rule is dependent upon speech rate (i.e., in slow speech /n/-insertion does not apply in (39b), but in fast speech /n/-insertion does apply there), whereas the OCP rule is not dependent upon speech rate. In chapter 4, I suggest a solution in which the /n/-insertion domains are prosodic domains constructed from the domains in which theta-roles are assigned. This solution is discussed in greater detail for other languages in chapters 4 and 5, where I also suggest that in fast speech these domains may be ignored and a default phrasing may be adopted. To outline precisely what I will suggest in chapter 4: in (42a) a theta-role is assigned to the object by the verb and to the subject by v. I propose that the domains that theta-roles are assigned in are relevant only to the phonology if the constituent stays in situ; hence only Younghee can project a boundary from the domain it receives its theta-roles in. This is shown schematically in (2.43). (2.43)

Structures for Korean syntax

a.  

k s´acin- l I

 

I



 N

ly h´ı-ka



J -









c’ik- s’-ta t t 

J







 O P  O



 '







b.

Q

-domains project L in slow speech

(that picture) (Younghee took)

c.

syntax  

k s´acin- l I

I

 



 N

ly h´ı-ka



J -

 









c’ik- s’-ta t t J





 O P  O



 

'





d.

Q

-domains ignored in fast speech

(that picture Younghee took)

The OCP rule on the other hand applies within the domain delimited by the Phase (Chomsky 1999). Both the adjunct k s´acin- l and the subject and verb y h´ı-ka I

I

J -

Domain Paradoxes

37

c’ik- s’-ta in (42a) form Phases, and so the OCP rule applies within these domains. But because the entire phrase t´o sae yep’ ta in (42b) forms a Phase, this phase is the domain of the OCP rule. The constituent structure for both of these structures are shown in (2.44). J

-

(2.44) a.

-

I

TP

NP

TP

that picture

vP NP Younghee

b.

v

9

took

VP





TP

NP

V

t

t

9





vP NP brother

v is handsome



9

VP t



In Korean, theta-domains are ignored in fast speech, only the later rule (/n/insertion) is sensitive to this use of the default structure without theta-domains being attended to by the phonology thus the string is bracketed into larger units. The data from Korean fast speech provide additional evidence that certain rules are sensitive to late bracketing of a string and others are sensitive only to morphosyntax.

Minimal Indirect Reference

38

2.5

Kimatuumbi: another violation of domain clustering

Kimatuumbi provides another illustration of a violation of domain clustering. In Kimatuumbi (Bantu), Odden (1996, 1995) describes two phonological rules which operate on distinct syntactic domains. Odden (1996) argues that a Vowel Shortening rule (VS) is sensitive to the right edge of X (2.45). $

(2.45) Vowel Shortening R

S

/ 0

Adapted from Odden (1995) / ...

 T

U

Y ...

 V 

S

This rule applies when underlying long vowels shorten in a word when the word they are in is not domain-final.  )

The application of the Vowel Shortening rule in Kimatuumbi is illustrated in examples (2.46)-(2.47).

(2.46) naa-k´ala/angite ch´oolya I-fried food ‘I fried food’

+

naa-k´alangite ch´oolya

Odden (1995)

According to Odden (1995), VS occurs on the verb in (2.46) because the verb is followed by an NP within VP and there is no intervening right edge of X between the V and the N. $

/ 0

(2.47) kik´olo/ombe cha/´ake mbwi¸/¸ig´a lya´angu E kik´olombe chak´e mbwi¸g´a lya´angu cleaning-shell of friend mine ‘the cleaning shell of my friend’

For example (2.47), Odden assumes the structure in (2.48) which has no right edges of maximal projections between the elements. Thus his proposal explains the VS domains in (2.47).  *

Domain Paradoxes

39

(2.48)

NP N shell

PossP Poss

NP

of

N

PossP

friend

mine

The second rule Odden (1996) discusses for Kimatuumbi (an SVO language) is This rule can be a phrasal tone insertion rule. The rule is stated in (2.49). restated as targeting the left periphery of a Phase (Chomsky 1999). 

 ,

 .

(2.49) Phrasal Tone Insertion (PTI): W 2

+

Adapted from Odden (1996)

H/

YP XP

X

The rule in (2.49) inserts a H-tone between a specifier/adjunct and anything else. Namely, the H-tone is inserted only when an XP is strictly adjacent to another element within YP, but if there is a trace present strict adjacency does not obtain and the rule does not apply. Because all three are adjuncts in (50a) the PTI rule applies between Friday, kiyogoyo and Mamboondo. Crucially, in (50a), the two nouns are left-dislocated and adjoined to IP. Here Friday is an adverb and Kiyogoyo and Mamboondo are leftdislocated and adjoined to TP. 

(2.50)

Odden (1995) +

a. i´ıjuma kiy´ogoyo Mamboondo naamp´ei l´ı i´ıjum´a kiy´ogoy´o Mamboond´o naamp´ei l´ı Friday kiyogoyo Mamboondo I-him-gave not ‘I did not give Mamboondo a kiyogoyo on Friday’ +

b. man´a naantumbil´e Mamboondo nduwae kuunnw´aaya man´a naantumbil´e Mamboond´o nduwae kuunnw´aaya if I-him-fell Mamboondo I-would him-nurse ‘If I had fallen on Mamboondo, I would have nursed him’

Minimal Indirect Reference

40

(50b) has a bit more complicated structure in that an empty complementizer allows strict adjacency between Mamboondo and the auxiliary verb as shown in (2.51). (Universally, in conditional structures there is a filled or unfilled C. This one is not filled as it would not be in the equvalent structure in English.)

(2.51)

CP

CP

CY Z

CY if

IP

IP

I

I

IY would

IY

VP him

had fallen on

nurse

VP Mamboondo

Observe that the domains for PTI and VS are, crucially, not always the same: In (52a), the word red ends a domain as indicated by a lack of shortening, but does not receive a phrasal-H (in addition the entire phrase does not receive a H because there is no phrase following it). Thus in (52a), there are two VS domains and no PTI domain, indicating that the domains for PTI and VS are not the same. (2.52) a. ((kik´olo/ombe kike´ele) shell red ‘My red shell’

b. (ku¸-su¸/u¸le (ngulu)) to-school large ‘to the large school’ 

'



" M #

: H

cha´angu) mine +

" M #

: H

+ " M #

: H

Odden (1995) kik´olombe kike´ele cha´angu

ku¸-su¸l´e ngulu´

Odden’s (1996) rules work for the data he discusses, but crucially violate Prosodic Hierarchy Theory and “domain clustering” because, on his account, the different rules target different structures and these structures are more numerous than one particular version of Prosodic Hierarchy Theory will allow.

Domain Paradoxes

41

In order to preserve Prosodic Hierarchy Theory there may be an alternative account. I will mention a few of these alternatives here. Alternative 1: one of these rules targets the Intonational phrase and the other targets the Phonological Phrase. This cannot be the case because there are several other rules which target the Intonational Phrase and are sensitive to prosodic considerations such as pause insertion (see Odden 1996: 255–265) and none of these rules act on the same domain as PTI or VS. Alternative 2: One of the rules targets the Prosodic Word and the other targets the Phonological Phrase. This cannot be the case because both rules operate on domains which are clearly larger than the Prosodic Word, as evidenced by (50a). Since neither of these alternatives are viable within Prosodic Hierarchy Theory I will discuss a recent account of Kimatuumbi which does not violate Prosodic Hierarchy Theory.

2.5.1

A solution to the domain problem in Kimatuumbi that obeys Prosodic Hierarchy Theory

Truckenbrodt (1995,1999) proposes a solution that predicts the difference between PTI domains and VS domains in Kimatuumbi, but also obeys Prosodic Hierarchy Theory. I will give an outline of Truckenbrodt’s account (discussed in greater detail in chapter 3) and then briefly discuss the shortcomings of this account. As will be discussed in chapter 3, Truckenbrodt (1999) proposes three OptimalityTheoretic constraints to account for the domains of application for VS and PTI in Kimatuumbi. The relevant constraints for the discussion in this chapter from Truckenbrodt’s account are: A LIGN -XP,R, ,R, A LIGN -XP,L, ,L: Each XP boundary is aligned with a Phonological Phrase (i.e., *( ) ). 1

1

T







4







W RAP -XP: Each lexically headed XP is contained within a single Phonological Phrase, i.e., is not split by a Phonological Phrase. Functional projections and traces are not constrained by W RAP -XP. (e.g., IP can be split into two Phonological Phrases) 

!

N ON R EC : No Phonological Phrase may contain another phonological phrase. (This is a constraint against recursivity of the phonological phrase.) In order to generate the output that Odden (1996) does for PTI without referring to two different syntactic domains, Truckenbrodt (1999) proposes a constraint which “misaligns” phrasal-H with the right edge of an XP (2.53).

Minimal Indirect Reference

42

(2.53) A LIGN (P,L;H,R): Align the left edge of every Phonological Phrase with the right edge of a H-tone. For Truckenbrodt, VS applies to all words except for the phrase-final one. In other words, VS does not occur when a word-boundary and a phrasal boundary coincide, i.e., as in Odden (1996), shortening can be blocked only by the right edge of a Phonological Phrase. Truckenbrodt argues that Kimatuumbi has recursive Phonological Phrasing because both VS and PTI are blocked from applying in (54a). (2.54)

Odden (1996) a. ((naamp´ei kik´oloombe) Mamb´oondo) I-him-gave shell Mamboondo ‘I gave Mamboondo the shell’ 



b. *(naamp´ei kik´oloomb´e) (Mamb´oondo) 

c. *(naamp´ei kik´olombe Mamb´oondo)





If the structure were not recursive as in (54b) PTI would wrongly be expected to apply. Alternatively, if there was no phrasal boundary between the two NPs as in (54c) VS would wrongly apply to the first NP. Because shortening is blocked and PTI does not apply in (54a), Truckenbrodt infers the presence of a recursive structure. N ONREC is violated in (54a) because the constraints A LIGN -XP,R and W RAP -XP are ranked high. Thus, A LIGN -XP,R and W RAP -XP are satisfied at the expense of N ONREC. This ranking is shown in the tableau in (2.55). (2.55)

 

V NP NP





☞a. ((V NP) NP) b. (V NP NP) c. (V NP) ) (NP) 1

1

1

1

1

A LIGN -XP,R W RAP -XP N ONREC * *! *!

This ranking of constraints also allows Truckenbrodt to account for (2.56), because here the recursive phrasing would account for the shortening domains as well as the lack of PTI on the adjective red. That is, because red is not followed by a left boundary, there is no environment for the rule of PTI.

Domain Paradoxes

43

(2.56) ((kik´oloombe kike´ele) shell red ‘My red shell’

" M #

: H

cha´angu) mine

+

kik´olombe kike´ele cha´angu

" M #

: H

However, there is a key stipulation about the syntax which is necessary for Truckenbrodt’s (1999) argumentation about the phrasing illustrated in (54a) to go through. He assumes the syntax in (2.57) for double complement constructions with Larsonian (1988) VP-shells which is in itself unproblematic, but Truckenbrodt must also argue that the lower VP is ignored by Wrap-XP because it contains a trace of head movement of the verb to the higher lexical VP. Because he suggests that the phonology cannot ‘see’ projections which contain traces, the only constituent boundaries that the phonology sees is the higher lexical VP. (Also, note that although this account uses constraints such as Align-XP which are indirect (refering only to the edges of XPs and thus flatter structure) it also contains a constraint (Wrap-XP) which is more direct in that it assumes that the phonology can see a constituent as a whole.) (2.57)

VP



V V

VP



9 



NP

V V

9

NP



As will be discussed in chapter 3 and 5, in double complement/applicative constructions in Bantu languages the indirect object (IO) is often considered to be generated in a functional projection (not a lexical one as he proposes) and verb movement is generally considered to occur (higher than V) before Spell-out (Marantz (1993), Keach and Rochemont (1994), inter alia). If verb movement to a higher projection is assumed, then there is no lexically headed projection which contains the verb, the IO and the DO. If there is no lowering and the verb is as morphologically complex as it appears in Bantu then it is unclear how the verb is so inflected without raising to a projection higher than VP. Thus, I argue that there is no way to get a recursive structure by ranking W RAP XP high. (In (2.58), what Truckenbrodt considers to be lexical projections are in boldface, there is verb movement, and the IO is in a functional projection.) (2.58) pro -V t V V 





 

[IO] [[ DO] ] 

IO) (IO)





DO) (DO)



 [



 



; \ 

'



Syntax By Align-XP,R By Wrap-XP

Minimal Indirect Reference

44

In (2.58) no violations of N ONREC are incurred by the satisfaction of A LIGN -XP and W RAP -XP. Thus including a reasonable assumption about the syntax of these languages in Truckenbrodt’s account makes it impossible to explain the difference between PTI domains and VS domains in Kimatuumbi. In addition, even if Truckenbrodt’s syntax for double complement constructions is the correct one, this account cannot explain the lack of VS on the complementizer in (2.59). (2.59) ((ke´enda) atell´ıik´e) if he-had-cooked ‘if he had cooked’ " M #

:

" M #

:

H

H

+

ke´enda atell´ıik´e

Odden (1996)

The predictions of Truckenbrodt’s analysis are shown in (2.60). Because there is a left bracket (of IP or VP) following the complementizer; PTI ought to apply, yet it does not. Thus, Truckenbrodt’s ranking of constraints in fact predicts wrong phonological domains. 

(2.60) ke´enda atell´ıik´e ke´enda (atell´ıik´e) *k´end´a atell´ıik´e 





 

+ L



(ke´enda atell´ıik´e) By Align-XP and Wrap-XP The VS and PTI predicted

The only way to predict the form in (2.60) is to assume an account with verb movement as shown in (2.61) and to argue, contra Truckenbrodt 1999, that traces are visible to the relevant rules. Specifically Truckenbrodt argues that constraints relating syntactic and prosodic categories do not apply to empty syntactic elements, but here they must if verb movement occurs. In this way the A LIGN constraint, which targets lexical projections, will not insert a boundary between the complementizer and the verb, and PTI on the complementizer in (2.59) is not predicted. )

(2.61)

CP Comp

IP Verb



VP t



Domain Paradoxes

45

There is still one problem, however, because even with verb movement, there is no right bracket between the complementizer and the verb in (2.60). Thus VS ought to apply according to Truckenbrodt’s analysis, but it does not. The only way to accomplish this on Truckenbrodt’s account is to make the complementizer a lexical category, which is a dubious stipulation. I argue on the contrary that no VS applies in (2.59) because the complementizer in (2.60) is a lexical exception to VS. On Truckenbrodt’s account, VS occurs after vocabulary insertion at the level of Phonological Phrasing. Thus it should not be possible for there to be lexical exceptions to VS, yet there are other exceptions to VS shown in (62a) and (62e). (These words are borrowings from Swahili, English or Arabic.) (2.62)

Odden (1996) a. ki¸tu¸u¸ngu¸u´¸ cha´angu onion my ‘my onion’ b. bo´oksi¸ ya´angu book my ‘my book’ c. mbwaa ywa´angu dog my ‘my dog’ d. sa´a ya´ang´u watch my ‘my watch’ e. baas´a ya´angu envelope my ‘my envelope’

In contrast with shortening, there are no lexical exceptions to PTI.

2.5.2

In support of Odden’s account

On my account both PTI and VS apply to the morphosyntactic parse (M ) (see (2.1) for a derivation which shows reference by rules to this parse.). As discussed in chapter 1, there is no special level of representation to which the phonology refers, rather, certain phonological processes are selectively sensitive to different pieces of syntax. 

Minimal Indirect Reference

46

Recall from the introduction also that the idea that the phonology is selectively sensitive is not a new one. Calabrese (1995) proposes something similar in his discussion of feature geometry for the visibility of certain features (p. 373).

Phonological rules differ with regard to the features that are visible to them: only marked features, only contrastive features, or all types of features.

I have suggested that, in the same way that certain degrees of feature specification are visible only to certain phonological rules, certain aspects of syntax are visible to only certain phonological rules. Specifically, PTI inserts a tone before an adjunct/specifier (2.63) and VS is sensitive to the right edge of X (2.64). $

/ 0

*

+

(2.63) i´ıjuma kiy´ogoyo Mamboondo naamp´ei l´ı i´ıjum´a kiy´ogoy´o Mamboond´o naamp´ei l´ı Friday kiyogoyo Mamboondo I-him-gave not ‘I did not give Mamboondo a kiyogoyo on Friday’ VS applies to all elements in (2.64) following Odden’s rule (2.45) and makes reference to a different syntactic domain than the PTI rule. (2.64) kik´olo/ombe cha/´ake mbwi¸/¸ig´a lya´angu E kik´olombe chak´e mbwi¸g´a lya´angu cleaning-shell of friend mine ‘the cleaning shell of my friend’

Now that I have discussed domain clustering violations, I will discuss some domains paradoxes and attempt to account for them using the same machinery used to account for the domain clustering violations.

2.6

L AYEREDNESS violations in Luganda

Hyman, Katamba, and Walusimbi (1987) exhibit violations of the Strict Layer Hypothesis in Luganda (Bantu). The data they discuss show that in Luganda there is a L-tone deletion rule that deletes L-tones between two H-tones. Hyman, Katamba, and Walusimbi (1987) refer to this process as H-plateauing (HP). ,

Consider example (2.65), here the L-tones on the verb are deleted when they are followed by a noun with a H-tone.

Domain Paradoxes

47

(2.65)

Hyman et al. (1987) a. t´u-l`ab-`aa` W´al´us`ı`ımb`ı we-see-indic Walusiimbi ‘We see Walusimbi’

+

t´u-l´ab-´a W´al´us`ı`ımb`ı

+

b. t´u-l`ab-y`e W´al´us`ı`ımb`ı t´u-l´ab-y´e W´al´us`ı`ımb`ı we-see-perf Walusiimbi ‘We have seen Walusiimbi’ c. tw-´aa´ -g´ee´ nd-´a t´ut`utu we-past-go-indic slowly ‘We went slowly’ d. tw-´aa´ -g´i-s´al-´a n´a o´ -b´u-sˆo we-past-it-cut-indic with IV-NC-knife ‘We cut it with knives’ e. t´u-l`ab-`aa` b´ı-k´op`o bi-n´en`e we-see-pres NC-cup NC-big ‘We see big cups’

+

t´u-l´ab-´a b´ı-k´op`o bi-n´en`e

For example, in (2.65) the low tones on the verb are changed to high when they are followed by a noun with high tones. For example, the high tone on cup changes the low on see to H in (65e). This rule does not apply when the noun has an initial vowel morpheme (IV) (2.66). The initial vowel morpheme correlates with wide focus; without the initial vowel only the object would be in focus. (2.66)

Hyman et al. (1987) a. t´u-l`ab-a e´ -b´ı-k´op`o we-see-pres IV-NC-cup ‘We see cups’ +

b. t´u-l`ab-a b´ı-k´op`o we-see-pres NC-cup ‘It is the cups we see’

+

t´u-l`ab-´a e´ -b´ı-k´op`o

t´u-l´ab-´a b´ı-k´op`o

There is also a rule of Vowel Shortening (VS) in Luganda which shortens vowels at the ends of a Clitic Group (67a) (Clitic Group is their terminology, we could also call this a recusive Prosodic Word). This rule is blocked from applying when a word with a final long vowel is followed by a clitic (67b). .

Minimal Indirect Reference

48

(2.67)

Hyman et al. (1987) +

a. t´u-ly-`a/`a t´u-ly-`a we-eat-indic ‘We are eating’ +

b. t´u-ly-`aa` =kˆo t´u-ly-´aa´ =kˆo we-eat-indic=little ‘We are eating a little’ Hyman et al. (1987) point out that the domains for HP and VS can be identical (68a), or the HP domain can be larger than the VS domain (68b), or the HP domains can be smaller than the VS domains (68c). This is an obvious problem for Prosodic Hierarchy Theory. (2.68)

Hyman et al. (1987) a. ( t´u-ly-´aa´ =kˆo we-eat-indic=of ‘We eat a little’ 

 \   ] #

b. ( t´u-ly-´a we-eat-indic ‘We eat dog’ 

c.

 \   ] #

)

; \ / #

H

mbw´a dog 

H

(te-t´u-ly-`aa` ) (=kˆo) Neg-we-eat-indic=of ‘We don’t eat any’ 

; \ / #

 \   ] #

)

; \ / #

H

Vincent Ochwo, p.c.

; \ / #  \   ] # H

Thus, in Luganda, because L AYEREDNESS requires that all smaller domains are properly contained in larger domains, domains for HP and VS clearly violate L AYEREDNESS. The evidence in (2.68) leads Hyman et al. to propose that the VS rule operates on a different autosegmental tier than the TP rule. In this way the rules can operate within different domains without violating the Strict Layer Hypothesis. They argue that both rules operate on the Clitic Group, but the tone rule operates on the tonal tier and the vowel length rule operates on the skeletal tier. This solution works for the problematic data; however, by allowing each tier to have associated with it its own domain, they permit that each feature to conceivably have a different domain, thus greatly complicating the task of acquisition.

2.6.1

A solution to the Luganda domain paradoxes

Suppose, however, that VS, like mutation, applies after a rebracketing of a clitic with its host and that HP, on the other hand, applies before this rebracketing.

Domain Paradoxes

49

Suppose further that the domain for VS is a rebracketed structure. If a vowel is domain-final it shortens. If in the late parse the clitic is bracketed with the verb, the domain for VS is the entire phrase in both (68c) and (68a). The HP rule is sensitive only to syntax and to the syntactic difference between (68a) and (68c). I surmise that in order to incorporate, the clitic should raise. However, in (68c) negation blocks the syntactic incorporation of the clitic ko into the XP. That is, for ko to be in the scope of negation it has to remain inside VP, otherwise the sentence would mean, ‘There is some we don’t eat’. This kind of blocking of movement across an operator is found in other languages as well. For example, descriptions in the scope of modal operators can be interpreted as applying to the actual world or to the possible world selected by the modal. In Greek and Italian, if the description is in the subjunctive (which needs to be licensed by a modal operator), then the description must be in the non-actual world, i.e., it cannot “raise” past the modal. But if the description is in the indicative, then it must (in Italian) or may (in Greek) refer to the actual world. An Italian example of this contrast is given in (69a)-(69d). 







(2.69)

Filippo Beghelli, p.c. a. Cerco una donna che parla Italiano I-am-looking-for a woman who speaks(INDIC) Italian ‘I’m looking for a woman who speaks Italian (presupposes there is such a woman)’ b. Cerco una donna che parli Italiano I-am-looking-for a woman who speaks(SUBJUN) Italian I’m looking for a woman who speaks Italian (no such presupposition)’ c. *?Non cerco alcuna donna che parla una lingua NEG I-am-looking-for any woman who speaks(INDIC) an exotic esotica language ‘I’m not looking for any woman who speaks an exotic language’ d. Non cerco alcuna donna che parli una lingua NEG I-am-looking-for any woman who speaks(SUBJUN) an exotic esotica language ‘I’m not looking for any woman who speaks an exotic language’

(69c) is somewhat worse than (69d) (at least for speakers who have a robust distinction between subjunctive and indicative). This occurs possibly because the negative polarity item (alcuna) “scopes out”, i.e., scopes higher than the inten-

Minimal Indirect Reference

50

sional verb cerco, which means it is not in the scope of the negation, but only on the non-specific reading.



For the Luganda data above, I suggest that the Phase is a domain for the TP rule and for all rules which make reference to syntax. Since VP is a Phase, TP applies within VP. This analysis is illustrated in (2.70) below. (2.70) a. b. c. d.

t´u-ly-`aa` kˆo t t t´u-ly-`aa` kˆo ( ) ( t´u-ly-`aa` kˆo ) (t´u-ly-´aa´ kˆo ) 







N

^ _

`















(2.71) a. te t´u-ly-`aa` kˆo b. te-t´u-ly-`aa` (kˆo) c. (te-t´u-ly-`aa` ) (kˆo) d. (te-t´u-ly-`aa` ) (kˆo) 

N

^ _

`

 '



 '



Sample derivation with clitic in VP syntax VP projects brackets default rule inserts matching brackets TP applies to these domains

Sample derivation with clitic not in VP syntax VP projects brackets default rule inserts matching brackets TP applies to these domains

In negative sentences the NP cannot raise (68c). It follows that an XP boundary intervenes between the NP and the verb, and there are two domains for tone spreading, which applies before rebracketing. Both (70a) and (71a) are subject to a morphological merger rule which rebrackets a clitic with its host, thus in the late parse the length domains are the same (2.68).

If the information relevant to the VS rule is added late, this predicts that in slow Luganda speech it might be the case that the clitic does not get rebracketed with the verb, and the long voweled form will not appear. This is true for Luganda: speech rate affects the shortening rule (72a-72b), but does not affect the tonal rules (72a-72b).

Domain Paradoxes

51

(2.72)

Vincent Ochwo, p.c. +

a. t´u-nyw´aa` -k´oo` t´u-nyw´aa´ -kˆo we-drink-little ‘we drink a little’

a / : # : ;   < H

+

b. t´u-nyw´a/`a-k´oo` t´u-nyw´a-kˆo we-drink-little ‘we drink a little’ c. e` mbw´aa y`o dog your ‘your dog’

+

d. e` mbw´a/a y`o dog your ‘your dog’

+

e` mbw´aa y`o

e` mbw´a y`o

: \ " b

: ;   < H

a / : # : ;