The English Left Periphery in Linearisation-based …...4.2 Previous analyses 64 4.2.1...

The English Left Periphery in Linearisation-based HPSG

Takafumi Maekawa

A thesis submitted for the degree of Doctor of Philosophy

in Language and Linguistics

Department of Language and Linguistics

University of Essex

2007

ii

Abstract

In many theories of syntax, linear order is inseparable from syntactic structure. In

Minimalism/Principles-and-Parameters theory, for example, the only information contained in

tree diagrams is that of constituent structure, and not of linear order since the latter can be

predicted from constituent structure (e.g., Kayne 1994). In contrast, recent years have seen

an emergence of a view that linear order is to a considerable extent independent from

constituency. Such an idea is most clearly manifested in a version of Head-Driven Phrase

Structure Grammar (HPSG; Pollard and Sag 1987, 1994), so-called ‘linearisation HPSG’. In

this framework, a linear sequence is analysed in terms of a level of ‘order domains’, which is

an ordered list of elements that often come from several local trees (e.g., Kathol 1995, 2000;

Pollard et al. 1994; Reape 1994, 1996). The objective of this thesis is to contribute to the

understanding of a number of issues on linear order by providing an analysis of the English

left periphery within linearisation HPSG.

More specifically, we provide a linearisation-based analysis of the following left

peripheral elements: wh-phrases, preposed negative phrases, preposed adverbials and

topic/focalised phrases. We explore their distributional possibilities, following Kathol’s

(1995, 2000, 2001) adaptation of ‘topological fields’, which has been traditionally employed

in analyses of German syntax (e.g., Drach 1937; Höhle 1986). English clause structure is

divided into six regions, and each element of a clause is marked for a particular region that it

is assigned to.

This thesis shows that linear precedence is more significant than constituent structure as the

determining factor in formulating linguistic generalisations. We regard it as a preferable

conclusion from the perspective that linear order is epistemologically prior to constituent

structure (Culicover and Jackendoff 2005, Culicover 2003, Kathol 2000, etc.).

Acknowledgements

First and foremost, I thank my supervisor Robert D Borsley. Bob has always influenced me

significantly throughout the entire course of my research at University of Essex. I have greatly

benefited from his valuable feedback on this thesis and everything else I have written.

I am grateful to my thesis examiners, Louisa Sadler and Steve Harlow. Their

comments and feedback have significantly improved this thesis.

I thank my supervisory board members, Martin Atkinson, Doug Arnold, Andrew Spencer,

for their comments and advice.

My thanks also go to the old and new members of the Constraint-based Linguistics

Discussion Group for interesting discussions: Doug Arnold, Bob Borsley, Louisa Sadler,

Andrew Spencer, Kakia Chatsiou, Maria Flouraki, Yasuhiro Kawata, Despina Kazana, Evita

Linardaki, Ryo Otoguro, Antonis Polentas, Gergana Popova, Mehran A Taghvaipour, Ian

Underwood, and Miriam Urgelles.

My special thanks are due to Shiro Wada for getting me interested in English linguistics in

the first place. I would like to thank Seisaku Kawakami and Yukio Oba for their

encouragement and stimulation. I am grateful to Kensei Sugayama in many ways, especially

for giving me an opportunity to write a chapter for his edited volume on Word Grammar.

Portions of the research culminating in this thesis have been presented at various

conferences and workshops. I am very grateful to the audiences who have provided

comments and discussions, most notably Anne Abeillé, Doug Arnold, Olivier Bonami, Rui

Pedro Chaves, Harald Clahsen, Claudia Felser, Danièle Godard, François Mouret, Andrew

Radford, Manfred Sailer, Yo Sato and Shûichi Yatabe. I am also grateful to the anonymous

reviewers for the conferences and the proceedings for their comments and feedback.

I gratefully acknowledge the generous financial assistance from the Department of

Language and Linguistics, University of Essex and The Alliance 05 Project, which allowed me

to travel to the conferences.

I wish to thank my friends for very enjoyable moments we have shared at Essex: Ibrahim

Acknowledgements iv

Almahboob, Gabriela Casillas, Takako Kondo, Abolaji S Mustapha, Fumiyo Nakatsuhara, Aya

Okamoto, Ryo Otoguro, Mohammad Rezai, Neal Snape, Kana Suzuki, Urara Myojin, Kalinka

Velasco Zarate, Saad Al-Amri, and the members of the University of Essex Fencing Club.

Some of them are my office mates, and I thank them for providing me with a very pleasant

atmosphere. I also thank Satoshi Hoshii, Alberto Moraglio, Fumiyo Nakatsuhara, and Kalinka

Velasco Zarate for welcoming me to their house and letting me stay there during my final stay in

Colchester for my viva.

Very special thanks go to my wife, Tomoko Maekawa. Without her support, patience

and love, this thesis would not have been written at all.

Lastly, but most importantly, I want to express my deepest gratitude to my parents,

Keiko and Kiyohiro Maekawa, for their warm and continuous support through all these years.

Contents

Abstract ii

Acknowledgements iii

Chapter 1: Introduction 1

Chapter 2: Head-Driven Phrase Structure Grammar – Theoretical Assumptions 4

2.1 Signs and types 5

2.2 Lexical entries 7

2.3 Immediate dominance schemata 8

2.4 Unbounded dependencies 11

2.5 Information structure 13

2.6 Linearisation HPSG 16

2.6.1 Background 16

2.6.1.1 ID/LP rules 17

2.6.1.2 Reape’s word order domain 18

2.6.2 (Partial) Compaction 22

2.6.3 Domain formation 26

2.6.4 Topological Fields 26

2.6.5 Positional effects in German 29

2.6.6 Order Domain-based SLASH Passing 34

2.7 Conclusion 36

Contents vi

Chapter 3: Linearisation-based approach to English clauses 37

3.1 Positional effects in English 37

3.2 Topological fields for English 39

3.3 Constraints on linear structure of finite clauses 42

3.3.1 Lexical constraints 42

3.3.2 Phrasal constraints 43

3.3.3 Constraints on internal syntax of clauses 46

3.3.4 Clausality 49

3.4 Clausal constructions 50

3.4.1 Matrix declarative clauses 52

3.4.2 Yes-no questions 53

3.4.3 Matrix non-subject wh-questions 53

3.4.4 Matrix subject wh-questions 54

3.4.5 Declarative verb-second clauses 55

3.4.6 Subordinate declarative clauses 56

3.4.7 Subordinate wh-questions 57

3.4.8 Relative clauses 58

3.4.9 Whether- and if-clauses 59

3.5 Summary 60

Chapter 4: Wh-questions and relative clauses 62 58

4.1 Introduction 63

4.2 Previous analyses 64

4.2.1 Principles-and-parameters theory 64

4.2.2 Earlier HPSG 65

4.2.3 Recent HPSG 67

4.2.4 Summary 68

4.3 Evidence for subject extraction 68

4.4 The who/whom distinction 70

4.5 Informal and prescriptive types of register 74

4.6 The non-prescriptive type of formal register 77

4.6.1 Linear structure 77

Contents vii

4.6.1.1 Extracted subjects in main clauses 77

4.6.1.2 Extracted subjects in subordinate clauses 81

4.6.1.3 Extracted non-subjects in main clauses 82

4.6.1.4 Extracted non-subjects in subordinate clauses 85

4.6.1.5 Long-extracted subjects 87

4.6.1.6 Differences from German 90

4.6.2 An account of the facts 91

4.6.3 Summary 94

4.7 An alternative approach 94

4.8 Supporting evidence 97

4.9 Conclusion 100

Chapter 5: Adverbial Preposing 101

5.1 Introduction 102

5.2 Configurational approaches 103

5.2.1 Minimalist/P&P analyses 103

5.2.1.1 Adjunction to IP and CP 104

5.2.1.2 Adjunction to TopP 105

5.2.1.3 Scene-setting projections 107

5.2.1.4 Lower projection in the CP domain 109

5.2.2 HPSG analyses 112

5.2.2.1 Pollard and Sag (1994) 112

5.2.2.2 Bouma et al. (2001) 115

5.2.2.3 Levine (2003) / Levine and Hukari (2006) 124

5.2.3 Conclusion 124

5.3 Differences between long- and short-fronted adverbials 125

5.3.1 Information structure 125

5.3.2 Blocking of wh-extraction 128

5.3.3 Restriction to root/root-like clauses 130

5.3.4 Summary and conclusion 131

5.4 Extracted phrases 133

5.5 Incidental adverbials 137

5.5.1 Incidentality 137

Contents viii

5.5.2 Incidental adverbials 138

5.5.3 Flexible positioning 138

5.5.4 Positioning constraints 141

5.6 An account of the facts 143

5.6.1 Information structure 143

5.6.1.1 Information structure of incidental adjuncts 143

5.6.1.2 Information structure of extracted adjuncts 143

5.6.1.3 Summary 145

5.6.2 Ordering 145

5.6.2.1 Incidental adjuncts 145

5.6.2.2 Ordering of extracted adjuncts 150

5.6.2.3 Summary 152

5.7 Conclusion 153

Chapter 6: Preposed negative expressions 154


6.2 Configurational approaches 156

6.2.1 Minimalist/Principles-and-parameters approach 156

6.2.1.1 Contrasting behaviour of wh- and negative expressions 158

6.2.1.2 Information structure in NI 160

6.2.1.3 Preposing of preverbal adverbials 162

6.2.2 Possible standard HPSG approach 162

6.2.2.1 Sentential negation 163

6.3 Constructional Constraints 164

6.4 Embedded Verb Second 175

6.4.1 Embedded V2 in Germanic languages 175

6.4.2 Linearisation-based analysis of embedded V2 177

6.5 An account of the facts 181

6.5.1 Information structure of NI 182

6.5.2 V2 with fronted wh-phrases 183

6.5.3 Preposing of preverbal adverbials 186

6.5.4 Clause-boundness 188

6.5.5 Sentential negation 190

Contents ix

6.6 Other facts 192

6.6.1 NI with the conjunction nor 192

6.6.2 Individual variation: unbounded extraction of negative phrases 194

6.7 Summary and concluding remarks 196

Chapter 7: Comparison with other approaches 198


7.2 Word Grammar approach 199

7.2.1 Outline of Word Grammar 200

7.2.2 WG approach to the phenomena 204

7.2.3 Problems 205

7.2.3.1 Weakness of the No-Tangling Principle 206

7.2.3.2 Headness of a wh-pronoun 208

7.2.3.3 An extractee preceding a complementiser 209

7.3 Chung and Kim’s (2003) linearisation-based analysis 210

7.3.1 Some predictions 212

7.3.2 Problems 214

7.4 Concluding remarks 216

Chapter 8: Concluding remarks 218

Appendix A: Phrase types 221

A.1 Type hierarchy 221

A.2 Constraints on phrase types 222

Appendix B: Clause types 225

B.1 Type hierarchy 225

B.2 Constraints on clause types 226

Contents x

Appendix C: Constraints on order domains 228

Appendix D: More on who and whom (Chapter 4) 229

D.1 Wilcock’s (1999) analysis of whom in pied-piping 229

D.2 Lasnik and Sobin’s (2000) approach to the who/whom distinction 231

D.3 Further data of who and whom 233

Bibliography 237

Chapter 1

Introduction

The objective of this thesis is to contribute to the understanding of a number of issues on

linear order in a linearisation-based approach in the framework of Head-Driven Phrase

Structure Grammar (HPSG; Pollard and Sag 1987, 1994). In many theories of syntax, linear

order is inseparable from syntactic structure. Thus, phrase structure rules specify both

dominance relations among mother and daughters and linear precedence relations among

daughters. It has been suggested in Minimalism/Principles-and-Parameters theory that the

only information contained in tree diagrams is that of constituent structure, and not of linear

order since the latter can be predicted from constituent structure (e.g., Kayne 1994). An

important consequence of this view is that all syntactic operations must be sensitive to

hierarchical structure, and should not be sensitive to word order.1 In the standard version of

HPSG, linear precedence (LP) rules determine word order and immediate dominance (ID)

rules define the combinations of elements (cf. GPSG; Gazdar et al. 1985). Linear order is a

property of phonology (represented as a value of PHONOLOGY attribute). However, the

linear order is still closely tied to constituent structure: the phonology of a set of sisters cannot

be separated by the phonology of a non-sister.

In contrast, recent years have seen an emergence of a view that linear order is to a

considerable extent independent from constituency. Such an idea is most clearly manifested

in a version of HPSG, so-called ‘linearisation HPSG’. In this framework, a linear sequence

is analysed in terms of a level of ‘order domains’, which is an ordered list of elements that

often come from several local trees (Chapter 2; see also, e.g., Kathol 1995, 2000; Kathol and

Pollard 1995; Müller 1995, 1997, 2004; Pollard et al. 1994; Reape 1994, 1996). An

important consequence of this approach is that syntactic constraints can be sensitive to linear

1 It has been proposed that some traditional syntactic operations, e.g., head movement, apply in Phonological

Form (PF).


order, not only to constituent structure; thus, it is possible to give a ‘linearisation-based’

approach to certain syntactic phenomena.2

This thesis will adopt this framework, and provide a linearisation-based analysis of the

left peripheral elements in English. More specifically, we will be concerned with the

following elements which typically occur in the left periphery: wh-phrases, preposed negative

phrases, preposed adverbials and topic/focalised phrases. Some examples of these

constructions are shown below.

(1) Wh-phrases

a. Who visited Mary?

b. Who did John visit?

c. I wonder who visited Mary.

d. I wonder who John visited.

(2) Preposed negative phrases

a. Under no circumstances would he accept their offer.

b. John swore that under no circumstances would he accept their offer.

(3) Preposed adverbials

a. Tomorrow it will rain.

b. He said that tomorrow it will rain.

(4) Topic/focalised phrases

a. Kim, Sandy loves.

b. Mary says that Kim, Sandy loves.

We will explore the distributional possibilities of these left-peripheral elements. The linear

organisation of the left periphery of clause structure has been of considerable current interest

in the broad fields of theoretical linguistics: e.g., HPSG (e.g., Chung and Kim 2003; Feldhaus

1999; Holler 2004; Kathol 1995, 2000, 2002; Richter and Sailer 1999, 2001),

2 The view that linear order is independent to a considerable extent from constituency, or that constituents may

be discontinuous, has also been proposed in other syntactic frameworks: e.g., Categorial Grammar (Bach 1983;

Dowty 1996; Hepple 1994; Morrill 1995), Dependency Grammar (Bröker 1998; Plátek et al. 2001), GPSG

(Pullum 1982; Zwicky 1986), Tree Adjoining grammar (Kroch and Joshi 1987; Rambow and Joshi 1994), and

others that posit tangled trees (McCawley 1982; Huck 1985; Ojeda 1987; Blevins 1994).


Minimalist/Principles-and-Parameters approach (e.g., Culicover 1991; Haegeman 2000a, b;

Haegeman and Guéron 1999; Rizzi 1996, 1997; and many others), Word Grammar (Hudson

2003b). This thesis will argue that many of the phenomena which earlier literature has tried

to explain in terms of configurational notions, such as hierarchical functional projections with

the mechanism of movement, should in fact be accounted for in linear terms.

The organisation of this thesis is as follows. Chapter 2 will outline the basic tenets of

the general framework of HPSG and introduce some important concepts and notations. The

same chapter will also introduce the particular version of HPSG which will be adopted

throughout this thesis: linearisation HPSG. In Chapter 3 we will first propose that the notion

of ‘topological fields’, which has been traditionally employed in analyses of German syntax

(e.g., Drach 1937; Höhle 1986), is also feasible in analysing English left periphery.

Following Kathol’s (1995, 2000, 2001) adaptation of the concept, English sentences will be

divided into regions, and each element of a clause will be marked for a particular region that it

is assigned to. Then we will give some basic constraints on linear order that will play an

important role throughout the thesis. In the same chapter we will also give an overview of

the linear organisation of various constructions in English. From Chapter 4 to 6 we will

provide a linearisation-based analysis of the constructions introduced above. In Chapter 4,

we will explore the left periphery of interrogative and relative clauses. We will argue that

initial wh-expressions in these constructions should be given a unified analysis as far as their

constituent structure is concerned, but linear structure of the sentences with wh-subjects

should be given a different treatment from that of the sentences with non-subject

wh-expressions. The main concern of Chapter 5 will be preposed adverbials, but we will also

give an analysis of topic and focalised phrases. In Chapter 6 we will look at preposed

negative phrases. It will be argued that negative preposing should be accounted for in pure

terms of linear sequence, irrespective of constituency. Chapter 7 will have a critical look at

other approaches to the left periphery; in particular, we will discuss Hudson’s (2003) Word

Grammar approach and Chung and Kim’s (2003) analysis in another version of linearisation

HPSG. Chapter 8 will summarise the whole thesis and give some concluding remarks.

Chapter 2

Head-Driven Phrase Structure Grammar: Theoretical Assumptions

Contents

2.1 Signs and types ....................................................................................................................5

2.2 Lexical entries......................................................................................................................7

2.3 Immediate dominance schemata........................................................................................8

2.4 Unbounded dependencies .................................................................................................11

2.5 Information structure .......................................................................................................13

2.6 Linearisation HPSG ..........................................................................................................16

2.6.1 Background.................................................................................................................16

2.6.1.1 ID/LP rules ............................................................................................................17

2.6.1.2 Reape’s word order domain ..................................................................................18

2.6.2 (Partial) Compaction .................................................................................................22

2.6.3 Domain formation ......................................................................................................26

2.6.4 Topological Fields.......................................................................................................26

2.6.5 Positional effects in German .....................................................................................29

2.6.6 Order Domain-based SLASH Passing .....................................................................34

2.7 Conclusion..........................................................................................................................36

Head-driven Phrase Structure Grammar (HPSG; Pollard and Sag 1987, 1994) is characterised as

a monostratal and thus nonderivational grammatical framework; it dispenses entirely with

multiple levels of syntactic representations and transformations that mediate among them. It is

termed ‘head-driven’ because heads contain information about the non-heads with which they

combine.

In this chapter we will first provide theoretical assumptions. Some important concepts

and notations of the general framework of HPSG will be introduced in 2.1 through 2.4. We will

also introduce the particular version of HPSG which will be adopted throughout this thesis: so

Chapter 2: Theoretical Assumptions

5

called ‘linearisation-based HPSG’.

2.1 Signs and types

Linguistic expressions in HPSG are modelled as feature structures of type sign, represented as a

complex of phonological, syntactic, and semantic information in terms of typed feature

structures. (1) shows an example of a feature structure of a sign.

(1)

( )

string-phonlist

context

content

speech-of-part

cat

loc

synsem

sign

PHON

CONX

CONT

COMPS

SUBJ

HEADCAT

LOCSYNSEM

The value of the feature PHONOLOGY (PHON) is of type list(phon-string), which represents

phonological information of a sign. The value of SYNTAX-SEMANTICS (SYNSEM) is of

type synsem, a feature structure containing syntactic and semantic information. The value of

LOCAL (LOC) contains the subset of syntactic and semantic information shared in long-distance

dependencies. The syntactic properties of a sign are represented under the path

SYNSEM|LOC|CAT(EGORY). The HEAD value contains information shared between a

phrase and its head, information such as parts of speech. As stated earlier, heads contain

information about the non-heads with which they combine. This information is specified in

valence features, whose value is a list of synsem objects. Valence features include the COMPS

feature, which indicates what kind of complements a head takes, and SUBJ feature, which

indicates what kind of subject a head takes. The semantic properties of a sign are represented

under SYNSEM|LOC|CONT(ENT). Contextual information is represented as a value of

CONTEXT (CONX). The value of this feature is an object of type context, which has the

feature INFORMATION-STRUCTURE (INFO-STRUC; See 2.5). Further features will be


6

introduced in the later chapters.

Sag (1997) and Ginzburg and Sag (2000) hypothesise that a rich network of constructions

with associated constraints is part of the grammars of natural languages. The type sign has the

immediate subtypes of word and phrase.

(2) Classification of sign

Lexical entries are descriptions of feature structures of the type word, while phrases of various

kinds are described by a feature structure of the type phrase. In the rest of this section we will

look at phrases, and words will be considered in the next section.

Phrases are subject to the following constraint.

(3) [ ]COMPS→phrase

This constraint states that phrases have an empty COMPS list. Phrases are classified into

subtypes (cf. Sag 1997; Ginzburg and Sag 2000; and other recent works). The partial hierarchy

of phrase types is given in (4).

(4) Partial classification of phrase

The hierarchy in (4) indicates that phrase has an immediate subtype headed-phrase (hd-ph).

This type has its own subtypes. The headed phrases are classified (at least) into

head-adjunct-phrase (hd-adj-ph), head-subject-phrase (hd-subj-ph), and

head-complement-phrase (hd-comp-ph). Other phrase types will be introduced in later

sign

word phrase

phrase

hd-ph

hd-adj-ph hd-comp-ph hd-subj-ph hd-filler-ph


7

chapters.

The type hierarchies allow properties that are shared between different phrasal types to be

spelled out just once: generalisations that hold for subtypes can be just specified for the

supertype. Pollard and Sag’s (1994) the Head Feature Principle (5a) can be understood as a

constraint on the phrasal type hd-ph, as shown in (5b).

(5) Head Feature Principle

a. The HEAD value of any headed phrase is structure-shared with the HEAD value of

the head daughter. (Pollard and Sag 1994: 34)

b.

→

]1[HEAD|CAT|LOC|SYNSEM|DTR-HD

[1]HEAD|CAT|LOC|SYNSEMph-hd

(Adapted from Müller 2002:7)

The HEAD-DAUGHTER (HD-DTR) feature refers to the head daughter. The Head Feature

Principle identifies the HEAD value of headed phrases with that of its head daughter.

Another such constraint is the Valence Principle.

(6) Valence Principle (Pollard and Sag 1994: 348)

In a headed phrase, for each valence feature F, the F value of the headed daughter is the

concatenation of the phrase’s F value with the list of SYNSEM values of the F-DTRS

value.

The Valence Principle states that in a given phrase the head daughter’s valence features specify

an element that is identified with the appropriate non-head sister of the head daughter.

2.2 Lexical entries

As noted above, lexical entries are descriptions of feature structures of the type word. This

section will give an example of a lexical entry of a verb, but only the portions which are relevant

to the discussions in this thesis will be provided. For more detailed discussion, see Chapter 1 of

Pollard and Sag (1994).


8

The lexical entry for a simple transitive verb like saw will have the following category

information.

(7)

NPCOMPS

NPSUBJ

][HEAD

CAT

finverb

(7) indicates that the finite verb saw takes a subject NP and a complement NP. As noted above,

the valence information is encoded as the values for the features SUBJ and COMPS. ‘NP’ is an

abbreviation for a feature structure description of the type synsem with the following

specification.

(8)

SPR

HEADCATLOC

noun

The feature value is sometimes displayed without the feature when there is no confusion. The

[fin] specification in (7) is an abbreviation for [VFORM fin], which indicates that the verb is

finite.

2.3 Immediate dominance schemata

In the conception of grammar adopted in this study, immediate dominance schemata are

understood as partial descriptions of feature structures of subtypes of phrase (Sag 1997;

Ginzburg and Sag 2001). In this section we will introduce three of such schemata. Further

schemata will be added later in this study.

Head-complement phrases are subject to the constraint in (9).

(9) [ ] [ ]

→[n],[1],COMPS

DTR-HD

[n]SS,,[1]SSDTRS-NH

K

K

wordph-comp-hd

The non-head daughters of a phrase are represented as the value of NONHEAD-DAUGHTERS

(NH-DTRS) feature. The head daughter is referred to by the HEAD-DAUGHTER (H-DTR)


9

feature. The integers in the square brackets, or ‘tags’, indicates that the same objects appears in

more than one position in the representation. In the above constraint for head-complement

phrases, the tags indicate that the synsem objects in the COMPS list of the head daughter

corresponds to the SYNSEM values of the non-head daughters. (9) states that a

head-complement phrase has a head daughter which is a word and non-head daughters whose

SYNSEM values are identical to the synsem objects that form the COMPS list of the head.

Head-subject phrases are subject to the constraint in (10).

(10)

[ ]

→

[1]SSDTR-NH

]1[SUBJDTR-HD

SUBJ

phraseph-subj-hd

This states that a head-subject phrase has an empty list as the value of the SUBJ feature and has a

head daughter which is a phrase, and the SYNSEM value of the non-head daughter is the synsem

object in the SUBJ list of the head.

Head-adjunct phrases are subject to the constraint in (11).

(11) [ ][ ]

→

[1]SSDTR-HD

[1]MODDTRS-NHph-adj-hd

(11) says that a head-modifier phrase has a head-daughter and a non-head daughter, and the

MOD value of the latter is identical to the SYNSEM value of the former.

Now let us examine how these schemata work. Let us consider the simple sentence in

(12).

(12) Mary plays the violin well

The phrase types and constraints that we have just outlined will give something like the

following structure to (12).


10

(13) Structure for (12)

Here and throughout, we follow abbreviatory conventions for displaying constituent structures

(Pollard and Sag 1994: 33): a phrasal sign is represented by a node labelled with (an abbreviation

of) the SYNSEM value of the sign; the daughters are represented at the end of labelled arcs

originating from that node; PHON values for signs appear below the corresponding nodes.

Other conventions of abbreviation that will be used in this study are the following.

(14) Abbreviation Abbreviated feature structure descriptions

VP

synsem

verb

SUBJ

HEADCATLOC

S

SUBJ

HEADCATLOC

verb

Let us examine (13) in detail. The lexical head plays requires an NP subject and an NP

complement. The lower VP plays the violin has two daughters: the lexical head plays as a head

Mary

|

[2]NP

COMPS

[2]SUBJ

[1]HEADVP

COMPS

[2]SUBJ

[1]HEAD[4]VP

[ ]

well

|

[4]MOD|HEAD

AdvP

plays

|

[3]COMPS

[2]SUBJ

[1]HEAD

verb

violinthe

[3]NP

COMPS

SUBJ

[1]HEADS


11

daughter and the complement NP the violin as a non-head daughter. The non-head daughter has

its SYNSEM value (indicated by the tag ‘[3]’) identical to the element on the COMPS list of the

head daughter, as required by constraint (9) on head-complement phrases. Due to the Valence

Principle in (6), the COMPS list of the mother VP is empty, and this is in accordance with

constraint (3) on phrases. The HEAD value (indicated by [1]) of the VP is identical to that of

the head daughter plays. This is due to the Head Feature Principle (5).

The higher VP plays the violin well has two daughters: the VP just considered is a head

daughter, and the adverbial phrase well is a non-head daughter. Constraint (11) on head-adjunct

phrases requires that an adjunct select the head it modifies: the MOD value of the latter is

identical to the SYNSEM value of the former. Note that the single element on the SUBJ list of

the mother VP (indicated by [2]) is identical to that of the head daughter VP; this is ensured by

the Valence Principle in (6). The Head Feature Principle (5) again necessitates the

structure-sharing of the HEAD value of the mother VP with that of the head daughter.

The top node has two daughters: the head daughter is the VP just considered, and the

non-head daughter John, which is a subject of the sentence. Conforming to the constraint on

head-subject phrases in (10) the SYNSEM value (indicated by [2]) of the subject NP is identical

to the single element on the SUBJ list of the head daughter. The SUBJ list of the mother node

is now empty, in accordance with the Valence Principle in (6). The identity of the HEAD value

of the mother S with that of the head daughter is again due to the Head Feature Principle (5).

2.4 Unbounded dependencies

In this section we will examine how HPSG analyses unbounded dependency constructions. A

typical example of such constructions is the bracketed string in (15).

(15) I wonder [who you saw].

This is a case of an embedded wh-question. Its tree diagram is sketched in (16).


12

(16) Structure for the bracketed string in (15)

The HPSG analysis of an unbounded dependency can be divided into three distinct parts: a

bottom, a middle, and a top, as labelled in (16). For the bottom of the unbounded dependency

constructions, we assume an empty category, which corresponds to what has been called ‘trace’

in Minimalism/Principles-and-Parameters theory, following the version of HPSG presented in

Chapters 1–8 of Pollard and Sag (1994). The lexical entry for a trace is as follows.

(17)

{}SLASH

{}REL

{}QUE

BIND-TO

{[1]}SLASH

{}REL

{}QUE

INHERITED

NONLOC

[1]LOC

SYNSEM

PHON

(Pollard and Sag 1994: 164)

A trace can occur as a complement of some head, as in (16).1 Then the LOC value will be

identical to whatever the selecting head specifies for that complement. The LOC value is 1 Levine (2003) and Levine and Hukari (2006) argue that a trace can occur as an adjunct and a subject of a head, and we will adopt this analysis in subsequent chapters.

[ ]

who

|

[1]LOC

NP

{[1]}SLASH|BIND-TO

{[1]}SLASH|INHERVP

[ ]{[1]}SLASH|INHER

[4]VP

saw

|

V

−

|{[1]}SLASH|INHER

[1]LOCNP

[ ]{}SLASH|INHER

S

you

|

NP

top

middle

bottom


13

identical to the single value of the SLASH set of the trace, as indicated by [1].

In the middle part of an unbounded dependency, the information contained in the trace is

propagated up to the mother node of phrasal signs. This is ensured by the Nonlocal Feature

Principle.

(18) Nonlocal Feature Principle (Pollard and Sag 1994: 164)

For each nonlocal feature, the INHERITED value on the mother is the union of the

INHERITED values on the daughters minus the TO-BIND value on the head

daughter.

In the case of (16), every node in the tree that dominates the trace inherits the single member of

the SLASH value set on the trace.

Finally, in the top part of an unbounded dependency, the SLASH value thus passed up the

tree is bound off or discharged by identification with the LOC value of the filler. This can be

formalised as a partial description of a feature structure of head-filler phrases, as in (19).

(19)

[ ][ ]

→

{[1]}SLASH|BIND-TO

}{[1],SLASH|INHERNONLOC

SUBJ

VFORMHEADCAT|LOC

SSDTR-HD

[1]LOC|SSDTRS-NH

K

fin

ph-filler-hd

(Adapted from Pollard and Sag 1994: 164)

(19) states that in head-filler phrases a finite sentence containing a trace is combined with a

phrase (i.e., a filler) whose local features are identical to those of the trace. The

TO-BIND|SLASH value on the head ensures that the value of the INHERITED|SLASH feature

is not shared by the mother, as specified in Nonlocal Feature Principle in (18).

2.5 Information structure

As stated in 2.1, a sign has information structure, which is represented as a value of its

INFO(RMATION)-STR(UCTURE), which is appropriate for the CONTEXT feature (Engdahl


14

and Vallduví 1996; Alexopoulou and Kolliakou 2002; De Kuthy 2002; and De Kuthy and

Meurers 2003. Its feature geometry reflects a focus-background structure of a sign. The

LINK and FOCUS (FOC) features are among those appropriate for INFO-STRUC. The LINK

feature is used for representing which part of a sign stands in some salient relationship to

elements that have been evoked in the prior discourse, and the FOC feature is used for

representing the focused part of the sign (See the references cited above; see also Birner and

Ward (1994) and Ward et al. (2002)). We assume, following Engdahl (1999: 186–187), that

each of INFO-STRUC features takes content objects (i.e., values of the CONT feature) as its

value.2 Furthermore, English uses intonational strategies to realise information structure

(Engdahl and Vallduví 1996: 7). Adopting Jackendoff’s (1972) and Engdahl and Vallduví’s

(1996) terminology, we call the intonation which characterises the focus the A-accent, and that

which characterises the link the B-accent. The A-accent corresponds to a simplex high pitch

accent (H*), and B-accent corresponds to a complex fall-rise pitch accent (L+H*). We assume

that English words are constrained by the following constraints.

(20) a.

−→

[1]FOC|STRUCINFO

[1]CONT

ACCENT|PHON a

word

b.

−

→

[1]LINK|STRUCINFO

[1]CONT

ACCENT|PHON b

word

(cf. Engdahl and Vallduví 1996: 11; Alexopoulou and Kolliakou 2002: 211)

(20a) states that if a word’s PHON|ACCENT value is a, the INFO-STRUC|FOC value is

structure-shared with the CONT value; this means that if a word has the A accent, it has the focus

interpretation. (20b) states that if a word’s PHON|ACCENT value is b, the

INFO-STRUC|LINK value is structure-shared with the CONT value; this means that if a word

has the B accent, it has the link interpretation. Finally, we adopt the following constraints to

capture the relation between INFO-STRUCT values of a phrasal sign and those of its daughters

or domain elements.3

2 See also De Kuthy (2002: 162) and De Kuthy and Meurers (2003: 103). 3 The constraint in (21) is very similar to Information Structure Instantiation Principle proposed by Engdahl and Valluduví (1996: 12) and Alexopoulou and Kolliakou (2002: 211).


15

(21) Either (a) if a daughter’s INFO-STRUC value is instantiated, then the mother inherits

this instantiation

or (b) if the FOCUS value of the rightmost domain object is instantiated, then the

FOCUS value of the mother is identified with its CONTENT value.

(21a) ensure inheriting of the INFO-STRUC values to the higher node in the cases of the link

and the narrow focus, and (21b) deals with the cases where the A accent can serve to mark focus

of more than just the accented word (See Jackendoff 1972; Reinhart 1995; Rochemont 1986;

Selkirk 1984, 1995, and many others). The former will be particularly important in the later

chapters.

To see how the above constraints work, let us consider the following sentence.

(22) John hates chocolate.


The object noun chocolate set has the A accent and therefore its CONT value is token-identical

with the FOC value, according to the constraint in (20a). The domain object for the focused

noun chocolate is rightmost in the order domain of the VP, and therefore the constraint (21b)

][3'],[2'],[1'DOM

[4]LINK

]6[FOCSTRUC-INFO

S

John

[4]LINK|STRUC-INFO

[4]CONT

ACCENT|PHON

]1[

NP

b

]'3[],'2[DOM

]6[FOC|STRUC-INFO

]6[CONTVP

chocolate

[5]FOC|STRUC-INFO

[5]CONT

ACCENT|PHON

]3[

NP

a

[ ]

hates

|

][2'DOM]2[

NP


16

allows the VP itself to be focused: the CONT value of the VP is token-identical with its FOC

value. The top S node inherits the FOC value of the VP, due to the constraint in (21a). This

indicates that the VP hates chocolate functions as the focus in the sentence. The subject noun

John has the B accent and the constraint in (20b) causes the structure-sharing of the CONT value

with the LINK value. Finally, the top S node inherits the LINK value of John, due to the

constraint in (21a). This indicates that the subject noun John functions as the link (i.e., topic) in

the sentence.

2.6 Linearisation HPSG

This chapter will introduce the version of ‘linearisation HPSG’ which we will be assuming in

this thesis. Although it is cast within the general framework of HPSG, the central tenet of

linearisation HPSG is that the domain within which constituent order is determined may be

larger than the local tree: linear order is independent from constituency (See, e.g., Kathol 1995,

2000; Kathol and Pollard 1995; Müller 1995, 1997, 2004; Pollard et al. 1994; Reape 1994, 1996).

The version which will be adopted in this thesis is largely based on the one presented by a series

of work by Kathol (1995, 2000, 2002; see also Borsley and Kathol 2001 and Kathol and Pollard

1995). Before going into the specifics, however, we will consider the background against

which linearisation HPSG has emerged. In 2.6.2 through 2.6.6, we will introduce some

important concepts and notations which will play an important role in the subsequent chapters.

2.6.1 Background

In this section we will review how linear order and its relationship with constituent structure

have been analysed in HPSG. We will start with the ID/LP rules which were proposed within

Generalised Phrase Structure Grammar (GPSG; Gazdar and Pullum 1981; Gazdar et al. 1985)

and have been adopted by standard HPSG (Pollard and Sag 1987) with certain modifications.

We will then move on to Reape’s (1993, 1994, 1996) approach to linearisation in terms of ‘word


17

order domains’, which have had a major influence on the subsequent work in linearisation

HPSG.

2.6.1.1 ID/LP rules

In many theories of syntax, linear order is inseparable from syntactic structure. Thus, phrase

structure rules specify both dominance relations among mother and daughters and linear

precedence relations among daughters. GPSG (Gazdar and Pullum 1981; Gazdar et al. 1985)

has separate rules for dominance relations and precedence relations. Immediate dominance (ID)

rules only capture immediate dominance relation, and linear order is captured by linear

precedence (LP) rules. The LP rules describe the permissible order of constituents in a local

tree, and a set of sisters cannot be separated by a non-sister. Linear order is still inseparable

from constituency in this respect.

The standard version of HPSG follows GPSG and adopts the separation of immediate

dominance and linear precedence (Pollard and Sag 1987). For HPSG, however, linear order is

a property of phonology, represented as a value of PHONOLOGY attribute. Pollard and Sag

(1987) propose the following constraint which states that the phonology of a phrasal sign is a

function of the sign’s daughters.

(24) Constituent Order Principle

( )

→

[1]DTRS

[1]tsconstituen-orderPHONphrase (Pollard and Sag 1987: 169)

The function order-constituents gives as its value a permutation (or disjunction of

permutations) of PHONOLOGY values of the sign’s daughter. The permissible permutations

are constrained by LP constraints. For example, Pollard and Sag (9897) propose the following

constraint for head-initial languages such as English.

(25) [ ] [ ]p+LEXHEAD (Pollard and Sag 1987: 172)

(24) states that a lexical head daughter is constrained to temporally precede the phonological

realisation of any of its sisters.


18

As in phrase structure rules and ID/LP rules in GPSG outlined above, (the phonology of) a

set of sisters cannot be separated by (the phonology of) a non-sister. This means that linear

order is inseparable from constituent structure in standard HPSG as well.4

2.6.1.2 Reape’s word order domain

Reape (1993, 1994, 1996) challenges configurational accounts of constituent order in which LP

constraints operate on local trees, and provides an alternative account in terms of what he terms

‘word order domains’. In Reape’s original proposal, order domains are represented by a feature

DOM on a phrase, whose value is a list of signs. In each local tree the order domain of the

mother is constructed from those of its daughter signs. The daughter signs are taken to the

DOM list of the mother, either as a list of signs (domain union), or as a singleton list containing

the list of signs (domain insertion), and they are then ‘sequence-unioned’ to form the mother’s

order domain. Informally, the sequence union (or, shuffle) relation permits interleaving of

elements of two lists in such a way as to preserve the order of each list. Building of an order

domain is formalised in the Domain Principle, given in (26).5

(26)

[ ]

[ ] [ ]

+

+

+

+

−−

⇒

]n[]1i[]i[]2[]1[DOM

[n]DOM

UNIONED,,

1][iDOM

UNIONED

UNIONED]i[,,UNIONED[2]DTRS-ARG

[1]DTR-FUN

DTRS

DTRS

oooooo

o

KK

K

K

structure-argument-functor

(Reape 1996: 225)

A sign with a negative value for the feature UNIONED is domain-inserted, and that with a

4 In their brief discussions on free word-order languages, Pollard and Sag (1987: 189) tentatively propose the Scrambling Principle, given below:

(i) ( )

[1]DTRS

[1]tsconstituen-interleavePHON

This principle allows constituents to be interleaved. They remain sceptical, however, whether any human languages are subject to this principle.

5 ‘o ’ is a notation for sequence union (shuffle).


19

positive value is domain-unioned into the order domain of the mother, and it is shuffled with

other domain elements contributed by its sisters.6 Domain insertion prevents intervening

materials coming from other constituents, whereas domain union permits this. An element

domain-unioned into the mother’s order domain corresponds to what is often referred to as

‘liberated’ (Zwicky 1986). In this approach, LP constraints are defined as well-formedness

conditions on word order domains. The phonological representation of a phrasal sign is

computed from the phonological representation of its domain elements. This is defined in the

following constraint.

(27) [ ][ ] [ ][ ]

⊕⊕→

n PHON,,1 PHON DOM

[n][1] PHON

K

Ksign-phrasal (Reape 1996: 225)

The PHON value of a phrasal sign is the concatenation of the PHON values of its domain

elements without changing the order.7

For illustration of the approach outlined above, let us consider a German cross-serial

subordinate clause in (28).

(28) daß es ihm jemand zu lesen versprochen hat

that it him someone to read promised has

‘that someone promised him to read it’ (Reape 1994: 157)

The NP es ‘it’ is the direct object of the verb zu lesen ‘to read’, ihm ‘him’ is the indirect object of

the past participle versprochen ‘promised’, and jemand ‘someone’ is the subject of the finite

auziliary hat. A phrase structure of this sentence is given in the tree in (29).

6 The value of the UNIONED feature is given either by a lexical specification or by language-specific principles. 7 ‘⊕ ’ is a notation for concatenation.


20

(29) Syntactic tree of (28)

S

[NP1 jemand] VP1 [V1 hat]

[NP2 ihm] VP2 [V2 versprochen]

[NP3 es] [V3 zu lesen]

In Reape’s term, (29) is a ‘syntactic tree’, which is an unordered tree that only displays

dominance relation.

The linearisation of (28) is given in the tree in (30).

(30) Domain tree of (28)

[S [NP3 es] [NP2 ihm] [NP1 jemand] [V3 zu lesen] [V2 versprochen] [V1 hat]]

[NP1 jemand] [V1 hat] [VP1 [NP3 es] [NP2 ihm] [V3 zu lesen] [V2 versprochen]

[NP2 ihm] [V2 versprochen] [VP2 [NP3 es] [V3 zu lesen]

[NP3 es] [V3 zu lesen]

In Reape’s term, this is a ‘domain tree’, which is an unordered tree with order domains displayed

as the nodes of trees. The dominance relation is the same as the syntactic tree in (29).8 In

each node the outer bracket represents the sign, and its contents are the sign’s order domain.

8 A close correspondence can be seen between the distinction between a syntactic tree and a domain tree, on the one hand, and the distinction between tectogrammatical structure and phenogrammatical structure (Dowty 1996; see also Curry 1961), on the other.


21

The infinitive zu lesen has already been inserted into its mother’s (V3) order domain. The VP2

es zu lesen is domain-unioned into its mother’s (VP1) order domain so that es and zu lesen can be

discontinuous. Likewise, VP1 es ihm zu lessen versprochen is domain-unioned into their

mothers’ (S) order domain so that jemand can intervene between ihm and zu lesen in the order

domain of S.

The following LP constraints are applied to order domains:

(31) a. NP precedes V.

b. A verb follows any verb that it governs.

These LP constraints licenses the well-formedness of the string found in (28), where all the

nouns appear before all the verbs, and each verb follows the verb it governs. Note that the

order of NPs is unconstrained. This allows for six possible order domains that correspond to

the following sentences to be licensed.

(32) a. es ihm jemand zu lesen versprochen hat

b. es jemand ihm zu lesen versprochen hat

c. ihm es jemand zu lesen versprochen hat

d. ihm jemand es zu lesen versprochen hat

e. jemand es ihm zu lesen versprochen hat

f. jemand ihm es zu lesen versprochen hat

There are some differences between the order domain-based approach to linearisation

outlined above and standard HPSG. As discussed in 2.6.1.1, linearisation in standard HPSG

reflects constituent structure in that LP rules apply to local trees and sisters cannot be separated

by a non-sister. In the linearisation-based approach, however, LP rules apply to order domains,

and sisters can be separated by a non-sister. This means that a constituent may have more

domain elements than daughters. The domain elements corresponding to a phrase can be

linearised discontinuously.

Another difference is that LP rules in standard HPSG determine the ordering of just the

phonological values of signs, whereas those in the order domain-based approach impose an

ordering on elements in an order domain.


22

An order domain-based approach to linearisation has been applied for various languages:

German (Kathol 1995, 2000, 2001; Kathol and Pollard 1995a,b; Müller 1995, 1997, 2002, 2004),

English (Kathol and Levine 1992, Chung and Kim 2003), Japanese (Calcagno 1993; Maekawa

2004; Yatabe 1996, 2001, 2003), Korean (Yoo 1993; Lee 1999), the Scandinavian languages,

Yiddish (Kathol 1995, 2000), Dutch (Kathol 1995, 2000; Campbell-Kibler 2002), Fox

(Crysmann 1999), French (Bonami et al. 1999), Polish (Borsley 1999; Kupść 1999),

Serbo-Croatian (Penn 1999), Warlpiri (Donohue and Sag 1999), Breton (Borsley and Kathol

2000), Portuguese (Crysmann 2000a, Crysmann 2002), Udi (Crysmann 2000b), Danish (Jensen

and Skadhauge 2001), Welsh (Borsley 1999, 2003, 2005a, 2005b), and Bulgarian (Jaeger 2003).

In the course of maturation, linearisation HPSG has been modified in various respects. In

the rest of this section, we will outline the modifications that will be adopted in the later

chapters.

2.6.2 (Partial) Compaction

As we discussed above, Reape assumes that order domains are a list of signs.9 It has been

pointed out, however, that signs are not desirable for elements of an order domain. First, a sign

includes information about constituency via its DAUGHTERS feature and information about

linear order via its DOM feature. There are no known phenomena, however, where

linearisation of linguistic objects depends on its phrase-structural or linear organisation (Pollard,

Kasper and Levine 1994; Kathol and Pollard 1995; Kathol 2000). Second, it is impossible for

lexical signs to introduce domain elements without giving rise to a cyclic feature structure as

follows.

(33)

[1]DOM

PHON

SYNSEM]1[

K

K

word

The cyclic structure of this sort incurs computational complexity, and it is preferable to avoid it

9 Donohue and Sag (1999), Kathol (2000: 97, 101), and Jaeger (2003) also assume that an order domain consists of an ordered list of signs.


23

(Kathol 2000: 98).

In order to eliminate redundant information from the domain elements, Kathol and Pollard

(1995) introduce a constraint which constructs domain elements from phonological and

syntactic/semantic information of a daughter’s sign.10 This is called ‘compaction’, which is

given the following informal description of compaction.11,12

(34)

( )

[ ] [ ]

⊕⊕

∧

≡

[n][4]PHON

[3]SYNSEM:]2[

[n]PHON,,[4]PHONDOM

]3[SYNSEM:]1[

]2[],1[compaction

K

K

obj-dom

sign

(Kathol and Pollard 1995)

Thus, compaction is defined as a mapping relation from a sign to a single domain object. A

domain object is constructed with only the information relevant for linearisation (phonological

and syntactic/semantic information) of the daughters’ sign: the SYNSEM value of the domain

object is inherited from the corresponding sign, and the phonological information is the

concatenation of the PHON values of the elements in the sign’s order domain. This revised

data structure for order domains involves neither of the problems mentioned above: it does not

involve information about phrase-structure or linear order, and it does not contain a cyclic

structure. The domain object thus constructed is integrated into the order domain of the mother

as a single element. Compaction thus prevents an intervening material from other constituents.

Kathol and Pollard (1995a) also introduces a related notion of ‘partial compaction’, which

subsumes the notion of compaction. Partial compaction prevents specifically marked domain

elements from being compacted with the remainder of the phrase. They illustrate this with the

phenomenon of extraposition in German such as (35).

10 Pollard, Kasper and Levine (1994) assemble phonological and syntactic/semantic information, as the values of the PHON and SYNSEM features, respectively, into a separate feature structure, of type node, which in turn is a value of an attribute NODE. The elements in order domains are of this type. The feature NODE and the type node are renamed as DOMAIN-OBJECT (DOM-OBJ) and domain-object (dom-obj), respectively, by Kathol (1993). See Kathol (1995: 104–106, 145, n39) for problems involved in the adoption of the DOM-OBJ feature. 11 Pollard and Kathol (1995) and Kathol (1995: 146) also provide a formal definition. 12 ‘⊕ ’ is a notation for concatenation.


24

(35) einen Hund füttern [der Hunger hat]

a dog feed that hunger has

‘feed a dog that is hungry’

The relative clause der Hunger hat has been extraposed out of the NP einen Hund der Hunger

hat. The argument NP gives rise to two domain elements, one for the relative clause der

Hunger hat, and one for the remainder of the NP einen Hund. Partial compaction is given the

following formal definition.

(36)

( )

( )( )]7[],5[join

[6][3],[5],shuffle[7]PHON

[4]SYNSEM[2]

[6]DOM

[4]SYNSEM]1[

]3[],2[],1[compaction-p

PHON

obj-dom

sign

∧

∧

∧

≡

The p-compaction relation holds of a sign [1], domain object [2] and a list of domain object [3]

only if [2] is the compaction of [1] with [3] liberated from the order domain of [1]. The

joinPHON operation takes a list of domain object and concatenates each PHON value. Kathol

and Pollard (1995a) defines ‘shuffle’ as follows: ‘The shuffle relation holds of n lists L1, …, Ln-1,

Ln, iff Ln consists of the elements of the first n-1 lists interleaved in such a way that the relative

order among the original members of L1 through Ln-1, respectively, is preserved in Ln.’.

Liberation of the relative clause as in (35) can be obtained by, for example, invoking this relation

when the NP combines with the verb. This can be expressed in the Head-Complement Schema

as in (37).


25

(37) [ ][5]DOM :][M

[ ][4]DOM :][H [1] :][C

( )( )

[ ][ ]( )

¬∨∧

+∧

∧

∧

SUBCAT

HEADSYNSEM]2[:[3]

EXTRASYNSEM:]3[

]5[],4[],3[,[2]shuffle

[3][2],[1],compaction-p

verb

list

The effect of the third clause attached is to ensure those elements marked as [EXTRA +] is

liberated from the rest of the NP. The fourth clause prevents extraposition out of clauses.13

The structure for (35) is shown in (38). The schema in (37) applies at the level where the NP

combines with the verb: the domain objects for einen and Hund are compacted, but that for the

relative clause is liberated and inserted into the domain of the VP.

(38) Extraposition via partial compaction (Kathol and Pollard 1995a)

( )( )]5[],4[,]3[,[2]shuffle

[3][2],[1],compaction-p

∧

∧

In the case where the list of liberated domain object is empty, the definition in (36) gives 13 Cf. Ross’ (1967) Right Roof Constraint.

+

EXTRA

S-REL]3[,V

,

NP[2][5]DOM

VP

hat Hunger derHund einen fütter

+

EXTRA

S-REL]3[,N

,DET

DOM

NP

]1[hat Hunger der

Hundeinen füttern

V]4[DOM

V

+

EXTRA

S-REL]3[,N

DOM

NP

hat Hunger derHund

[ ]

einenDOM

DET

+

V,

NP,

RELDOM

EXTRA

S-REL

hatHungerder

[ ]

HundDOM

N


26

the effect of total compaction (and domain insertion). Thus, total compaction can be regarded

as a subcase of partial compaction.

The notion of (partial) compaction has been adopted by much subsequent work (Kathol

1995: 306ff, 2000: 100; Yatabe 1996; Lee 1999; Penn 1999; Campbell-Kibler 2002).14

2.6.3 Domain formation

In Reape’s original proposal, the value of the feature DOM is a list of signs: as formalised in the

Domain Principle in (26), the order domain of the mother is constructed from its daughters’ signs.

In Kathol’s version adopted in this study, however, the order domain elements are domain

objects, which are distinct from signs. Thus it is not possible to adopt Reape’s Domain

Principle. In the versions of Kathol and Pollard (1995) and Kathol (1995, 2000), formation of

order domain is done by some ID schemata. We have also seen an example of such a schema,

the Head-Complement Schema (Kathol and Pollard 1995), in (37). In the present study,

however, ID schemata are understood as partial descriptions of feature structures of subtypes of

phrase, as discussed in 2.3. Therefore, domain formation will also be incorporated into the

constraints on phrasal types. See Chapter 3 for details.

2.6.4 Topological Fields

Research on German syntax has traditionally approached the basic clausal organisation of the

language not only from the perspective of constituent structure but also in terms of ‘topological

fields’ (e.g., Drach 1937; Höhle 1986). The last decade has seen a renewal of interest in this

notion, especially within the framework of Head-Driven Phrase Structure Grammar. In

Kathol’s version of domain-based approach to linear order (Kathol 1995, 2000, 2001), any

German sentence is divided into regions, and each element of a clause is marked for a particular

region that it is assigned to. At least four such regions, or fields, can be assumed for German:

14 In Penn’s (1999: 170) system, a domain object corresponds only to a word, and compaction to an element larger than a word is represented by the structure-sharing of a C(OMPACTION) feature. Therefore, compaction is not a map from sign to a domain object.


27

first, second, third and fourth.15,16 The position assignment of elements in three major types of

clause in German is summarised in (39) (Kathol 2001: 50).

(39) Topological Fields for German

first second third fourth

(i) daß

that

Lisa eine Blume

Lisa a flower

gepflanzt hat

planted has

(ii) was

what

Lisa

Lisa

gepflanzt hat

planted has

(iii)

hat

has

Lisa eine Blume

Lisa a flower

gepflanzt

planted

(iv) eine Blume

a flower

hat

has

Lisa

Lisa

gepflanzt

planted

In a subordinate clause such as (i) and (ii), the finite verb and a verb that it governs occur in

fourth position, and the subject and the complement appear in third position. In (i) second

position contains the complementiser daß ‘that’, but in (ii), the wh-word (was ‘what’) occupies

the same position. These sentences illustrate the verb-final clause structure. In a root clauses

such as (iii) and (iv) second position holds the finite verb. In a verb-initial clause (iii) first

position is empty. In a verb-second clause such as (iv), first position contains one of the verb’s

arguments or modifiers. Root wh-interrogatives, therefore, include a wh-element in first

position.

The topological field which a particular domain element belongs to is represented by

means of subtyping dom-obj (cf. Kathol 2000).17 We can assume that maximal subtypes of

dom-obj are first, second, third and fourth, each of which are understood as belonging to first,

second, third and fourth position, respectively, of the order domains. 15 Kathol (2000) adopts the following terminology for these regions: vorfeld ‘initial field’ (vf), linke Satzklammer ‘left sentence bracket’ (cf), Mittelfeld ‘middle field’ (mf), and rechte Satzklammer ‘right sentence bracket’ (vc). 16 The fifth field, Nachfeld ‘post-field’, is usually assumed for extraposed elements in German, but it is ignored here. 17 In the version presented in Kathol (1995), position class is represented as a value of the feature TOPO(LOGY), which is appropriate for the type dom-obj


28

(40) The type hierarchy for domain objects

dom-obj

first second third fourth fifth

For example, the German complementiser daß ‘that’ is given the following lexical

description

(41) [ ]

[2]SYNSEM

[1]PHONDOM

HEAD]2[SYNSEM

]1[PHON

second

c

word

This description specifies that the domain object for daß is of the type second, which means that

the complementiser daß occupies second position of the order domain.18

The linear ordering among the fields is imposed by means of the following LP constraint

(Kathol 1995: 138; 2000: 79; 2001: 50).

(42) Topological Linear Precedence Constraint:

fourththirdsecondfirst ppp

Thus, assignment of constituents to a particular field in a sentence is in many cases enough to

determine the word order of the sentence. Cardinality restrictions on position classes are also

represented in terms of LP constraints as follows (Kathol 2001: 51).

(43) Topological Uniqueness Conditions:

a. firstfirst p

b． secondsecond p

These constraints entail that there may be only one element in each of the position classes first

and second.

LP rules make primary reference to syntactic categories in standard HPSG, while positional 18 The assignment of domain objects to topological fields is subject to not just lexical constraints like (41), but also constructional constraints. Such constraints will be introduced in Chapter 3 and later chapters.

daß


29

assignment is in many cases enough to determine the linear order of the clause in this topological

field-based approach.

2.6.5 Positional effects in German

One of the important consequences of assuming topological fields for German sentences is that

they provide the basis for accommodating situations where there are important similarities

between elements with a quite different constituent structure status. In German,

complementisers and finite verbs show a complementary distribution.

(44) a. Liest Hans die Zeitung?

reads Hans the newspaper

‘Does Hans read the newspaper?’

b. ob Hans die Zeitung liest

whether Hans the newspaper reads

‘whether Hans read the newspaper’

c. * ob liest Hans die Zeitung

whether reads Hans the newspaper

In environments such as yes-no questions, finite verbs can occur in the clause-initial position, as

in (44a). However, the presence of the complementiser requires the finite verb to occur in the

clause-final position, as shown by the contrast between (44b) and (44c).

Not only complementisers, wh-element in the initial position of subordinate clause requires

the finite verb to occur clause-finally.

(45) a. Ich will wissen, [was Hugo liest].

b. * Ich will wissen, [was liest Hugo].

In (45a) the wh-expression was is in the initial position of the subordinate clause, and the finite

verb liest is in the clause-final position. The verb-second ordering is impossible in the presence

of a wh-expression, as (45b) shows.


30

Finally, it is not possible for a complementiser and a wh-element co-occur in Standard

German.19

(46) * Ich will wissen, [was daß/ob Hugo liest].

Kathol’s (2000, 2001) approach to the phenomena is as follows. As we have discussed

above, the lexical complementiser daß is lexically assigned to second position (See (41)). In

addition, the position assignment of finite verbs is specified disjunctively as either second or

fourth in their lexical entries. The schematic representation of the entry for the finite verbs

pflanzte is given in (47).

(47)

∨

pflanztePHONDOM

fourthsecond

If the positional assignment of a finite verb were resolved to second, then this position would be

filled by two elements in the case of the embedded clauses with a lexical complementiser, as

represented in (48).

(48) *

[ ] [ ] [ ]

acc

Blume eine

third

nom

Lisa

third

fin

pflanzte

secondsecond

NP

,

NP

,

V

,

DOM

This order domain, however, violates the Topological Uniqueness Condition (43b), which

requires second position to be filled by only one element. Thus, constraints (43b) and (47)

interact to force a finite verb to occur in fourth position when second position is already

occupied by a complementiser.

(49)

[ ] [ ] [ ]

fin

pflanzte

fourth

acc

Blume eine

third

nom

Lisa

third

second

V

,

NP

,

NP

, DOM

Since there is no complementiser in verb-first clauses, a finite verb can occupy second position.

19 Some dialects of German allow a complementiser and a wh-element to co-occur. See Chapter 4 and 6; see also Bayer (1984a,b) and Kathol (1995, 2000).

daß

daß


31

(50)

[ ] [ ] [ ]

acc

Blume eine

third

nom

Lisa

third

fin

pflanzte

second

NP

,

NP

,

V

DOM

The positional assignment of wh-phrases is indeterminate and specified as either first or second.

In root clauses, second position is already filled by the finite verb, and hence the only available

position to the filler is first, as in (51).

(51)

Lisa

third

pflanzte

second

was

first,, DOM

In subordinate clauses, the wh-phrases are assigned to second position in the absence of a

complementiser.

(52)

pflanzte

fourth

Lisa

third

was

second,, DOM

Wh-phrases in subordinate clauses are obligatorily assigned to second position because of the

following constraint.

(53) Finite Clause Constraint (Kathol 2000: 117)

In every finite clause in German, the second topological field is instantiated.

Due to the Finite Clause Constraint (53), order domains of the following sort are unavailable

since second position is not instantiated.

(54) *

,..., DOM

Lisa

third

was

first

The placement options of finite verbs, wh-phrases, and complementisers can be summarised as in

(55). The commonality of wh-phrases and complementisers outlined above follows thus

directly from the assumption that they compete for a single syntactic position, i.e., second, which

may only be filled by a single element due to (43b).


32

(55) Positional possibilities in German

finite verb

[first], [second], [third], … [third], [fourth], … [fourth], …

wh-phrase complementiser

Supporting evidence for an approach along these lines can be found in the commonality of

wh-phrases and complementisers (Kathol 2000, 2001). The standard transformational account

of clause structures with fronted wh-expressions in embedded clauses assumes that the

wh-phrase moves to the [Spec, CP] position.

(56) [CP [Spec was] [C' ∅ [IP Lisa gepflanzte hat]]]

what Lisa planted has

‘what Lisa planted’

It is also assumed that complementisers occupy the Comp head position. Those

complementisers can be phonologically empty, as in (56) above, or overt, as in (57).

(57) [C' [Comp daß] [IP Lisa eine Blume pflanzte]]

that Lisa a flower planted

‘that Lisa planted a flower.’

An approach along these lines does not allow any possibility of positional uniformity of fronted

wh-phrases and complementisers in subordinate clauses. Evidence has been provided, however,

for the uniformity (Reis 1985; Höhle 1992). First, complementisers and wh-fillers may be

coordinated (Reis 1985: 301).

(58) Ich habe erfahren [daß] und [in wen] er sich verliebt hat

I have learned that and in who he self enamoured has

‘I learned that he fell in love and who he fell in love with.’

In (58) the complementiser daß and the wh-phrase in wen are conjoined in the subordinate

clause.


33

Second, bysyllabic lexical complementisers and wh-expressions share phonological

properties in subordinate clauses (Reis 1985: 303). Bysyllabic lexical complementisers have

stress on the second syllable. Examples include obwóhl ‘although’, sobáld ‘as soon as’, or

trozdém ‘although’ in its use as a subordinator.

(59) Ich mag Hans [trozdém/*trótzdem er mich beleidigt hat].

I like Hans although he me insulted has

Wh-expressions allow stress assignment on either the first or second syllable in main clauses as

in (60a). (60b) shows, however, that stress is only allowed on the second syllable in

subordinate clauses.

(60) a. Wárum/Warúm hat Lisa das getan?

Why has Lisa that done

‘Why did Lisa do that?’

b. Ich weiss nicht [warúm/*wárum Lisa das getan hat]

I know not why Lisa that done has

The third piece of evidence involves verum focus. A pitch accent on the finite verb in

German V-first and V-second clauses emphasises the truth of the proposition expressed by the

sentence in declarative clauses (Höhle 1992).

(61) Sie HÖRT mir zu

She listens me to ‘She does listen to me’

A similar emphasis of the clausal propositional content arises if a pitch accent is on a

complementiser.

(62) Aber du siehst doch [DASS sie mirzuhört]

but you see PART that she meto.listen

‘But you can see that she DOES listen to me.’

A pitch accent on a wh-expression in a subordinate clause has the same effect:

(63) Jetzt möcht ich wissen [WEN du reingelegt hast]


34

now would.like I know who you fooled have

‘Now I’d like to know who you DID fool.’

In a matrix question, only the finite verb in the second position may be the bearer of verum focus,

but not the initial wh-expression.

(64) a. * WEN hast du reingelegt?

Who have you fooled

b. Wen HAST du reigelegt?

who have you fooled ‘Who DID you fool?’

Fourth, inflectional morphology, which expresses agreement or finiteness features, occurs

on complementisers and initial wh-phrases in embedded environments (Reis 1985: 301).

(65) a. weil-ste/dass-te endlich kommst

because-2ND.SG/that-2ND.SG finally come-2ND.SG

b. warum-ste/wann-ste kommst

why-2ND.SG/when-2ND.SG come-2ND.SG

These pieces of data suggest that fronted wh-phrases and complementisers in subordinate

clauses have common characteristics. The current standard transformational account of

German clause structure cannot accommodate them since it assumes two different positions for

fronted wh-phrases and complementisers; wh-phrases are in the [Spec, CP] position while

complemtisers are in the Comp head position.

2.6.6 Order Domain-based SLASH Passing

Penn (1999) observes that wh-elements extracted from embedded clauses in Serbo-Croatian

often do not form complete tectogrammatical constituents. In this language, second-position

clitics can attach either to the initial syntactic constituent or to the first prosodic word. In the

latter case, a syntactic constituent can be discontinuous. In (66), the clitic si attaches to u koji

‘in which’, a prosodic word which just corresponds to part of a prepositional phrase [u [koji


35

grad]].

(66) a. U koji si grad znao da je stigao?

in which CL-2s city knew COMP CL-3s arrived

‘Into which city did you know he has arrived?’ (Penn 1999: 179)

b. U koji si znao da je stigao grad?

in which CL-2s knew COMP CL-3s arrived city (Penn 1999: 159)

Since wh-extraction involves the SLASH mechanism, 20 the fact requires giving up the

assumption that LOCAL feature structures are the appropriate value of SLASH (Pollard and Sag

1994). He assumes that SLASH takes list(list(dom_obj)) as its value. In his system, a type

dom_obj contains the PHON feature and the information about compaction, represented by a

structure-sharing of a C(OMPACTION) feature. A domain object does not bear any

tectogrammatical information, such as the SYNSEM attribute.

(66a) is represented as in (67).21 The sign that provides the extracted domain objects, u

koji and grad, is a sister of the verb stigao. At the point of combination with the verb, the

DOM list of the complement is placed on INH|SLASH. The INH|SLASH value is passed up

along the tree to the higher nodes. At the point of combination with si, the domain objects are

removed from INH|SLASH, and they obtain a topological field assignment in the DOM list of

the mother: u koji is assigned to pre-cf and grad to post-cf.

20 Penn (1999) limits the application of the SLASH mechanism to dependencies that cross a clause boundary. See also Kathol (1995: 182–183). 21 In Penn’s (1999) approach to Serbo-Croatian clitic placement, topological fields are arranged into a hierarchy.


36

(67) Representation for (66a) (Penn 1999: 180)

2.7 Conclusion

In this chapter we have looked at theoretical assumptions of HPSG, and some important concepts

and notations have been introduced. Finally, we have looked at the central features of

‘linearisation-based HPSG’.

[ ] [ ]

[ ][ ] [ ][ ]

gradPDOM,kojiu PDOMPROSODY

SLASH|INH

grad,kojiu DOM[ ][ ]stigaoDOM

[ ][ ] [ ]

grad,kojiu SLASH|INH

stigaoDOM[ ][ ]jeDOM

[ ] [ ][ ] [ ]


stigao,jeDOM

[ ][ ] [ ]


stigao je daDOM

[ ] [ ][ ] [ ]


stigao je da,znaoDOM

[ ][ ]daDOM

[ ][ ]znaoDOM

[ ][ ]siDOM

[ ] [ ] [ ] [ ] [ ]

SLASH|INH

stigao je da ,znao ,grad ,si ,kojiu DOM

pre-cf cf post-cf rf pre-cf cf rf

ef

matrix

Chapter 3

Linearisation-based approach to English clauses

Contents

3.1 Positional effects in English..............................................................................................37

3.2 Topological fields for English ...........................................................................................39

3.3 Constraints on linear structure of finite clauses.............................................................42

3.3.1 Lexical constraints .....................................................................................................42

3.3.2 Phrasal constraints.....................................................................................................43

3.3.3 Constraints on internal syntax of clauses.................................................................46

3.3.4 Clausality ....................................................................................................................49

3.4 Clausal constructions ........................................................................................................50

3.4.1 Matrix declarative clauses .........................................................................................52

3.4.2 Yes-no questions ..........................................................................................................53

3.4.3 Matrix non-subject wh-questions..............................................................................53

3.4.4 Matrix subject wh-questions .....................................................................................54

3.4.5 Declarative verb-second clauses................................................................................55

3.4.6 Subordinate declarative clauses................................................................................56

3.4.7 Subordinate wh-questions .........................................................................................57

3.4.8 Relative clauses...........................................................................................................58

3.4.9 Whether- and if-clauses ..............................................................................................59

3.5 Summary ............................................................................................................................60

3.1 Positional effects in English

In Chapter 2 we observed that wh-expressions, complementisers and finite verbs compete for a

single position in German. As noted by Kathol (2002), similar positional competition effects


can be found in English as well. This means that it is possible to analyse English clauses in

terms of topological fields in a parallel fashion with German. This is not surprising in the

light of the fact that English is one of the Germanic languages along with German and the

Scandinavian languages, for which topological analyses have been provided. In this section

we observe the positional effects in English and compare them with German counterparts.

First, yes-no questions in main clauses exhibit subject-auxiliary inversion, but embedded

yes-no questions disallow it.

(1) a. Has she thrown out the garbage?

b. I wonder whether she has thrown out the garbage.

c. * I wonder whether has she thrown out the garbage.

The above examples show a complementary distribution between the complementiser whether

and the finite auxiliary has, which indicates that they compete for a single position, as in

German.

(2) a. Liest Hans die Zeitung?


‘Does Hans read the newspaper?’

b. ob Hans die Zeitung liest

whether Hans the newspaper reads

‘whether Hans read the newspaper’

c. * ob liest Hans die Zeitung

whether reads Hans the newspaper

In environments such as yes-no questions, finite verbs can occur in the clause-initial position,

as in (2a). However, the presence of the complementiser requires the finite verb to occur in

the clause-final position, as shown by the contrast between (2b) and (2c). Thus, in German,

complementisers and finite verbs show a complementary distribution.

Second, non-subject wh-questions exhibit subject-auxiliary inversion only in root

clauses, as illustrated in (3a). The inversion is not possible in subordinate clauses, as in (3b).


(3) a. I wonder what Sandy ate.

b. * I wonder what did Sandy eat.

The examples in (3) show that the wh-element what and the auxiliary did compete for a single

position. In German, not only complementisers, wh-element in the initial position of

subordinate clause requires the finite verb to occur clause-finally.

(4) a. Ich will wissen, [was Hugo liest].

b. * Ich will wissen, [was liest Hugo].

In (4a) the wh-expression was is in the initial position of the subordinate clause, and the finite

verb liest is in the clause-final position. The verb-second ordering is impossible in the

presence of a wh-expression, as (4b) shows.

Third, it is not possible for a complementiser and a wh-element to co-occur.

(5) * I wonder what that/whether Sandy ate

The same phenomenon can be found in standard German.

(6) * Ich will wissen, [was daß/ob Hugo liest].

This indicates that a wh-expression and a complementiser compete for a single position.

The above data show English exhibits positional competition effects that are similar to

those in German: in English wh-expressions, complementisers and finite auxiliary verbs

compete for a single position.

3.2 Topological fields for English

The conclusion we have reached in the last section leads to the prospect that English clauses

can be modelled in a parallel fashion with those in German (and the Scandinavian languages)

in terms of topological fields. As we saw in Chapter 2, Kathol (2000: 115) summarises the

distribution of wh-expressions, complementisers and finite verbs in German as in (7).


(7) finite verb

[first], [second], [third], … [third], [fourth], … [fourth], [fifth],…


Kathol (2002) summarises the distribution of these elements as in (8).

(8) finite auxiliary

[first], [second], [third], [fourth], … [fourth], [fifth],…


(7) and (8) show that there is a parallel distribution in German and English, except for the fact

that the verbal elements that occupy the second position (and the fourth position when the

second position is filled by something else) are finite verbs in German but finite auxiliaries in

English.1,2

As shown in (8), we assume at least five topological fields for English: first, second,

third, fourth and fifth. They are represented by means of subtyping dom-obj (Chapter 2).

(9) The type hierarchy for domain objects in English (version 1)

dom-obj

first second third fourth fifth

The linear ordering among the fields is imposed by the following LP constraint.

1 This is justified by the ungrammaticality of sentences such as the following

(i) a. * At no time went John to London.

b. * Where went John?

In these ungrammatical sentences, the second element is a finite non-auxiliary verb. 2 Following Kathol (2002), we assume that in so-called ‘locative inversion’ constructions have the locative phrase in third and the grammatical subject in fifth. See Kathol (2002). See also Kathol and Levine (1992) for

relevant discussions.


(10) Topological Linear Precedence Constraint for English (version 1)

fifthfourththirdsecondfirst pppp

The distribution of domain elements we will assume for English finite clauses can be

summarised as follows.

(11) Distribution of domain elements in English

first Matrix non-subject wh-phrases, Preposed negative phrases, etc.

second Finite auxiliary verbs in subject-auxiliary inversion (SAI) sentences,

Complementisers, Subordinate non-subject wh-phrases

third Subjects

fourth Finite verbs in non-SAI-sentences

fifth Complements of the finite verb

The topological organisation of domain elements suggested above for English is similar to that

suggested for German by Kathol (1995, 2000, 2001). However, they differ with respect to

the following points. First, wh-non-subjects are in first in matrix clauses and in second in

subordinate clauses, but wh-subjects are always in third; there is no such asymmetry between

wh-subjects and wh-non-subjects in German.3 Second, there can be an arbitrary number of

non-fronted arguments and adjuncts in the third position in German while in English the same

position can hold just one element: the grammatical subject.4 Third, verbal complexes are in

fourth in German, but English does not have verbal complexes; fourth is instantiated by only

finite verbs in non-SAI sentences. Finally, complements of the finite verb are assigned to

fifth; in German, fifth is the position for extraposed elements.

As in the case of German discussed in Chapter 2, cardinality restrictions on position

classes are represented in the following LP constraints (Kathol 2001: 51).


a. firstfirst p

3 In-situ wh-constituents are handled in the same way as non-wh-constituents; subject in-situ wh-phrases are in third, and complement in-situ wh-phrases are in fifth (cf. Ginzburg and Sag 2000). 4 Following Kathol (2002), we assume preverbal adverbials such as always in John always knew the answer is in the same position as the verbal head it modifies; in this case, in fourth.


b． secondsecond p

These constraints entail that there may be only one element in each of the position classes first

and second. The positional competition observed above is due to (12b).

Later in this chapter we will provide a topological analysis of major English clauses, but

before that, some constraints will be introduced in the next section.

3.3 Constraints on linear structure of finite clauses

In this section, some lexical, phrasal and constructional constraints on the linear organisation

of English finite clauses will be introduced.

3.3.1 Lexical constraints

The position class of complementisers is lexically specified. For example, complementisers

that and whether have something like the lexical descriptions in (13).

(13) a.

[ ][ ]

second

austinian

c

DOM

[2]CONTSST-ARG

[2]CONT

HEADSYNSEM

thatPHON

(version 1; cf. Ginzburg and Sag 2000: 46; Kathol 2000: 151)

b.

[ ][ ]

second

npropositio

questionc

DOM

[2]CONTSST-ARG

[2]PROPCONT

HEAD

SYNSEM

whetherPHON

(version 1; cf. Ginzburg and Sag 2000: 214; Kathol 2000: 156)

The position class of the complementisers is specified as second. We follow Ginzburg and

Sag (2000) with respect to other aspects of conplementisers: (i) a complementiser takes a

sentential complement and heads a complementiser phrase (contra Pollard and Sag 1994,


Kathol 2000, etc.); and (ii) complementiser that shares the CONT value austinian with its

sentential argument (Ginzburg and Sag 2000: 46). The type austinian is a supertype of

proposition and outcome. Therefore that allows both indicative and subjunctive complement

clauses. The CONT value of whether is question. Questions involve a proposition in their

construction (Ginzburg and Sag 2000: 121). This is indicated by the feature PROP of the

type question, which takes proposition as its value. The PROP value of whether is identical

to the CONT value of the sentential complement. See Ginzburg and Sag (2000) for further

details.

Finite auxiliaries have the following lexical descriptions, in which their position class is

specified disjunctively as either second or fourth.

(14)

+

∨

AUX

VFORMHEAD|DOM

fin

verb

fourth second

L

Finite non-auxiliary verbs occupy fourth position, as described in (15).

(15)

−AUX

VFORMHEAD|DOM

fin

verb

fourth

L

3.3.2 Phrasal constraints

Some phrasal constraints impose positional constraint on their order domain elements.

(16) [ ]

→

][n' ][2' ]1[DOM

[n] ]2[DTRS-NH

[1]DOMDTR-HD

oKoo

oKoph-hd

The constraint in (16) states that in phrases of the type headed-phrase (hd-ph), the DOM value

of the mother is a sequence union of the head’s DOM value and the compaction of non-head

daughters. In the rest of this thesis we will introduce the following convention: when the

compaction relation is not explicitly stated, a primed tag is used for domain objects and a tag


without a prime for the corresponding sign. For example, [n'] is the domain object for the

sign tagged [n].

One of its subtype, head-complement-phrase (hd-comp-ph), is subject to the constraint

in (17), in addition to the usual constraint which determines the combination of the head and

the complement daughters (Chapter 2).

(17)

→

Ko

KK

[2]DOM

[1],,DTRS-NHph-comp-hd (Revised in Chapter 6)

Where: ( ) ≡[2][1],compaction

[ ]fifth

nelist

sign

:[2]

DOM:]1[

∧

∨

:[2]

DOM:]1[

∧

sign

(17) states that in head-complement phrases the domain objects for non-head daughters are of

the type fifth; the second disjunct in the attached clause allows object traces not to have

domain objects.5

Another subtype of hd-ph, head-subject-phrase (hd-subj-ph) is subject to the constraint

in (18), in addition to the usual constraint which determines the combination of the head and

the subject daughter (Chapter 2).

(18)

→

Ko ]2[DOM

[1]DTR-NHph-subj-hd


[ ]third

nelist

sign

:[2]

DOM:]1[

∧

∨

:[2]

DOM:]1[

∧

sign

(18) states that the domain object of the non-head daughter of a head-subject phrase is of the

type third; the second disjunct in the attached clause allows subject traces not to have domain

5 The constraint on hd-comp-ph will be revised in Chapter 6.


objects.

Let us see how these constraints work. Consider the following simple example.

(19) The girl plays the violin.

The finite verb plays has the subject NP the girl and the object NP the violin. This sentence

has the structure as in (20).


Here and throughout this thesis, the following stands for the sign tagged [1] with its SYNSEM

value [2].

(21) [ ]]1[

[2]NP

The finite verb plays combines with the NP complement via the constraint in (17); the

non-head daughter [3] is compacted, and the resulting domain element [3'] is assigned to fifth

in the VP node. The position for the finite verb is lexically determined to be fourth due to the

lexical entry in (15). The VP combines with the subject NP via the constraint in (18); the

non-head daughter [1] is compacted, and the resulting domain element [1'] is assigned to third

[ ] [ ][ ]girl,theDOM[1]

[4]NP

[ ]

plays]'2[DOM

[5]COMPS

[4]SUBJ

]6[HEADV

fourth

finverb

[ ] [ ][ ]violin,theDOM]3[

[5]NP

][3'],[2',

girl the][1'DOM

[4]HEADS

third

violinthe]'3[],'2[DOM

COMPS

[2]SUBJ

]1[HEADVP

fifth


in the top S node. Thus, linear organisation of the example (19) is (22).

(22)

violinthe

,plays

,girl the

DOMfifthfourththird

The rest of the thesis will occasionally employ schematic representations of the following sort

when other information than linear organisation does not matter.

(23) [DOM <[third the girl], [

fourth plays], [

fifth the violin]>]

(22) and (23) are only alternative representations of the same thing.

Further lexical and phrasal constraints will be introduced in the rest of this thesis.

3.3.3 Constraints on internal syntax of clauses

A linear organisation of clause structure is not only governed by lexical and phrasal constraints

introduced above. Following Kathol (2000: 143), a linear structure of finite clauses is subject

to constraints on each clause type, which are characterised as subtypes of the type finite-clause.

The hierarchical organisation of the types is shown in (24).

(24) Classification of English clauses in terms of their internal syntax

The type finite-clause is subject to the following constraint.

finite-clause

root subordinate

v1 v2

sai non-sai


(25)

[ ]

→

KK ,,DOM

VFORMHEAD

SUBJ

CAT

third

fin

vclause-finite

A finite clause is [SUBJ < >] and is headed by a finite verb. Furthermore, a finite clause

should have an order domain element whose type is third (cf. Kathol 2002). Grammatical

subjects fill third position, due to the phrasal constraint (18) on head-subject phrases.

Finite clauses in English are first of all classified depending on whether or not they are

root or subordinate clauses. The type root is subject to constraint (26).

(26)

∨→KK

L

,[1]HEAD

,DOM

[1]HEAD|

fourth secondroot

(cf. Kathol 2000: 147, 2001: 57)

Constraint (26) requires that clauses of the type root have the finite verbal head whose position

is either second or fourth.

Subtypes of root are classified according to whether or not they involve subject-auxiliary

inversion. The clause type subject-auxiliary-inversion (sai) satisfies constraint (27).

(27)

[ ]

+

→KK

L

,AUX[1]HEAD

,DOM

[1]HEAD|

secondsai

(27) states that in sai clauses an auxiliary verb is a head, which is in second position.

Each subtype of the type sai is classified according to whether or not it includes a

particular kind of element in first position. A v1 clause does not include such an element,

while a v2 clause does. These clause types are constrained by the following constraints.

(28) a. [ ][ ]K,DOM secondv1 → (Kathol 2000: 147)

b. [ ]]2[]1[

,[2]LOCSUBJ

,]1[LOC

DOM

≠∧

→

Ksecondfirst

v2

In both v1 and v2, the element in second position is auxiliary verbal head, due to constraint

(27). (28a) states that the element in second position is the first element of the order domain


in v1 clauses. (28b) specifies the following two points: (i) there is a domain element in first;

and (ii) the element in first does not share the LOC value with a single member of the SUBJ

list of the element in second.6

The requirement (ii) is to exclude a subject in initial position of verb-second structures

in English. One might argue that (29) is a case of a verb-second structure, with the subject in

first position and the non-auxiliary verb liked in second position.

(29) Who liked it?

In English, however, non-auxiliary verbs cannot be in second position. Consider the verb-first

structure in (30).

(30) * Like you Mary?

In this ungrammatical sentence the non-auxiliary verb like is in initial position and it is in

second position. We can thus conclude that the non-auxiliary liked in (29) is not in second

position, and the sentence is not a verb-second structure.

The type non-sai is another subtype of root, and it is subject to constraint (31).

(31) [ ] [ ][ ]elist

thirdfirstsai-non

=∧

⊕⊕⊕⊕→

]1[

]1[DOM LL

(31) specifies that in non-sai clauses, there is no element in second position. The constraints

in (26) and (31) interact to require that non-sai clauses have the verbal head in fourth.

We assume constraint (32) for clauses of the type subordinate.

(32) [ ][ ] [ ] [ ]nesetnesetc

esubordinat

REL QUE HEAD [1]

,[1],DOM

∨∨=∧

→ K

(32) requires that the initial element in the order domain of subordinate clauses is either a

complementiser ([HEAD c]), an interrogative wh-expression ([QUE neset]), or a relative

expression ([REL neset]).

6 Throughout this thesis the inequality symbol ‘≠ ’ requires that the tagged items must not be token-identical.


3.3.4 Clausality

Finite clauses are also classified in terms of their semantics; i.e., sentence modes such as

declarative, interrogative, and imperative. The constraints are organised as in (33).

(33) Classification of English clauses in terms of their semantics

(cf. Kathol 1995, 2000, 2001, 2002)

Each type of clause in English is first of all classified depending on whether it is declarative or

interrogative.

(34) a. [ ]austinianedeclarativ CONT→

b. [ ]questioniveinterrogat CONT→ (cf. Kathol 2001: 59)

The constraint in (34a) requires declarative clauses to be interpreted either as a proposition or

an outcome (3.3.1), and (34b) requires interrogatives to have a question interpretation. The

type interrogative is further classified into wh and polar.

(35) a.

→

neset

questionwh

PARAMSCONT

b.

→

{}PARAMSCONT

questionpolar (a,b from Kathol 2001: 59)

The feature PARAM(ETER)S (Ginzburg and Sag 2000: 121; Kathol 2000: 144) represents a

set of wh-operators. Wh-questions have a nonempty PARAMS set (35a), but polar questions

have an empty PARAMS set (Ginzburg and Sag 2000: 213) (35b).

interrogative declarative

wh polar

finite-clause


3.4 Clausal constructions

In order to formulate generalisations about the shared properties of diverse expressions, the

version of HPSG assumed here includes a cross-classifying multidimensional hierarchy of

constructional types with associated constraints, following the recent development of HPSG

(Ginzburg and Sag 2000; Green and Morgan 1996; Kathol 2000, 2001, 2002; Kim and Sag

2002; Sag 1997); each type inherits constraints from its supertypes. In Kathol’s version,

clausal expressions are classified with respect to two dimensions: ‘internal syntax’ and

‘clausality’, which will be adopted in this study. Each maximal clausal type (i.e., types with

no subtypes) inherits constraints from both dimensions. They also interact with the lexical

and phrasal constraints to determine the positioning of the order domain elements. The type

hierarchy of major clausal types that we propose for English is sketched in (36). In the rest

of this section, we will look at each clausal type in turn.


(36) finite-clause

clausality internal syntax

root subordinate

interrogative rel sai

ns-sub su-sub

decl wh polar v1 v2 non-sai

v2-decl v2-wh v1-pol non-sai-wh non-sai-decl sub-pol sub-decl ns-sub-wh ns-rel su-sub-wh su-rel


3.4.1 Matrix declarative clauses

Clauses of the type non-sai-decl are normal matrix declarative clauses, illustrated by the

following sentence.

(37) The girl plays the violin. [= (19)]

Following Kathol (2002), we propose the positioning of the order domain elements of

non-sai-decl clauses as follows.

(38)

violinthePHON,

playsPHON

[1]HEAD,girl thePHON

DOM

CONT

VFORM]1[HEAD

fifthfourth

third

austinian

fin

verb

decl-sai-non

Clauses of this type inherit the constraints imposed on all the supertypes, non-sai (31), root

(26) and declarative (34a). Constraint (34) on declarative states that clauses of this type

should have a propositional content. Constraint (26) on root requires that clauses of this type

should have a finite verb, which constraint (31) on non-sai requires to be in fourth. The

complement of the finite verb, the violin in (37), is assigned to fifth position, due to the

constraint in (17). The subjects are in third as determined by the phrasal constraint in (18).

We also assume that the constraint in (17) assigns non-finite verb that is complement of an

auxiliary to fifth. Thus, the schematic representation of the order domain of (39a) is (39b).

(39) a. John can dance.

b. [DOM <[third John], [

fourth can], [

fifth dance] >]

Recall that in 3.3.2, the top node of the sentence in (37) was characterised as an instance

of hd-subj-ph introduced in (18). To be precise, therefore, (37) is a subtype of both

non-sai-decl and hd-subj-ph. In the version of linearisation HPSG adopted here, all clausal

signs are thus characterised in terms of multiple inheritance from the constraints of both

phrasal types and constructional types. In the rest of the thesis, however, we will only

mention clausal types in typed feature structures of signs, as exemplified in (38).


3.4.2 Yes-no questions

Clauses of the type v1-pol inherit all the constraints imposed on root (26), sai (27), v1 (28a),

interrogative (34b) and polar (35b). The finite auxiliary head is in second, and it is the first

element of the order domain. The clauses of this type have a question interpretation, and the

CONT|PARAM value is empty. Thus, the sentence Is the girl coming has the following

representation.

(40)

comingPHON,

girl thePHON,

[1]HEAD

isPHONDOM

{}PARAMSCONT

VFORM]1[HEAD

fifththirdsecond

question

fin

verb

pol-v1

The auxiliary verb is is in second, and it is the initial element of this order domain. The

positioning of the rest of the order domain elements is determined by (17) (for the

complements) and (18) (for the subjects).

3.4.3 Matrix non-subject wh-questions

A clause type v2-wh is a subtype of both wh and v2. This type inherits the constraints

imposed on root (26), sai (27), v2 (28b), interrogative (34b) and wh (35a).

(41) What did Sandy eat?

The positioning of the wh-expression is determined by the following constraint on the type

v2-wh.

(42)

→ KK ,

QUE,DOM

neset

firstwh-v2 (version 1)

Constraint (42) states that in v2-wh clauses the wh-expression is in first position. The linear


organisation of (41) is as follows.

(43)

eatPHON,

SandyPHON,

[1]HEAD

didPHON,

{[2]}QUE

whatPHONDOM

{[2]}PARAMSCONT

VFORM]1[HEAD

fifththirdsecondfirst

question

fin

verb

wh-v2

The wh-expression what is first. Constraint (27) on sai requires the finite auxiliary to be in

second. The subject is in third due to the phrasal constrain in (18) on head-subject-phrases.

The non-finite verb eat is a complement of did, so constraint (17) on

head-complement-phrases specifies its position as fifth.

3.4.4 Matrix subject wh-questions

Example (44) is an instance of the type non-sai-wh.

(44) Who wrote the editorial?

Clauses of this type inherit the constraints imposed on all the supertypes, non-sai (31), root

(26), interrogative (34b), and wh (35a): they have a finite verbal head in fourth position, they

have a question interpretation, and their CONT|PARAMS value is non-empty. The

wh-subjects are fillers, and its position is determined by the following constraint,

independently of head-subject phrases.

(45)

→ KK ,

QUE,DOM

neset

thirdwh-sai-non

(45) states that non-sai-wh clauses have a wh-expression in third. Thus who in (44) occupies

third position. The finite verb is in fourth due to constraint (31) on non-sai clauses.

Constraint (17) on head-complement-phrases specifies the position of the editorial as fifth.


(46)

editorial thePHON,

[1]HEAD

wrotePHON,

{[2]}QUE

whoPHONDOM

{[2]}PARAMSCONT

VFORM]1[HEAD

thirdfourththird

question

fin

verb

wh-sai-non

Note the difference in positioning of wh-subjects from wh-nonsubjects discussed in 3.4.3: the

former is in third while the latter is in first.

We will give a detailed discussion of this positioning of wh-expressions in Chapter 4.

3.4.5 Declarative verb-second clauses

Sentence (47a) and the clauses in the square brackets in (47b–e) are instances of clauses of the

type v2-decl.

(47) a. On no account will I write a paper.

b. [So slowly did the workmen get on with their work] that they were dismissed.

(Hawkins 1986: 169)

c. Abby can play more instruments [than can her father]. (Merchant 2003)

d. When Bill smokes, [all the more does Susan hate him].

e. ? The more Bill smokes, [the more does Susan hate him].

(d,e from Culicover and Jackendoff 2005: 515)

Clauses of this type inherit the constraints imposed on all the supertypes: root (26), sai (27),

v2 (28b) and declarative (34a). In addition, v2-decl bears the following constraint.

(48) { }

→ KK ,

QUE,DOM

firstdecl-v2 (cf. Kathol 2001: 59)

(48) states that a v2-decl clause does not have a wh-expression in first position. Clauses of

this type have something like the following structure.


(49)

paper a writePHON,

IPHON,

[1]HEAD

willPHON,

{}QUE

account noon PHONDOM

CONT

VFORM]1[HEAD


npropositio

fin

verb

decl-v2

Chapter 6 will provide a detailed discussion of this clause type.

3.4.6 Subordinate declarative clauses

Clauses of the type sub-decl are subordinate declarative clauses, illustrated by the clauses

marked with that below.

(50) a. Bill knows [that Mary bought a pen].

b. Kate is surprised [that John can dance].

The type sub-decl is a subtype of declarative (34a) and ns-sub, which is a subtype of

subordinate (32). The type ns-sub is subject to the following constraint.

(51) [ ] [ ][2] [1]

,]2[LOCSUBJ

,, HEAD|CAT]1[LOC

DOM

≠∧

¬→

KKfourth

verb

second

sub-ns

(version 1)

This constraint states that clauses of the type ns-sub should have an initial domain object of

the type second, and the latter is not a verb and does not correspond to the subject of the finite

verb in fourth. We propose that sub-decl is constrained by the following constraint.

(52)

→K,

CONT

[1]HEADDOM

[1]HEAD

austinian

cdecl-sub

(52) requires that in subordinate declarative clauses, a complementiser ([HEAD c]) has

austinian (Ginzburg and Sag 2000: 46) as a CONT value (3.3.1), and it is the initial element of

the order domain. The complementiser is required to be in second due to its lexical entry

(13). We further introduce a phrasal type cp-ph in (53), which is a subtype of hd-ph. This

constraint specifies that the complementiser is unioned into the order domain of the sister.


(53) [ ]

[ ]

→

][2]'1[DOM

]2[DOMDTRS-NH

HEAD[1]DTR-HD

o

c

ph-cp

(53) states that the domain object for the head daughters of a cp-ph is sequence-unioned into

the order domain of the non-head daughter.7 The positioning of the order domain elements of

sub-decl clauses is as follows.

(54)

pen aPHON,

boughtPHON,

MaryPHON,

PHON

[2]CONT

[1]HEADDOM

[2]CONT

VFORM]1[HEAD

fifthfourththird

that

second

austinian

fin

v

decl-sub

The elements other than the complementiser are assigned to their respective position in the

same way as the declarative matrix clauses (3.1.1).

3.4.7 Subordinate wh-questions

We now turn to subordinate wh-questions. Examples are the clauses in the square brackets in

(55).

(55) a. I wonder [who will write the editorial].

b. I wonder [who Mary will see].

The subordinate clauses in these examples have the following representations.

(56) a.

seePHON,

[1]HEAD

willPHON,MaryPHON

,

{[2]}QUE

whoPHONDOM

{[2]}PARAMSCONT

VFORM]1[HEAD

fifthfourth

thirdsecond

question

fin

verb

wh-sub-ns

7 Throughout this thesis, the symbol ‘ o ’ indicates a relation of sequence-union (‘shuffle’).


b.

[ ]

editorial thewritePHON,

]3[LOCSUBJ

[1]HEAD

willPHON,

]}2{[QUE

]3[LOC

whoPHONDOM

{[2]}PARAMSCONT

VFORM]1[HEAD

fifth

fourththird

question

fin

verb

wh-sub-su

The position of the wh-phrase in (56a) is determined by the constraint for ns-sub in (51): it is

in second since it does not correspond to the subject of the finite verb in fourth. The position

of the wh-phrase in (56b) is determined by the following constraint for su-sub, which is a

subtype of the type subordinate.

(57) [ ][2] [1]

,]2[LOCSUBJ

,,]1[LOC

DOM

=∧

→

KKfourththird

sub-su

This constraint states that clauses of the type su-sub have an initial domain object of the type

third, and it corresponds to the subject of the finite verb in fourth. In (56a,b) the position of

the finite auxiliary head is resolved to fourth (See constraints (14) and (15)) because second

position is not available: in wh-subject clauses, third is at the left edge of the clause; in

non-subject wh-clauses second is already occupied by the wh-phrase.

We will give a detailed discussion of this positioning of wh-expressions in Chapter 4.

3.4.8 Relative clauses

Let us consider the linear organisation of relative clauses.

(58) a. the person [who Mary likes]

b. the person [who wrote the editorial]

Relative clauses are of the type rel.

(59) [ ][ ]

→

K,{[1]}RELDOM

NPMODHEAD [1]rel (cf. Sag 1997: 451; Kathol 2000: 154)


Following Pollard and Sag (1994) and Sag (1997: 448), REL takes a set of referential indices

as its value. (59) states that the initial element of the order domain of a rel clause must have

a REL value containing only one index, which is the same index as the relative clause’s MOD

value. Relative clauses with a relative subject are of the type su-rel, which is a subtype of

su-sub (57), and the relative subject is in third. Relative clauses with a relative non-subject

are of the type ns-rel, which is a subtype of ns-sub (51), and the relative phrase is in second.

The linear organisation of sentences (58a) and (b) is something like (60a) and (b), respectively.

(60) a.

[1]HEAD

likesPHON,MaryPHON

,

]}2{[REL

whoPHONDOM

CONT

NPMOD

VFORM]1[HEAD

[2]

fourththird

second

fin

verb

rel-ns

L

b.

editorial thePHON,

[1]HEAD

wrotePHON,

]}2{[REL

whoPHONDOM

CONT

NPMOD

VFORM]1[HEAD

[2]

fifthfourththird

fin

verb

rel-su

L

We will give a detailed discussion of this positioning of relative expressions in Chapter 4.

3.4.9 Whether- and if-clauses

Subordinate clauses in examples such as (61) are instances of sub-pol.

(61) I wonder [whether Kim left].

Clauses of this type inherit the constraints imposed on all the supertypes, subordinate (32),

ns-sub (51), interrogative (34b) and polar (35b): they have an initial element in second in the

order domain, they have a question interpretation, and their CONT|PARAMS value is empty.

In addition, sub-pol has its own constraint.


(62)

→K,

{}PARAMSCONT

]1[HEAD|CAT

DOM

[1]HEAD

question

cpol-sub

(62) states that the complementiser head is the initial element of the order domain, and its

CONT value is question whose PARAMS value is empty. The complementisers that can

head this type of clause are whether and if.

The representation of the subordinate clause of (61) is as follows.

(63)

leftPHON,

KimPHON,

[3]CONT

[1]HEAD

whetherPHONDOM

{[2]}PARAMS]3[CONT

VFORM]1[HEAD

fourththird

second

question

fin

verb

pol-sub

Constraint (53) requires that the complementiser should be inserted to the order domain of the

sister. The complementiser is in second due to the lexical constraint in (13) and constraint

(32) on subordinate clauses. The lexical entry in (14) for finite verbs states that they occupy

either second or fourth. In (63) the finite verb should be in fourth since second position has

been already filled by the complementiser. The positioning of the subject is determined by

constraint (45).

3.5 Summary

In this chapter we have argued that positional competition effects among wh-expressions,

complementisers and finite auxiliary verbs can be found in English (Kathol 2002). Then we

have given some constraints in topological terms that can accommodate the positional

competition. We have also provided some phrasal and constructional constraints that


determine the distribution of domain elements in English finite clauses. Finally, we have

looked at some major construction types of English finite clauses.

Chapter 4

Wh-expressions∗

Contents

4.1 Introduction .......................................................................................................................63

4.2 Previous analyses...............................................................................................................64

4.2.1 Principles-and-parameters theory ............................................................................64

4.2.2 Earlier HPSG..............................................................................................................65

4.2.3 Recent HPSG ..............................................................................................................67

4.2.4 Summary .....................................................................................................................68

4.3 Evidence for subject extraction........................................................................................68

4.4 The who/whom distinction ................................................................................................70

4.5 Informal and prescriptive types of register ....................................................................74

4.6 The non-prescriptive type of formal register..................................................................77

4.6.1 Linear structure..........................................................................................................77

4.6.1.1 Extracted subjects in main clauses ........................................................................77

4.6.1.2 Extracted subjects in subordinate clauses .............................................................81

4.6.1.3 Extracted non-subjects in main clauses.................................................................82

4.6.1.4 Extracted non-subjects in subordinate clauses ......................................................85

4.6.1.5 Long-extracted subjects ........................................................................................87

4.6.1.6 Differences from German......................................................................................90

4.6.2 An account of the facts...............................................................................................91

4.6.3 Summary .....................................................................................................................94

4.7 An alternative approach ...................................................................................................94

4.8 Supporting evidence..........................................................................................................97

4.9 Conclusion........................................................................................................................100

∗ Portions of this chapter have already been published as Maekawa (2004).

Chapter 4: wh-expressions

63

4.1 Introduction

In this chapter, we will turn to the linear structure of the left periphery of wh-interrogative and

relative clauses. Some typical examples are given in (1) and (2).1

(1) a. Who visited Mary?

b. I wonder who visited Mary.

c. the man who visited Mary

(2) a. Who did John visit?

b. I wonder who John visited.

c. the man who John visited

In the literature, two types of analyses have been provided for the position of wh-expressions

in (1) and (2). In one type of analysis, all the wh-expressions in (1) and (2) are extracted

from a lower position. Another type of analysis assumes that there is an asymmetry between

wh-subjects as in (1) and wh-non-subjects as in (2): the former are in situ, like normal subjects

like John in (3), while the latter are extracted.

(3) John visited Mary.

In the first type of analysis, all initial wh-expressions are treated alike as extracted, but in the

second type, there is an asymmetry between wh-subjects and wh-non-subjects.

The recent work has given rather convincing evidence that all initial wh-expressions are

extracted (Bouma et al. 2001; Levine and Hukari 2006). In this chapter, we will also adhere

to this view. However, we will argue that linear structure of the left periphery of the

sentences with wh-subjects is different from that of the sentences with non-subject

wh-expressions. Some evidence for this separation will be provided.

The organisation of this chapter is as follows. In 4.2 we will review some previous

analyses of the left periphery of the wh-subject and wh-non-subject constructions. 4.3 will

provide some evidence for the extraction analysis of wh-subjects. In 4.4 we will discuss the

distinction between who and whom, and argue that their rather complex behaviour poses a

1 That-relatives will be treated as wh-relatives in the rest of this chapter.


64

problem for the unified treatment of wh-fillers. In the same section we will show that if we

distinguish three separate register types, that is, informal type, prescriptive type, and

non-prescriptive type, the apparent complexity of the data is restricted to just non-prescriptive

type, and who/whom in the other two types show a rather straightforward behaviour. Section

4.5 will show that the general framework of HPSG can accommodate the who/whom

distinction in the informal and prescriptive types without any additional theoretical apparatus

beyond those proposed in previous work. In section 4.6 it will be shown that a

linearisation-based constraint can provide a straightforward account for the quite puzzling

distribution which who and whom show in the non-prescriptive type. In 4.7 we will consider

an alternative analysis, and argue for the original analysis by giving supporting evidence in 4.8.

Section 4.9 is the conclusion.

4.2 Previous analyses

In this section we examine various treatment of wh-subjects in the literature.

4.2.1 Principles-and-parameters theory

According to Chomsky (1986: 48–52, 58–60), the wh-subject remains in [Spec, IP] and the

auxiliary does not undergo inversion.

(4) [CP C [IP who has fixed the car]]

Non-subject wh-expressions, however, moves to [Spec, CP], and the auxiliary verb undergoes

inversion.

(5) [CP who2 did1 [IP John t1 visit t2]

Thus, wh-subjects and wh-non-subjects are given separate analyses. The wh-subjects in (64)

extracted out of the subordinate clauses are analysed in terms of movement to [Spec, CP] of

the higher clause.


65

(6) a. Who do you think visited Mary?

b. I wonder who you think visited Mary.

c. the man who you think visited Mary

In (6), who is extracted out of the subordinate clause and appears in the clause-initial position

of the higher clause, as in (7).

(7) [CP who2 do1 [IP t1 you think [CP … t2 …]]

Thus, in this analysis, wh-subjects in (1) and those in (6) are treated separately: the former is

in [Spec, IP], and the latter is in [Spec, CP].

More recent work in principles-and-parameters theory tends to give a homogeneous

structural analysis to subject and non-subject wh-expressions: all wh-expressions moves to

[Spec, CP]. A construction with a wh-subject is given something like the following structure

(Rizzi 1996; Pesetsky and Torrego 2001).

(8) [CP who [IP t has fixed the car]]

Unlike the structure in (4), the structure in (8) places wh-subject in line with wh-non-subjects

(5) and long-extracted wh-subjects (7).

4.2.2 Earlier HPSG

In earlier work in HPSG (Pollard and Sag 1994) examples as in (1) are given a slash-less

Subject + VP structure (Gazdar 1981).

(9) Slash-less analysis of (1a)

[ ]{}SLASH

S

Who

|

NP

[ ]

Mary visited

{}SLASH

VP


66

The structure in (9) is exactly the same as that given for normal declarative sentences such as

John visited Mary. On the other hand, non-subject wh-expressions are treated as a filler,

which is a sister of a slashed S (See Chapter 2 for details).

(10) Slash analysis of (2a)

Thus, Pollard and Sag (1994) give separate structures to wh-subjects and wh-non-subjects. In

particular, they suggest that wh-subjects are not extracted and occur in situ.

Long-extracted subjects as in (6) are given a separate analysis. Pollard and Sag (1994)

gives a structure in (11) to these cases.

(11) Pollard and Sag’s (1994) analysis of subject extraction out of embedded clauses

[ ]{}SLASH

S

[ ]

Who

|

[1]LOC

NP

[ ]

John visit did

{[1]}SLASH

S

[ ]{[1]}SLASH|INHER

S [ ]

Who

|

[1]LOC

NP

[ ]{[1]}SLASH|INHER

VP

[ ][ ]

left

[1]LOCSUBCAT

VP[fin]

[ ]

claim

|

]}1{[SLASH|INHER

V

did

|

V

Kim

|

NP

S


67

The head verb claim in (11) is derived from a lexical rule (12) which applies to the verb

governing the clause from which the subject is extracted.

(12) Subject Extraction Lexical Rule (SELR; Pollard and Sag 1994: 349)

[ ]

[ ]

{[1]}SLASH|INHER

,

{}SLASH|INHER

[1]LOCSUBJVP

,COMPS

X

,,COMPS

YSUBJX

⇓

KK

KK unmarked

(12) says that any English verb that selects a nonsubject S complement gives rise to a new

lexical entry that selects an unslashed VP complement. The specification [unmarked] is an

abbreviation of [MARKING unmarked], which means that the sentential complement does not

have a complementiser. The LOC value of the subject of the sentential complement is passed

up with the SLASH mechanism, and it is matched with the wh-expression which is higher in

the tree. In this case the wh-expression is a filler which is a sister of a slashed S.

Thus, Pollard and Sag gives separate analyses to the constructions in (1), (2) and (6).

However, it is possible to say that wh-non-subjects and long-extracted wh-subjects are in

parallel in that they are both fillers and sisters of a slashed constituent.

4.2.3 Recent HPSG

More recent work in HPSG tends to give a homogeneous structural analysis to the

wh-expressions in (1), (2) and (6) (Bouma et al. 2001; Ginzburg and Sag 2000; Levine and

Hukari 2006). They are all treated as fillers, which are sisters of slashed constituents. The

top of the unbounded dependency in the sentences in (1), (2) and (6) is schematically

displayed as in (13).


68

(13) The top of the unbounded dependency constructions in (1), (2) and (6)

4.2.4 Summary

We have seen that in both Principles-and-Parameters theory and HPSG, two types of analyses

have been provided for wh-subjects. Both types of analysis assume that wh-non-subjects in

(2) and wh-subjects in (6) are extracted, but they have different claims on the status of

wh-subjects in (1). Chomsky (1989) and earlier HPSG claims that they are in situ, and recent

P&P and recent HPSG analyses claim that they are extracted. In the next subsection, we will

observe some data which suggest that the latter type of analysis is on the right track.

4.3 Evidence for subject extraction

Evidence has recently been put forth that the subject NP is in fact extracted in the above

examples and the verb heading such a construction is slashed (Bouma et al. 2001; Levine and

Hukari 2006). First, it is possible to coordinate clauses whose subjects have been extracted

with those in which there is an object gap.

(14) a. Robin is the only person who I like and likes me back.

b. Robin is the only person who likes me and I like back.

If these examples show coordinations of unlike categories, an S with a gap and a VP, then they

are predicted to be ill-formed. However, the fact that they are grammatical suggests that the

[ ]

who

|

[1]LOC

NP

[ ]{[1]}SLASH|INHER

[ ]{}SLASH|INHER

S


69

conjuncts are like categories with a trace in each.

(15) a. Robin is the only person who [I like __] and [__ likes me back].

b. Robin is the only person who [__ likes me] and [I like __ back].

This leads to the conclusion that subject extraction involves a gap in parallel with object

extraction.

The second evidence comes from constructions with parasitic gaps. Pollard and Sag

(1994: 195) capture the contrast seen in the examples in (16) with the Subject Condition given

in (17).

(16) a. * That was the rebel leader who rivals of __ assassinated the British consul.

b. That was the rebel leader who rivals of __ assassinated __.

(17) Subject Condition (Pollard and Sag 1994: 350)

A lexical head’s subject can be slashed only if one of its complements is.

In the ill-formed example in (16a), the subject of the lexical head assassinated is slashed, but

the complement is not. (16b) is well-formed since the complement is also slashed. Now let

us consider the following example, where the lexical verb believe has a slashed subject.

(18) Robin is someone that even friends of __ believe should be carefully watched

The Subject Condition requires that the complement clause should be carefully watched

should be slashed. The only element missing in this clause is the subject. This leads to the

conclusion that the complement clause contains a subject gap.

Third, in languages where extraction is locally registered along the extraction path,

subject extraction is indicated in the same way as extraction of other arguments (Hukari and

Levine 1996a,b). Let us consider the following examples in Chamorro, where verbs bear

morphology indicating agreement with arguments that either contain or are themselves

extracted elements. The special morpheme is indicted as WH in the glosses.

(19) a. Hayi f-um-a’gasi i kareta

who WH.SU-wash the car

‘Who washed the car?’


70

b. Hayi si Juan ha-sangan-i hao [f-um-a’gasi i kareta]

who UNM Juan E3S-say-DAT you WH.SU-wash the car

‘Who did Juan tell you washed the car?’

(Bouma et al. 2001)

The main clause subject in (19a) and the embedded subject in (19b) trigger the same

morphological registration. These examples can be accounted for if we assume that the

subject is slashed both in (19a) as well as (19b).

These pieces of data show that subject wh-expressions are extracted and subject

extraction is in line with non-subject extraction.

In this section we have observed evidence that an extraction analysis is superior to an

in-situ analysis. It is important to note that the extraction analysis put short subject

extraction in (1) in line with long subject extraction in (6). However, there is evidence that

the two types of subject extraction should be given separate analyses. In the rest of the

chapter, we argue that the distinction between who and whom observed in a certain register

cannot be handled with a unified analysis.

4.4 The who/whom distinction

It has been traditionally accepted as a prescriptive rule that who is the form for a subject and

subject complement and whom is the form for a verbal or prepositional object. This rule

would require that who should be employed in the following sentences.

(20) a. Who/*whom wrote the editorial?

b. the man who/*whom came to dinner

In (20) who is a subject of the following finite verb, and therefore whom is prohibited. The

prescriptive rule would also require the occurrence of who in the following examples.

(21) a. We feed children who/*whom we think are hungry.


71

b. the man who/*whom I believe has left.

c. the man who/*whom it was believed had left.

In (21) who is a subject of the lower clause, so whom is excluded.

With regard to non-subject positions, however, there is an alternation between who and

whom. As illustrated by the following examples, whom alternates with who as object of a

verb or preposition in main clauses (22), embedded clauses (23), and in situ (24). The

prescriptive rule would predict the occurrence of whom, not who, in these contexts.

(22) a. those whom/who we consulted.

b. someone whom/who we can rely on

c. He didn’t say whom/who he had invited.

(23) a. Whom/who did you meet?

b. Whom/who are you referring to?

(24) a. Who will marry whom/who?

b. Who is buying a gift for whom/who?

c. It was whom/who?2

The important point that we should note is that the prescriptive rule only works in the formal

register. In the informal register, speakers do not stick to this rule and they use who in any

syntactic environment.

The formal/informal register distinction introduced above can account for the occurrence

of whom and the impossibility of who in (25).

(25) a. To whom/*who are you referring?

b. someone on whom/*who we can rely

2 The copular verb be requires an accusative complement, except for the formulaic use of nominative as in It

was I.

(i) a. In this picture, the person in the purple shorts is me/*I.

b. It was just us/*we

See Sobin (1997) and Lasnik and Sobin (2000) for details.


72

In (25), who/whom is in the complement position of a fronted PP. The impossibility of who

in this position will be able to be attributed to the fact that this kind of construction, i.e.,

pied-piping, is confined to the formal register. Given that the construction itself is in the

formal register, the prescriptive rule captures the occurrence of whom in (25) since it is a

prepositional object.

Thus, if we assume separate rules for the formal and the informal register, we can keep

the prescriptive rule for the formal register; for the informal register, who is the only available

form.

There is, however, a striking fact about the formal register: for many speakers, the

distribution of who and whom does not conform to the prescriptive rule. They allow an

alternation of who and whom for the subject of the lower clause in (21).

(26) a. We feed children who/whom we think are hungry.

b. the man who/whom I believe has left.

c. the man who/whom it was believed had left.

As we noted above, the prescriptive rule would predict only the occurrence of who in such a

syntactic environment. It seems that not all native speakers of English accept this use of

whom; for example, Quirk et al. (1985: 368) cites the following example as hypercorrection.

(27) * The ambassador, whom we hope will arrive at 10 a.m., …

They also mention, however, that this kind of use of whom is ‘common’ (1985: 368), and it is

indeed acceptable for many English native speakers.3 In these sentences whom occurs in a

position where its source is the subject of a lower finite clause. If we just assumed the above

prescriptive rule for the formal register, it would lead to the wrong prediction that who is the

only form that appears in such a syntactic context. A satisfactory analysis of the who/whom

distinction in the formal register should be able to ensure that some native speakers of English

accept whom and others reject it in (26); the latter category can be said to manage to conform

to the prescriptive rule.

3 See Jespersen (1924; 1927), Swan (1995), Lasnik and Sobin (2000), Huddleston and Pullum (2002), etc.


73

On the basis of the observation so far, the distribution of who and whom can be

summarised as in (28).

(28) Distribution of who and whom by register type

Formal Environments

non-

prescriptive

prescriptive

Informal

Obj in a fronted PP N/A

Non-subj in embedded clauses

Non-subj in main clauses

Non-subj in situ

whom

Subj of a lower clause

whom

Subj of the first following V who

who

who

We assume that there are two registers: formal and informal. We further assume that there

are two types for the formal register: the prescriptive type and the non-prescriptive type.

Thus we have three types of register: prescriptive, non-prescriptive and informal. (28) makes

it clear that each of the three register types has its own version of the who/whom distribution.

The informal register employs who in every syntactic environment except for the object

position of a fronted PP in pied-piping constructions, which is not available in this register.

In the prescriptive type of formal register, whom is employed in all the non-subject

contexts and who is employed for subjects, whichever clause it is originated from, the upper or

the lower clause (i.e., (20) and (21)). What we should note here is that for this type the

choice of who works in the same way as assignment of nominative case; any theory of

filler-gap dependencies would predict that a filler associated with a gap in the lower clause has

the case that is assigned to the position of the gap. We will see in the next section that the

who/whom distinction in these register types does not pose any problem to HPSG.

Turning to the non-prescriptive type, whom is employed in all cases except where a filler

is the subject of the first following V: whom is used for a filler that corresponds to the subject

of the lower clause (i.e., (26)). This would be totally unexpected if the non-prescriptive type


74

were governed by the same constraints as the prescriptive type. We should also note that this

situation is not consonant with the unified analysis of short-extracted and long-extracted

subjects: this register type has the following paradigm.

(29) a. the man who/*whom has left

b. the man whom/*who I believe has left

In the unified analysis, there is no way to differentiate the distribution of who and whom in

(29). A separate analysis should therefore be provided for who and whom in this type.

We will look at the informal and the prescriptive types in the next section, and show that

no additional theoretical apparatus will be needed beyond those proposed in previous work in

HPSG to give an account for who/whom. The non-prescriptive types poses a problem to

HPSG, to which we will turn in section 4.6.

4.5 Informal and prescriptive types of register

As discussed in the last section, the informal register employs who only. We can give the

following description to this lexical item (cf. Wilcock 1999: 383).

(30)

REGSTR

CASE

who PHON

informal

case

Following Wilcock (1999), we represent register variation in terms of the feature REGISTER

(REGSTR), which is appropriate for CONTEXT. The REGSTR feature takes a value of sort

register, which has two subtypes, formal and informal.

The underspecification of the CASE value in (30) indicates that the informal register

always employs who whatever case it has. Thus, the occurrence of who in (20) to (24) is

captured by this constraint.

(31) a. Who/*whom wrote the editorial? [= (20a)]


75

b. We feed children who/*whom we think are hungry. [= (21a)]

c. those who/*whom we consulted. [= (22a)]

d. Who/*whom did you meet? [= (23a)]

e. Who will marry who/*whom? [= (24a)]

f. To whom/*who are you referring? [= (25a)]

Who in (31a,b) is nominative, and that in (31c,d,e) is accusative. The constraint in (30)

licenses these occurrences of who since its CASE value is underspecified and is compatible

with both nominative and accusative. The unavailability of whom in the informal register

can be accounted for by assuming that this register does not employ this lexical item

whatsoever. The impossibility of who in pied-piping in (25) can be attributed to the fact that

the formal status of pied-piping conflicts with the [REGSTR informal] specification of who.

Wilcock (1999) has provided an argument along the same lines. Wilcock’s (1999) analysis of

pied-piping will be summarised in Appendix D.

Let us turn to the prescriptive type of formal register. As discussed earlier, who

appears not only in an informal style but also in a formal style when it is a subject of the

nearest following verb as in (20), and when it is a subject of the lower clause as in (21).

(32) a. Who/*whom wrote the editorial? [= (20a)]

b. We feed children who/*whom we think are hungry. [= (21a)]

In these syntactic environments, whom is excluded. In all the non-subject environments,

however, whom is employed.

(33) a. those whom/*who we consulted. [= (22a)]

b. Whom/*who did you meet? [= (23a)]

c. Who will marry whom/*who? [= (24a)]

d. To whom/*who are you referring? [= (25a)]

The distribution of who and whom in this type can be formalised along the same lines as an


76

ordinary case assignment.4 We propose that the grammar of the prescriptive type of formal

register includes the following constraints.

(34) a. who (prescriptive type)

REGSTR

CASE

who PHON

formal

nom

b. whom (prescriptive type)

REGSTR

CASE

whom PHON

formal

acc

Who in (32a) is nominative, so it is licensed by (34a). (34b), which only licenses use of

whom when accusative, excludes whom from this environment. The SLASH mechanism

requires the LOC value of the filler to be the same as that of the gap, and therefore a filler

associated with a gap in lower clause is assigned the case that is assigned to the position of the

gap. In the case of who in (32b), the filler has nominative case since the SLASH mechanism

ensures that it has the same LOC value and hence the same case as the gap. Thus, these two

constraints and the HPSG view of unbounded dependencies capture the occurrence of who in

the prescriptive type of formal register, in such examples as (20) and (21). Whom in (33)

occurs in positions where accusative nominal is expected. Therefore, the lexical constraint

(34b) licenses whom in these positions, but who is excluded due to (34a).

In this section, we have shown that existing, independently motivated theoretical

apparatus within HPSG can capture the who/whom distribution in the prescriptive and

informal types. In the next section, we will move on to the non-prescriptive type of formal

register in which who and whom show an apparently puzzling behaviour as discussed in the

earlier sections.

4 For the HPSG literature on case, see Heinz and Matiasek (1994), Meurers (2000), Pollard (1994), Przepiórkowski (1999), etc.


77

4.6 The non-prescriptive type of formal register

The characteristics of the non-prescriptive type of formal register are illustrated by the

following minimal pair.

(35) a. the man who/*whom has left

b. the man whom/*who I believe has left [= (29)]

It is impossible to adopt the case marking strategy proposed for the prescriptive type in the last

section since the SLASH mechanism would allow the CASE value of the both types of subject

to have the same range of choice. It would not be possible to distinguish them in

configurational terms: as we discussed in 4.3, they should be given a unified treatment as

extracted elements: both in (35) are head-filler phrases.

We look at the pair in (35) from the point of view of linear order. In the first sub-section,

we will examine linear structure of the clauses in which who/whom occurs. In 4.6.2 we will

provide some constraints to determine the distribution of who and whom.

4.6.1 Linear structure

4.6.1.1 Extracted subjects in main clauses

In our analysis, the example in (36) has the representation in (37).

(36) Who wrote the editorial?

Let us look at combination of constituents first. The VP has two daughters and its order

domain contains two DOM elements, one for wrote and one for the editorial which has been

compacted to form a single element. The VP combines with the subject trace (Levine and

Hukari 2006) to form the lower S. Traces do not have a domain object (3.3.2; see also

Kathol 1995), so there is nothing corresponding to the trace in the order domain of the lower S.

The top S node has two daughters, the wh-subject filler and the lower S. Its order domain

contains three domain objects, which are for who, wrote and the editorial, respectively. This


78

analysis is consonant with the conclusion we reached in 4.3: wh-subjects are fillers.


In Chapter 3, we argued that matrix subject wh-questions are instances of the type non-sai-wh.

The clause of this type inherits all the constraints from the following types: finite-clause,

interrogative, wh, root, non-sai and non-sai-wh.

{ }

NP

editorial the]'3[,

V

wrote]'2[,

NP

who]'1[DOM

SLASH|INHER

COMPS

SUBJS

fifthfourththird

[ ][4]LOC]1[

NP

][3'],[2'DOM

{[4]}SLASH|BIND-TO

{[4]}SLASH|INHER

COMPS

SUBJS

][3'],[2'DOM

COMPS

[5]SUBJVP

__

|

{[4]}SLASH|INHER

[4]LOC

[5]NP

[ ]]3[[6]NP

][2'DOM

[6]COMPS

[5]SUBJV


79

(38) Type hierarchy for non-sai-wh

The important constraints for the present purpose are root, non-sai and non-sai-wh. The

following is the constraint for root.

(39)

∨→KK

L

,[1]HEAD

,DOM

[1]HEAD|

fourth secondroot

This constraint requires that clauses of the type root should have the finite verbal head whose

position is either second or fourth. (40) is the constraint for non-sai.

(40) [ ] [ ][ ]elist

thirdfirstsai-non

=∧

⊕⊕⊕⊕→

]1[

]1[DOM LL

(40) states that in non-sai clauses there is no element in second position. The constraints for

root (39) and non-sai (40) interact to require that non-sai clauses should have the verbal head

in fourth. In the case of (36) this position is filled by wrote because finite non-auxiliary

verbs occupy fourth position, as described in (41) (See Chapter 3).

(41)

−AUX

VFORMHEAD|DOM

fin

v

fourth

L

The constraint in (42) is for non-sai-wh.

(42)

→ KK ,

QUE,DOM

neset

thirdwh-sai-non

finite-clause

root

non-sai

interrogative

wh

non-sai-wh


80

(42) states that non-sai-wh clauses have a wh-expression in third. In order to ensure that

wh-subject fillers are positioned in third, we need the following constraint.5

(43)

[ ][ ]

[ ]

→

nelistfinverbthird

ph-filler-su-hd

[A][4]LOCSUBJ

[2]HEAD

[3]LOC]'1[DOM

[2]HEADDTR-HD

]1[DTRS-NH

oo

[4] ]3[ =∧

(43) is a constraint for head-subject-filler-phrase (hd-su-filler-ph), a subtype of head-filler-ph,

and specifies the following: in hd-su-filler-ph, the LOC value of the non-head daughter is

token-identical with that of the single element in the SUBJ list of the lexical verbal head, and

the domain object for the non-head daughter is of the type third. This means, among other

things, that the subject wh-filler in (36) is assigned to third. Let us have a closer look at the

order domain of the top S node.

(44)

[ ]

K,

[4]LOCSUBJ

V

wrote ,

[4]LOC

NP

who DOM

fourththird

This representation is consonant with the constraint in (43): the LOC value of who is identical

with that of the single element in the SUBJ list of the verbal head wrote. The position of the

NP the editorial is determined by the constraint in (45) on head-complement phrases.

(45)

→

Ko

KK

[2]DOM

[1],,DTRS-NHph-comp-hd (Revised in Chapter 6)


[ ]fifth

nelist

sign

:[2]

DOM:]1[

∧

∨

:[2]

DOM:]1[

∧

sign

5 The symbol ‘ o ’ notates a relation of sequence-union in this thesis.


81

(45) states that in head-complement phrases the domain objects for non-head daughters are of

the type fifth; the second disjunct in the attached clause allows object traces not to have

domain objects. Thus, according to the constraints on order domains, who is assigned to

third, wrote to fourth, and the editorial to fifth.

4.6.1.2 Extracted subjects in subordinate clauses

Constituent structure of subordinate clauses with extracted subjects, exemplified by the

sentences in (46), is the same as the main clause counterparts (See (37)).

(46) a. I wonder who wrote the editorial.

b. the man who will write the editorial.

However, they are subject to different linear constraints from (37). Embedded wh-question is

a type of subordinate-wh, and relative clause is a type of rel. The former inherits all the

constraints from finite-clause, interrogative, wh, and subordinate, and the latter from

finite-clause, rel, and subordinate.

(47) Type hierarchy for subordinate-wh and rel

We assume constraint (48) for clauses of the type subordinate.

finite-clause

subordinate interrogative

wh

subordinate-wh rel



82

(48) [ ][ ] [ ] [ ]nesetnesetc

esubordinat

REL QUE HEAD [1]

,[1],DOM

∨∨=∧

→ K

(48) requires that the initial element in the order domain of subordinate clauses is either a


expression ([REL neset]). Due to (41), the finite non-auxiliary verb wrote in (46a) is in

fourth. Recall, however, that the position of auxiliary verbs is specified disjunctively as

either second or fourth.

(49)

+

∨

AUX

VFORMHEAD|DOM

fin

v

fourth second

L

The position of the auxiliary will in (46), however, is resolved to fourth. The constraint in

(48) requires that there should be nothing to the left of who, whose position is determined by

(43): in head filler phrases, if the LOC value of the non-head daughter is identical with that of

the single element in the SUBJ list of the lexical verbal head, then the domain object for the

non-head daughter is of the type third. This means that who is in third. Therefore, second

is not available. The rest of the sentences is determined by the constraint on

head-complement phrases in (45): the editorial in (46a) and write the editorial in (46b) are

complements of write and will, respectively, so they are in fifth. To summarise, linear

organisation of the embedded clauses in (46) is something like the following.

(50) a. [DOM <[third who], [

fourth write], [

fifth the editorial]>]

b. [DOM <[third who], [

fourth will], [

fifth write the editorial]>]

4.6.1.3 Extracted non-subjects in main clauses

Let us turn to extraction of non-subjects as in the sentence in (2a), repeated below.

(51) Who did John visit? [= (2a)]

The structure for the sentence in (51) is represented in (52).


83


The verb visit combines with an object trace to form a VP, which contains only a domain

element for visit since a trace does not contribute a domain element. The VP combines with

the auxiliary verb did via the head-complement schema; hence it is compacted to form a

domain element [4] and placed in fifth position in the ordered domain of the higher VP. The

subject John is combined with the higher VP via the head-subject schema (53) to form the

lower S.

{ }

VP

visit]'4[,

NP

John]'3[,

V

did]'2[,

NP

who]'1[DOM

SLASH|INHER

COMPS

SUBJS


[ ][8]LOC]1[

NP

][4'],[3'],[2'DOM

{[8]}SLASH|BIND-TO

{[8]}SLASH|INHER

COMPS

SUBJS

][4'],[2'DOM

{[8]}SLASH|INHER

COMPS

[5]SUBJVP

[ ]]3[[5]NP

][2'DOM

[6]COMPS

[5]SUBJV

]'4[DOM

{[8]}SLASH|INHER

COMPS

]5[SUBJ

[6]VP

[ ]

visit]'4[DOM

]7[COMPSV

__

|

{[8]}SLASH|INHER

[8]LOC

[7]NP


84

(53)

→

Ko ]2[DOM

[1]DTR-NHph-subj-hd


[ ]third

nelist

sign

:[2]

DOM:]1[

∧

∨

:[2]

DOM:]1[

∧

sign



objects. Finally, the lower S combines with the filler who.

This clause type is an instance of the type v2-wh, which is constrained by interrogative,

wh, finite-clause, root, sai, v2 and v2-wh.

(54) Type hierarchy for v2-wh

The constraints for sai, v2 and v2-wh are significant for the present purpose. They are given

in (55)–(57). (55) determines the position of the auxiliary verbal head did.

(55)

[ ]

+→

KK

L

,AUX[1]HEAD

,DOM

[1]HEAD|

secondsai

(55) states that in sai clauses an auxiliary verb, which is a head, is in second. (56) is a

finite-clause

root

sai

interrogative

wh

v2-wh

v2


85

constraint for v2 clauses.

(56) [ ]]2[]1[

,[2]LOCSUBJ

,]1[LOC

DOM

≠∧

→

Ksecondfirst

v2

(56) states that in v2 clauses there is an element in first and its LOC value is not

token-identical to that of a single member of the SUBJ list of the element in second. (57)

specifies the position of the wh-expression.

(57)

→ KK ,

QUE,DOM

neset

firstwh-v2 (version 1)

(57) states that in v2-wh clauses there is a wh-expression in first. The position of a

non-subject wh-filler is determined by the following constraint.

(58)

[ ][ ]

[ ]

∨

→

nelistfinverbsecondfirst

ph-filler-ns-hd

[A][4]LOCSUBJ

[2]HEAD

[3]LOC]'1[DOM

[2]HEADDTR-HD

]1[DTRS-NH

oo

[4] ]3[ ≠∧ (version 1)

(58) is a constraint for head-nonsubject-filler-phrase (hd-ns-filler-ph), a subtype of

head-filler-phrase, and specifies the following: in hd-ns-filler-ph, the LOC value of the

non-head daughter is not token-identical with that of the single element in the SUBJ list of the

lexical verbal head, and the domain object for the non-head daughter is of the type first or

second. The position of the filler should be resolved to first because the constraint in (56)

specifies that there should be an element in first in v2 clauses.

4.6.1.4 Extracted non-subjects in subordinate clauses

The subordinate clauses in the sentences in (59a) and (b) are of the type subordinate-wh and

rel, respectively.


86

(59) a. I wonder who John visited.

b. the man who John will visit

The subordinate clauses in (59) have structure in (60).

(60)

The verb visited combines with the object trace to form a VP, which combines with the subject

John. The resulting S combines with the filler who. Due to the lexical entry in (41), the

finite non-auxiliary verb visited in (59a) is in fourth. The head-subject schema requires the

subject to be in third. The position of the filler is partially determined by the constraint

hd-ns-filler-ph in (58), which specifies that the position of non-subject fillers is either first or

second. We assume the following constraint for clause of the type subordinate.

{ }

VP

visited]'3[,

NP

John]'2[,

NP

who]'1[DOM

SLASH|INHER

COMPS

SUBJS

fourththirdsecond

[ ][6]LOC]1[

NP

][3'],[2'DOM

{[6]}SLASH|BIND-TO

{[6]}SLASH|INHER

COMPS

SUBJS

][3'DOM

{[6]}SLASH|INHER

COMPS

[4]SUBJVP

[ ]]2[[4]NP

[ ]

visited]'3[DOM

]5[COMPSV

__

|

{[6]}SLASH|INHER

[6]LOC

[5]NP


87

(61)

¬

→ KK , HEAD

,DOMverb

secondesubordinat

This constraint states that subordinate clauses should have a domain object of the type second,

and the latter should not be a verb. This means that second position should be filled by either

a complementiser or wh- or relative expression. Therefore, who should appear in second, not

in first.

In at least one dialect of English, embedded wh-questions can satisfy the constraint in

(61) with something which is not a wh-expression. In Belfast English, wh-expressions occur

with a complementiser in subordinate clauses.

(62) a. I wonder which dish that they picked.

b. They didn’t know which model that we had discussed.

c. They asked which book that I had chosen. (Henry 1995: 107 (a,b), 120 (c))

The constraint in (58) leaves positional assignment of extracted non-subjects indeterminate

between first and second position. We assume that the complementiser of this dialect is

lexically determined to be in second as in Standard English. When it occurs, therefore,

position assignment of the non-subject wh-expression should be resolved to first because

second is filled by the complementiser. Thus, the embedded question in (62a), for example,

have the following order domain.

(63) [DOM <[first which book],[

second that],[

third I],[

fourth had],[

fifth chosen]>]

This linear structure satisfies the constraint in (61) since second position is occupied by a

complementiser.

4.6.1.5 Long-extracted subjects

Let us finally look at long-extracted wh-subjects in the sentence in (64).

(64) a. Who do you think visited Mary?

b. I wonder who you think visited Mary.


88

c. the man who you think visited Mary

In (64), who is extracted out of the subordinate clause and appears in the clause-initial position

of the higher clause. We first look at (64a). The structure for this sentence is represented in

(65). Let us first look at combination of constituents. The lowest VP has two domain

elements, one for visited and one for Mary. This VP combines with the subject trace (Levine

and Hukari 2006) to form the lowest S. Traces do not have a domain object (Kathol 1995),

so there is nothing corresponding to the trace in the order domain of the mother node. The

combination of the S with the verb think is governed by the head-complement schema (45).

Domain objects for the non-head daughter of head-complement phrase are totally-compacted.

Thus, the S visited Mary becomes compacted to form a single domain element when it

combines with the verb think. The auxiliary verb do combines with the VP complement think

visited Mary via the head-complement schema, and therefore the latter becomes compacted to

form a single element in the order domain of the mother VP. This VP combines with the

subject you via the head-subject schema to form an S, which combines with the wh-subject

filler via the head-filler schema. The order domain of the top S contains three domain

objects, which are for who, do, you, and think visited Mary.

The linear organisation is almost the same as that of non-subject extraction clauses

discussed in 4.6.1.3. The constraint in (55) determines that the auxiliary verbal head do is in

second. The constraint in (58) requires that the position for the subject wh-filler should be

either first or second. In this case it is assigned to first to satisfy both of the constraints in

(56) and (57): the former constraint for v2 requires that there should be an element in first

position in verb-second clauses, and the latter constraint requires that there should be a

wh-expression in first. The embedded clause is a complement of think, so the constraint in

(45) on head-complement phrases assigns it to fifth. Position of the VP think visited Mary is

fifth due to the constraint in (45) on head-complement phrases; think is a complement of the

finite verb do. The positioning of the matrix subject (you) is determined by the constraint for

head-subject phrases in (53).


89

(65)

Let us move on to subordinate wh-expressions in (64b,c), repeated in (66).

(66) a. I wonder who you think visited Mary. [= (64b)]

b. the man who you think visited Mary [= (64c)]

{ }

VP

Mary tedthink visi]'4[,

NP

]'3[,

V

do]'2[,

NP

who]'1[DOM

SLASH|INHER

COMPS

SUBJS

fifth

you

thirdsecondfirst

[ ][5]LOC]1[

NP

][4'],[3'],[2'DOM

{[5]}SLASH|BIND-TO

{[5]}SLASH|INHER

COMPS

SUBJS

][4'],[2'DOM

{[5]}SLASH|INHER

COMPS

[6]SUBJVP

[ ]]3[[6]NP

][2'DOM

[7]COMPS

[6]SUBJV

]'13[],'10[DOM

{[5]}SLASH|INHER

COMPS

SUBJ

]4[

[7]VP

[ ]

think]'10[DOM

]8[COMPSV

][12'],[11'DOM

{[5]}SLASH|INHER

COMPS

SUBJ

[8]S

]13[

[ ] [ ]

Mary]'12[,visited]'11[DOM

COMPS

]9[SUBJ

VP

__

|

{[5]}SLASH|INHER

[5]LOC

[9]NP


90

The linear organisation is almost the same as that of non-subject extraction clauses discussed

in 4.6.1.4. The position of the filler is either first or second due to the constraint

hd-ns-filler-ph in (58). The constraint in (61) requires that second position should be filled

by either a complementiser or wh- or relative expression. Therefore, who should appear in

second, not in first.

Let us summarise the discussion so far. Short-extracted subjects, unlike extracted

non-subjects, are in third, as in (67a). Long-extracted subjects, however, are like extracted

non-subjects: in main clauses they are in first (67b), and in subordinate clauses they are in

second (67c).


fourth wrote], [

fifth the editorial]>]?

b. [DOM <[first who], [

second do], [

third you], [

fourth think], [

fifth visited Mary]>]

c. [DOM <[second

who], [third you], [

fourth think], [

fifth visited Mary]>]

4.6.1.6 Differences from German

In English, there is an asymmetry between the placement of extracted non-subjects and

extracted subjects. While the former instantiate either first or second, as in (67b,c), the latter

can only occur in third. Such an asymmetry cannot be found in other Germanic languages

(e.g., Kathol 1995, 2000). In other Germanic languages, all wh-fillers (indeed all fillers) are

in first in main clauses and second in subordinate clauses. Thus, the linear organisation of

the sentence in (68a) is something like (68b) (Kathol 2000: 158). The subordinate

counterpart is given in (68c).

(68) a. Wer niest?

who sneezes ‘Who sneezes?’

b. [DOM <[first wer], [

second niest] >]

c. [DOM <[second

wer], [fourth

niest] >]

The extracted wh-subject wer ‘who’ is in first in the main clause example in (68b), and the

subordinate clause counterpart in (68c) has wer in second. The sentence in (69a) is a


91

wh-question with a non-subject wh-expression. Its order domain is represented in (69b), and

its subordinate counterpart is displayed in (69c) (Kathol 2000: 114–115).

(69) a. Was hat Lisa gesehn

what has Lisa seen ‘Who did Lisa see?’

b. [DOM <[first was], [

second hat], [

third Lisa], [

fourth gesehn] >]

c. [DOM <[second

was], [third Lisa], [

fourth gesehn], [

fourth hat] >]

The positioning of the wh-expression and the finite verb in (69b,c) is exactly the same as

(68b,c): both the subject and non-subject wh-expressions are in first in the main clauses but in

second in the subordinate clauses, and the finite verbs are in second in main clauses but in

fourth in the subordinate clauses.

In 4.7, we will consider the possibility that the English extracted subjects have the same

positioning as (68b,c).

4.6.2 An account of the facts

We are now in a position to account for the who/whom distribution in the non-prescriptive type

of register. We assume that the grammar of this register type include the following lexical

constraints for who and whom, instead of (34a,b) for the prescriptive type.

(70) a. who (non-prescriptive type)

who

third

formal

PHON DOM

REGSTR

b. whom (non-prescriptive type)

¬

whom

third

formal

PHON

DOM

REGSTR

The lexical description (70a) allows who to occur only in third. Due to the lexical


92

description (70b) for whom, it is allowed to occur anywhere else.

We can now account for the fact that who is possible but whom is not possible in (35a),

which is repeated in (71).

(71) the man who/*whom has left [= (35a)]

From the discussion in 4.6.1, (71) has the following linear structure.

(72) a. [DOM <[third who/whom], [

fourth has], [

fifth left]>]

In this clause type, third is the position for extracted subjects. For this position, only who is

legitimate because of the constraints in (70): whom is lexically determined to occur in

positions other than third, so it is not allowed in third.

Now let us turn to the use of whom in (35b).

(73) the man whom/*who I believe has left [= (35b)]

From the discussion in 4.6.1, the top S node of (73) has the following linear structure.

(74) [DOM <[second

whom], [third I], [

fourth believe], [

fifth has left]>]

In this clause type, second is the position for extracted subjects. For this position, who is not

allowed since it can only occur in third. On the contrary, whom is lexically determined to

occur in positions other than third, so it can occur in second.

Not only those cases which typically signal the non-prescriptive type of formal register,

the constraints in (70) can also capture the occurrence of whom in (22) to (25). Let us look at

each case.

(75) a. those whom/*who we consulted. [= (22a)]

b. Whom/*who did you meet? [= (23a)]

c. Who will marry whom/*who? [= (24a)]

d. To whom/*who are you referring? [= (25a)]

From the discussion in 4.6.1, positional assignment of the elements in each of these sentences

is as follows.

(76) a. [DOM <[second

whom/*who], [third we], [

fourth consulted]>]

b. [DOM <[first whom/*who], [

second did], [

third you], [

fifth meet]>]


93

c. [DOM <[third

Who], [fourth

will], [fifth marry whom/*who]>]

d. [DOM <[ first To whom/*who], [

second are], [

third you], [

fifth referring]>]

Due to the constraints in (58) and (61), the wh-expression in (76a) is in second. The

constraints in (56) and (58) specifies that the wh-expression in (76b) is in first. For (76c) we

assume the following combination of elements: the in-situ wh-expression is combined with its

selecting head marry, and the resulting VP is a complement of the auxiliary will, so the

constraint in (45) for head-complement phrases requires the VP to be compacted and

positioned in fifth; therefore the wh-expression in situ is not in third. The constraint in (58)

specifies that the PP including the wh-expression in (76d) is in first. In all these cases, the

wh-expressions in question are in positions other than third; therefore who is excluded and

whom is employed in these environments.

Our analysis can also accommodate the following use of who.

(77) A: What did Kim do?

B: What did who do?

The utterance B is an example of an echo question.6 We assume that in-situ wh-expressions

combine with the selecting head in the same way as a normal noun phrases (Ginzburg and Sag

2000): in this particular case, who combines with its head via the head-subject schema.

(78)

→

Ko ]2[DOM

[1]DTR-NHph-subj-hd


[ ]third

nelist

sign

:[2]

DOM:]1[

∧

∨

:[2]

DOM:]1[

∧

sign [= (53)]



6 See Ginzburg and Sag (2000: 255ff) for details of an HPSG treatment of echo questions.


94

objects. Therefore, who in (77B) is in third.

(79) [DOM <[first what], [

second did], [

third who], [

fourth do]>]

Thus, who is employed in this case.

4.6.3 Summary

In this section, we have provided an account for the seemingly puzzling distribution of

who/whom in the non-prescriptive type. The lexical descriptions of who (70a) and whom

(70b) incorporate the specification of the position where they should occur: who is restricted to

third while whom is specified to occur in the positions other than third. What is significant is

that the who/whom distinction is not a matter of case marking in the non-prescriptive formal

register, and that makes it possible to accommodate the occurrence of whom in the cases where

nominative case is normally expected, as in (35b) (repeated in (73)).

4.7 An alternative approach

We have proposed different linear structures for clauses with extracted subjects and those with

extracted non-subjects. Extracted subjects are in third both in main and subordinate clauses,

as in (80a). Extracted non-subjects are in first in main clauses (80a), but in second in

subordinate clauses (80b).


fourth wrote], [


b. [DOM <[first what], [

second did], [

third you], [

fourth write]>]

c. [DOM <[second

what], [third you], [

fourth write]>]

One might argue that it would be desirable to give a unified treatment to wh-fillers, assigning

all of them to first position in main clauses, and second position in subordinate clauses.

(81) a. [DOM <[first who], [

fourth wrote], [


b. [DOM <[first what], [

second did], [

third you], [

fourth write]>]


95

This might be favourable from the cross-linguistic perspective as well because it makes

English more like other Germanic languages, where all wh-fillers can be analysed in first in

main clauses and second in subordinate clauses (See the last paragraph of 4.6.1).

Let us consider an alternative analysis where English counterparts have structures which

parallel these German cases.7 The distinction between who and whom cannot depend on the

difference in position they occur because they are now in the same positions: first position in

main clauses, and second position in subordinate clauses. The alternative constraints would

be something like the following.

(82) Alternative analysis of who and whom

a. [ ]

[ ]whoPHON]2[

][[4]LOCSUBJ

][[1]HEAD

[3]LOC

whoLME]2[DOM

]1[HEAD

↓

nelistA

finverboo

[4] ]3[ =∧

b. [ ]

[ ]whomPHON]2[

][[4]LOCSUBJ

][[1]HEAD

[3]LOC

whoLME]2[DOM

]1[HEAD

↓

nelistA

finverboo

[4] ]3[ ≠∧

Here we assume that a word has the feature L(EXE)ME, which represents the lexeme which

the word form instantiates (Ackerman and Webelhuth 1998). This feature enables us to

assume that who and whom are distinct phonological realisations of the same lexeme: <who>.

(82a) states that that the phonology of the lexeme <who> is <who> if its LOC value is

identical with that of the single element in the SUBJ list of the lexical verbal head. (82b)

7 It will be impossible to assume exactly the same structure for English as German because non-auxiliary verb cannot appear in first position in English.

(i) * Reads Hans the newspaper?

In this ungrammatical sentence the non-auxiliary verb reads is in first. This structure is possible in German.

(ii) Liest Hans die Zeitung?



96

states that the phonology of the lexeme <who> is <whom> if its LOC value is not identical

with that of the single element in the SUBJ list of the lexical verbal head.

Let us now try to account for the fact that who is possible but whom is not possible in

(35a) repeated in (71) and in the following.

(83) the man who/*whom has left [= (35a), (71)]

In this alternative analysis (83) has the following linear structure.


<who>], [fourth

has], [fifth left]>]

The lexeme <who> in second corresponds to a subject of the verbal head has. Therefore,

(82a) predicts that the phonology of the lexeme is <who>.

The use of whom in (35b), repeated in (73) and below, can be accounted for by (82b).

(85) the man whom/*who I believe has left [= (35b), (73)]

From the above discussion, the top S node of (85) has the following linear structure.

(86) [DOM <[second

<who>], [third I], [

fourth believe], [

fifth has left]>]

The lexeme <who> in first does not correspond to a subject of the verbal head believe.

Therefore, (82b) predicts that the phonology of the lexeme is <whom>.

It seems that the constraints in (82) can also capture the occurrence of whom in (22) to

(25), repeated in (75) and below.

(87) a. those whom/*who we consulted. [= (22a), (75a)]

b. Whom/*who did you meet? [= (23a), (75b)]

c. Who will marry whom/*who? [= (24a), (75c)]

d. To whom/*who are you referring? [= (25a), (75d)]

In the alternative analysis, positional assignment of the elements in each of these sentences is

as follows.


<who>], [third we], [

fourth consulted]>]

b. [DOM <[first <who>], [

second did], [

third you], [

fifth meet]>]

c. [DOM <[first Who], [

fourth will], [

fifth marry <who>]>]


97

d. [DOM <[ first To <who>], [

second are], [

third you], [

fifth referring]>]

In all these cases, the lexeme <who> does not corresponds to the subject of the verbal head.

The constraint in (82b) predicts that the PHON value is <whom>, and this is a correct

prediction as can be seen in (87).

This alternative analysis can also accommodate the use of who in (77), repeated in (89).

(89) A: What did Kim do?

B: What did who do? [= (77)]

Here who correspond to the subject of finite verbal head did. Therefore the occurrence of

who in this case is in consonant with the constraint in (82a).

It appears, then, the alternative analysis can capture all the who/whom distinctions we

have been concerned with. In the next section, however, we will see some evidence that

suggests that this alternative is not desirable and the analysis given in 4.6 is preferable.

4.8 Supporting evidence

We can find some evidence for the positioning of extracted subjects in Belfast English. We

have already seen that in this dialect, that can co-occur with wh-expressions in embedded

questions (4.6.1.4).


b. They didn’t know which model that we had discussed.

c. They asked which book that I had chosen. [= (62)]

The embedded question in (90c) has the following order domain.

(91) [DOM <[first which book],[

second that],[

third I],[

fourth had],[

fifth chosen]>]

[=(63)]

As in Standard English, the complementiser that is in second. The constraint in (58) leaves

the positioning of extracted non-subjects indeterminate between first and second position. In


98

this case, however, its position is resolved to first because second has been already filled.

(92) shows, however, that wh-subjects cannot occur with that.

(92) * I wonder which theory that makes the best prediction. (Henry 1995: 120)

Let us maintain our assumptions on positioning of extracted subjects and complementisers:

the wh-subject which theory occupies third position, and the complementiser that is in second.

(93) * [DOM <[third which theory], [

second that], [

fourth makes], [

fifth best prediction]>]

Then the ungrammaticality of (92) can be accounted for by the LP constraint for position

classes, which is repeated in (94).

(94) fifthfourththirdsecondfirst pppp

The permutation in (92), where the wh-subject precedes the complementiser, violates the

constraint in (94) which predicts that the former follows the latter. It seems, then, that the

claim that extracted subjects are in third is on the right track.8

Now let us see how the alternative analysis given in 4.7 might analyse the subordinate

clause of (92). Under this analysis, (92) will have the following linear organisation.

(95) [DOM <[first which theory], [

second that], [

fourth makes], [


Here the extracted subject which theory is in first, in parallel with the non-subject

wh-expression which book in (91). Then, the permutation in (95) is completely consistent

with the LP constraint in (94), so there is no way to account for its ungrammaticality. Thus,

our original analysis is superior to the alternative in that it can accommodate the Belfast

English data.

The second piece of evidence comes from the contrast between Belfast English and the

dialects of German. (96) is an example from Bavarian German.

8 In order to rule out an order domain such as (i) where the complementiser that precedes the wh-expression

which theory, we need to assume the LP rule in (ii).

(i) [DOM < [second

that], [third which theory], [

fourth makes], [


(ii) [ ] [ ]cneset HEADQUE p


99

(96) I woaß ned wann daß da Xaver kummt

I know not when that the Xaver comes (Bayer 1984b: 24)

As in Belfast English in (90), Bavarian German allows wh-expresions to precede

complementisers. The embedded clause in this sentence will have the following order

domain (cf. Kathol 2000: 119).

(97) [DOM <[first wan],[

second daß],[

third da Xavar],[

fourth kummt] >]

As in standard German, positional assignment of extracted phrases is indeterminate between

first and second position. Here the complementiser fills second position, so the non-subject

wh-expression is assigned to first. As we have already discussed, in German, there is no

asymmetry in positioning between subject and non-subject wh-fillers: all sorts of wh-fillers are

assigned to either first or second (See 4.6.1.6). Assuming that this is the case in Bavarian

German as well, this dialect allows subject wh-fillers to appear in first position. Then it is

predicted that it can co-occur with complementisers in second position, contrary to Belfast

English. This prediction is borne out.

(98) a. I woaß net wer daß des tõa hod

I know not who that that done has

‘I don’t know who did that.’ (Bayer 1984a: 212)

b. Es is no ned g’wieß wea daß kummt

it is yet not sure who that comes

‘It is not yet sure who will come.’ (Bayer 1984b: 24)

In (96), the subject wh-filler wer/wea ‘who’ occurs with the complementiser daß ‘that’.

The example in (98) is given the following order domain (Kathol 2000: 119).

(99) [DOM <[first wer],[

second daß],[

third des],[

fourth tõa], [

fourth hod]>]

Like the non-subject filler in (96), the positioning of the subject wh-filler wer is indeterminate

between first and second position. Here the complementiser fills second position, so the

subject filler appears in first.

This gives supporting evidence for the difference between English and German in the


100

position of extracted subjects: in German, both extracted subjects and extracted non-subjects

are in first or second, but in English, extracted subjects are in third although extracted

non-subjects have the same positional possibilities as German (i.e., first or second). The

contrast between (92) and (98) clearly argues for our original claim given in 4.6 that extracted

non-subjects and extracted subjects are constrained by different positional constraints.

4.9 Conclusion

In this chapter, we have been concerned with the linear position of subject-fillers. We argued

that they occupy a rather different position from non-subject fillers: short-fronted

subject-fillers are in third position whereas and long-fronted subject-fillers and non-subject

fillers are in first or second position. As evidence for this claim, we have focused on the

distribution of the English interrogative/relative pronouns who and whom. It was argued that

the constraints in (70) are responsible for the use of who and whom in the non-prescriptive

type of formal register. The constraints are repeated in the following.

(100) a. who (non-prescriptive type)

who

third

formal

PHON DOM

REGSTR

[= (70a)]

b. whom (non-prescriptive type)

¬

whom

third

formal

PHON

DOM

REGSTR

[= (70b)]

These constraints for who and whom can provide a straightforward account of the quite

puzzling distribution which these lexical items show in the non-prescriptive type. This

provides evidence for linear organisation proposed for the clause types in which they occur.

We look at more issues on who and whom in Appendix D.

Chapter 5

Adverbial Preposing∗∗∗∗

Contents

5.1 Introduction .....................................................................................................................102

5.2 Configurational approaches ...........................................................................................103

5.2.1 Minimalist/P&P analyses.........................................................................................103

5.2.1.1 Adjunction to IP and CP......................................................................................104

5.2.1.2 Adjunction to TopP .............................................................................................105

5.2.1.3 Scene-setting projections ....................................................................................107

5.2.1.4 Lower projection in the CP domain ....................................................................109

5.2.2 HPSG analyses..........................................................................................................112

5.2.2.1 Pollard and Sag (1994)........................................................................................112

5.2.2.2 Bouma et al. (2001).............................................................................................115

5.2.2.3 Levine (2003) / Levine and Hukari (2006) .........................................................124

5.2.3 Conclusion.................................................................................................................124

5.3 Differences between long- and short-fronted adverbials.............................................125

5.3.1 Information structure ..............................................................................................125

5.3.2 Blocking of wh-extraction........................................................................................128

5.3.3 Restriction to root/root-like clauses........................................................................130

5.3.4 Summary and conclusion ........................................................................................131

5.4 Extracted phrases............................................................................................................133

5.5 Incidental adverbials.......................................................................................................137

5.5.1 Incidentality ..............................................................................................................137

5.5.2 Incidental adverbials................................................................................................138

5.5.3 Flexible positioning ..................................................................................................138

∗ This chapter is an extended version of Maekawa (2006a).

5: Adverbial Preposing 102

5.5.4 Positioning constraints.............................................................................................141

5.6 An account of the facts....................................................................................................143

5.6.1 Information structure ..............................................................................................143

5.6.1.1 Information structure of incidental adjuncts .......................................................143

5.6.1.2 Information structure of extracted adjuncts ........................................................143

5.6.1.3 Summary .............................................................................................................145

5.6.2 Ordering....................................................................................................................145

5.6.2.1 Incidental adjuncts ..............................................................................................145

5.6.2.2 Ordering of extracted adjuncts ............................................................................150

5.6.2.3 Summary .............................................................................................................152

5.7 Conclusion........................................................................................................................153

5.1 Introduction

In this chapter we deal with preposed adverbials. The following sentences are typical

examples.

(1) a. The next day she sold her house.

b. With a hammer he broke the computer.

Minimalist/P&P approaches have analysed preposed adverbials as occupying a fixed position

in the tree configuration, such as adjoined to IP, or occupying the specifier position of CP or

TopP. In Pollard and Sag’s (1994) version of HPSG, preposed adverbials are adjoined to the

S which they modify. More recent versions of HPSG assume that they are fronted by the

SLASH mechanism in the same way as arguments (Bouma et al. 2001; Kim and Sag 2002).

In these frameworks a rigid link between linear order and constituent structure determines the

linear position of such adverbials in the sentence-initial position. Let us call this type of view

on linear order ‘configurational’. The alternative analysis within linearisation-based HPSG

assumes a more flexible relation between linear order and constituent structure.


This chapter will consider the possibility of providing a detailed analysis of adverbial

preposing constructions within these two views on linear order. I will argue that there is a

body of data which suggests that adverbial preposing do not work as the configurational

approaches predict. I will then show that linearisation-based HPSG can provide a fairly

straightforward account of the facts.

The organisation of this chapter is as follows. In the next section we present the

relevant data. In section 5.2 we outline the configurational approaches to adverbial

preposing constructions, and point out problems. In section 5.3 we will provide detailed

descriptions of the differences between long and short fronting of adjuncts, and argue that a

type of short-fronted adjunct, namely, ‘incidental adjuncts’, have a different properties from

the other type, namely, extracted adjuncts. We will discuss extracted adjuncts in section 5.4.

Our analysis of incidental adjuncts, partly based on Bonami et al.’s treatment of incidentals,

will be given in section 5.5. In section 5.6, we will see how our approach to fronted

adverbials handles the data observed in earlier sections. Section 5.7 is the conclusion.

5.2 Configurational approaches

In this section we consider how the configurational approaches to linear order might work.

In 5.2.1 we will look at Minimalism/Principles-and-Parameters theory, and then in 5.2.2 we

will look at HPSG.

5.2.1 Minimalist/P&P analyses

There are at least four types of analyses proposed for explaining the facts about adverbial

preposing. The division depends on the position in constituent structure to which preposed

adverbials are assigned.


5.2.1.1 Adjunction to IP and CP

In much previous work which assumes clause structure along the lines of the one in (2a),

adverb preposing in English is assumed to be adjunction to IP (or S) within CP, as shown in

(2b) (Baltin, 1982; Culicover, n.d.: 34 (105); Rochemont 1989).

(2) a. [CP C [IP I [VP V …]]]

b. [CP C [IP ADV [IP I …]]]

Baltin (1982: 25) gives (3) as evidence that preposed adjuncts adjoin to the IP.

(3) It’s obvious that, although he’s a nice guy, John isn’t too bright. (Baltin 1982: 25)

Here the preposed adverbial although he’s a nice guy is between the complementiser that and

the subject John, and it would be reasonable to say it is adjoined to the IP.

However, the sentences in (4) are problematic.

(4) a. [IP During the holidays [CP for what kind of jobs would you go into the office?]]

b. * [CP What [C will [IP during the holidays [IP you do]]]]

c. * [CP On no account [C will [IP during the holidays [IP I write a paper]]]]

This approach predicts that (4a) is ungrammatical because the adverbial is adjoined to CP, and

(4b) and (4c) are grammatical because the adverbial is adjoined to IP.

One might argue, as Baltin (1982: 26) does, that preposed adverbials are adjoined to the

CP when co-occurring with a wh-element. Wh-elements are normally assumed to be in the

specifier position of CP. Hence it might be reasonable to assume that the preposed adverbial

in (4a) is adjoined to the CP, as shown in (5).

(5) [CP During the holidays [CP for what kind of jobs would you go into the office?]]

However, arguments along these lines lead to an incorrect prediction. An embedded

wh-question is a CP. If adverbials could be adjoined to the CP, as in (5), the structure in (6)

should be felicitous. However, the sentence is ungrammatical.

(6) * I was wondering [CP during the holidays [CP for what kind of jobs you would go


into the office.]]

Moreover, there is still no way to predict the ungrammaticality of (4b,c).

Thus, the IP/CP adjunction analysis of preposed adjuncts has problems.

5.2.1.2 Adjunction to TopP

The recent trend within Minimalism/Principles-and-Parameters theory is to split CP into a

number of separate projections. Rizzi (1997) proposes the following structure for the left

periphery of a clause.

(7) [ForceP Force0 [TopP* Top

0 [FocP Foc

0 [TopP* Top

0 [FinP Fin

0 [IP …

ForceP encodes the clause types and determines whether the clause is a question, a declarative,

an exclamative, a relative, a comparative, etc. The head of ForceP carries information which

is accessible to a higher selector. FinP is a projection whose head carries the features for

(non-)finiteness. The specifier of FocusP hosts a focalised constituent and its head hosts the

focus feature. The specifier of TopicP hosts the fronted topic and its head hosts a topic

feature. Much work within the Split-CP hypothesis assumes that a wh-element is in the Spec

of FocP in main interrogative clauses and in the Spec of ForceP in embedded interrogatives.

Since the subordinate questions are selected by a matrix predicate, Force is associated with the

wh-feature (Culicover 1991; Rizzi 1997; Haegeman 2000a,b). In embedded wh-questions,

therefore, the wh-element moves to [Spec,ForceP] to establish a spec-head relation with the

wh-feature.1 Preposed negative phrases are assumed to be in the Spec of FocP in both main

and embedded clauses (See also Haegeman 2000a,b). Furthermore, Rizzi (1997: 300)

assumes that preposed adverbials are adjoined to the higher TopP. It is also assumed that

unlike Italian English does not have the lower TopP.

The structure in (7) fails to capture the grammaticality of the sentences in (8).

(8) a. [FocP For what kind of jobs [TopP during the holidays would you go into the office?]]

1 This makes the auxiliary inversion unnecessary.


b. [FocP On no account [TopP during the holidays will I write a paper]].

c. John said that [FocP on no account [TopP during the holidays would he write a

paper]].

These structures are incompatible with the structure in (7). In all these cases, a constituent

which is assumed to be adjoined to TopP follows a constituent which is assumed to be in [Spec,

FocP]. Moreover, these sentences do not satisfy the WH-Criterion and Negative Criterion,

which are defined as in (9) and (10), respectively (See, e.g., May 1985; Motitz and Valois

1993; Haegeman 1995, 2000a: 123, 2000b: 23; Haegeman and Zanuttini 1991, 1996; Rizzi

1996).

(9) The WH-Criterion

a. A wh-operator must be in a Spec-head configuration with an X0[+wh]

b. An X0[+wh] must be in a Spect-head configuration with a wh-operator.

(10) The Negative Criterion

a. A negative operator must be in a Spec-head configuration with an X0[+neg]

b. An X0[+neg] must be in a Spec-head configuration with a negative operator.

These criteria require that the wh-phrase or negative phrase should be in a spec-head

configuration with a [+wh] or a [+ neg] feature which is assumed to appear with the auxiliary.

In (8), however, these requirements are not met: the wh-/negative phrases are not in a

spec-head configuration with an auxiliary verb. Thus, Rizzi’s (1997) analysis would predict

their ungrammaticality, which is not borne out.

Thus, the TopP analysis of preposed adverbials has some problems.2

2 In addition, Rizzi’s (1997) system predicts that in languages with the lower topic position like Italian, the fronted adverbials should be able to follow elements in Spec of FocP, such as interrogative wh-constituents.

Haegeman (2004a: 183) points out that this prediction is not borne out. In the following Italian sentence, the

wh-constituent che cosa is followed by the fronted adverbial rapidamente, but this is ungrammatical.

(i) * Che cosa, rapidamente, hanno fatto?

‘what, rapidly, did they do?’


5.2.1.3 Scene-setting projections

Haegeman (2000a: 143ff) postulates an additional functional category for preposed adjuncts in

the split CP structure. They are generated as specifiers of Sc(ene-Setting) P(rojections),

projected by Sc0. ScP is assumed to dominate FocP (Haegeman 2000a: 146). ScP gives

information on the scene, such as time, place, etc. ScP is recursive. Sc0 heads can merge

with each other, forming a complex head such as [Sc1 Sc2i-Sc1] through head-to-head

movement, as in (11).

(11) [ScP1 Next week [Sc1' [Sc1 Sc2i-Sc1] [ScP2 if I have time [Sc2' [Sc2 ti] [IP I will …]]]]]

In (11), the time adjunct next week and the conditional adjunct if I have time are generated as

specifiers of separate Sc projections. The Sc2 is incorporated into the Sc1 to form a complex

head, which defines a single scene.

Let us look at the sentences in (8), in which the Wh-Criterion or Negative Criterion

appears not to be satisfied since the adverbials intervene between the wh-/negative phrase and

the auxiliary with a [+ wh]/[+ neg] feature. Haegeman assumes something like the following

structure for them.


(12) Structure for (8a) (Haegeman 2000a: 149)

Haegeman assumes that the wh-/negative phrase first moves to the Spec of FocP and

subsequently moves up to the Spec of ScP1. She further assumes coindexation between the

head and its specifier (2000a: 150). Thus, the trace of the wh-/negative phrase in the Spec of

Foc is coindexed with the Foc head. Let us call the index j. The wh-/negative phrase then

moves to the Spec of ScP1, and its complex head [ScP1 Sc2i-Sc1] obtains the index j because it

is now in a spec-head relation with an element indexed j. Consequently, [ScP1 Sc2i-Sc1] and

Foc share the same index j. This enables to form a representational head chain

<<Sc2i-Sc1,ti>,Foc>.3 The Foc, the lowest member of the chain, carries a wh-feature, and the

3 Haegeman (2000a) follows Rizzi’s (1986: 66) following characterisation of chain:

C = <x1, …, xn> is a chain iff, for 1 < i < n, xi is the local binder of xi + 1, where

(i) x is a binder of y iff, for x, y = a category, x and y are coindexed, and x c-commands y;

PP

For what

kind of jobs

[+WH]

ScP1

Sc'1

Sc1

Sc2i Sc1

ScP2

PP

during the

vacation

Sc'2

Sc2

|

ti

FocP

PP

|

Foc'

Foc

|

would

FinP

|

…


head (i.e., highest member) of the chain, Sc2i-Sc1, has the wh-phrase as its specifier. The

wh-phrase satisfies the WH-Criterion by being in a long-distance spec-head relationship with

the auxiliary in Foc. Thus, the wh- and the negative phrases in (8) satisfy the WH/NEG

Criterion even if there is an intervening adverbial in the Spec of Sc2.

It is clear that this analysis requires a quite complicated mechanism. If it is possible to

give a simpler analysis, that would be preferable.

5.2.1.4 Lower projection in the CP domain

Haegeman (2002a, 2004a) makes some modifications for the Split-CP structure proposed by

Rizzi (1997) (see (7)). As a motivation for this modification, she pays attention to the

difference between what she calls ‘central adverbial clauses’ and ‘peripheral adverbial

clauses’.

(13) a. If it rains we will all get terribly wet and miserable.

b. If [as you say] it is going to rain this afternoon, why don’t we just stay at home

and watch a video? (Haegeman 2002a: 117)

The conditional clause in (13a) specifies a sequential and causal relation between the two

events: that of raining and that of getting wet and miserable. The conditional clause in (13b)

expresses a premise that leads to the question expressed in the main clause. The first type of

adverbial clause is analysed as inserted in the IP of the matrix clause, and hence called

‘central’, while the second type is analysed to be attached outside the matrix CP, and therefore

‘peripheral’. Haegeman proposes that peripheral adverbial clauses have ‘illocutionary force’,

but central adverbial clauses do not. She also claims that the presence of Force is directly

correlated with illocutionary force (2002a: 159). This leads to the hypothesis that central

adverbial clauses are structurally deficient in lacking Force while peripheral adverbial clauses

display the full clausal structure as root clauses do (See also Hooper and Thompson 1973).

(ii) x is the local binder of y iff x is a binder of y and there is no z such that z is a binder of y, and z is not a

binder of x.


Haegeman (2002a, 2004a) proposes that the three types of clause have the following

structures.

(14) a. Central adverbial clauses

[SubP Sub0 [ModP* Mod

0 [FinP Fin

0 [IP …]]]

b. Peripheral adverbial clauses

[SubP Sub0 [ForceP Force

0 [TopP* Top

0 [FocusP Focus

0 [ModP* Mod

0 [FinP Fin

0 [IP

…]]]]]]

c. Root clauses

[ForceP Force0 [TopP* Top

0 [FocusP Focus

0 [ModP* Mod

0 [FinP Fin

0 [IP …]]]]]]

Here there are two new additions to the set of empty heads. Mod(ifier) hosts adjuncts

(Haegeman 2002a: 148, 2004a: 181; See also Rizzi 2001).4,5 Sub(ordinator) serves to

subordinate the clause, and hosts complementisers and wh-elements in embedded clauses

(Haegeman 2002a: 159, 2004a: 168). As shown in (14a), Force is missing in central

adverbial clauses since it lacks illocutionary force. Haegeman (2002a: 161) assumes that Top

and Foc depend on Force. Therefore, (14a) includes neither Top nor Foc. (14b,c) show that

peripheral adverbial clauses and root clauses have the full clausal structure. Haegeman

(2004a) suggests that (14) is supported by the following contrastive acceptability concerning

fronted arguments included in them.

(15) a. If these problems we cannot solve, there are many others that we can tackle

immediately.

b. * If these exams you don’t pass you won’t get the degree.

(Haegeman 2004a: 159–160)

Topicalisation is allowed in peripheral adverbial clauses, as in (15a), but it is not allowed in

central adverbial clauses as in (15b). This contrast between the two types of clauses might

4 More precisely, the Mod hosts short-moved adjuncts. Long-moved adjuncts and topicalised arguments are in the Spec of the TopP. 5 Haegeman (2002b) suggests the left edge of IP, and Haegeman (2004b) suggests FinP for the site of preposed adverbials. See also Rizzi (1997: endnotes 26, 30, 32) for the FinP analysis.


follow from the assumption that Top is missing, along with Force and Foc, in central adverbial

clauses. However, adverbial preposing is possible in both of these types of clause.

(16) a. If with all these precautions you have not succeeded, why do you want to try

again?

b. If with these precautions you don’t succeed, you should try again next week.

(Haegeman 2002a: 149)

Thus, the fact that adverbial preposing is possible in central adverbial clauses means that the

position of preposed adverbials should be lower than Force, Top and Foc.

However, this analysis makes incorrect predictions as well. First, it predicts that

adjuncts cannot precede wh-phrases. The sentences in (17), however, are grammatical.

(17) a. During the holidays for what kind of jobs would you go into the office?

b. During the holidays on no account will I write a paper.

c. John said that during the holidays on no account would he write a paper.

Second, the ungrammaticality of (18a,b) does not follow from (14). Here FocP dominates

ModP, and there is nothing to prevent the preposed adverbial from following the wh-phrase

and the preposed negative phrase.

(18) a. * [FocP What [Foc will [ModP during the holidays [IP you do]]]]

b. * [FocP On no account [Foc will [ModP during the holidays [IP I write a paper]]]]

One might argue that it would be possible to add a higher projection, such as

Scene-setting projections to the inventory in (14). However, this would lead to the same

problems as the analysis outlined in 5.2.1.3.

There is another problem. What Force really deals with is sentence type, such as

declarative, interrogative, and so on (Bob Borsley, p.c.). It should not be mixed up with

illocutionary force, which determines the type of speech act that the sentence is in. For

example, interrogatives can be used to ask questions but they can also be used in other ways:

e.g., the sentence type of ‘Would you like to close the window?’ is an interrogative, but its


illocutionary force is a directive, not a question (Huddleston and Pullum 2003: 62).

Therefore, Haegeman’s (2002a: 159) proposal that the presence of Force is directly correlated

with illocutionary force is questionable.

Thus, the ModP analysis of preposed adverbials has some problems.

5.2.2 HPSG analyses

In this section, we look at three types of analysis of preposed adjuncts which have been

proposed in the framework of HPSG: Pollard and Sag’s (1994) S adjunction analysis, and the

long-distance dependency analyses of Bouma et al.’s (2001) and Levine (2003) / Levine and

Hukari (2006).

5.2.2.1 Pollard and Sag (1994)

Pollard and Sag (1994: 385) analyse preposed adjuncts as matrix modifiers, which are simply

adjoined to the clause that they modify.

(19) Tomorrow, will Dana go to Spain?

Pollard and Sag (1994) propose that adjuncts select their head. They introduce the head

feature MODIFIED (MOD) for this purpose. The SYNSEM value of the lexical entry for the

preposed adverbial tomorrow includes the following information at least.

(20)

SMOD HEAD | CATadv

The partial lexical entry (20) says that this adverb selects S as its head. An adjunct with this

information and its head combine via the ID schema called ‘Schema 5’ (Pollard and Sag

1994:56).

(21) (Schema 5) a phrase with DTRS value of sort head-adjunct-structure

(head-adj-struc), such that the MOD value of the adjunct daughters is


token-identical to the SYNSEM value of the head daughter.

(Pollard and Sag 1994:56)

In (19) the adverbial tomorrow and the rest of the clause make an instance of

head-adjunct-structure, whose structure is schematically represented in (22).

(22) Head-adjunct structure for Tomorrow will Dana go to bed?

The tag [1] indicates the structure-sharing of the value of the MOD feature of the adverb and

the SYNSEM value of the S.

Chapter 9 of Pollard and Sag (1994) gives a separate treatment to long adverbial fronting.

They posit the Adjunct Extraction Lexical Rule.

(23) Adjunct Extraction Lexical Rule (Pollard and Sag 1994: 387)

[ ]{ }

[ ]{ }

⇓

[3]ARG-SOA|CONTENT

]3[:[2]MOD'Y'SLASH|INHER

[2], ,COMPS

SLASH|INHER

, SUBJ'[2]V' ,COMPS

KK

KK

The sentence with a long fronted adjunct (24) has a structure as in (25).

(24) Yesterday I believe Kim left.

In the structure in (25), the verb believe has a lexical entry produced by the Adjunct Extraction

Lexical Rule (23). The LOC value of the preposed adjunct yesterday is [3]. As represented

in the structure of believe, the MOD value of this LOC value is [2], which is the SYNSEM

S1

[1]S2 ADV

[MOD [1]]

Tomorrow will Dana go to Spain?


value of the embedded clause Kim left. Long fronted adjuncts are thus ensured to modify the

lower clause.

(25) Structure for (24) (Pollard and Sag 1994: 387)

(26) is problematic to Pollard and Sag’s (1994) treatment of short adverbial fronting.

(26) * I was wondering [S during the holidays [S for what kind of jobs you would go into

the office.]]

In (26), the preposed adverbial modifies and combines with an S, conforming to the schema in

(22). However, (26) is ungrammatical.

One might introduce the head feature INDEPENDENT-CLAUSE (IC) to accommodate

this. This feature is to distinguish between independent clauses and the others (Ginzburg and

Sag 2000: 45). Consider (27) for example.

(27) a. Problems of this sort, our analysis would never account for.

b. * She subtly suggested [problems of this sort, our analysis would never account for].

The topicalised sentence in (27a) is an independent clause (i.e., [IC +]), hence its head

V''

ADV''

[LOC [3]]

yesterday

V''

[INHER|SLASH [3]]

[1]N''

|

I

V''

[SUBJ <[1]>, INHER|SLASH {[3]}]

V0

[SUBJ <[1]>, COMPS <[2]>,

INHER|SLASH {[3][MOD [2]]}]

|

believe

[2]V''

Kim left


daughter our analysis would never account for has [IC +]. A clause has the [IC −]

specification in an embedded environment, and hence the embedded clause in (27b) is [IC −].

Thus, if a clause is specified as [IC −], it cannot be an independent clause and must be an

embedded clause. If we assumed that preposed adverbials selected only the S with the [IC +]

specification, the ungrammaticality of (26) might be accounted for since the adverbial

modifies the embedded wh-question, i.e., [IC −]. However, arguments along these lines

would make an incorrect prediction. Let us consider (28).

(28) I was wondering [S[IC −] for what kind of jobs [S[IC −] during the holidays [S[IC −] you

would go into the office.]]]

In (28) the preposed adverbial occurs in the clause with the specification [IC −]. This means

that the S that the adverbials modify is also [IC −] because the Head Feature Principle ensures

that the HEAD value of the mother is structure-shared with the head value of the head

daughter. This example is then predicted to be ungrammatical. However, it is grammatical.

5.2.2.2 Bouma et al. (2001)

The notable feature of Bouma et al.’s (2001) treatment of adverbials is that they are selected for

by a lexical head by the same mechanism which accounts for the selection of complements

(Abeillé and Godard 1997; Kim and Sag 2002; van Noord and Bouma 1994; Przepiórkowski

1999a, 1999b).6 Bouma et al. (2001) assume that the ARG-ST list contains all and only the

selected arguments of a lexical head. They also introduce the DEPS list for specifying the list

of dependents of a lexical head, including synsems of selected arguments and adverbials. The

relationship between ARG-ST and DEPS is defined by the following constraint (Bouma et al.

2001: 42).

6 Kim and Sag (2002: 384) assume that at least locative, temporal, and manner adverbials are selected as optional complements by the verbal head.


(29) Argument Structure Extension

⊕

⇒

]2[KEY|CONT

[1]ST-ARG

[2]KEY

[3]HEADMOD[1]DEPS

[3]HEAD

listverb

This constraint allows a verb’s DEPS list to contain any number of adverbials along with the

verb’s true argument. The MOD|HEAD value of the adverbial is identified with the HEAD

value of the verb (tagged [3] in (i)), so the selectional relation between the modifier and the

modified element is ensured (See 5.2.2.1). The value of the feature KEY is the semantic

relation introduced by the lexical head of the phrase, which is passed up from the head of a

phrase to the mother (Bouma et al. 2001: 18).7 The adverbial’s MOD|KEY value is unified

with the KEY relation introduced by the verb.

The Argument Realisation Principle defines the relationship between DEPS lists and the

valence features (e.g., SUBJ(ECT) and COMP(LEMENT)S) (Bouma et al. 2001: 12; see also

Ginzburg and Sag 2001: 171).

(30) Argument Realisation

⊕

−⇒

[2][1]DEPS

)-([2]COMPS

[1]SUBJ

ssgaplistword

(30) ensures that all dependents, except for a gap-ss, are realised on the appropriate valence

lists, and hence are selected by a head.8 Constraints (29) and (30) combine to ensure that the

synsem of adverbials may appear in the COMPS list of the verb.

With these assumptions, adjunct-preposing can be given a unified treatment with

argument topicalisation: both constructions are combination of the filler and the slashed

construction (Bouma et al. 2001; Kim and Sag 2002). Bouma et al. (2001: 47) gives the

structure in (32) for (31).

(31) On Tuesday Sandy visits Leslie.

7 Bouma et al (2001) adopt the semantic architecture of Minimal Recursion Semantics (Copestake et al 1999). 8 A gap-ss is allowed to appear in the SUBJ list



V

In (32) the DEPS of the verb visit contains an adverbial element, whose synsem is specified as

a gap-ss. Gap-ss, a subtype of synsem, is specified to have a nonempty value for the feature

SLASH. Its LOC value corresponds to its SLASH value.

(33)

⇒

{[1]}SLASH

[1]LOCss-gap

Note that the gap-ss in the representation of visit, which is in the DEPS list, does not appear in

a COMPS list. The nonempty SLASH value is incorporated into the verb’s SLASH value by

the SLASH Amalgamation constraint (Bouma et al. 2000: 20).

{ }

WH

[1]LOCPP

On Tuesday

[2]NP

Sandy

[3]NP

Leslie { }{ }

[1]SLASH

[1]SLASH

[1]LOC[3],[2],DEPS

[3]COMPS

[2]SUBJ

{ }[ ][1]SLASH

S

{ }[ ][1]SLASH

VP

{ }[ ]SLASH

S

{ }[ ][1]SLASH

S

On Tuesday Sandy

[3]NP

Leslie

visits


(34) SLASH Amalgamation

[ ] [ ]

( )

−∪∪

⇒

]0[[n][1]SLASH

[0]BIND

[n]SLASH,,[1]SLASHDEPSCATLOC

K

K

word

SLASH Amalgamation ensures that if a dependent is slashed, its head is also be slashed. The

verb’s SLASH value is projected upward in a syntactic tree from the head daughter to mother,

due to the SLASH Inheritance (Bouma et al. 2000: 20).

(35) SLASH Inheritance

[ ]

⇒

[1]SLASHDTR-HD

[1]SLASH-- phvalhd

This constraint is defined as a constraint on head-val-phrases, a type of phrase involving head

and complement or subject daughters. The termination of this transmission occurs at an

appropriate point higher in the tree, and a head-filler phrase is one possibility at the top of the

dependency: a dislocated constituent (specified as [LOC [1]] in (32)) combines with the head

that has the property specified in the constraint for hd-fill-ph (Bouma et al. 2000: 21).9,10

(36) { }[ ]

⇒

[3]SLASH

[1]LOCDTRS-HD-NON

[1][2]SLASHDTR-HD

]3[[2]SLASH

SUBJ

U

U

ph-filler-hd

(36) states that the SLASH value of a head-filler phrase is the SLASH value of the head

daughter minus the LOC value of the filler plus the SLASH value of the filler.

However, Bouma et al.’s (2001) approach to adverbial preposing faces some problems.

First, (26), which is problematic for Pollard and Sag’s (1994) early HPSG analysis, is

problematic for Bouma et al. (2001) as well.

(37) * I was wondering [S during the holidays [S for what kind of jobs [S you would go

9 The symbol denotes disjoint set union, which is the same as set union except that the disjoint union of two sets with a nonempty intersection is undefined (Bouma et al 2001: 21). 10 Kim and Sag (2002: 386) treats usually in the following example as adjunction to S.

(i) Usually Kim will not run.

+

+


into the office.]]] [= (26)]

There is nothing to prevent a filler from combining with an embedded wh-question. As in

the case of Pollard and Sag (1994), the [IC +] specification for the head-daughter can exclude

(37), but this gives rise to another problem.


would go into the office.]]] [= (28)]





Second, Bouma et al.’s (2001) extraction analysis of adjunct preposing entails that

adjunct preposing and argument topicalisation are basically the same, but there is an important

difference between the two. A preposed adverbial does not give rise to contradictions in the

environment of VP ellipsis (39a) (see Shaer 2003a: 247, 2003b: 458–459, 2004a: 314, 2004b:

385), but a topic does (39b).

(39) a. At five, John finally signed the form, and Mary did too – though she, in fact, did it

a bit later. (Adapted from Shaer 2004a: 314)

b. # The picture, John saw, and Mary did too – though she, in fact, saw another picture.

In (39a) the meaning of the preposed adverbial is not part of the antecedent for VP ellipsis; the

interpretation of did too is the same as the antecedent VP finally signed the form, and hence

the time when Mary signed the form does not have to be five o’clock, which does not cause

contradiction with though she, in fact, did it a bit later. In (39b) the only possible

interpretation of the second conjunct is that Mary saw the same picture as John. This cause

contradiction with though she, in fact, saw another picture. The representation of the

conjunction in (39b) will be something like the following.


(40) Representation for (39b)

In (40) the filler NP that picture is shared between the first and the second conjunct. Therefore,

the interpretation of the second conjunct is that Mary saw the same picture which John saw.

This conforms to the fact observed in (39b).

Let us consider an alternative analysis of (39b). In this representation, that picture is

the filler just for the first constituent. It is interpreted as part of the antecedent for VP ellipsis

because the LOC information about the filler is contained in the antecedent VP. Therefore,

the interpretation of the second conjunct is that Mary saw the same picture which John saw.

This again conforms to the fact observed in (39b).

[ ][1]LOC

NP

That picture

{ }[ ][1]SLASH

S

[2]NP

|

John

{ }[ ][1]SLASH

S { }[ ][1]SLASH

S

NP

|

Mary

Conj

|

and

{ }[ ] SLASH

S

{ }[ ][1]SLASH

VP

did too { }

{ }

[1]SLASH

[1]SLASH

[1]LOC[2],DEPS

COMPS

[2]SUBJ

V

{ }[ ] SLASH

S

{ }[ ][1]SLASH

VP

|

saw


(41) Alternative representation for (39b)

Now, let us consider the possibility that (39a) has a parallel structure. In (42), as is the

picture in (40), the adverbial at five is a filler and shared between the first and the second

conjunct. Therefore, the only possible interpretation of the second conjunct should be that

Mary signed the form at five. This predicts that (39a), whose though-clause states that Mary

signed the form later than five o’clock, should give rise to contradiction. This is clearly a

wrong prediction because (39a) does not contain any contradiction at all. If the preposed

adverbial at five in (39a) is a filler like the picture in (39b), this difference between these two

sentences is not expected to occur.

[ ]]1[LOC

PP

that picture

[2]NP

|

John

{ }[ ]]1[SLASH

S

{ }[ ]SLASH

S

NP

|

Mary

Conj

|

and

{ }[ ] SLASH

S

{ }[ ]SLASH

VP

did too

{ }[ ]SLASH

S

VP

{ }{ }

[1]SLASH

[1]SLASH

[1]LOC[2],DEPS

COMPS

[2]SUBJ

saw


(42) Extraction analysis of (39a)

Let us consider an alternative analysis of (39a). In this representation, at five is the filler just

for the first constituent. It is interpreted as part of the antecedent for VP ellipsis because the

LOC information about the filler is contained in the antecedent VP. Therefore, the only

possible interpretation of the second conjunct should be that Mary signed the form at five.

This is a wrong prediction because (39a) does not contain any contradiction at all.

{ }[ ][1]LOC

PP

At five

{ }[ ]]1[SLASH

S

[2]NP

|

John

{ }[ ]]1[SLASH

S { }[ ]]1[SLASH

S

{ }[ ]]1[SLASH

VP

NP

|

Mary

Conj

|

and

{ }[ ] SLASH

S

{ }[ ]]1[SLASH

VP

did too

{ }{ }

[1]SLASH

[1]SLASH

[1]LOC[3],[2],DEPS

[3]COMPS

[2]SUBJ

V [3]NP

the form

|

signed

{ }[ ] SLASH

S

{ }[ ]]1[SLASH

S

{ }[ ]]1[SLASH

VP


(43) Alternative extraction analysis of (39a)

Bouma et al.’s (2001) analysis gives a unified treatment of clause-internal fronting and

long-distance extraction of adverbials (Bouma et al. 2001: 45), the only difference being

where the SLASH inheritance terminates. The third problem for their analysis is that they

cannot capture the fact that there are important differences between the two types of adverbials.

We will observe the relevant data in section 5.3. A satisfactory analysis of adverbial

preposing should be able to capture these differences, but this is not a simple matter for

Bouma et al.’s (2001) unified analysis.

[ ]]1[LOC

PP

at five

[2]NP

|

John

{ }[ ]]1[SLASH

S

{ }[ ]SLASH

S

NP

|

Mary

Conj

|

and

{ }[ ] SLASH

S

{ }[ ]SLASH

VP

did too

{ }[ ]SLASH

S

{ }{ }

[1]SLASH

[1]SLASH

[1]LOC[3],[2],DEPS

]3[COMPS

[2]SUBJ

signed

V

{ }[ ]]1[SLASH

VP

[3]NP

the form


5.2.2.3 Levine (2003) / Levine and Hukari (2006)

Let us turn to the analysis of adjunct fronting developed by Levine (2003) / Levine and Hukari

(2006). It is assumed that adverbials in adjoined positions can extract leaving a trace behind.

(26), which is problematic for Pollard and Sag (1994) and Bouma et al. (2001), is

problematic for Levine (2003) / Levine and Hukari (2006) as well.

(44) * I was wondering [S during the holidays [S for what kind of jobs [S you would go

into the office.]]] [= (26)]

There is nothing to prevent a filler from combining with an embedded wh-question. As in

the case of Pollard and Sag (1994), the [IC +] specification for the head-daughter can exclude

(37), but this gives rise to another problem.


would go into the office.]]] [= (28)]





5.2.3 Conclusion

In this section, we have discussed how earlier analyses of adjunct fronting work, and have

pointed out problems that they are confronted with. In these analyses, the linear position of

fronted adjuncts is identified by their position in constituent structure: as an adjunct of a

constituent it modifies, a specifier of some maximal projection, or a sister of a constituent with

a trace/gap. In the next section, it will be argued that the type of fronted adjunct which we

have considered so far should be differentiated from another type. It will become clear that

the failure of the configurational approaches stem from the fact that they do not take such a


distinction into account. In particular, relatively free positioning of such adjuncts do not

conform to the characterisation provided by the configurational analyses.

5.3 Differences between long- and short-fronted adverbials

As promised at the end of 5.2.2.2, we will look at differences between short-fronted adverbials

and long-fronted adverbials. As we will see, the contrast is not between short- and

long-fronting, but between extracted adverbials and non-extracted adverbials,

5.3.1 Information structure

In this section, we will see that short-fronted adjuncts share some properties with fronted noun

phrases and long-fronted adjuncts, but the former should be differentiated from the latter two

in important respects. We will look at the shared properties first. (46) is a typical example

of focus movement of an NP.

(46) ROBIN I really dislike (Culicover 1991: 34)

In this sentence, the fronted element is stressed (i.e., A-accented; Jackendoff’s 1972; Engdahl

and Vallduví 1996; see Chapter 2) and given narrow focus. (47) is a typical example of

topicalisation of an NP.

(47) Robin, I really dislike. (Culicover 1991: 34)

In this sentence, the fronted element is B-accented (Jackendoff’s 1972; Engdahl and Vallduví

1996; see Chapter 2) and functions as a topic. In (48), the short-fronted adjuncts last year

and in those days have narrow focus, and hence they have heavy stress.

(48) a. LAST year we were living in St. Louis.

b. In THOSE days we drove a nice car. (Culicover 1991: 34)


In (49), the adjuncts on Monday and on Tuesday function as a topic or a ‘link’ (Birner and

Ward 1998) in that the fronted adjunct stands ‘in some salient and relevant relationship to

elements that have been evoked’ in the prior discourse (Ward et al. 2002: 1368).

(49) This is John’s schedule. On Monday, he plays a match in London and meets the

press. On Tuesday, he plays a match in Sheffield… (Shaer 2004a: 325)

These properties are shared by long fronted adjuncts as well. Let us consider (50).

(50) a. * Tomorrow I promised that he would be there.

b. Around midnight I promised that he would be there. (Ross 1986: 179)

With normal intonation, (50a) is ungrammatical and (50b) is unambiguous with the adverbial

around midnight construed only with the matrix clause. However, Ross (1986: 180–181)

states that ‘[i]f the preposed adverbials bear heavy stress, […,] [(50a)] becomes grammatical,

and [(50b)] becomes ambiguous, for the adverbial can modify the lower verb as well as the

main verb’. If Ross’ observation is correct, the long fronted adjuncts can carry narrow focus.

In (51), the adverbial on Monday functions as a topic in the same way as (49).

(51) We’ve decided our schedule for the next week. On Monday we have arranged

that we have dinner in a restaurant in town.

Now let us turn to the differences between short-fronted adjuncts and fronted noun

phrases and long-fronted adjuncts. Short-fronted adjuncts do not necessarily have to be a

topic or a narrow focus. First, consider (52).

(52) a. A man came in. Carefully he lit a pipe.

b. A man came in. With a hammer he broke the window. (Frey 2005: 107)

The fronted adjuncts carefully and with a hammer in (52a,b) do not stand in any relationship to

elements evoked in the prior discourse. Second, short fronted adjuncts can occur in a

sentence focus context.

(53) A: What happened?

B: Five minutes ago, my car broke down.


A sentence with sentence focus can be an answer to What happened? The fact that a

sentence with a short-fronted adjunct can be an answer to this question indicates that such an

adjunct does not have a topic nor a narrow focus interpretation.

On the other hand, the following data indicates that the preposed noun phrase cannot be

part of wide focus.


B1: John broke the computer.

B2: # The computer(,) John broke.

B1 has SVO word order: it can carry a sentence focus, as illustrated by the fact that it can be a

felicitous answer to What happened? B2, with the preposed noun phrase, cannot be a

felicitous answer to the question requiring a sentential focus domain.

Long fronted adjuncts do not occur in such contexts as (52) and (53), either.

(55) a. A man came in. With a hammer he broke the window.

b. A man came in. #With a hammer I think he broke the window.

(Frey 2005: 107)


B: # With a hammer I think he broke the window.

The data in (55) and (56) suggest that long-fronted adjuncts cannot be part of a broad focus

domain, unlike short-fronted adjuncts.

The above observation suggests that there are three types of short-fronted adjuncts, but

two types for fronted noun phrase and long-fronted adjuncts. The table in (57) summarises

the differences.


(57) Information-structural status of fronted adverbials and fronted noun phrases

Topic Narrow focus

Part of broad

focus

Short-fronted

adjuncts

Yes Yes Yes

Long-fronted

adjuncts

Yes Yes No

Noun phrases Yes Yes No

Thus, short-fronted adjuncts can be a topic, a narrow focus and a part of a broad focus domain,

but long-fronted adjuncts and noun phrases can just be a topic and a narrow focus.

5.3.2 Blocking of wh-extraction

Haegeman (2003) shows that long fronted temporal adjuncts pattern like fronted arguments

with some respects. The observations in this and the following subsection depend on her

work.

It is difficult for fronted arguments to follow a fronted wh-phrase (See also Baltin 1982;

Rizzi 1997). If the wh-phrase is a non-subject, fronted noun phrases degrade the

grammaticality of the sentence.

(58) ?? the student to whom, your book, I will give tomorrow

(Haegeman 2003a: 642, (3))

In the case of wh-subjects, the sentence is ungrammatical.

(59) a. * This is a man who liberty would never grant to us. (Rizzi 1997:307, (71b))

b. * I wonder who, this book, would buy around Christmas. (Rizzi 1997:307, (76a))

In Haegeman’s (2003) terms, fronted arguments ‘block wh-extraction’.


However, if the wh-phrase is followed by a short-fronted adverbial, the sentence is fully

grammatical.

(60) a. John Prescott is the person who in future t will be in charge of major negotiations.

(Haegeman 2003a: 641, (1c))

b. the student to whom, tomorrow, I will give your book

(Haegeman 2003a: 642, (3))

Thus, short-fronted adjuncts do not block wh-extraction.

Now let us look at long adjunct fronting. The fronted adjuncts in (61), next year, on

Tuesday and during my time as university president, cannot be construed with the lower

clause.11

(61) a. ??/* This is the linguist who next year t expects that all his students will have a job.

b. I called up my mother, who, on Tuesday, I had told it is likely that Sandy will visit

Leslie.

c. These are the investigators who, during my time as university president, I told

there were never any illegitimate appropriations of money.

(Haegeman 2003a: 643–644)

This means that long-fronted adjuncts cannot follow fronted wh-phrases: they block

wh-extraction.

The above observations are summarised in (62).

(62)

Blocking of wh-extraction

Short-fronted adjuncts No

Long-fronted adjuncts Yes

Noun phrases Yes

11 (61b) and (61c) are grammatical with the interpretation that the adverbials modify the higher clause.


Thus, again, long fronted adjuncts behave like fronted arguments, rather than short-fronted

adjuncts. It is possible to say that positioning of short-fronted adjuncts is relatively free

compared with the others in that the former can follow fronted wh-phrases while the latter

cannot.

5.3.3 Restriction to root/root-like clauses

Argument fronting is restricted to root clauses or clauses with root behaviour (Emonds 1970,

2000; Rutherford 1970; Hopper and Thompson 1973; Andersson 1975; Green 1976;

Haegeman 1984a,b, 2001, 2002a,c; Maki, Kaiser, and Ochi 1999, Heycock 2002)

(63) a. * If these exams you don’t pass, you won’t get the degree.

b. * While her book Mary was writing this time last year, her children were staying

with her mother.

c. * When her regular column she began to write for the Times, I thought she would

be OK. ((a)–(c) Haegeman 2003a: 642, (4)–(6))

The sentences in (63) have a fronted argument in a non-root clause, and they are

ungrammatical.12

However, short adjunct fronting can occur in non-root clauses as well.

(64) a. If next week you cannot get hold of me, try again later.

b. While around this time last year Mary was writing her book, her children were

staying with her mother.

c. When last month she began to write a regular column for the Times, I thought she

would be OK. (Haegeman 2003a: 642, (4)–(6))

These sentences show that short adjunct fronting is grammatical in the same environment as

12 Argument fronting is possible in adverbial clauses with root like properties.

(i) His face not many admired, while his character still fewer felt they could praise. (Quirk et al. 1985:1378)

In (i) the connective while gets an interpretation similar to and (Hornstein 1990:206), and therefore the

while-clause behaves like a root clause (Haegeman 2003a: 642).


(63).

Turning to long fronted adjuncts, (65) shows that they resist non-root environments.

(65) If this afternoon they say that it will rain, we won’t go. (Haegeman 2003a: 644)

The fronted adverb this afternoon is only construed with the higher clause, which means that

in such non-root environments as the if-clause in (65), long fronting of adjuncts is impossible.

The above observations are summarised in (66).

(66)

Restriction to root/root-like clauses

Short-fronted adjuncts No

Long-fronted adjuncts Yes

Noun phrases Yes

Thus, again, long-fronted adjuncts behave like fronted arguments, rather than short-fronted

adjuncts. Again, it is possible to say that positioning of short-fronted adjuncts is relatively

free compared with the others in that the former can follow complementisers while the latter

cannot.

5.3.4 Summary and conclusion

Our observations in 5.3.1 to 5.3.3 are summarised in (67).


(67)

Topic Narrow focus

Part of

broad focus

Blocking of

wh-extraction

Root/root-like

clauses

Short-fronted

adjuncts

Yes Yes Yes No No

Long-fronted

adjuncts

Yes Yes No Yes Yes

Noun phrases Yes Yes No Yes Yes

(67) clearly shows that long-fronted adjuncts and fronted arguments pattern alike, and

short-fronted adjuncts are separate from them. As stated in 5.2.2.2, this is a challenge to a

unified treatment of adjunct/argument extraction by Bouma et al. (2001) and Sag (2005).

The fact that long-fronted adjuncts and fronted arguments behave in parallel strongly

suggests that they are one and the same. We assume therefore that they are manifestations of

a single extraction phenomenon, which should be handled by the SLASH mechanism.

Once it is established that the long fronting of adverbials is due to the SLASH

mechanism, there should also be cases where the SLASH value is matched with a filler within

the same clause as the trace. That is, there should be a variant of short adjunct fronting with

the same effects as argument fronting and long adjunct fronting. We therefore propose to

split what has been called ‘short-fronted adjuncts’ into two types, one of which has exactly the

same properties as long-fronted adjuncts and fronted argument: information-structural status

as either a topic or a narrow focus, blocking of wh-extraction, and restriction to root/root-like

clauses. Another type, which we will call ‘incidental adjuncts’, constitute part of a broad

focus domain, does not block wh-extraction, and does not have restriction to root/root-like

clauses. The overall picture will then look like (68).


(68)

Topic Narrow focus

Part of

broad focus

Blocking of

wh-extraction

Root/root-like

clauses

Incidental

adjuncts

No No Yes No No

Extracted

phrases

Yes Yes No Yes Yes

Here, included in the category of ‘extracted phrases’ are fronted arguments, long-fronted

adjuncts, and short-fronted adjuncts with a topic or a narrow-focus interpretation.

It will have now become clear that the failure of configurational analyses discussed in

section 5.2 is due to the fact that they are not aware of the distinction between extracted

adjuncts and incidental adjuncts, and the peculiar properties of the latter.

In the next section we will provide an analysis of extracted phrases, including extracted

adjuncts. Then in section 5.5 we will give an analysis of incidental adjuncts.

5.4 Extracted phrases

In this section, we will give an analysis of fronted arguments, long-fronted adjuncts, and

short-fronted adjuncts with a topic or a narrow-focus interpretation. They are all instances of

extraction. We make the following assumptions. First, adverbials in adjoined positions can

extract (Levine 2003; Levine and Hukari 2006). In this case, fronted adjuncts are fillers

combined with a slashed clause (Pollard and Sag 1994; see also Hukari and Levine 1995).

Second, a filler with an empty REL and QUE value is given either a narrow focus or a topic

interpretation.

Sentence (69) with a topic in the initial position has a representation as in (70).

(69) Your book, you should give to Paul.


(70)

Paul togive

]'4[,should

]'3[,you

]'2[,bookyour

]'1[DOM

{}SLASH

]6[LINK|STRUC-INFO

VFORM|HEAD

Paul togive shouldyou book your PHON

fifthfourththirdfirst

fin

decl-sai-non

As discussed in Chapter 2, the INFO-STRUC feature is appropriate for the CONTEXT feature.

The LINK and FOCUS features are among those appropriate for INFO-STRUC. We assume,

following Engdahl (1999: 186–187), that each of INFO-STRUC features takes content objects

(i.e., values of the CONTENT feature) as its value. In (70) the filler NP your book has the

link interpretation, and its CONT value is identified with its LINK value. This value, tagged

[6], is inherited to the LINK value of the top S, due to the constraint in (71a) introduced in

Chapter 2.

(71) Either (a) if a daughter’s INFO-STRUC value is instantiated, then the mother

][4'],[3'],[2'DOM

{[5]}SLASHS

][4'],[3'DOM

{[5]}SLASH

[7]SUBJVP

[ ]]2[[7]NP

|

you

]'3[DOM

]}5{[SLASH

]8[COMP

]7[SUBJV

|

should

[ ]{[5]}SLASH]4[

[8]VP

give to Paul

[6]CONT

[6]LINK|STRUC-INFO[5]LOC]1[

NP|

your book


inherits this instantiation

or (b) if the FOCUS value of the rightmost domain object is instantiated, then

the FOCUS value of the mother is identified with its CONTENT value.

(71a) ensure inheriting of the LINK value to its mother node. As a result, the filler functions

as a link (i.e., topic).

Note that the filler your book occupy first position. This is due to two constraints

which have been introduced so far. One is (72), which was introduced in Chapter 4.

(72)

[ ][ ]

[ ]

∨

→

nelistfinverbsecondfirst

ph-filler-ns-hd

[A][4]LOCSUBJ

[2]HEAD

[3]LOC]'1[DOM

[2]HEADDTR-HD

]1[DTRS-NH

oo

[4] ]3[ ≠∧ (version 1)

(72) is a constraint for head-nonsubject-filler-phrase (hd-ns-filler-ph), a subtype of

head-filler-phrase, and specifies the following: in hd-ns-filler-ph, the LOC value of the

non-head daughter is not token-identical with that of the single element in the SUBJ list of the

lexical verbal head, and the domain object for the non-head daughter is of the type first or

second. Furthermore, another constraint is the following constraint on non-sai clauses

introduced in Chapter 3.

(73) [ ] [ ][ ]elist

thirdfirstsai-non

=∧

⊕⊕⊕⊕→

]1[

]1[DOM LL

(73) specifies that in non-sai clauses, there should be no element in second position. The

constraint in (72) allows the filler in (70) to be assigned to first or second position, but it ends

up in first position since second position is not available due to the constraints in (73).

We assume that long-fronted adjuncts are fillers. Therefore sentences with long

adjunct fronting are given essentially the same analysis as sentences with argument fronting

such as (70). Thus, a sentence On Saturday I think Dana will go to Spain, with narrow focus

on the initial adverbial on Saturday, will be given a representation as in (74).


(74)

Spain togo willDana]'4[,

think]'3[,

I]'2[,

Saturdayon ]'1[DOM

{}SLASH

]6[FOCUS|STRUC-INFO

VFORM|HEAD

Spain togo willDana think ISaturday on PHON


fin

decl-sai-non

The long-fronted adjunct on Saturday is a filler, whose LOC value is identified with the

SLASH value extracted out of the embedded clause. It is given a narrow focus interpretation

since the CONT value of the adjunct PP is identified with the INFO-STRUC|FOCUS value.

According to (71a), the same value is inherited to the same feature of the mother sign. The

filler is in first position in the order domain of the top node.

[6]CONT

[6]FOCUS|STRUC-INFO[5]LOC]1[

PP

On Saturday

][4'],[3'],[2'DOM

{[5]}SLASHS

][4'],[3'DOM

{[5]}SLASH

[7]SUBJVP

[ ]]2[[7]NP

|

I

]'3[DOM

]}5{[SLASH

]8[COMP

]7[SUBJV

|

think

[ ]{[5]}SLASH]4[

[8]S

Dana will go to Spain


5.5 Incidental adverbials

In this section we will provide our analysis of a type of short-fronted adjuncts which can occur

in all-sentence focus contexts.

5.5.1 Incidentality

We will now introduce the notion of ‘incidentality’ (Bonami and Godard 2003; Bonami,

Godard and Kempers-Manhe 2004).13 This is concerned with a prosodic property.

Adverbials are incidental when they have a special prosody which sets them apart from the

rest of the sentence. Preposed adverbials clearly have incidentality in this respect since they

have ‘comma intonation’. Moreover, incidentals have some flexibility with respect to

positioning. For example, Bonami and Godard (2003) state that a French sentence (75) can

contain incidentals in the positions indicated with dots.

(75) • Paul • a • envoyé • ses voeux • à ce vieil ami •.

Paul has sent his wishes to this old friend

‘Paul sent his best wishes to this old friend of his.’

(Bonami and Godard 2003: 2)

This is also characteristic of adverbials which we are concerned with. (76) shows that the

adverbial at five can occur in various positions.

(76) a. At five, John finally signed the form.

b. John finally, at five, signed the form.

c. John finally signed the form, at five. (Adapted from Shaer 2004a: 314)

Moreover, it has comma intonation wherever it occurs. It is thus reasonable to assume that

13 Bonami and Godard (2003) and Bonami, Godard and Kempers-Manhe (2004) distinguish incidentality from

‘parentheticality’. The latter term denotes the semantic/pragmatic property. Adverbials have a parenthetical

interpretation when their semantic/pragmatic contribution is not integrated into the proposition which the

sentence asserts.


these adverbials are manifestations of the same incidental adverbial.14

5.5.2 Incidental adverbials

We assume that incidental adverbials have the following description.

(77) Description of incidental adverbials

+

[3]FOCUS

[2]LINKSTRUC-INFO

[1]CONT

VPMOD

INCIDHEAD

PHON phon-incidental

, where [1] ≠ [2] and [1] ≠ [3]

The PHON value specifies that they have incidental phonology (comma intonation), and

[INCID +] specifies that they are incidentals (Bonami and Godard 2003: 10).15 Incidentals

are ordinary adjuncts in constituent structure (Bonami and Godard 2003: 11). We assume

that they are VP adverbials.16 In addition, we specify that incidental adverbials should be

neither link nor narrow focus. ‘[1] ≠ [2] and [1] ≠ [3]’ in (77) specifies that the CONTENT

value of incidental adjuncts is not token-identical with the LINK and the FOCUS value. This

information allows the occurrence of incidental adverbials as a part of a wide focus domain.

5.5.3 Flexible positioning

We assume above that incidentals are ordinary adjuncts that modify a VP. To ensure that

incidental adverbials are linearised in various positions of the sentence they modify (see (76)),

we assume, along with Bonami and Godard’s (2003: 12), that such modifiers are

domain-inserted into the domain of the VP they modify. This is constrained by the following

14 Shaer (2003a: 247, 2003b: 458–459, 2004a: 314, 2004b: 385) call such adverbials ‘orphans’ (McCawley 1982,

Espinal 1991; Haegeman 1988). 15 Phrases phonologically fully integrated into the rest of the clause have the [INCID −] specification.

16 See Levine (2003) and Levine and Hukari (2006) for discussions.


constraint on head-incidental-adjunct-phrase (hd-incid-adj-ph), which is a subtype of

hd-adj-ph.17,18

(78) [ ][ ]

+→

KK ,][1',DOM

INCID[1]DTRS-NH

incidentalph-adj-incid-hd

This constraint states that the domain object for the non-head daughter of phrases of the type

hd-incid-adj-ph is of the type incidental. The type hierarchy for domain objects is modified

in the following way.

(79) The type hierarchy for domain objects in English (version 2)

dom-obj

incidental first second third fourth fifth

The type incidental is not constrained by the Topological LP constraint introduced in Chapter

3.



Therefore, incidental adjuncts can have a rather free positioning (but see 5.5.4).

We assign the following representation to (76b).

17 Ordinary, non-incidental, adjuncts, such as those illustrated below, are handled via

head-normal-adjunct-phrase (hd-normal-adj-ph), which is another subtype of hd-adj-ph.

(i) a. Thoe probably bought flowers. (Ernst 2002: 18)

b. She laid out the plan roughly. (Ernst 2002: 42) 18 There are some constraints imposed on positioning of non-incidental adjuncts, as illustrated by the following

examples. However, we do not address these issues in the thesis.

(i) a. Theo probably cleverly bought flowers.

b. * Theo cleverly probably bought flowers. (Ernst 2002: 19)


(81) Structure for (76b)

The adverbial at five is a modifier of the VP signed the form, and the former is inserted into

the order domain of the latter. As stated above, domain objects of the type incidental are not

constrained by the LP constraint in (80). This allows incidental adjuncts rather free ordering

possibilities, as illustrated in (82).

(82) a. [DOM <[incidental

at five ], [third John], [

fourth signed], [

fifth the form]>]

b. [DOM <[third John], [

fourth signed], [

fifth the form], [

incidental at five ]>]

Thus, an approach along the lines of Bonami and Godard (2003) can give a unified treatment

of the adverbials in various positions while maintaining the assumption that they are all VP

modifiers. In the present approach, then short-fronted adjuncts are incidentals which are in

[ ][ ]

form the]'4[,

signed]'3[,

fiveat ]'2[,

John]'1[DOM

]5[CONT

[5]FOCUSSTRUC-INFO

]VFORMHEAD

form thesigned fiveat John PHON

fifthfourthincidentalthird

fin

decl-sai-non

[ ][1]

NP

|

John

[ ]][4'],[3'],[2'DOM

VP

[ ]][4'],[3'DOM

[7]VP

+

[7]MOD

INCID

PHON

]2[

PPphon-incid

|

at five [ ]][3'DOM

V

|

signed

[ ][4]

NP

|

the form


the initial position of a sentential order domain.

5.5.4 Positioning constraints

We argued in the last section that incidental adverbials can occur in various positions in a

sentence, as exemplified in (76) (See also (82)). This does not mean, however, that they are

unconstrained in their positioning. First, incidental adverbials cannot occur in the position

between the verb and its nominal object.

(83) * John signed at five the form.

As shown in (75), incidental adverbials can be in this position in French. Taking this fact

into account, we do not want to rule out (83) by any universal principle. By stipulation, we

propose the following constraint

(84) [ ] [ ]−→

INCID]1[,

MOD],1[,HEAD,DOM KK

none

nominalverb

This constraint bars incidentals from occupying the position between a verb and a noun phrase.

(84) rules out incidental adverbials between the verb and its object, as in (83). In 5.6.2.1,

we will see one more case where (84) plays a role.

Second, as in Standard German, Standard English does not allow anything in

subordinate clauses to come before complementisers or fronted wh-phrases.

(85) a. He said that tomorrow it will rain.

b. * He said tomorrow that it will rain.

This restriction is not universal. For example, the variety of English spoken in Belfast allows

wh-phrases to precede a complementiser that.


b. They didn’t know which model that we had discussed. (Henry 1995:107)

The following example, cited from Hudson (2003: 640), is from Greek (Tsimpli 1990).


(87) Mu-ipe to vivlio oti edhoso sti Maria.

to-me-he-said the book that he-gave to-the Mary

‘He said that he gave the book to Mary.’

In this example, the topicalised object to vivlio ‘the book’ which belongs to the subordinate

clause precedes the complementiser oti ‘that’. Due to the fact that the pattern in (85b) is

possible in some languages, we will not make this restriction a universal principle. We assume

the following constraint for English, which requires that subordinate clauses have restricted order

domains (cf. Subordinate Leftness Constraint in Kathol 2000: 120).

(88) [ ][ ] [ ] [ ]nesetnesetc

esubordinat

REL QUE HEAD [1]

,[1],DOM

∨∨=∧

→ K

(88) requires that the initial element in the order domain of a subordinate clause is either a


expression ([REL neset]). The order domains of the embedded clause of the examples in

(85) are represented as follows.19

(89) a. [DOM < [second

that], [incidental

tomorrow], [third it], [

fourth will], [

fifth rain]>]

b. * [DOM <[ incidental

tomorrow], [second

that], [third it], [

fourth will], [

fifth rain]>]

(89b), in which the complementiser is preceded by an adverb, is excluded because it violates

the Subordinate Leftness Constraint in (88).

Thus, we assume two constraints, (84) and (88), to constrain order domains of standard

English sentences.

19 As we discussed in Chapter 3, complementisers are in the same order domain as their complement clauses.

(i) [ ]

[ ]

→

][2]'1[DOM

]2[DOMDTRS-NH

HEAD[1]DTR-HD

o

c

ph-cp

This constraint states that the domain object of the head daughters of a cp-ph is inserted to the order domain of

the non-head daughter.


5.6 An account of the facts

In this section we will consider how our analysis outlined above accommodates the data that is

problematic for the earlier HPSG analyses of fronted adjuncts.

5.6.1 Information structure

5.6.1.1 Information structure of incidental adjuncts

As discussed in section 5.5, incidental adverbials have the following description.

(90) Description of incidental adverbials

+

[3]FOCUS

[2]LINKSTRUC-INFO

[1]CONT

VPMOD

INCIDHEAD

PHON phon-incidental

, where [1] ≠ [2] and [1] ≠ [3] [= (77)]

This definition of incidental adverbials states that incidental adverbials are neither a narrow

focus nor a topic. This constraint captures the fact that such adverbials occur in a sentence

focus context such as (52) and (53), repeated in (91) and (92), respectively.

(91) a. A man came in. Carefully he lit a pipe.

b. A man came in. With a hammer he broke the window. [= (52)]


B: Five minutes ago, my car broke down. [= (53)]

5.6.1.2 Information structure of extracted adjuncts

Now let us see how our analysis of extracted adjuncts deals with the data. As we discussed

in section 5.4, constructions with an extracted phrase has something like the following


descriptions.

(93)

Paul togive

]'4[,should

]'3[,you

]'2[,bookyour

]'1[DOM

{}SLASH

]6[LINK|STRUC-INFO

VFORM|HEAD

Paul togive shouldyou book your PHON


fin

decl-sai-non

[= (70)]

(94)

Spain togo willDana

]'4[,think

]'3[,I

]'2[,Saturdayon

]'1[DOM

{}SLASH

]6[FOCUS|STRUC-INFO

VFORM|HEAD

Spain togo willDana think ISaturday on PHON


fin

decl-sai-non

[= (74)]

In 5.3.1, we saw that fillers have either a topic or a narrow focus interpretation. This analysis

can accommodate the data in (48)–(51). We analyse the fronted adjuncts in (48)–(51) as

fillers. Those in (48) and (50), repeated here in (95) and (96), respectively, are

narrow-focused.

(95) a. LAST year we were living in St. Louis.

b. In THOSE days we drove a nice car. [= (48)]

(96) a. Tomorrow I promised that he would be there.

b. Around midnight I promised that he would be there. [= (50)]

Recall that the fronted adjuncts in (96) can modify the embedded clause if they are stressed.

The fronted adjuncts in (49) and (51), repeated in (97) and (98), are instances of topicalised

filler.

(97) This is John’s schedule. On Monday, he plays a match in London and meets the

press. On Tuesday, he plays a match in Sheffield… [= (49)]

(98) We’ve decided our schedule for the next week. On Monday we have arranged


that we have dinner in a restaurant in town. [= (51)]

In our analysis, a filler is only allowed to be a topic or a narrow focus. They cannot be part

of a broad focus domain. This accounts for the unacceptability of (55b) and (56).

(99) A man came in. #With a hammer I think he broke the window. [= (55b)]


B: # With a hammer I think he broke the window. [= (56)]

5.6.1.3 Summary

Incidental adjuncts occur in a sentential focus context whereas extracted adjuncts have a

narrow focus or a topic interpretation. Our analysis of fronted adverbials given in (90) and

(93)/(94) can accommodate such a contrast, which is problematic to Bouma et al.’s (2001)

analysis of extracted adjuncts.

5.6.2 Ordering

5.6.2.1 Incidental adjuncts

We assumed in 5.5.3 that they are inserted into the order domain of their head. Although

they are VP modifiers, relatively free ordering is allowed by sequence-union, unless they

violate constraints (84) and (88), which are repeated in (101) and (102), respectively.

(101) [ ] [ ]−→

INCID]1[,

MOD ],1[ ,HEAD,DOM KK

none

nominalverb [= (84)]

(102) [ ][ ] [ ] [ ]nesetnesetc

esubordinat

REL QUE HEAD [1]

,[1],DOM

∨∨=∧

→ K . [= (88)]

As we discussed above, the constraint (101) rules out the ungrammatical examples in (83), but


it also rules out (103), given in 5.2.1 as problematic cases for the IP adjunction and the ModP

analyses of adverbial fronting.

(103) a. * What will during the holidays you do.

b. * On no account will during the holidays I write a paper. [= (4b,c), (18a,b)]

In these examples, the incidental adverbials occur between a finite auxiliary verb and its

subject. They are correctly excluded by the constraint in (101), which prevents incidental

adjuncts from occupying the position between a finite verb and a noun phrase.

In the ungrammatical sentence (104), the incidental adjunct occupies the position before

the wh-phrase.

(104) * I was wondering during the holidays [for what kind of jobs you would go into the

office]. [= (6), (26), (37), (44)]

The DOM representation of the subordinate clause of (104) is (105).

(105) [DOM <[incidental

during the holidays], [second

for what kind of jobs], [third you], [

fourth

would], [fifth go into the office]>]

The first domain element in (105) is the incidental adjunct, violating the constraint (102) in

(i.e., (88)). The ungrammaticality of (104) is due to this constraint.

In our analysis, the initial positioning of incidental adjuncts, as in (106), is just one of

possible alternative linearisation patterns.

(106) a. During the holidays for what kind of jobs would you go into the office?

[= (4a), (17a)]

b. During the holidays on no account will I write a paper. [= (17b)]

c. John said [that during the holidays on no account would he write a paper].

[= (17c)]

In verb-second clauses, the wh- and the negative phrase in (107) occupy first position, and the

auxiliary would and will are in second position.

(107) a. For what kind of jobs during the holidays would you go into the office?


b. On no account during the holidays will I write a paper.

c. John said [that on no account during the holidays would he write a paper]. [= (8)]

These clauses have the following DOM value.

(108) [DOM <[first on no account], [

incidental during the holidays], [

second will], [

third I],

[fourth

write a paper]>]

This positioning of incidental adverbial does not violate LP constraint (84).20

The embedded interrogative clause of (109) has the DOM representation given in (110).

(109) I was wondering [for what kind of jobs during the holidays you would go into the

office]. [= (28), (38), (45)]

(110) [DOM <[second

for what kind of jobs], [incidental

during the holidays], [third you],

[fourth

would], [fifth go into the office]>]

The wh-phrase occupies second position in embedded clauses. Therefore, the finite auxiliary

is assigned fourth position. The incidental adjunct is between the wh-phrase and the subject

in third. This positioning of incidental adverbial does not violate LP constraint (84).

The facts that we observed in (60) are accounted for along the similar lines. The data is

repeated in (111).

(111) a. John Prescott is the person who in future will be in charge of major negotiations.

b. the student to whom, tomorrow, I will give your book. [= (60)]

Thus, these examples are given the following DOM representations.


incidental in future], [

fourth will], [

fifth be in charge of major

negotiations]>]

20 For the treatment of the complementiser that introducing a verb-second clause, as in (107c), see Chapter 6.

There we will propose that such a complementiser occupies a position before first, which we will call pre-first.

A more precise representation for (107c) is then something like the following.

(i) [DOM <[pre-first

that], [first on no account], [

incidental during the holidays], [

second would], [

third he], [

fourth write a

paper]>]

The positioning of other elements is exactly the same as (108), and the incidental adjunct during the holidays is

between first and second positions.


b. [DOM <[second

to whom], [incidental

tomorrow], [third I], [

fourth will], [

fifth give your

book]>]

Subject wh-phrases are assumed to occupy third position, along with normal (non-wh) subjects

(See Chapter 4 for details). Non-subject wh-phrases are in second. Incidental adverbials

follow them. Again, this positioning does not violate LP constraint (84).

The fact that short adverbial fronting is not restricted to root/root-like clauses, as

opposed to argument fronting and long adjunct fronting, can be accounted for in the same way.

(113) a. If next week you cannot get hold of me, try again later.

b. While around this time last year Mary was writing her book, her children were

staying with her mother.

c. When last month she began to write a regular column for the Times, I thought she

would be OK. [= (64)]

In our assumption, complementisers occupy second position in subordinate clauses.

Therefore, we have the following representations for the subordinate clause of these sentences.


if], [incidental

next week], [third you], [

fourth cannot], [

fifth get hold of

me]>]

b. [DOM <[second

while], [incidental

around this time last year], [third Mary], [

fourth was],

[fifth writing her book]>]

c. [DOM <[second

when], [incidental

last month], [third she], [

fourth began], [

fifth to write a

regular column for the Times]>]

The incidental adjuncts are between the complementiser and the subject NP. Again, this

positioning of incidental adverbial does not violate LP constraint (84).

The data in (115) provides supporting evidence for our analysis of incidental adjuncts.

Shaer (2004a: 314) notes that preposed adverbials in these sentences can all behave as

elements independent of the antecedent VP in the environment of VP ellipsis (see also Shaer

2003a: 247, 2003b: 458–459, 2004b: 385).



a bit later.

b. John finally, at five, signed the form, and Mary did too – though she, in fact, did it

a bit later.

c. John finally signed the form, at five, and Mary did too – though she, in fact, did it

a bit later. ((a)–(b) adapted from Shaer 2004a: 314)

In our treatment of incidental adverbials, the representation in (116) is given to the conjoined

part of not only (115a) but also (115b,c).

(116) Constituent structure for the conjoined part of the sentences in (115)

(116) shows, the incidental adverbial at five is not included in the antecedent VP. This allows

the interpretation that Mary signed the form some other time. Hence the occurrence of the

though-clauses in (115) is permitted.

{ }[ ]SLASH

[1]S

{ }[ ]SLASH

S

NP

|

Mary

Conj

|

and

{ }[ ] SLASH

S

{ }[ ]SLASH

VP

did too

{ }[ ]SLASH

S

{ }[ ] SLASH

S { }[ ]SLASH

S

{ }[ ][1]MOD

PP

NP

|

John

{ }[ ]SLASH

[1]VP

signed the form


5.6.2.2 Ordering of extracted adjuncts

Let us see how our analysis of extracted phrases given in 5.4 handles the ordering patterns.

Let us start with fronted arguments. As we have seen in (58) and (59), fronted arguments

cannot occupy the position after the fronted wh-phrase. The data is repeated here for

convenience.

(117) ?? the student to whom, your book, I will give tomorrow. [= (58)]

(118) a. * This is a man who liberty would never grant to us.

b. * I wonder who, this book, would buy around Christmas. [= (59)]

The representations of the DOM value of the embedded clause of (117) and (118a) are given

in (119).


to whom], [first your book], [

third I], [

fourth will], [

fifth give tomorrow]>]

b. [DOM <[third who], [

first liberty], [

fourth would], [

fifth never grant to us]>]

In (119a), the relative wh-phrases to whom and who are in second and third, respectively.

The fronted arguments are assumed to occupy first. In these examples, however, they follow

the wh-phrase. This ordering violates the following Topological LP Statement first




This constraint states, among other things, that elements in first should precede those in

second and those in third. The representations in (119) both violate this constraint, so (117)

and (118a,b) are ungrammatical.21

The fact that the fronted adjuncts in (61), repeated in (121), cannot be construed with the

21 One might argue that the ungrammaticality of the sentences in (117) and (118) is the result of the requirement

that head-filler phrases only allow a singleton SLASH value (Steve Harlow, p.c.). However, the

ungrammaticality of the examples in (122), where a complementiser and a filler co-occur, cannot be handled

along these lines. The linear analysis proposed here will be necessary for dealing with these cases.


lower clause can be accounted for along the same lines.

(121) a. ??/* This is the linguist who next year t expects that all his students will have a job.

b. I called up my mother, who, on Tuesday, I had told it is likely that Sandy will visit

Leslie.

c. These are the investigators who, during my time as university president, I told

there were never any illegitimate appropriations of money. [= (61a,b,c)]

In our analysis, long fronted adjuncts are fillers, and they occupy first position. The

wh-phrase in (121a) is in third, and those in (b) and (c) are in second. Thus, the permutations

in (121) are prohibited for the same reason as (117) and (118): they violate Topological LP

Statement (120).

The fact that argument fronting and long adjunct fronting is restricted to root/root-like

clauses can be accounted for in the same way. The subordinate clauses in (122) (123) are

non-root clauses. In (123), this afternoon cannot be interpreted to modify the lower clause.

(122) a. * If these exams you don’t pass, you won’t get the degree.

b. * While her book Mary was writing this time last year, her children were staying

with her mother.

c. * When her regular column she began to write for the Times, I thought she would

be OK. [= (63)]

(123) If this afternoon they say that it will rain, we won’t go. [= (65)]

In our analysis, complementisers occupy second position in subordinate clauses. Therefore,

we have the following representations for the subordinate clause of the sentences in (122) and

(123).


if], [first these exams], [

third you], [

fourth don’t], [

fifth pass]>]

b. [DOM <[second

while], [first her book], [

third Mary], [

fourth was], [

fifth writing this time

last year]>]

c. [DOM <[second

when], [first her regular column], [

third she], [

fourth began], [

fifth to


write for the Times]>]

d. [DOM <[second

if], [first

this afternoon], [third they], [

fourth say], [

fifth that it will rain]>]

In these structures, the complementisers in second is followed by a fronted argument in first,

which violates (120).

In 5.2.2.2, we noted that the contrast between incidental adjuncts and fronted arguments

seen in (125) is problematic to Bouma et al.’s (2001) unified treatment of arguments and

adjuncts. Let us see how our analysis handles this matter.


a bit later.

b. # The picture, John saw, and Mary did too – though she, in fact, saw another picture.

[= (39a,b)]

We have already noted that in (125a), represented as in (116), the incidental adverbial at five is

not included in the antecedent VP, and this allows the interpretation that Mary signed the form

some other time. We have kept the usual assumption that extracted arguments are fillers.

This means that the picture in (125b) has its trace as a complement of the verb saw. The

trace and the verb make a VP, which is the antecedent of the elliptical VP in the second

conjunct. Therefore, the only possible interpretation of the second conjunct is that Mary saw

the same picture as John. This cause contradiction with though she, in fact, saw another

picture. Thus, our analysis can accommodate the contrast between an incidental adverbial

and a fronted argument, which is problematic to Bouma et al.’s (2001) analysis.

5.6.2.3 Summary

In our analysis, incidental adverbials are inserted into the order domain of their head.

Although they are VP modifiers, relatively free ordering is allowed by sequence-union, unless

they violate constraints (84) and (88). This property of incidental adjuncts allow us to

accommodate all the data problematic to earlier analyses outlined in section 5.2. On the


other hand, long fronted adjuncts are analysed as fillers. They are therefore in position first,

and given a narrow-focus or a topic interpretation. With this characterisation of long fronted

adverbials we can accommodate the differences from incidental adverbials.

5.7 Conclusion

In this chapter, we first argued that there is a body of data which suggests that adjunct fronting

does not work as the earlier, ‘configurational’, approaches predict. The configurational

approaches face difficulties since they express linear order in terms of constituent structure

and information available therein: fronted adjuncts are identified by a position in constituent

structure (i.e., adjunct, specifier, or sister of a constituent with a trace/gap). That is, they do

not take into account the distinctions between extracted adjuncts, involving long-fronted

adjuncts and fronted arguments, and ‘incidental’ adjuncts. We showed that these two types of

adjuncts should be differentiated since they have different ordering patterns and

information-structural status.22 In our linearisation-based analysis, extracted adjuncts are

fillers which occupy first position in sentences; incidental adjuncts are not categorised into any

position class, which enables them to have a rather free positioning unless they violate the two

English-particular constraints. This characterisation of fronted adverbials can provide a

fairly straightforward account of the facts that are problematic for the configurational

approaches.

22 There is another type of initial adjuncts.

(i) a. Clearly, they saw the sign.

b. Rudely, she left. (Ernst 2002: 43)

They are sentential modifiers and we assume that they adjoin to the top S node via hd-normal-adj-ph (Footnote

17 in 5.5.3). There are variations in the interpretations of such adverbs, depending on which position of a

sentence they occur.

(ii) a. They saw the sign clearly.

b. She left rudely. (Ernst 2002: 43)

The sentences in (ii) have manner reading, and we assume that the adverbs adjoin to the verb. Thus, we handle

the variations in the interpretation seen in (i) and (ii) in configurational terms.

Chapter 6

Preposed Negative Expressions*

Contents

6.1 Introduction .....................................................................................................................155

6.2 Configurational approaches ...........................................................................................156

6.2.1 Minimalist/Principles-and-parameters approach .................................................156

6.2.1.1 Contrasting behaviour of wh- and negative expressions.....................................158

6.2.1.2 Information structure in NI .................................................................................160

6.2.1.3 Preposing of preverbal adverbials .......................................................................162

6.2.2 Possible standard HPSG approach.........................................................................162

6.2.2.1 Sentential negation ..............................................................................................163

6.3 Constructional Constraints ............................................................................................164

6.4 Embedded Verb Second .................................................................................................175

6.4.1 Embedded V2 in Germanic languages ...................................................................175

6.4.2 Linearisation-based analysis of embedded V2 ......................................................177

6.5 An account of the facts....................................................................................................181

6.5.1 Information structure of NI ....................................................................................182

6.5.2 V2 with fronted wh-phrases ....................................................................................183

6.5.3 Preposing of preverbal adverbials ..........................................................................186

6.5.4 Clause-boundness .....................................................................................................188

6.5.5 Sentential negation ...................................................................................................190

6.6 Other facts........................................................................................................................192

6.6.1 NI with the conjunction nor ....................................................................................192

6.6.2 Individual variation: unbounded extraction of negative phrases........................194

6.7 Summary and concluding remarks ...............................................................................196

* This chapter is an extended version of Maekawa (2006b).

Chapter 6: Preposed negative expressions

155

6.1 Introduction

In this chapter we will look at the negative inversion (NI) constructions. The sentences in (1)

are typical examples.

(1) a. Under no circumstances will he eat raw spaghetti.

b. No race could Lewis win.

c. With no job would Mary be happy.

The most plausible approach to NI constructions in major current syntactic theories is to analyse

the initial negative expression as a sister of the rest of the clause: negative expressions can be a

modifier of the rest of the clause, as in (1a), or they can be a sister of a constituent containing a

gap/trace, as in (1b,c). In the latter case the relationship between the negative expression and

the gap/trace is represented in terms of movement (Minimalist/Principles-and-Parameters

approaches) or the SLASH feature (HPSG), in the same way as in wh-interrogatives (2a) and

topicalisation sentences (2b) (Culicover 1991; Haegeman 2000a,b; Rizzi 1997; etc).

(2) a. What did they handed to the baby?

b. That toy, they handed to the baby.

In the linearisation framework, on the other hand, it is possible to analyse negative preposing in

terms of the linear sequence, irrespective of constituency.

I will consider the possibility of providing a detailed analysis of negative preposing in NI

constructions within these two views on linear order. I will argue that there is a body of data

which are problematic for configurational approaches, but linearisation-based HPSG can provide

a fairly straightforward account of the facts.

The organisation of this chapter is as follows. In the next section we will outline the

configurational type of approach to NI constructions, and then we will look at data that is

problematic for the configurational approach. 6.3 presents an analysis of NI constructions in

terms of linearisation. In 6.4 we will consider embedded NI constructions. In 6.5 we will see

how our analysis can deal with the problematic data for the configurational approaches. 6.5.5


156

provides an analysis of other related data. 6.7 summarise the discussion and gives concluding

remarks.

6.2 Configurational approaches

In much previous work it has been argued that the initial negative expression is in a specifier

position of a certain functional category and establishes a spec-head configuration with a verb

that moves to the head position (Culicover 1991; Haegeman 1995, 2000a,b; Haegeman and

Guéron 1999; Haegeman and Zanuttini 1991; Rizzi 1996, 1997; Rizzi and Roberts 1996; and

Roberts and Roussou 2002). In 6.2.1 we consider how this type of approach might work. We

will then provide pieces of data which are problematic for the configurational approach.

6.2.1 Minimalist/Principles-and-parameters approach

Rizzi (1997) proposes the following articulated structure for the left periphery of clause

structure.

(3) [ForceP Force0 [TopP* Top

0 [FocP Foc

0 [TopP* Top

0 [FinP Fin

0 [IP …

The traditional CP is first decomposed into two functional projections, ForceP and FinP: ForceP

encodes the illocutionary force of the clause, and FinP is a projection whose head carries the

features for (non-)finiteness.1 He also argues for the existence of other functional heads and

projections between these two: FocP and (recursive) TopP. The specifier of FocP hosts a

focalised constituent and its head hosts the focus feature. The specifier of TopP hosts the

fronted topic and its head hosts a topic feature. Within this view, wh-questions are given

something like the following analysis.

(4) [FocP which booki [Foc willj [IP you tj read ti]]]

1 What Force really deals with is sentence type, such as declarative, interrogative, and so on (Bob Borsley, p.c.).


157

The wh-phrase moves out of IP to the specifier of FocP. The movement of the auxiliary to Foc

is then triggered by the WH-criterion, which checks a feature of the wh-expression with a verb in

a spec-head configuration (see, e.g., Rizzi 1996, 1997; Haegeman 2000a,b).

The positioning of negative expressions and the accompanying subject-auxiliary inversion

in NI are seen as parallel to the positioning of wh-expressions and the accompanying

subject-auxiliary inversion in interrogatives. It is assumed that the initial negative expression is

in [Spec,FocP] of a functional head Foc (Rizzi 1997: 317; Haegeman 2000a: 126; Haegeman

2000b: 26; see also Culicover 1991: 12, 15).2 NI constructions are given something like the

following representation.

(5) [FocP Not a single paperi [Foc didj [IP he tj finish ti on time]]]

In (5) the negative expression is in [Spec,FocP], and the auxiliary verb carrying the NEG-feature

has moved to Foc0 to satisfy the Negative Criterion (Haegeman 1995, 2000a,b; Haegeman and

Zanuttini 1991; Rizzi 1996: 73–74; Rizzi 1997: 315–318).3 Thus, NI constructions are

analysed in the same way as wh-questions: the wh-phrase in (4) and the negative expression in

(5) are in [Spec,FocP] and they are in a spec-head configuration with the auxiliary in Foc.

Unlike main clauses, NI constructions do not look so much like wh-questions in

subordinate clauses. Compare the following examples.

(6) a. * I wonder what did Robin see.

b. I said that not once had Robin raised his hand.

(6) shows that wh-questions do not involve subject-auxiliary inversion in subordinate clauses

while NI constructions do. Since the subordinate questions are selected by a matrix predicate,

the highest head of the CP domain, Force, is associated with the wh-feature (Culicover 1991;

Rizzi 1997; Haegeman 2000a,b). In embedded wh-questions, therefore, the wh-element moves

to [Spec,ForceP] to establish a spec-head relation with the wh-feature. This makes the auxiliary

inversion unnecessary. On the other hand, the embedded NI clauses are not selected by a

2 Haegeman (2000a: 126) assumes that a focus feature associated with the negative expression triggers preposing.

3 The Negative Criterion is defined as follows (Haegeman 2000a: 123; Haegemann 2000b: 23):

(a) A NEG-operator must be in a Spec-Head configuration with an X-[NEG]

(b) An X-[NEG] must be in a Spect-Head configuration with a NEG operator.


158

matrix predicate, so Force is not associated with the NEG feature (Haegeman 2000a: 135). As

is the case for main clauses, the negative expression occupies [Spec,FocP], and the Negative

Criterion triggers movement of the auxiliary to Foc0. The complementiser that can cooccur

with the element in [Spec,FocP] since the former is in Force0.

To summarise, the configurational analysis outlined above gives a parallel analysis to main

wh-questions and NI sentences: the initial wh- and negative expression occupy [Spec,FocP].4

(7) a. Wh-question: [FocP which booki [Foc willj [IP you tj read ti]]]

b. NI: [FocP Not a single paperi [Foc didj [IP he tj finish ti on time]]]

In the following two subsections, we will look at a body of data which are problematic for this

analysis.

6.2.1.1 Contrasting behaviour of wh- and negative expressions

The analysis outlined above predicts that initial negative expressions in NI constructions always

behave like wh-expressions in wh-interrogatives. However, a body of data illustrates the

contrasting behaviour of wh-expressions and negative expressions.

First, wh- and negative expressions can co-occur in main clauses, as long as the former

precedes the latter.5

(8) a. What under no circumstances would John do for Mary?

4 Culicover (1991) assumes the following clause structure.

(i) [CP C0 [PolP Pol

0 [IP I

0 …

NegP moves into the specifier position of PolP.

(ii) [PolP [Spec NegP] [Pol Neg] [IP …I…]]

He assumes that Neg is a morpheme that must cliticise to another head. However, the configuration in (ii) does not

have such a head for Neg to cliticise to. This induces the raising of the head of IP to Pol.

(iii) [PolP [Spec NegP] [Pol Neg]+Ii [IP …ti…]]

The configuration in (iii) represents subject-auxiliary inversion. In embedded clauses, the complementiser is in the

C0 position.

(iv) [CP C [PolP [Spec NegP] [Pol Neg]+Ii [IP …ti…]]

Neg can still appear in Pol, and embedded negative inversion is possible. This analsysis is similar to Rizzi’s

(1997): Culocover’s C and Pol correspond to Rizzi’s Force and and Foc, respectively. Therefore, all the criticisms

to Rizzi discussed below can be applied to Culicover as well. 5 I would like to thank Bob Borsley and Neal Snape for the grammaticality judgements of these sentences.


159

b. * Under no circumstances what would John do for Mary?

c. Where under no circumstances would John go for a holiday?

d. * Under no circumstances where would John go for a holiday?

The assumption that they are in a single position [Spec,Foc] leads to the prediction that they

should not co-occur (Haegeman 2000a: 134; Haegeman 2000b: 46). This is not borne out,

however, as the examples cited above illustrate.6

Second, the unbounded extraction of wh-phrases is grammatical, but unbounded extraction

of a negative phrase is unacceptable for many speakers (Sobin 2003: 184–185). Let us consider

the following pair.

(9) a. What did you say [that Mary will eat t]?

b. Never again did I say [that Mary will eat clams]. ((b) from Sobin 2003: 184)

The sentence in (9a) has a wh-phrase, which is extracted out of the embedded clause. For Rizzi

(1997), it should be possible to give the same analysis to (9b), where the initial negative phrase is

extracted out of the embedded clause. However, if the unbounded extraction of a negative

expression were grammatical, never again should be able to modify the lower clause, and (9b)

should have the same meaning as (10).

(10) I said [that never again will Mary eat clams]. (Sobin 2003: 184)

However, (9b) does not have the same meaning as (10). In the former, never again modifies

only the matrix clause while in the latter it modifies only the embedded clause.

The contrast in (11) may illustrate the same point.

(11) a. What did Bill say that Mary remembered to bring.

b. ?? Not a penny did I say that Mary remembered to bring. (Sobin 2003: 185)

6 Haegeman (2000a,b) cites the following examples as evidence that the wh-phrase and the negative expression

compete for the same position [Spec,FocP].

(i) a. * In no way, why would Robin volunteer?

b. * Why, in no way would Robin volunteer? (Haegeman 2000a: 134)

(ii) a. * On no account where should I go?

b. * Where on no account should I go? (Haegeman 2000b: 46)

However, my informants do not find the (b) examples ungrammatical.


160

The unbounded extraction of wh-phrases, as in (11a), is grammatical, but as (11b) illustrates, the

unbounded extraction of a negative phrase is very difficult.7 If wh-interrogatives and NI

constructions have parallel analysis, the sentences in (11) should elicit similar judgements.

However, this is not the case.

These pieces of data show that there is no reason to think that negative preposing in NI

should be given a parallel analysis with wh-fronting, and that the configurational analysis of NI

outlined in section 6.2 is dubious. This suggests that an alternative analysis is needed in which

NI sentences and wh-interrogatives are treated rather differently.

6.2.1.2 Information structure in NI

We saw above that in the configurational approach, the initial negative expression in NI occupies

the specifier position of a functional head Foc. Many proponents of this approach assume that

the preposed element in the sentences of the following type occupies the same position

(Culicover 1991; Rizzi 1997; Haegeman 2000a,b).

(12) To ROBIN I gave a book. (Culicover 1991: 34)

The preposed element with focus stress (in capitals) is assumed to be in the [Spec,FocP] position.

It is important to note the fact that (12) can be used to answer the question (13a), but cannot be

used to answer (13b).

(13) a. To whom did you give a book?

b. What happened?

The question in (13a) requires an answer with constituent focus on a recipient PP, and (13b)

requires an answer with the whole-sentence focus. The fact that (12) can only answer (13a)

indicates that the initial constituent in [Spec,FocP] is the only possible scope of focus.

If the initial negative expression in NI sentences occupies the position [Spec,FocP], it is

expected to have the same scope of focus as the preposed element in (12). The following data,

cited by Culicover (1991: 34) and Haegeman (2000b: 34), might appear to give evidence to this.

7 There seems to be variation in native speakers’ judgements. We will discuss this later in this chapter.


161

(14) a. Did you see anyone?

b. No, not a single person did I see. (Culicover 1991: 34)

An answer to a yes-no question serves as a test for focushood of a constituent (e.g., Chomsky

1971; Jackendoff 1972; Rochemont 1986). The fact that an NI sentence serves as an answer for

the yes-no question (14a) indicates that the initial negative expression is focused and has new

information.

However, there is also evidence that NI sentences as a whole can convey new information

(Sobin 2003: 205ff). Let us consider the following examples from Sobin (2003: 206).

(15) a. * Because never again will I endure such a speech, I left.

b. I left because never again will I endure such a speech

c. * That rarely does Mary eat seafood will surprise everyone.

d. It will surprise everyone that rarely does Mary eat seafood.

e. * That never again would Mary eat seafood was inferred by everyone.

f. Everyone inferred that never again would Mary eat seafood.

g. * Since never does Mary eat seafood, Bill served chicken.

h. (?) Bill served chicken, since never does Mary eat seafood.

In English, an element with new information normally follows old information. Thus, if a

subordinate clause comes before a main clause, it means that the subordinate clause is associated

with old information; if a subordinate clause comes after the main clause, it means that the

subordinate clause provides new information. In ungrammatical sentences in (15a,c,e,g), an

embedded NI clause comes before a main clause. The ungrammaticality of these sentences is

due to the fact that an NI construction is not compatible with an old information position. In

the sentences in (15b,d,f,h), an NI clause is in a new information position. They are

grammatical since NI constructions convey new information.

To summarise, NI sentences are ambiguous with respect to the domain of focus: they have

either a narrow focus on the initial negative expression as in (14b), or a wide focus on the whole

sentence as in (15b,d,f,h). This fact is problematic for the configurational approach since it


162

predicts that only the constituent in [Spec,FocP] is focused; it does not predict the wide focus

pattern.

6.2.1.3 Preposing of preverbal adverbials

There is another problem for the assumption that both the initial negative expression in NI and

the preposed focus as in (12) occupy the specifier position of a functional head Foc.

The following pair might appear to show that the adverb never moves to the [Spec,Foc]

position from the preverbal position in NI constructions.

(16) a. I have never seen a ghost.

b. Never have I seen a ghost.

If the movement from the preverbal position to [Spec,Foc] were possible, nothing would prevent

other preverbal adverbs, such as merely and almost in (17), from moving to the same position, in

the form of focus movement as in (12).

(17) a. Kim merely opened the door.

b. Kim almost found the solution.

(18) shows, however, that preverbal adverbs cannot be preposed (Jackendoff 1972; Bouma et al

2001; Kim and Sag 2002).

(18) a. * Merely Kim opened the door. (Kim and Sag 2002: 386)

b. * Almost Kim found the solution. (Adapted from Bouma et al 2001: 45)

This contrasting behaviour of never and other preverbal adverbials means that the assumption

that both the initial negative expression in NI and the preposed focus move to [Spec,FocP] is

problematic.

6.2.2 Possible standard HPSG approach

There seems to be no analysis in standard HPSG on preposed negative expressions, but the most


163

plausible approach to NI constructions in this framework is to treat them as a kind of unbounded

dependency construction. With such an assumption, the negative expression is a sister of a

constituent containing a gap, and the relationship between them is represented in terms of the

SLASH feature, in the same way as in wh-interrogatives (2a) and topicalisation sentences (2b).

(19) Possible standard HPSG analysis of ‘With no job would Mary be happy.’

The standard HPSG approach to NI predicts that it shares the properties with typical UDCs such

as wh-interrogatives or topicalisation. In 6.2.1, however, we saw a body of evidence that initial

negative expressions do not behave like wh-interrogatives.

Moreover, we will see below that initial negative expressions do not behave like initial

topic expressions, either.

6.2.2.1 Sentential negation

This property will become clear if we compare NI sentences with sentences with an initial topic

with constituent negation. The latter type of sentence is exemplified by (20).

(20) Not long ago Jack attended a party,

Here the negative expression is topicalised. NI sentences always have a sentential negation

(Liberman 1975; Klima 1964: 271ff, 306ff; Haegeman 1995: 72ff, 2000a,b; Rudanko 1982), but

sentences such as (20) negation is confined to the initial topic. First, the NI sentence in (91a)

admits neither tags, while the constituent negation sentence in (91b) does not.

(21) a. Not often does Jack attend parties and neither does Jill.

b. * Not long ago Jack attend a party and neither did Jill. (Rudanko 1982: 350)

[ ]{}SLASH|INHER

S

[ ]

job nowith

|

[1]LOC

NP

[ ]

happy beMary would

{[1]}SLASH|INHER

S


164

Second, the NI sentence in (92a) takes non-negative tags, while the constituent negation sentence

(92b) takes negative tags.

(22) a. Not often does Jack attend parties, does he/*doesn’t he?

b. Not long ago Jack attended a party, didn’t he/*did he? (Rudanko 1982: 350)

Third, the initial negative expression not often in the NI sentence in (93a) license the negative

polarity item any. (93b) shows that the constituent negation does not license any.

(23) a. Not often does Jack attend any parties.

b. * Not long ago Jack attended any parties. (Rudanko 1982: 350)

Fourth, the NI sentence cannot be coordinated with tags introduced by so, while the constituent

negation sentence can.

(24) a. * Not often does Jack attend parties, and so does Bill.

b. Not long ago John bought a house, and so did Bill. (Haegeman 1995: 73)

These pieces of data show that NI sentences have a sentential negation. This indicates that

initial negative expressions do not behave like topic expressions.

6.3 Constructional Constraints

In this section we look at how our linearisation-based approach deals with NI constructions. In

Chapter 3 (3.3.5), we characterised an NI sentence as an instance of declarative verb-second

clause. This clause type inherits all the constraints from finite clause, root, sai, v2, decl and

v2-decl. In the following we will see the last three constraints which are particularly relevant to

the present discussion.

In the internal syntax dimension of finite clauses, which is related to the placement of the

finite verbal head in linear structure, the constraints are organised as in (25).


165

(25) Classification of English clauses in terms of their internal syntax

Each type of clause in English is first of all classified according to whether or not they are root or

subordinate clauses. The root clauses are further classified according to whether or not they

involve subject-auxiliary inversion. A clause of the type subject-auxiliary-inversion (sai)

satisfies constraint (26).

(26)

[ ]

+

→KK

L

,AUX[1]HEAD

,DOM

[1]HEAD|

secondsai

(26) states that in sai clauses an auxiliary verb is a head, which is in second position. The

specification that the second element is an auxiliary verb is justified by the ungrammaticality of

sentences such as the following, in which the second element is a finite non-auxiliary verb.

(27) a. * At no time went John to London.

b. * Where went John?

Each subtype of clause of the type sai is classified according to whether or not it includes a

particular kind of element in first position. A v1 clause does not include such an element, while

a v2 clause does, constrained by the following constraint.

(28) [ ]]2[]1[

,[2]LOCSUBJ

,]1[LOC

DOM

≠∧

→

Ksecondfirst

v2

internal-syntax

root subordinate

v1 v2

sai non-sai


166

(28) states that in v2 clauses there is an element in first and its LOC value is not identical to that

of a single member of the SUBJ list of the element in second.

In the clausality dimension, finite clauses are classified in terms of their semantics; i.e.,

sentence modes such as declarative, interrogative, and imperative. The constraints are

organised as in (29).

(29) Classification of English clauses in terms of their clausality

Each type of clause in English is first of all classified depending on whether it is interrogative or

declarative. The former is further classified into wh-interrogative and polar interrogative.

Clauses of the latter type, decl, are constrained by (30) (See Chapter 3 for details).

(30) [ ]austinianedeclarativ CONT→

Following Kathol (2000, 2001, 2002), each maximal clausal type inherits both from an

‘internal syntax’ type and from a ‘clausality’ type (see also Sag 1997 and Ginzburg and Sag

2000). Thus, a clause type v2-decl is a subtype of both decl and v2.

clausality

interrogative

decl wh polar


167

(31) Type hierarchy for v2-decl

In addition to all the constraints imposed on its superytpes decl and v2, a clause of the v2-decl

type satisfies the following constraint.

(32) { }

→ KK ,

QUE,DOM

firstdecl-v2 (cf. Kathol 2001: 59)

(32) specifies that a v2-decl clauses does not have a wh-expression in first position.

English has much more restriction on what kind of expressions can occur in first position

of V2 declarative clauses than other Germanic languages. Some examples are given below in

italics.

(33) a. So slowly did the workmen get on with their work that they were dismissed.

(Hawkins 1986: 169)

b. Abby can play more instruments than can her father. (Merchant 2003)

c. When Bill smokes, all the more does Susan hate him.

d. ? The more Bill smokes, the more does Susan hate him.

(c and d from Culicover and Jackendoff 2005: 515)

finite-clause

v2-decl

internal-syntax

root

v2

sai

clausality

decl


168

It is therefore reasonable to consider that each of the sentences in (33) is of a particular sentence

type, which can be characterised as a subtype of v2-decl. Let us assume the following

hierarchy for such subtypes.

(34) Subtypes of v2-decl

These subtypes are classified according to what kind of element occupies first position. A

negative-inversion clause satisfies the constraint in (35), in addition to all the constraints

imposed on its supertype v2-decl.

(35)

+

→ K,NEG

DOMfirst

inversion-negative

This constraint states that a clause of the type negative-inversion has a negative expression (i.e.,

those with the [NEG +] specification) in first position.

Let us now consider how the above constraints work for characterising NI constructions.

The structure of an NI sentence in (36) is given in (37)

(36) On no account will I write a paper.

The non-finite verb write is combined with the object NP a paper. The resulting VP combines

with the finite auxiliary verb will to form a VP, which in turn is combined with the PP modifier

on no account. The resulting VP combines with the subject I.

v2-decl

negative- inversion

so- inversion

than- inversion …


169


{ }[ ]{ }

paper a write]'4[,

I]'3[,

AUX

]V[

will][2',

NEG STORE

WH

account noon ]'1[ DOM

CONT

VFORMHEAD

+

fifththird

fin

secondfirst

npropositio

fin

verb

inversion-negative

The higher VP and the adjunct PP combine to form a head-adjunct phrase, which is subject to the

following constraint (Chapter 2).

(38) [ ][ ]

→

[1]SSDTR-HD

[1]MODDTR-NHph-adj-hd

(38) states that the MOD value of the non-head daughter of the head-adjunct phrase is

structure-shared with the SYNSEM value of the head daughter (e.g., Pollard and Sag 1994: 56;

Kim 2000: 11). However, this constraint does not specify anything about the position of the

negative PP. We assume head-negative-adjunct-phrase (hd-neg-adj-ph) as a subtype of

+

[8]MOD

NEG]1[

PP

]'2[DOM

[6]COMPS

[5]SUBJV

[ ] [ ]

paper a,writeDOM

COMPS

]5[SUBJ

]4[

[6]VP

]'4[],'2[DOM

COMPS

]5[SUBJ

[8]VP

[ ]]3[

[5]NP

]'4[],'2[],'1[DOM

COMPS

]5[SUBJVP


170

hd-adj-ph.

(39)

+→

NEGDOMDTRS-NH

firstph-adj-neg-hd

(39) states that in a phrase of the type hd-neg-adj-ph, the single non-head daughter, which is a

negative expression, is in first. We follow Borsley and Jones (2005: 195) in marking negative

words, such as no, nobody, nothing, etc., as [NEG +].8 We also assume the following: (i) if

non-n-words have n-words as their specifiers, they have the same value for NEG; (ii)

prepositions and their complement has the same value for NEG (Borsley and Jones 2005: 198);

and (iii) the value for NEG is inherited from the head to its mother. The PP on no account in

(37) becomes [NEG +] in the following way.

(40) Inheritance of the NEG value

The n-word no is [NEG +] by definition. The noun account is [NEG +] because of its specifier

no. The mother NP inherits [NEG +] from its head noun. The preposition on is [NEG +]

because it must have the same value for NEG as its complement no account. Finally, the

mother PP inherits [NEG +] from its head preposition.9 The constraint in (39) requires the VP

8 We assume that seldom, rarely and only in the following example are also [NEG +].

(i) Seldom/rarely/only on two occasions have I heard anything like that. 9 The mother VP of the PP on no account does not share the NEG value with the PP because the PP is not the head

of the mother VP.

PP

[NEG +]

NP

[NEG +]

DET

[NEG +]

|

no

N

[NEG +]

|

account

P

[NEG +]

|

on


171

modifier on no account to occupy first position.10 The negative expression in first position

satisfies the constraint in (32) on v2-decl. The auxiliary is in second due to the constraint in

(26) for sai. The other domain elements occupy the appropriate position in the order domains

according to the constraints given in earlier chapters.

Let us turn to preposing of noun phrases as in (41).

(41) No race could Lewis win.

The structure for (41) is (42).


The NP no race is [NEG +] according to our assumption that if non-n-words have n-words as

10 It is also possible to adjoin negative expressions as an ordinary adjunct. In (i), the negative expression on no

account adjoins the lower VP to form an ordinary head-adjunct phrase.

(i) I [VP will [VP on no account [VP write a paper]]].

[ ] [ ]

+

race,noDOM

NEG]1[

[7]NP

]'4[DOM

]7[COMPS

]5[SUBJV

]'4[],'1[DOM

COMPS

]5[SUBJ

[6]VP

]'2[DOM

[6]COMPS

[5]SUBJV

]'4[],'2[],'1[DOM

COMPS

]4[SUBJVP

[ ]]3[

[4]NP

win]'4[,

Lewis]'3[,

could]'2[,

race no]'1[DOM

COMPS

SUBJ

S



172

their specifiers, they have the same value for NEG, and the value for NEG is inherited from the

head to its mother. It combines with the verb win to form a VP, which in turn combines with

could. However, we would face two problems if we kept the current formulation of constraints

on head-complement phrases.

(43)

→

Ko

KK

[2]DOM

[1],,DTRS-NHph-comp-hd (version 1)


[ ]fifth

nelist

sign

:[2]

DOM:]1[

∧

∨

:[2]

DOM:]1[

∧

sign

(43) states that in head-complement phrases the domain objects for non-head daughters are of the

type fifth; the second disjunct in the attached clause allows object traces not to have domain

objects. The first problem is that the constraint in (43) on hd-comp-ph requires that the NP

complement no race should incorrectly be assigned to fifth although it should eventually be in

clause-initial position. The second problem is that the lower VP composed of win and no race

would be compacted to form a single element when it combines with could, so it would be

impossible to liberate no race from win. Thus, we need to revise our assumption on

head-complement phrases. We assume the following type hierarchy for head-complement

phrases.

(44) Subtypes of hd-comp-ph

The supertype, hd-comp-ph, is subject to the phrasal constraint which has already been


hd-neg-comp-ph hd-normal-comp-ph

hd-comp-ph


173

(45) [ ] [ ]

→[n],[1],COMPS

DTR-HD

[n]SS,,[1]SSDTRS-NH

K

K

wordph-comp-hd

(45) states that a head-complement phrase has a head daughter which is a word and non-head

daughters whose SYNSEM values are identical to the synsem objects that form the COMPS list

of the head. One of the subtypes, hd-neg-com-ph, is constrained by the following constraint.

(46) [ ][ ]

+→

KK

KK

,NEG]1[,DTRS-NH

,][1',DOM firstph-comp-neg-hd

(46) states that phrases of the type hd-neg-comp-ph have a negative expression as a non-head

daughter, whose domain object is assigned to first. This solves the first problem: the

combination of win and no race is governed by this constraint, and the non-head daughter no

race is assigned to first.

To solve the second problem, we will employ the partial-compaction (p-compaction)

operation introduced in Chapter 2. The operation ‘p-compaction’ is defined as in (47) (Kathol

and Pollard 1995a).

(47)

( )

( )( )]7[],5[join

[6][3],[5],shuffle

[7]PHON

[4]SYNSEM[2]

[6]DOM

[4]SYNSEM]1[

]3[],2[],1[compaction-p

PHON

obj-dom

sign

∧

∧

∧

≡

The p-compaction relation holds of a sign [1], domain object [2] and a list of domain object [3]

only if [2] is the compaction of [1] with [3] liberated from the order domain of [1]. The

joinPHON operation takes a list of domain object and concatenates each PHON value. Kathol

and Pollard (1995a) define shuffle as follows: ‘The shuffle relation holds of n lists L1, …, Ln-1, Ln,

iff Ln consists of the elements of the first n-1 lists interleaved in such a way that the relative order

among the original members of L1 through Ln-1, respectively, is preserved in Ln.’. Liberation of

negative expressions as in (42) can be obtained by invoking this relation when the VP

combines with the verb. This can be expressed in the constraint on hd-normal-comp-ph as in


174

(48).

(48) [ ]

→

[1]DTRS-NH

[4]DOMDTR-HD

[5]DOM

ph-comp-normal-hd

where: ( )[3][2],[1],compaction-p

∧ ( )]5[],4[,[3],[2]shuffle

[ ]+∧ NEG:]3[

¬∨∧

SUBJ

HEAD SYNSEM:]2[:]3[

verb

[ ]fifth:]2[∧

∨

:]3[

:]2[

DOM:]1[

∧

∧

sign

The effect of the second clause of the first disjunct is to ensure that negative expressions are

liberated from the rest of the VP. The third clause of the first disjunct prevents extraction of

negative expressions out of clauses. The fourth clause of the first disjunct requires the domain

element [2] to be of the type fifth.11 The second disjunct in the attached clause allows object

traces not to have domain objects.

We can now return to (42). If we assume that the combination of could and the lower VP

is governed by the constraint in (48), the domain object for no race is liberated from the rest of

the VP, and the latter is assigned to fifth. As discussed above, the domain object for no race has

already been assigned to first, due to the constraint for hd-neg-com-ph in (46).

Preposition stranding, exemplified by (49), can be accounted for along the same lines.

(49) None of the students did John talk to.

Here none of the students is liberated from its prepositional head to. Constituent structure of

this sentence is the following.

(50) [S John [VP did [VP talk [PP to [NP none of the students]]]]]

11 The constraint in (48) will be revised for some speakers.


175

Due to the constraint in (46), the complement NP none of the students is assigned to first when it

combines with the prepositional head to. According to the constraint in (48), the resulting PP is

partially-compacted when it combines with the verb talk to form a VP so that none of the

students is liberated from to. When the VP combines with the auxiliary did to form the higher

VP, (48) comes to play again: the lower VP is partially-compacted to liberate none of the students,

which, as discussed above, has already been assigned to first; the rest of the lower VP (i.e., talk

to) is assigned to fifth.

6.4 Embedded Verb Second

It is possible to embed a negative inversion construction.

(51) John swore that under no circumstances would he accept their offer.

It will be possible to say that (51) is a case of embedded V2, which can be seen more broadly in

other Germanic languages. In this subsection we consider how the linearisation-based approach

accommodates such a case. Before that, let us consider the general properties of embedded V2

clauses in Germanic languages.

6.4.1 Embedded V2 in Germanic languages

German allows embedded verb-second complement clauses with some group of verbs.

(52) a. Sie sagte, sie wolle keine Bücher kaufen

she says she wants no books buy

b. * Johan bezweifelt, morgan wird Maria früh aufstehen

John doubts tomorrow will Mary get up early

c. * Johan bedauert, dieses Buch habe ich gelesen

John regrets this book have I read

((a) from Heycock 2005, (b)–(c) from Vikner 1995: 72)


176

Embedded V2 clauses in German are different from those in Mainland Scandinavian in that the

former should not have an overt complementiser. Compare (52a) with (53)

(53) * Karl sagte, daß dieses Bucher sollte ich gelesen haben

Karl says that this book should I read have

(Holmberg and Platzack 1995: 79)

The embedded V2 clause in (52a) has no complementiser, but that in (53) has a complementiser

daß, which makes the latter sentence ungrammatical. A complementiser is allowed only when

the embedded clause is not V2.

(54) a. Sie sagte, daß sie keine Bücher kaufen wolle

she says that she no books buy wants

b. * Sie sagte, daß sie wolle keine Bücher kaufen

she says that she wants no books buy


Mainland Scandinavian languages also allow embedded verb-second clauses to occur as a

sentential complement of some group of verbs.

(55) a. Jan sa att den här boken borde jag ha last

Jan said that this here book should I have read

b. * Jan beklagar att den här boken hade jag last

Jan regrets that this here book had I read


(55) shows that the complement clause of bridge verbs (e.g., sa ‘said’ in (55a)) can have a

verb-second order, but that of non-bridges cannot (e.g., beklagar ‘regrets’ in (55a)). Embedded

V2 is grammatical only if the complementiser is present, as can be seen from the following

Danish examples.

(56) a. Hun sagde at vi skulle ikke købe denne bog

she said that we should not buy this book


177

b. ?? Hun sagde vi skulle ikke købe denne bog

she said we should not buy this book

Thus, embedded V2 is possible in German and Mainland Scandinavian languages, but as a

sentential complement of some group of verbs.

6.4.2 Linearisation-based analysis of embedded V2

Kathol (1995, 2000: 152–154) gives an account of embedded V2 in terms of order domains.

His analysis is based on German, but he applies it to embedded V2 in Mainland Scandinavian in

the later chapter (Kathol 2000: 272–278). As discussed above, embedded V2 (i) has

verb-second order and (ii) has neither initial wh-phrase nor complementiser in German. The

first requirement can be met by requiring the clauses to be of the type root-decl: declarative

clauses with the finite verb in second position. The second condition can be met by requiring

that the clauses should be [MARKING unmarked]. The MARKING feature is appropriate for

the type category, which in turn is a value for the CAT(EGORY) feature. The MARKING

value is constrained by the following constraint.

(57) Marking Constraint for German (Kathol 2000: 118)

A clause is marked as subordinate ([MARKING marked]) if and only if the cf element in

its order domain is either

・・・・ a complementiser or

・・・・ a wh/d-phrase.

Thus, a clause with the [MARKING unmarked] specification should have neither initial

wh-phrase nor complementiser in cf position.12 Kathol (2000) formulates these requirements

via the following special schema: Head-V2-Complement Schema.

12 The correspondence of Kathol’s (2000) terminology for position classes to that adopted in this thesis is as

follows: vf = first, cf = second, mf = third, vc = fourth, nf = fifth.


178

(58) Head-V2-Complement Schema (Kathol 2000: 153)

This schema combines a head that requires a finite unmarked clause as its complement with such

a complement, the latter of which is further constrained to be of the type root-decl and to occur

in nf position.13 Thus, the governing verb glaubt and the V2 complement clause die Erde sei

flach of the German sentence in (59a) are combined via this schema, and the order domain of the

top S node is represented as in (59b).

(59) a. Otto glaubt die Erde sei flach.

Otto believes the earth is flat ‘Otto believes the earth is flat.’

b. [DOM <[vf <Otto>]],[

cf <glaubt>],[

nf <die Erde sei flach>]>]

Kathol (2000: 272ff) applies this analysis to embedded V2 clauses in Mainland

Scandinavian languages as well. The following examples are from Swedish.

(60) a. Han sa [ att Lisa hade trolingen rest till Rom ].

he said that Lisa had probably gone to Rome

‘He said that Lisa had probably gone to Rome.’

b. Han sa [ att Lisa trolingen hade rest till Rom ].

he said that Lisa probably had gone to Rome

‘He said that Lisa had probably gone to Rome.’

13 Müller (2006) criticises this treatment of embedded V2 clauses. The Head-V2-Complement Schema (58) not

just combines a head with the complement clause which it selects through the SUBCAT list, but also restricts the

complement clause to the type root-decl. This clausal type is a subtype of declarative and v2, and it inherits the

constraints for v2 and declarative. Both of them refer to the values of the DOM feature. Note that this feature is

appropriate for sign and hence outside of the feature structures contained under SYNSEM. This means that the

clause type of complements cannot be properly selected by heads through the SUBCAT list. As Müller (2006: 8)

argues, however, it would be more desirable if we could have an analysis with a single schema, rather than

stipulating more than one head-complement (or head-argument) schemata. A solution to this problem has to be left

for future research.

nf

sign

][3'[1]DOM

SUBCAT|

o

LL

]1[DOM

] ,S[]2[,SUBCAT| unmarkedfin

sign

LL

[2]SYNSEM]3[

decl-root


179

The order domain of (60a) and (60b) are represented in (61a) and (61b), respectively.

(61) a. [DOM <[rc att],[

vf Lisa],[

cf hade],[

of trolingen rest till Rom]>]

b. [DOM <[cf att],[

mf Lisa], [

mf trolingen],[

vc hade],[

of rest till Rom]>]

In the non-V2 embedded clause in (61b), the complementiser occupies cf, and the finite verb is

in vc. In (61a), however, the complementiser belongs to the additional position class, labelled

rc, which precedes vf. As cf position is not occupied by the complementiser, the finite verb can

come to the position. The MARKING value is constrained by the Marking Constraint for

Mainland Scandinavian.

(62) Marking Constraint for Main Scandinavian (Kathol 2000: 266)

a. A clause is strongly marked as subordinate ([MARKING s-marked]) if and only if

the cf element in its order domain is either

・・・・ a complementiser, or

・・・・ a wh/d-phrase.

b. A clause is unmarked marked as subordinate ([MARKING unmarked]) if the cf

element in its order domain is either

・・・・ the clausal head, or

・・・・ an adverbial.

In (61a), an element in cf is the finite verb hade, which is the head of the clause. Therefore,

(62b) requires that the embedded clause in (61a) bear the specification [MARKING unmarked].

Let us now turn to English cases. Embedded V2 in Present-day English is similar to

Mainland Scandinavian and German.

(63) a. She has often said that under no circumstances would she vote for Quayle.

(Vikner 1995: 84)

b. John swore that under no circumstances would he accept their offer.

(Authier 1992: 330)

c. I found out that never before had he had to borrow money

(Hopper and Thompson 1973: 480)


180

(64) a. * John doubts that under no circumstances will Mary get up early.

b. * John is sorry that under no circumstances will I read this book.

((a)–(b) from Vikner 1995: 85)

c. * John regretted that never had he seen Gone with the Wind.

(Authier 1992: 334)

d. * Macbeth whispered that not a word did Banquo speak at the banquet.

(Fischer et al 2000: 113)

The group of verbs in (63) can take a V2 complement. The group of verbs in (64), however,

cannot. As in Mainland Scandinavian and German, only a subgroup of verbs taking a sentential

complement can have a V2 complement clause. Like Mainland Scandinavian but unlike

German, the complementiser is obligatory if a V2 complement clause appears.

(65) She believed *(that) never would he let her down. (Heycock 2005)

It seems, then, that Present-day English patterns like Mainland Scandinavian languages.

Therefore, it is reasonable to give the same analysis to embedded V2 in both Present-day English

and Mainland Scandinavian languages.

(66b) is the order domain representation of the V2 complement in (66a).

(66) a. John swore that under no circumstances would he accept their offer.

b. [DOM <[pre-first

that], [first

under no circumstances], [second

would], [third

he], [fifth

accept

their offer]>]

As in the Scandinavian case in (61a), the complementiser belongs to the additional position class,

which we now call pre-first. The type hierarchy for domain objects is modified in the

following way.

(67) The type hierarchy for domain objects in English (final version)

dom-obj

incidental pre-first first second third fourth fifth


181

The domain objects of the type pre-first precede those of first. The Topological Linear

Precedence Constraint for English introduced in Chapter 3 is modified in the following way.

(68) The Topological Linear Precedence Constraint for English (final version)

fifthfourththirdsecondfirstfirst-pre ppppp

There is a preposed negative phrase in first position. As second position is not occupied by the

complementiser, the finite verb would can come to the position. The MARKING value is

constrained by the same constraint as for Mainland Scandinavian: (62b). As the element in

second is the head of the clause, this constraint requires that the embedded clause in (61a) bear

the specification [MARKING unmarked].

The lexical descriptions for complementisers that and whether, introduced in Chapter 3,

are modified in the following way.

(69) a.

[ ][ ]

∨

second first-pre

austinian

c

DOM

[2]CONTSST-ARG

[2]CONT

HEADSYNSEM

thatPHON

(final version)

b.

[ ][ ]

∨

second first-pre

npropositio

questionc

DOM

[2]CONTSST-ARG

[2]PROPCONT

HEAD

SYNSEM

whetherPHON

(final version)

The position class of the complementisers is now underspecified as pre-first or second.

6.5 An account of the facts

We will now look at how the constraints introduced above can accommodate the properties of

the NI construction outlined in 6.2, which are problematic for the configurational analysis.


182

6.5.1 Information structure of NI

Let us start with the ambiguity of NI sentences discussed in 6.2.1.2: they may have a narrow

focus on the initial negative expression as in (14b), or they may have a wide focus on the whole

sentence as in (15b). We argued that the configurational approach can capture only the narrow

focus pattern.

(14) a. Did you see anyone?

b. No, not a single person did I see. (Culicover 1991: 34)

(15) a. * Because never again will I endure such a speech, I left.

b. I left because never again will I endure such a speech (Sobin 2003: 206)

In the present approach this ambiguity can be accommodated quite easily. We propose the

following as an additional constraint on the negative-inversion type.

(70)

∨

→

K,[1]CONT|SS

DOM

[1]FOC|STR-INFO

[1]CONT|SS

[1]FOC|STR-INFO

first

inversion-negative

(70) states that the FOC value of negative-inversion is structure-shared with either the CONT

value of the sign or the CONT value of the domain element in first position. As discussed in

Chapters 2 and 5, a sign has information structure, which is represented as a value of its

INFO(RMATION)-STR(UCTURE). Its feature geometry reflects a focus-background structure

of a sign. Each of those features takes content objects as its value (Engdahl 1999: 186–187).

The value of the FOC feature is structure-shared with the CONTENT value of the focused part

of the sign. In the first disjunct of constraint (70), the CONT value of the clause is

structure-shared with the FOC value of the clause, and this captures the wide focus pattern in

(15b). In the second disjunct, the CONT value of the first domain element is structure-shared

with the FOC value of the clause. This captures the situation where the first domain element


183

functions the focus of the sentence, and accommodates the narrow focus pattern in (14b).14

6.5.2 V2 with fronted wh-phrases

Let us turn to the set of data discussed in section 6.2.1.1. The data in (8), which is repeated in

(71), shows that the fronted wh-element and negative expression do not show a complementary

distribution.

(71) a. What under no circumstances would John do for Mary?

b. * Under no circumstances what would John do for Mary?

c. Where under no circumstances would John go for a holiday?

d. * Under no circumstances where would John go for a holiday? [= (8)]

This fact would be surprising if the fronted wh-element and negative expression occupied one

and the same position, as the configurational analysis assumes. It would also be different from

languages such as German, where only one element is allowed in the initial position in V2

clauses. The examples in (72), cited from Hawkins (1985: 166), show the German cases; (72a)

is ungrammatical since it contains two elements before the finite verb.

(72) a. * Gestern das Auto verkaufte Fritz an einen Händler.

yesterday the car sold Fritz to a dealer

b. Gestern verkaufte Fritz das Auto an einen Händler.

yesterday sold Fritz the car to a dealer

c. Das Auto verkaufte Fritz gestern an einen Händler.

the car sold Fritz yesterday to a dealer.

Here we must recall the differences between German and English discussed in 6.4.1: the

complementiser cannot appear in embedded V2 in German while the complementiser is

obligatory in English and Mainland Scandinavian counterpart. Kathol (1995, 2000) gives the

representations for order domains of V2 clauses of these three languages.

14 Which disjunct of the constraint (70) is appropriate for the interpretation of a particular negative inversion

sentence depends on which context the sentence occurs. We do not address this issue in this thesis.


184

(73) a. German embedded V2

[DOM <[vf sie], [

cf wolle],[

mf keine Bücher],[

vc kaufen]>]

b. Swedish embedded V2

[DOM <[rc att],[

vf Lisa],[

cf hade],[

of trolingen rest till Rom]>] [= (61a)]

c. English embedded V2

[DOM <[pre-first

that],[first

under no circumstances],[second

would],[third

he],[fifth

accept

their offer]>] [= (66b)]

Let us assume that this extra position, which Kathol (1995, 2000) call rc, is available in matrix

V2 clauses as well at least in English. We propose that in cases like (71a,c) the wh-element

occupies this position, called pre-first. Thus, the order domain of the top S node of (71a) is as

follows.

(74)

Maryfor do,

John,

would,

cescircumstan

nounder ,what

DOM


first-pre

The position assignment of the rest of the clause is exactly the same as normal cases of V2

clauses: the negative expression is in first, the finite auxiliary is in second, and so on.

To accommodate constructions such as (71a,c), we need to revise the constraint on v2-wh

provided in Chapter 3.

(75)

∨→ KK ,

QUE,DOM

neset

first first-prewh-v2 (final version)

(75) states that in v2-wh clauses there is a wh-expression in pre-first or first. This is one of

subtypes of the type v2, which is constrained by constraint (28), repeated in (76).

(76) [ ]]2[]1[

,[2]LOCSUBJ

,]1[LOC

DOM

≠∧

→

Ksecondfirst

v2 [= (28)]

(76) states that in v2 clauses there is an element in first and its LOC value is not token-identical

to that of a single member of the SUBJ list of the element in second. The constraint on


185

head-nonsubject-filler-phrase (hd-ns-filler-ph), first introduced in Chapter 4, is also modified to

include pre-first position as an option for the position assignment of non-subject fillers.

(77)

[ ][ ]

[ ]

∨∨

→

nelistfinverb second first first-pre

ph-filler-ns-hd

[A][4]LOCSUBJ

[2]HEAD

[3]LOC]'1[DOM

[2]HEADDTR-HD

]1[DTRS-NH

oo

[4] ]3[ ≠∧ (final version)

(77) states that in hd-ns-filler-ph, the LOC value of the non-head daughter is not token-identical

with that of the single element in the SUBJ list of the lexical verbal head, and the domain object

for the non-head daughter is of the type pre-first, first or second.

Now let us get back to (74). It is clear that (74) satisfies both (75) and (76): as a v2-wh

clause it has a wh-expression in position pre-first, and as a v2 clause, it has a negative expression

in first position.

The same phenomenon occurs in relative clauses as well.15

(78) something which, under no circumstances, would John do for Mary

The constraint for ns-sub, which was originally introduced in Chapter 3, is revised in the

following way.

(79) [ ] [ ][2] [1]

,]2[LOCSUBJ

,, HEAD|CAT[1]LOC

DOM

≠∧

¬

∨

→KK

fourth

verb

second first-pre

sub-ns

(final version)

This constraint states that clauses of the type ns-sub should have an initial domain object of the

type pre-first or second, and this domain object is not a verb and does not correspond to the

subject of the finite verb in fourth. The order domain of (78) is represented as in (80).

(80) [DOM <[pre-first

which], [first

under no circumstances], [second

would], [third

John],

[fifth

do for Mary]>]

15 This example was provided by Steve Harlow.


186

Here, second position has already been occupied by the finite auxiliary would, and the domain

object of the relative pronoun is resolved to pre-first.

Finally, let us consider (81).

(81) What, during the holidays, under no circumstances would John do for Mary?

Here the incidental adjunct during the holidays is between the wh-phrase in pre-first position and

the negative phrase in first position. In Chapter 5, we have argued that the ordering of

incidental adverbials is rather free because they are not constrained by the Topological Linear

Precedence Constraint (68).16

6.5.3 Preposing of preverbal adverbials

As we saw in 6.2.1.3, preverbal adverbials normally cannot be preposed, as illustrated by (82),

but a preverbal adverbial never can be in the initial position of an NI sentence, as shown in (83).

(82) a. (*Merely) Kim (merely) opened the door.

b. (*Almost) Kim (almost) found the solution.

c. (*Never) I have (never) seen a ghost.

(83) Never have I seen a ghost.

As we have already discussed, this fact cannot be handled by Rizzi’s (1997) approach. Let us

consider how our analysis might deal with it.

We should note that our analysis of NI sentences does not include any specification about

the status of the negative expressions in constituent structure. This means that initial negative

expressions do not necessarily have to be even fillers. Suppose that positional assignment of

preverbal adverbs is lexically specified to be fourth, but that of negative adverbials such as never

16 The following example might appear to be incompatible with the analysis proposed (Steve Harlow).

(i) Who, under no circumstances, during the holidays, would leave the children alone?

The wh-subject who is in third (Chapter 4), but it is followed by a negative expression which should be in first.

Note, however, that there is no subject-auxiliary inversion here. We assume that this is because the negative

expression is not in first, and that it is an incidental adjunct. Note that the constraint for hd-incid-adj-ph introduced

in Chapter 5 does not exclude the possibility that the non-head daughter is a negative expression.

(ii) [ ][ ]

+→

KK ,][1',DOM

INCID[1]DTRS-NH



187

is lexically indeterminate between first and fourth.

(84) a.

[ ]

fourthDOM

VPMOD

almostPHON

b.

[ ]

∨ fourthfirst DOM

VPMOD

neverPHON

The structure for the sentence in (83) is something like (85). The adverb never combines with

the VP seen a ghost as ordinary preverbal adverbials do, and the resulting VP combines with

have to form a highest VP have never seen a ghost. In (85), the domain object for never is

liberated when the VP tagged [11] is partially-compacted. Once liberated, the adverb can

satisfy the constraints on v2-decl (32), which states that a negative expression can occur in first

position, since as shown in (84b), never can be placed in first. The adverb never is ‘in situ’ in

constituent structure as a sister of a VP it modifies, but it is in the sentence-initial position in the

order domain of the top S node.


188


( )][1'],[4'[7],compaction-p∧

( )]10[,]'2[,][1',][4'shuffle∧

6.5.4 Clause-boundness

In this subsection we will give an account to the fact that negative preposing in NI is

clause-bound (6.2.1.1). The relevant data is repeated here.

(86) a. What did Bill say that Mary remembered to bring.

b. ?? Not a penny did I say that Mary remembered to bring. [= (11)]

(87) a. I said [that never again will Mary eat clams]. [= (10)]

b. Never again did I say [that Mary will eat clams]. [= (9b)]

In our analysis, this phenomenon is captured by the constraint in (48), which is repeated in (88).

]'1[DOM

[5]MODAdvP

]'2[DOM

[11]COMPS

[6]SUBJV

[ ] [ ]

ghost a]'9[,seen]'8[DOM

COMPS

]6[SUBJ

[5]VP

]4[

]'9[],'8[],'1[DOM

COMPS

]6[SUBJ

[11]VP

]7[

]'4[],'2[],'1[]10[DOM

COMPS

]6[SUBJVP

[ ]]3[

[6]NP

ghost aseen ]'4[,

I]'3[,

have]'2[,

never]'1[DOM

S



189

(88) [ ]

→

[1]DTRS-NH

[4]DOMDTR-HD

[5]DOM

ph-comp-normal-hd


∧ ( )]5[],4[,[3],[2]shuffle

[ ]+∧ NEG:]3[

¬∨∧

SUBJ

HEAD SYNSEM:]2[:]3[

verb

[ ]fifth:]2[∧

∨

:]3[

:]2[

DOM:]1[

∧

∧

sign

[= (48)]

The third clause of the first disjunct prevents extraction of negative expressions out of clauses.

If never again is liberated from the rest of the subordinate clause at the point where the latter

combines with the matrix verb said, the resulting structure is something like the following.

(89) *

+SUBJ

HEAD

clamseat Mary willthat ,,

NEG

againnever DOMverb

fifthfirst

K

The element in fifth has the same structure as the second disjunct of the fourth attached clause in

(88). Therefore, never again is prohibited to be liberated out of the embedded clause.

Instead, the NI clause plus the complementiser that is compacted at the point where it

combines with the matrix verb said.


190

(90) Total compaction of an embedded clause

The resulting structure in (90) displays the matrix VP for (87a).

In 6.6.2 we will discuss the fact that unbounded extraction of negative expressions is

acceptable for some native English speakers.

6.5.5 Sentential negation

Our approach can accommodate the well-known fact that NI sentences always have a sentential

negation (6.2.2.1). First, the NI sentence in (91a) admits neither tags, while the constituent

negation sentence in (91b) does not.

(91) a. Not often does Jack attend parties and neither does Jill.

b. * Not long ago Jack attended a party and neither did Jill. (Rudanko 1982: 350)

Second, the NI sentence in (92a) takes non-negative tags, while the constituent negation sentence

(92b) takes negative tags.

(92) a. Not often does Jack attend parties, does he/*doesn’t he?

b. Not long ago Jack attended a party, didn’t he/*did he? (Rudanko 1982: 350)

Third, the initial negative expression not often in the NI sentence in (93a) licenses the negative

polarity item any.17 (93b) shows that constituent negation does not license any.

(93) a. Not often does Jack attend any parties.

b. * Not long ago Jack attended any parties. (Rudanko 1982: 350)

17 For an HPSG treatment of licensing of negative polarity items, see Tonhauser (2001).

][1'DOM

[3]COMPS

clamseat

,Mary

,will

,againnever

,that

DOM

[3]S

]2[ fifththirdsecondfirstfirst-pre

clamseat Mary again willnever that ]'2[,

said]'1[DOM

VP

fifthfourth


191

Fourth, the NI sentence cannot be coordinated with tags introduced by so, while the constituent

negation sentence can.

(94) a. * Not often does Jack attend parties, and so does Bill.

b. Not long ago John bought a house, and so did Bill. (Haegeman 1995: 73)

These pieces of data show that NI sentences have a sentential negation.

We can formalise the requirement of sentential negation in NI as an additional constraint to

the type negative-inversion.

(95)

[ ]

→

{}STORE

,}{[1]STORE

,DOM

[1]QUANTS

KKquant-neg

firstinversion-negative

Following de Swart and Sag (2002) and Borsley and Jones (2004) we assume that negative

expressions have a negative quantifier in storage. It is assumed that the quantifiers in storage

are retrieved from storage at a clausal node which determines their scope. (95) states that in a

negative-inversion clause, the negative quantifier in storage in the initial negative expression

should be structure-shared with one of the elements in the QUANTS list at the immediately

containing clause.18 Let us see how this constraint works. We assume that the adverbial

phrase not often has a negative quantifier as the CONTENT value, which is in storage (i.e.,

identical to the STORE value). This quantifier is retrieved at the S level and incorporated into

the value of QUANTS, following the constraint in (95). Thus, the whole sentence constitutes

the scope of never.

18 This constraint is compatible with quantifier retrieval either at lexical or phrasal level.


192

(96)

6.6 Other facts

In this section we will see that the present proposal is compatible with other important properties

of NI sentences.

6.6.1 NI with the conjunction nor

As illustrated by (97), the initial negative expression in NI constructions can be the conjunction

nor.

(97) a. Mary neither spends her vacations at the seashore nor does she go to the mountains.

(Culicover 1999: 55)

b. He did not receive any assistance from the authorities nor did he believe their

assurance that action would soon be taken.

c. The house could hardly be called red, nor was brown the right word.

d. The little creature cried and laid down, nor could all our breathing raise it.

[ ] [ ][ ]pariesany attend]'7[,does]'5[DOM

[2]VP

[ ]JackPHON]6[

[3]NP

]'7[],'5[],'4[DOM

]}1{[STORE

]3[SUBJVP

[ ]

[ ]

oftennot ]'4[DOM

]1[STORE

]1[CONT

[2]MODAdvP

quant-neg

]'7[],'6[],'5[],'4[DOM

{}STORE

]1[QUANTSS


193

e. I remained silent, nor did he speak a single word.

((b)–(e) Mazzon 2004: 104–105)

The configurational analysis assumes that the initial negative expression in NI is in [Spec,FocP].

As a conjunction, however, nor is not involved in an unbounded dependency relation.

Therefore it would be difficult to assimilate it to wh-expressions, which are always in an

unbounded dependency relation.

In our treatment of NI sentences outlined in 6.3, the initial positioning of negative

expressions is constrained only in terms of the linear sequence. This means that there is no

specification about the status of the negative expressions in constituent structure. Another

feature of our treatment of NI is that there are just a few constraints on the internal property of

the initial negative element: (32) stating that it should not contain a wh-expression, and (35)

stating that it should contain a negative expression. Other aspects of the internal structure are

underspecified. This means that any syntactic category can in principle be allowed in the first

position in NI. For example, (1a) contains an adverbial phrase, (1b) an NP and (1c) a PP.

The examples are repeated here for convenience.

(98) a. Under no circumstances will he eat raw spaghetti.

b. No race could Lewis win.

c. With no job would Mary be happy. [= (1)]

The absence of any restriction on constituent structure status and the underspecification of

the internal structure allow the occurrence of nor in position first as in (97), although it is a

conjunction and is not involved in an unbounded dependency relation. The nor-clause in (97a)

is given the following order domain representation.

(99) ,,,DOM

mountains the to go

fourth

she

third

does

secondfirst

nor

The conjunction nor is a negative expression, so the constraint in (35) for the type

negative-inversion licenses its occurrence in first position of the clausal domain, however


194

fashion it combines with the rest of the clause in constituent structure.19,20

6.6.2 Individual variation: unbounded extraction of negative phrases

In 6.2.1.1 we noted that unbounded extraction of a negative phrase is unacceptable for many

speakers (Sobin 2003: 184–185), and we gave a linearisation-based approach to this

phenomenon in 6.5.4. However, there are some native speakers of English who accept

unbounded extraction of a negative phrase. The examples in (100a,b) are given to the author by

one such speaker, and other examples cited in the literature are given in (100c) to (h).

(100) a. Not a single one of the boys would Michael Jackson admit that he had slept with.21

b. Never again do I think that she will eat clams.

c. Nothing did the doctor say the baby must eat. (Cormack and Smith 2000)

d. No such chemicals did he know that there were in the bottle.

e. No such car did they ever ask whether I had seen.

f. No such car would they have preferred it if I had bought.

g. No other colours did he think they had ever painted their car.

h. No theory did Ernie interview any natives who accepted or contact any foreigners

who rejected. ((d)–(h), Postal 1998)

For example, the initial negative phrase not a single one of the boys in (99a) is a complement of

the preposition with, which belongs to the embedded that-clause. In the present analysis,

liberation of negative expressions is due to the constraint on hd-normal-comp-ph in (48), which

is repeated below.

19 We will not discuss the combinatorial relation between the conjunction nor and the rest of the clause in

constituent structure since it is of little importance to the discussion in the present thesis. 20 It is assumed that the clause after nor is uncompacted although other conjunctions are different in this respect.

21 I would like to thank Andrew Radford for providing me with these examples.


195

(101) [ ]

→

[1]DTRS-NH

[4]DOMDTR-HD

[5]DOM

ph-comp-normal-hd


∧ ( )]5[],4[,[3],[2]shuffle

[ ]+∧ NEG:]3[

¬∨∧

SUBJ

HEAD SYNSEM:]2[:]3[

verb

[ ]fifth:]2[∧

∨

:]3[

:]2[

DOM:]1[

∧

∧

sign

[= (48)]

This constraint, as it is, excludes (100), due to the third clause of the first disjunct which prevents

extraction of negative expressions out of clauses.

We assume that the existence of this particular clause is subject to individual variation:

those speakers who do not accept (100) have this specification in their constraint on

hd-normal-comp-ph, but those who accept (100) do not have it. For the latter type of speaker,

the constraint is as follows.

(102) [ ]

→

[1]DTRS-NH

[4]DOMDTR-HD

[5]DOM

ph-comp-normal-hd


∧ ( )]5[],4[,[3],[2]shuffle

[ ]+∧ NEG:]3[

[ ]fifth:]2[∧

∨

:]3[

:]2[

DOM:]1[

∧

∧

sign


196

[cf. (48), (101)]

It is worth noting that long fronting of negative phrases cannot be accounted for in terms of

the SLASH mechanism as in the normal cases of extraction; as we have discussed, preverbal

adverbs cannot be fronted with SLASH mechanism.

(103) a. (*Merely) Kim (merely) opened the door.

b. (*Almost) Kim (almost) found the solution.

c. (*Never) I have (never) seen a ghost. [= (82)]

This means that never again in (100b), a preverbal adverb, is fronted without the SLASH

mechanism. Our analysis in terms of partial compaction can capture the fact.

6.7 Summary and concluding remarks

Let us summarise this chapter. We first looked at how the configurational analysis within

Minimalism/Principles and Parameters theory deals with NI constructions, and then provided

some pieces of data that are problematic for the approach. We provided a linearisation-based

approach to NI. We proposed that NI sentences are of the clause type negative-inversion, and

that this is a subtype of a type v2-decl, which in turn is a subtype of both v2 and decl. Thus NI

sentences should satisfy all the constraints on these clause types. Of considerable importance

here is that these constraints concern just the sentence type and the internal syntax of the clause,

so do not impose any restriction on constituent structure. The initial placement of a negative

expression is thus treated as a linearisation phenomenon in a clausal order domain.22 It was

shown that this approach can accommodate all the data problematic for the configurational

approach.23

22 Culicover (1999: 162–165) suggests that NI should be analysed in terms of the linear sequence, but he does not

offer an actual analysis. 23 The present approach is somewhat similar to Sobin’s (2003) analysis within Minimalism in that the initial

positioning of the negative expression does not involve an unbounded dependency. Sobin (2003) posits the clause

structure shown below, which includes a simpler CP layer.

(i) [CP … [AgrP … [NegP … [TP … [VP …

The negative expression associated with NI constructions is located in [Spec,NegP]. Thus, an NI construction is


197

The present analysis accommodates not just the construction-specific properties of NI

sentences but also the regularities that they share with other constructions. The use of

hierarchically organised network of clausal types allows us to have constraints of any level of

generality. The present approach can thus capture the distinctive properties of NI sentences

without missing any generalisations.

given an analysis such as (ii).

(ii) [CP [AgrP [NegP never again [Neg ØNeg] [TP [T will] [VP he [V' …

There is no attraction of the verb to the negative expression (i.e., no Negative Criterion). The apparent inversion is

impeded movement where the elements involved (verb and subject) fail to arrive at the normal declarative surface

positions. There are two problems, empirical and conceptual. First, this approach predicts a full grammaticality

of sentences like (iii).

(iii) % Beans, never in my life will I eat.

Secondly, in this approach, [Spec,AgrP] in NI constructions is empty, which violates the Extended Projection

Principle. Sobin provides a couple of possible solutions, but the mechanisms involved require further

development.

Chapter 7

Comparison with other approaches∗∗∗∗

Contents

7.1 Introduction .....................................................................................................................198

7.2 Word Grammar approach..............................................................................................199

7.2.1 Outline of Word Grammar .......................................................................................200

7.2.2 WG approach to the phenomena..............................................................................204

7.2.3 Problems .....................................................................................................................205

7.2.3.1 Weakness of the No-Tangling Principle..............................................................206

7.2.3.2 Headness of a wh-pronoun ..................................................................................208

7.2.3.3 An extractee preceding a complementiser ..........................................................209

7.3 Chung and Kim’s (2003) linearisation-based analysis.................................................210

7.3.1 Some predictions ........................................................................................................212

7.3.2 Problems .....................................................................................................................214

7.4 Concluding remarks........................................................................................................216

7.1 Introduction

There are two ways to represent the relationship between individual words: dependency

structure and phrase structure. The former is a pure representation of word-word relationships

while the latter includes additional information that words are combined to form constituents. If

all work can be done just by means of the relationship between individual words, phrase

structure is redundant and hence dependency structure is preferable to it. It would therefore be

worth considering whether all work can really be done with just dependencies. There is indeed

∗ Portions of this chapter have been published as Maekawa (2005a).


a family of frameworks, collectively called ‘dependency grammar’, which uses with just

dependencies. The purpose of section 7.2 is to take a critical look at treatment of linear order

within an instance of dependency grammar, and compare it with our linearisation HPSG

approach. We will particularly focus on a version of Word Grammar (WG) developed by

Hudson (2003b). In 7.3 we will turn to Chung and Kim’s (2003) analysis in linearisation-based

HPSG, and point out some problems.

7.2 Word Grammar approach

Hudson (1995) compared WG with Pollard and Sag’s (1994) version of HPSG and showed

some advantages of WG; in particular, he argued that phrase structure could be removed from

HPSG in favour of dependency structure. His criticism was mainly on Pollard and Sag’s

treatment of fronted adjuncts. As we discussed in Chapter 5, there is no straightforward way of

explaining the linear order asymmetries between main clauses and subordinate clauses in

Pollard and Sag’s (1994) version of HPSG.

(1) a. Who had ice-cream for supper?

b. For supper who had ice-cream?

(2) a. Who had ice-cream for supper is unclear.

b. * For supper who had ice-cream is unclear.

A fronted adjunct can precede a fronted wh-element in a main clause, but this is not possible in

an embedded clause. Hudson (1995) offers an alternative analysis in the framework of WG, and

the idea developed there has been further developed in Hudson (1999) and Hudson (2003b).

In this section we will discuss how the WG approach deals with the phenomena in

question. We will then point out that it faces some problems. Before looking at the WG analysis,

however, we will briefly outline how word order, wh-constructions and extractions are treated in

WG in 7.2.1.


7.2.1 Outline of Word Grammar

In WG word order is controlled by two kinds of rule: general rules that control the geometry of

dependencies and word-order rules that control the order of a word in relation to other word(s):

its landmark(s) (Hudson 2003a). In simple cases a word’s landmark is its parent: the word it

depends on. In the cases where the word has more than one parent, only the ‘higher’ parent

becomes its landmark (Promotion Principle; see Hudson 2003a). For example, let us consider a

sentence It was raining. The raised subject it depends on two verbs, was and raining, so it has

two parents. In this case, eligible as its landmark is was. This is because raining depends on

was, so the latter is the ‘higher’ of the two. In a WG notation, It was raining is represented as

shown below.

(3)

s r

It was raining.

s

The fact that it is the subject of the two verbs is indicated by the two arrows labelled ‘s’ (subject).

The arrow labelled ‘r’ indicates that raining is a ‘sharer’ of was. This is so named since it shares

the subject with the parent verb. In a notation adopted here, the dependencies that do not

provide landmarks are drawn below the words. Therefore, one of the ‘s’ arrows, the one from

raining to it, is drawn below the words. We thus pick out a sub-set of total dependencies of a

sentence and draw them above the words. This sub-set is called surface structure. Word-order

rules are applied to it, and determine the positioning of a word in relation to its landmark or

landmarks. Thus, the surface structure is the dependencies which are relevant for determining

word order. A word-order rule specifies that a subject normally precedes its landmark, and

another rule specifies that a sharer normally follows its landmark, as illustrated by the

representation in (3).

Among the rules that control the surface structure, the No-Tangling Principle is the most


important for our purpose: dependency arrows in the surface structure must not tangle.1 This

principle excludes the ungrammatical sentence (4b).

(4) a. He lives on green peas.

b * He lives green on peas.

The dependency structures of this pair are shown in (5).

(5) a.

He lives on green peas

b.

* He lives green on peas

(5b) includes tangling of the arrows. Its ungrammaticality is predicted by the No Tangling

Principle.

Let us turn to the WG treatment of wh-interrogatives. Consider the dependency structure

of What happened? for example. As in the case of an ordinary subject such as it in (3), the

grammatical function of the wh-pronoun what to the verb happened is a subject. Therefore,

what depends on happened, and this situation can be represented as follows.

(6)

s

What happened?

On the other hand, Hudson (1990: 361–382; 2003b) argues that the verb is a complement of the

wh-pronoun and thus depends on it.

1 In the current version of WG (Hudson 2003a), the No-Tangling Principle has been replaced with Order Concord,

whose effects are essentially the same as its predecessor. In this thesis we will refer to the No-Tangling Principle.


(7)

c

What happened?

The evidence for the headness of wh-pronoun includes the following phenomena (Hudson

2003b). First, the pronoun can occur without the verb in sluicing constructions.

(8) a. Pat: I know he’s invited a friend.

Jo: Oh, who [has he invited]?

b. I know he’s invited a friend, but I’m not sure who [he’s invited].

Second, the pronoun is what is selected by the higher verb. In (9) wonder and sure require a

subordinate interrogative clause as their complement. For a clause to be subordinate

interrogative, the presence of either a wh-pronoun, or whether or if is required.

(9) a. I wonder *(who) came.

b. I’m not sure *(what) happened.

Third, the pronoun selects the verb’s characteristics such as finiteness and whether or not it is

inverted. (10) illustrates that why selects a finite or infinite verb as its complement, but when

only selects a finite verb.

(10) a. Why/when are you glum?

b. Why/*when be glum?

(11) indicates that why selects an inverted verb as its complement whereas how come selects a

non-inverted verb.

(11) a. Why are you so glum?

b. * Why you are so glum?

c. * How come are you so glum?

d. How come you are so glum?

(12) illustrates that what, who and when selects a to-infinitive, but why does not.

(12) I’m not so sure what/who/when/*why to visit.


Hudson (2003b) argues that all of these phenomena are easily accounted for if the wh-pronoun is

a parent of the next verb. In the framework of WG, therefore, there is no reason to rule out any

of (6) and (7); the sentence is syntactically ambiguous. Thus, in What happened? what and

happened depend on each other, and the dependency structure may be either of (13a) and (13b).

(13)

s

What happened?

c

s

What happened?

c

Thus, wh-interrogatives may involve a mutual dependency. In (13b), happened is the head and

the dependency labelled ‘s’ is put in surface structure. In (13a), however, what is the head, and

the dependency labelled ‘c’ (complement) is put in surface structure.

Finally, we outline how extraction is dealt with in WG. Let us consider (14a) with an

fronted adjunct in the sentence initial position.

(14) a. Now we need help.

b. We need help now.

The fronted adjunct now would otherwise follow its parent need as in (14b), but it is precedes it.

This situation is represented in WG by adding an extra dependency ‘extractee’ to now. ‘x<’

(15)

Now we need help.

The arrow from need to now is labelled ‘x<, >a’, which means ‘an adjunct which would

o

x<, >a

s


normally be to the right of its parent (‘>a’) but which in this case is also an extractee (‘x<’)’.

Thus the adjunct now is to the left of the parent verb need.

7.2.2 WG approach to the phenomena

With this background in mind, let us now turn to the asymmetry between main and subordinate

clauses in question: adverb-preposing is not possible in subordinate interrogatives although it is

possible in main interrogatives.

(16) a. Now what do we need?

b. * He told us now what we need.

As stated above, a wh-pronoun and its parent are mutually dependent. In (16a) do is the

complement of what whereas what is the extractee of do. Thus, the dependency structure for

(16a) would be either of (17a) and (17b). In the former, what is the parent, and the dependency

labelled ‘c’ from what to do is put in surface structure. In the latter, however, do is the parent

and the dependency labelled ‘x<’ from do to what is put in surface structure. The fronted

adjunct now is labelled ‘x<, >a’, and precedes its parent do. As the diagram show, the ‘x<, >a’

arrow from do to now tangles with the vertical arrow in (17a). Thus, it violates the No Tangling

Principle. On the other hand, there is no tangling in (17b), so it is the only correct WG analysis

of (16a).

(17) a.

Now what do we need?

c x<,>a

c s

x<,o

c s


b.

Now what do we need?

Let us turn to the subordinate wh-interrogative in (16b). In (16b) what is the object and the

extractee of need while need is the complement of what. It has the structure represented in (18).

What is the clause’s subordinator and it has to be the parent of the subordinate clause. The

dependency labelled ‘c’ should be put in surface structure since if the arrow labelled ‘x<, o’ were

in the surface structure, what has two parents and it violates the No-Dangling Principle: words

should not have more than one parent in surface structure (Hudson 2003a). As the diagram

show, the arrow from need to now is tangled with the one from told to what. Unlike the main

clause case in (17), it has no alternative structure, so (16b) is ungrammatical.

(18)

told us now what we need

Thus, WG can capture the linear order asymmetries of the main and subordinate clauses in terms

of dependencies in surface structure and general principles on dependencies.

7.2.3 Problems

Although the WG analysis looks successful in accommodating the asymmetry between the main

and subordinate clauses, there are some weaknesses, to which we turn to in this section.

x<, >a c

c s

x<,o

x< s

o c

s

x<, o

c x<, >a


7.2.3.1 Weakness of the No-Tangling Principle

As surveyed above, the WG approach states that adjunct preposing is possible out of main

wh-interrogatives because the fronted adjunct avoids violation of the No-Tangling Principle due

to the fact that it is a co-dependent of the wh-element (Hudson 2003b: 636). The argument

along these lines would suggest that extraction is allowed as long as it does not violate the

No-Tangling Principle. However, there are cases in which extraction out of embedded

wh-interrogative is excluded although it does not violate the No-Tangling Principle. The data

comes from negative inversion constructions illustrated by (19).2

(19) Under no circumstances would I go into the office during the vacation.

In WG, a preposed negative expression, such as under no circumstances in (19), is a kind of

extractee (Hudson 2003a), so we should expect that it behaves like a fronted adjunct. As

expected, preposed negative expressions cannot be extracted out of the subordinate

wh-interrogatives, as illustrated by the following examples.

(20) a. * Lees wonders under no circumstances at all why would Robin volunteer.

b. * I wonder only with great difficulty on which table would she put the big rock.

(a,b from Chung and Kim 2003)

A WG approach would suggest that this is due to the No-Tangling Principle.

(21)

c x<, >a c

c s

wonders under no circumstances why would Robin volunteer

x<, >a

x<, >a

x<, >a

2 The examples in this section are cited from Haegeman (2000) unless otherwise indicated.


With this in mind, let us consider the main wh-interrogative clause. A WG approach would

predict that preposing of a negative expression is possible out of main wh-interrogatives because

it should not involve violation of the No-Tangling Principle, as in the case of adjunct preposing.

(22)

x<, >a c

x< s

In no way why would Robin volunteer?

c s

x<, >a

However, it is actually ungrammatical.

(23) a. * In no way, why would Robin volunteer?

b. * Only with great difficulty on which table would she put the big rock?

(a,b from Chung and Kim 2003)

Here the preposed negative expression precedes the wh-element. Note that the situation is

completely on a parallel with the case of the adjunct preposing like (16a), which is repeated

here.

(24) Now what do we need?

As we have seen, (24) is grammatical because it does not violate the No-Tangling Principle.

However, the sentences in (23) are ungrammatical though they do not violate the same principle.

This makes the analysis in terms of the No-Tangling Principle less plausible.

As we saw at the outset of this section, word order is controlled by two kinds of rule in WG:

general rules, such as the No-Tangling Principle that control the geometry of dependencies and

word order rules that control the order of a word in relation to its landmark or landmarks

(Hudson 2003a). Someone might suggest that the No-Tangling Principle is simply irrelevant in

(23) and that a word-order rule could exclude the ill-formed order. However, there are some

problems in this approach. Let us suppose that WG has a rule which exclude the NEG < WH

order, where NEG stands for a preposed negative expression and WH for a wh-expression. This


rule correctly excludes (23). It is natural to suggest that the same rule could apply not only to

main clauses but also to subordinate clauses. As predicted, the subordinate clauses with the

same elements in the same order as (23) are ungrammatical. This is actually illustrated by (20),

repeated below for convenience.


b. * I wonder only with great difficulty on which table would she put the big rock.

[= (20a,b)]

Now we should recall, however, that they can be excluded by the No-Tangling Principle; a

preposed negative expression is extracted from the subordinate wh-interrogative clause, as

depicted in (21). The situation is entirely on a parallel with (16b). We now have two possible

ways to exclude (25): the No-Tangling Principle and the word-order rule. This sort of

redundancy would not be preferable in any linguistic theories.

It might be possible to account for (25) by the No-Tangling Principle, and the

corresponding main clauses by a word-order rule. It is clear, however, that we cannot adopt

this approach: it forces the word-order rule to refer to main clauses. Note that WG does not have

a unit larger than a word, so it does not recognise clauses (Hudson 2003a). It does not, therefore,

have a way to distinguish main and subordinate clauses, apart from the assumption that the latter

has a subordinator and a parent outside of the clause (Hudson 1990: 375–376). It is, then,

impossible for WG rules to refer to any clause.

It seems, then, that the No-Tangling principle is not effective enough to capture the

preposing facts.

7.2.3.2 Headness of a wh-pronoun

Recall that the most important assumption for a WG approach is that the wh-pronoun is the

parent of the subordinate wh-interrogatives. We should note that this assumption itself is not

without problem. Consider examples in (26a) and (26b); the former is the one cited by Hudson


himself as a problematic data for his analysis (Hudson 1990: 365).3

(26) a. Which students have failed is unclear.

b. Who shot themselves is unclear.

In WG treatment of wh-pronoun, which and who are not only the subject of have and shot,

respectively, but also the subject of is. The verb should agree in number with its subject, so

have/shot and is should both agree with which/who. Which in (26a) should share its plurality

with students since the former is a determiner of the latter; who in (26b) should share its

plurality with themselves since the former is the antecedent of the latter. This does not explain

the morphology of the copula verb in both sentences, which requires the singular subject. This

analysis would predict the sentences like the following to be grammatical.

(27) a. * Which students have failed are unclear.

b. * Who shot themselves are unclear.

The copular verb is are, not is, agreeing with its subject which in (a) and who in (b). These

sentences are, however, ungrammatical. Thus, the assumption that the wh-pronoun is the parent

of the subordinate interrogatives has weakness.

7.2.3.3 An extractee preceding a complementiser

We should also note that there are some cases where an extractee is allowed to precede the

complementiser. The following examples are from Ross (1986).

(28) a. Handsome though Dick is, I’m still going to marry Herman.

b. The more that you eat, the less that you want.

In (28a), the first clause is the subordinate clause, and the adjective handsome, a complement of

is, is in front of the complementiser though. In (28b) the more, which is an object of eat and

want, is followed by the complementiser that.4 It would be natural to assume the fronted

3 (26b) was provided to me by Bob Borsley (p.c.)

4 (28b) is not acceptable to some speakers (Borsley 2004).


elements in these examples to be an extractee in WG’s terms; but if so, the dependency arrow

from the verb to the extractee would tangle with the vertical arrow to the complementiser, and

hence the resulting structure in (29) violates the No-Tangling Principle.5

(29)

Clever though Kim is, …

It seems, then, that a WG approach to the asymmetry between main and subordinate

wh-interrogatives has some problems.

7.3 Chung and Kim’s (2003) linearisation-based analysis

Chung and Kim (2003) give an analysis of English left peripheral elements in linearisation

HPSG. A notable feature of this approach is the topological organisation of domain elements.

They assume that the left periphery of English clause structure can be divided into three regions:

marker field, topic field and focus field, which are abbreviated as 1, 2 and 3, respectively. The

positional assignment is determined by the following constructional constraints.

(30) a. topic-ph:

[ ]

−

+

+→

INV

IC

VFORM

,TOPIC

][DOMfin

2H

b. foc-cl:

[ ] [ ]+

+→ IC ,

WH/NEG

][DOMH

3

5 The data in (28) could be accommodated in WG if we assumed a dependency relation between the

complementiser and the extractee (Borsley (p.c.); and Sugayama (p.c.)). Needless to say, however, an argument


c. embed-wh-cl:

[ ] [ ]−

+→− INV ,

WH

][DOMIC H

1

(Chung and Kim 2003: 73–74)

Wh-elements are assigned to position 3 in main clauses, and those in embedded (interrogative

and relative) clauses are put in position 1. Topic elements are always assigned to position 2, and

the operators are always assigned to position 3. Thus, left peripheral elements in English have

the following distributions.

(31) Distribution of English left peripheral elements6

Marker field

1

Topic field

2

Focus field

3

Main clause topic wh/neg-op

Embedded clause wh/comp topic neg-op

(Chung and Kim 2003: 73)

An embedded wh-phrase competes for position 1 with a complementiser. This competition

accounts for the fact that these two elements never co-occur in Standard English (cf. Chomsky

and Lasnik 1977). This competition is captured by the Uniqueness Condition.

(32) Uniqueness Condition

i < i (where i = {1, 2, 3}) (Chung and Kim 2003: 76)

This constraint entails that there should not be more than one element in positions 1, 2 and 3.

They assume another linear precedence constraint: the Topological Linear Precedence

Constraint.

(33) Topological Linear Precedence Constraint

1 < 2 < 3 (Chung and Kim 2003: 74)

(33) states that the elements in position 1 should precede those in 2, which should in turn

precede those in 3.

along these lines would need to clarify the nature of this apparently ad hoc grammatical relation. 6 In (31), ‘neg-op’ stands for ‘negative operator’, which corresponds to preposed negative expressions in negative


7.3.1 Some predictions

Now let us consider how this approach might accommodate the asymmetry between main and

subordinate wh-interrogatives. As introduced above, Chung and Kim’s (2003) approach

assumes that a topic is in position 2 and a wh-element is in position 3 in main clauses. The

crucial assumption of their approach is that what we have called adjunct fronting is treated as a

kind of topicalisation. Thus, in their treatment, now in (34) is a topic.

(34) Now what do we need?

The adverb now is therefore subject to the constraint in (30a) and is required to be in 2. The

wh-phrase what is assigned to 3 due to the constraint in (30b), which determines the position of

wh-phrases in main clauses.

(35) [DOM <[2 now], [

3 what],…>]

The representation in (35) is in consonant with the Topological Linear Precedence Constraint in

(33). Hence, (34) is grammatical.

Let us turn to embedded clauses, where nothing can precede a wh-expression.

(36) * He told us now what we need.

The constraint in (30c) states that wh-elements are assigned to position 1 in embedded clauses.

The constraint in (30a) requires that topic elements are always in position 2, no matter whether it

is embedded or not. Thus, the left periphery of the embedded clause in (36) has the following

order domain.

(37) * [DOM <[2 now], [１ what],…>]

(37) violates the Topological Linear Precedence Constraint since its domain element marked 2

precedes that marked 1. This explains the ungrammaticality of (36). Thus, Chung and Kim’s

(2000) approach can accommodate the asymmetry between main and embedded clauses with

respect to a topic/fronted adjuncts and a wh-element.

Their analysis can also predict the grammaticality of (38a), whose representation for the

inversion constructions.


order domain is (38b).

(38) a. I was wondering for which job, during the vacation, I should go into the office.

b. [DOM <[1 for which job], [

2 during the vacation],…>]

This linear organisation is consistent with the constraint in (33).

They can also account for the ungrammaticality of the following sentence.

(39) * To whom, a book like this, would you give?

Under their treatment, wh-phrases are in 3 in main clauses and topics are in 2. Therefore the left

periphery of this sentence has the following representation of the order domain.

(40) * [DOM <[3 to whom], [

2 a book like this],…>]

Here, the element which is assumed to be in 3 precedes that which is assumed to be in 2. This

representation violates the Topological Linear Precedence Constraint in (33). Chung and Kim’s

(2003) analysis thus correctly predicts that (39) is ungrammatical.

Constraint (30b) states that wh-elements and preposed negatives are both assigned to

position 3 in main clauses. This accounts for the ungrammaticality of the sentence in (41a).

(41) a. * In no way, why would Robin volunteer?

b. * [DOM <[3 in no way], [

3 why],…>]

Due to the constraint in (32), there should not be more than two elements in one identical

position.

A wh-phrase is assigned to position 1 in embedded clauses while preposed negatives are

assigned to position 3, embedded or not. This accounts for the grammaticality of (42a); its

linear organisation shown in (42b) does not violate the constraint in (33).

(42) a. Lees wonders why under no circumstances would Robin volunteer.

b. [DOM <[1 why], [

3 under no cirsumstances],…>]

(43a) is correctly excluded since its order domain (43b) violates the constraint in (33).


b. * [DOM <[3 under no cirsumstances], [

1 why]…>]


Thus, a linearisation-based HPSG approach by Chung and Kim (2000) can provide an

account for many pieces of data. However, their approach faces some problem, which we will

turn to below.

7.3.2 Problems

There are at least two problems in Chung and Kim’s (2003) analysis. Firstly, they give a unified

treatment of fronted adjuncts and topics. Secondly, they assume that wh-phrases and preposed

negative phrases always occur in the same position in main clauses. We will look at these

problems in more details below.

Chung and Kim’s (2003) approach faces a problem when we consider the cases where

wh-phrases are followed by fronted adjuncts in main clauses.

(44) For what kind of jobs during the vacation would you go into the office?

Under their treatment, wh-phrases are in 3 in main clauses and fronted adjuncts/topics are in 2.

Therefore the left periphery of this sentence has the following representation of the order

domain.

(45) * [DOM <[3 for what kind of jobs], [

2 during the vacation],…>]

Here, the element which is assumed to be in 3 precedes that which is assumed to be in 2. This

representation violates the Topological Linear Precedence Constraint in (33). Chung and Kim’s

(2003) analysis thus predicts ungrammaticality of (44), which is contrary to the fact.7 In the

version of linearisation HPSG presented in this thesis, the left periphery of the sentence in (44)

is given the following order domain representation (Chapter 5).

(46) [DOM <[first

for what kind of jobs], [during the vacation],…>]

The non-subject wh-expression is assigned to first position in main clauses, as we argued in

Chapter 5. The adjunct during the vacation is an incidental phrase, which is not categorised into

7 Chung and Kim (2003) assume that (44) is ungrammatical. Our informants judge it grammatical, as noted in

Chapter 5.


any position class, and therefore has a rather free positioning.

Secondly, Chung and Kim (2003) assumes that wh-phrases are in 1 in embedded clauses,

and topic phrases are in 2, as in main clauses. This predicts that (47a) is grammatical.

(47) a. * I wonder to whom this book, Bill should give.

b. [DOM <[1 to whom], [

2 this book],…>]

As we noted in Chapter 5, however, topic phrases cannot follow wh-phrases.8 In our analysis,

wh-phrases are in second in embedded clauses, and topic phrases are in first.

(48) * [DOM <[second

to whom], [first

this book],…>]

The representation in (48) is ill-formed since the element which is supposed to be in second

precedes the element which is supposed to be in first. This correctly predicts the

ungrammaticality of (47a).

Chung and Kim (2003) proposed that wh-phrases and preposed negatives are assigned to 3

in main clauses, due to the constraint (30b). This predicts that they do not co-occur.

(49) a. Why, in no way would Robin volunteer?

b. [DOM <[3 why], [

3 in no way],…>]

The constraint in (32) requires that that there should not be more than two elements in one

identical position. The linear organisation in (49b) violates this constraint. As noted in Chapter

6, however, our informants judge sentences such as (49a) as grammatical. In our analysis, the

left periphery of the sentence in (49a) is represented in the following way.

(50) [DOM <[pre-first

why], [first

in no way],…>]

As discussed in Chapter 6, non-subject wh-expressions can be assigned to pre-first position.

Therefore, the wh-expression in this position can co-occur with the preposed negative

expression in first position.

It seems, then, that Chung and Kim’s (2003) approach to the left periphery of the English

clauses has some problems.

8 Chung and Kim (2003) assumes that (47) is acceptable. Our informants judge it ungrammatical.


7.4 Concluding remarks

In this chapter, we first looked at a WG approach (Hudson 2003b) to the asymmetries between

main and embedded clauses with respect to the elements in the left periphery of a clause, and

argued that it has some problems in dealing with the relevant facts. As we discussed at the

outset of this paper, dependency structure is simpler than phrase structure in that the former only

includes information of the relationship between individual words, but the latter involve

additional information about constituency. Other things being equal, simpler representations

are preferable to more complex representations. This might lead to the conclusion that WG is

potentially superior to HPSG. The discussions in this chapter have shown, however, that the

dependency-based analysis in WG cannot account for the linear order facts. Note that we have

argued throughout the preceding chapters that constituency-based analyses are not satisfactory,

either. These two frameworks follow the traditional distinction between the rules for word order

and the rules defining the combinations of elements.9 However, the rules for word order are

applied to local trees in constituency-based approaches and to dependency arrows in WG.

Sisters must be adjacent in the former whereas in the latter the parent and its dependent can only

be separated by elements directly or indirectly depend on one of them. This means that the

linear order is still closely tied to the combinatorial structure. That these frameworks cannot

accommodate certain linear order facts suggests that neither dependency structure nor phrase

structure is appropriate as the locus of linear representation. We have shown throughout this

thesis that linearisation HPSG analysis gives a satisfactory account of linear order of elements in

the left periphery. This conclusion suggests that we need to separate linear order from

combinatorial mechanisms more radically than the above traditional separation of the rules.

In 7.3, we had a critical look at Chung and Kim’s (2003) version of linearization HPSG

approach. As discussed in 7.3.1, their analysis can account for many pieces of data. However,

we found that it is problematic in two respects: they give a unified analysis to topicalisation and

adjunct fronting; and they assume that wh-phrases and preposed negative phrases are always in

9 In constituency-based grammars such as HPSG, these two rule-types are linear precedence rules and immediate

dominance rules.


the same position in main clauses. In 7.3.2 we saw that our version of linearisation HPSG can

provide a satisfactory account of the data.

Chapter 8

Concluding remarks

In the preceding chapters, we investigated the linear organisation of left periphery of English

clause structure. Following Kathol’s (1995, 2000, 2001) adaptation of ‘topological fields’,

which has been traditionally employed in analyses of German syntax (e.g., Drach 1937; Höhle

1986), we divided English sentences into five regions, and marked each element of a clause

for a particular region that it is assigned to.

From Chapter 4 to 6 we developed a linearisation-based analysis of particular

constructions in English. In Chapter 4, we looked at the left periphery of interrogative and

relative clauses. We first noted that initial wh-expressions in these constructions are

extracted and should be dealt with in terms of the SLASH mechanism. However, the

distribution of who and whom observed in non-prescriptive formal register gives evidence that

linear position of wh-subjects is different from that of non-subject wh-expressions. This

disparity between configurational position and linear position gives strong support of our view

that linear order is independent from constituent structure.

In Chapter 5 we first argued that earlier ‘configurational’ analyses cannot give a

satisfactory account of the behaviour of fronted adjuncts. These approaches face difficulties

since they identify the linear position of fronted adjuncts with their position in constituent

structure. We showed that extracted adjuncts (long-fronted adjuncts and fronted NP

arguments) and ‘incidental’ adjuncts should be differentiated. In our linearisation-based

analysis, extracted adjuncts occupy first position of sentences, and incidental adjuncts are not

categorised into any position class. This characterisation of fronted adverbials can provide a

fairly straightforward account of the facts which are problematic for the configurational

approaches.

In Chapter 6, we provided a linearisation-based approach to preposed negative phrases

in negative inversion (NI) constructions. We argued that NI does not impose any restriction


on constituent structure. The initial placement of a negative expression is thus treated as a

linearisation phenomenon in a clausal order domain.

In Chapter 7 we had a critical look at other approaches to the left periphery. First we

discussed Hudson’s (2003b) Word Grammar approach. Word Grammar represents linear

order on the basis of word-word relationships. We argued that this approach cannot provide a

satisfactory account of the facts which our linearisation-based approach can accommodate

quite easily. We also looked at Chung and Kim’s (2003) another approach in linearisation

HPSG, and argued that they include some problems.

Throughout the thesis, we have employed just a small number of combinatory schemata,

whereas we proposed a number of linear constraints. The preceding chapters showed that

arguments along these lines are on the right track. This suggests that linear precedence is

more significant than constituent structure as the determining factor in formulating linguistic

generalisations. We regard it as a preferable conclusion from the perspective that linear order

is epistemologically prior to constituent structure (Culicover and Jackendoff 2005, Culicover

2003, Kathol 2000, etc.).

One might argue for assuming just a flat constituent structure and against separating

linear order from constituent structure. This is essentially a view taken by Culicover and

Jackendoff (2005) and Culicover (2003) to handle linearisation phenomena such as heavy shift,

scrambling, etc. Note, however, that we have dealt with some phenomena which force us to

keep the distinction between the two levels. On the one hand, a single position can be

occupied by elements with different kinds of constituent structure status: first position can be

occupied by fillers and preposed negatives, and second position can be occupied by embedded

wh-fillers, complementisers, and finite auxiliary verbs. On the other hand, elements with the

same constituent structure status can be assigned to different positions in order domains:

wh-subjects and wh-non-subjects are fillers but the former are in third and the latter are in first

in matrix clauses and second in subordinate clauses. Moreover, there are some phenomena

which can be characterised only in either constituent structure or order domains. On the one

hand, incidental adjuncts have a specific constituent structure status as modifiers, but their

linear status is not so specific: they are not categorised into any topological field. On the


other hand, preposed negatives in negative inversion constructions are only constrained in

terms of the linear sequences, and there is no specification about their constituent structure

status. It seems, then, that we would miss important generalisations if we just had a flat

constituent structure.

Appendix A: Phrase types

A.1 Type hierarchy

(1) Classification of phrase

hd-neg-comp-ph hd-normal-comp-ph

phrase

hd-ph

hd-adj-ph hd-comp-ph hd-subj-ph hd-filler-ph

hd-incid-adj-ph hd-neg-adj-ph hd-normal-adj-ph

cp-ph

hd-su-filler-ph hd-ns-filler-ph


222

A.2 Constraints on phrase types

(2) [ ]COMPS→phrase

(3) [ ]

→

][n' ][2' ]1[DOM

[n] ]2[DTRS-NH

[1]DOMDTR-HD

oKoo

oKoph-hd

(4) Head Feature Principle

a. The HEAD value of any headed phrase is structure-shared with the HEAD value of

the head daughter.

b.

→

]1[HEAD|CAT|LOC|SYNSEM|DTR-HD

[1]HEAD|CAT|LOC|SYNSEMph-hd

(5) Valence Principle (Pollard and Sag 1994: 348)

In a headed phrase, for each valence feature F, the F value of the headed daughter is

the concatenation of the phrase’s F value with the list of SYNSEM values of the

F-DTRS value.

(6)

[ ]

→

[1]SSDTR-NH

]1[SUBJDTR-HD

SUBJ

phraseph-subj-hd

(7)

→

Ko ]2[DOM

[1]DTR-NHph-subj-hd

where: ( ) ≡[2][1],compaction

[ ]third

nelist

sign

:[2]

DOM:]1[

∧

∨

:[2]

DOM:]1[

∧

sign


223

(8) [ ] [ ]

→[n],[1],COMPS

DTR-HD

[n]SS,,[1]SSDTRS-NH

K

K

wordph-comp-hd

(9) [ ]

→

[1]DTRS-NH

[4]DOMDTR-HD

[5]DOM

ph-comp-normal-hd


∧ ( )]5[],4[,[3],[2]shuffle

[ ]+∧ NEG:]3[

¬∨∧

SUBJ

HEAD SYNSEM:]2[:]3[

verb

[ ]fifth:]2[∧

∨

:]3[

:]2[

DOM:]1[

∧

∧

sign

(10) [ ][ ]

+→

KK

KK

,NEG]1[,DTRS-NH

,][1',DOM firstph-comp-neg-hd

(11) [ ][ ]

→

[1]SSDTR-HD

[1]MODDTRS-NHph-adj-hd

(12) [ ][ ]

+→

KK ,][1',DOM

INCID[1]DTRS-NH


(13)

+→

NEGDOMDTRS-NH

firstph-adj-neg-hd

(14) [ ]

[ ]

→

][2]'1[DOM

]2[DOMDTRS-NH

HEAD[1]DTR-HD

o

c

ph-cp


224

(15)

[ ][ ]

→

{[1]}SLASH|BIND-TO

}{[1],SLASH|INHERNONLOC

SUBJ

VFORMHEADCAT|LOC

SSDTR-HD

[1]LOC|SSDTRS-NH

K

fin

ph-filler-hd

(16)

[ ][ ]

[ ]

→

nelistfinverbthird

ph-filler-su-hd

[A][4]LOCSUBJ

[2]HEAD

[3]LOC]'1[DOM

[2]HEADDTR-HD

]1[DTRS-NH

oo

[4] ]3[ =∧

(17)

[ ][ ]

[ ]

∨∨

→

nelistfinverb second first first-pre

ph-filler-ns-hd

[A][4]LOCSUBJ

[2]HEAD

[3]LOC]'1[DOM

[2]HEADDTR-HD

]1[DTRS-NH

oo

[4] ]3[ ≠∧

Appendix B: Clause types

B.1 Type hierarchy

(1) finite-clause


root subordinate

interrogative rel sai

ns-sub su-sub

decl wh polar v1 v2 non-sai

v2-decl v2-wh v1-pol non-sai-wh non-sai-decl sub-pol sub-decl ns-sub-wh ns-rel su-sub-wh su-rel

negative- inversion


B.2 Constraints on clause types

(2)

[ ]

→

KK ,,DOM

VFORMHEAD

SUBJ

CAT

third

fin

vclause-finite

(3)

∨→KK

L

,[1]HEAD

,DOM

[1]HEAD|

fourth secondroot

(4)

[ ]

+

→KK

L

,AUX[1]HEAD

,DOM

[1]HEAD|

secondsai

(5) [ ][ ]K,DOM secondv1 →

(6) [ ]]2[]1[

,[2]LOCSUBJ

,]1[LOC

DOM

≠∧

→

Ksecondfirst

v2

(7) [ ] [ ][ ]elist

thirdfirstsai-non

=∧

⊕⊕⊕⊕→

]1[

]1[DOM LL

(8) [ ][ ] [ ] [ ]nesetnesetc

esubordinat

REL QUE HEAD [1]

,[1],DOM

∨∨=∧

→ K

(9)

¬

→ KK , HEAD

,DOMverb

secondesubordinat

(10) [ ][2] [1]

,]2[LOCSUBJ

,,]1[LOC

DOM

=∧

→

KKfourththird

sub-su

(11) [ ] [ ][2] [1]

,]2[LOCSUBJ

,, HEAD|CAT[1]LOC

DOM

≠∧

¬

∨

→KK

fourth

verb

second first-pre

sub-ns

(12) [ ]austinianedeclarativ CONT→

(13) [ ]questioniveinterrogat CONT→

(14)

→

neset

questionwh

PARAMSCONT


(15)

→

{}PARAMSCONT

questionpolar

(16) [ ][ ]

→

K,{[1]}RELDOM

NPMODHEAD [1]rel

(17)

∨→ KK ,

QUE,DOM

neset

first first-prewh-v2

(18)

→ KK ,

QUE,DOM

neset

thirdwh-sai-non

(19) { }

→ KK ,

QUE,DOMfirst

decl-v2

(20)

→K,

CONT

[1]HEADDOM

[1]HEAD

austinian

cdecl-sub

(21)

→K,

{}PARAMSCONT

]1[HEAD|CAT

DOM

[1]HEAD

question

cpol-sub

(22)

+

→ K,NEG

DOMfirst

inversion-negative

(23)

∨

→

K,[1]CONT|SS

DOM

[1]FOC|STR-INFO

[1]CONT|SS

[1]FOC|STR-INFO

first

inversion-negative

(24)

[ ]

→

{}STORE

,}{[1]STORE

,DOM

[1]QUANTS

KKquant-neg

firstinversion-negative

Appendix C: Constraints on order domains

(1) The type hierarchy for domain objects in English

dom-obj

incidental pre-first first second third fourth fifth


a. firstfirst p

b． secondsecond p

(3) The Topological Linear Precedence Constraint for English

fifthfourththirdsecondfirstfirst-pre ppppp

Appendix D: More on who and whom (Chapter 4)

D.1 Wilcock’s (1999) analysis of whom in pied-piping

The impossibility of who in pied-piping in (1) is due to the fact that the formal status of

pied-piping conflicts with the [REGSTR informal] specification of who, along the lines of

Wilcock (1999; cf. Paolillo 2000).

(1) a. To whom/*who are you referring?

b. someone on whom/*who we can rely

This appendix will summarise Wilcock’s (1999: 384ff) approach to this issue.

Wilcock (1999) notes systematic covariation between register and nonlocal features of

preposition. This is formalised as lexical constraints in which register restrictions are

associated with PP construction subtypes.

(2) a. { }{ }

{ }

→

formal

prep

prep-rel

REGSTR

SLASH

[1] REL

QUE

HEAD

b. { }{ }

{ }

→

formal

prep

prep-que

REGSTR

SLASH

REL

[1] QUE

HEAD

(2) requires prepositions with non-empty REL (2a) and non-empty QUE (2b) to have the

formal register. The combination of these lexical constraints with the Register

Amalgamation Constraint (3) provides an account for the distribution of who/whom in (1).

(3) Register Amalgamation Constraint (Wilcock 1999: 382)

[ ] [ ]

[1] REGSTR

[1] REGSTR,,[1] REGSTR ST-ARG

→

Lword


(3) is a lexical constraint that ensures the amalgamation of contextual information from a

word’s arguments.

(4) is the constituent structure for the filler PP of (1a).

(4)

{ }

formal REGSTR

[1] QUE

whomto, PHON

PP

{ }

{ }

formal

prep-que

[3] REGSTR

[1] QUE

[3] REGSTR

[1] QUE]2[ ST-ARG

to PHON

{ }

formal[3] REGSTR

[1] QUE

[1] INDEX

whom PHON

[2]NP

The SLASH Amalgamation Constraint requires that the non-empty QUE of whom should be

amalgamated into the QUE value of with. The preposition has thereby a non-empty QUE, so

constraint (2b) requires it to have the formal register. The Register Amalgamation

Constraint (3) requires the REGSTR value of the argument to be unified with that of the head.

This requirement is indeed satisfied here since whom is lexically specified as [REGSTR

formal].1

Let us turn to ungrammaticality of who in (1). The representation of the head of the

filler PP in (1a) is something like the following.

(5) * { }

{ }

formal

informal

prep-que

[3] REGSTR

[1] QUE

REGSTR

[1] QUE]2[ ST-ARG

to PHON

The SLASH Amalgamation Constraint requires the non-empty QUE of who to be

1 In order to ensure that a phrase inherits the REGSTR values of its daughters, Wilcock (1999: 377) introduces

the Contextual Head Inheritance Principle, which states that in a head-nexus-phrase and a head-adjunct-phrase

the phrase’s CONTEXT is by default token-identical to that of its contextual head daughter.


amalgamated into the QUE value of to, which is tagged [1] in (5). The preposition has

thereby a non-empty QUE, so constraint (2b) requires it to have the formal register. However,

the REGSTR value of who cannot be unified with that of with: informal and formal,

respectively. This is a violation of the Register Amalgamation Constraint (3).

D.2 Lasnik and Sobin’s (2000) approach to the who/whom distinction

Here we consider how another approach might accommodate the who/whom distinction. A

recent attempt to provide a theoretical account of the relevant facts is Lasnik and Sobin

(2000).2 They argue that who is the basic form of the wh-pronoun, which can check either

nominative (NOM) or accusative (ACC) case. The suffix -m of whom is assumed to be

associated with an additional ACC feature and has to be checked independently of the ACC

feature associated with the stem who. This additional ACC feature carried by the suffix is

checked by the rules with the status of ‘grammatical viruses’, characterised as

extra-grammatical devices, entirely independent of ordinary case marking mechanisms.

They serve to license prestige forms. Rule (6) licenses the occurrence of whom as object of a

verb or preposition, as in (1) and (7).

(6) The Basic ‘whom’ Rule (Lasnik and Sobin 2000: 354)

If: [V/P] who- -m

[ACC] [ACC]

1 2 3

then: check ACC on 3

(7) a. Who will marry whom/who?

b. Who is buying a gift for whom/who?

c. It was whom/who?3

2 See also Kayne (1984) and Radford (1988).

3 The copular verb be requires an accusative complement, except for the formulaic use of nominative as in It


Rule (8) licenses the occurrence of initial whom in any type of wh-construction where the

wh-pronoun functions as the object of a verb (9), stranded preposition (10), or the subject of an

embedded clause (11).

(8) The Extended ‘whom’ Rule (Lasnik and Sobin 2000: 359)

If: who- -m … NP, where

[ACC]

1 2 3

a) 3 is the nearest subject NP to 2, and

b) ‘…’ does not contain a V which has 1–2 (a single word whom) as its subject,

then: check ACC on 2.

(9) a. those whom we consulted.

b. Whom did you meet?

c. He didn’t say whom he had invited.

(10) a. someone whom we can rely on

b. Whom are you referring to?

(11) a. We feed children whom we think are hungry.

b. the man whom I believe has left.

c. the man whom it was believed had left.

The unacceptable occurrences of whom in (12) are ruled out by the fact that they are not

compatible with the sequential arrangement of (6) or (8).

(12) a. Who/*whom wrote the editorial?

b. the man who/*whom came to dinner

was I.

(i) a. In this picture, the person in the purple shorts is me/*I.

b. It was just us/*we

See Sobin (1997) and Lasnik and Sobin (2000) for details.


However, their approach involves some problems. First, it is not clear whether the

who/whom distinction should be treated as a matter of case. Two different forms of a lexeme

should not necessarily be seen as two different case forms. If they are not realisations of case,

it will not be necessary to assume that the stem who- and the affix -m have two different cases.

Other things being equal, it would be preferable not to have such a counter-intuitive

assumption.

Second, as Lasnik and Sobin (2000: 362) themselves note, (8) is fairly complex;

especially it includes the stipulations about 3 and about what can appear between 2 and 3. A

rule that is acquired in a special way may be more complex than an ordinary grammatical rule,

and, as they suggest (2000: 362), such complexity may be a reason for being a prestige usage.

Complexity, however, is a potential source of suspicion, and it is indeed suspicious in this case

since the stipulations included are questionable. First, it is not obvious how ‘the nearest

subject NP to 2’ is to be identified within Principles and Parameters assumptions. Next, their

analysis includes the stipulation about what can appear between 2 and 3: the V should be a

theta-role assigner and must not be an auxiliary verb. It is not clear why a theta-role

assigning ability is relevant here. Our HPSG analysis is clearly simpler which is free of any

questionable stipulations.

D.3 Further data of who and whom

We present some further data of the distribution of who and whom, which seems to be captured

within the general framework of HPSG. First, let us consider the following sentence.

(13) For who/whom?

This is an instance of so-called sluicing. It looks as if (13) were a fronted PP, like those in

such a pied-piping context as in (14), with the rest of the clause sluiced.

(14) For whom/*who do you do it?

Who is not allowed in such a context. Thus, (13) might be a potential problem to our analysis.


We consider, however, that PP-fronting is not involved in (13a). The following example

from Culicover (1999: 136) illustrates this point.

(15) This is the review of a book by someone important, but I can’t remember {by

whom/*by whom it is a review of a book}

(15) shows that it is possible to have sluicing without genuine PP-fronting. This fact

indicates that by whom which stands alone is not a fronted PP. Therefore, it will be possible

to analyse (13) independently of the constructions such as (14), so we will analyse (13a) along

the same lines as an ordinary prepositional phrase such as (7b): just another case of the

who/whom alternation as a distinction between a informal and formal register.

Second, we will turn to another sluicing example.

(16) Who/*whom for?

This is the case which Culicover (1999: 130ff) calls sluice-stranding. It looks as if (16) is

composed of fronted who and the preposition for as in the following example, with the

remainder of the sentence sliced.

(17) Who/whom do you do it for?

In such a normal preposition-stranding context as (17), both who and whom are possible.

Thus, (16) might be a potential problem to our approach if it were really a sluiced version of

(17). We should note the fact argued for by Culicover (1999) that sluice-stranding is not

fully general, but a highly idiosyncratic construction. Let us just look at two pieces of

evidence. First, the combinations of wh-words with prepositions in sluice-stranding contexts

are fixed to some extent. Culicover (1999: 131ff) notes that who allows sluice-stranding with

about, by, for, from, of, on, to and with and what allows sluice-stranding with at, about, for,

from, of, on and with. This will be highly likely if the construction itself is sui generis.

Second, consider the following examples from Culicover (1999: 136).

(18) a. This is the review of a book by someone important, but I can’t remember {who

by/*?who it is a review of a book by}.


b. He was writing a book about someone, but I couldn’t figure out {*who about/who

he was writing a book about}.

(18a) indicates that the normal preposition-stranding is ungrammatical whereas

sluice-stranding is grammatical when the relevant components are who and by. In the case of

who and about, as in (18b), however, sluice-stranding is ungrammatical while the normal

stranding is grammatical. This data shows that it will be difficult to relate these two types of

constructions. On the basis of these pieces of data, we assume that sluice-stranding is an

idiosyncratic construction.4 What we should note here is the fact that the ‘sluice-stranding

construction’ is informal in style.5 Therefore, it is not compatible with the formal style of

whom. The register specification of the construction is compatible to that of who. This is

why only who can occur in the sluice-stranding construction.

Finally, let us turn to the fact that whom is more acceptable in a less formal register in

embedded clauses than in main clauses. This is illustrated by the examples in (19), which are

not formal but not fully informal.

(19) a. Hugh wasn’t impressed with this ingratiating barman whom Roddy had raked up.

b. Award-winning journalist Nelson Keece (Gary Busey) is coldly detached from his

chosen subject, serial killer Stefan (Arnold Vosloo), whom he catches in the act of

murder.

To deal with these cases, we need to have a more fine-grained classification of register than

that given earlier. We assume that the type formal has two subtypes, fully-formal and

semi-formal. We can formulate the new classification in the following hierarchy.

(20) register

formal informal

fully-formal semi-formal

4 For more details, the readers are referred to Culicover (1999).

5 I would like to thank Bob Borsley for pointing out this fact to me.


There are thus three maximal subtypes for register: fully-formal, semi-formal and informal.

We introduce a further constraint in (21).

(21) [ ] [ ]formal-fully

wh-v2 REGSTR[1]

whomPHON[1]DTRS-NH→

(21) states that the REGSTR value of whom, specified as formal, is resolved to fully-formal if

it is the non-head daughter of clauses of the type v2-wh. This explains the considerably more

formal status of whom in main clauses. Since constraint (21) does not apply for [IC −]

clauses, the REGSTR value of whom remains formal in embedded clauses. This entails that

whom can appear in a semi-formal as well as fully-formal register, which accounts for the

occurrence of whom in less formal style sentences in (19). This contrast between main

clause interrogatives and relatives matches Huddleston and Pullum’s (2002: 465) description:

‘[t]he formal feel of whom is most apparent in main clause interrogatives’.

Bibliography

Abeillé, Anne, and Danièle Godard. 1997. The Syntax of French Negative Adverbs. In

Danielle Forget, Paul Hirschbühler, France Martineau, and María L. Rivero (eds.).

Negation and Polarity: Syntax and Semantics. Amsterdam: John Benjamins. 1–17.

Agbayani, Brian. 2000 Wh-subjects in English and the Vacuous Movement Hypothesis.

Linguistic Inquiry 31. 703–713.

Alexopoulou, Theodra,, and Dimitra Kolliakou. 2002. On linkhood, topicalization and clitic

left dislocation. Journal of Linguistics 38. 193−245.

Bach, Emmon. 1983. On the relationship between word-grammar and phrase-grammar.

Natural Language and Linguistic Theory. 65–89.

Baltin, Mark. 1982. A landing site for movement rules. Linguistic Inquiry 13. 1–38.

Bayer, Josef. 1984a. COMP in Bavarian syntax. The Linguistic Review 3. 209–274.

Bayer, Josef. 1984. Towards an explanation of certain that-t phenomena: The COMP-node in

Bavarian. In Wim de Geest and Yvan Putseys (eds.). Sentential Complementation.

Dordrecht: Foris Publications. 23–32.

Belleti, Adriana, and Rizzi, Luigi (eds.). 1996. Parameters and Functional Heads. Oxford:

Oxford University Press.

Birner, Betty, and Gregory Ward. 1994. Information Status and Noncanonical Word Order in

English. Amsterdam/Philadelphia: John Benjamins.

Blevins, James P. 1994. Derived constituent order in unbounded dependency constructions.

Journal of Linguistics 30. 349–409.

Bonami, Olivier, Danièle Godard, and Jean-Marie Marandin, 1999. French Subject Inversion

in Extraction Contexts. in Gosse Bouma, Erhard Hinrichs, Geert-Jan M. Kruijff, and

Richard Oehrle (eds.). Constraints and Resources in Natural Language Syntax and

Semantics. Stanford: CSLI publications. 21–40.

Bonami, Olivier, and Danièle Godard. 2003. Incidental adjuncts: an overlooked type of

adjunction. Handout given at the 10th International Conference on Head-Driven Phrase

Structure Grammar, Michigan State University.

Bibliography 238

Bonami, Olivier, and Danièle Godard, and Brigitte Kempers-Manhe. 2004. Adverb

classification. In Francis Corblin and Henriëtte de Swart (eds.). Handbook of French

Semantics. Stanford: CSLI Publications. 143–184.

Borsley, Robert D. 1999a. Weak auxiliaries, complex verbs, and inflected complementizers in

Polish. In R. D. Borsley and A. Przepiórkowski (eds.). 29–59.

Borsley, Robert D. 1999a. Mutation and constituent structure in Welsh. Lingua 109. 263-300.

Borsley, Robert D. 2003. Agreement, mutation and missing NPs in Welsh. Unpublished paper.

University of Essex. (http://privatewww.essex.ac.uk/~rborsley/Agreement-paper.pdf)

Borsley, Robert D. 2004. An Approach to English Comparative Correlatives. In Stefan Müller

(ed.). Proceedings of the HPSG04 Conference. Stanford: CSLI Publications.

Borsley, Robert D. 2005a. On the superficiality of Welsh agreement and related matters.

Unpublished paper. University of Essex.

(http://privatewww.essex.ac.uk/~rborsley/superficial-agreement-3-paper.pdf)

Borsley, Robert D. 2005b. A new approach to Welsh negation. Unpublished paper. University

of Essex. (http://privatewww.essex.ac.uk/~rborsley/new-negation.pdf)

Borsley, Robert D. and Kathol, Andreas. 2000. Breton as a V2 language. Linguistics 38–4.

665–710.

Borsley Robert D., and Przepiórkowski, Adam (eds.). 1999. Slavic in Head-Driven Phrase

Structure Grammar. Stanford: CSLI Publications.

Bouma, Gosse, Rob Malouf, and Ivan A. Sag. 2001. Satisfying Constraints on Extraction and

Adjunction. Natural Language and Linguistic Theory 19. 1–65.

Bröker, Norbert. 1998. Separating surface order and syntactic relations in a dependency

grammar. Proceedings of the 17th International Conference on Computational

Linguistics (COLING) and the 36th Annual meeting of the Association for

Computational Linguistics (ACL). Montreal. 174–180.

(http://www.aclweb.org/anthology/P98-1026)

Büring, Daniel. 2004. Negaitve Inversion. To appear in the Proceedings of NELS 2004.

Calcagno, Mike. 1993. Toward a linearization-based approach to word order variation to

Japanese. In Andreas Kathol and Carl Pollard (eds.). Papers in Linguistics, 26−45.

Bibliography 239

Campbell-Kibler, Kathryn. 2002. Bech’s Problem, Again: Using Linearization on Dutch

R-pronouns. In Frank Van Eynde and Lars Hellan and Dorothee Beermann (eds.). The

Proceedings of the 8th International Conference on Head-Driven Phrase Structure

Grammar. Stanford: CSLI Publications. 87–102.

Cann, Ronnie, Claire Grover, and Philip Miller (eds.). 2000. Grammatical Interfaces in HPSG.

Stanford: CSLI Publications.

Chomsky, Noam. 1971. Deep structure, surface structure, and semantic interpretation. In

Daniel Steinberg and Leon Jacobovits (eds.). Semantics. Cambridge: Cambridge

University Press. 183–216.

Chomsky, Noam. 1986. Barriers. Cambridge, MA: MIT Press.

Chomsky, Noam. 2004. Beyond explanatory adequacy. In Adriana Belletti (ed.). Structures

and Beyond. Oxford: Oxford University Press. 104–131.

Chomsky, Noam, and Lasnik, Howard. 1977. Filters and control. Linguistic Inquiry 8.

425–504.

Chung, Chan, and Kim, Jong-Bok. 2003. Capturing word order asymmetries in English

left-peripheral constructions: A domain-based approach. In Stefan Müller (ed.).

Proceedings of the 10th International Conference on Head-Driven Phrase Structure

Grammar. Stanford: CSLI Publications. 68–87.

Copestake, Ann, Dan Flickinger, Carl Pollard, and Ivan A. Sag. 1999. Minimal Recursion

Semantics: An Introduction. Stanford University and Ohio State University.

Cormack, Annabel, and Neil Smith. 2000. Fronting: the syntax and pragmatics of ‘focus’ and

‘topic’. UCL Working Papers 20. 387–417.

Crysmann, Berthold. 1999. Morphosyntactic Paradoxa in Fox. In Gosse Bouma and Erhard

Hinrichs and Geert-Jan Kruiff and Richard Oehrle (eds.). Constraints and Resources in

Natural Language Syntax and Semantics. Stanford: CSLI publications. 41–60.

Crysmann, Berthold. 2000a. Syntactic Transparency of Pronominal Affixes. In R. Cann, C.

Grover and P. Miller (eds.). 77–96.

Bibliography 240

Crysmann, Berthold. 2000b. Clitics and Coordination in Linear Structure. In Birgit Gerlach

and Janet Grijzenhout (eds.). Clitics in Phonology, Morphology, and Syntax. John

Benjamins. 121–159.

Crysmann, Berthold. 2002. Constraint-Based Coanalysis: Portuguese Cliticisation and

Morphology-Syntax Interaction in HPSG. Saarbrücken Dissertations in Computational

Linguistics and Language Technology, number 15. Saarbrücken: Deutsches

Forschungszentrum für Künstliche Intelligenz and Universität des Saarlandes.

Culicover, Peter W. 1991. Topicalization, inversion, and complementizers in English. Ms. The

Ohio State University.

Culicover, Peter W. 1999. Syntactic Nuts. Oxford: Oxford Univesity Press

Culicover, Peter W. 2000. Concrete Minimalism, branching structure, and linear order.

Proceedings of GLiP-2 (Generative Linguistics in Poland). Warsaw.

Culicover, Peter W. 2002. Branching and structure. In Bohumil Palek (ed.). Linguistics and

Phonetics 2000. Prague: Charles University Press.

Culicover and Nowak, Andrzej. 2003. Dynamical Grammar. Oxford: Oxford University Press.

Culicover, Peter W., and Ray Jackendoff. 2005. Simpler Syntax. Oxford: Oxford University

Press.

Curry, Haskell B. 1961. Some logical aspects of grammatical structure. Structure of Language

and its Mathematical Aspects: Proceedings of the Twelfth Symposium in Applied

Mathematics. Providence: American Mathematical Society. 56–68.

De Kuthy, Kordula. 2002. Discontinuous NPs in German − A Case Study of the Interaction of

Syntax, Semantics and Pragmatics. Stanford: CSLI Publications.

De Kuthy, Kordula, and Walt Detmar Meurers. 2003. The secret life of focus exponents, and

what it tells us about fronted verbal projections. In Stefan Müller (ed.). Proceedings of

the HPSG03 Conference. Stanford: CSLI Publications. 97–110.

Donohue, Cathryn, and Ivan A. Sag. 1999. Domains in Warlpiri. Ms. Stanford University.

Dowty, David R. 1996. Towards a minimalist theory of syntactic structure. Bunt and van

Horck. 1996. 11–62.

Drach, Erich. 1937. Grundgedanken der deutschen Satzlehre. Frankfurt: Diesterweg.

Bibliography 241

Engdahl, Elisabet. 1999. Integrating pragmatics into the grammar. In Lunella Mereu (ed.).

Boundaries of Morphology and Syntax. Amsterdam/Philadelphia: John Benjamins.

175–194.

Engdahl, Elisabet and Enric Vallduví. 1996. Information packaging in HPSG. In Claire

Grover. and Enric Vallduví (eds.). Edinburgh Working Papers in Cognitive Science, vol

12: Studies in HPSG. 1−31.

Ernst, Thomas. 2002. The Syntax of Adjuncts. Cambridge: Cambridge University Press.

Espinal, María Teresa. 1991. The representation of disjunct constituents. Language 67.

726–762.

Feldhaus, Anke. 1999. German wh-interrogatives in HPSG. In V. Kordoni, (ed.). 1999.

459–493.

Frey, Werner. 2005. Pragmatic properties of certain German and English left peripheral

constructions. Linguistics 43. 89–129.

Gazdar, Gerald, Ewan Klein, Geoffrey Pullum, and Ivan Sag. 1985. Generalized Phrase

Structure Grammar. Oxford: Basil Blackwell.

Gazdar, Gerald. 1981. Unbounded dependencies and coordinate structure. Linguistic Inquiry

12. 155–184.

Ginzburg, Jonathan and Sag, Ivan A. 2000. Interrogative Investigations. Stanford: CSLI

Publications.

Georgia M. Green and Jerry L. Morgan. 1996. Auxiliary Inversion and the notion of ‘default

specification’. Journal of Linguistics 32. 43–56.

Haegeman, Liliane. 1988. Parenthetical adverbials: the radical orphanage approach. In Chiba,

Shuki, Akira Ogawa, Yasuaki Fuiwara, Norio Yamada, Osamu Koma and Takao Yagi

(eds.). Aspects of Modern English linguistics: Papers Presented to Masatomo Ukaji on

his 60th Birthday. Tokyo: Kaitakusha. 232–54.

Haegeman, Liliane. 1995. The Syntax of Negation. Cambridge: Cambridge University Press.

Haegeman, Liliane. 2000a. Inversion, non-adjacent-inversion, and adjuncts in CP.

Transactions of the Philological Society 98. 121–160.

Haegeman, Liliane. 2000b. Negative preposing, negative inversion, and the split CP. In

Bibliography 242

Laurence R. Horn and Yasuhiko Kato 2000: Negation and Polarity. Syntactic and

Semantic Perspectives. Oxford: Oxford University Press. 21–61.

Haegeman, Liliane. 2002a. Anchoring to speaker, adverbial clauses and the structure of CP. In

Simon Mauck and Jenny Mittelstaedt (eds.). Georgetown University Working Papers in

Theoretical Linguistics. Volume 2. 117–180.

Haegeman, Liliane. 2002b. Sentence-medial NP-adjuncts in English. Nordic Journal of

Linguistics 25. 79–108.

Haegeman, Liliane. 2003a. Notes on long adverbial fronting in English and the left periphery.

Linguistic Inquiry 34: 640–649.

Haegeman, Liliane. 2003b. Speculations on adverbial fronting and the left periphery. In

Jacqueline Guéron and Liliane Tasmowski (eds.). Temps et Point de Vue/Tense and Point

of View. Université Paris X. 329–365.

Haegeman, Liliane. 2004a. Topicalization, CLLD and the left periphery. In Shaer et al. (eds.).

157–192.

Haegeman, Liliane. 2004b. The syntax of adverbial clauses and its consequences for

topicalisation. Martine Coene, Gretel De Cuyper and Yves D’Hulst (eds.). Antwerp

Papers in Linguistics 107: Current Studies in Comparative Romance Linguistics. 61–91.

Haegeman, Liliane, and Guéron, J. 1999. English Grammar. Oxford: Blackwell.

Haegeman, Liliane, and Zanuttini, R. 1991. Negative heads and the negative criterion. The

Linguistic Review 8. 233–251.

Hawkins, John A. 1986. A Comparative Typology of English and German. London: Croom

Helm.

Heinz, Wolfgang and Johannes Matiasek. 1994. Argument structure and case assignment in

German. In Nerbonne, J. et al. (eds.). 199–236.

Henry, Alison. 1995. Belfast English and Standard English: Dialect Variation and Parameter

Setting. New York/Oxford: Oxford University Press.

Hepple, Mark. 1994. Discontinuity and the lambek calculus. Proceedings of the 15th

Conferenceon Computational Linguistics (COLING-94). Kyoto.

Heycock, Caroline. 2005. Embedded root phenomena. In Martin Everaert, Henk Van

Bibliography 243

Riemsdijk, Rob Goedemans, Bart Hollebrandse (eds.). The Blackwell Companion to

Syntax. Vol. II. Oxford: Blackwell.

Holler, Anke. 2004. How the left-periphery of a wh-relative clause determines its syntactic and

semantic relationships. In B. Shaer, W. Frey and C. Maienborn (eds.). 143–260.

Holler, Anke. 2005. On non-canonical clause linkage. In Stefan Müller (ed.). Proceedings of

the HPSG05 Conference. Stanford: CSLI Publications.

Holmberg, Anders, and Christer Platzack. 1995. The Role of Inflection in Scandinavian Syntax.

New York/Oxford: Oxford University Press.

Höhle, Tilman. 1986. Der Begriff ‘Mittelfeld’, Anmerkungen über die Theorie der

topologischen Felder. In Albrecht Schöne (ed.). Kontroversen, alte und neue: Akten des

7. Internationalen Germanisten-Kongresses, Göttingen 1985, Band 3, Niemeyer,

Tübingen. 329–40.

Höhle, Tilman. 1992. Über Verum-Fokus im Deutschen. In Joachim Jacobs (ed.).

Informationsstructur und Grammatik, Linguistische Berichte, Sonderheft 4. Oplanden:

Westdeutscher Verlag. 112–141.

Hooper, Joan and Sandra Thompson. 1973. On the applicability of root transformations.

Linguistic Inquiry 4. 465–498.

Hornstein, Norbert. 1990. As Time Goes By. Cambridge, MA: MIT Press.

Huck, Geoffrey and Almerindo Ojeda (eds.). 1987. Discontinuous Constituency. New York:

Academic Press.

Huck, Geoffrey. 1985. Exclusivity and discontinuity in phrase structure grammar. West Coast

Conference on Formal Linguistics 4. 92–98. Stanford: CSLI Publications.

Huddleston, Rodney, and Geoffrey K. Pullum. 2002. The Cambridge Grammar of the English

Language. Cambridge: Cambridge University Press.

Hudson, Richard A. 1976. Arguments for a Non-transformational Grammar. University of

Chicago Press.

Hudson, Richard A. 1990. English Word Grammar. Oxford: Blackwell.

Hudson, Richard A. 1995. HPSG without PS?

(http://www.phon.ucl.ac.uk/home/dick/unpub.htm.)

Bibliography 244

Hudson, Richard A. 1999. Discontinuity. (http://www.phon.ucl.ac.uk/home /dick/discont.htm.)

Hudson, Richard A. 2003a. An Encyclopedia of English Grammar and Word Grammar.

(http://www.phon.ucl.ac.uk/home/dick/enc-gen.htm.)

Hudson, Richard A. 2003b. Trouble on the left periphery. Lingua 113. 607−642.

Hukari, Thomas E. and Levine, Robert D. 1995. Adjunct extraction. Journal of Linguistics 31.

195–226.

Hukari, Thomas E. and Levine, Robert D. 1996. Phrase structure grammar: the next generation.

Journal of Linguistics 32. 465–496.

Jackendoff, Ray. 1972. Semantic Interpretation in Generative Grammar. Cambridge, MA:

MIT Press.

Jaeger, T. Florian. 2003. Topics First! In- and outside of Bulgarian wh-interrogatives. In Stefan

Müller (ed.). Proceedings of the 10th International Conference on Head-Driven Phrase

Structure Grammar. Stanford: CSLI Publications. 181–202.

Jensen, Per Anker and Peter Skadhauge. 2001. Linearization and diathetic alternations in

Danish. In W. D. Meurers and T. Kiss (eds.). 111–140.

Jespersen, Otto. 1924. The Philosophy of Grammar. Chicago: University of Chicago Press.

Jespersen, Otto. 1927. A Modern English Grammar on Historical Principles, Part III. London:

Allen and Unwin.

Kathol, Andreas. 1995. Linearization and coordination in German. In A. Kathol and C. J.

Pollard (eds.). 117–151.

Kathol, Andreas. 1995. Linearization-Based German Syntax. PhD thesis, Ohio State

University.

Kathol, Andreas. 2000. Linear Syntax. Oxford: Oxford University Press.

Kathol, Andreas. 2001. Positional effects in a monostratal grammar of German. Journal of

Linguistics 37. 35–66.

Kathol, Andreas. 2002. Linearization-based approach to inversion and verb-second

phenomena in English. In Proceedings of the 2002 LSK International Summer

Conference Volume II: Workshops on Complex Predicates, Inversion, and OT Phonology.

223–234.

Bibliography 245

Kathol, Andreas and Robert D. Levine. 1992. Inversion as a linearization effect. In Amy

Schafer (ed.). Proceedings of the North East Linguistic Society 23. Department of

Linguistics, University of Massachusetts: GLSA. 207–221.

Kathol, Andreas, and Carl Pollard. 1995a. Extraposition via complex domain formation. In

33rd Annual Meeting of the ACL, 174–180. San Francisco: Morgan Kaufmann.

Kathol, Andreas and Carl J. Pollard. 1995b. On the left periphery of German subordinate

clauses. In Proceedings of the Fourteenth West Coast Conference on Formal Linguistics.

Stanford University: CSLI Publications/SLA.

Kathol, Andreas and Carl J. Pollard (eds.). 1995. Papers in Syntax. OSU Working Papers in

Linguistics 42. Ohio State University: Department of Linguistics.

Kayne, Richard S. 1984. Connectedness and Binary Branching. Dordrecht: Foris Publications.

Kayne, Richard S. 1994. The Antisymmetry of Syntax. Cambridge, MA: MIT Press.

Kim, Jong-Bok and Ivan A. Sag, 2002. Negation without head-movement. Natural Language

and Linguistic Theory 20. 339–412,.

Klima, Edward S. 1964. Negation in English. In Fodor Jerry A. and Jerrold J. Katz (eds.). The

Structure of Language. Englewood Cliffs, NJ: Prentice-Hall. 246–323.

Koizumi, Masatoshi. 1995. Phrase Structure in Minimalist Syntax. PhD. thesis, MIT.

Kordoni, Valia (ed.). 1999. Tübingen Studies in Head-Driven Phrase Structure Grammar.

Arbeitsberichte des SFB 340, number 132. Tübingen: Universität Tübingen.

Kroch, Anthony S. and Aravind K. Joshi. 1987. Analyzing extraposition in a tree adjoining

grammar. In G. J. Huck and A. E. Ojeda, (eds.).

Kupść, Anna. 1999. Haplology of the Polish Reflexive Marker. In R. D. Borsley and A.

Przepiórkowski (eds.). 91–124.

Lasnik, Howard. and Nicholas Sobin. 2000. The who/whom puzzle: on the preservation of an

archaic feature. Natural Language and Linguistic Theory 18. 343–371.

Lee, Sun-Hee. 2001. Argument composition and linearization: Korean complex predicates and

scrambling. OSU Working Papers in Linguistics 56. 53–78

Levine, Robert D. and Thomas E. Hukari. 2006. The Unity of Unbounded Dependency

Constructions. Stanford: CSLI Publications.

Bibliography 246

Liberman, Mark. 1974. On conditioning the rule of Subj.-Aux. inversion. Ellen Kaisse and

Jorge Hankamer (eds.). Papers from the Fifth Anual Meeting of North Eastern Linguistic

Society. 77–91.

Maekawa, Takafumi. 2004. Constituency, word order and focus projection. In Stefan Müller

(ed.). Proceedings of the 11th International Conference on Head-Driven Phrase

Structure Grammar. Stanford: CSLI Publications. 168–188.

Maekawa, Takafumi. 2005a. WG surface structures and HPSG order domains. In Kensei

Sugayama, and Richard Hudson (eds.). Word Grammar: New Perspectives on a Theory

of Language Structure. London: Continuum. 145–67.

Maekawa, Takafumi. 2005b. An HPSG approach to the who/whom puzzle. In Stefan Müller

(ed.). Proceedings of the HPSG-2005 Conference. Stanford: CSLI Publications.

290–310.

Maekawa, Takafumi. 2006a. Long and short adjunct fronting in HPSG. In Stefan Müller (ed.).

Proceedings of the HPSG-2006 Conference. Stanford: CSLI Publications. 212–227.

Maekawa, Takafumi. 2006b. Configurational and linearization-based approaches to negative

inversion. In Olivier Bonami and Patricia Cabredo Hofherr (eds.). Empirical Issues in

Syntax and Semantics 6. 227–247.

Mazzon, Gabriella. 2004. A History of English Negation. Harlow: Longman.

McCawley, James D. 1988. Parentheticals and Discontinuous Constituent Structure. Linguistic

Inquiry 13. 91–106.

Merchant, Jason. 2003. Subject-Auxiliary inversion in comparatives and PF output constraints.

In Kerstin Schwabe and Susanne Winkler (eds.). The interfaces: Deriving and

Interpreting Omitted Structures. Amsterdam/Philadelphia: John Benjamins. 55–77.

Meurers, Walt Detmar. 2000. Raising spirits (and assigning them case). Groninger Arbeiten

zur Germanistischen Linguistik 43, 173–226.

Meurers, Walt Detmar, and Tibor Kiss (eds.). 2001. Constraint-Based Approaches to

Germanic Syntax. Stanford: CSLI Publications.

Morrill, Glynn V. 1995. Discontinuity in categorial grammar. Linguistics and Philosophy 18.

175–219.

Bibliography 247

Müller, Stefan. 1995. Scrambling in German – Extraction into the Mittelfeld. In Proceedings of

the tenth Pacific Asia Conference on Language, Information and Computation. City

University of Hong Kong. 79–83.

Müller, Stefan. 1997. Yet another paper about partial verb phrase fronting in German.

Research Report RR-97-07. Deutsches Forschungszentrum für Künstliche Intelligenz,

Saarbrücken.

Müller, Stefan. 2002. Complex Predicates: Verbal Complexes, Resultative Constructions, and

Particle Verbs in German. Stanford: CSLI Publications.

Müller, Stefan. 2004. Continuous and discontinuous constituents? A comparison between

syntactic analyses for constituent order and their processing systems. Research on

Language and Computation 2. 209−257.

Müller, Stefan. 2006. Elliptical constructions, multiple frontings, and surface-based syntax. In

Gerhard Jäger, Paola Monachesi, Gerald Penn and Shuly Winter (eds.). Proceedings of

Formal Grammar 2004. Stanford: CSLI Publications.

Nerbonne, John, Klaus Netter, Carl J. Pollard (eds.). 1994. German in Head-Driven Phrase

Structure Grammar. Stanford: CSLI Publications.

van Noord, Gertjan and Bouma, Gosse. 1994. Adjuncts and the processing of lexical rules.

In Fifteenth International Conference on Computational Linguistics (COLING ’94).

250–256.

Ojeda, Almerindo. 1987. Discontinuity, multidominances and unbounded dependency in

generalized phrase structure grammar. In G. J. Huck and A. E. Ojeda, (eds.). 257–182.

Paolillo, John. 2000. Formalizing formality: an analysis of register variation in Sinhala.

Journal of Linguistics 36, 215–259.

Penn, Gerald. 1999. Linearization and WH-extraction in HPSG. In R. D. Borsley and A.

Przepiórkowski (eds.). 149–182.

Pesetusky, David and Torrego, Esther. 2001. T-to-C movement: causes and consequences. In

Michael Kenstowics (ed.). Ken Hale, a life in language. Cambridge, MA: MIT Press.

355–426.

Pollard, Carl. 1994. Toward a unified account of passive in German. In Nerbonne, J. et al.

Bibliography 248

(eds.). 273–296.

Pollard, Carl, Kasper, Robert and Levine, Robert. 1994. Studies in constituent ordering:

Towards a theory of linearization in Head-driven Phrase Structure Grammar. Research

Proposal to the National Science Foundation, Ohio State University.

Pollard, Carl, and Sag, Ivan A. 1989. Information-Based Syntax and Semantics. Stanford:

CSLI Publications.

Pollard, Carl, and Sag, Ivan A. 1994. Head-Driven Phrase Structure Grammar. Chicago:

University of Chicago Press.

Postal, Paul. 1998. Three Investigations of Extraction. Cambridge, MA: MIT Press.

Przepiórkowski, Adam. 1999a. On complements and Adjuncts in Polish. In R. D. Borsley and

A. Przepiórkowski (eds.). 183–210.

Przepiórkowski, Adam. 1999b. On case assignment and ‘adjuncts as complements’. In

Webelhuth et al. (ed.). 1999. 231–245.

Pullum, Geoffrey. 1982. Free word order and phrase structure rules. Proceedings of the North

Eastern Linguistic Society 12. 209–220.

Rambow, Owen and Aravind Joshi. 1994. A formal look at dependency grammars and

phrase-structure grammars, with special consideration of word-order phenomena. In L.

Wanner (ed.). Current Issues in Meaning-Text-Theory. London: Pinter.

(http://arxiv.org/abs/cmp-lg/9410007.)

Randolf, Quirk, Sidney Greenbaum, Geoffrey Leech, and Jan Svartvik. 1985. A

Comprehensive Grammar of the English Language. London: Longman.

Radford, Andrew. 1988. Transformational Grammar. Cambridge: Cambridge University

Press.

Reape, Michael. 1994. Domain union and word order variation in German. In John Nerbonne,

Klaus Netter, Carl J. Pollard (eds.). German in Head-Driven Phrase Structure Grammar.

Stanford: CSLI Publications. 151−98.

Reape, Michael. 1996. Getting things in order. In Harry Bunt and Arthur van Horck (eds.).

Discontinuous Constituency. Berlin/New York: Mouton de Gruyter, 209–253.

Bibliography 249

Reinhart, Tanya. 1995. Interface Strategies. OTS Working Papers in Linguistics. University of

Utrecht.

Reis, Marga. 1985. Satzeinleitende Structuren im Deutschen. Über COMP, Haupt-, und

Nebensätze, w-Bewegung und die Doppelkopfanalyse. In Werner Abraham (ed.).

Erklärende Syntax des Deutschen. Tübingen: Gunter Narr. 271–311.

Richter, Frank, and Manfred Sailer. 1999. Lexicalizing the left periphery of German finite

sentences. In V. Kordoni (ed.). 116–154.

Richter, Frank, and Manfred Sailer. 2001. On the left periphery of German. In W. D. Meurers

and T. Kiss (eds.). 257–300.

Rizzi, Luigi. 1996. Residual V-second and wh-criterion. In A. Belleti and L. Rizzi (eds.).

63–90.

Rizzi, Luigi. 1997. On the fine structure of the left-periphery. In Liliane Haegeman (ed.).

Elements of Grammar. Dordrecht: Kluwer Academic Publishers. 281-337.

Rizzi, Luigi. 2001. Relativized Minimality. In Mark Baltin and Chris Collins (eds.). The

Handbook of Contemporary Syntactic Theory. Oxford: Blackwell. 89–110.

Rizzi, Luigi and Ian Roberts. 1996. Complex inversion in French. In A. Beletti and L. Rizzi

(eds.). 91–116.

Roberts, Ian and Anna Roussou. 2002. The EPP as a condition on the tense dependency. In

Peter Svenonius (ed.). Subjects, Expletives, and the EPP. Oxford: Oxford University

Press. 125–155.

Rochemont, Michael. S. 1986. Focus in Generative Grammar. Amsterdam/Philadelphia: John

Benjamins.

Rudanko, Juhani. 1982. Towards a description of negatively conditioned subject operator

inversion in English. English Studies 63. 348–359.

Ross, John R. 1986. Infinite Syntax! New Jersey: Ablex Publishing Corporation.

Sag, Ivan A. 1997. English relative clause constructions. Journal of Linguistics 33. 431−484.

Selkirk, Elizabeth. 1984. Phonology and Syntax. Cambridge, MA: MIT Press.

Selkirk, Elizabeth. 1995. Sentence prosody: Intonation, stress, and phrasing. In J. A.

Goldsmith (ed.). The Handbook of Phonological Theory. London: Basil Blackwell.

Bibliography 250

550−569

Shaer, Benjamin 2003a ‘Manner’ adverbs and the “association” theory: Some problems and

solutions. In: E. Lang, C. Maienborn and C. Fabricius-Hansen (eds.). Modifying

Adjuncts. Berlin: Mouton de Gruyter. 211–260.

Shaer, Benjamin. 2003b. An ‘orphan’ analysis of long and short adjunct movement in English.

In Gina Garding and Mimu Tsujimura (eds.). Proceedings of the West Coast Conference

on Formal Linguistics 22. 450–463.

Shaer, Benjamin. 2004a. Left/right contrasts among English temporal adverbials. In Jennifer R.

Austin, Stefan Engelberg, and Gisa Rauh (eds.). Adverbials. The interplay between

meaning, context, and syntactic structure. Amsterdam: Benjamins. 289–332.

Shaer, Benjamin. 2004b. On the syntactic status of certain fronted adverbials in English. In

Pawel M. Nowak, Corey Yoquelet, and David Mortensen (eds.). Proceedings of the

Berkeley Linguistics Society 29, 379-390.

Shaer, Benjamin and Frey, Werner. 2004 ‘Integrated’ and ‘non-Integrated’ left-peripheral

elements in German and English’. In B. Shaer, W. Frey and C. Maienborn (eds.). 465–

502.

Shaer, Benjamin, Frey, Werner and Maienborn, Claudia (eds.). 2004. Proceedings of the

Dislocated Elements Workshop, ZAS Berlin, November 2003 (2 Volumes). ZAS Papers in

Linguistics 35.

Sobin, Nicholas. 1997. Agreement, default rules, and grammatical viruses. Linguistic Inquiry

28. 318–343.

Sobin, Nicholas. 2003. Negative inversion as nonmovement. Syntax 6. 183–212.

Swan, Michael. 1995. Practical English Usage. Oxford: Oxford University Press.

de Swart, Henriette, and Sag, Ivan A. 2002. Negation and negative concord in Romance.

Linguistics and Philosophy 25. 373–417.

Tonhauser, Judith. 2001. An approach to polarity sensitivity and negative concord by lexical

underspecification. In Dan Flickinger and Andreas Kathol (eds.). Proceedings of the

7th International HPSG Conference. Stanford: CSLI Publications. 285–304.

Tsimpli, Dimitra-Irini. 1990. The clause structure and word order of Modern Greek. UCL

Bibliography 251

Working Papers in Linguistics 2. 226–258.

Vikner, Sten. 1995. Verb Movement and Expletive Subjects in the Germanic Languages. New

York: Oxford University Press.

Ward, Gregory, Betty Birner, and Rodney Huddleston. 2002. Information packaging. In R.

Huddleston and G. K. Pullum. 2002. Chapter 16. 1363–1447.

Webelhuth, Gert, Koenig, Jean-Pierre, and Kathol, Andreas (eds.). 1999. Lexical and

Constructional Aspects of Linguistic Explanation, Stanford: CSLI Publications.

Wilcock, Graham. 1999. Lexicalization of context. In G. Webelhuth et al. (eds.). 373–387.

Yatabe, Shûichi. 1996. Long-distance scrambling via partial compaction. In Masatoshi

Koizumi, Masayuki Oishi and Uli Sauerland (eds.). Formal Approaches to Japanese

Linguistics 2 (MIT Working Papers in Linguistics 29). Cambridge, MA, 303−317.

Yatabe, Shûichi. 2001. The syntax and semantics of left-node raising in Japanese. In Dan

Flickinger and Andreas Kathol (eds.). Proceedings of the 7th International Conference

on Head-Driven Phrase Structure Grammar. Stanford: CSLI Publications. 325−344.

Yatabe, Shûichi. 2003. Does scrambling in Japanese obey the Coordinate Structure Constraint?

Proceedings of the 126th Meeting of the Linguistic Society of Japan. Linguistic Society

of Japan, Kyoto. 262−267.

Yoo, Eun-Jung. 1993. Subcategorization and case marking in Korean. In A. Kathol and C. J.

Pollard (eds.). 178–198.

Yoshimoto, Kei. 2000. A bistratal approach to the prosody-syntax interface in Japanese. In

Ronnie Cann, Claire Grover, and Phillip Miller (eds.). Grammatical Interfaces in

Head-Driven Phrase Structure Grammar. Stanford: CSLI Publications. 267–282.

Yoshimoto, Kei. 2003. A linear approach to multiple clause embedding. In Jong-Bok Kim and

Stephen Wechsler (eds.). The Proceedings of the 9th International Conference on

Head-Driven Phrase Structure Grammar. Stanford: CSLI Publications. 439–458.

Zwicky, Arnold. 1986. Concatentation and liberation. Papers from the 22nd Regional Meeting,

Chicago Linguistic Society. Chicago: Chicago Linguistic Society. 65–74.

The English Left Periphery in Linearisation-based …...4.2 Previous analyses 64 4.2.1...

Documents

Transcript of The English Left Periphery in Linearisation-based …...4.2 Previous analyses 64 4.2.1...