syntactic issues in the english imperative

INFORMATION TO USERS

This manuscript has been reproduced from the microfilm master. UMI

films the text directly from the original or copy submitted. Thus, some

thesis and dissertation copies are in typewriter face, while others may be

from any type of computer printer.

The quality of this reproduction is dependent upon the quality of the

copy submitted. Broken or indistinct print, colored or poor quality

illustrations and photographs, print bleedthrough, substandard margins,

and improper alignment can adversely afreet reproduction.

In the unlikely event that the author did not send UMI a complete

manuscript and there are missing pages, these will be noted. Also, if

unauthorized copyright material had to be removed, a note will indicate

the deletion.

Oversize materials (e.g., maps, drawings, charts) are reproduced by

sectioning the original, beginning at the upper left-hand comer and

continuing from left to right in equal sections with small overlaps. Each

original is also photographed in one exposure and is included in reduced

form at the back of the book.

Photographs included in the original manuscript have been reproduced

xerographically in this copy. Higher quality 6” x 9” black and white

photographic prints are available for any photographs or illustrations

appearing in this copy for an additional charge. Contact UMI directly to

order.

UMIA Bell & Howell Information Company

300 North Zed) Road, Ann Arbor MI 48106-1346 USA 313/761-4700 800/521-0600

R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission.

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.

UNIVERSITY OF CALIFORNIA

SANTA CRUZ

SYNTACTIC ISSUES IN THE ENGLISH IMPERATIVE

A dissertation submitted in partial satisfaction of the requirements for the degree of

DOCTOR OF PHILOSOPHY

in

LINGUISTICS

by

Eric Potsdam

December 1996

The dissertation of Eric Potsdam is approved:

[dim Aissen, Chair

JarxiesMcCIoskey )

Donka Farkas

Dean of Graduate Studies


UMI Number: 9714966

UMI Microform 9714966 Copyright 1997, by UMI Company. All rights reserved.

This microform edition is protected against unauthorized copying under Title 17, United States Code.

UMI300 North Zeeb Road Ann Arbor, MI 48103


V o l . 1


Copyright © by

Eric Potsdam

1996


Abstract

Syntactic Issues in the English Imperative

Eric Potsdam B.S., Tufts University

M.S., Stanford University Directed by: Professor Judith Aissen

This dissertation investigates central syntactic issues in the English imperative clause type

and argues that the imperative has largely regular syntactic behavior within a conventional

conception of English clause structure. The work focuses on dominant places in the

grammar where it has been argued that imperative syntax irreducibly differs from that of

more-thoroughly-analyzed finite clauses. Cumulatively, the resolution of these issues

provides a structural analysis of the English imperative.

Chapter 2 investigates the contrasting behavior of auxiliaries in imperatives versus finite

clauses. The differences are argued to follow from the empirically-supported claim that the

two are in different structural positions. Imperative auxiliaries, like finite main verbs but

unlike finite auxiliaries, never raise from V° to 1°. There is thus a principled and unified

explanation for the contrasts.

Chapter 3 investigates imperative subjects in English. It argues that noun phrases such as

everybody in Everybody jump! are syntactic subjects with canonical subject properties. In

particular, they are not vocatives. The null imperative subject is also analyzed and asserted

to be structurally represented by a null pronominal.


Chapters 4 and 5 investigate the phrase structure of non-neutral imperatives with don’t and

emphatic do. A wide range of evidence from VP ellipsis, syntactic and semantic parallels

with finite clauses, adverb placement, floating quantifiers, negative preposing, and

topicalization converges on the result that do and don't are inflectional heads that are base

generated in 1° and optionally undergo inversion to C \ The analytical behavior is thus

argued to assimilate to that seen in superficially similar finite interrogatives.

Chapter 6 examines formal imperatives with do not which do differ from corresponding

finite interrogatives, in word order and subject options. The chapter demonstrates that the

structural differences result from distinct patterns of T-to-C0 inversion. While interrogative

inversion is obligatory and syntactically driven, imperative inversion is semantically

motivated, occurring only with negation and emphasis. The contrasting behavior is

analyzed using the core mechanisms of checking theory from the Minimalist Program. The

chapter concludes with a discussion of the syntactic basis for distinguishing clause types in

English.


V

Table of Contents

Acknowledgm ents........................................................................................................viii

In troduction................................................................................................................. 11. The Domain of Investigation................................................................................ 1

1.1. The Imperative Clause Type.................................................................... 21.2. Summary of the Data....................................................................... 4

2. Overview of the Dissertation................................................................................ 10

Chapter 1. Clausal Syntax...................................................................................... 151. Introduction.......................................................................................................... 152. English Clause Structure...................................................................................... 15

2.1. Checking Theory and Movement............................................................. 193. Adverbs.............................................................................................................. 21

3.1. Adverb Placement: Jackendoff 1972......................................................... 223.2. A Syntactic Proposal................................................................................ 26

3.2.1. Syntactic Consequences.............................................................. 293.3. Adverb Licensing............................................................................... 31

3.3.1. IP-Adjoined Adverbs.................................................................. 353.3.2. Licensing from C°....................................................................... 38

4. Sentential Negation....................................................................................... 465. VP Ellipsis........................................................................................................... 49

5.1. Lobeck 1995............................................................................................ 515.2. A Licensing Condition............................................................................. 69

5.2.1. The Licenser............................................................................... 695.2.2. The Structural Relationship........................................................ 83

5.3. Confirmation of the Licensing Condition......................................... 885.4. Conclusions............................................................................................. 93

6. Floating Quantifiers...............................................................................................101

Chapter 2. Verb Raising........................................................................................... 1061. Introduction........................................................................................................... 1062. Background...........................................................................................................1093. Diagnostics............................................................................................................ 112

3.1. Adverb Placement..................................................................................... 1133.2. Negation....................................................................................................1153.3. VP Ellipsis................................................................................................1173.4. Summary.................................................................................................. 137

4. Imperatives............................................................................................................ 1384.1. Adverb Placement..................................................................................... 1394.2. Negation....................................................................................................1424.3. VP Ellipsis................................................................................................143

5. Subjunctive Clauses....................................................................................... 1505.1. Adverb Placement..................................................................................... 1525.2. Negation....................................................................................................154


5.2.1. Negation in Subjunctive Clauses..................................................1545.2.2. Sentential versus Constituent Negation........................................ 1565.2.3. Negation and Verb Raising.................................................. 162

5.3. VP Ellipsis..................................................................................................1656. Conclusion............................................................................................................. 171

C hapter 3. Im perative Subjects.........................................................................1751. Introduction............................................................................................................1752. Discourse Roles..................................................................................................... 1783. Vocatives................................................................................................................185

3.1. Distinguishing Vocatives and Imperative Noun Phrases............................1853.1.1. Intonation...................................................................................1873.1.2. Referent...................................................................................... 188

3.1.2.1. quantiflcational and indefinite vocatives................. 1943.1.2.2. INPs and the addressee..............................................197

3.1.3. Anaphora...................................................................................... 1993.2. Conclusions................................................................................................201

4. Imperative Subjects............................................................................................... 2044.1. Syntactic Status.......................................................................................... 205

4.1.1. Beukema and Coopmans 1989.................................................... 2214.2. Semantic Restrictions.................................................................................225

4.2.1. Restrictions in Form.................................................................... 2264.2.2. Discourse Roles...........................................................................2304.2.3. The Addressee and the Imperative Subject...................................236

4.3. Summary................................................................................................... 2455. The Null Imperative Subject..................................................................................246

5.1. Interpretation of the Null Subject............................................................... 2475.2. Identification of pro ................................................................................... 257

5.2.1. The Role of [PERSON] in the Identification of pro...................2625.2.2. The Identification of Imperative pro ..................................... 267

5.3. Consequences......................................................................................... 269Appendix. Patterns of Pronominal Agreement............................................................ 272

Chapter 4. The Status of Imperative D o(n't) ..................................................... 2831. Introduction...........................................................................................................2832. Two Hypotheses...................................................................................................285

2.1. The Auxiliary Analysis..............................................................................2852.2. The Imperative Particle Analysis...............................................................287

3. Evidence................................................................................................................ 2883.1. VP Ellipsis................................................................................................ 2893.2. Let ’ -Constructions...................................................................................2943.3. Syntactic and Semantic Symmetry............................................................ 303

3.3.1. Syntactic Parallels: Imperative Subject Position...........................3033.3.2. Semantic Parallels: Emphatic do...................................................311

4. Closing Arguments......................................................................................... 3194.1. Zhang 1990............................................................................................... 3194.2. Cohen 1976............................................................................................... 3234.3. Summary...................................................................................................327

Appendix. Imperative-Like Constructions with Let.................................................... 327


Chapter 5. The Phrase Structure of Imperatives............................................... 3361. Introduction........................................................................................................... 3362. The Hypotheses.................................................................................................... 338

2.1. The VP-Internal Subject Analysis............................................................. 3392.2. The CP Analysis................................................................................3412.3. The Articulated Infl Analysis........................................................... 342

3. Evidence Against the VP Hypothesis.................................................................... 3433.1. Multiple Auxiliaries...................................................................................3453.2. Floating Quantifiers.................................................................................. 3483.3. VP Ellipsis................................................................................................3513.4. Adverb Placement..................................................................................... 353

4. The CP/FP Hypotheses................................................................................. 3574.1. Topicalization............................................................................................3594.2. Inversion-Triggering Adverbials.............................................................. 370

5. Summary............................................................................................................... 387

Chapter 6. The Status of Imperative Do N o t ......................................................3911. Introduction...........................................................................................................3912. The Problematic Data............................................................................................394

2.1. Schmerling 1977...................................................................................... 3962.2. Akmajian 1984....................................................................................4012.3. Beukema and Coopmans 1989.................................................................406

3. A Structural Resolution......................................................................................... 4093.1. A Subject Position............................................................................ 4103.2. The Sequence do not.................................................................................420

3.2.1. A Diagnostic for the Position of d o ..............................................4213.2.2. The Position of do in Formal Imperatives.................................... 427

4. Imperative I°-to-C°................................................................................................4314.1. A Characterization of Imperative Inversion.............................................. 4314.2. An Analysis..............................................................................................434

5. Conclusions.......................................................................................................... 439

References.................................................................................................................... 443


“Minds are like parachutes, they only function when open.”

Santa Cruz bumper sticker

“Minds are like parachutes, using them only slows you down.”

anonymous

Acknowledgments

Eventually you just have to stop thinking and write, so it all starts here where I would like

to briefly express my thanks to those who have made my time in Santa Cruz so

tremendous. I feel very lucky to have been given the opportunity to get my introduction to

linguistics at UCSC and to learn from its uniformly superb faculty. I would like to thank

them for their top-notch teaching examples, high standards of scholarship, honest

encouragement, and the numerous opportunities they have provided me. In particular, the

syntax community: Judith Aissen, Sandy Chung, Jorge Hankamer, and Jim McCloskey,

have most directly and, it is hoped, positively influenced my thinking and teaching.

This dissertation owes much to its committee. Judith Aissen could not have been a better

advisor and committee chair. Everything that passed by her was scrutinized and critiqued

with great care and thoroughness and interactions inevitably pushed things towards clarity,

rigor, and better organization—all qualities that I value highly. At other points, her

consistent, solid advice throughout my career as a graduate student was also invaluable.

Jim McCloskey’s insights into linguistic theory, ability to see implications for the big


picture, and encyclopedic knowledge of the literature are beyond admiration. He has helped

me to push conclusions to a level of maximum generality and to see the significance, and

insignificance, of many of the results arrived at in development of this work. I owe Donka

Farkas for her guidance in understanding some of the semantic issues surrounding

imperatives. I appreciated her patience as I grappled with the central but unfortunately

largely neglected semantic questions and their relation to the syntax. In the ideal world,

more of her advice would have been incorporated. More generally, I thank the entire Santa

Cruz linguistic community for numerous discussions, constructive previewing of material,

and countless grammaticality judgments in areas where I no longer seem to have any. I

would also like to use this opportunity to recognize previous professors and mentors, none

of them linguists, who have had a great influence on my intellectual development: A. B.

Perlman (Mechanical Engineering, Tufts University), Lee Edelman (English, Tufts

University), Thomas Kane (Applied Mechanics, Stanford University), and John

Hodgkinson (Douglas Aircraft Company).

This dissertation would not have come together if it weren’t for my last housemate, Carol,

who provided a near perfect environment for its creation. I thank her for keeping me very

well-humored and wonderfully well-fed during the final two years of my studies and for

sharing with me a family that I am completely embarrassed to say that I know, and enjoy.

My other adopted family, the Olneys, has provided me with countless weekends of busy

fun—if shoveling snow can be said to be fun—when I surely would otherwise have been

idle and unproductive. My own family I owe infinitely much for their unwavering support

and confidence in whatever pursuits I have chosen over the years. There’s nothing, except

for maybe buying another ‘66 Mustang, that they ever said I couldn’t do. So, I’ve never

known what was supposed to be impossible and, consequently, I’ve never known not to


X

try it. With regard to linguistics, I am grateful for their rarely asking what I’m doing but at

the same time approving of it, whatever it is exactly.

A dissertation isn’t the best place for ones best friends. They know who they are, how

much I have relied on them, and how much they are with me wherever I am. Hopefully

they will never read this work—there are too many other things to do together and, as

always, too little time.

And thanks to all the trees . . . . Here come just a few of them.


1

Introduction

This dissertation is a close investigation of central syntactic issues in the En2 argues that the

imperative has largely regular syntactic behavior within a conventional conception of

English clause structure. The claims of this work explicitly contrast with early generative

treatments which ascribed exceptional structure to the imperative, with concomitant

complications in the grammar.

This introduction provides an overview of the syntactic investigation to follow. Section 1

defines the empirical domain under study: the imperative is taken to be a

morphosyntactically-identifiable clause type that is the canonical realization of a directive

speech act. Defined in this way, a clear body of data is delimited. These data are then

previewed in advance of their analysis in chapters 1 through 6. Section 2 gives an overview

of the contents of these chapters. For each chapter, it summarizes the primary issue to be

investigated, its proposed analytical account, and the consequences for imperative syntax

and its relation to general English clause structure.

1. The Domain of Investigation

This section discusses the body of data that will be investigated in the dissertation and how

that body of data forms a coherent target of inquiry. Section 1.1 develops what is meant

syntactically by the term imperative while section 1.2 surveys the data that fall into this

domain.


2

1 .1 . The Imperative Clause Type

The range of sentences that have been called imperatives in the literature is astoundingly

large. The reason for this variety is that imperative is often defined in semantic terms.

Under a semantic characterization, any construction with the relevant illocutionary force or

pragmatic effect may be classified as an imperative, irrespective of its form. Such a

description is fundamentally one based on linguistic use and it defines an imperative as a

SPEECH ACT TYPE (see the philosophical literature including Austin 1962 and Searle

1969). For example, such an approach potentially labels all of the examples in (1) as

imperatives because they all have approximately the same force in a certain situation. I will

refer to this speech act type as DIRECTIVE but will not attempt to explicate it (but see chapter

3, section 4 as well as Katz and Postal 1964, Broadie 1972, Stockwell et. al. 1973,

Downes 1977, Schmerling 1982, Davies 1986a, and Hamblin 1987).

(1) a. Close the window!

b . You should close the window!

c. Will you close the window, please?

d. The window ought to be closed, John.

While such a classification is useful in describing language’s communicative potential and

actual use, it does not make any claims about the syntactic structure of such sentences.

Typically, however, there is one particular sentence form which is considered the canonical

syntactic realization of a given semantic clause type and this allows us to relate a speech act

type to the syntax. I will define an IMPERATIVE to be the syntactic sentence form which is

the canonical realization of a directive speech act type. In addition to being unified by a

canonical use, imperatives also share core syntactic or morphological properties. (2) is a set

of criteria for the English imperative, adapted from Davies 1986a:7, which I will use to


3

further characterize the English imperative clause type. Under this view, the imperative is a

SYNTACTIC CLAUSE TYPE. See Davies 1986a:chapter 1 for considerable discussion.

(2) Characteristics of the English Imperative Clause Type

a. lack of tense inflection

b. optionality of subject

Given (2) and the more extensive discussion in Davies 1986a, an imperative is

characterized by the highest verbal head or auxiliary being uninflected for tense.

Additionally, the subject may be absent without apparent syntactic consequences

elsewhere. Summarily, the object of investigation in this work is the imperative as defined

in (3).

(3) Imperative Syntactic Clause Type

An imperative is the canonical realization of a directive, with the

morphosyntactic properties in (2)

The difference between speech act type and syntactic clause type can be further exemplified

by considering questions, which will also be of importance in this study. All of the

examples in (4) are questions, as a speech act type, because they function as a request for

information.

(4) a. Is he here yet?

b. Who will Jack stay with?

c. He’s here already?

d. Jack will stay with WHO?


4

Within this semantic domain, we can identify a syntactic clause type, the INTERROGATIVE,

(4)a and b, which is semantically a question but is unique among questions in having

certain particular syntactic properties which distinguish them: they license negative polarity

items in certain circumstances and require subject-auxiliary inversion, for example.

Questions are thus a speech act type whose canonical realization is the syntactic clause type

interrogative. Interrogatives themselves are in turn a subtype of finite clause.

It should be noted that this characterization of the imperative is specific to English. While

speech act types are likely to be universal under the assumption that all languages are put to

roughly the same uses, no such claim is made about syntactic clause types, or their

relationships to speech act types. Further partitions in both speech act types and clause

types will yield different mappings between form and function, which are not under study

here.

1.2 . Summary of the Data

Crucial to the investigation of the imperative here is the recognition of a wide range of

grammatical imperative data—wider in fact than traditionally assumed. The data are

surveyed below. Much of the data and many of the empirical observations which the

dissertation explores and analyzes are originally due to Davies’ (1986a) thorough and

systematic exploration of the English imperative.

Ordinarily, the English imperative consists only of a verb phrase, in the non-technical

sense: a verb in its bare form plus internal arguments and adjuncts, if any. The subject is

usually understood to be second person you but is not overt. I will call examples as in (5)

SUBJECTLESS IMPERATIVES, taking subject here in a non-theoretical sense to mean a


5

phonetically overt noun phrase that is canonically interpreted as the agent or actor of the

imperative verb.

(5) a. Hoist the sails!

b. Mind your business!

c. Be happy!

English is typologically unusual in also allowing many kinds of overt subjects in

imperatives. The term SUBJECT IMPERATIVE describes the examples in (6) in which there is

an overt noun phrase filling the role of agent or actor.

(6) a. You take out the trash!

b. Everyone take out a pencil!

c. One of you fetch me that ladder!

d. Someone get me an aspirin!

e. Those in the front back away from the barricade!

Besides the expected you in (6)a, quantified phrases, second person partitives, indefinite

pronouns, and demonstratives are also regularly found, as in (6)b through e. Chapter 3

discusses the syntactic status of these noun phrases and the apparent semantic restrictions

that hold of them.

Subjects also appear in emphatic and negative imperatives, which I will collectively call

NON-NEUTRAL IMPERATIVES. Syntactically these are formed by appending subjectless

imperatives with do and don't, respectively, as in (7) and (8).

(7) a. Do be more careful!

b . Do try some of the dandelion salad!


6

(8) a. Don’t be so selfish!

b. Don’t forget my birthday!

A claim often made is that subjects are rather restricted in emphatic and negative imperatives

(Bolinger 1967; Stockwell et. al. 1973; Schmerling 1977,1982; Akmajian 1984; Beukema

and Coopmans 1989; Zhang 1990). Schmerling 1977 and Stockwell et. al. 1973, for

example, take the position that they may never occur with do, citing the ungrammatically

of examples like those in (9), in particular with you as subject.

(9) a. (*)Do YOU/SOMEBODY sit down!

b. (*)SOMEBODYdo open the window!

In negative imperatives, the standard generalization, implicitly assumed in Stockwell et. al.

1973 among others, is that a subject is permitted and it follows don’t, as in (10). Examples

with the subject preceding don't, as in declarative sentences, are not considered.

(10) a. Don’t EVERY one leave!

b . Don’t you give me any lip!

It is such limited observations about the position of imperative subjects that typically force

restrictive analyses of the imperative in which they are viewed as an idiosyncratic

construction and their syntax is divorced from standard assumptions about English clause

structure (Schmerling 1977, 1982; Culicover 1976). In an importantly thorough

investigation of the English imperative, Davies 1986a re-evaluates the data regarding

subject possibilities with do and don’t. The work considers the full range of subject plus

do(n’t) combinations and concludes that both orderings are permitted. Instances of each are

given in (11) through (14).


7

(11) a. Do SOMEone help him quickly! DOASUBJECT

b . Do AT LEAST SOME of you give it a try!

(12) a. SOMEone do answer the phone! SUBJECTAD0

b . Those with children do bring them along!

(13) a. Don’t anyone touch my stuff! DCWTASUBJECT

b . Don’t you misbehave while we’ re gone!

(14) a. Those with luggage don’t leave it unattended! SUBJECTADCWT

b. Girls go into the hall, BOYS don’t move!

Furthermore, Davies 1986a argues for the acceptability of do even with you, a combination

which most researchers consider completely impossible.1*2

(15) a. For heaven’s sake, of all people, DO YOU give me some support!

b . Do AT LEAST YOU have a go, even if the others won’t!

D avies I986a:90 suggests a functional explanation for the general oddness of do with you as subject. In the acceptable examples, you is not being used as a marker to signal authority over the addressee, as is typical, but, rather, functions contrastively. It contrasts the addressee to certain other people. Adverbs like at least which improve the acceptability of the examples make this contrast explicit. Do not occurring with you is a consequence of the rarity of situations in which both persuasive do and contrastive you are warranted.2Such examples were clearly acceptable in recent stages of English. Poutsma 1928:107 cites numerous text examples:

(i) a.b.

c.

d.

Lucy, do you watch! (Sheridan, The Rivals, Act I, scene 2, 1775)Do you give me a minute’s calm attention without looking at Rick!

(Dickens, Bleak House, chapter XXIV, 1853) Well, do you go, at any rate . . . . Do you go down.

(Trollope, Castle Richmond, chapter VI, 1860) He must want a secretary. He would be shy at an offer from me. Do you hint it if you get a chance. (Meredith, Ormont, chapter m , 1894)


8

For the present, I simply assert the acceptability of the above data. These data will be more

fully investigated, particularly with respect to subject options, in later chapters.3

Negative imperatives may also be formed with uncontracted do not, (16). I will call such

examples FORMAL IMPERATIVES.

(16) a. Do not pass up these deals!

b. Do not walk on the grass!

c. Do not forget!

It is universally claimed that subjects are impossible in formal imperatives. Chapter 6 more

carefully investigates this claim and argues that, although subjects are typically degraded in

this environment, there are clear contrasts depending upon where the subject is placed. If

the subject is before or after do not, the data are marginal, or inconsistently acceptable, (17)

and (18). Only when the subject occurs between do and not are the data robustly

ungrammatical, (19).

(17) a. I know I’ve done wrong but I can’t survive on my own.

?Oh please, SOMEbody do not desert me!

b . ?SOMEone do not abandon the gate! The fight is not yet lost and we must

maintain the security.

3The two subject positions with respect to do(n’t) are not simply free variants of one another. Davies 1986a notes that the difference between the two subject position options, before or after do(n’t), tends to correlate with the reading on the subject. The order do(n’t)^subject places contrastive focus on the subject while the order subjectAdofn’rJ is neutral. I leave this issue for future investigation. General restrictions in imperative subjects are discussed in chapter 3 and seem to be operative in all of these examples. The important claim to be relied upon in this work is that both word order patterns are acceptable.


9

(18) a. ?Do NOT ANY of you touch that cake! It’s for the wedding and if any of it is

missing heads will roll,

b . ?DO not YOU, of all people, insult me in this heinous and base manner!


*Do somebody not desert me!

b . *Do someone not abandon the gate!

Lastly, I will have occasion to appeal to imperatives with the auxiliaries perfective have and

progressive/passive be in investigating imperative clause structure. It is sometimes claimed

that imperatives with auxiliaries are ungrammatical. Akmajian, Steele, and Wasow 1979

and Takezawa 1984, for example, do not accept imperatives with perfective have in (20).

For most, however, such examples are acceptable.4 (21) illustrates imperatives with

progressive and passive be. Davies explicitly argues that there are in general no syntactic

constraints against auxiliaries in imperatives and I accept the conclusion.

(20) a. Please have read at least the introduction for Monday’s meeting!

b . Have waited at least an hour before going swimming!

c . Do have reached a decision regarding the matter!

4Bolinger 1967 convincingly argues that they are grammatical, citing acceptable examples along with an experiment in which a passage containing a perfective have imperative was submitted to twelve native English speakers. Only three people noted the imperative and none marked it as un-English. Davies 1986a: 16 makes the same point.


10

(21) a. Be paying attention when the instructor looks up!

b . Be waiting for me at the comer!

c. Be tested, just to be safe!

d. Don’t be going so soon!

e. Don’t be fooled by his phony credentials!

2. Overview of the Dissertation

The dissertation is organized around various central syntactic issues: particular places in the

grammar where it has been argued that 1) imperative syntax irreducibly differs from that of

better studied finite clauses and 2) ad hoc syntactic mechanisms are thus required. Each

chapter addresses roughly one outstanding point about imperative syntax, as described

below. The following briefly describes the contents and conclusions of the individual

chapters.

Chapter 1 lays out the syntactic framework in which the analyses are embedded,

approximately the Principles and Parameters Theory of the early 1990’s including the

innovation of checking theory from the recent Minimalist Program. It also develops specific

analyses of adverb placement, verb phrase ellipsis, negation, and floating quantifiers which

are central to much of the clause structure investigations and argumentation in later

chapters.

Chapter 2 investigates the position of verbal elements, auxiliaries and main verbs, in

imperatives and other English clauses. Current syntactic theory posits that verbal elements

typically appear either within the verb phrase or higher in an inflectional projection. This

latter position is often modeled derivationally as V°-to-r movement of the verbal head. In

English tensed clauses, only finite auxiliaries undergo V-to-I* verb raising. This in turn is


11

typically taken as the reason why they, unlike main verbs, do not require or permit support

do in the presence of affirmation and negation, (22).

As is well-known, auxiliaries in imperatives contrast with finite auxiliaries in requiring do

in the same environments. The pattern of grammaticality for imperatives in (23) is the

reverse from that of auxiliaries in (22)a and matches that of main verbs in (22)b.

This differing behavior of auxiliaries in finite clauses, (22)a, versus imperatives, (23), is

sometimes taken as an indication that imperatives irreducibly differ from finite clauses: they

require do even with the auxiliaries have and be. Chapter 2 argues that the difference has an

explanation in terms of verb raising. Diagnostics for verb raising using adverb placement,

negation, and VP ellipsis are developed and used to demonstrate that auxiliaries in

imperatives, like finite main verbs, do not undergo V0-to-I°. Consequently, the parallel

behavior is not unexpected. The mechanisms that ensure that do will appear with negation

and affirmation in finite clauses with verbs that do not move into 1° can then potentially be

used to achieve the same result in imperatives. There is little reason to allow imperative

clause structure to differ from standard conceptions, at least in this domain.

Chapter 3 investigates the status of the italicized noun phrases in (24).


(22)

♦cooked not did not cook

(23) f *Be not 1 . . , a- jDo not be}late!

b' {D “ not!hS}elIeftusamesstocleanuP!

12


b . Those in front move away from the barricade!

c. Don’t anyone get in my way!

d. Somebody do help him, he’s drowning!

Although such noun phrases appear to be subjects, numerous works attempt to restrict

them in motivated, although exceptional, ways, resulting in their being typically non

subject-like. The chapter argues that they are subjects, and in particular not vocatives. This

is a syntactically unsurprising claim since ordinary clause structure provides a structural

subject position and typically requires that it be filled. The syntactic and semantic behavior

of imperative subjects is thus argued not to be an exceptional feature of the English

imperative but to follow from the way an imperative is used.

Chapter 4 initiates the investigation of imperative phrase structure by considering the status

of the two formatives do and don't in emphatic and negative imperatives, (25) and (26)

respectively.

(25) a. Do have some more tea!

d. Do at least YOU try to understand where I’m coming from!

(26) a. Don't be bullied into signing!

d . Don’t you mention it further!

The chapter considers whether these elements are best analyzed as imperative particles or

inflectional heads, as their transparent similarity to finite forms might suggest. It is argued

based on patterns of VP ellipsis, word order, and syntactic parallels with related

let ’ -constructions that an analysis in which they are integrated into clausal syntax as

inflectional heads is superior. They behave like the corresponding finite forms and can be


13

profitably analyzed with existing mechanisms that are independently needed for finite

clause structure. This conclusion provides further support for the claim that imperatives

have largely predictable syntax. The elements do and don't fit expectedly into general

English clause structure.

Chapter 5 turns to the phrase structure of the imperatives in (25) and (26) and investigates

the structural position of the subject noun phrase and the heads do and don’t. It considers

and rejects—based on evidence from VP ellipsis, adverb placement, floating quantifiers,

and imperatives with auxiliaries—a hypothesis in which imperative subjects are VP-

intemal. Instead, the chapter concludes, based on a close investigation of negative

preposing and topicalization, that the examples have a syntactic structure parallel to that of

superficially similar, more thoroughly investigated, and syntactically unexceptional yes/no

questions. The chapter thus claims that imperatives appeal to the well-established process

of I°-to-C° movement seen elsewhere in English.

Chapter 6 concludes with an investigation of the syntax of formal imperatives, repeated in

(27).


b . Do not walk on the grass!

c. Do not forget!

These examples and similar ones have typically been at the center of debates over the

syntactic analysis of imperatives. They display interesting restrictions on subject

possibilities and a general syntactic inertness that has made their assimilation to general

English clause structure particularly difficult. In contrast to the conclusions of chapter 5,

they are a place in the grammar where imperatives and otherwise similar interrogatives


14

seemingly clearly differ. In particular, the formal imperatives in (28) with the subject

intervening between do and not are robustly ungrammatical in comparison to interrogatives

with this word order, (29), which are fully acceptable.

(28) a. *Do you not desert me!


(29) a. Do you not like artichokes?

b . Does somebody not have a pitchfork that I could borrow?

Despite this difference, the chapter nevertheless argues that the syntax of formal

imperatives is largely regular. The above contrast is real but is the reflex of patterned and

systematic difference between I°-to-C° in the two clause types. While interrogative

inversion in (29) is obligatory and triggered by a purely syntactic requirement, imperative

inversion is semantically driven and is in fact not present in the corresponding examples,

(28). Imperative inversion, the chapter argues, occurs with emphatic do and don’t, as

concluded in chapter 5, but not with pleonastic do, in (28). The examples are thus

ungrammatical because the head movement that would derive the word order is

unmotivated. Thus, while imperatives are different from interrogatives in this domain, the

chapter shows that the difference has a principled analysis within the theory of clause

structure. Consequently, even in this domain, imperative syntax is unexceptional and

analyzable within a conventional model of clausal structure—the thesis throughout the

dissertation.


15

Chapter 1

Clausal Syntax

1. Introduction

This chapter presents the theoretical preliminaries for the investigation of the syntax of the

English imperative which is to follow. The analysis of the imperative is embedded in a

theory of syntax roughly corresponding to the Principles and Parameters Theory of the

early 1990’s. In addition, I import the innovation of checking theory from the Minimalist

Program (Chomsky 1993, 1994, 1995). To the greatest extent possible, however, I attempt

to formulate the results of this work in a theory-neutral way. Section 2 presents the main

theoretical assumptions about English clause structure. Sections 3 through 6 investigate

narrower domains of English syntax: adverb placement, negation, VP ellipsis, and floating

quantifiers. These phenomena are central to much of the syntactic argumentation in later

chapters.

2 . English Clause Structure

The X phrase structure schema in (l)a and b along with the stipulation that English is left-

headed and has specifiers on the left allows only for phrase structures as in (l)c.


16

( 1) X Principles

a. XP-» Spec X

b. X’ -> X YP

Following numerous researchers in taking I(nfl) to be the head of a sentence, I assume that

it too is projected according to (l)c. The specifier of DP is the canonical position for the

subject in English and its presence is motivated by something like the Extended Projection

Principle (Chomsky 1981) or a strong D-feature of inflection (Chomsky 1993). This yields

(2) as the scheme for an English clause.1

(2)

subject I’

f ^ ^ V PI

V’

V (complement)

In addition to structures that obey the X schema, I allow for the following adjunction

structures. Heads may adjoin to other heads under head movement, yielding a structure as

in (3) (order of adjunction aside). With Chomsky 1995,1 assume that all head movement is

of this sort, although later chapters will not always show the internal details of this

operation.

1 In this structure and what follows, subjects are shown in spec,IP. This should not be taken as a claim about their base position. A VP-intemal base position (The VP-intemal subject hypothesis of Kuroda 1988, Koopman and Sportiche 1991) is compatible with all the analyses presented but explicitly adopting one does not benefit the exposition so I leave it out of the representations. In future phrase markers I will also omit the X’ level when not relevant.


Maximal projections may adjoin to intermediate levels, X’, and other maximal projections,

XP, but are restricted by the Adjunction Prohibition in (4) (Chomsky 1986; McCloskey

1992, 1996a). Informally, the Adjunction Prohibition disallows adjunction to arguments of

a lexical head, phrases that are s(emantically)-selected. It permits adjunction to

complements of functional projections, which are not s-selected.

(4) Adjunction Prohibition (McCloskey 1992:11)

Adjunction to a phrase which is s-selected by a lexical head is prohibited

Yes/no or polarity questions in English, whose word order shows inversion of the subject

and the first auxiliary or modal in the main clause, require additional structure above IP to

house the inverted auxiliary. The inversion which relates interrogatives with their

declarative counterparts is an example of the head movement and adjunction in (3). It is

movement of 1° to the head of a higher projection, CP. Interrogative formation is thus I°-

to-C° (Koopman 1984, Chomsky 1986). (5) schematizes the analysis.

(5) Interrogative Inversion in English

Regarding the specific lexical content of the verbal and inflectional projections, I follow

Chomsky 1957, Emonds 1976, Lobeck 1987 and numerous others in taking the modals,


CP

CA ^

C Ij subject

IP

I’

VP

18

must, can, should, etc., to be lexically specified as being of category 1°. Also of category 1°

are forms of support do, which I take to be directly inserted into 1° when needed. Accepting

a long tradition dating back to Ross 1969, I take the auxiliaries have and be to be

syntactically of category V°. Categorially, they are the same as main verbs such as walk,

seem, or rain. When the auxiliaries are finite, they obligatorily undergo V°-to-I° head

movement; main verbs do not undergo V°-to-r (Jackendoff 1972, Emonds 1976, Pollock

1989, Chomsky 1991, Lightfoot and Homstein 1994b, Rohrbacher 1994, and many

others). This is investigated more thoroughly in chapter 2. These lexical specifications are

summarized in (6).

(6) Categorial Specification of English Verbal Elements

1°: modals, support do

V°: have, be, main verbs

Under these assumptions, the structure of a sentence containing a modal and multiple

auxiliaries is (7).

(7) a. Zack might have been waiting

DP

Zack I

might V

have V

waiting


19

2 .1 . Checking Theory and Movement

Chomsky 1993, 1995 and the Minimalist Program give a very prominent position to the

lexicon and morphosyntactic features which are encoded there (see Marantz 1995, Lasnik

1995b for summary). The lexicon, in feeding syntactic structure, supplies only fully

inflected elements—for example verbs which are fully specified for person, number, and

tense. This information—inflectional, Case, and purely formal syntactic properties—in

addition to being reflected morphologically on the lexical items, is represented by features

on the head. These features place an important role in the syntax because, for a derivation

to succeed, they must be ‘verified’ as being appropriate for the structure in which they were

inserted. This verification is called FEATURE CHECKING and it must occur within a defined,

local configuration, in a position that depends upon the particular feature being checked. In

particular, feature checking takes place within the CHECKING DOMAIN of the head whose

features are being checked. On a non-technical level, phrases in the checking domain of a

head H° are a subset of those dominated by HP, as illustrated in (8). The checking domain

of H°, which is boldfaced, includes phrases that are in the specifier of HP (ZP), adjoined

within HP (XP), adjoined to the complement of H° (QP), or adjoined to H itself (J).

(8) HP

U P ^ ^ H P

Z P ^ ^ H ’

X P ^ ^ H ’

YP

YP

The relevant definitions and applications to (8), from Chomsky 1993:15-17 and Chomsky

1995:177-178,299,326 are as follows: First a dominates P if and only iff every segment of

J H QP


20

a dominates (3 (Chomsky 1986). The domain of a head a is the set of categories dominated

by the least full-category maximal projection dominating a that are distinct from and do not

contain a . In (8), the domain of H° is J, ZP, XP, YP, QP, and everything that they

dominate. The complement domain of a is the subset of the domain reflexively dominated

by the complement of a. The complement domain of H° is the lower YP and everything it

dominates. This does not include QP since not every segment of YP dominates it. The

residue is the domain less the complement domain. The residue in (8) is J, ZP, XP, QP,

everything they dominate, and the upper YP segment. The minimal residue, or checking

domain, is the smallest subset K of the residue such that for any T in the residue, some p in

K reflexively dominates T. The checking domain of H° is J, ZP, XP, and QP. These

definitions incorporate conclusions from chapter 4 of Chomsky 1995:326, in which the

checking domain excludes phrases adjoined to the maximal projection of the head.2

Checking can, in principle occur at any time in a derivation; however, not all features have

the same status regarding checking. If a feature is STRONG, then it is required to be checked

overtly, before the sentence is pronounced at P(honological) F(orm). If the feature is

w e a k , its checking will necessarily be delayed, by the principle of Procrastinate (Chomsky

1995:198), until the covert syntax, after the derivation branches to the level of L(ogical)

F(orm). All features must ultimately be checked because they otherwise cause the

derivation to crash. Strong features must additionally be checked before PF because they

are illicit PF objects. Feature checking serves to drive many, if not all, movement

operations since all strong features must move to be checked in accordance with (9)

2Phrases adjoined to the specifier of the head are also in the checking domain as are phrases adjoined to the specifier of the specifier of the head and phrases adjoined to this specifier, etc. Since this is clearly a shortcoming with the definition of checking domain, I ignore these positions.


21

(Chomsky 1995:232): a categorial feature, a feature associated with a certain category (i.e.

V, D, N, etc.), moves to check a feature of a functional head.

(9) If a feature F is strong, then F is a feature of a nonsubstantive category and

F is checked by a categorial feature.

Examples of feature checking include subjects raising to spec,IP (checking of the strong D-

feature of 1° by the subject DP) and V°-to-r raising (checking of a strong V-feature of 1° by

a verb). I°-to-C° movement illustrated in (5) is taken to be checking of some strong

categorial feature of C° by a modal or auxiliary.

3 . Adverbs

In this section, I offer an analysis of the syntax of English adverbs.3 The primary aim of

the discussion is to provide an explicit system in which to investigate adverb placement as a

reliable probe on English syntax. Section 3.1 presents observations from Jackendoff 1972

regarding possible adverb positions in English finite sentences. In section 3.2, these

generalizations are adapted to the more recent syntactic structures under consideration and I

propose an explicit description of the distribution of adverbs in English clauses. Section

3.3 attempts to provide an analytical basis for the distributional patterns. The analysis

proposes that adverbs are licensed by lexical heads in a local domain of the head and this

3The analyses that follow are only concerned with the syntactic category adverb— words that typically end in -ly—except where noted. I leave open the possibility that the analysis could be extended to adverbials more generally. While such a move seems ultimately desirable, other kinds of adverbials, for example prepositional phrases, noun phrases, and clausal adjuncts do not always have the same distribution as roughly synonymous adverb phrases. This is particularly true for clause-internal positions. Jackendoff 1972:94 points out the similarity between sentence-level adverbs and prepositional phrases that function like sentence adverbs. He shows that adverb/prepositional phrase pairs such as probably/in all probability have the same syntactic distribution. This is not the case, however, for many VP-level adverb phrases and PPs such as the pair soon/at 6:00.


22

licensing accounts for their realized structural positions. Although the relative position of

adverbs and verbal elements is most often used as an indicator of verb raising (e.g. Pollock

1989), such accounts are rarely explicit as to what assumptions allow the diagnostic to

succeed. By developing and defending a reasonable hypothesis about adverb placement in

English, this section gives support to the use of adverbs as a reliable probe for verb

movement, subject position, and other issues involving the interaction of word order and

syntactic structure.

3.1 . Adverb Placement: Jackendoff 1972

Jackendoff 1972 develops a classification of adverbs in English based on their positional

distribution in a clause. Except for a small class of typically non -ly adverbs, all adverbs in

English can appear in what is termed auxiliary position—a position that is somewhere

between the subject and main verb, neither clause-initial nor clause-final. According to

Jackendoff s typology, these adverbs can potentially have two interpretations: a manner

interpretation or a sentence interpretation. Regarding adverbs with the latter, they are

propositional modifiers (McConnell-Ginet 1982). Manner adverbs, in contrast, function as

modifiers of predicates. Ernst 1984 analyzes traditional manner adverbs as contributing the

meaning “there is something about the action, state, etc. denoted by the verb that can be

described by the adverb.” In contrast to sentential adverbs, which modify a situation,

manner adverbs modify an attribute of a verb. In Jackendoff* s syntactic analysis, manner

adverbs are at the VP level while sentence adverbs modify S. For this reason, I will use the

terms VP-ADVERB and S-ADVERB to label the two classes. Examples of each are given in

(10) and (11), respectively (Jackendoff s (3.12) and (3.7)). Some adverbs are ambiguous

between the two interpretations, behaving as either a VP-adverb or an S-adverb. (12)a

illustrates the ambiguity (Jackendoff s (3.1)); it has either of the meanings in (12)b and c.


23

(10) Stanley

(11) Horatio

completelyeasily

handily

evidentlyprobablycertainly

apparently

ate his Wheaties.

lost his mind.

VP-ADVERB

S-ADVERB

BOTH INTERPRETATIONS(12) a. John jciumsUy} droPPed CUP of coffee.

{clever 1ciumsyj •l°lin t0 drop his cup of coffee.

{clever 1 clumsy/'

The positional distribution in finite clauses that Jackendoff proposes for the two adverb

classes is illustrated in (13) through (20). Most of the examples are taken from Jackendoff

1972:chapter 3. S-adverbs may appear clause-initially, (13), immediately following the

subject, (14), or to the immediate right of a modal or finite4 auxiliary verb, (15).

(13) a. Horatio has lost his mind.

b. Max was climbing the walls of the garden.

a. Horatio j p ^ ^ / y j has l°st his mind.

b. Max j / ^ p ”/yj was climbing the walls of the garden.c. George probably will lose his mind.

d. George j ^ ^ J y J ate up the cabbage.

(14)

4Throughout, I use finite to mean bearing tense, where the tenses in English range over past and non-past. Under this conception, only indicative clauses are finite since only they contain tensed forms, o f which the modals are assumed to be instances (Gazdar, Pullum, and Sag 1982).


24

(15) a. George will probably lose his mind.

b. Horatio has jprofofe/y} *ost m“1£-

c. George was j U robaM y^\ rt^nec by tornado.

S-adverbs may not appear to the right of a non-finite auxiliary, (16).

(16) a. *George has been ru“iec by tornado,

b. *George is being ^pparently\ tailed by ra L

VP-adverbs, in contrast, may appear clause-finally, in (17), or to the left of the main verb,

in (18). This latter position is independent of what might precede the adverb; any

combination of modal and auxiliaries is permissible, as illustrated.

(17) a. Stanley ate his Wheaties j co^ ^ yj.

b . Bob scaled the wall J •(18) a. Stanley ate his Wheaties.

b. George was mined by the tornado.

c. George has read tbe *300*c-

d. George might have tost his mind.

e. Peter could have been rescued.

VP-adverbs are not permitted clause-initially, (19), or to the left of modals or auxiliaries, in

(20).

(19) * Stonley ate his Wheaties.


25

(20, a. -George {“ «

b- w {c 7 f f i .

was ruined by the tornado,

has read the book.

c. *George might ^°^!tirely^\ ^ave *ost mind.

d. *Peter could | ej^ ^ ^ y /y J have been rescued.

e. *John will have rapidly been beaten by Bill.

These positional distributions are summarized in (21) and (22).

(21) Positional Distribution ofS-Adverbs

a. clause-initial

b . immediately following the subject

c . to the immediate right of a modal or finite auxiliary5

5Emst 1984 argues that S-adverb interpretations are not restricted as Jackendoff 1972 claims and that, in addition to the above, they may also be found after multiple auxiliaries. Some of his examples in (i) do seem acceptable. They may be compared with the minimally different examples in (ii) in which the adverbs are moved farther left to a more usual position. It is unclear why examples such as (i) are so infrequent or difficult to create. In general, S-adverbs do not sound grammatical when they are as far right as in (i)— as Jackendoff originally maintained and I will idealize here.

(i) a. The library staff has been stupidly filing all those articles on Raising in the child psychologysection.

b. Egbert might have been cleverly abducted to get an interview with the guerrillas.c. Millicent must have been very clearly disqualified by the RUSSIAN judge, or else the

committee would not have protested so hard.d. She may have actually been only aiming for the vice-president’s job.

(ii) a. The library staff has stupidly been filing all those articles on Raising in the child psychologysection.

b. Egbert cleverly might have been abducted to get an interview with the guerrillas.c. Millicent must very clearly have been disqualified by the RUSSIAN judge, or else the

committee would not have protested so hard.d. She may actually have been only aiming for the vice-president’s job.


26

(22) Positional Distribution ofVP-Adverbs

a. clause-final

b. to the immediate left of the main verb6

Particularly in light of footnotes 5 and 6 ,1 note that this distribution is an idealization. It is

nevertheless a useful and otherwise realistic one and I will adopt it for what follows.

Examples that do not conform to it are rare and often forced.

This distribution is schematically illustrated in (23), which shows the possible adverb

positions in a string of a subject followed by optional auxiliaries followed by a main verb.

For orientation purposes, j ^ u x } *s position of 1° in current conceptions.

(23) S-ADV SU S-ADV ( { ™ j j x } ) S'ADV <AUX> VP-ADV V VP-ADV

A proposal to translate this description into a concrete syntactic analysis is conspicuously

absent in the recent linguistic literature. I develop one in the following section.

3.2. A Syntactic Proposal

To sketch an account that is descriptively adequate I bring in three assumptions. The first is

the finite clause structure we are working with, repeated below. The second is that subjects

can only be located in specifier positions, for example spec, VP or spec,IP.

6Again, Ernst’s (1984) proposal is less restrictive. In addition to permitting VP-adverbs next to the main verb, it allows them to the left of the passive auxiliary, as in (i). I will ignore this in the formulation above.

(i) Bobby will have handily been beaten by Billy Jean.


27

(24)

subject I’

V complement

The third is that adverbs are adjoined elements, attaching either to X’ or XP.7 The excluded

alternative is that adverbs are located in specifier positions. Two desirable consequences

follow from an adjunction analysis of adverbs. First, adjunction accounts for the possibility

of adverb iteration. This would be unavailable if adverbs were in specifier positions since

phrases have only one specifier. The examples of multiple adverbs, in (25), seem to be

acceptable, particularly with appropriate intonation, a claim also made in Ernst 1984.

Second, on the assumption that heads do not generally agree with adjuncts, we also

account for the observation that languages do not typically show agreement with adverbs,

in contrast to arguments. Again, if adverbs occurred in specifier position, one might expect

to see spec-head agreement involving an adverb.

(25) a. She seldom ever opens the cabinet.

b . He apparently never merely skims papers but rather reads each one carefully.

c. ?The burglars evidently probably broke in the back door.

d . ?The mechanic simply carefully separated the two pieces.

e. ‘He gladly always took people up for rides.’ (SF Chronicle, April 12, 1996)

7One or the other o f these is typically assumed in most analyses. For example, Johnson 1991 and Bowers 1993 assume adjunction to X’ while Pollock 1989, Iatridou 1990 and Chomsky 1991 assume adjunction to XP.


2 8

(26) and (27) constitute a proposal that will place adverbs in the correct structural positions

as described above. It is not the only set of possibilities and, more importantly, it does not

unambiguously determine adverb placement in any given situation. On the other hand, it is

largely descriptively adequate and within the set of assumptions above it can identify

specifier positions and heads that are overtly realized.

(26) Syntactic Distribution ofS-Adverbs

a. left adjunction to IP

b . left adjunction to I’

c. left adjunction to the topmost VP

(27) Syntactic Distribution o f VP-Adverbs

a. left or right adjunction to main verb V’

The proposal places adverbs in the positions indicated in (28). The reader can check that

these are just those desired given Jackendoff s description. Right adjunction of a

VP-adverb is not shown since it transparently places the adverb in clause-final position.


29

(28)

(S-adverb) IP

subject

(S-adverb) I’

(S-adverb) VP

(VP-adverb) V’

main

Below I consider this proposal in more detail. Before ending this section, I show that the

distributional generalizations make two correct predictions regarding the interaction of

adverbs with other aspects of English syntax. In the next section, I attempt to demonstrate

that, despite appearances, the distributional patterns have some syntactic and semantic

generality and are not simply a scattered collection of adjunction positions randomly

grouped together so as to obtain the right facts.

3.2.1. Syntactic Consequences

Two desirable syntactic consequences of the proposal are illustrated in (29) through (31).

(29) confirms that S-adverbs must always appear to the left of a VP-adverb in the auxiliary

position. This follows from (26) and (27) because an adjunction to (the topmost) VP will

always be above a V’ adjunction site.


30

(29) a. Hulk Hogan [evidently]s [completely]yp annihilated his opponent,

b . *Hulk Hogan [completely] yp [evidently]s annihilated his opponent.

(30) and (31) illustrate that VP-adverbs are necessarily interpreted inside VP ellipsis

structures while S-adverbs need not be. When the antecedent of a null VP contains a

VP-adverb, the interpretation of the elided VP obligatorily contains the adverb, as in (30).

In the case of S-adverbs, this is not necessary; the adverb may be excluded from the

interpretation, as it is in (3 l)a through c for pragmatic reasons. (3 l)d and e illustrate that an

interpretation with the S-adverb is also available.

(30) a. Helga easily won her race and Sophie will 0 too.

0 = easily win her race, *win her race

b . Johnny blatantly disobeys the baby sitter because his sister does 0.

0 = blatantly disobeys the baby sitter, *disobeys the baby sitter

c . Daisy thoroughly cleaned the pans since the automatic dishwasher couldn’10.

0 = thoroughly clean the pans, *clean the pans

(31) a. Sonny will certainly get elected if Ron does 0.

0 = get elected, *certainly get elected

b . Due to the traffic, we will unfortunately miss the opening credits, but those who

were there early won’t 0.

0 = miss the opening credits, ^unfortunately miss the opening credits

c. The Mafia allegedly set the hotel on fire because the owner didn’t 0.

0 = set the hotel on fire, *allegedly set the hotel on fire


31

d . The Joneses will probably buy a Mercedes because the Smiths will 0.

0 = buy a Mercedes, probably buy a Mercedes

e. No logician would knowingly state a falsehood even though a politician might 0.

0 = state a falsehood, knowingly state a falsehood

These examples are accounted for under the assumption that VP ellipsis targets VPs and an

S-adverb adjoined to VP creates a two segment category, either of which is a possible

antecedent for VPE. Adjunction to V’ in the case of VP-adverbs does not create a

structurally ambiguous antecedent for VPE. There is only one VP that can be the antecedent

and the VP-adverb is necessarily internal to it. Both sets of facts, without additional

assumptions, would seem to rule out adjoining VP-adverbs to VP instead of V’.

3 .3 . Adverb Licensing

A primary desideratum for any syntactic proposal regarding adverb placement is that it

reflect the apparent link between an adverb’s meaning, or what it modifies, and its syntactic

position. This goal is a specific instantiation of what Sportiche 1988 formulates as the

Adjunct Projection Condition in (32).

(32) Adjunct Projection Condition (Sportiche 1988)

If some semantic type X modifies some semantic type Y, and X and Y are

syntactically realized as x and y, x is projected adjacent to either y or the

head of y.

Sportiche’s proposal, while intuitively desirable, is programmatic in the sense that it does

not specify what the actual syntactic realization(s) of the semantic modification are. This

would hinge largely on the interpretation of “adjacent” in (32). The above proposal

stipulates what these realizations are for adverbs but is compatible with Sportiche’s claim


32

that it is the head of a modified element that plays a central role in the actual syntactic

outcome. If we look more closely at the set of positions for S- and VP-adverbs, we see that

each group clusters around a particular head, 1° or V° respectively. (33) and (34) illustrate

the two situations more closely.

(34) VP

(VP-adverb) V’

V’ (VP-adverb)

(S-adverb)

(S-adverb)

V,main

(S-adverb) VP

Accepting that such groupings have some validity, I will develop the hypothesis that the

two adverb types are actually licensed in their possible syntactic positions by the respective

heads (see Travis 1988 for a similar proposal and Bowers 1993 for an application of the

idea). S-adverbs are licensed by 1° and VP-adverbs are licensed by V°. Intuitively, main

VPs function predicatively as actions or states and are modified by manner (VP-) adverbs;

IPs create a proposition from VP and, hence, are appropriately modified by prepositional

(S-) adverbs. The head which licenses each adverb type thus reflects the semantics of the

adverb, as modifying either a proposition or an action/state—and conversely.

The proposal is that an adverb is licensed by a head and it must occur in a position that is

syntactically associated to the head via some relation, call it R. It remains to specify what R

is so as to capture the syntactic positions in (33) and (34). I will develop the idea that R can


33

be formulated as “in the checking domain of the head”.8 The set of positions in which

adverbs appear in (33) is almost coextensive with the set of phrases that are in the checking

domain of 1° as defined in section 2.1. Informally, the checking domain of a head includes

its specifier, phrases adjoined within its projection, and those adjoined to the complement

of the head. We may say, then, that S-adverbsAT-adverbs must modify IP/VP within the

checking domain of the head r/V°. (26) and (27) repeat the syntactic distribution of

adverbs to evaluate how the checking domain relation accounts for it.



b. left adjunction to I’

c. left adjunction to the topmost VP

(27) Syntactic Distribution o f VP-Adverbs


The analysis correctly predicts the adjunction possibilities for VP-adverbs in (27).

Adjunction to V’ is within the checking domain of V°. Adjunction to VP is not. Exactly the

configurations encoded in the structure in (34) fall out as desired. Although a VP-adverb

adjoined to the complement of V° would also be in the checking domain, such a structure is

ruled out by the Adjunction Prohibition.

Now consider the distribution of S-adverbs in (26), with the relevant head being 1°. An

adverb adjoined to the VP selected by 1° will be in the checking domain of 1° so we account

for (26)c, which previously seemed like a rather odd restriction since reference to the

8Although I will not do so here, a workable, and perhaps more straightforward, proposal could also be developed taking R to be government (Rizzi 1990).


34

‘topmost’ VP is otherwise ad hoc. An adverb adjoined to I’, in (26)b, is also transparently

in the checking domain of 1°. The IP-adjoined position, (26)a, is incorrectly ruled out

because it is not in the checking domain of 1°. I turn to this problem below.

One additional position in the checking domain of 1° not considered in the distributional

discussion above is right adjunction in general. The checking domain hypothesis predicts

that S-adverbs should adjoin on the right as well as the left, as was the case with

VP-adverbs. This is in fact permitted. Typically there is a pause required before the adverb,

(35). Certain S-adverbial phrases do not require a pause. Thus the prediction is generally

borne out, although details remain to be understood.

(35) a. Horatio has lost his mind, probably.

b . Casey thinks that there are guerrillas in the rose garden, evidently.

c. Louis had rid the city of rats, supposedly.

d. She then played the Toccata and Fugue in C Minor, quite surprisingly.

(36) a. They’ll win most likely.

b . His plan will destroy us in all probability.

c. The new measures will destroy us for certain.

Returning to (26)a, however, we incorrectly do not permit S-adverb adjunction to IP since

an adverb in such a position will not be in the checking domain of a head. Without

modifying the definition of checking domain, two options present themselves: either 1)

adopt the view that the IP-adjoined site is not a base-generated adverb position, the adverb

is licensed elsewhere in the structure and moved to the IP-adjoined position, or 2) license

adverbs adjoined to IP from C° or some higher head. I briefly explore both possibilities

below and adopt the second proposal.


35

3.3.1. IP-Adjoined. Adverbs

Jackendoff 1972:66 observes that the IP-adjoined, or clause-initial, position for S-adverbs

sounds less acceptable in some embedded contexts, (37)a and (38)a, compared to the

situation in which the adverb is lower in the structure, (37)b and (38)b and c.

(37) a. ?George says that evidently Bob has disappeared,

b . George says that Bob has evidently disappeared.

(38) a. ?I won’t come because [jp probably [jp my mother is sick ] ]

b . I won’t come because [jp my mother [j. probably [r is sick ] ] ]

c. I won’t come because [jp my mother is [yp probably [yp sick ] ] ]

Since there are numerous movement operations which are restricted in embedded clauses,

he concludes from these facts that this adverb position is likely derived via movement. In

particular, it seems reasonable to assimilate this hypothesized movement to English

Topicalization. For some, this construction also sounds odd in embedded contexts. This is

illustrated in (39) and (40). The judgments given are relative.

(39) a. (?)George says that, this article, he won’t read again,

b . This article, I won’t read again.

(40) a. (?)I believe that [jp the books [jp I gave away to some friends ] ]

b . The books, I gave away to some friends.

Like the left-peripheral adverb position, Topicalization is also analyzed in (40) as

adjunction to IP following Baltin 1982 and Lasnik and Saito 1992. The two phenomena

can thus be unified by assuming that the IP-adjoined adverb position is simply adverb

topicalization. Both topics and clause-initial adverbs would then be instances of a single


36

movement operation: fronting to an IP-adjoined position. Such a move has independent

motivation. Gundel 1988:154 demonstrates (contra Chomsky 1995) that topicalization of

adverbials is a process that must exist in the grammar anyway. This claim is illustrated by

the data in (41). These examples seem to require long-distance construal of the italicized

adverbial with the embedded clause. Under the assumption that adverbials are base

generated in the clauses they modify, the derivation of such examples must involve

topicalization of the adverbial from the embedded clause as illustrated in (41)c.

(41) a. Around midnight, I promised he would be there.

b . Yesterday, Jim claims that he wrote three chapters.

c. [jp next week} [jp George said [cp that he’d talk about quantifiers *,•]]]

Not all people get the readings in (41) and long-distance movement of adverbials is not

uncontroversial. Given the indicated judgments however, its existence is supported by the

observation that such data seem to obey island constraints, a standard diagnostic for

movement. The examples in (42) contrast minimally with those in (41) in that the adverbial

can absolutely not be construed with the lower clause when it is in an island—a complex

NP in (42)a, an adjunct clause in (42)b, and a w/t-island in (42)c. The examples are

ungrammatical on the indicated interpretation.

(42) a. *Around midnight, I made a promise that he would be there.

(cf. I made a promise that he would be there around midnight.)

b . * Yesterday, Jim is tired because he wrote three chapters.

(cf. Jim is tired because he wrote three chapters yesterday.)

c. *[jp next week; [Ip George wondered [cp how he’d have time to talk about

quantifiers f(- ] ] ]

(cf. George wondered how he’d have time to talk about quantifiers next week.)


37

The data lead to the conclusion that adverb topicalization to the IP-adjoined position must

exist in the grammar on independent grounds. Analyzing all IP-adjoined adverbs as

topicalization would reduce the number of base-generated positions required for them while

introducing no new syntactic mechanisms. It would also eliminate the need to explain how

EP-adjoined adverbs are licensed. The adverbs would be licensed in their base-generated

position, before movement.

On the other hand, there is one piece of evidence that suggests that another solution should

also be available. The double topicalization construction illustrated in (43) argues against

taking all instances of IP-adjoined adverbs as fronted topics. Topicalization of more than

one XP is typically quite unacceptable (but see Culicover 1996 for a differing view), an

instance of the more general pattern of strong ungrammatically that results from the

combination of two A movements within the same clause.

(43) a. *On the table, this book, John put.

b . *To the Salvation Army, the old National Geographies, you should donate.

Examples of single topicalization with an IP-adjoined adverb, in (44) and (45), may

perhaps seem degraded, but not nearly to the same extent as the sentences in (43). Under

the assumption that all IP-adjoined adverbs are derived via movement, the data in (44) and

(45) should have a similar status to (43) since they are syntactically, structurally identical.9

That the examples below are only slightly degraded may be taken to indicate that they are a

distinct phenomenon from double topicalization.

9Under the view that double topicalization is acceptable (Culicover 1996), the difference in grammaticality is no doubt attributable to the difference in ease of processing. When an argument is topicalized, the hearer must link it up with the verb of which it is an argument. Topicalized S-adverbs require no such processing effort provided they have not been moved from a lower clause. Thus, the contrast here may well have independent motivation once multiple topics are permitted.


38

(44) a. Apparently, in his mattress, John hid his money,

b . ?In his mattress, apparently, John hid his money.

(45) a. Perhaps, the house on the Cape, they’re going to put up for sale,

b . ?The house on the Cape, perhaps, they’re going to put up for sale.

Independent of the analysis that such examples would receive, their status does not seem to

warrant assimilating them to the clearly ungrammatical cases in (43). I conclude from these

limited observations that the IP-adjoined adverb position is potentially a base-generated

one, in addition to possibly being derived via movement. To account for this, I will offer

an analysis by which the IP-adjoined adverb is licensed by the c-commanding head C°, an

option which cannot be excluded in principle anyway.

Given the structure in (46), it is clear that an adverb adjoined to IP will be in the checking

domain of C° just as an adverb adjoined to VP was in the checking domain of 1°. This

provides a mechanical way of licensing IP-adjoined adverbs illustrated in (47).

(46) CP

e

c IP

AdvP IP

(47) a. Perhaps it will stop raining soon.

b . Conceivably he went to the wrong restaurant to meet us.

c. Probably you would be more successful in a bigger company.

3.3.2. Licensing from C°

Licensing clause-initial adverbs from C° leads to two expectations. First, since a phrase

adjoined to C’ would also be in the checking domain of C°, we expect to find adverbs in


39

this position, between the specifier of CP and C°. Second, since 1° and C° are distinct

licensers, we introduce the possibility that an adverb could be licensed by only one of the

two heads. Such an adverb, if licensed only by C°, would only appear clause peripherally.

If licensed only by 1°, the adverb would appear immediately following the subject or in the

auxiliary position, but not clause-initially. Both of these expectations seem to be borne out

and are illustrated below.

The examples in (48) confirm the prediction of C’-adjoined adverbs. I assume that

questions are CPs in which the specifier of CP is occupied by a vWi-phrase and the head C°

contains the inverted auxiliary. With reference to the structure in (46), C' adjunction places

an adverb between these two elements as illustrated for (48)g. Although the examples are

not all perfect, my consultants also do not judge them fully ungrammatical.

(48) a. ?Who possibly can we call at this hour of the night?

b . Which of them apparently does he not like?

c. ?Who ever has he criticized except Jan?

d. ?Where conceivably could one find a good buy on snow tires?

e. Bill is arguably more qualified that Tracy, Tony, Tim, Tina,. . .

?Who arguably is he not more qualified than?

f. ?Who stupidly did you invite by mistake?

g. ?[cp where [c < most likely [c> will [jp the spy meet his contact? ] ] ] ]

The second expectation is that there will exist adverbs which are licensed by only one of

either 1° or C \ I present two potential instantiations below. The first adverb type involves

what I will call E(xtent)-adverbs, exemplified by just, merely, simply, etc. They are

licensed only by 1°. The second case involves bare-NP adverbs which may appear clause-

initially but not clause-intemally. They are reasonably licensed only by C°.


40

In addition to S- and VP-adverbs, Jackendoff 1972 describes a third class of adverbs

which have the distribution of neither. Adverbs like merely, hardly, or scarcely do not

assimilate into either category on syntactic or semantic grounds. I will call them

E(XTENT)-AD VERBS since they approximately describe the extent or degree to which a

situation holds. Jackendoff gives their distribution as necessarily occurring somewhere

between the subject and the main verb. (49) and (50) show immediate post-subject and

post-modal positions, respectively. (51) illustrates that clause-initial position is not

possible. (52) is Jackendoff s (3.141). As he observes, only clause-internal positions and

not clause-peripheral ones are acceptable for E-adverbs.

(49) a. He simply is incapable of it.

b . The raccoons scarcely have touched our garbage.

c. They hardly should worry about that.

(50) a. He is simply incapable of it.

b . The raccoons have scarcely touched our garbage.

c . They should hardly worry about that.

(51) a. * Simply he is incapable of it.

b . * Scarcely the raccoons have touched our garbage.

c. * Hardly they should worry about that.

(52) a. *Merely John will have been beaten by Bill.

merely will have been , T , will merely have been , „ ,b - Iohn ?wUl have merely been beaten ** BlU-

will have been merely

c. *John will have been beaten by Bill merely.


41

(53) extends the description of adverb placement to account for the clause-internal

positioning of E-adverbs. This potentially places them in any of the positions indicated in

(54). The structure shown is for (49)c and (50)c.

(53) Syntactic Distribution o f E-Adverbs

a. left adjunction to I’

b . left adjunction to VP or V’

(54) IP

DP I’

they (hardly) I’

should (hardly) VP

(hardly) V’

worry about that

(53) gives a description of the distribution of E-adverbs. It remains to turn this into a

formal licensing scheme for the adverb. I maintain that the distribution in (53) can be

captured by allowing E-adverbs to be licensed by 1° and V°, but not C°. Such an analysis

would provide support for the idea that 1° and C° are distinct licensing options. For

E-adverbs, 1° licensing will place them in the following adjunction sites: I’ adjoined and

topmost VP adjoined. V° licensing permits the adverb to adjoin to V’ and to a VP that is the

complement of some V°. These possibilities are schematized in the structure in (55) and are

exactly the ones proposed in (53). The reader can verify that they are the positions


42

determined in (53) and that they permit the data in (49) through (52).10 Crucially, the

clause-initial position, repeated below in (51), and the C’-adjoined position, (56), are ruled

out.

(55)

subject

(E-adverb)

VP

(E-adverb) VP

(E-adverb)

(E-adverb)

(E-adverb)

(51) a. * Simply he is incapable of it.

b . * Scarcely the raccoons have touched our garbage.

c . * Hardly they should worry about that.

*®The proposal does not prevent right adjunction of E-adverbs, which must transparently be prohibited so that these adverbs do not appear clause finally. I have no explanation for this restriction. Kayne 1994 unilaterally rules out right adjunction and is one potential path to pursue. I will not do so here because it would require complete rethinking of other aspects of adverb placement The fact that E-adverbs can be licensed by any verb, including auxiliaries, while VP-adverbs can only be licensed by predicative main verbs is also unaccounted for. It is undoubtedly due to their differing semantics. Given that there is no unified theory of the meaning of E-adverbs, I will not pursue these problems.


43

(56) a. *Who simply will he not invite?

b . *Why hardly did Zoe talk to you?

c. *What just didn’t the students understand?

E-adverbs do not appear in the clause-initial position (IP-adjoined) or between a w/i-phrase

and an inverted auxiliary (C’-adjoined) because they are not licensed by C°. They

nevertheless appear in the positions which are licensed by 1°. They thus provide evidence

for C° and 1° being distinct licensers.

A class of adverbials that contrast with E-adverbs in plausibly being licensed by C° but not

by 1° are bare-NP adverbs (Larson 1985, Emonds 1987). Larson 1985 demonstrates that a

restricted set of noun phrases including temporals (numerous examples such as tomorrow,

last week, next Christmas, and noun phrases headed by the common noun time), locatives

(headed by the common noun place), directionals (headed by way or direction), and

manner NPs (headed by way) may function as adverbial modifiers. (57) through (60)

illustrate their distribution. They may appear clause-initially, in (57), and clause-finally, in

(58). Structurally, the data illustrate left/right adjunction to IP as schematized in the c

examples.

(57) a. Tomorrow I will start my new diet.

b . Next Christmas the Smiths are going to Hawaii.

c . [jp last week [jp the government expelled several alleged spies ] ]

(58) a. I will start my new diet tomorrow.

b . The Smiths are going to Hawaii next Christmas.

c. [j-p [jp the government expelled several alleged spies ] last week ]


44

Other positions for these adverbs are degraded. They may not appear lower in the clause,

either immediately to the right of the subject in (59), the I’-adjoined position, or to the right

of a modal, in (60), the topmost VP-adjoined position. They are also not possible lower in

the VP, (61), where they would be adjoined to embedded VP or V’.

(59) a. *1 tomorrow will start my new diet.

b . *The Smiths next Christmas are going to Hawaii.

c. *[jp the government [r next week [r is expelling several alleged spies ] ] ]

(60) a. *1 will tomorrow start my new diet.

b . *The Smiths are next Christmas going to Hawaii.

c . *the government might [yp next week [yp expel several alleged spies ] ]

(61) a. *1 will be tomorrow starting my new diet.

b . *The Smiths could be next Christmas going to Hawaii.

c. *according to sources, the government might [yp have [yp [y/yp last week

[y, expelled several alleged spies ] ] ] ]


45

Summarily, these bare-NP adverbs may not appear in any of the positions that would be

licensed by 1°. They only appear adjoined to IP. This is compatible with their being

licensed solely by

To summarize, I have attempted to explain the syntactic distribution of adverbs by licensing

them in the checking domain of individual, semantically contentful heads: C°, 1°, and V°.

The latter two are clearly motivated; the licensing of adverbs by C° is also suggested.

Before leaving the realm of adverb licensing, it is worth pointing out that this use of the

checking domain is perhaps somewhat unorthodox. The checking domain was intended to

serve as a domain in which abstract features are checked in a defined local relationship and

morphologically realized as agreement. In adopting the checking domain for adverb

licensing, we predict that a language might show overt agreement between a licensing head

and an adverb adjoined within the head’s checking domain. This is certainly not the case

for English and I know of no such language elsewhere. This would seem to be an

11 There are two difficulties with this analysis for which I have no explanation, calling it into doubt. First, the proposal predicts that these adverbials should appear in the C’-adjoined position, (i) shows this prediction to be wrong, although some speakers do accept the data.

(i) a. */?Who tomorrow can we call to fix the plumbing?b . */?Which o f them last year did she date?c. */?Where next weekend will you ever find a hotel room?

Second, the clause-final position seems analyzable as adjunction at some level lower than IP. The negative polarity item any is licensed in these bare-NP adverbs by sentential negation but only in the clause-final position. This is surprising if the two positions are structurally identical, modulo direction of adjunction.

(ii) a. *Any day this week, you won’t find an available room in this town, b. You won’t find an available room in this town any day this week.

A more exact characterization of the placement of these bare-NP adverbs may be that they must be clause peripheral.^Issues for future research are why the adverb classes under investigation, particularly E-adverbs and bare- NP adverbs, have the distribution that they do and whether the syntactic distributions have a more fundamental semantic basis.


46

inappropriate application of the checking domain, as originally conceived. It seems

ultimately necessary to distinguish a head’s checking domain, a domain in which features

are actually checked, from some, still local, domain in which thematic roles are assigned

and adverbs are licensed. The former would seem to be the more restricted of the two as it

does not include adjoined positions except for direct adjunction to the head.

4 . Sentential Negation

Negation in English behaves in many ways like an adverb, although in most recent

syntactic works it is not treated as such. I will adopt the view that sentential negation in

English, realized as the lexeme not, is the head of its own projection in the syntax (Kayne

1989, Pollock 1989, Zanuttini 1991, Potsdam in press). Negative sentences as in (62) have

the general structure in (63) (I consider contracted negation -n’t below).

(62) a. Bill did not look well.

b . You should not miss this movie.

c . He cannot get here any earlier.

(63) IP

subject I’

f ^ ^ N e g P

modal Neg VP

not

Zanuttini 1991 discusses in depth the position of Neg(ative) P(hrase)s in English and also

concludes that there is a NegP immediately dominating VP. More accurately, since her

proposal is couched within a complex Infl scenario in which two functional projections,

AgrP and TP, replace the formerly unitary IP, Zanuttini proposes that there are actually two


47

NegPs in English. Neg2P occurs between TP and VP and corresponds to the NegP in (63).

The head of this Neg2P is not. NegjP occurs between AgrP and TP and is headed by the

dependent morpheme -n't. Modals and auxiliaries located in T° may move to Neg^ to

adjoin to -n't and form the negative elements don’t, shouldn't, etc. I will adopt the least

controversial part of her proposal, namely, that Neg2P exists and is located immediately

above VP. I will further assume that the syntactic position of this NegP is constant across

clauses. I do not adopt the second negative phrase, NegtP, for two reasons. First, it

crucially depends upon a split Infl scenario, which I do not adopt. Second, I find the

‘lexical’ analysis of contracted forms in Zwicky and Pullum 1983 particularly compelling.

They argue that -n ’t is not a syntactic clitic as in Zanuttini’s analysis but rather an

inflectional affix. Their evidence is based on semantic and syntactic idiosyncrasies of these

auxiliaries which make a clitic analysis of -n 't unlike other clearer cases of cliticization. I

will henceforth use the terms NEGATIVE MODALS and NEGATIVE AUXILIARIES to refer to

the two classes consisting of can’t, shouldn't, etc. and isn’t, hadn’t, etc. Morphologically

distinct negative modals are found in numerous languages, for example the languages of

West Africa (Zwicky and Pullum 1983, Koopman 1992).

Zwicky and Pullum’s analysis can be adapted to a clause structure in which sentential

negation is obligatorily represented with NegP by taking the negative modals and

auxiliaries to be formed in the lexicon and appended with the feature [NEG] which must be

checked by a corresponding feature in NegP. This will force the presence of NegP in all

cases of sentential negation, even when there is no negative morpheme per se. Otherwise,

negative modals and auxiliaries are analyzed no differently than their positive counterparts.

In particular, they have the same category status. Negative modals in (64) are analyzed

according to the structure in (65). They are inserted into 1° and the [NEG] feature raises

from Neg° to be checked.


48

(64) a. Bill didn’t look well.

b . You shouldn’t miss this movie.

c . He can’t get here any earlier.

(65) IP

subject I’

r ^ ^ " ^ N e g P

I Negj tj VPI

negativemodal

Negative auxiliaries have the analysis illustrated in (67). again, syntactically, they behave

no differently from their positive counterparts with the exception of the presence of NegP

in the clause. In raising to 1° (see chapter 2), they pass through Neg° to check a [NEG]

feature before ultimately checking tense in 1°.

(66) a. We aren’t satisfied.

b . Somebody hasn’t authorized this.

c. The answer isn’t obvious.

IP

subject V

I NegP

Negj I t{ VP

(67)

Neg XP

negativeauxiliary

This account of English negation attributes the presence of sentential negation to a unified

syntactic source: NegP. At the same time, it takes advantage of Zwicky and Pullum’s 1983


49

observations regarding the morphological realization of negation. The analysis is

compatible with Laka’s (1990) claim that NegP is actually one instantiation of ZP, a phrase

that specifies the polarity (positive or negative) of a clause. Neg0 is merely one form of Z°.

For clarity, I will continue to use NegP unless a more general claim is being made.

5. VP Ellipsis

VP ellipsis (VPE) in English, illustrated in (68), is a process that allows a constituent

consisting of a verb, its arguments, certain adjuncts, and, optionally, accompanying

auxiliary verbs to be missing under ‘identity’ with a like constituent elsewhere in the

discourse (Hankamer and Sag 1976; Sag 1980; Zagona 1988a, 1988b; McCloskey 1991;

Lobeck 1987, 1991, 1992, 1995).

(68) a. Joe will taste the food if Mikey does.

b . Matt is moving to Finland and Sophie might also be.

Since auxiliaries plus main verbs and their internal arguments analytically constitute verb

phrases, the phenomenon is labeled VP ellipsis.13 VPE is seen in other languages besides

13More generally, it has been claimed (Steele et. al. 1981, Williams 1984, Baltin 1995) that VP ellipsis needs to be viewed as predicate ellipsis given examples such as (i). The missing constituent, marked by 0, is interpreted as an AP, DP, or PP, depending on the complement of be in the antecedent.

(i) a. Snoopy might be [^ p mean to the new puppy ] and the cat might be 0 too.b. Mrs. Witherspoon is [^ p a notary public ] and her husband might be 0 as well.c. John is [pp in trouble ] and, if you don’t behave, you will be 0 too.

Under the analysis of the copula be here, there is no movement that would allow us to claim that the missing constituent in any of these cases is a VP, although such mechanisms are in principle available. See for example Lobeck 1995 or earlier analyses such as Akmajian and Wasow 1975 and Iwakura 1977 in which there is a rule of non-finite be raising which moves certain forms of non-finite be to the left, out of VP, leaving behind a VP with no head. Although the label Predicate Ellipsis or Post-Auxiliary Ellipsis (Sag 1980) would thus be more accurate, I will continue to refer to the phenomenon as VP ellipsis because it is


50

English, for example, Brazilian Portuguese (Chao 1988), Chinese (Chao 1988), Hebrew

(Doron 1990, Borer 1995, Goldberg 1996), Irish (McCloskey 1991), and Malay (Lobeck

1992), although it is not found in all languages.

In this section I develop an analysis of VP ellipsis with which to investigate the phrase

structure of imperatives and other clause types. VPE is particularly useful as a probe on the

position of elements near the IP/VP interface. The section begins with a presentation and

examination of the analysis of VPE in Lobeck 1995. Within a Government-Binding

approach, Lobeck 1995 discusses cross-categorial ellipsis and offers a restrictive condition

on the licensing and identification of elided categories, including null VPs. The analysis

assimilates the licensing and identification of elided VPs to that of other zero elements such

as null subjects. Like them, a null VP is required to be in a particular locality relationship

with an agreeing head. After laying out Lobeck’s system, I proceed to point out some

empirical and theoretical difficulties that suggest that some of the analytical particulars are

untenable. This careful look at Lobeck’s system has two valuable purposes, however. It

lays out the core body of data to be accounted for in any syntactic analysis of VPE and it

yields valuable insights into the conditions under which VP ellipsis is permitted in English.

From this foundation, I proceed to argue for a revised condition that builds on Lobeck’s

approach. The new Licensing Condition requires a null VP to be licensed by an inflectional

head that is in the same s-projection (Abney 1987) as the null VP. I show how this

accounts for the full range of Lobeck’s data in a more explanatory and theoretically more

unified manner.

verbs that primarily concern us here. The observations that are made below will be seen to readily extend to a more general view of ellipsis.


51

Before continuing, I introduce some terminology for later use. It has been observed that VP

ellipsis behaves much like an anaphoric process: there is an antecedent VP and a null VP

whose interpretation is dependent upon the antecedent. The elided VP in the examples

below is marked by the symbol 0 and I will call it the TARGET. The ANTECEDENT, or

TRIGGER, is bracketed below and determines how the target VP is interpreted.

(68) a. Joe will [yp taste the food ] if Mikey does 0.

b . Matt is [yp moving to Finland ] and Sophie might also be 0.

5 .1 . Lobeck 1995

Lobeck 1995 (adopting a suggestion from Wasow 1972) takes null VPs to be non-NP

pronominal empty categories: null pronouns similar to pro of null argument languages such

as Italian. Chao 1988 and Lobeck 1992 present evidence that elided VPs do have

pronominal properties despite being non-NP in nature. They observe that, like pronouns,

empty VPs allow split antecedents, may have pragmatically controlled (non-linguistic)

antecedents, and obey the Backwards Anaphora Constraint (Langacker 1969), although the

first two of these are perhaps controversial (Hankamer and Sag 1976). Given that null VPs

can be analyzed as pronominal elements, Lobeck and Chao represent a null VP as in (69)a,

where the elided VP is a null pronominal element. I will use the modified representation in

(69)b.

(69) a. VP b. VPI I

pro 0

A central goal of Lobeck’s approach is to assimilate the possibility of VP empty categories

to that of other base-generated null pronominals. The primary mechanism for

accomplishing this is the licensing and identification requirement on pro given in (70). It


52

will be referred to as the Licensing and Identification Condition (LIQ or just the Condition.

It specifies the surface structure conditions under which pro, both DP and non-DP, can

appear.

(70) Licensing and Identification o/pro (Lobeck 1995)

At surface structure, a non-arbitrary, empty pronominal must be 1) properly

head-governed and 2) governed by an X° specified for strong agreement

The licensing portion of the Condition is accomplished via the proper head government

clause and identification is done through government by an agreeing head. These

relationships are defined according to Rizzi’s (1990) system of Relativized Minimality

(RM), in (71) through (73). The definitions for c-command and m-command that I assume

are from Chomsky 1986 in (74) through (76).

(71) Relativized Minimality

X a-govems Y only if there is no Z such that

(i) Z is a typical potential a-govemor for Y,

(ii) Z c-commands Y and does not c-command X

(72) Head Government

X head-govems Y iff

(i) X = {A, N, P, V, H[+tense]}

(ii) X m-commands Y

(iii) no barrier intervenes

(iv) Relative Minimality is respected


53

(73) Proper Government

X properly a-govems Y iff X a-govems Y within the most immediate

projection

(74) C-Command

a c-commands 3 iff a does not dominate P and every ythat dominates a

dominates P

(75) M-Command

a m-commands p iff a does not dominate P and every y, y a maximal

projection, that dominates a dominates P

(76) Condition on Domination

a is dominated by P only if it is dominated by every segment of p

The intuition behind the RM system is that the government relation holds between a head

and another element provided that, other things being equal, there is no closer head which

could serve as a governor. Proper head government is a particularly strict form of

government which must obtain in the c-command domain of a head. Thus, in (77), the

head X governs both its complement ZP and its specifier YP because they are in the

m-command domain of the head but X properly governs only its complement ZP because

X’ defines the immediate projection of X and the specifier YP is outside of this domain.

Rizzi considers Chomsky 1986 and ultimately follows Cinque 1990 in defining a BARRIER

as an element not directly selected by a category nondistinct from [+V]. The

(77) XP

Y

X ZP


54

nondistinctness specification makes the maximal projections selected by the functional

heads 1° and C° not barriers, in addition to the complements of V' and A0. Consequently, a

head governs an element that it c-commands provided no barrier intervenes and Relativized

Minimality is not violated.

Lobeck further allows for a government to hold between a moved head and elements in the

domain of its original position via the Government Transparency Corollary (GTC), from

Baker 1988.

(78) Government Transparency Corollary (Baker 1988)

A lexical category which has an item incorporated into it governs everything

which the incorporated item governed in its original structural position

The intuition behind the GTC is that head movements such as T-to-C” inversion or

T°-to-Agr° feature checking should not reduce the government domain of the moved head.

The GTC serves to restore to a moved head its original government domain, in particular,

its deep structure complement. The configuration of concern and the results that the GTC

achieves are illustrated in (79). Hj is a head that has undergone head movement to adjoin to

the higher head position H2. The original government domain of H( is the set {ZP, WP,

W°}: the specifier of HjP, the complement of Hj and the head of the complement of Hj.

The government domain of H2 is limited to {YP, HjP, tHj} by Minimality. The GTC

extends the government domain of H2 to include the union of these two sets, its own

government domain plus the government domain of the head that adjoined to it.

Specifically, H2 governs the deep structure complement of Hj, WP. The government

domain of H2 incorporating the effects of the GTC is boldfaced in the structure.


55

(79)

YP

H2 H t ZP H .’

WP

Given these definitions and conventions, the Condition in (70) licenses a null VP via

proper government by a V° or 1° head which may or may not have independently

undergone head movement. This part of the condition then differs little, if at all, from the

Empty Category Principle which regulates the appearance of movement-derived empty

categories. I now turn to the second part of the Condition, the identification of pro.

As has been long recognized in the literature on null arguments, the possibility of a null

argument is tied to the presence elsewhere of features that can identify the missing

argument. For nominal arguments, this identification is typically in the form of person and

number agreement morphology on or near the verb. For null VPs, the type of agreement

that can identify the empty category is less obvious. The strong agreement required by the

Identification Condition is defined in (80) and (81), drawing on Emonds 1987.

(80) Agreement (Lobeck 1995)

An X° is specified for agreement iff its features must be ‘shared’ with those

of another head or phrase under government

(81) Strong Agreement (Lobeck 1995)

An X° is specified for strong agreement iff X°, or the phrase or head with

which X° agrees, morphologically realizes agreement in a productive

number of cases


56

If we combine the definition of agreement in (80) with that of government in (72), a head H

may potentially agree with its specifier, its complement, or the head of its complement. Call

this agreeing element J. The feature F shared between H and J will depend upon the

properties of H. The head H is, additionally, specified for strong agreement and endowed

with the ability to license pro if H or J morphologically realizes the shared feature F.

Lobeck 1995 asserts that the relevant agreement feature for VPE is [+tense], since only

tensed clauses in English allow a null VP.14 [±tense] is a feature that is typically realized on

a verbal head but it is assumed to be located in 1° where it is subject to feature checking.

Morphological tense appears on either 1° or the head of the VP it selects. Taking [+tense] as

a morphologically realizable agreement feature, we have that 1° is specified for strong

agreement under (80) and (81) because the conditions are met: 1° governs its complement

VP and either 1° itself or V° morphologically realizes the feature [+tense]. As a result, we

derive Lobeck’s claim that it is 1° that identifies pro. This conclusion supports proposals

about VPE going back to Bresnan 1976, who first implied that the AUX position was

crucially involved in the licensing of VP ellipsis. In the discussion following the explication

of Lobeck’s proposal, I return to this central observation and attempt to justify it and make

it more precise. Before this, however, I survey the core cases of VPE in English and

evaluate how Lobeck’s proposal accounts for the facts. To summarize, the necessary pieces

of the analysis are The Government Transparency Corollary in (78), the definition of

Strong Agreement in (81) and the Licensing and Identification Condition on pro in (70).

14Ellipsis is clearly also possible in non-finite complements in (i). Lobeck offers an analysis of this possibility that relies on the presence of tense in the predicate which selects the infinitival complement. I will not consider Lobeck’s analysis of such examples here.

(i) a. John doesn’t care about going to the parade but Bill hopes to 0.b. I’ll do it if you really want me to 0.


57

Consider first the classic grammatical case of VP ellipsis in (82)a whose second conjunct

has the structure in (82)b. In this example, and the ones below, it is necessary to confirm

that the empty category is both licensed and identified.

(82) a. John didn’t leave but Mary should/did 0.

b. IP

r

Mary I

should/did [+tense]

The empty category in (82) is properly head governed by 1° which satisfies the licensing

requirement. The identification clause is likewise satisfied by 1°. 1° is specified for strong

agreement because it morphologically realizes tense. In the case of did, this is transparent

because of the past tense form of the verb. For modals such as should, Lobeck adopts the

hypothesis that they are base-generated in 1° and realize tense, following Pollock 1989 and

Chomsky 1993 (see Gazdar, Pullum, and Sag 1982 for evidence).15 Being specified for

strong agreement, 1° may identify the null VP. (82) thus satisfies both conditions on the

appearance of pro and the result is grammatical. Here and in all of the grammatical cases to

follow, the empty category is licensed and identified by the same head, 1°, although such a

situation is not strictly required by (70).

The minimally different example in (83) without the auxiliary is transparently

ungrammatical because the empty category is not identified. 1° may very well be specified

15More precisely, since Lobeck actually employs a split Infl analysis in which IP is broken into TP and AgrP, modals are base-generated in TP from where they raise to AgrP for feature checking. It is hoped that maintaining the conceptually simpler IP analysis does not alter Lobeck’s fundamental proposals.


58

for [+tense] and thus properly govern the null VP and license it; however, 1° cannot be said

to morphologically realize tense. It is therefore not specified for strong agreement and does

not identify the null VP.

(83) a. *John didn’t leave but Mary 0.

b. *John didn’t leave but [jp Mary [j, I[+tns] [yp 0 ] ] ]

Next consider the grammatical example in (84), which is an elliptical interrogative

counterpart to (82). Under the assumption that interrogatives are derived via I°-to-C°

movement of the auxiliary/modal, we do not want such head movement to adversely

interfere with the licensing and identification by [+tense] in 1°. It should be evident that this

is the role of the Government Transparency Corollary. The GTC allows the empty category

in (84)b to be governed just as in (82) above. C° properly governs the null VP via the GTC

and so licenses it. C° also identifies the empty category because it is specified for the strong

agreement feature [+tense] by having had 1° adjoin to it. The representation assigned to

head movement in (84)b follows Lobeck 1995.

(84) a. John didn’t leave but should/did Mary 0?

b. CP

C ^ ^ ^ I P

should/didj DP I’[+tense] / \

Mary I VPI Iq 0

The example in (85) receives a fully parallel account. Its structure in (85)b differs from the

standard case in (82) only in that the auxiliary that is in 1° at the surface has moved there

rather than having been base-generated there. Its analysis is identical to that given for (84),


59

with the exception that the operative head movement is V°-to-I°. Again, the GTC allows the

[+tense]-specified 1° to properly govern the empty VP and agree with it via the trace of

movement in V \

(85) a. Mary is leavmg and John is/has 0 too.

John I

is/has j V[+tense]

Now consider the examples in (86), due to Bresnan 1976, which illustrate that VP ellipsis

is ungrammatical with aspectual verbs such as start, begin, or continue which Lobeck takes

to select bare VP complements, following Emonds 1985. The putative structure of (86)b is

in (87).

smoke I

(86) a. *Because Mary continued 0, John also started speaking French,

b . *Fire began pouring out of the building and then smoke began 0.

(87) * IP

VP

V VPI I

began 0[+tense]

Pro in (87) is licensed according to the Licensing and Identification Condition because it is

properly head governed by the verb. Furthermore, the verb began is transparently specified


60

for tense, leading us to conclude that the empty category is also identified since it is

governed by a head specified for strong agreement. It would seem that an incorrect

prediction is made. In fact, however, Lobeck agrees that the Condition is satisfied at

surface structure in this example. The difficulty arises when the main verb begin raises to 1°

at LF to check its tense feature. The LF is as in (88). It is ruled out by assuming that “a

feature can identify only one empty category during a derivation (Lobeck 1995:149)”. At

surface structure, in (87), the feature [+tense] identified the VP pro. At LF, when the verb

has raised into 1°, it is “blocked from also identifying the matrix VP” which Lobeck

assumes without argument must likewise be identified.

(88) * IP (LF)

smoke I

began V [+tense] I

The assumption necessary to exclude the data in (86) is stated in (89) since it causes

difficulties for the analysis elsewhere.

(89) Restriction on Feature Identification (Lobeck 1995:149)

A feature may identify only one empty category during a derivation

In particular, if we return to the example in (85), repeated below, we see that the surface

structure of the example is identical to the LF of (88) modulo individual lexical items.

Given the restriction in (89) needed to account for the aspectual verb case, the structure in

(85)b should also be ill-formed and the example ungrammatical because the feature [+tense]

must identify both VPs. This is contrary to fact.


61

(85) a. Mary is leaving and John is/has 0 too.

b. IP

DP F

John I VP

is/hasj V VP [+tense] | |

0

It seems evident that the assumption in (89) restricting the ability of a feature to license a

null VP is problematic in one case or another. If one were to assume that the restriction did

not hold or that the main VP didn’t need to be identified at LF, the ungrammaticality of (86)

would remain unaccounted for. Adopting the restriction, on the other hand, rules out

grammatical examples. A further difficulty for the account of the aspectual verb data is that

the Licensing and Identification Condition is assumed by Lobeck to apply at surface

structure,16 so LF operations such as Verb Raising should be irrelevant.

The analysis does correctly predict that inserting a lexical element into 1° in the examples in

(86) will not save them, (90). (86) is repeated below for reference.

(90) a. *Because Mary must continue 0, John also started speaking French.

b . *If fire begins pouring out of the building then smoke might begin 0 too.

(86) a. ^Because Mary continued 0, John also started speaking French,


16‘‘S-structure, the level at which licensing and identification conditions on ellipted categories must be met (Lobeck 1995:162)”.


62

(91)

smoke I

might V [+tense] |

begin

The examples in (90) are still ungrammatical, despite the presence of an additional potential

identifying head. The problem is no longer that the verb can raise to 1° at LF and create the

need for identification of a second VP. The modal in 1° blocks Verb Raising. It is the case,

however, that identification is now impossible. In the examples in (90), which will have

the structure in (91), Relativized Minimality prevents the [+tense] modal in 1° from

governing the empty category. A closer potential governor exists, V°, even though it is

unable to actually identify the empty category. The data in (90) are thus correctly excluded.

Even if the account of the ungrammaticality of (86) were to go through, however, it seems

intuitively unappealing that (90) and (86) should receive such different explanations.

As is also well known, VPE may optionally leave one or more auxiliaries when it applies:

(92) a. John could have been studying Spanish and Bill could have been studying

Spanish too.

b . John could have been studying Spanish and Bill could have been 0 too.

c. John could have been studying Spanish and Bill could have 0 too.

d. John could have been studying Spanish and Bill could 0 too.

(92)d receives an account as in (85) where the immediately adjacent I°[+tense] properly

governs the empty category and licenses and identifies it. In (92)b and c, by contrast, the

heads adjacent to the null VP, been and have respectively, are untensed verbs which are


63

incapable of identifying the empty category. There is apparently no way for the empty

category to be identified since 1°, a head specified for strong agreement, is too far away to

govern. Relativized Minimality strictly excludes 1° from governing the empty category

because of the closer V° governors. Lobeck’s way out of this difficulty is illustrated in

(93), for (92)b. The key analytical device is that all auxiliaries, both finite and non-finite,

move overtly to 1° at surface structure.

(93)

Bill I

could V

have been

Each non-finite auxiliary undergoes head movement, adjoining to the next highest head in

accordance with the Head Movement Constraint. The auxiliary complex finally adjoins to 1°

as shown. The motivation for such movement is taken to be checking of aspectual features.

The desirable consequence of such a structure is that VP ellipsis is then permitted. The

empty VP will be licensed and identified by 1° via the Government Transparency Corollary.

On the other hand, other motivation for such head movement is absent and the negative

empirical consequences seem overwhelming. The constituent structure in (93) closely

resembles that proposed in Chomsky 1957 and later argued for in Akmajian and Wasow

1975. Emonds 1976, Iwakura 1977, and McCawley 1988 present numerous arguments

against it drawing on a wide range of phenomena: There Insertion, parenthetical placement,

vWnc/i-pronominalization, and Coordination. Furthermore, it does not easily fit into

conceptions of adverb placement or negation developed here. On a theoretical level, the


64

structure in (93) also runs afoul of the restriction on a single feature identifying multiple

empty categories. The feature [+tense] is required to identify three VPs from Lobeck’s

viewpoint.17 Finally, I will present two additional difficulties with this multiple head

movement analysis.

First, the structure in (93), without some additional mechanism, derives the incorrect result

when r-to-C0 interrogative inversion applies. Movement of the head 1° derives (94)a rather

than the grammatical (94)b.

(94) a. *Could have been Bill studying Spanish?

b . Could Bill have been studying Spanish?

Second, the proposal requires an unmotivated V°-to-D° movement operation in the case of

Poss-ing gerunds. Akmajian and Wasow 1975 observe that VP ellipsis is found in Poss-

ing gerunds, illustrated in (95) from Iwakura 1977, although it is rather restricted.18

(95) a. Which bothers you more: John’s having been arrested or Bill’s having been 0?

b . Ford’s having been examined by a psychiatrist was just as unreasonable as

Nixon’s having been 0.

Abney 1987 proposes a structure similar to the one in (96) for Poss-mg gerunds. They are

fundamentally a D(eterminer) head that selects a VP complement.

17A further theoretical problem derives from Lobeck’s (1995:162) assumption that only unchecked features can identify empty categories. She crucially uses this to explain the fact that French and German do not have VP ellipsis. If we assume that only unchecked features identify pro then we must assert that English auxiliaries, which raise to 1° from where they can identify an empty category, do not raise in order to check features. If this were the motivation for raising, pro could not be identified because the strong agreement features would have been checked.18Abney 1987:244 asserts that VPE in Poss-/ng gerunds is impossible but the examples here and below seem acceptable.


65

(96) a. Bob’s quitting

b.

Bob D

quitting

s[+poss]

Lobeck 1995 independently accounts for ellipsis in noun phrases when preceded by a

possessive as in (97) with the assumption that [+poss] is a strong agreement feature. A

head that is specified as [+poss], namely the determiner’s, shows strong agreement and

identifies an empty category. Since it also properly governs the empty category, the

Licensing and Identification Condition is satisfied and ellipsis is permitted.

(97) a. Although John’s friends were late to the rally, Mary’s 0 were on time,

b . John caught a big fish but Mary’s 0 was bigger.

Abney’s structure combined with Lobeck’s theory of ellipsis in DPs correctly predicts that

a kind of VP ellipsis will be seen in gerunds, (98). These have a structure given by (99).

D° properly governs pro, satisfying the licensing condition. D° also identifies pro because it

governs the empty category with the strong agreement feature [+poss].

(98) a. Bob’s quitting the firm was more surprising than Henrietta’s 0.

b . Billy’s helping to clean the house was appreciated and Betty’s 0 was as well.

(99) DP

D’

VPBetty D

[+poss]


66

If we return to the examples in (95), repeated below, in which auxiliaries are stranded

under VPE, they can only be accounted for in Lobeck’s theory if we move the auxiliaries

into the licensing head. That is, parallel to the VP ellipsis examples with multiple

auxiliaries, a rule moving non-finite auxiliaries into the functional projection must be

posited so as to create a situation in which the head specified for strong agreement, D°, is

also the governor of pro. This is the V°-to-D° head movement illustrated in (100).

(95) a. Which bothers you more: John’s having been arrested or Bill’s having been 0?

b . Ford’s having been examined by a psychiatrist was just as unreasonable as

Nixon’s having been 0.

( 100)

Nixon D

[+poss]having been

The argument against the Lobeck 1995 analysis centers on this needed operation. In

contrast to V0-to-I° movement, V°-to-D° movement is unmotivated. It does not exist

independently and D° is an unlikely locus of aspectual features for checking; yet, it is

apparently required to predict the grammaticality of these examples.

As a last case, I look at VP ellipsis when negation is present. As is well-known, a null VP

immediately preceded by sentential not is grammatical, (101). Under Lobeck’s theory, not

must be the licensing head which we can see by considering the structure of the examples.


67

most I

(101) a. I definitely want to go to the fashion show although my husband might not 0.

b . A clown is coming to the party but a mime is not 0.

c . Some of the guests tried the appetizers but I think that most did not 0.

(102) illustrates the second conjunct of (101)c and what is necessary for the Licensing and

Identification Condition to rule them in.

(102) IP

r

NegP

Neg’

Neg VPI I

not 0

Relativized Minimality prevents I°[+tense] from governing the empty category because

there is a closer potential governor Neg\ Not properly governs pro and thus licenses it.

Since it is the closest governor, it must also identify pro for the example to be grammatical.

This requires that not be specified for strong agreement—a condition which does not

obviously obtain. Lobeck 1995:157 claims that Neg0 is specified for strong agreement

because it is in a spec-head agreement relationship with an operator in its specifier (not

shown in the structure). Lobeck uses Rizzi’s (1990) arguments that opacity effects

associated with negation are a result of an A operator in spec,NegP. Despite not

“morphologically realizing agreement” as the definition for strong agreement in (81)

requires, the spec-head relationship is apparently sufficient to allow Neg° to identify an

empty category. At the same time, Szabolcsi and Zwarts 1993 convincingly argue that the

weak island effects of negation discussed in Rizzi 1990 receive a more explanatory account

in semantic terms since they are invariant across languages, independent of the syntactic


68

analysis of negation as specifier, head, or adverbial. This weakens the support for

spec-head agreement in NegP. Finally, the data in (103) would seem to force one to say

that there also exists a positive polarity A operator, presumably also in spec,IP. The

examples indicate that so, which Laka 1990 takes to be a realization of 1°, licenses a null

VP as shown in the structure in (103)d. It must therefore also be specified for strong

agreement. While the existence of such an operator is not impossible, its motivation is

unclear.

(103) a. I didn’t mess with your stuff. You did so 0.

b. You weren’t invited. I was so 0.

c. The dog can’t reach the pork chops on the counter. It can so 0.

d. IP

I’

IP

can

i iSO 0

To summarize, Lobeck’s analysis has three primary difficulties: 1) contradictory

restrictions on the identification of pro, in (89), 2) unnatural constituent structures related to

head movement, and 3) poorly motivated assumptions about the realization of agreement in

negation structures. Consequently, the account is of limited use for investigating VP

ellipsis in other clause types. It requires structures and assumptions which are at odds with

major portions of this work. For these reasons, I will step back from the analytical details.

Instead, I will attempt to extract from Lobeck’s thorough examination of VPE the dominant


69

generalizations based on the data surveyed above. These observations will form the basis

for a revised licensing condition developed below.

5 .2 . A Licensing Condition

A primary intuition behind Lobeck’s analysis is that null VPs are allowed as a consequence

of being in some structural relationship to an appropriate head. Furthermore, for Lobeck it

is filled 1° that always plays the central role in the licensing, a claim also expressed in

Roberts’ (1990) analysis of related VP Fronting. In the analysis that I will develop I

crucially rely on Lobeck’s fundamental claim. I will argue that an inflectional head must

license an elided VP. My analysis will differ from hers in that the relationship that I claim

holds between the head and the null VP is less local than government. To this end, I

consider two separate but related issues: 1) what is the necessary licensing element for VPE

(subsection 5.2.1) and 2) what is the structural relation that obtains between this licensing

element and the elided VP (subsection 5.2.2).

5.2.1. The Licenser

In all the grammatical examples of VPE that we have seen thus far, two syntactic elements

have always been present: an overt element in 1° and a verbal head that is traditionally

labeled [+AUX].19 This co-occurrence is a consequence of my assumptions regarding the

English verbal system in finite clauses. The elements that are inserted directly into 1° are all

[+AUX] and, if 1° is unfilled and a [+AUX] element is present immediately below 1°, it will

move into 1°. This ambiguity explains the two main approaches that have been taken in the

^Analytically, the elements that are [+AUX] are the modals, periphrastic do, all uses of be, perfective have, main verb have (British English), and so-called quasi-modals, dare, need, etc. See Huddleston 1980, Quirk et. al. 1985, and Warner 1993:11 for criteria and lists of members.


70

literature regarding what the licensing element is in VPE structures: 1° or X °[+A U X ].

Lobeck’s work of course represents the former approach as do Zagona 1988a, 1988b and

Lopez 1994. Gazdar, Pullum, and Sag 1982 and Kim 1995, in contrast, take the left-

context element of VPE to be a [+AUX] element. What both approaches share is the

assumption, formulated in (104), that the licenser is an overt, zero-level category. Most

works that have considered the syntactic licensing requirements on elided VPs (e.g. Sag

1980; Gazdar, Pullum, and Sag 1982; Lobeck 1987, 1995; Zagona 1988a, 1988b; L6pez

1994) adopt such a stance. I take it to be relatively uncontroversial.

(104) Head Condition on VP Ellipsis Licensing

An elided VP must be licensed by a morphologically realized head

In (105), I summarize these two hypotheses regarding the identity of a null VP licenser.

Both conform to the Condition in (104). The Infl analysis takes an 1° node overtly filled by

a morphologically-independent lexical item to be the licensing element for a null VP. The

[+AUX] analysis asserts that the licenser is a head specified [+AUX].

(105) Two Hypotheses about VPE Licensers

1) In fl HYPOTHESIS: the licenser is a non-affixal element in 1°

2) [+AUX] HYPOTHESIS: the licenser is an element with the feature [+AUX]

A crucial case for distinguishing between the T-licenser analysis and the [+AUX]-licenser

analysis will be one in which one or the other, but not both, of these elements is present.

As stated previously, finite clauses are of no help because both elements are always

simultaneously present in any grammatical VPE example. In English, being in 1° entails

[+AUX] but the reverse does not obtain—[+AUX] elements do not always move to 1°, even

when it is not otherwise overtly occupied. Chapter 2 demonstrates that this situation holds


71

in subjunctive clauses. Thus, subjunctive complements and their interaction with VPE are

investigated below. What we will see is that ellipsis is generally impossible despite the

presence of a [+AUX] auxiliary. This observation suggests that the licenser is not correctly

characterized as [+AUX] since the presence of a [+AUX] element is not sufficient. The Tnfl-

licenser hypothesis predicts the unavailability of ellipsis since 1° is not filled. Surprisingly,

the introduction of an overt inflectional element, not, rescues the ungrammatical patterns,

further supporting the Infl hypothesis. The data thus argue that the necessary element for

licensing VPE is 1°, supporting Lobeck’s general approach. The presence of X °[+ a u x ] is

not sufficient. Upon reaching this result, the data will require that we be more precise about

the identity of the licensing element.

Subjunctive complements, italicized in (106), are found with a limited set of English

predicates such as insist, demand, be important, be necessary, etc. and they are

characterized, at least in the present tense, by a verb phrase headed by a verb in its bare

form.

(106) a. We insist that Bob take the leftover vodka with him.

b . It is important that you be punctual.

c. It is necessary that the foundation be inspected before we proceed.

Subjunctive clauses are the right kind of test case for investigating the identity of the

licenser of an elided VP because, as I will immediately demonstrate in previewing data

from chapter 2, subjunctive 1° is never overtly filled. None of the [+AUX] elements found

in 1° in finite clauses—modals, periphrastic do, and raised auxiliaries—are permitted in

subjunctive 1°. Auxiliaries may indeed be present in subjunctive complements but they do

not occupy 1°.


72

(107) shows that modals are impossible in subjunctive clauses.

(107) a. *He demanded that the successful candidates can speak German.

a. *The police require that the spectators must stand behind the barricade.

Periphrastic do is also not permitted in subjunctive complements. Do is neither necessary

nor possible with emphatic affirmation, (108), or negation, (109). It is well known that

negation in subjunctives is expressed with not alone, (110).

(108) a. ?*Contrary to what the polls say, we suggest that Jimmy do run for re-election.

b . ?*I requested that she do be more assertive as she is quite competent.

c . ?*Mom demanded that you do be careful.

(109) c. *Who suggested that hedo not don’t

doesn'tact so aloof if he hopes to find a wife?

d . * Jack asks that we j ^ Q ^ j cut down his bean stalk just yet.

(110) a. Who suggested that he not act so aloof if he hopes to find a wife?

b . Jack asks that we not cut down his bean stalk just yet.

c. The sign requests that one not be loitering during curfew hours.

d . The librarian insisted that the book not be returned in shabby condition.

e . The queen desires that you nor be inattentive.

The third class of elements that may occupy 1° in finite clauses, the auxiliaries have and be,

are found in subjunctive clauses, in (111) and (112), respectively. Chapter 2 argues that

these auxiliaries are still in VP, however, and have not moved into 1° as they do in finite

clauses.


73

(111) a. After eating, it is imperative that one have waited at least an hour before going

swimming.

b . The committee asks that you have studied the manuscript beforehand.

(112) a. We insist that from now on you be waiting outside,

b . The public demands that buses be on time.

I briefly review the evidence here. Two diagnostics for verb movement are word order with

respect to sentential negation and sentential adverbs. Auxiliaries that have undergone verb

movement are able to precede either of these elements while auxiliaries that are still in VP

cannot. Since auxiliaries do not raise in subjunctive clauses, perfective have and be in

subjunctive complements do not grammatically appear to the left of sentential not, in (113),

or sentential adverbs, in (114). Similar finite clauses, in which the auxiliaries do raise, are

given for comparison and are fully grammatical. In each example, the sequence

auxiliaryA/iot/adverb is italicized.

(113) a. *In the interest of matrimonial bliss, the counselor suggests that you be not

keeping secrets from your wife.

(cf. We want to believe that you are not keeping secrets from your wife.)

b . *The association urges that he be not examined by that quack.

(cf. He was not examined by that quack.)

c. *Humility requires that one be not proud.

(cf. They are not proud.)


74

d . (?)My parents suggested that the baby sitter have not left a mess in the kitchen

for them to clean when they get back.20

(cf. Fortunately, the baby sitter has not left a mess.)

(114) a. *The sales manual requires that all agents be definitely paying attention to the

customers’ complaints and taking note of them during the exchange.

(cf. The agents were definitely paying attention to the customer’s complaints and

taking note of them.)

b . *It is recommended that you be normally approved by the committee before

coming to the seminar.

(cf. Participants are normally approved by the committee before coming to the

seminar.)

c . *Protocol requires that the attendees be normally standing when the guest of

honor walks in.

(cf. Attendees are normally standing when the guest of honor walks in.)

d. ?It is mandatory that everybody have certainly read at least the introduction.

(cf. Everybody had certainly read at least the introduction.)

20Johnson 1988 demonstrates that the behavior of negation with respect to perfective have in subjunctives is exceptional and examples like this one with not following have are sometimes acceptable. One interpretation of the data is that have is higher in the structure than be in comparable examples. The sentential adverb example in (i) that will be introduced in the text shortly supports the same conclusion. For some people, it is not completely ungrammatical either, hence the ? judgment.

(i) ?It is mandatory that everybody have certainly read at least the introduction.(cf. Everybody had certainly read at least the introduction.)

The subjunctive data are part of a larger body of facts, discussed in Lobeck 1987 and Johnson 1988, pointing to the inadequacy of treating perfective have structurally the same as auxiliary be. The phenomenon is further discussed in chapter 2.


75

I conclude that 1° is never overtly filled in subjunctive clauses. I will claim in chapter 2 that

it is filled with an independent zero subjunctive modal, following Roberts 1985:40 and

Lasnik 1995b. The hypothesized structure of a subjunctive complement is (115), where

Msbj is a null subjunctive modal. The structure corresponds to the complement clause

(112)b.

(112) b . The public demands that buses be on time.

(115) CP

C ^ ^ I P

I*

VP

Msbj V PP

that DP

[+aux]I on tune

be

These conclusions interact with predictions regarding the possibility of VPE in subjunctive

clauses. The Infl hypothesis predicts that VPE should be unavailable since no licenser is

present in (115). 1° is not overtly filled. The [+AUX] hypothesis will permit ellipsis because

of the possible presence of a [+AUX] element, be in (115).

Ellipsis patterns in subjunctive complements have not been previously documented to my

knowledge. What we find first is that, unsurprisingly, it is not permitted if there is no head

preceding the elided VP. The data in (116) are thus compatible with the Head Condition on

VP Licensing in (104) and are parallel to the finite clause case, repeated in (83).


76

(116) a. *Kim needn't be there but it is imperative that the other organizers 0.

b . *Ted didn’t want to vacation in Hawaii but his agent suggested that he 0.

c . *We think that Mary should present her case to the committee and we ask that

Bill 0 too.

(83) *John didn’t leave but Mary 0.

Ellipsis is also not permitted, however, when auxiliaries are stranded, in contrast to the

situation in finite clauses seen earlier. This is the crucial case. Even when auxiliaries are

present, (117), the examples are ungrammatical. For comparison, similar finite clause

examples are in (118). Both the relative and absolute grammaticality judgments are

important. Even for those who marginally accept some of the examples in (117), they are

poor in comparison to the corresponding data in (118).

(117) a. *We can’t count on Josh to be waiting for us at the airport so we request that you

be 0 instead.

b . *My running partner is training for the marathon and it’s imperative that I be 0

too.

c. *The bridges were repaired before the engineers could even insist that the

supporting structure be 0 first.

d. *If the whole system is to be on time, it is necessary that the connecting flights

be 0 as well.

e. *By the time Wanda finishes, it is necessary that Bob have 0 too.

f . *When the laborers have come to a decision, it is important that the leader have 0

as well.


77

(118) a. We can’t count on Josh to be waiting for us at the airport so it would be nice if

y°u {woddbe} 0 tastMd-

b . My running partner is training for the marathon and I {sho^ 0 too.

c. The bridges were repaired before the engineers could even show that the

supporting structure must be 0 first.

d. If the whole system is to be on time, then the connecting flights 0 as

well.{ Jjoo 1will have/ 0 10°-

f . When the laborers have come to a decision, it is hoped that the leader

{mighrhave} 0 38 weU'

Examining the structure of the examples in (117) clarifies how this provides an argument

for the Infl-licenser hypothesis and against the [+AUX]-licenser hypothesis. The partly

elided subjunctive complement of request in (117)a has the putative structure in (119).


be 0 instead.

(119) * CP

I’

VP

Msbj V VP[+aux] |

I 0be

that DP

you I


78

If the [+AUX] element be were the licenser of the null VP in (119), the example would be

grammatical, contrary to fact. Under the Infl hypothesis, the lack of an overt 1° element

accounts for the ungrammaticality; there is no licenser. Subjunctive clauses thus indicate

that [+AUX] is not always able to license a null VP.21

At the same time, neither analysis would seem to be able to account for the earlier

observation that not licenses a null VP. It was seen in the discussion of Lobeck 1995 that

not may precede an ellipsis site, repeated in (101). Ellipsis is not in fact always

ungrammatical in subjunctive clauses. The same grammatical pattern with negation results

there as well, (120).

2 'Lobeck 1987, 1995 and Zagona 1988a derive an argument for Infl licensing of null VPs from the phenomenon of Auxiliary Reduction (AR) (Zwicky 1970, Kaisse 1983, Selkirk 1984, Lobeck 1987) illustrated in (i). The finite auxiliaries have, has. and is phonologically reduce and merge with the preceding overt element.

(i) a. John’s been rather melancholy lately, b. We’ve never seen this happen before.

AR is relevant to VPE because, as is well known, VPE is excluded following a reduced auxiliary:

(ii) a. *Pam is writing a movie review and Tim’s 0 too.b. *He says he’s happy and she says she’s 0 too.c. *Joe read the material and I’ve 0 too.

Assuming that such auxiliaries cliticize to the preceding material (Kaisse 1983, Zwicky and Pullum 1983)by moving out o f I* and adjoining to the preceding maximal projection, the ungrammaticality of theexamples follows because I* is transparently not filled. Under Lobeck’s analysis in particular, it does notproperly govern the elided VP. This argument does not distinguish the above two hypotheses, however,because both can appeal to the cliticization process as being the source of the ungrammaticality of (ii),irrespective of the categorial or featural specification(s) of the reduced element. Furthermore, it is notuncontroversial that AR involves syntactic cliticization. Alternative analyses of exist which account for (ii) as a violation of prosodic requirements (Sells 1983, Selkirk 1984). This fact, and others, follow from the theory of metrical phonology (see Sag and Fodor 1995 for an overview). Under this view, the status of the reduced element as I* or [+AUX] is also immaterial.Zagona 1988b provides an argument for Infl licensing based on the behavior of VPE in ro-infinitival clauses. Because I have found much dialectal variation in the crucial data in this complex domain, I have not presented the arguments based on it but, rather, have attempted to present independent arguments for the same conclusion. Zagona’s results support the general line of analysis here.


79



c. Some of the guests tried the appetizers but I think that most did not 0.

(120) a. Kim needs to be there but it is better that the other organizers not 0.

b . Ted hoped to vacation in Liberia but his agent suggested that he not 0.

c. We think that Mary should present her case to the committee but we will ask that

Bill not 0.

Lobeck assumed that not was the licensing element in (101), although the details of the

account were unconvincing. The data in (120) confirm this conclusion, particularly in light

of the minimally different but unacceptable cases in (116) without not. It seems evident

from the examples in (116) and (117) that the subjunctive modal cannot license a null VP;

otherwise, there would be no explanation for the data’s ungrammaticality. The only place

where we can pin the grammaticality difference is on the head not. The subjunctive clauses

with ellipsis have the schematic structure in (121). Not is the licensing head for the null

VP.

( 121)

subject

Msbj Ne§

not

Additionally, ungrammatical examples with auxiliaries parallel to (117) can reasonably be

salvaged with not:


80

(122) a. John was fooled by Geraldo’s smooth sales pitch so it’s imperative that the rest

of us not be 0.

b . Isolated flights can be late but it is necessary that the connecting flights not be 0.

c. If the laborers haven’t come to a decision, it’s important that the leaders not have

0 either.

Not’s dramatic role in licensing null VPs can additionally be seen in other types of non-

finite clauses. Contrasts paralleling the above are given in (123) and (124) for small

clauses, from Williams 1994, and tenseless why suggestions, respectively. I conclude that

not is implicated as a potential VPE licenser.

(123) a. I consider Bill intelligent and I consider Sally not 0.

b . *1 consider Bill intelligent and I consider Sally,

(124) [You could be Casper the Friendly Ghost for Halloween.]

a. Yeah, why not (be) 0!

b. *Nah, why be 0!

This ability of not to license a null VP provides a second argument against the [+AUX]

hypothesis. Both analyses must admit not as a possible licenser, requiring a revision to the

original hypotheses:

(125) Two Hypotheses about VPE Licensers (revised)

1) INFL HYPOTHESIS: the licenser is a non-affixal element in 1° or Neg0

2) [+AUX] HYPOTHESIS: the licenser is an element with the feature [+AUX]

or Neg°

The statement regarding possible licensers under the [+AUX] hypothesis is a disconnected

list: Neg” or an element with the feature [+AUX]. These two elements do not form a natural


81

class and make no predictions about what other elements might also be licensers. This is

clearly an undesirable account. On the other hand, the licensers under the Infl hypothesis,

1° and Neg°, do form a natural class. On an informal level, they are both functional

inflectional heads. One could reasonably predict whether other heads will be licensers.

The situation we arrive at is that 1° and Neg0 license a null VP while V°[+AUX] does not.

In order to characterize the actual licensers more exactly, it is necessary to make a

distinction between these two classes. When we examine the structure of a typical clause,

in (126), the necessary characterization is clear enough, although perhaps somewhat

difficult to formalize: the possible licensers belong to the inflectional layer, which is made

up of the projections between CP and the topmost lexical projection of the clause. By

lexical projections I mean those which are traditionally decomposed by the features

[±N ,±V ]: AP, VP, NP, and PP.22 (127) defines an INFLECTIONAL PROJECTION. Their

heads are licensing elements in VPE contexts.

22Numerous other phrases have been claimed to exist in this inflectional region, among them agreement, mood, aspect, tense, and polarity. If their existence were assumed, one would expect that they too could license null VPs. On agreement and tense phrases see Pollock 1989 and the large body of work that has resulted from his proposal. Laka 1990 proposes the existence of a polarity phrase ZP which subsumes negation and affirmation. If Laka’s is right about emphatic so being a realization of Z* that alternates with Neg’, then the observation that so appears to license an elided VP is correctly predicted, in (i).

(i) Billy doesn’t like artichokes. Yes, he does so 0 !

A mood phrase is also found in Laka 1990. Aspect phrases have also been proposed in the literature.


82

(126)

subject I’

Neg VP [+aux]

V VP[+aux]

V XP

(127) Inflectional Projection

An inflectional projection a is an X-compatible projection such that

1) a is the complement of C°, or

2) a is the complement of an inflectional projection and is not a lexical

projection

With this definition, I offer the preliminary formulation of the licensing condition on VPE

in (128). In order for a null VP to be licensed it must be c-commanded by an independent,

morphologically-realized, inflectional head. The Licensing Condition in (128) incorporates

the Head Condition of (104) and the Infl hypothesis of (125).

(128) Licensing Condition on VPE (preliminary version)

An elided VP must be c-commanded by an overt, non-affixal inflectional

head

In this section using English subjunctive data I have shown that the an Infl-licenser analysis

for VPE is superior to one which takes the licenser of null VPs to be an element with the

feature [+A U X]. In the following section I turn to an investigation of the structural

relationship that holds between the licenser and the ellipsis site. The data repeated in (92),


83

in which the licenser and ellipsis site are bold faced, indicate that it is not a completely local

one. In (92)b in particular, we can see that maximal projections may intervene between the

licenser could and the null VP 0. The two are not adjacent. Under normal definitions then,

the relationship cannot be either of the local relationships commonly assumed: government

or sisterhood. There is nevertheless a close connection between the two and I address this

issue below.


Spanish too.

b . John could have been studying Spanish and Bill could have been 0 too.



5.2.2. The Structural Relationship

While (128) is not a completely local constraint since the licensing head and the null VP

need not be sisters given the data in (92)b and c, it must also not be unconstrainedly non

local in terms of the material that may intervene between the two. For example, we

obviously do not want a modal in the root clause to license an otherwise illicit null VP in a

lower complement:

(129) *Picasso is painting the walls but the landlord shouldn’t know [cp that I asked

[cp that he 0 ] ]

The only projections that can appear between the functional head and the ellipsis site are

those of auxiliaries. There is clearly a precise kind of locality and I would like to explore an

account of it in terms of Abney’s (1987) concept of a s(emantic)-projection and

Grimshaw’s (1991) notion of extended projection. I develop an analysis of English


84

clauses, based on Aissen 1995, which integrates auxiliaries into the system and permits

them to form an s-projection with the immediately dominating inflectional projections and

the main predicate they introduce. The ultimate goal is to have a situation in which we can

reasonably allow an inflectional projection to “see through” these optional verbal

projections to license null VPs.

The intuition behind the proposals in Abney 1987 and Grimshaw 199123 is that certain

orderings of functional and lexical projections are widely pervasive and form a cohesive

unit while other head-complement combinations are completely unattested. In the verbal

domain, clauses of the form CP/IP/VP are commonly assumed but other combinations do

not appear: *CP/VP, * IP/CP/VP, etc. The functional head C° always takes an IP

complement and 1° a VP complement, while the syntactic complement of the lexical head V°

has no such restriction. V° may select CP, EP, VP, DP, or PP: various clausal and non-

clausal arguments. Abney argues that there are syntactic relations between all heads and

their complements but that the relationship between a thematic lexical head and its

complement is fundamentally different from the one between a functional head and its

complement. This latter relationship Abney labels F(UNCnONAL)-SELECTION and its main

property is that it is typically unique and predictable. Functional heads will be distinguished

by Abney’s feature [+F]. The functional elements C° and 1° are clearly [+F] and f-select IP

and VP, respectively. I will suggest that this is also a property of V°[+AUX] and that they

too are [+F] functional heads.

23I will more closely follow the exposition in Abney 1987 because of its wider generality. Grimshaw’s system is intricately dependent upon categorial features which ultimately introduce unnecessary complexity for my purposes.


85

This syntactic predictability of the complement of functional heads corresponds to semantic

predictability in some sense as well. Abney describes f-selection as the “passing on” of the

descriptive content of the complement and it is clear, in an informal way, that the

combining of a functional head with its complement does not result in a new semantic entity

in the same sense that combining a lexical head with its complement does. For example,

taking the noun phrase the daisies to eat the daisies via verbal complementation is in some

way different than going from eat the daisies to didn ’t eat the daisies or might eat the daisies

which involve f-selection.

Abney encodes the semantic transparency of f-selection by recognizing S(EM A NTIC)-

PROJECTIONS in addition to the more usual C(ATEGORY)-PROJECTION. A c-projection is

the familiar projection of a head that shares categorial features: the c-projections of V° are

V’ and VP, the maximal c-projection of 1° is IP, and so on. The s-projection of a head, on

the other hand, is an extended path along which the head’s descriptive content is passed on.

That is, in the typical case it will be a c-projection plus any functional c-projections that

dominate it. The maximal c-projection of V° is CP, that of 1°, CP as well. Formally, we

have (130).

(130) S(emantic)-Projection (Abney 1987:57)

P is an s-projection of a iff

1) P = a, or

2) p is a c-projection of an s-projection of a , or

3) p f-selects an s-projection of a

Abney’s term s-projection corresponds to Grimshaw’s extended projection. They differ

only in how they are composed. A typical verbal s-projection is bold-faced in (131); its

head is V j0. Under the definition in (130), the complement of this verb is not in the


86

s-projection because it is not f-selected. A verb V2° that takes this CP as a complement will

also not be part of the s-projection for the same reason. It will constitute the head of a

separate s-projection.

(131)

(132) is a structure that incorporates auxiliaries into the system in a way that seems natural.

I hypothesize that auxiliaries are functional heads that f-select their complements and they

are part of the s-projection formed from the main verb which they dominate. Aissen (1995)

formulates a parallel proposal for Tzotzil auxiliaries within Grimshaw’s system. Formally,

the auxiliaries have and be are [+F].24 We can more generally capture this claim with a

feature redundancy statement: [+AUX] id [+F]— if something is specified [+AUX] then it is

also [+F]. With respect to the property of selecting a unique complement, the auxiliaries

pattern largely as expected. Only copular be is exceptional since it takes AP, VP, DP, and

PP.25 The remaining auxiliaries transparently f-select only VP.

24Abney himself (1987:63) classifies auxiliaries as [-F], although it is not clear why.25Gazdar, Klein, Pullum, and Sag 1985 argue that copular be selects an XP complement unspecified for category status. It need only be a predicative expression. Such an approach might permit us to view be as selecting one kind of complement.


Neg VP [+aux]

V[+aux]

VP

V XP

It is not unreasonable to claim that auxiliaries are functional in nature. Abney 1987 ascribes

the properties in (133) to functional elements, although he stresses that they are not

criterial. In addition to selecting a unique complement, auxiliaries also satisfy several of the

other properties. The match is not exact; however, it is suggestive and the hypothesis is

worth exploration. Auxiliaries are at least ‘weakly’ functional. At the same time, their

categorial specification of V° recognizes their verbal status (Ross 1969, Pullum and Wilson

(133) Characteristics o f Functional Elements (Abney 1987:64)

1) functional elements constitute closed lexical classes2) functional elements are generally phonologically and morphologically dependent3) functional elements permit only one complement, which is generally not an

argument. The arguments are CP, PP, and DP. Functional elements select IP, VP, NP.

4) functional elements are usually inseparable from their complement5) functional elements lack what I [Abney] will call “descriptive content”. Their

semantic contribution is second-order, regulating or contributing to the interpretation of their complement. They mark grammatical or relational features, rather than picking out a class of objects.

Although such a proposal might seem contradictory within current syntactic frameworks

because I am claiming that a head is both lexical and functional, it need not be. It certainly

has a natural interpretation in theories of grammar which do not adhere to categorial

1977).


88

divisions (for example Lakoff and Johnson 1980, Lakoff 1987). Heine 1993 specifically

considers this issue with respect to auxiliaries and argues for a continuum going from full

verbs on one end to grammatical tense-aspect-mood markers on the other. Auxiliaries fall in

the middle of this analog chain. While we do not need to go so far as to accept that there are

no categorial distinctions in syntax, there seems to be nothing contradictory about

maintaining that lexical elements have one set of properties and functional elements have

another set of properties and that some elements, auxiliaries, have both sets of properties.

This is coherent as long as the properties are not inherentiy contradictory.

If we accept that auxiliaries are functional categories, their inclusion in a clause does not

create a new s-projection. This allows a straightforward revision of the requirement on

VPE that captures the desired degree of locality: the licensing inflectional head and the

empty category must be in the same s-projection.

(134) Licensing Condition on VPE (final version)


head within the same s-projection

In the final section, I demonstrate the empirical adequacy of the condition in (134) for the

range of data that has appeared thus far.

5 .3 . Confirmation of the Licensing Condition

The data used to explore Lobeck’s analysis of VPE are correctly accounted for by the

Licensing Condition in (134) with no additional stipulations. The examples are repeated in

large part below. I evaluate the data with respect to the two requirements inherent in the

condition: 1) that an appropriate licensing element be present and 2) that it be within the

s-projection of the null VP.


89

Both parts are clearly satisfied in (82) and (92)b through d. The c-commanding licenser in

all cases is the modal or form of do located in the inflectional head 1°. The null VPs are also

part of the extended projection of the inflectional head as required. In (82) and (92)d, they

are directly f-selected by 1°, creating an s-projection. In (92)b and c, c-projections of the

auxiliaries intervene but they are still part of the same s-projection because they are

functional heads by hypothesis. (92)b has the structure in (92)e with the s-projection bold

faced.

(82) John didn’t leave but Mary should/did 0.


Spanish too.

b. John could have been studying Spanish and Bill could have been 0 too.



e. IP

DPZ ZBill I

[+F]I v

could [+aux,+F]I v

have [+aux,+F]I

been

(83)a is ungrammatical because it fails the licenser requirement. There is no overt

inflectional head to serve as licenser, as can be seen in (83)b.


90

(83) a. *John didn’t leave but Mary 0.

b . *John didn’t leave but [jp Mary [r I[+tns] [yp 0 ] ] ]

Lobeck recognizes that T-to-C0 head movement, in which 1° was filled at an earlier

syntactic level, must not count to rule out examples where there is a null VP. (84)a, shown

with its structure, must be licensed according to the Licensing Condition. It is if we use the

assumed definition of c-command and the condition on domination, repeated below. 1°

c-commands the null VP because the first category dominating it is CP. C° does not

dominate 1° because not every segment dominates 1°. The elided VP and the licenser are

still within the same s-projection. This is indicated by the bold-facing of the s-projection.

(84) a. John didn t leave but should/did Mary 0?

C DP

should/ Mary t; did

(74) C-Command

a c-commands P iff a does not dominate P and every y that dominates a

dominates P

(76) Condition on Domination

a is dominated by P only if it is dominated by every segment of P

Examples with raised auxiliaries, (85), are similarly accounted for. The licensing element is

the auxiliary which has moved into 1°. Again, the null VP and the licenser are transparently

in the same extended projection.


91

(85) a. Mary is leaving and John is/has 0 too.

DP

John I

is/has

The Licensing Condition also correctly accounts for Bresnan’s examples in which VPE

fails with verbs that take VP complements, repeated in (86) and (90). What is going to

crucially rule out these examples is the restriction that the licenser and the null VP be in the

same s-projection. In such cases, they are not, accounting for the ungrammaticality. This

can be seen by examining the structure of a typical case in (91) corresponding to (90)b.

(86) a. ^Because Mary continued 0, John also started speaking French,


(90) a. *Because Mary must continue 0, John also started speaking French.

b . *If fire begins pouring out of the building then smoke might begin 0 too.

(91) * IP

smoke I

might V

Part of the Licensing Condition is fulfilled in (91): there is an overt inflectional head,

namely, the modal in 1°. What does not obtain is that it is in the same s-projection as the

null VP. The head of the s-projection to which the modal belongs is the lexical verb begin.


92

Because the verb is not functional, [+F], its VP complement is not included in the

s-projection. The null VP constitutes a separate s-projection. Thus, the VP complement to a

non-functional, lexical head will never be able to elide independent of whether or not there

is a licenser available. The account of Bresnan’s original examples in (86) is the same. The

data are correctly predicted.

Lastly, the Licensing Condition accounts for the ellipsis data involving not. In finite

clauses, (101), there are two licensers, the modal and the Neg head not. Both inflectional

heads are within the s-projection of the elided VP.



c. Some of the guests tried the appetizers but I think that most did not 0.

The subjunctive data, repeated in (120), was the critical test case because the overt modal is

not present. The only available licenser is not. The examples are grammatical because there

is a licenser and it is within the s-projection of the null VP. (135) indicates the structure of

the elided clause in (120)c.

(120) a. Kim needs to be there but it is better that the other organizers not 0.

b . Ted hoped to vacation in Liberia but his agent suggested that he not 0.

c . We think that Mary should present her case to the committee but we will ask that

Bill not 0.


What we see is that the descriptive Licensing Condition in (134) correctly predicts the core

VPE data.

5 .4 . Conclusions

Before summarizing, I offer some commentary concerning the Licensing Condition and its

status in the theory.

First, it should be observed that the Licensing Condition readily extends cross-linguistically

in a way that a constraint based on licensing by the feature [+AUX] does not. While there

has been a great deal of research into the possible universal status of the category AUX

(Steele et. al. 1981), the hypothesis is by no means uncontroversial and in recent theorizing

has been largely replaced by the introduction of functional heads like I(nfl) and T(ense).

Summarily, it seems evident that not all languages have a syntactic class of [+AUX] heads.

The [+AUX] hypothesis makes a more modest prediction regarding the cross-linguistic

presence of [+AUX] elements: if a language has VP ellipsis then it will necessarily have a

class of [+AUX] elements to license them. But even this narrower implicational claim does

not seem to hold. In at least two languages, Irish (McCloskey 1991) and Hebrew (Doron,

1990, Goldberg 1996), VPE is claimed to be fully productive after main verbs which have

raised to an inflectional head, in the absence of an auxiliary. Auxiliaries in these languages


94

are not necessary to license a null VP and we have already seen that in English they are not

sufficient. The subjunctive data, (120), in which negation served as the licenser of the null

VP demonstrated that ellipsis succeeds in the absence of any verbal element. The larger

picture thus further suggests that auxiliaries are not implicated in the licensing of null VPs.

To maintain the [+AUX] hypothesis, one would be forced to the alternative that the licensing

of null VPs is done on a language-specific basis: the [+AUX] hypothesis is correct for

English but not applicable to other languages.

On the other hand, analyses of most languages postulate an organization of phrase structure

in which lexical projections are dominated by an inflectional layer. Under such a scenario,

VPE is in principle possible and is independent of the existence of a class of auxiliaries in

the language. It is dependent upon the presence of an overtly occupied inflectional head, the

situation claimed to hold for both Irish and Hebrew, mentioned above.26 As a result, if the

proposal is correct, it explicitly supports a conception of clause structure in which there are

inflectional heads such as 1° in the syntax. Without such an assumption, the licensers of

null VPs in each language are a potentially unconnected set of elements which seem to have

no unified description either within the language or across languages more generally. As in

the past, the need to list members involved in a particular syntactic phenomenon can

indicate that a generalization is being missed. Under the Infl analysis, the

description/generalization is claimed to be maximally simple: the unifying property is that

they occupy an inflectional head position.

26Cross-linguistically, this cannot be the only requirement for ellipsis to be present. As Lobeck 1995 demonstrates, languages in which all verbs move to 1°, such as French (Emonds 1978, Pollock 1989), may still lack VPE. I have no proposal for predicting whether or not a language whose VPs meet the structural aspects of the Licensing Condition will actually permit null VPs.


95

Second, implicating inflectional heads in the formal licensing of ellipsis has potentially

positive consequences for syntactic analyses of ellipsis embedded within split Infl clause

structures (Pollock 1989 and most recent work within the Minimalist Program). It is

permitted that VPE may actually be, more generally, ellipsis of categories larger than VP,

provided that these categories are complements of certain inflectional heads. This is a

desirable result in that numerous recent proposals claim that elements such as main verbs

and direct objects that are traditionally thought to be within VP have actually moved outside

of it overtly (Johnson 1991; Chomsky 1993; Bowers 1993; Koizumi 1993; Kratzer 1993;

Runner 1994,1995; Lasnik 1995a). If direct objects are outside of VP at surface structure,

as claimed, but they are still elided under ellipsis, then the target of VPE cannot be a VP. It

must be some larger constituent which contains the material commonly present in a deep

structure verb phrase. The present proposal admits this possibility.

At the same time, one would want to ask why inflectional heads should be implicated in the

licensing of empty VPs. The Licensing Condition on VPE is in appearance a formal

syntactic requirement which does not, as yet, follow from anything in the theory. It is not a

semantic identification condition in the sense that it helps to recover the meaning of the

elided VP. It belongs to the syntactic module proper unless motivation can be found for its

existence. I will leave (134) as it stands, recognizing that, while it seems necessary, it is

nevertheless somewhat mysterious.

Third, the introduction of a notion like s-projection into the formulation of syntactic

constraints seems necessary in a theory of syntax that espouses the presence of numerous

functional heads. A recurring difficulty in recent syntactic analyses is the observation that

these projections, although present, are nevertheless “invisible” to many syntactic

mechanisms: movement, selection, agreement, and locality phenomena (Grimshaw 1991).


96

Acknowledging the theoretical status of s-projections permits a straightforward way of

dealing with this theoretically-induced difficulty.

Fourth, it is important to place this proposal within the context of previous analyses of the

syntactic licensing of VPE, beyond Lobeck’s work. In this way it will clearly be seen as

support for and an extension of existing research programs. Neither piece of the Licensing

Condition, the identity of the licenser nor its locality relationship with the null VP, is

without grounded precedent. The intuition that the element responsible for licensing null

VPs is an inflectional head has a long tradition in the earlier works of both Lobeck (1987,

1991, 1992) and Zagona (1982, 1988a,1988b), as well as Chao (1988). Roberts 1990 is

particularly explicit about the unique role of overt 1° in licensing the related movement

phenomenon of VP Fronting. Regarding inflectional heads more generally, Lobeck 1995 as

we have seen recognizes the importance of Neg° in VPE licensing. Lopez 1994 capitalizes

upon the role of Neg° and argues that the licensing element for VPE in both English and

Spanish is always precisely I \ the head of a polarity phrase IP (Laka 1990) which

includes negation as one of its instantiations. Lopez’s (1994) proposal shares with this one

the recognition that other inflectional heads besides 1° can license a null VP.

The various proposals have differed more radically on how the licenser and the ellipsis site

are related. In Lobeck 1995 we saw that hypothesizing movement of the auxiliaries into

1° permitted the licenser and the null VP to be connected by the very local relationship of

government. Zagona 1988a, 1988b too propose that government is the proper relationship;

however, the work achieves government over intervening maximal projections through

coindexation and consequent government by a chain rather than movement. A primary

intuition in the proposal is that auxiliaries are transparent to government by 1°. Lastly,

Lopez 1994 assumes as well that government is the relevant relationship, adopting


97

Zagona’s insight. In Lopez’ system, government across the auxiliaries is permitted by

appealing to the notion of projection system, fundamentally an s-projection consisting of

the main verb phrase and dominating auxiliary verb phrases, and stipulating that maximal

projections are not barriers to government inside a projection system. The driving force

behind Zagona’s and Lopez’ proposals is to allow VPE licensing to be reduced to the

Empty Category Principle. This brings us to the account presented here, repeated below,

which can be seen as a fusion of many of the basic claims of the previous accounts.

(134) Licensing Condition on VPE



A primary difference is that the Licensing Condition is less theory-dependent, neither

appealing to government nor implicating specific functional heads.

Finally, it must be pointed out that the discussion of VPE has been simplified and idealized

to a certain extent. As is well known, there are various additional restrictions on the

auxiliary sequences that permit and do not permit ellipsis. What such cases indicate clearly

is that the Licensing Condition will be a necessary but not sufficient condition in most

dialects of English. I will illustrate two of these additional restrictions below. I have not

attempted to incorporate them into the above account because I believe that they are rather

parochial in nature. The restrictions are neither robust nor common to all dialects of

English. They are thus reasonably handled at a different level of analysis.

Sag 1980, Steele et. al. 1981, and Zagona 1988a observe that certain VPE interpretations

are unavailable where one would not expect them to be based on a naive syntactic view of

VPE. Perfective have creates such a situation. The example in (136) from Zagona


98

1988a: 103 is ungrammatical, at least in some idiolects, on the given interpretation, although

it has another meaning where perfective have is not interpreted in the elided VP. Perfective

have, even when it has not raised into 1°, cannot for whatever reason be interpreted at the

ellipsis site despite being within a VP. Zagona accounts for the unavailable reading in (140)

using syntactic mechanisms.

(136) a. *John might have read that chapter and Bill could 0.

0 = have read that chapter

ok: 0 = read the chapter

It is evident from the example in (137), a text example from the comic strip Peanuts, that

such a restriction cannot be maintained for all speakers. The example contains numerous

examples of VPE and two VP antecedents are available: have been great and be great. They

are bold-faced and italicized respectively. Parallel to the example in (136), only the latter

should be an actual antecedent. In the interpretation of the ellipses in (137)a through e, also

marked by italicization, this is indeed the case. In fact, this is the only sensible

interpretation. In the final frame, (137), the interpretation of the null VP is different and

uses the entire bold-faced VP which includes perfective have. Observe that using only the

italicized VP makes much less sense in the given context. The example is thus of the same

form as (136) and illustrates that whatever accounts for (136) is not a fully general feature

of English VPE. Given that it only holds in certain dialects, or perhaps only under certain

semantic conditions, it is reasonable not to embed its analysis deeply within the Licensing

Condition on VPE.


99

(137) [Snoopy is talking to Woodstock]

a. ‘You and I are a lot alike.. Just a common bird and a common dog.. Of

course, if we had wanted to 0 , we could have been great. .

b . But we didn’t want to 0 . .

c . But if we had wanted to we could have 0 . .

d. But we didn’t need to 0 . .

e. But we could have 0 . .

g. Or could we 0?

A second case illustrating the idiosyncratic nature of VPE is what I will call the Stranded

-ing Constraint. Numerous researchers (Akmajian and Wasow 1975, Pullum and Wilson

1977, Iwakura 1977, Huddleston 1978, Akmajian, Steele, and Wasow 1979, Steele et. al.

1981) observe that VPE is ungrammatical when the elided VP immediately follows an

auxiliary in the -ing form. (138) from Akmajian, Steele, and Wasow 1979 illustrate this

with be*be sequences in finite clauses.

(138) a. *John is being noisy and Bill is being 0 too.

b . ♦John was being watched by the FBI and Bill was being 0 too.

c. *John isn’t being very assertive now but he should be being 0.

It also obtains in gerunds, which it was suggested earlier exhibit a kind of VPE. The

examples in (139) are from Akmajian and Wasow 1975 and Iwakura 1977.

(139) a. *Which bothers you more: John’s having been arrested for drug dealing, or

Bill’s having 0.

b . *Which bothers you more: John’s having taken drugs, or Bill’s having 0.

c. *Ford’s being examined by a psychiatrist was just as unreasonable as Nixon’s

being 0.


100

Finally, it obtains in British dialects (annotated by the diacritic £) in which do may be

inserted into the position of the elided VP (Pullum and Wilson 1977:761):

(140) a. £1 didn’t know the answer but John may do.

(cf. I didn’t know the answer but John may 0.)

b . *£I’m not working but Bill may be doing.

c . £1 haven’t seen him but John may have done.

(cf. I haven’t seen him but John may have 0.)

d . *£If you’ re not working you should be doing.

The parochial nature of this constraint is evidenced by the fact that, according to

Huddleston 1978 and Gazdar, Pullum, and Sag 1982, all of the above examples are

grammatical in some dialects of British English. Huddleston 1978:46-47 offers the

additional acceptable examples in (141).

(141) a. When is the building going to be demolished?

£It already is being 0.

b. Is John unkind then?

£Not usually, but he is being 0 at the moment.

c . £Sam was being examined by a psychiatrist at that time and Bill was being

0 too.

d. £He’s always being teased about it. I don’t think he likes being 0.

e. £I’ve been Rex’s mistress for some time now, and I shall go on being 0, married

or not.

Warner 1993 discusses additional restrictions of this sort.


101

To summarize, in this section I have developed a constraint on the licensing of elided VPs

in English based on the Licensing and Identification Condition from Lobeck 1995.1 have

argued for its correctness using data from finite declarative clauses and subjunctive

complements. Chapter 2 shows that it receives further support from the behavior of

imperatives.

6 . Floating Quantifiers

Sportiche 1988 develops a novel theory of floating quantifiers which crucially takes

advantage of the VP-intemal subject hypothesis (Kuroda 1988, Koopman and Sportiche

1991). In this section, I lay out Sportiche’s basic analysis of Quantifier Float (QF) for

English as far as it will be relevant for the syntactic issues under consideration. The

phenomenon of QF provides a valuable window on clausal syntax and will be recruited in

the investigations of imperative subjects and clause structure in later chapters.

Sportiche’s theory of QF is more appropriately described as quantifier stranding. Floating

quantifiers like all and both may be stranded in any position that the noun phrase they

modify passes through in the derivation. Floating quantifiers should then be canonically

associated with derived arguments—any argument that is not in its deep structure position

at surface structure. QF will be particularly relevant for subjects if Bobaljik and Jonas 1996

are correct in claiming that all subjects must move overtly. Since subjects are assumed to

originate in the specifier of VP and to move to a higher functional specifier which is their

surface position, intermediate specifiers are potential stranding positions for floated

quantifiers. I consider the details of the proposal below.

Following Shlonsky 1991, the structure of a noun phrase like all the answers is as in

(142). All and both are heads of category Q° which take full DP complements.


102

(142) Elsie has all the answers

aU D

the answers

Quantifier stranding is stranding of the QP by movement of the DP. It proceeds via the

specifier of QP, which naturally accounts for the agreement between the DP and its

modifying quantifier that many languages show (Shlonsky 1991, Merchant 1996). This

intermediate landing site also allows the extraction from QP, an operation that would

otherwise violate Subjacency if it occurred directly from the complement of Q° position.

The derivation for The rebels might have all fled is in (143) and that for The rebels might all

have fled is in (144). In the former, the quantifier is stranded in the base, specifier of VP

position. In the latter, it is stranded in an intermediate specifier of VP position.27

27It should be noted that under the system o f clause structure adopted here, there is no derivation for The rebels all might have fled. This is because there is transparently no specifier position between the subject and the modal, as reference to the above structures will show. One might thus interpret this as an argument for an exploded clause structure in which the unitary IP is decomposed in to two or more functional projections, typically Agrs P and TP (Pollock 1989), and subjects occupy the higher specifier position (spec,Agrs P). There would then be a specifier position, the lower specifier (spec,TP), for the quantifier. However, it is often assumed that the specifier of TP is unavailable in English (Chomsky 1993, 1995, Bobaljik and Jonas 1996). If this is indeed the case, the example is still unexplained.


103

(143)

(144)

The rebels might have all fled.

IP

DPu I’

the rebels I VP

might tk V’

V VP

have QPA

Q’A

Q I

an

The rebels might aU have fled.

IP

V5I

VI

fled

DPb I’

the rebels I VP

V’might QP;A

Q’A

Q tk have t;I

all

VP

V’I

VI

fled

These two cases clearly show that when a subject is separated from its modifying

quantifier, it can no longer be in its base position.

Under a stranding analysis, The rebels all fled, in (145)a, is also an example of the same

phenomenon. Constituency tests in (146)—conjunction, clefting, placement in object

position, and topicalization—all show that the string the rebels all is not a constituent. The


104

analysis given to it is (145)b. Most generally, when a full DP precedes its associated

quantifier {all or both), we can conclude that the quantifier has been stranded by leftward

movement of DP.

(145) a. The rebels all fled,

b. IP

DPk r

the rebels I VP

QP V’A *k Q’

1V

A |Q *k1

all

fled

(146) a. *The rebels all and their leader fled.

b. *It’s the rebels all who fled.

c. *We hid from the rebels all.

d. *The rebels all, we saw.

The one exception to this claim is pronouns. They may optionally appear to the left of all or

both and still form a constituent with the quantifier (Postal 1974, Maling 1976), in (147).

Analytically, the pronoun is able to move into the specifier of QP and not proceed any

farther (see Shlonsky 1991 for discussion). It thus forms a constituent with the quantifier

as in (148).

COORDINATION

CLEFTING

OBJECT PLACEMENT

TOPICALIZATION


105

(147) a. The queen invited you both,

b. We sold them all.

(148) QP

D P ^ >

D Q tkI I

you all

The sequence you all, which appears often in imperatives, is thus not an indication of QF

and will not be used in demonstrating the implications of QF.

An important consequence of Sportiche’s analysis for argumentation in later chapters is that

QF is a diagnostic for movement. A non-pronominal noun phrase cannot be in its base-

generated position if it precedes a modifying quantifier. The appearance of a noun phrase to

the left of all or both necessarily indicates that the noun phrase is no longer in its base

position, spec,VP.

The primary alternative to Sportiche’s analysis is one in which the quantifiers are treated as

adverbs (Klein 1976, Dowty and Brodie 1984, Kayne 1984, Doetjes 1992, Baltin 1995,

Bobaljik 1995). They would have roughly the distribution of E-adverbs discussed earlier.

If Sportiche’s analysis were rejected in favor of an adverbial approach, then the arguments

that appeal to it would be reduced to ones relying on adverb placement. Because of the

novel and convincing nature of the stranding analysis, I will adopt it and point out the

consequences that it has for imperative syntax.


106

Chapter 2

Verb Raising

1. Introduction

As the starting point for the analysis of imperative clause structure, this chapter investigates

the phenomenon of Verb Raising, or V°-to-I° movement, in English. In finite clauses, the

position of verbal heads, in V° or 1°, has a wide range of highly visible syntactic

consequences and this leads to the expectation that the determination of the position of main

verbs and auxiliaries in imperatives will usefully inform the structure of that clause type as

In particular, a primary motivation for looking at this domain is that it will help to make

sense of an often-made claim about imperative syntax (see for example Quirk et. al. 1985):

imperatives irreducibly differ from finite clauses because they require a form of do even

with the auxiliaries have and be. The contrast is illustrated in (1) and (2). The negative

imperatives in (1) require do with have and be while the negative finite clauses in (2) do not

allow do with the two auxiliaries.

well.

(1)

Do not have ♦Have not > left us a mess to clean up!


107

(2) a. We |

b. They

*do not bel . were not Jf*do not have

| late.

have not j left.

Whether or not a principled explanation for this contrast exists within the theory has

important implications for the syntax of imperatives. If none is forthcoming, then an

exceptional structure and analysis might be warranted. On the other hand, if the contrast

can be reduced to more familiar syntactic ground, then there is little reason to allow

imperative clause structure to differ from standard conceptions, at least in this domain. This

chapter demonstrates that, in current terms, the difference has an explanation in terms of

Verb Raising (VR), whether or not the auxiliaries under consideration undergo head

movement from V° to 1°.

As mentioned in chapter 1, the following state of affairs obtains with regard to VR in finite

clauses: auxiliaries raise from V° to 1° but main verbs remain in situ, in V°. At the same

time, support do appears with main verbs but not auxiliaries. Thus there is the well-known

correlation between the presence of do and absence of V°-to-I° raising. If there is no verb

raising at all in imperatives, as will be shown, the occurrence of do with have and be in (1),

despite not paralleling the behavior seen in (2) with auxiliaries in finite clauses, is not

unexpected. Under the assumption that imperative do is the same support do found in finite

clauses, a conclusion to be argued for in chapters 4 and 6, the fact that auxiliaries do not

raise in imperatives actually predicts the above pattern. Auxiliaries in imperatives will

behave syntactically like non-raising finite main verbs—which they do as (3) shows in

comparison to (1).

(3)


108

The contrast illustrated in (1) and (2) is claimed to not be a significant syntactic

idiosyncrasy. What remains to be determined is only why auxiliaries do not in fact raise in

imperatives.

The chapter is also an investigation of the prominence of VR in the grammar of English

more generally. It looks at the presence or absence of VR in other English clause types

beyond finite clauses where its presence is most evident. A fundamental empirical question

that arises is: where in the grammar of English do we see Verb Raising? An accurate

answer will permit an evaluation of the relative place and importance of VR in the syntax of

English. The conclusion of the chapter with regard to this issue is that VR is confined to a

very small, albeit visible comer of English: finite auxiliaries. Given that it appears nowhere

else, English is, for all practical purposes, a non-Verb Raising language. In this light, the

analytical burden is not to explain why raising is absent in numerous situations, in

particular imperatives, but rather to explain why it occurs in this one instance. Its absence

in imperatives becomes completely unsurprising, supporting the general claim that

imperative syntax is not exceptional.

The remainder of this chapter is organized as follows. Section 2 discusses the history

behind the Verb Raising analysis. Section 3 then develops three diagnostics for the

presence or absence of VR that rely on adverb position, sentential negation, and verb

phrase ellipsis (VPE). Analyses of these phenomena were developed in chapter 1.

Although they are regularly used as diagnostics for Verb Raising, the assumptions behind

their application and interpretation are rarely explicitly stated. One goal of this section is to

formally lay out the conditions under which these phenomena are valid tests for raising.

The three diagnostics are discussed with respect to finite clauses and are shown to yield the

expected results in such cases: only finite auxiliaries undergo VR. Sections 4 and 5 apply


109

the diagnostics to imperatives and subjunctive complements, two clause types that do not

evidence any kind of overt lexical element which might be taken to otherwise block the

movement under investigation. Section 4 considers imperative clauses. The conclusions of

this section build on and confirm Lasnik’s (1981) observation that Verb Raising is

systematically absent in imperatives. Section 5 investigates the syntax of subjunctive

complements, an area that has been largely ignored. It demonstrates that VR is absent there

as well. Section 6 concludes based on the preceding investigations that nowhere else in

English besides finite clauses do we see Verb Raising. This is the primary statement of this

chapter. The result makes raising of tensed auxiliaries in declaratives the exception rather

than the rule and I propose, in agreement with Baker 1991, that, whatever its ultimate

source, it is not a core phenomenon in English syntax. English is fundamentally a language

without Verb Raising (Emonds 1978). Imperatives are a particular instantiation of this

result and the section also discusses the consequences for imperative syntax.

2 . Background

Syntactic research in the 1960’s on English quickly noticed that the modals, perfective

aspect have, and all forms of be collectively behaved differently in numerous contexts

compared with main verbs. As Ross 1969 and Jackendoff 1972 remark, the disjunction modals]

« have repeatedly appeared in transformational formulations, indicating that a be

generalization was being missed. A widely accepted solution to this inadequacy has been

labeled Verb Raising in the literature (see Jackendoff s (1972:77) Have-Be Raising for

English, Emonds’ (1978:165) Finite Verb Raising for French, also Emonds 1976, Pollock

1989, Chomsky 1992, Lightfoot and Homstein 1994a, and references cited therein).

Summarily, Verb Raising accounts for the above disjunction by maintaining that the three

elements occupy the same structural position in the surface syntax, although their individual


110

routes to this position are not necessarily the same. Modals and auxiliaries are taken to

originate in syntactically distinct positions within a structure. The auxiliary have and all

forms of be are main verbs (Ross 1969, Pullum and Wilson 1977, Zagona 1988a). They

are heads of category V8 that project to VP. Modals are of category I(nflection)° (Chomsky

1957, Emonds 1976, Lobeck 1987). Assuming the clause structure in (4), the unitary

behavior and distributional similarities of modals and auxiliaries are captured by giving

them the same surface position, 1°. Modals are base-generated, in 1° and auxiliaries are

moved there. This movement is known as Verb Raising—V-to-I8 head movement in

current versions of Govemment-Binding/Principles and Parameters theory.

(4) IP

subject

complement

In the discussions that follow I will assume the informal characterization of Verb Raising in

(5). Verb Raising is movement of a verbal element from within VP into the head position of

the inflectional projection immediately dominating and outside this VP, illustrated in (4).

For a clausal structure as in (4), this target position will be 1°. Movement to other

inflectional projections, if they are present, may also occur but the relevant piece here is that

Verb Raising is at least raising out o f VP:1

1In a clause structure in which IP is not unitary (Pollock 1989, Chomsky 1991) and all verbs move to at least the lowest inflectional head (e.g. Johnson 1991, Bowers 1993, Koizumi 1993, and others), Verb Raising would need to be characterized as movement to the highest inflectional head, rather than as movement out of VP. The absence of VR would be characterized as movement to at most an intermediate inflectional head (see Rohrbacher 1994:9). If a complex Infl scenario were adopted, the discussion could be modified accordingly.


I l l

(5) Verb Raising

Verb Raising is movement of the highest verb in a stack of one or more VPs

to the lowest inflectional head that c-commands it. The verb raises out of

VP.

Verb Raising is restricted by something like the Head Movement Constraint in (6) (Travis

1984:131, Chomsky 1986:71, Baker 1988) so that it is strictly local. The verb moves to the

next immediate X° that c-commands it. The HMC need not be, and probably is not, a

primitive of the theory. It is useful as a descriptive restriction that appears to hold of head

movement.

(6) Head Movement Constraint

An X° may only move into a Y° which properly governs it

From this it follows that Verb Raising is only at issue for the first, or highest, verb in a

stacked VP structure as was illustrated in (4) and stated explicitly in (5) because lower

verbs will be unable to skip over intermediate verbal heads to land in 1°. In what follows, I

will only be concerned with making generalizations about this topmost verb. No functional

projections will be assumed between stacked VPs. The inflectional layer immediately

dominates only the highest VP.

Emonds 1978 and Pollock 1989 demonstrate that in finite clauses Verb Raising is lexically

restricted in English but fully productive and obligatory in French. Other Romance

languages show VR as well (Lightfoot and Homstein 1994b citing Torrego on Spanish and

Efthimiou and Homstein for Modem Greek, Belletti 1994 and Rohrbacher 1994 for Italian,

and Rohrbacher 1994 citing Rouveret 1989 for European Portuguese). Vikner 1995

systematically investigates Verb Raising in the Germanic languages claiming that it is found


112

in finite clauses in Icelandic, Yiddish, and some Swedish dialects but is absent in standard

Swedish, Norwegian, Danish, and spoken Faroese.

3. Diagnostics

The claim that a particular language or clause type exhibits Verb Raising is usually argued

for on the basis of word order facts: relative positioning of the verb with respect to such

items as adverbs, negation, floating quantifiers, shifted objects, and/or clitics. Once

diagnostics have been developed to establish Verb Raising in a particular clause type,

provided that the diagnostics can be made sufficiently precise, they can be used to identify

Verb Raising in other environments. In this section, I develop and discuss three diagnostics

for verb position in English finite clauses using adverbs, negation, and VP ellipsis. The

first two have been used to argue for Verb Raising in a variety of languages while the third

is largely English specific. The diagnostics establish that auxiliary have, be, and pleonastic

do raise (or are base-generated in 1° in the case of do) in finite clauses but that main verbs

do not. In the sections that follow this one, I apply these diagnostics to imperatives and

other English clause types. What we will ultimately see is that, in the grammar of English,

Verb Raising only occurs with auxiliaries in finite declarative clauses.

The Verb Raising analysis is a consequence of the assumption that all instances of a verb

form, perfective have and be in the case of English, are base-generated as heads of VP. It

should be pointed out, however, that the Verb Raising diagnostics themselves actually

indicate only the surface position of the verb forms. They are silent on the base position,

which is a theory-internal assumption. If it is empirically concluded that a verb form is in 1°

at the surface and it is assumed that its base-generated position is inside VP, then

movement must be invoked to connect the two levels of representation. Thus, although the


113

conclusions will be largely couched in derivational terms (i.e. in terms of movement), it is

not strictly necessary to analyze the facts in this manner. The more theory-neutral results of

the diagnostics are conclusions about the surface syntactic position of the elements under

investigation (although these too are unavoidably also somewhat theory-dependent). Within

the present framework, this point is particularly appropriate for the support auxiliary do. In

the examples below, we will be able to show that at the surface do is in 1°, like the modals

and finite have/be. This is not a commitment to an analysis of pleonastic do in which its

base-generated position is V° and it undergoes raising. In fact, I claim that do does not

originate as a verbal head but, rather, is inserted directly into 1°. Consequently, it is not a

target of VR since it is not a V° but it still satisfies the diagnostics because its surface

position coincides with the landing site of VR.

3 .1 . Adverb Placement

Chapter 1 developed a syntactic analysis of the positional distribution of adverbs in English

clauses assuming the structure in (4). The description of their placement was given as (7)

and (8).



b. left adjunction to I’

c . left adjunction to the topmost VP

(8) Syntactic Distribution ofVP-Adverbs


Adopting these positional restrictions, we can use adverbs as a diagnostic for Verb Raising.

Because VP-adverbs appear only within a main verb phrase, very low in a structure, they


114

are of limited use in the more general case in which we are also concerned with the position

of auxiliaries. The only verbs that VP-adverbs will ever c-command are main verbs. I thus

turn directly to S-adverbs. S-adverbs crucially appear at the IP A T boundary that is of

interest. Since such adverbs can be no lower than the left edge of the uppermost VP,

anything to an S-adverb’s left will be outside of VP. This yields a diagnostic for Verb

Raising. If a verb or auxiliary can appear to the left of an S-adverb then it can no longer be

within VP. It must be outside of VP. The ordering verbAS-adverb, then, is indicative of

Verb Raising. If the order is not permitted, no Verb Raising has taken place; S-adverbAverb

will be the only possible order. This test using S-adverbs confirms the well-known

observations that main verbs do not raise, in (9), but the auxiliaries do, as (10) shows for

have, and (11) shows for progressive, passive, and copula be. (12) indicates that do must

also be outside VP.

(9) a. *The principal talked probably to the parents,

b . The principal probably talked to the parents.

(10) a. She has evidently misunderstood my question.

b . They have probably surrendered by now.

c. The bell boy has certainly taken care of that.

(11) a. The police are probably surrounding the warehouse as we speak.

b . The inhabitants were understandably bothered by the rioting outside.

c . Catherine was apparently a sports fanatic.

(12) a. He DID certainly misunderstand me.

b . Max DID probably behave badly as usual.

c . The butler DID apparently have an alibi.


115

The adverb diagnostic for Verb Raising can be summarized as follows:

(13) Adverb Diagnostic for Verb Raising

A verb has undergone Verb Raising if and only if it may appear to the left of

an S-adverb

3 . 2 . Negation

It was concluded in chapter 1 that English negation is structurally represented by NegP

headed by not which appears between IP and VP as in (14).

(14) IP

subject V

NegP

Neg VP

not

Such a structure makes negation very similar to adverbs in its positional properties and it

can be used to probe Verb Raising in a similar way as S-adverbs were in the preceding

section. Since NegP is outside of VP, it provides a diagnostic for verb movement. Crucial

to this test is the assumption inherent in (14) that the position of sentential negation is fixed.

The unmarked state of affairs will have negation to the left of any element that is

VP-intemal. Thus, any verb to the right of not has not undergone raising out of VP. We

can confirm that main verbs in finite clauses do not raise by observing that only the order

not followed by main verb is grammatical:

(15) a. Sampson did not attend the baby shower,

b . *Sampson attended not the baby shower.


116

The opposite pattern obtains for the set of elements that appear in the inflectional head:

perfective have in (16), the three forms of be in (17) through (19), and pleonastic do in

(20). In each case, the verb appears to the left of negation, unlike the main verb in (15)

which is to the right of it.2

(16) a. *That junker not have/has run in a million years,

b . That junker has not run in a million years.

(17) a. * Kathy not beAvas getting her hair done,

b. Kathy is not getting her hair done.

(18) a. *My koalas not beAvere misplaced by the luggage handlers,

b . My koalas were not misplaced by the luggage handlers.

(19) a. *Mom’s scarf not be/is on the candelabra,

b . Mom’s scarf is not on the candelabra.

(20) a. *The raccoons not did destroy the garbage,

b . The raccoons did not destroy the garbage.

The diagnostic for verb movement that negation provides is stated in (21).

(21) Negation Diagnostic for Verb Raising

A verb has undergone Verb Raising if and only if it can appear to the left of

sentential not

2Such examples would seem to be clear violations of the Head Movement Constraint since the auxiliaries transparently move over the head Neg* to land in I*. One o f two assumptions must be in error. Either the HMC is not a true description of all head movement phenomena or not is not the head of its own projection (see Ernst 1992, Chomsky 1995, Lasnik 1995b for discussion).


117

3.3 . VP Ellipsis

In languages that have it, VP Ellipsis can be used as a diagnostic for Verb Raising since it

is sensitive to whether or not given material, a verb in particular, is contained in a VP.

First, since the interpretation of an elided VP, the target, depends upon a VP constituent

elsewhere in the discourse, the trigger, VPE can readily identify whether or not a given

verb is part of a VP or has moved out. Second, if the complex interaction between I" and

the licensing of VPE outlined in the discussion of VPE in chapter 1 is correct, the

phenomenon provides direct evidence for claiming that an element is in 1°. Below, I

develop these two facets of VPE as diagnostics for Verb Raising.

Consider the consequences of verb raising in an ellipsis antecedent. On a strict view in

which ellipsis involves some sort of identity of lexically overt VPs, if a verbal element in

the antecedent clause has raised out of VP into 1°, it is not contained in a VP constituent and

should not be available to be part of an ellipsis trigger. The target cannot then be interpreted

as containing this element. Conversely, if a verbal element has not raised into 1° and is still

in VP, it is potentially part of a licit VP antecedent and the interpretation of a target VP may

include the verbal element. The available interpretations of a target, then, shed light on the

syntactic structure of the antecedent and, in particular, whether or not verb raising has

occurred. This is seen more clearly by considering the schematic example below. (22)

shows a stack of one or more VPs whose topmost verb is Vj and whose main verb is Vm

(specifiers are not shown). Consider the strings Y j = V i aV 2a . . .AXP and

Y2=V2a . AXP as potential ellipsis antecedents. If Vi raises, it is not contained in a VP

and thus cannot be part of a VP antecedent. Y j may therefore not be a trigger because it

does not correspond to a VP constituent; only Y2 still constitutes a full VP constituent. If,


118

on the other hand, Vi does not raise, it remains part of VP! so it is available as an

antecedent, as is Y2 still.

(22) r

I VP!

Yj is the string VtAV2A. . -AXP

Y2 is the string V2A. . .AXP

As a consequence, if a verb can be interpreted within a target, it cannot have undergone

raising in the antecedent structure. The contrapositive is also true: if a verb has raised in an

antecedent structure, it cannot be interpreted within a VPE target. This observation will

form the basis for half of the diagnostic presented below. Before presenting the diagnostic,

however, I turn to address a potential objection to the above interpretation.

An objection to the above might center on the observation that the antecedent clause still

contains a trace of raising. This trace might be able to stand in for the lexical element which

has vacated VP. If it can, then we would permit the interpretation that the above discussion

claimed would be absent. For example, in (22), the base position of V j contains a trace

after raising which might allow VPj to be an antecedent including Vj in its interpretation.

The general issue to be resolved is the status of traces in elided VPs and whether a trace

might be in some way equivalent to the moved element itself in terms of its interpretability.

Lasnik 1995b explores the issue and it is further investigated in Potsdam to appear. The


119

latter concludes that the answer is dependent upon the type of trace involved; X° traces and

XP (noun phrase) traces behave differently. Specifically, X° traces may not be

independently interpreted in elided VPs while XP traces may. This conclusion allows us to

maintain the view of the interaction of ellipsis and verb raising developed above. In what

follows, I summarize the evidence that supports this claim. See Potsdam to appear for

further discussion.

We can be more specific about what it means for a trace in an elided VP to be interpretable

by considering two issues which characterize the degree of dependence of such a trace on a

phonetically overt antecedent: 1) whether the trace may be free (not bound) and 2) if it is

not free, whether it may have a different binder from the one in the antecedent clause. This

first question is a rephrasing of the issue of whether a trace may stand in for its antecedent

in an ellipsis site. If it can, then it is interpretationally equivalent to its binder and does not

need a binder to be interpreted elsewhere. The second question provides a looser view on

the interpretation of traces. Perhaps the trace must be bound but is able to pick up a new,

distinct antecedent. For X° traces, the working hypothesis above was that the answer to the

first question is ‘no’—an X° trace without its binder has no interpretation. The data to be

presented below indicate that the answer to the second question is also ‘no’—an X° trace

cannot find a new binder in the target clause. In contrast, the answer to both of these

questions for XP traces is ‘yes’, which I illustrate immediately.

First, the data in (23) show that XP traces can be interpreted unbound in elided VPs. In the

examples, a trace of topicalization or w/z-movement in an antecedent VP need not be bound

in the corresponding target clause. In each case, the antecedent VP is bracketed and

contains a trace of A-movement. When this VP appears in the target clause, in the position


120

indicated by 0, the trace is not bound but is nevertheless grammatically interpreted as its

binder in the antecedent. (24) illustrates the structures for (23)a.

(23) a. Shoes, my son refuses to [yp wear t ].

I do 0 too.

b . Linguistics, I [yp like t ].

Yeah, I can’t imagine who wouldn’t 0.

c. Actually, it’s the SPE stuff which they [yp really hate t ].

Well, they used to 0, but I’m not sure they do 0 anymore.

d. John named a country which he wants to [yp visit t ], and given the amount of

traveling he does, I’m sure he will 0.

(24) a. Shoes, my son refuses to wear. I do too.

b.

shoes DP

my son I

refuses to wear t;

In the structure in (24)b, the bold-faced antecedent VP contains a trace of the topicalized DP

shoes. In the target clause, however, there is no binder at all for the trace; rather, it is

understood as the original noun phrase.3 These data thus illustrate that traces of XP

3 Here and elsewhere I attempt to remain neutral as to whether VPE is deletion at PF or reconstruction/copying at LF. By LF, both theories would seem to countenance equivalent representadons for elided VPs. In particular, the trace in the structure above will be present in (at least) the LF representadon of the target clause under both theoretical approaches.


121

movement can be interpreted at an ellipsis site.4

Second, XP traces are less restricted in elided VPs in that, when they are bound, they do

not need to have the same binder in both the antecedent and target clauses. In the examples

below, the trace in the target clause is bound, but not by the same noun phrase that binds it

in the antecedent. This situation obtains for both A-movement of XPs in (25)a through d5

and A-movement in (25)e and f. The distinct binders are italicized.

4The same situation cannot easily be created with A-movement, at least in part because of Case considerations. A-movement, such as passive, will typically leave a trace that is not Case-marked. When the antecedent VP is duplicated at the target site, the trace will be without Case, making the example potentially ungrammatical for independent reasons. Despite this complication, examples of the relevant type can be found in the literature and are claimed to be grammatical. If the data in (i), variously from Sag 1980, Dalrymple 1991, and Fiengo and May 1994, are acceptable, they make the same point as the above A -movement cases.

(i) a. This law restricting free speech should be [yp repealed t ] by Congress, but I can assure you that it won’t 0 .

b. It should be [yp noted t ], as Max did 0, that Fermat’s last theorem has not yet been proven.c. Botanist: That can all be [yp explained t ].

Mr. Spock: Please do 0.d. In March, four fireworks manufacturers asked that the decision be [yp reversed t ], and on

Monday the ICC did 0.e. A lot of this material can be [yp presented t in a fairly informal and accessible fashion ], and

often I do 0. (Chomsky 1982:41)5Not all A-movement cases containing the desired configuration are as acceptable as the above. Sag 1980 cites similar examples which he indicates that his consultants found very degraded:

(i) a. A: What did Harry take a picture of?B: An elephant.A: What did Bill 0?B: A tiger,

b. C: What was John able to take a picture of?D: An elephant.C : What was Harry able to 0?D: A tiger.

Further research will elucidate the differences between the examples in (i) and those above. Fiengo and May 1994 indicate that issues of contrastiveness are involved. One apparent difference is that the traces in the text examples are bound by antecedents which are more clearly D-linked (Pesetsky 1987, Cinque 1990, Rizzi 1990).


122

(25) a. Chicken, she’ll [yp eat t ], but ostrich, she won’t 0.

b. There was one thing in his talk which I [yp liked ti ] but there were about twenty

things which I didn’t 0.

c. Please tell me who I should [yp call t ] and who I shouldn’t 0.

d. ?Frank is easy to [yp like t ] but his twin brother is easy not to 0.

e. The students were [yp arrested t ], and the instructors were 0 too.

f . Pam [yp seemed t to enjoy the party ] and her sister did 0 too.

(26) illustrates the structure for (25)a. Although the traces are bound in both the antecedent

and target clauses, the trace in the former is bound by the DP chicken while the trace in the

latter is bound by the DP ostrich.

(26) a. Chicken, she’ll eat, but ostrich, she won’t 0.

b. IP

chicken DP ostrich DP

she I she I

won t

Summarily, XP traces have great freedom of interpretation in an elided VP. They may pick

up a new binder in the target or they may remain unbound and substitute interpretationally

for their original binder.

This latter conclusion is also reached in Lasnik 1995b. The first of two arguments in that

work is based on the example in (27).

(27) Linguistics, I like t, and you should 0 too.


123

The example does not convincingly makes the desired point, however, because an

alternative analysis of the sentence is available in which the trace in the antecedent VP

[yp like t ] is bound in the target clause. (28) illustrates a possible structure for (27). The

topic linguistics is extracted across the board from both conjunct clauses and binds both

traces simultaneously. On one analysis of (27), then, the trace in the target clause is not

being independently interpreted.

(28)

linguistics IP

DPZ X

VP you I

should

The second argument in Lasnik 1995b is based on the existential sentences in (29), which

were considered grammatical to a reasonable extent. Under the assumption that be takes a

VP small clause out of which the subject may raise, this VP also contains a trace. If it can

be recycled in the target clause, then there is evidence that traces may be interpreted as their

antecedents at an ellipsis site. The data are not obviously grammatical, however, (all of my

consultants reject the examples) and, although it is not immediately clear why, the fact that

they are at least degraded for most people make them a weak argument for the point at

hand.

(29) a. (*)Someone is [yp t in the garden ], isn’t there 0?

b . (*)Someone will be t in the garden, won’t there 0?


124

Thus, although Lasnik 1995b concludes that XP traces can substitute for their binders in

elided VPs, the data do not convincingly support this in the way that the examples in (23)

do. More importantly, Lasnik 1995b took the XP trace data to be representative of the

behavior of all traces and extended the conclusion to X° traces. The contrasting behavior of

X° traces, immediately below, indicates that this generalization is unjustified.

First, it was hypothesized above that X° traces must be bound in an elided VP in contrast to

the XP traces in the data in (23). Second, I demonstrate that X° traces must have the same

binder in both the antecedent and target clauses, again in contrast to the behavior of XP

traces in (25). The empirical reflex of this restriction is that the raised verbs in ellipsis

antecedent and target clauses must to match. That this situation actually obtains has been

asserted for at least two languages which exhibit both verb raising and VP ellipsis. Doron

1990 shows that it holds for Hebrew VPE. McCloskey in preparation demonstrates that it

is strikingly the case for Modem Irish. I discuss the Irish case and then illustrate that the

restriction can reasonably also be seen in English.

McCloskey 1996a argues that the VSO word order characteristic of Irish finite clauses is a

result of the verb fronting from V° to 1° with the subject and complements remaining

VP-intemal. On this view, the transitive Irish clause in (30)a has the structure in (30)b.

(30) a. Cheannaigh siad teach.

buy.PAST they house

“They bought a house.”


125

b. IP

I VP

cheannaighj DP V’(bought)

siad tj DP(they) y \

teach (house)

McCloskey 1991 first argued that Irish has a process of VP ellipsis similar to English.

From the structure in (30), though, it can be seen that the surface result of eliding a VP in

Irish will be quite different from English because everything but the finite verb is

VP-intemal and, hence, will elide. This is illustrated in (31). VP ellipsis in Irish leaves only

the finite verb.

(31) a. Cheannaigh siad teach?

buy .PAST they house

“Did they buy a house?”

b. Cheannaigh. c. Nfor cheannaigh.

buy.PAST not.PAST buy.PAST

“Yes, they did.” “No, they didn’t.”

Relevant to the discussion here is that McCloskey in preparation indicates that there is a

strict identity requirement between the verb in the antecedent clause and the stranded verb in

the target clause. While the discourse above would seem to be semantically coherent if the

response to the question “Bought they a house?” were “No, sold (they house)” using a

verb different from the one in the question, this is completely impossible. Specific

examples are given in (32). The structure of (32)b is more closely investigated in (32)c.


126

Their ungrammatically is claimed to follow from the fact that the binder of the verbal trace

in the target clause is not the same as the binder in the antecedent.

(32) a. *Leigh mean dan achrnor thuig0.

read.PAST I the poem but not.PAST understand.PAST

read I the poem but not understood

b. *Bhris siad an chathaoir agus dheisigh 0 ina dhiaidh sin.

break.PAST they the chair and repair.PAST after-that

broke they the chair and repaired 0 after that

IP IP

VP

bhriSj DP (broke) / \

V5

siad t; (they)

DP

an chathaoir (the chair)

II

dheisighk(repaired)

VP

The same situation exists in British English, which has two verbs, possessive have and

copula be, which undergo V°-to-I0 and which have transparently similar argument

structures.6 The crucial data are given in (33). In each case, the bracketed VP antecedent

contains a trace of the raised verb have. This VP may not, however, be recycled in the

target clause even though the trace would be bound by a form of be. The reason is claimed

to be because the binders are different. The structure of (33)a is made explicit in (34).

6See Hoekstra 1994 and references therein for proposals that the argument structure of have and be is not as transparent as assumed here. If those analyses turn out to be correct, the data would not transparently illustrate the desired point.


127

(33) a. *1 haven’t a dependable friend, unless you are 0.

b . *Have you a good dentist? Yes, my cousin is 0.

c. Have you to be at the wedding rehearsal?

*Yes, I am 0, at six. I’m needed to organize the guests.

(34)

you I

haven tj V

tj a dependable friend

As with the Irish data, ellipsis in (34) fails because the new binder of the verbal trace in the

target clause, a form of be, is not the same as the binder in the antecedent clause, a form of

have. In other respects, the VP is appropriate to serve as an antecedent. If the binders are

made to be the same lexical item, the results are grammatical:

(35) a. I haven’t a dependable friend. Have you 0?

b . Have you a good dentist? No, but my cousin has 0.

c . Have you to be at the wedding rehearsal?

Yes, I have 0, at six. I’m needed to organize the guests.

To summarize, X° traces seem to be invisible for the purposes of VPE interpretation. They

cannot stand in for the raised verb and, if they are present in an elided VP, they must be

bound by the same verb as in the antecedent clause. This behavior contrasts with that of XP

traces in elided VPs which did not need to be bound in the ellipsis target site and did not

need to carry over their binders. Potsdam to appear concludes from these differences that

X° and XP traces are interpretationally distinct. As a result, we can maintain the view of

VPE above that the available interpretations of an elided VP depend upon the presence or


1 2 8

absence of verb raising in the antecedent clause. Having justified the plausibility of this first

verb raising diagnostic, I now return to a second use of VPE in investigating verb raising.

The second contribution of VPE in diagnosing VR relies on the constraint formalized in

chapter 1 that an overt element in 1° must precede the elided material. The account is

repeated in (36). Crucially, a null VP must be licensed by a filled 1° within the same s-

projection.




Adopting this restriction, we expect that modals, do, and any verb or auxiliary that raises to

an inflectional projection will license an immediately following null VP. This licensing

restriction provides additional diagnostic power to VP ellipsis in the following way. A verb

that has undergone Verb Raising can license an elided VP because it moves into 1° while an

unmoved one cannot license ellipsis. Returning to the stacked VP configuration in (22),

repeated below, where 1° is unfilled and considering VP2 as a potential target site, ellipsis

will be possible there only if V j raises out of VPj to 1°. If V[ does not raise, there will be

no overt element in 1° to license the target site and the constraint will not be satisfied,

leading us to expect ungrammaticality.


129

In summary, Verb Raising reduces the possible ellipsis antecedents because it removes

material from within the VP. With respect to target VPs, it can actually increase VPE

possibilities because it moves verbal elements into 1° to license VPs that otherwise could

not be elided. The diagnostic for Verb Raising that VPE provides can be summarized in

(37).

(37) VP Ellipsis Diagnostic for Verb Raising

a. If a verb can be interpreted as part of an elided VP then it has not

undergone Verb Raising in the VPE antecedent

b. A verb can license a null VP within its extended projection if and only if

it has undergone Verb Raising

Below I use VPE as conceived to confirm that main verbs in English do not raise to 1° but

perfective have, all forms of be, and pleonastic do do. The data are presented in pairs. The

first group of sentences illustrates that the particular verb has or has not raised in the

antecedent. This affects the possible interpretation(s) of the target. If the example is

grammatical on the indicated interpretation, the verb/auxiliary has not raised. The second

group looks at the ellipsis site. Grammaticality indicates that raising has taken place because

the verb/auxiliary is in 1° to license the null VP.


130

(38) and (40) illustrate that main verbs do not raise out of VP. In (38), the main verbs,

which head the verb phrases identified as ‘0 =’, remain in the VP antecedent and are

eligible to be interpreted in the target. A typical antecedent has the structure in (39), which

corresponds to (38)c. The main verb is part of a VP constituent since it has not raised to 1°.

This correctly allows an interpretation of the target which includes the main verb.

(38) a. Joe criticized them and Bill did 0 too.

0 = criticize them

b . If Tigger does 0, Robin will also bounce across the stream.

0 = bounce across the stream

c . Mabel got to the leftovers before I could 0.

0 = get to the leftovers

(39) structure of the antecedent clause

DP

Mabel I

to the leftovers

The example in (40) illustrates that, since main verbs do not raise at the ellipsis site either,

they cannot license an elided VP from 1° as required. In (40)a, the inability of the main verb

to license the empty category shows that it has not raised. The remaining examples in (40)b

through f make the same point using main verbs that are argued to take VP complements

(Lobeck 1992, Iveland 1993): aspectual verbs (b through d) and causative verbs (e and f).

If we assume that the main verbs directly preceding the ellipsis sites are not in 1° then the


131

ungrammaticality of the examples follows. The VP complements cannot be grammatically

elided. It is not the case that an antecedent is unavailable; an appropriate VP is bracketed in

each case. The ellipsis fails for purely formal, syntactic licensing reasons. The target clause

in an example like (40)c will have the structure in (41), with 1° unoccupied.7

(40) a. *Nathaniel read the book and Herman j rej^j 0! too.

b . * Although Emily finished 0, Tony only just [yp started reading War and Peace ].

c . *Because Diane continued 0, Jack also started [yp studying Wolof ].

d . *Maria claimed that she couldn’t [yp eat more cake ] but kept 0 anyway.

e. *We had the gardener [yp mow the lawn ] but we should have had the kids 0.

f . ??Although she never makes John 0, Mary makes Sue [yp write her papers on

the computer ].

7Not all examples similar the ones above seem ungrammatical, as Lobeck (1992), who appeals to the patterns originally noticed by Bresnan 1976, suggests. For some aspectual verbs such as start, finish, and continue, this is likely due to the observation that they can be intransitive as in (i).

(i) a. Don’t start until I say so.b. He continued despite my warnings.

These cases cannot be VP ellipsis, however, since the interpretation of the “missing” complement can be easily pragmatically controlled, a characteristic not shared by VP Ellipsis (Hankamer and Sag 1976; Schachter 1977, 1978; Chao 1988). If such readings can be ruled out through context, ellipsis is generally unsatisfactory with the aspectuals.Some examples with causative verbs, particularly make and let, are also quite acceptable:

(ii) a. They made Max clean the toilets but they can’t make me 0 .b. ‘I was going to be a neo-deconstructivist but mom wouldn’t let me 0 .’ (Calvin and Hobbes

July 12, 1995)

I follow Gazdar, Pullum, and Sag 1982 in analyzing such examples as Null Complement Anaphora and not VP Ellipsis.


132

(41) structure of the target clause

* CP

C IP

because DP I’

Diane I VP

V VP

continued 0

(42) and (44) show the same situations for perfective have. Because have undergoes Verb

Raising, the pattern of grammaticality is reversed from that of main verbs, which do not

raise. Examples in the first set are ungrammatical while those in the second set are

grammatical. This directly contrasts with the judgments in (38) and (40), which are

grammatical and ungrammatical, respectively. In (42), have has raised in the antecedent

clause and is unavailable to be part of an antecedent VP for the target. This is illustrated

with the structure in (43), corresponding to (42)a. The data are bad on the indicated

interpretation, though they may be acceptable with another meaning. Have cannot be part of

the elided VP’s meaning because it is not contained in a VP constituent in the antecedent

clause, as seen in (43).8

8Lasnik 1995b proposes an alternative, morphologically-based, explanation for these data. In that analysis, VPE requires strict morphological identity of VPs at some appropriate level:

(i) Identity Requirement on VPE (Lasnik 1995b)A form o f a verb V can only be deleted under identity with the very same form. Forms of be and have are introduced into syntactic structures already fully inflected. Forms of ‘main’ verbs are created out of lexically introduced bare forms and independent affixes.

Observe that the given morphological realization that is required in the elided VPs below is the bare form have, not has as in the antecedent. Since auxiliary have and be are inserted into structures fully inflected, there is never a point at which there is the required morphological identity in the above examples, so they are ruled out.As appealing as such an account might be, there are counterexamples to it, in (ii), in which ellipsis is grammatical but the italicized auxiliaries are not o f the same form.


133

PERFECTIVE HAVE

(42) a. *John has read the book and Bill might 0 too.

0 = have read the book

b . * Although Betsy couldn’t 0, Mary has known their secret for a long time.

0 = have known their secret

c. *Lloyd has practiced his lines every day this week and you should 0 as well.

0 = have practiced your lines


Jo h n I

hasj V

read the book

The have sentences in (44) are acceptable, in contrast to the main verb examples in (40),

under the hypothesis that have raises to 1° in the target clause and licenses the ellipsis site.

The structure of such examples is exemplified by (45), for (44)c. Have has the same

position as in (43) above.

(ii) a. John may be questioning our motives, but Peter hasn’t 0.0 = been questioning our motives

b. ?Mary is being examined and Jack really should 0 also.c. We could have been great. . . but we didn’t need to 0. (Peanuts cartoon)

Potsdam to appear discusses additional empirical difficulties with the account of VPE in (i).


134

(44) a. Lucky had reached the pot of gold before Sean had 0.

b . The baby llama will go anywhere its mother has 0.

c. Mary has seen them naked and I think Bill has 0 too.


IP

r

BUI I

Grammaticality with the three forms of be, progressive, passive, and copula, in (46)

through (52) pattern with perfective have, not main verbs, indicating that they too undergo

Verb Raising. (46), (47), and (48) illustrate the case where be has raised out of VP in the

antecedent clause and cannot be part of the antecedent. The data are bad on the relevant,

indicated reading where the auxiliary is included in the interpretation of the elided VP. See

(42) and (43) for a paraUel explanation and structure with perfective have. The structure of

the antecedent clause in the three cases is given by (49), in which the form of be is in 1° and

is not part of a VP.

PROGRESSIVE BE

(46) a. *Scrooge is saving money and you should 0 too.

0 = be saving money

b . ^Everyone is working on his project for tomorrow so John must 0 too.

0 = be working on his project

c. * Although Klaus might 0, the rest of the team doesn’t look like it is trying.

0 = be trying


135

PASSIVE BE

(47) a. *Madonna was invited and Elvis might 0 too.

0 = be invited

b. *Skippy was elected because you couldn’t 0.

0 = be elected

c. *Smith was electrocuted on May 23, five months after Jones had 0.

0 = been electrocuted

d. *Our daughter is noticed by everyone if she can 0.

0 = be noticed by everyone

COPULA BE

(48) a. *John is allergic to all the yellow trees that are blooming and I might 0 too.

0 = be allergic to all the yellow trees that are blooming

b. *Betsy is a caring woman and if you gave Rhonda a chance you might find that

she can 0 too.

0 = be a caring woman

c. *Her son is perfect because Mother Jones expects that he should 0.

0 = be perfect


IP

subject V

I ' ^ ^ V P

bij V ^ ^ X PI


136

(50), (51), and (52) demonstrate that be can license a null VP at the target site, a result

which is compatible with it being in 1°, as was the case for have in (44) and its tree in (45).

(53) indicates the general structure of the target clause in the examples. Be raises into 1°

from its base position inside VP.

PROGRESSIVE BE

(50) a. Cinderella is doing her chores and I see you are 0 too.

b . Because Newt is 0, all the other Republicans will vote for the bill too.

c. No one else will support the new candidate despite the fact that the mayor is 0.

PASSIVE BE

(51) a. If Chip was accepted at State University then Muffy was 0 too.

b . The other client was told everything I was 0.

c. The bridges were repaired before the roads were 0.

COPULA BE

(52) a. You are in deep trouble and your friend is 0 too.

b . Because the flume ride is frightening, Jack wouldn’t go on the roller coaster

thinking that it was 0 too.


IP

subject I’

r ^ > pbej V ^ ^ X P

I I^ 0


137

The behavior of pleonastic do matches that of the auxiliaries have and be with the exception

that I do not assume that do starts out as the head of a verb phrase. Do is inserted directly

into 1° like the modals. For this reason and the fact that it is semantically vacuous, do will

never have an interpretation in an ellipsis target. The first test illustrated above, then, is not

applicable. The second diagnostic can be carried out, however, since I claim that do is in 1°.

If this is so, it will license an elided VP to its right. (54) confirms this well-known fact. Do

performs its role as a dummy tense bearer in 1° and licenses a null VP. Note that the

examples are ungrammatical without do. The target clause has a structure as in (55).

PLEONASTIC DO

(54) a. Boxer auditioned for the choir and his roommate did 0 too.

b . Although Mary did 0, no one else will try the guacamole ice cream.


Mary I

3 . 4 . Summary

I take the above data to establish the claim that in finite clauses main verbs do not undergo

Verb Raising but auxiliary have, progressive be, passive be, and copula be do. Pleonastic

do is also outside of VP. In the next two sections I explore the possibility of raising in

other contexts using the adverb, negation, and VP ellipsis diagnostics. The data indicate

that nowhere else in the grammar of English do auxiliaries raise. The tests are presented

together here.


138

(13) Adverb Diagnostic for Verb Raising

A verb has undergone Verb Raising if and only if it may appear to the left of an

S-adverb

(21) Negation Diagnostic for Verb Raising

A verb has undergone Verb Raising if and only if it can appear to the left of

sentential not

(37) VP Ellipsis Diagnostic for Verb Raising

a. If a verb can be interpreted as part of an elided VP then it has not

undergone Verb Raising in the VPE antecedent

b. A verb can license a null VP within its extended projection if and only if

it has undergone Verb Raising

4 . Imperatives

Lasnik 1981, Beukema and Coopmans 1989, Pollock 1989, and Zhang 1990 all note on

the basis of limited data that imperative verbs including auxiliaries never undergo Verb

Raising; they always remain inside VP. Their claim is based on a cursory look at the

position of negation in imperatives. This section more thoroughly supports their

observation using the three tests developed above. New evidence from applying the

S-adverb and VPE diagnostics to imperatives confirms that there is, in fact, no raising. The

observations support the negation facts that are cited in the above works, which are also

more thoroughly established. Imperatives thus contrast with finite clauses in not evidencing

any verb movement out of VP, (56). On the other hand, do in imperatives is like pleonastic

do in finite clauses: it is outside of VP. This in itself supports the claim that there is no


139

raising since this is a primary situation in which do appears: when raising into 1° is

unavailable. The behavior of do is also illustrated below.

(56) Verb Raising in Imperatives

Verbs/Auxiliaries are always within VP in imperatives. Pleonastic do is

outside VP.


The adverb test in (13) indicates that a verb has not raised if it cannot appear to the left of an

S-adverb. The configuration verb/auxiliaryAS-adverb is given below for main verbs, (57),

the three forms of be, (58) through (62), and perfective have, (63). In all cases, the data are

ungrammatical with the adverb following the verb. Thus, the diagnostic indicates that there

is no raising of these verbs. Since we assume that the adverb can appear no lower than the

left edge of VP and the data indicate that the verb/auxiliary cannot appear to the adverb’s

right, a position that would be outside VP, it cannot be the case that the verb/auxiliary has

raised into 1°. It must still be in VP, the only alternative. The reverse order S-adverbAverb

is given for comparative judgments; the examples are significantly improved.

MAIN VERB

(57) a. *Interview certainly with the companies that you’re interested in!

(cf. Certainly interview with the companies that you’re interested in!)

b . * Write definitely to your uncles!

(cf. Definitely write to your uncles!)

Taking (58) as illustration for the three uses of be, the examples are ungrammatical when

be is to the left of the adverb. The adverb can, at best, be adjoined to VP, requiring the

structure in (60) (traces not shown). For the given word order, however, be would need to


140

be in 1°, as shown. Given the ungrammaticality of the data, this cannot be the case. Hence,

the structure in (60) must be incorrect because of the raised auxiliary. In actuality, be must

still be in V \ as we are led to expect by the grammaticality of examples in (59) which

illustrate the alternative S-adverbAauxiliary order.

PROGRESSIVE BE

(58) a. *Be absolutely listening to the customers’ complaints and taking note of them!

b . *Be definitely paying attention to his every word!

(59) a. Absolutely be listening to the customers’ complaints and taking note of them!

b . Definitely be paying attention to his every word!

(60) * IP

D P ^ ^ T

I ^ ^ V P

bej AdvP VP

definitely t; VP

Vmain DP

paying attention

PASSIVE BE

(61) a. *Be normally approved by the committee before coming to the seminar!

(cf. Normally be approved by the committee before coming to the seminar!)

b . *Be perhaps examined, just to be on the safe side!

(cf. Perhaps be examined, just to be on the safe side!)


141

COPULA BE

(62) a. *Be normally in your seats before the guest of honor walks in!

(cf. Normally be in your seats before the guest of honor walks in!)

b . *Be definitely in bed by nine o’clock!

(cf. Definitely be in bed by nine o’clock!)

PERFECTIVE HA VE?

(63) a. ?Have certainly read at least the introduction!

b . ?Have normally waited at least an hour before going swimming!

The data with pleonastic do behaves differently and confirms that do is in 1°. This is

compatible with the assumption that it is base-generated there. The diagnostic lead us to

expect that if do is outside VP then it may appear to the left of an S-adverb. (64) confirms

this prediction and it is the opposite pattern of grammaticality seen with main verbs, be, and

aspectual have above. The italicized adverb in each example is adjoined to the topmost VP

and the result is grammatical with the pleonastic to its left.

^The data with the adverb following auxiliary have repeated in (i) seem more natural than the corresponding data with be and the contrast with the S-adverbAauxiliary order in (ii) seems less strong.

(i) a. ?Have certainly read at least the introduction!b. ?Have normally waited at least an hour before going swimming!

(ii) a. Certainly have read at least the introduction!b. Normally have waited at least an hour before going swimming!

If the examples in (i) were fully acceptable, they would be interpreted according to the diagnostic to indicate that have raises in imperatives. Lobeck 1987:72-78 reaches a similar conclusion regarding non-finite occurrences of have based on its interaction with adverb placement, VP ellipsis interpretation, and VP fronting. The work concludes that non-finite have is actually base-generated in I* with the modals. I will resist this conclusion based largely on the results of the other diagnostics to follow. Johnson 1988 also observes numerous subtle irregularities with have but argues against generating non-finite have in I* based on the variable position of not in subjunctive and infinitive clauses. He suggests that have is a bound morpheme which must affix onto another head. This unusual behavior of perfective have is addressed later.


142

PLEONASTIC DO

(64) a. Do normally prepare for class better than this!

b . Do definitely come to our house warming bash!

In summary, Lasnik’s (1981) observation that imperative verbs do not raise, which was

based on the position of negation in imperatives, is confirmed. I turn to negation in the next

section and VP ellipsis in the final one.

4 .2 . Negation

The test for Verb Raising using sentential negation says that there is Verb Raising if and

only if the verb can appear to the left of sentential negation. As is well known, this is not

how negative imperatives are formed. Neither main verbs nor auxiliaries, in (65) through

(69), can be followed by not or the negative affix -n 't. Only do, in (70) may precede

negation. This again confirms that it is outside VP.

MAIN VERB

(65) a. *Touch n’t/not that glass!

b . *Open n’t/not that box!

c. *Talk n’t/not to the kid in the comer!

PERFECTIVE HAVE

(66) a. *Please have n’t/not left a mess in the kitchen for us to clean when we get back!

b . *Have n’t/not just skimmed over his critique!

c. *Have n’t/not eaten everything before we sit down to supper!


143

PROGRESSIVE BE

(67) a. *Be n’t/not going so soon!

b. *Be n’t/not sleeping in Professor Zok’s lectures!

c. *Be n’t/not keeping secrets from your wife!

PASSIVE BE

(68) a. *Be n’t/not fooled by his chicanery!

b. *Be n’ t/not examined by that quack!

c. *Be n’t/not elected to that dead end position!

COPULA BE

(69) a. *Be n’t/not proud!

b . *Be n’t/not a cheapskate!

c. *Be n’t/not in such a hurry, you’ll just get into an accident!

PLEONASTIC DO

(70) a. Don’t/Do not touch that!

b . Don’t/Do not have left a mess in the kitchen for us to clean when we get back!

c. Don’t/Do not be going so soon!

d. Don’t/Do not be examined by that quack!

e. Don’t/Do not be proud!

4.3 . VP Ellipsis

Lastly, the VP Ellipsis test support the results of the previous two tests. This diagnostic for

verb movement gives us two expectations for imperatives if it is the case that verbs and

auxiliaries are always in VP as has been seen thus far. First, the entire string of verbal

elements in an imperative can serve as the antecedent for VPE. This follows because there


144

is no raising; the entire string corresponds to a VP constituent. Imperatives will thus

contrast with finite clauses with raised elements. In the latter, the topmost auxiliary is not

inside VP and so is not available to be part of the antecedent VP and interpreted in the

target. The first set of examples in each group below, (71), (73), (75), (77), and (79),

demonstrates this for main verbs, the three forms of be and perfective have. The verbal

element does not raise in an imperative so that it can in fact be interpreted in the elided VP.

For comparison, the second set of examples in each group, (72), (74), (76), (78), and

(80), illustrates the finite clause case. If there is also no raising, as with main verbs, then

the pattern of grammaticality is the same. If finite clauses have raising, as with auxiliaries,

then the judgments are reversed.

MAIN VERB

(71) a. Run away if you can 0!

b . Doit because you should 0!

c . Eat a goldfish and I will 0 too!

d. Try the fudge after Lee has 0!

e . Swim in the deep end only if a supervisor is 0!

(72) a. Gretchen ran away since she could 0!

b. Max does that because he can 0!

c . If you eat a goldfish then I will 0 too!

PROGRESSIVE BE

(73) a. Be waiting for me outside if you can 0!

b . Be sitting at the video phone at 9 and I will 0 too!

c. Be working whenever the boss says to 0!


145

(74) a. *John usually is waiting outside if he can 0!

0 = be waiting outside

b . *Marsha was sitting at the video phone at 9 but Jan couldn’10!

0 = be sitting at the video phone

PASSIVE BE

(75) a. Be elected if you can 0!

b . Be aligned by a chiropractor and I will 0 too!

c. Be escorted by famous people whenever you can 0!

(76) a. *John was elected and you can 0 too.

0 = be elected

b . *Kurt was aligned by a chiropractor but Melissa is afraid to 0.

0 = be aligned by a chiropractor

c. *Because Madonna is escorted by famous people, Cher feels that she must 0 too.

0 = be escorted by famous people

COPULA BE

(77) a. Be on time if you can 0! (Warner 1985:66)10

b . Be in bed before I have to tell you to 0!

c. Be cooperative and I will 0 too!

d. Be all you can 0!

e. Be a ghost this Halloween if you never have 0 before!

10Wamer 1985, in the context of a GPSG analysis of the English auxiliary system, claims to be the first to discuss such examples.


146

(78) a. *The kids are never quiet even if they can 0.

b . *Spike is rarely in bed before we tell him to 0.

c . *Grice says that we are cooperative but sometimes I can’t 0.

d. *Bill is always all he can 0.

PERFECTIVE HAVE11

(79) a. Please have studied the manuscript beforehand if you can 0!

0 = have studied the manuscript

b . Please have cleaned up part of the mess and I guarantee I will 0 too!

0 = have cleaned up part of the mess

c. Please have memorized the responses unless you’re sure your opponent won’t 0!

0 = have memorized the responses

(80) a. *He has studied the manuscript beforehand and you should 0 too.

0 = have studied the manuscript beforehand

b . *If he has cleaned up part of the mess by Tuesday then I guarantee I will 0 too.


c. *Bob has memorized the responses because he’s sure his opponent will 0 too!


The second expectation arising from the interaction of VPE and the hypothesized lack of

raising in imperatives is that VPE in imperatives with just a main verb or the auxiliaries

should be unavailable. Remember that VPE is possible only if there is an element in 1° to

license the elided VP. Accepting that auxiliaries never raise in imperatives, this option

1 1 These two sets of examples will not clearly contrast for all speakers. This is because for some there isgeneral difficulty in retrieving an antecedent VP headed by perfective have. This restriction was discussed in chapter 1, section 5.4. For me, there is a clear difference.


147

should be uniformly ruled out. This prediction is confirmed in (81) through (85) for main

verbs, have, and be. In each set of examples, 1° is lexically unfilled and, consequently,

ellipsis is impossible. Corresponding examples with non-imperatives are given for

contrast. Lasnik 1995b makes this same observation in a footnote.

MAIN VERB

(81) a. *We need to have someone clean the pool. I know, have John 0!

b . ??Someone should empty the dishwasher. Well, make the kids 0!

c. * Jack started reading the poem, now Jimmy keep 0!

PERFECTIVE HAVE

(82) a. * You should tidy up before the guests arrive. Yes, definitely have 0!

(cf. You should tidy up before the guests arrive. At least we hope that you will

have 0.)

b . *The Smiths might have reached a decision regarding the matter. Jane, please

have 0 too!

(cf. The Smiths might have reached a decision regarding the matter so we want

Jane to have 0 also.)

c. *This is a difficult exam. The teacher suggests that you have prepared for it. So,

do as he says and have 0!

PROGRESSIVE BE

(83) a. *Cinderella is doing her chores so, if you know what’s good for you, be 0 too!

(cf. Cinderella is doing her chores and I see you are 0 too.)

b. *Huey is doing his homework now. When I come home, you be 0 too!

(cf. Huey is doing his homework now and when I come home, I hope that you

are 0.)


148

PASSIVE BE

(84) a. *They were examined by that nurse so you be 0 too!

(cf. They were examined by that nurse because I was 0.)

b . *Tiffany makes sure she is seen with the right crowd. It couldn’t hurt your

career. You be 0 too!

COPULA BE

(85) a. *We can’t have any latecomers. You should definitely be on time. Absolutely no

excuses, be 0!

(cf. You should definitely be on time. No excuses! I was 0.)

b . *With the flu going around, you can’t be too careful. Especially since you work

with so many people, be 0!

c. *Despite the baseball strike, it’s fun to be a sports fan. So, be 0!

When we look at do, in (86), the behavior is different. Pleonastic do shows clear evidence

of being outside VP. Although we can’t test whether it can be part of a target, again since it

is semantically empty and would contribute nothing to the interpretation of an elided VP,

we can determine if it licenses a null VP. This seems to be the case; ellipsis in imperatives

with do and don’t is grammatical. Schmerling 1977 seems to be the first to discuss these

facts. The work notes the grammaticality of Don’t! as an elliptical imperative. Akmajian

1984 and Quirk et. al. 1985 attest to the acceptability of Do! in addition.12 Beyond Do! and

Don’t! in (86)a and b, there exist data as in (86)c through j in which subjects and/or

auxiliaries follow do(n’t). These examples are correctly predicted to be grammatical under

12See chapter 4 for evidence that such examples are instances of VPE and not frozen expressions.


149

an analysis in which do(n’t) is, or was, in 1° and the additional stranded elements are

licensed as part of the same extended projection as the empty VP.

PLEONASTIC DO

(86) a. Did we say you could draw on the walls? All right, then, don’t 0!

b . We want everyone to come, so those who can, please/by all means do 0!

c. Rick walked out of the lecture, but don’t everyone else 0, please!

d . Billy didn’t tell mom what I did, so don’t you 0 either!13

e. Someone needs to volunteer to help with the church bazaar.

Oh yes, please, do someone 0!

f. Mom said you shouldn’t be playing in the fireplace, so don’t be 0!

g. Don’t you be seen with those losers and don’t anyone else be 0 either!

h. Don’t YOU be examined by a doctor, and don’ t YOU be 0, either.

(Schmerling 1977)

i. ‘This blushing dip isn’t shy, so don’t you be.’

{The New Basics Cookbook, p. 25)

j . The movers told us to be careful with the dining table.

Yes, do be 0! It’s a priceless heirloom,

k. Do you think they’ve seen the mess?

Well, it depends on whether they’ve looked in the closet already.

Oh, please don’t have 0!

To summarize, verbal elements, main verbs and auxiliaries, do not undergo Verb Raising

in imperatives.

13Warner 1985:48 claims that don't you! is ungrammatical; however, it seems that it merely needs to be put in some context to make it acceptable.


150

5. Subjunctive Clauses

Very little has been written about the syntactic structure of subjunctive clauses to my

knowledge (but see Stockwell et. al. 1973, Johnson 1988, and Zanuttini 1991). A

hypothesis that I will put forth here in keeping with the theme of this current work is that

subjunctive clauses are structured as other clauses are. Subjunctive complements,

illustrated in (87), are complement clauses, obligatorily CP. This is suggested by the

required presence of the head that, in (88), which I take to be a subjunctive C°. The CP

immediately dominates IP, which selects VP.

(87) a. It is imperative that passengers remain behind the yellow line.

b . He suggests that you be more receptive.

c. It is necessary that the children go to bed before 8.

(88) a. It is necessary that you be here,

b . *It is necessary you be here.

Chapter 1 demonstrated that modals and periphrastic do are completely impossible in

subjunctive clauses. With the modal data in (89) and (90), the difficulty seems purely

syntactic as the examples have grammatical paraphrases without modals.

(89) a. *He demanded that the successful candidates can speak German.

b . He demanded that the successful candidates be able to speak German.

(90) a. *The police require that the spectators must stand behind the barricade,

b . The police require that the spectators stand behind the barricade.

Attempts to use do, with negation or emphasis, are equally unsuccessful. Neither do with

negation in (91) nor emphatic assertion, (92), is acceptable. As discussed in chapter 1 and


151

below, subjunctive clauses must be negated with not alone. There seems to be no syntactic

way to emphasize a subjunctive complement. The ideas expressed in (92) must be done

with stress and intonation.

(91) a. *Everyone suggested that he don’t/doesn’t act so aloof if he hopes to find a wife,

b . *Jack insists that we don’t cut down his bean stalk just yet.

(92) a. ?*Contrary to what the polls say, we suggest that Jimmy do run for re-election.

b . ?*I requested that she do be more assertive as she is quite competent.

c . ?*Mom demanded that you do be careful.

d . *The manufacturer urges that the buyer do not remove the label.

These restrictions are on elements that normally appear in 1° and one could adopt one of

two reasonable alternatives to account for the syntactic patterns: subjunctive 1° is either I)

filled with a null element or 2) absent altogether. Although saying that IP is absent is a

workable solution and is proposed in Zanuttini 1991,1 will not explore it here since it

would require complete reworking of earlier assumptions. Instead, I will assume that

English subjunctive 1° is a zero modal, Msbj-, a proposal originating in Roberts 1985:40 and

also asserted in Lasnik 1995b. Lasnik notes that the lack of ^o-support seen above

indicates that Msbj- is not an affix but a morphologically independent head. The

hypothesized structure is then as in (93), similar to what I have assumed elsewhere for

English clause structure.


V complement

Having an independent subjunctive modal in 1° suggests that overt verb raising will not be

seen in subjunctives since the head position is occupied. The data below, using the adverb,

negation, and ellipsis diagnostics that have been developed, confirm this absence of Verb

Raising in subjunctive clauses. Given the structure in (93), the results are expected.

Zanuttini 1991:97 makes similar observations based superficially on the lack of subjunctive

verbal morphology as does Lasnik 1995b based on limited VPE data and Beukema and

Coopmans 1989 based on the position of negation.


If there is no verb raising, the adverb diagnostic indicates that the relevant verb will be

unable to appear to the left of an S-adverb. This is confirmed in (94), (95), (96), (97), and

(98) for main verbs, the three uses of be, and perfective have respectively. In all the

examples, the judgments are degraded if the italicized S-adverb is to the right of the relevant

verb or auxiliary. The examples given for comparison in parentheses contain the adverb in

a position to the left of the verbal element.


153

MAIN VERB

(94) a. *My teacher insists that I work normally on my homework before bedtime.

(cf. My teacher insists that I normally work on my homework before bedtime.)

b . *The guide urges that one go certainly to the local museum.

(cf. The guide urges that one certainly go to the local museum.)

PROGRESSIVE BE

(95) a. ??The sales manual requires that all agents be definitely paying attention to the

customers’ complaints and taking note of them during the exchange.14

(cf. The sales manual requires that all agents definitely be paying attention to the

customer’s complaints and taking note of them.)

b . *During his lectures, it is crucial that we be absolutely paying attention to his

every word.

(cf. It is crucial that we absolutely be paying attention to his every word.)

PASSIVE BE

(96) a. *It is recommended that you be normally approved by the committee before

coming to the seminar.

(cf. It is recommended that you normally be approved by the committee before

coming to the seminar.)

b . ♦The doctor proposed that the patient be probably examined a second time, just

to be on the same side.

(cf. The doctor proposed that the patient probably be examined a second time.)

14The variable and partially acceptable status of some of the examples may conceivably be attributed to Ernst’s (1984) observation that S-adverbs are sometimes acceptable inside VP.


154

COPULA BE

(97) a. ??DipIomatic protocol requires that the attendees be normally standing when the

guest of honor walks in.

(cf. Diplomatic protocol requires that the attendees normally be standing when

the guest of honor walks in.)

b . ??The tyrannical baby sitter required that the kids be definitely in bed by eight

o’clock.

(cf. The tyrannical baby sitter requires that the kids definitely be in bed by eight

o’clock.)

PERFECTIVE HAVE

(98) a. ?It is mandatory that everybody have certainly read at least the introduction,

b . ?It is important that she have normally waited at least an hour before going

swimming.

5 .2 . Negation

This section applies the negation diagnostic to confirm the absence of raising in subjunctive

complements. Complicating the demonstration is the potential confusion that can arise over

the status of not in subjunctive clauses. I take a brief detour to demonstrate that the not that

appears in subjunctives does represent sentential negation. It can then be used in the

negation test.

5.2.1. Negation in Subjunctive Clauses

As is well known, negation in subjunctive clauses always takes the form not and never n ’t.

This fact is illustrated in (99) and (100) for main verbs and the auxiliaries and the results

are very robust.


155

(99) a. The neighbors asked that we not party so loudly.

b . The sign requests that one not be loitering during curfew hours.

c. The librarian insisted that the book not be returned in shabby condition.

d . The queen desires that you not be inattentive.

e. The workers requested that I not have ruined their new sidewalk before it even

dries.

(100) a. *The neighbors asked that we don’t party so loudly.

b. *The sign requests that one j^ j loitering during curfew hours.

c. *The librarian insisted that the book j^ken' returned in shabby condition.

{don.11 bel ben> t f inattentive.

*The workers requested that I haven^t6} ru“ie< t ie*r new sidewalk before ite.

even dries

Given that negation cannot appear as n 7, it is important to confirm that the not that does

appear in (100) is sentential negation and not constituent negation. Whereas sentential

negation heads its own projection above VP, the latter I take to be simply adjoined to an

appropriate constituent. The structures I assume for the two kinds of negation are in (101).

(101)a follows Pollock 1989, Zanuttini 1991, and the discussion in chapter 1 in taking not

to be the head of NegP immediately above VP. (101)b assimilates constituent negation to

the analysis of VP-adverbs.


156

(101) a. sentential negation b. constituent negation

NegP VP

Assuming hypothetically that subjunctive clauses have no inflectional projection(s), it

would not be unreasonable to claim that subjunctive clauses could not be negated except at

the constituent level via (101)b. Using tests for sentential versus constituent negation,

however, we can verify that negation in subjunctive clauses is in fact sentential. In what

follows I lay out these tests and their application to subjunctive clauses. With the sentential

nature of subjunctive not firmly established, we can go on to apply the negation diagnostic

forVR.

5.2.2. Sentential versus Constituent Negation

Iatridou 1990 discusses two diagnostics for sentential versus constituent negation which

rely on the scope ambiguities that can arise between negation and various adjoined

elements. Assuming with Ladusaw 1988 that the scope of negation is determined by its

position in surface structure, the two negations will have different scopal domains since

they are structurally distinct. This can be seen more clearly in (101). Given these structures

which show an XP adjoined to VP, constituent negation in (101)b will not be able to

c-command a VP-adjoined element. Sentential negation will be able to take scope over such

elements since they are within the c-command domain of Neg° in (101)a. The crucial

observation for the two tests below is that sentential negation can take scope over

VP-adjuncts while constituent negation cannot. Consequently, when the adjunct contains


157

elements that can give rise to scope ambiguities, the ambiguities will only appear with

sentential negation.

If a VP adjunct containing a quantified XP is within the scope of negation, two readings

result depending upon whether or not the quantifier is interpreted inside or outside the

scope of negation. Thus, Iatridou’s example in (102) is ambiguous with the interpretations

in (103). In (103)a, negation takes scope of the quantificational many phrase, what I will

call the NOT > MANY interpretation. The reverse obtains in (103)b.

(102) John did not play football for many years.

(103) a. John played football for not many years. NOT > MANY

b. For many years, John didn’t play football. MANY > NOT

An example with constituent negation, (104) (Iatridou’s (69)), is unambiguous. It has only

the reading in (103)b where not is within the scope of the quantifier.

(104) John has been not playing football for many years. MANY > NOT

Turning to subjunctive clauses, we see that parallel examples in (105) are ambiguous.

Crucially, the wide scope reading of not (narrow scope reading of the quantifier) is

available. This corresponds to the structure for sentential negation in (101)a with not

c-commanding the adjunct. To illustrate, (105)b has both interpretations in (106).

(105) a. The recruiters would have preferred that John not have played football for many

years.

b . Her parents suggested that Willa not get married for two years.

c. Judy’s doctor insisted that she not walk for four days.


158

(106) a. Her parents suggested that Willa get married for not just two years.

NOT> TWO

b . Her parents suggested that Willa wait two years before getting married.

TWO > NOT

Iatridou’s second test relies on Linebarger 1987 and ambiguities found with because-

adjuncts. Johnston 1994 argues that because-clauses illustrated in (107) are structurally

ambiguous. They may be right-adjoined to VP or IP. This gives rise to the ambiguity

spelled out in (108) for (107)c in which negation may, but need not, take scope over the

adjunct.

(107) a. The Browns had to sell the cow because they were short on cash.

b . Mike passed up dessert because he was trying to give up sweets.

c. Marion hasn’t been selling drugs because the demand changed.

(108) a. Marion hasn’t been selling drugs and the reason is that the demand changed.

b . Marion has been selling drugs but not for the reason that the demand changed.

The reading in (108)a, which Johnston 1994 calls the Negated Head reading (NH), results

when the adverbial is not c-commanded by negation. It is adjoined to IP which is outside

the c-command domain of Neg°. The Negated Adjunct reading (NA) in (108)b arises

because negation has scope over the because-clmse. In this case it is necessarily adjoined

to VP. Looking at an example of constituent negation, the ambiguity disappears; (109) has

only the Negated Head reading in (108)a in which the negation does not c-command the

adverbial.15

15The because-adjunct could be adjoined to either IP or VP. Neither position is c-commanded by constituent negation, as the above structure indicates.


159

(109) Marion has been not selling drugs because the demand changed.

Returning to negated subjunctive clauses, the data in (110) are ambiguous indicating again

that the negation is sentential not constituent negation. (110)c, for example, has both

readings in (111). Most importantly, the reading in (11 l)b in which not has scope over the

because-clause is available.16

(110) a. We proposed that Mrs. Brown not sell the cow because she needed the income.

b. The broker urged that Marty not sell his shares because the market was unstable.

c. His wife suggested that Marion not sell drugs because the demand has changed.

(111) a. His wife suggested that Marion not sell drugs for the reason that the demand has

changed.

b. His wife suggested that Marion sell drugs but not for the reason that the demand

has changed.

A third test, due to Jim McCloskey, relies on the licensing of negative polarity items (NPIs)

and is similar to Iatridou’s diagnostics. NPIs are elements such as anybody, ever, yet, etc.

which must appear within the c-command domain of negation or an appropriately negative

element (Ladusaw 1992, Progovac 1994). We saw with the above that VP adjuncts are

within the scope, and hence c-command domain, of sentential negation but not constituent

negation. Given the surface licensing condition on NPIs, they should only be compatible

with sentential negation when they appear within an adjunct. This accounts for the contrast

in (112). The NPI any is licensed in (112)a because it is c-commanded by sentential not.

Constituent negation in (112)b is too low in the structure to license the NPI.

I6These examples also have an irrelevant third reading in which the because-clause is construed with the matrix verb.


160

(112) a. John has not been playing football for any great length of time,

b . * John has been not playing football for any great length of time.

If negation in subjunctive clauses is indeed sentential negation, we expect NPIs to appear in

adjuncts. This prediction is confirmed in (113).

(113) We require that our new students not have studied dance for any significant

length of time before enrolling.

A fourth test for distinguishing sentential and constituent negation relies on their different

behavior in VP ellipsis (VPE) contexts. As discussed in chapter 1, an elided VP must be

licensed by an appropriate inflectional head within the extended projection of the null VP.

The work showed that a lexically overt head of IP, or NegP if present, dominating the VP

empty category licenses ellipsis. (114) repeats several examples that show sentential not

may immediately precede an ellipsis site.

( 114) a. I definitely want to go to the fashion show although my husband might not 0.


c. My car passed the smog test but Henry’s did not 0.

In this regard, the two types of negation contrast. Constituent negation may not introduce a

null VP (Johnson 1988, Ernst 1992, Kim and Sag 1995). This reasonably derives from its

adverbial status. Baker 1971 and Sag 1978 both observe that left-adjoined adverbs in

general may not immediately precede an ellipsis site. This accounts for the

ungrammaticality of the examples in (115) because they contain instances of constituent

negation.


161

(115) a. *Some of the students have been studying but some have been not 0.

b . *Goof off while we’re gone, but understand that when we arrive home you will

be not 0.

This difference permits an identification of not as sentential or constituent. If the not under

investigation can grammatically precede a null VP, it must be sentential negation. Returning

to subjunctives and the determination of the status of not, we will see below that ellipsis in

subjunctive clauses is generally unacceptable. A primary and striking exception is when

negation is present as illustrated in (116), first pointed out in chapter 1. The appearance of

not permits an otherwise impossible null VP. This is predicted if negation in subjunctives is

sentential negation but is completely unexpected if it is constituent negation. Not in the

examples below is the head Neg° and it licenses the elided VP. See Potsdam in press for

further discussion.

(116) a. You could take the last piece of candy but I heartily suggest that you not 0.

b . Ted wanted to vacation in Afghanistan but his agent suggested that he not 0.

c. Because the Republicans say to raise taxes, the Democrats insist that we not 0.

d . We think that Mary should present her case to the committee but we will ask that

Bill not 0.

From these four diagnostics, listed in (117), we can conclude that negation in subjunctive

clauses is indeed sentential negation.

(117) Constituent Versus Sentential Negation Diagnostics

1) interaction between negation and quantified adjuncts

2) interaction between negation and because-adjuncts

3) interaction between negation and NPI in VP adjuncts

4) licensing of VP Ellipsis


162

The not that appears in subjunctive clauses like the ones repeated in (99) can thus

straightforwardly be used to confirm that verb raising does not take place in subjunctive

clauses.

(99) a. The neighbors asked that we not party so loudly.

b . The sign requests that one not be loitering during curfew hours.

c. The librarian insisted that the book not be returned in shabby condition.

d. The queen desires that you not be inattentive.

e. The workers requested that I not have ruined their new sidewalk before it even

dries.

5.2.3. Negation and Verb Raising

The examples below apply the Negation test for Verb Raising. This is illustrated for main

verbs, (118), and the three forms of be, (119) through (121). Since the verbs under

consideration can absolutely not appear to the left of not, raising is unavailable. Roberts

1985 also uses negation to claim that there is no raising in such contexts.

MAIN VERB

(118) a. *The sales staff insisted that my son touch not that glass.

b . *Zeus demanded that Pandora open not that box.

c . *The teacher asked that we talk not to the kid in the comer.

PROGRESSIVE BE

(119) a. *Hilda requested that the guests be not going so soon.

b . *Professor Zok asks that we be not sleeping during his lectures.

c. *In the interest of matrimonial bliss, the counselor suggests that you be not

keeping secrets from your wife.


163

PASSIVE BE

(120) a. *It is advised that everyone be not fooled by his chicanery.

b . *For his own sake, the association urges that he be not examined by that quack.

c . *We advise that you be not elected to that dead end position.

COPULA BE

(121) a. *Humility requires that one be not proud.

b . *His mother is always urging that he be not a cheapskate.

c . *The highway patrol recommends that you be not in such a hurry, you’ll only

get into an accident.

The data with perfective have appear in (122). They sound better than expected with have

to the left of not. Johnson 1988 also observes the acceptability of not after non-finite have

in subjunctive clauses. Assuming that not in these examples is still Neg° which is not

appearing lower in the structure, I conclude that have is able to appear outside of VP and,

furthermore, above Neg°.17 This is an unexpected result. Perfective have in subjunctives

seems to be able to optionally appear outside of VP according to the negation diagnostic.

Note that the standard order of not preceding the auxiliary, in (123), is still possible, and

perhaps preferred. The test forces us to conclude that movement of have to 1° may

optionally occur in subjunctive clauses.

*7The b example contains a because-cfoust and is ambiguous, indicating sentential negation.


164

PERFECTIVE HAVE

(122) a. ?My parents suggested that the baby sitter have not left a mess in the kitchen for

them to clean when they get back.

b. ?I urged that Mary have nor just skimmed over his critique simply because of his

attitude.

c. ?Who demanded that we have not eaten everything before sitting down to

supper?

(123) a. My parents suggested that the baby sitter not have left a mess in the kitchen for

them to clean when they get back.

b . I urged that Mary not have just skimmed over his critique simply because of his

attitude.

c. Who demanded that we not have eaten everything before sitting down to supper?

If the results of these diagnostics are consistent with one another, we expect that the adverb

test would have yielded the same conclusion. The adverb data are repeated below in (98)

and the reader can verify that the results were not, in fact, completely ungrammatical,

though the contrasts are subtle. The unusual behavior of perfective have in subjunctives

will also appear with VP ellipsis in the next section.

PERFECTIVE HAVE

(98) a. ?It is mandatory that everybody have certainly read at least the introduction,

b . ?It is important that she have normally waited at least an hour before going

swimming.


165

5 .3 . VP Ellipsis

The behavior of subjunctive clauses with respect to ellipsis generally parallels that seen

with imperatives. The two expectations from the VPE diagnostic are illustrated below.

First, when a verb does not raise in a VPE antecedent, it may be interpreted as part of an

elided target VP. Main verbs and be do not raise in subjunctive clauses and thus verify the

claim. This is illustrated in (124) through (127). The examples are grammatical because the

subjunctive verb/auxiliary is still in the VP which can then serve as part of the antecedent.

MAIN VERB

(124) a. I suggest that you [yp write to Aunt Martha ] first and then I will 0 later.

0 = write to Aunt Martha

b . Mr. Hines urged that Billy save his money and hinted that John should 0 as well.

0 = save his money

c. It is necessary that the builders finish before the cleaners do 0.

d. The king decreed that everyone obey him but it is clear that Robin Hood won’t 0.

PROGRESSIVE BE

(125) a. We demand that from now on you [yp be waiting outside ] if you can 0.

0 = be waiting outside

b . It is required that a clerk be sitting at the desk when the manager can’10.

0 = be sitting at the desk

c. It is mandatory that we all be working if the boss says to 0.


166

PASSIVE BE

(126) a. After his mother insisted that Bob [yp be examined by a real doctor ], he finally

realized that he should 0.

0 = be examined by a real doctor

b . The graduate students demanded that they be treated with respect but I doubt that

they will 0.

0 = be treated with respect

c . It is important that I be seen with famous linguists whenever I can 0.

0 = be seen with famous linguists

d. We don’t advocate that you two be married by that priest, but if you absolutely

must 0, go ahead.

COPULA BE

(127) a. The public demands that buses [yp be on time ] but that alone doesn’t ensure that

they will 0.

0 = be on time

b . I’ 11 ask again that those brats be quiet if they can 0.

0 = be quiet

c . It is imperative that you be in bed before I have to tell you to 0 again.

0 = be in bed

d . It is suggested that the union be cooperative and management will 0 too.

0 = be cooperative

The data in (128) with perfective have perhaps also have the indicated reading; although the

judgments are rather unclear in comparison to the above. Their grammaticality would be

compatible with have remaining inside VP.


167

PERFECTIVE HAVE

(128) a. ??The committee asks that you have studied the manuscript beforehand if you are

able to 0.

0 = have studied the manuscript

b . ??By dinner it is necessary that the kids have cleaned up part of the mess and I

guarantee that by then I will 0 too.


c. ??We advise that the contestant have memorized the responses unless he is sure

his opponent won’t 0.


At the same time, their ungrammaticality receives an explanation as part of the more general

phenomenon in which VPs headed by non-finite have unexpectedly cannot always serve as

VPE antecedents. Numerous researchers (Akmajian, Steele, and Wasow 1979, Steele et.

al. 1981, Lobeck 1987, Zagona 1988a) observe that interpreting an elided VP using a VP

antecedent whose head is non-finite have is infelicitous when we would otherwise not

expect it to be. Thus, for some (though not all) speakers, (129) is ungrammatical on the

indicated interpretation. Chapter 1, section 5.4 discussed the parochial nature of this

restriction. Whatever ultimately rules out (129) would account for the ungrammaticality of

(128) because, under current assumptions, the two antecedent clauses have the same

structure, modulo the contents of 1°. Assimilating the ungrammaticality of (128) to that of

the example in (129) predicts that those who can get the given reading in (129) will also

accept the data in (128). The interpretation of the data in (128), then, is inconclusive

regarding the potential raising of have in subjunctives.


168

(129) *John might have read that chapter, and Bill could 0 too.

0 = have read that chapter

The second diagnostic using VPE involves looking at the VPE target. No raising into 1°

leads to the expectation that a null VP will not be licensed in the target clause because 1° will

not be lexically filled. Again, the data in (130) through (133) confirm this for main verbs

and the three uses of be. The examples are ungrammatical because the constraint on the

licensing of elided VPs is violated. It should be noted that these facts do not provide

independent evidence for the lack of verb raising in subjunctive complements because they

were themselves used to motivate the Licensing Condition on VPE, (36). The full set of

facts are repeated here primarily for completeness. They are confirmed by the adverb and

negation diagnostics and thus are not the sole evidence for the conclusion that follows.

MAIN VERB

(130) a. *Nathaniel read the book and the teacher suggested that Herman | re^ j 0I too.

b . *It was proposed that Diane continue 0 since Jack started studying Wolof.

c . *Maria said she couldn’t eat more cake but her mother insisted that she keep 0

anyway.

d. *We had the gardener mow the lawn this time but I propose that we have the

kids 0 next time.

e. ??It is important that you never make John 0 but you can make Sue climb on the

roof.


169

PROGRESSIVE BE


be 0 instead.

b . *1 don’t know if Mary is taking care of the plants while we’re gone so I would

ask that you be 0.

c. *My running partner is training for the marathon and it’s imperative that I be 0

too.

PASSIVE BE

(132) a. *The doctor didn’t think that Jesse needed to be examined by a specialist but he’s

going to suggest that Joan be 0.

b . *Bill was interviewed before he quit and I want to insist that you be 0 too.

c. * Jack’s room was given a thorough cleaning and I suggest that yours be 0 too.

d . *The clients were told everything I suggested that they be 0.

e. *The bridges were repaired before the engineers could even insist that the

supporting structure be 0 first.

COPULA BE

(133) a. *You should be happy with the results even though I can’t insist that you be 0.

b. *If the whole system is to be on time, it is necessary that the connecting flights

be 0 as well.

The perfective have data contrast with the above in that have seems to at least marginally

license a null VP. The examples in (134) seem more acceptable than those in (131) to (133)

with be.


170

PERFECTIVE HAVE

(134) a. ?When the laborers have come to a decision, it is important that the leader have 0

as well.

b . ?By the time Wanda finishes, it is necessary that Bob have 0 too.

c. ?Jim memorized the piece and I suggest that, by next lesson, you have 0 too.

If they are acceptable, they again force the conclusion that have is in 1°, from where it can

license ellipsis. This was clearly not the case for the auxiliary be. Because this conclusion

does not impact the remainder of this study, I will not attempt a close account of the facts.

It is not obvious that simply allowing auxiliary have to move to 1° would be empirically

justified; the data are more complex. The subjunctive clause data that I have introduced are

part of a larger body of facts, discussed in Lobeck 1987 and Johnson 1988, pointing to the

inadequacy of treating non-finite perfective have the same as auxiliary be.

It is worth observing in the above context that subjunctive clauses with perfective have are

generally uncomfortable and usually have less strained paraphrases without have. This

awkwardness certainly influences the judgments on such examples when additional

syntactic complexities are involved. Further, it may be that this uncertainty is compensated

for by appealing to judgments on similar finite clauses. The minimal difference in

morphology between have (subjunctive) and has (finite) would facilitate this shift, as

would the loss of subjunctive clauses in some dialects. This interference would serve to

further make the judgments difficult.

In summary, Verb Raising is not present in the syntax of subjunctive clauses, with the

possible exception of perfective have. This observation was accounted for analytically by

positing a null modal in subjunctive 1°; however, the more theory-neutral conclusion is that

verbs and auxiliaries are within VP in English subjunctive complements.


171

6. Conclusion

In this chapter I have laid out three tests for diagnosing movement of a verbal head from the

VP into the inflectional projections of a clause, Verb Raising. The diagnostics, relying on

adverb position, negation, and VP ellipsis, were then used to demonstrate that VR is absent

in two major, but understudied, clause types: imperatives and subjunctive complements.

Superficially, raising might have been expected in these places because of the lack of overt

inflectional heads that could be taken to be preventing the movement. Nevertheless, no

head movement was evidenced.18 From these observations, it seems evident that VR is

limited to a rather small comer of the grammar of English: finite forms of the auxiliaries

have and be, which I will dub finite auxiliary raising (FAR). I would like to conclude that

Verb Raising in English is highly restricted, albeit highly visible. In the larger picture,

English is a language without VR, a conclusion that accords with Emonds 1978 original

comparative investigation of the phenomenon. It patterns with the mainland Scandinavian

languages in which VR is completely absent (Rohrbacher 1994, Vikner 1995) and contrasts

with French and other Romance languages in which VR is frilly productive (Emonds 1978,

Pollock 1989, Rohrbacher 1994). The implication of this conclusion for the present study

is the following: if VR is essentially absent in English, there is no actual expectation of

seeing VR. The analytical burden is not to explain why auxiliaries don’t raise in a particular

situation, for example imperatives, but rather only to state why raising does occur in the

18(i) illustrates using the negation diagnostic that raising is also not found in infinitival complements. Neither a main verb in a nor an auxiliary in b can precede sentential not. This is unsurprising, however, since to reasonably occupies the Infl position and blocks VR.

(i) a. *[ For Bruce to win not ] would be a tragedy, b. *[ For there to be not an election ] is likely.


172

one case. Since this is not fundamentally a study of finite clauses, I will not attempt this.19

To repeat, the underlying assumption in the remainder of this work is that the grammar of

English lacks the syntactic mechanisms that trigger overt Verb Raising, however they are

ultimately formulated.

With regard to imperative syntax, this lack of Verb Raising can explain the often-cited

observation, from the introduction, that imperatives differ from finite clauses in requiring a

form of do even with the auxiliaries have and be, (1) versus (2) repeated below.

This difference can be tied to a more fundamental one: the auxiliaries have and be raise to 1°

in finite clauses but not in imperatives. With respect to their co-occurrence with do, then,

auxiliaries in imperatives should pattern with finite main verbs, in (3), which also do not

19Baker 1991 articulates a structured system of core grammar versus peripheral grammar which is explicitly designed to account for the parochial nature of FAR. The work largely concludes that it is not clear that the motivation for FAR is due to deep theoretical principles rather than simple lexical arbitrariness. Baker argues this position in great depth in considering Pollock’s (1989) and Chomsky’s (1991) analysis of the phenomenon. In weighing descriptive richness against theoretical restrictiveness. Baker determines that the analytical cost o f accounting for English auxiliary behavior is too high if its analysis is admitted into the core grammar. He thus concludes that FAR is a peripheral rule and not an epiphenomenon reflecting deeper principles. His analysis has the effect o f moving the account of auxiliary raising into a less central, language-particular portion of the grammar. This is one possible approach towards explaining FAR. See also Pollock 1989 and Lasnik 1995b.


(1)[Do not bel , „ . ( *Be not } late!/Do not havel ,I *Have not | e

(2)f*do not bel | were not J

(3) We do not arrive *We arrive not

173

raise. While the raising of finite auxiliaries makes the insertion of do into 1° unnecessary or

impossible since 1° is filled, the lack of raising in imperatives does not rule out do from

appearing since 1° is unfilled. Furthermore, the fact that do is forced to appear, just as with

finite main verbs, suggests that Do-Support is involved. Given mechanisms predicting the

appearance of do in finite clauses when there is no raising, that (1) patterns with (3) and not

(2) is expected. Imperatives can be accounted for with the same mechanisms. While this

would seem to require an explicit analysis of Do-Support, I will resist this. I want to

remain independent of the particulars and assert on the following: whatever the correct

analysis of do in finite clauses is, it extends, at no cost, to imperatives because their

syntactic behavior is fully parallel. In chapters 4 and 6 ,1 return to further and more fully

justify this common derivational history for do in the two clause types. The Verb Raising

investigation alone, in revealing one parallel, lends initial credibility to the claim to be

argued for in chapter 4 that do is an inflectional head, just as in finite clauses.

One argument for this assertion will come from the imperative ellipsis data introduced in

this chapter. They are particularly important and have numerous interesting consequences

that will be explored in later chapters. The position taken here is that imperative ellipsis is

possible only when do or don't is present, (136), and ungrammatical otherwise, (135). The

relevant contrast is repeated below.

(135) a. *You should tidy up before the guests arrive. Yes, definitely have 0!

b . *Cinderella is doing her chores so, if you know what’s good for you, be 0 too!

c . *They were examined by that nurse so you be 0 too!

d. *Despite the baseball strike, it’s fun to be a sports fan. So, be 0!


174


b . We want everyone to come, so those who can, please/by all means do 0!

c. Rick walked out of the lecture, but don’t everyone else 0, please!

d. Mom said you shouldn’t be playing in the fireplace, so don’t be 0!

e. The movers told us to be careful with the dining table.

Yes, do be 0! It’s a priceless heirloom.

In conjunction with chapter l ’s theory of VPE, the data argue that do and don’t are

inflectional heads and not imperative particles because they so clearly save ungrammatical

VPE data (chapter 4). The examples also suggest that imperative subjects are in the

canonical subject position, spec,IP since they can be stranded under ellipsis and so must be

outside of VP (chapters 3 and 5). Lastly, they support the claim that the appearance of do is

a result of Do-Support applying in imperatives (chapter 6).

If we are willing to accept that the adverb and negation data alone affirm the lack of raising

in imperatives, then the ellipsis pattern can be taken as independent support for the formal

account of VP ellipsis in chapter 1. These data illustrate that the crucial syntactic element in

licensing a null VP is indeed an inflectional head and not merely an auxiliary element. Like

subjunctive clauses, imperatives provide a test case in which an auxiliary may be present

but it is not in 1°, so the inflectional layer is lexically unfilled. Since the auxiliary is not in

1°, ellipsis is not possible, as predicted. The appearance of do or don’t in an inflectional

head licenses the ellipsis site and renders the data grammatical.


Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.

UNIVERSITY OF CALIFORNIA

SANTA CRUZ

SYNTACTIC ISSUES IN THE ENGLISH IMPERATIVE

A dissertation submitted in partial satisfaction of the requirements for the degree of

DOCTOR OF PHILOSOPHY

in

LINGUISTICS

by

Eric Potsdam

December 1996

The dissertation of Eric Potsdam is approved:

__Jo±2__Q ±z.[dim Aissen, Chair

A \arnelMcCloskey )

Donka Farkas

1 raduate Studies


V o l . 2


175

Chapter 3

Imperative Subjects

1. Introduction

The analyses of the English imperative in this work treat the overt noun phrases that appear

italicized in the imperatives in (1) as ordinary subjects.

(1) a. Tom take out the trash!

b . Everyone write to Congress about this atrocity!

c . Someone get me an aspirin!

d . Those in front step away from the barricade!

In this chapter, I more closely investigate these phrases, which I will neutrally call

IMPERATIVE NOUN PHRASES (INPS), and I offer evidence to support the assumption that

they are subjects. Numerous works attempt to restrict INPs in motivated, although

exceptional, ways, resulting in their being typically non-subject-like. This chapter will

conclude that the syntactic and semantic behavior of INPs is not an exceptional feature of

the English imperative and that they deserve the label ‘subject’. The result is in keeping

with the general theme of this work: that the English imperative is not a syntactically

idiosyncratic construction. We thus remove another obstacle to assimilating the imperative

to more general English clause structure.


176

Section 2 presents background information regarding the identification of entities in a

discourse. I will use the cover term DISCOURSE ROLES to refer to the different groups of

people which are potentially involved in a verbal exchange, mainly speaker and hearers.

The vocabulary and distinctions introduced will be helpful in talking about the uses of INPs

and vocatives. In Section 3 ,1 investigate vocatives and consider the reasonable hypothesis

that INPs as in (1) are vocatives. Claiming that INPs are actually vocatives would avoid the

need to recognize imperative subjects, a marked category cross-linguistically. There are

clear differences between the two, however, even in English where there is no

morphological difference and I offer several diagnostics specific to English that can be used

to differentiate them. Although the tests are not conclusive in every case, their combined

power can rule out vocative status in any given instance. The section concludes that

vocatives are entities syntactically and semantically distinct from INPs—not a new result

(Downing 1969, Stockwell et. al. 1973, Schmerling 1975, Davies 1986a), but a result that

is worth establishing because such confusion is often likely. Section 4 turns to establishing

the actual subject status of the NPs under investigation. It begins with a discussion

regarding the notion of ‘subject’ within the espoused theory of syntax (McCloskey to

appear). Subjects cross-linguistically have certain canonical characteristics and the section

demonstrates that INPs are largely unexceptional when evaluated with respect to these

properties. I show that the expected syntactic behavior, traditionally taken for granted, does

obtain. I then consider formal restrictions that have been claimed to hold of the imperative

noun phrase. These semantic restrictions have often been the motivation for giving INPs

unique status and for appealing to unusual syntactic mechanisms (Beukema and Coopmans

1989, Platzack and Rosengren 1994). A survey of possible INPs reasonably demonstrates

that there are few, if any, real constraints on INPs that cannot be explained with reference

to the use of the imperative instead. I thus agree with Davies 1986a in concluding that these


177

putative restrictions are not encoded in the grammar in a formal way. Showing that such

restrictions are only apparent removes a primary road block to assimilating imperative

clauses to more general English clause structure. I conclude in section 4 that INPs are

syntactically and semantically subjects.

Accepting that INPs are subjects, Section 5 analyzes the ZERO, NULL, or EMPTY SUBJECT

that is more typical of imperatives. I will atheoretically call such examples SUBJECTLESS

IMPERATIVES.

(2) a. Take out the trash!

b . Write to Congress about this atrocity!

c. Get me an aspirin!

d . Step away from the barricade!

This absence of an overt subject is striking and central to many theoretical issues because

such empty subjects are almost universally allowed in imperatives, independent of the

presence or absence of zero subjects elsewhere in a language’s grammar (Sadock and

Zwicky 1985). Section 5 considers the representation and interpretation of the null subject,

arguing that it is a null pronominal, as found in numerous so-called pro-drop languages. Its

interpretation and distribution follow from characteristics of the imperative. This is a

potentially new and significant result as it introduces a large and important body of data that

I claim is relevant for the theory of null pronominals but that has previously been ignored.

Summarily, the chapter argues that the noun phrases illustrated in (1) and the

phonologically null subject underlying the examples in (2) are subjects in the standard

sense and thus do not deserve special syntactic analyses.


178

2 . Discourse Roles

In order to more precisely talk about the function of various noun phrases found in the

imperative, it is necessary to define the various entities in a discourse that such noun

phrases typically refer to. In this section I introduce some terminology to talk about the

discourse functions of argument noun phrases and vocatives. The terms are an extension of

the traditional discourse roles of speaker and hearer. Several researchers (e.g. Goffman

1975, Goodwin 1981, Clark and Carlson 1982, Levinson 1983) have observed the

inadequacy of recognizing only speaker and hearer, however; so I propose a more fine

grained distinction and offer further support for it later in the chapter by showing its utility

in talking about imperative subjects and vocatives.

Conversations among more than two people proceed largely as an exchange of information

between various actors. During the course of the interaction, the roles played by individual

actors may change as the source and direction of information flow change. I conceive of the

discourse as including at least the DISCOURSE ROLES in (3), which I define below.

(3) Discourse Role Relations

PARTICIPANTS

SPEAKER

ADDRESSEE SIDE-PARTICIPANTS

I will be considering only entities in the intended audience and will call them

PARTICIPANTS: those who are intentionally party to, as well as receive and interpret, the

utterances of the speaker. The participants are the traditional actors in a conversation and

include the speaker, the addressee, and the side-participants. The SPEAKER is the issuer of

the utterance and s/he presents the utterance to the remainder of the audience, the HEARERS.


179

Hearer is a traditional role but here I subdivide it into two types. The ADDRESSEE is the

specific target of what is being said, the one for whom the speaker most directly designs

his utterances. The addressee may be a group of entities. SlDE-PARTIClPANTS are other

hearers, the participants less the speaker and addressee. They crucially differ from the

addressee in that they are not the direct target of the speaker’s utterance, although they are

hearers. They are part of the discourse still but they are not the addressee. As previously

stated, the roles that the actors assume will typically change throughout a discourse and in

the usual case a participant takes on at some point each of the roles of speaker, addressee,

and side-participant. Of the three roles, speaker and addressee can be distinguished as

always being required in a discourse; side-participants are optional.

I rely on Clark and Carlson 1982 in justifying and defining these terms more precisely.

This reference should be consulted for argumentation that at least the three roles designated

above are necessary, in particular, the distinction between addressee and side-participant.

Crucial to their definitions is the assumption that all utterances in a conversation involve an

illocutionary speech act (Austin 1962, Searle 1969). An illocutionary act is the act of

making a statement, request, order, promise, etc. in uttering a sentence by virtue of a

communicative force associated with the utterance. The speaker is defined as the agent or

source of this illocutionary act, the person who performs it. In the unmarked case, the

speaker’s utterance will have a conventional force which s/he intends to convey. If the

speaker is uttering a declarative, it is the act of informing. If an imperative is being uttered,

it is the directive act that is associated with the imperative. The hearers are those whom the

speaker intends to ‘take part in’ the illocutionary act. That is, they are chosen audience

members whom the speaker wishes to have recognize his illocutionary intentions. Among

the hearers, the addressee is privileged in being the direct target of the illocutionary act. The

addressee is the only required target and the one to whom the act is aimed. In the case of a


180

statement, the addressee is the one being informed by the speaker. While (any part of) the

information in the statement may be new to other participants, the speaker’s intentions are

that at least the addressee be targeted. If s/he intends for others to be informed, they too

become addressees. In the case of an imperative, the addressee is the actor or actors to

whom the directive is issued, the one who is to act to bring about some state of affairs. The

roles are summarized in (4).1

(4) Participant Roles (after Clark and Carlson 1982)

a. HEARERS: intended understanders of the illocutionary act

b . SPEAKER: source o f the illocutionary act

c . ADDRESSEE: target of the illocutionary act

d. SIDE-PARTICIPANT: non-speaker, non-addressee participants

To illustrate, consider a conversation taking place on a bus between three people: Al, Bea,

and Cici. (5) is a part of the exchange.

*An alternative way of explicating these roles is in terms o f the concept of a common ground (Grice 1975; Stalnaker 1974, 1979; Heim 1988). The COMMON GROUND as defined by Stalnaker is the set of presuppositions that participants in a conversation all share. It is a set o f mutually-held background assumptions. Participants in a discourse share a common ground and are working towards specifying the body of information in it. Rather than targeting participants with an illocutionary act, as above, the speaker’s goal is to update and further the common ground. Under this view as well, the addressee is the most directly affected participant. When a statement is made, the common ground is changed in complex ways. The addressee is the one who is to update her/his knowledge base to make it compatible with the common ground. Although an explication of discourse roles in such terms has the advantage that the machinery is explicit in the literature, it is less helpful in the current context because it does not easily extend beyond assertions to imperatives (Heim 1988:290).


181

(5) [A1 to Bea, in front of Cici:]

At Bea, Cici and I saw a great movie last night.

PARTICIPANTS: A l, Bea, Cici

SPEAKER: A l

HEARERS: Bea, Cici

ADDRESSEE: Bea

SIDE-PARTICIPANT: Cici

We can identify the various roles in this brief conversation. Collectively, Al, Bea, and Cici

are the participants. The illocutionary force of the utterance is one of informing (a

constative in Austin’s (1962) sense). Al is the speaker since he is the agent of the act, the

source of the information. Bea is the addressee as target of the informing and Cici is a side-

participant. It is evident that Bea and Cici do not have the same status as hearers since

informing Bea of (5) is felicitous, whereas, informing Cici of (5) is not—she is already

aware of the information.2 This is the motivation for making a distinction among hearers

and distinguishing addressee from side-participants. Only the former is the target of the

speaker’s illocutionary act or, alternatively, only her/his knowledge is intentionally affected

by the speaker. At the same time, although only Bea is the target of the illocutionary act,

Cici must nevertheless record the fact that Al has informed Bea of their doings the previous

evening. It is in this sense that she is still a hearer/participant and understands Al’s

illocutionary intent.

2In terms of a common ground, informing Bea has the affect of requiring her to update her knowledge base so as to be compatible with the common ground. Cici is not required to take this action, although she is passively affected by the new common ground.


182

The discourse roles introduced above are pragmatic terms that are useful for explicating the

structure of discourse and the roles that various participants take on. They do not

necessarily have direct grammatical reflexes.3 Nevertheless, it is worth looking at how

these roles relate to the grammatical agreement features of person and number: first (1),

second (2), and third (3), and singular (sg.) and plural (pi.). The English pronoun system

is presented in (6) and the relationships are in (7). The examples below explicate the uses

of these pronouns and their various reference possibilities. The roles of speaker and

addressee are central to the definitions, although the correspondence is not direct in all

cases. The relationship between the discourse roles and pronominal reference involves

inclusion rather than identity: the referent of a first person pronoun is restricted in that it

must include the speaker; the referent of a second person pronoun must include the

addressee, but not the speaker; and a third person pronoun’s referent must not include

either the speaker or the addressee. Summarily, if the referent of a noun phrase includes the

speaker, it will be first person; otherwise, if the referent includes the addressee, it will be

second person; otherwise, the noun phrase will be third person.

3For example, Benveniste 1966 and Farkas 1988 argue that speaker and addressee, realized as first and second person pronouns respectively, are united in various syntactic phenomena. According to the discourse structure above, however, they do not form a natural class. On the other hand, they are still unique in being the only required discourse roles and can thus be differentiated in that manner. It is not clear whether the discourse structure above needs to explicitly encode this grouping.


183

(6) English Pronominal System

a. 1 person singular: I, me, my, mine, myself

b . 1 person plural: we, us, our, ours, ourselves

c. 2 person singular: you, your, yours, yourself

d. 2 person plural: you, your, yours, yourselves

e. 3 person singular he/she/it, him/her, his/her s/its, himself/herself/itself

f . 3 person plural: they, them, their, theirs, themselves

(7) Relationship between Discourse Roles and Person/Number Specifications

a. 1st person, singular: speaker

b . 1st person, plural: speaker plus others

c. 2nd person, singular: addressee

d . 2nd person, plural: addressee plus others not the speaker

e. 3rd person: neither speaker nor addressee

The first person singular (lsg.) pronouns in (6)a clearly refer to the speaker, as indicated in

(7)a. The corresponding first person plural (lpl.) forms in (6)b do not simply represent

plural speakers, as might be expected, but rather the speaker plus any others, who may be

the addressee or not, (7)b. The pronouns in (8)a are lsg. and show speaker reference. The

pronouns in (8)b through e show various lpl. reference possibilities. The pronouns in (8)b

are lpl. referring to plural speakers alone. (8)c pronouns are lpl. referring to the speaker

and addressee. In (8)d the pronouns refer to the speaker and others neither the addressee

nor another speaker. The pronoun in (8)e refers to the speaker, addressee, and a third

person. Some of the examples in (8) are from Quirk et. al. 1985.


184

(8) a. I hurt myself.

b . We, the undersigned, do hereby relinquish all rights to the property.

c. We bet on the wrong horse, Willy.

d. Mary and I, we can take care of ourselves, thank you.

e. You, Ann, and I, we are working ourselves to death.

A parallel pattern occurs with second person, which will be of interest below in discussing

vocatives. (7)c indicates that singular second person pronouns (2sg.) in (6)c refer only to

the addressee, as in (9)a. Again, second person plural (2pl.) refers not to plural addressees

but to the addressee plus others who are not the speaker, (7)d. (9)b shows the pronoun

referring to multiple addressees. (9)c illustrates reference to the addressee plus another

person who is neither speaker nor addressee.

(9) a. You should behave yourself, Bobby.

b. You ought to proud of yourselves, soldiers.

c. You and John, you' 11 have to fend for yourselves.

Lastly, the third person pronouns (3sg./3pl.) in (6)e and f normally refer to

anyone/anything that is neither the speaker nor the addressee as indicated in (7)e. The

interpretation of the examples in (10) precludes including the speaker or addressee in the

referent of the pronoun.

(10) a. He has hurt himself evidently.

b . Those players, they just throw themselves into the game with such abandon.


185

3 . Vocatives

Virtually all of the analyses of imperatives within the Standard Theory, since Chomsky

1975, take the imperative noun phrases under consideration here to be subjects. Thome

1966 is perhaps alone in arguing that they are vocatives. Despite this, the hypothesis that

these noun phrases are vocatives is a reasonable one. Vocatives naturally show up in

sentences with directive force, in the same positions, and analyzing INPs as vocatives

would eliminate the need for imperative subjects in English. After all, imperative subjects

are cross-linguistically highly marked. For this reason, I will consider the English vocative

more carefully below and give several diagnostics that differentiate vocatives and INPs, as

seems necessary. Such tests, although implicit in many discussions of INPs, have never

been carefully laid out. Section 3.1 presents the behavior of vocatives and the various tests

for identifying them. The main conclusion in section 3.2 will be that the vocative is a

distinct syntactic entity from INPs and we must search elsewhere for an analysis of INPs. I

will use the section to summarize the behavior of vocatives and apply the appropriate

diagnostics to illustrate that the INP that follows do(n’t) in negative imperatives is not a

vocative.

3 .1 . Distinguishing Vocatives and Imperative Noun Phrases

The question that I will be considering in this section is: given a noun phrase in an

imperative, can we confidently determine whether or not it is a vocative. If we can, then it

opens the possibility that a noun phrase under consideration which does not meet standards

of vocativehood is a subject instead. By vocative, I mean the noun phrases italicized in (11)

which are typically used to call someone’s attention or address them in a particular manner

(Zwicky 1974).


186

(11) a. Bond, do you have a light?

b . Slowpoke, we’d better get moving very soon.

c. Sir, the table is prepared.

d. You, why are there still toys scattered on the floor?

e. You with the glasses, there are seats closer to the front.

f . Those with tickets, the auditorium is now open.

g . The boy without a tie, that kind of attire won’t do.

The sections below lay out three diagnostics for identifying vocatives: intonation, reference

restrictions and anaphoric agreement. The results of these tests applied to imperatives

indicate that there is another kind of noun phrase that is possible in an imperative beyond a

vocative. The two have different distributions, syntactic behavior, and discourse function.

The conclusion is that INPs cannot be equated with vocatives in all cases.

In developing the diagnostics, I will first consider vocatives in declarative sentences. This

domain is independent of imperatives and assures that we are not inadvertently confusing

vocatives and INPs, since none of the latter will initially be involved. This methodology

aims to avoid the circularity of establishing a set of diagnostics which depend upon one

another for verification and have no proven independent validity. To the extent possible, I

will also avoid examples that could potentially be understood with an appositive reading or

left dislocation reading—that is, a peripheral, vocative-like phrase that is coreferential with

another noun phrase in the sentence. Once a characteristic of vocatives is established with

declarative clauses, I will then show that it carries over to imperatives and that there are

imperative noun phrases that do not have the particular property.


187

3.1.1. Intonation

Perhaps the best known property of vocatives is their distinct intonation contours. They

constitute a separate intonational phrase from the rest of the sentence and are typically

accompanied by a rising intonation. Informally, vocatives are followed by a pause that is

not present with subjects. Subjects and other arguments are a part of their clause in a way

that vocatives are not. In the data, this is reflected with punctuation; vocatives are set off by

commas. Some examples are given in (12).

(12) a. Mary, is there a performance this evening?

b . The man with the newspaper, what does it say about tomorrow’s weather?

c. Waiter, we need more rolls please.

d. Everyone, we’re leaving in fifteen minutes.

McCawley 1988 analyzes vocatives as not being part of the sentence even though they are

integrated into the string of words. He asserts that they don’t act as a syntactic unit within

the sentence despite occurring so linearly. They function more like interjections. This

would account for their separate intonation contour. It should be noted that the examples in

(12) are not really English if said without the intonation break. Vocatives obligatorily

constitute a separate intonational phrase;

(13) a. *Mary is there a performance this evening?

b . *The man with the newspaper what does it say about tomorrow’s weather?

c . *Waiter we need more rolls please.

d. *Everyone we’re leaving in fifteen minutes.

This vocative intonation can be applied to INPs, as illustrated in (14).


188

(14) a. Mary, go to the store!

b . You with the newspaper, read us the headlines!

c. Everyone, put your pencils down now!

There are instances of INPs, however, which may be pronounced with no intonational

break between them and the rest of the imperative:

(15) a. Mary go to the store and Bob stay here and help me with the dinner!

b . The man with the newspaper read us the headlines!

c . Everyone leave in fifteen minutes!

These noun phrases are integrated into the (single) intonational phrase of the main clause.

They may have the same intonation as ordinary declarative clause subjects, given in (16)

for comparison.

(16) a. Mary should go to the store and Bob will stay here and help me with the dinner.

b . The man with the newspaper just read us the headlines.

c. Everyone is to leave in fifteen minutes.

To the extent that intonation can be reliably determined, it indicates that not all INPs are

vocatives. Some must be distinct. While vocatives must form a separate intonational phrase

on their own, other INPs need not. If INPs were always vocatives, the lack of an

intonation break in the imperative examples in (15) would be unexplained.

3.1.2. Referent

I consider here the referent of a vocative and claim that it corresponds exactly to the notion

of addressee introduced above. By recognizing the distinction among the various audience

members as we have done, we can correctly and precisely characterize the vocative. With


189

this description of a vocative’s referent, I turn to imperatives and show that there are some

INPs that do not refer to the addressee and are, in fact, unacceptable as vocatives in

declarative clauses. This is surprising if INPs are simply vocatives and it again argues for

their distinctness.

It seems to be unarguably and even definitionally true that the vocative picks out the

addressee as defined above (Downing 1969; Zwicky 1974; Haverkate 1978; Levinson

1983; Davies 1986a, 1986b; Hamblin 1987; and others). When a vocative is present, it

serves to identify the person(s) to whom the utterance is specifically addressed. I

summarize this in (17).

(17) Vocative

A vocative necessarily refers exclusively to the addressee

In systems that don’t recognize at least the discourse roles that were introduced above, the

notion of addressee that seems necessary for defining vocatives is conflated with hearers

more generally. Such a position is typified by Downes 1977, who refers to “the hearer of

the utterance, the addressee”. The role of hearer, which includes in its referent both the

addressee and side-participants of the discourse structure in (3), is inappropriately broad as

the definition of a vocative, however. I return to the bus scenario repeated below to

illustrate the inadequacy of such a claim.

(5) [A1 to Bea, in front of Cici:]

Al: Bea, Cici and I saw a great movie last night.

HEARERS: Bea, and Cici

ADDRESSEE: Bea


190

In this conversation, it was determined that the hearers were Bea and Cici. The addressee

was claimed to be Bea alone, who is also explicitly the vocative in Al’s utterance in (5). In

order to accommodate the hypothesis that the vocative refers to the hearers, we must claim

that, in (5), the speaker has restricted his audience to Bea alone, excluding Cici among

others. In other words, the participants in this conversation are, temporarily, really only

Bea and the speaker, Al. Such a hypothesis, however, makes the unreasonable claim that

Cici is no longer a hearer, that she is not a party to the statement that Al is making and not

intended to understand it. This is neither intuitively nor pragmatically acceptable. As

discussed earlier, we need to recognize that Cici is a central part of the conversation despite

not directly being the target of the information. This is reflected in the fact that her

knowledge about the conversational context will need to include the fact that Bea now

knows (5). If this were not the case, any continuation of the conversation by Cici

acknowledging this, such as “I wish we had called you to join us”, would be infelicitous—

contrary to fact. The alternative is to accept that Cici remains a hearer but the vocative does

not pick out the hearers, only a portion of it, the addressee.

I now present two additional sets of data that support the proposed characterization of the

vocative. In the first set, the impossibility of various vocative forms is predicted given

(17). In the second, an unusual agreement pattern with vocatives in older forms of English

is adduced as superficial support.

(18) presents several cases of potential vocatives that are ungrammatical. The vocatives are

unacceptable for the reason that the noun phrase in each case is unable to refer to the

addressee.


191

(18) a. *Hey, you and Fred, did no one say to stay out of the construction zone?

b . * Your children, can the play area be kept clean?

c . * He/Him, will there be a protest this evening?

d . *Nobody, that man just rode off with my bicycle!

According to (7) and the previous discussion of grammatical person specification, second

person pronouns obligatorily include the addressee in their referent. If the pronoun is

singular, its referent is identical to the addressee. This leads to the conclusion that a

vocative noun phrase may be the second person singular pronoun since it picks out

precisely the addressee but that no vocative will be able to contain a second person

pronoun. In such a case the vocative’s referent would need to be dependent upon itself and

would not be able to refer solely to the addressee.4 In (18)a, to illustrate, the singular

second person pronoun you of you and Fred is the addressee. If you picks out the

addressee, then, you and Fred cannot also do so. You and Fred will therefore be ruled out

as a vocative. (18)b is similarly excluded. The vocative your children refers to the

addressee’s children and therefore cannot be the addressee itself. It will correctly be an

unacceptable vocative.

The third person pronoun in (18)c, repeated below, is also not a licit vocative.

(18) c . * He/Him, will there be a protest this evening?

The discussion on pronominal person above determined that third person pronouns referred

to entities which were neither the speaker nor the addressee. Given this latter restriction, the

4Formally this may be ruled out by the accessibility requirement (Chomsky 1981) or the i-within-/ condition. More generally, circularity of reference must be excluded.


192

use of a third person pronoun as a vocative is transparently prohibited. There is no way for

it to correspond to the addressee.

Lastly, (18)d is excluded for yet a different reason. Nobody is clearly an inappropriate

addressee:

(18) d . *Nobody, that man just rode off with my bicycle!

The set of people that nobody picks out is empty. Consequently, the addressee will not be

able to appear in this set and the negative quantifier nobody, and others like it, will always

be ungrammatical as vocatives. This yields the following clear contrast between the subject

in (19)a and the vocative in (19)b. I will refer to this as the nobody-test: negative

quantifiers cannot be vocatives.

(19) a. Nobody should get his clothes dirty.

b . *Nobody, you should get your clothes dirty.

The identity between vocative the vocative’s referent and the addressee is further supported

by languages that have grammaticized the relationship to some degree. Remembering that

the grammatical realization of the addressee is second person, we might expect the vocative

to show second person agreement in some situations, for example, when a vocative is

modified. This was the case with older forms of English. In the Middle and Early Modem

English examples in (20) from Visser 1963a:92, a relative clause on a vocative could show

second person subject-verb agreement where Modem English would use third person


193

agreement.5 The dates of the examples are given in parentheses. They show that the

connection between the vocative and grammatical second person, and hence the addressee,

is particularly close. A framework in which to understand this pattern of agreement is

offered in the appendix.

(20) a. Lazarus, my sonne, whech art (2 person) ther brothyr ... (ca. 1450)

b . O limed soul, that, struggling to be free, art (2 pers.) more engaged!... (1602)

c. Pope, who couldst (2 pers.) make Immortals,... (1742)

d. God! Who didst (2 pers.) name the day ... (1813)

These examples and the predicted illicit vocatives above support the claim that a vocative

does refer to the addressee.

Before moving on, I investigate in the next subsection several interesting examples of

vocatives that are apparently problematic for this analysis and the definition in (17). I then

return to imperatives and show that, based on the characterization argued for above, there

are INPs that cannot be vocatives.

5Downing 1969 claims the same pattern for French, citing the example in (i) from the Port Royal Grammar of 1676.

(i) Soleilqui voyez toutes choses...sun which see.2pl. all.plural things ‘Sun who sees all th ings,. . .’

Jim McCloskey (personal communication) indicates that Irish shows a similar pattern:

(i) A M hdirea bhfuil an fhidil leat, ...VOC Mary COMP is the fiddle with.yoa “Mary who’s carrying the fiddle,...”


194

3.1.2.1. quantificational and indefinite vocatives

The examples of vocatives above are all grammatical because the italicized noun phrases are

definite and identify their referent in a given context. They allow the vocative to pick out the

addressee. Examples, repeated in (11), include names or epithets in ( ll)a , b and c,

appropriate pronouns in d and e, and definite descriptions in f and g.

(11) a. Bond, do you have a light?

b. Slowpoke, we’d better get moving very soon.

c. Sir, the table is prepared.

d. You, why are there still toys scattered on the floor?

e. You with the glasses, there are seats closer to the front.

f. Those with tickets, the auditorium is now open.

g- The boy without a tie, that kind of attire won’t do.

Given this observation, it is surprising that the quantificational and indefinite noun phrases

in (21) can also be used vocatively. Such noun phrases are arguably non-referential and do

not pick out a known entity in the discourse. It is not clear in what sense they identify the

addressee that the definition in (17) requires.

(21) a. Will you please follow me, everyone.

b . All of you, the teacher said to stop messing around and listen up.

c . A doctor ought to be called, someone.6

^The judgments in literature on examples involving someone as a vocative are inconsistent Davies 1986a marks such examples as ?. Downing 1969 and Thome 1966 claim that they are ungrammatical. I will agree with Stockwell et. al. 1973 and Quirk et. al. 1985 and assert that they are acceptable, as my consultants confirm.


195

d. I need to borrow a jacket, one o f you.

e. I want my map back, whichever o f you has it.

f . The last one to leave, the lights really shouldn’t stay on overnight.

g. Anyone who isn ’t busy, the storeroom could use a little straightening up.

h. Any o f the girls, can you please go find out what is holding Sue up.

i. ?A fourth-year student, can you please assist with the surgery?

j . Whoever wrote on the walls, the principal would like the drawings cleaned up.

A key to understanding such examples is to observe that these vocatives do not have an

open domain of possible reference but are restricted to ranging over the discourse

participants, intended hearers of the utterance. Many of the examples in fact sound best

with an overt partitive o f you which makes the domain restriction to the participants more

transparent. These vocatives thus correspond to an entity or entities among the participants,

the difficulty is that this entity is not referentially identified. I will suggest that this entity is

nevertheless the addressee. Recall the definition of addressee, from (4), as being the target

of the utterance’s illocutionary act. Such a definition allows the addressee to remain under

determined, as long as the illocutionary act is directed at someone. If the speaker does not

have enough information to name the addressee but believes that one exists, an indefinite or

quantificational vocative should be appropriate. I will explicate this idea using (21)f and the

expanded scenario:

(21) f . [Kay, Lee, and Mike are working late at the office. The boss leaves, saying:]

The last one to quit, the lights really shouldn’t stay on overnight.

The illocutionary act associated with (21)f on the most conventional interpretation is a

request to turn out the lights. The hearers here are Kay, Lee, and Mike; they are the ones

whom the speaker intends to understand the utterance. The boss is the speaker, but the


196

crucial question is who is the addressee, the target of the request? Three answers seem

possible. It is either 1) the (entire group of) hearers, 2) no one, or 3) a single one of them,

albeit unidentified, the one who will quit last. Given the definition of addressee in (4), we

are forced to adopt the latter option. The request in (21)f is not asked of the entire hearer

group or none of them; it is clearly directed at precisely one of them—specifically, the last

one to quit. The addressee in (21)f will therefore be the last person to quit working. This is

also precisely the vocative, the last one to quit. The difficulty with this account is that there

is an addressee who is nevertheless not referentially identified at the time of the utterance.

In (21)f, the addressee is one of Kay, Lee, or Mike, but we cannot say which one. The

reason that the examples in (21) are acceptable, then, is that the vocative does identify the

addressee. We must give up on the commonly accepted idea that the addressee can be

referentially determined in all situations. It appears that description of the addressee is

sufficient.

An observation of cross-linguistic interest is that this pattern is not found in all languages.

While some languages, such as English, German, and Irish, do permit such examples,

others do not. In the Romance languages, or Greek illustrated below,7 definite vocatives

parallel to (11) are permitted, (22), but non-referential vocatives are impossible, (23).

7Thanks to Anastasia Giannakidou for insightful help with these and other Greek data in this chapter. The left-peripheral noun phrases can be identified as vocatives because they are separated from the rest of the sentence by an intonation break, lack the definite article which is otherwise required in most contexts, and/or have overt vocative morphology.


197

(22) a. |p ^ j e}, anikse tin porta!

{pauLvoc} °Pen-imP-2sg- ^ door “You/Paul, open the door!”

b. Pedhia, anikste tin porta!children open.imp.2pl. the door “Children, open the door!”

(23) a. *Kathenas, ela edho!everyone come.imp.2sg. here

“Everyone, come here!”

b. * j voithise me me ta piata!

{“ r a ^ y . vT } telp.unp.2sg. me with the dishes“Somebody, help me with the dishes!”

c. *Enas apo sas, ela sto grafio mu amesos!one from you come.imp.2sg. to.the office my immediately

“One of you, come to my office immediately!”

d. *1 epivates ia Athina, katevite edho!the passengers for Athens get-off.imp.2pl. here

“Passengers for Athens, get off here!”

It appears that the generalization in these languages is that vocatives noun phrases must be

referential. They must pick out an identifiable addressee with no quantificational structure

permitted. I leave investigation of this proposal for future work. The difference exemplified

by English versus Greek is an interesting contrast that ultimately deserves an explanation.

3.1.2.2. INPs and the addressee

I now return to imperatives to demonstrate that INPs and vocatives are distinct with respect

to referential possibilities. To do so, it suffices to show that there are grammatical INPs that

are not the addressee. If an INP can be a non-addressee, then it cannot be a vocative. Three

such illicit vocatives were discussed in (18), repeated below as (24). These noun phrases


198

could not function as vocatives because they could not be identical with the addressee. In

direct contrast, (25) illustrates that they are fine as INPs.


b . * Your children, can the play area be kept clean?

c . * Nobody, that man just rode off with my bicycle!

(25) a. You and Fred stay out of the construction zone!

b . Your men guard the front while we creep round to the back!

c. Nobody ride off with my bicycle!

The grammaticality difference with the nobody-test is particularly compelling. As before,

nobody is clearly not grammatical as a vocative, (24)c, but it is fully acceptable as an INP,

in (25)c. Such examples support the claim that INPs and vocatives are distinct. We thus

have a test for vocativehood: if an INP refers to a non-addressee then it is not a vocative.

This diagnostic will not identify every INP as a non-vocative since many INPs are also the

addressee and the test is thus silent; however, if one wishes to construct examples that rule

out a vocative interpretation the method is straightforward.8 Below I turn to a final

additional diagnostic.

8The reverse situation also exists; namely, there are noun phrases that are acceptable as vocatives but not as INPs. Such examples also demonstrate that INPs and vocatives are distinct. The test is of limited use, however, because such NPs belong to a restricted set and are readily identifiable. Quirk et. al. 1985 and Davies 1986b survey many such cases that include occupations (waiter, driver, doctor, etc.), epithets {dear, sweetie-pie, honey, coward, idiot, asshole, etc.), and general nouns {brother, buddy, partner, etc.). These NPs are acceptable vocatives, (i), but not declarative clause subjects, (ii).

(i) a. Waiter, are our dinners ready yet?b. Dear, where did we put the matches?c. Partner, I think the saloon’s closed.


199

3.1.3. Anaphora

It is an often cited observation (Downing 1969, Stockwell et. al. 1973, Cohen 1976,

Davies 1986a, Zhang 1990 among many others) that imperative subjects and vocatives

differ in the patterns of pronominal agreement which they trigger. I lay out these patterns

below and discuss how they demonstrate the distinctness of the two noun phrases under

consideration. The pattern that will emerge is that vocatives are restricted to second person

agreement but there are noun phrases in imperatives which systematically allow third

person agreement in addition to second person. Since vocatives do not permit this former

option, such noun phrases must be distinct from vocatives.

The declarative examples in (26) show that vocatives are restricted to triggering second

person agreement on pronouns with which they are coreferential. Coreference in the data is

indicated by italicization. Although all the vocatives are syntactically third person, they

require resumed reference with a second person pronoun. This is partly unsurprising given

the characterization of the vocative as identical to the addressee.

(ii) a. * Waiter is bringing our drinks.b. *Dear is in the kitchen.c. *Partner moseyed up to the bar.

(iii) illustrates that the same NPs are also bad in imperatives without vocative intonation. In these examples, then, they cannot be vocatives.

(iii) a. * Waiter bring our drinks!b. *Dear come here!c. * Partner round up the horses!


200

(26) a. Bond, do(es)you* /

* /tehave a light?

b. Those with tickets, may take j^ g ^ -J seats.

c. ■S’ir, we have table prepared.

d. Dave, we’d better get to a doctor.

e. Ladies and gentlemen, I welcome to the show.

The same agreement pattern is possible in imperatives with INPs. They may trigger second

person pronominal agreement as illustrated in (27). INPs differ from the vocatives in (26),

however, in that third person agreement is also possible. (26) showed that this is

impossible for vocatives.

(27) a. Somebody lend me | - ^ , r | handkerchief!

b . Everyone pick up | - ^ rJ bill on the way out!

c. One o f the boys test while I wait!

d. Drivers have money ready!

e. Nobody forget j ^ - ^ J lunch for the picnic tomorrow!

f . Whoever knows the answer put up hand!

g . People interested in the project give me a way in which to contact !

Unless there is something special about the imperative that allows a vocative to trigger third

person agreement in them, it seems reasonable to conclude that the INPs are not always

vocatives. In particular, when a third person coreferential pronoun is allowed, the INP is

not a vocative.


201

One might interpret the agreement pattern as follows: vocatives permit exclusively second

person pronouns and non-vocative INPs permit only third person coreference. If we accept

this characterization, then the examples in (27) are systematically ambiguous between

vocative and non-vocative readings. Unfortunately, the facts do not permit such a simple

description. In example (27)e in particular the negative quantifier nobody, which we saw

could not be a vocative, may agree with either second or third person pronouns.

Additionally, none of the examples requires the characteristic vocative intonation in order to

get the second person pronoun. The correct pattern is apparently that vocatives are

restricted to second person agreement but non-vocative INPs—which I will argue later are

imperative subjects—may trigger second or third person agreement. I offer an account of

this agreement pattern in the appendix.

3 .2 . Conclusions

To summarize, this section has demonstrated that INPs are not always vocatives. In

particular, three diagnostics, outlined in (28), can be used to distinguish the two. Vocatives

have a separate intonation contour from the remainder of the sentence, trigger only second

person on coreferential pronouns, and must uniquely refer to the addressee in discourse.

Some INPs are not so restricted. They have the intonation of clausal subjects, need not

refer only to the addressee, and can trigger second or third person agreement.

(28) VOCATIVE SUBJECT

Again, I conclude from this evidence that INPs are distinct from vocatives. In the section to

follow, I will pursue the hypothesis that they are in fact subjects. Before doing so,

separate intonational phrase identical to the addressee agreement is second person

intonationally integrated into clause not restricted in discourse role agreement is second or third person


202

however, I turn to a examination of the noun phrases that follow do and don’t in non

neutral imperatives, in (29) and (30). I use the above diagnostics to demonstrate that they

also belong in this class of non-vocative INPs. Consequently, they too may be included in

the investigation of the next section.

(29) a. Don’t EVERY one leave!


c . Don’t anyone touch my stuff!

(30) a. Do SOMEone help him quickly!

b . Do AT LEAST SOME o f you give it a try!

c. Do AT LEAST YOU have a go, even if the others won’t!

First, with regard to intonation, it is evident that the italicized noun phrases above do not

constitute a separate intonational phrase. There is no pause before or after the noun phrase.

This is not to say that such an intonation is impossible. Davies offers (31)a to show that

post-don't noun phrases may be vocatives and, in that case, they have vocative intonation.

Other examples are also given in (3 l)b through d.

(31) a. Don’t, my dear friends, contribute to the campaign!

b . Don’t, you cheapskates, even suggest that I should pay for this!

c. Do, everyone, come to the opening night gala!

d. Do, you clod, be more sympathetic next time!

Thus it seems that both vocatives and non-vocatives are possible in this position. At the

same time, the intonation difference between (29) and (30) versus (31) is quite clear and

supports the independence of these INPs from vocatives.


203

Looking at the referent of these noun phrases confirms the non-unity. Consider the

meaning difference between (31)a and (32), below. In the former, the vocative necessarily

corresponds to the addressee, the set of people characterized by the speaker as his dear

friends. (31)a is an admonition to the addressee to not contribute to the campaign. (32)

contains a non-vocative noun phrase and is also an admonition to the speaker’s friends.

The group to which this noun phrase refers, however, need not be the addressee. The most

natural interpretation of (32) is one in which the speaker’s friends as a group contrast with

someone else’s and, as such, the noun phrase refers to a clear subset of the addressee. The

two examples thus provide a clear contrast and indicate that the noun phrases under

consideration are not always vocatives.

(31) a. Don’t, my dear friends, contribute to the campaign!

(32) Don’t my friends contribute to the campaign!

Further examples of noun phrases which do not correspond exactly to the addressee by the

above discussion (see (24) and (25)) and so are not vocatives are given in (33).

(33) a. Don’t you and Fred go playing in the rose garden!

b . Don’t YOUR men behave like that!

c. %Don’t nobody touch nothin’!

d . ?Do at least your OLDER children come camping with us!

Lastly, with respect to anaphora, post-do(n’t) noun phrases readily permit third person

agreement, in (34). This was seen to be impossible with vocatives, which allowed only

second person agreement. The italicized noun phrases in (34) cannot be vocatives.


204

(34) a. Don’t everybody raise his hand at once!

b. Don’t anyone cut himself!

c. Don’t those with cars forget that they promised to drive tomorrow!

d. Do SOMEONE lend me his shirt to stop this bleeding!

I conclude that noun phrases following do(n't) are also not vocatives and have the same

status as the other non-vocative INPs.

4 . Imperative Subjects

Having established what INPs are not, in this section I propose what INPs are—subjects,

unsurprisingly. The approach will be two pronged: syntactic and semantic. Section 4.1 first

considers what it means to be a (syntactic) subject within the theory, following closely the

discussion in McCloskey to appear. I demonstrate that, syntactically, imperative subjects

are syntactically parallel to uncontroversial subjects of other clause types in having certain

canonical properties and undergoing characteristic processes. I argue against the syntactic

analysis of INPs in Beukema and Coopmans 1989 which, although recognizing the

subject-like status of INPs, ascribes non-standard syntactic operations to them. The

conclusion is that INPs behave syntactically like subjects. Section 4.2 considers the

semantic range of INPs and argues that this too is not uncharacteristic of subjects.

Conventional wisdom (e.g. Schmerling 1982, Zwicky 1988, Beukema and Coopmans

1989, Platzack and Rosengren 1994) has it that the imperative subject is limited to certain

kinds of noun phrases. Such putative restrictions pose a prima facie impediment to

assimilating INPs to subjects because normal subjects are not so restricted. These

restrictions have in fact often been used to support a non-subject status for INPs. This

work agrees with Davies 1986a that these restrictions are only apparent and it illustrates this

claim with a range of data categorized in two ways. First, I present various INP


205

possibilities differentiated by the semantics of the INP: definiteness and referentiality. INPs

do not show restrictions with regard to the semantic nature of the noun phrase. Second, I

present a range of INPs corresponding to a variety of discourse roles: addressee, hearer,

participant, non-participant. Here too there appear to be no obvious restrictions on possible

INPs. The wide range of data gathered in this section illustrates that, semantically, then,

INPs are also unexceptional. The conclusions of section 4 permit an assimilation of INPs

to subjects. This is a primary result of the chapter.

4 .1 . Syntactic Status

In order to evaluate the claim that INPs are indeed subjects, it is necessary to clarify what is

meant by subject in our syntactic theory. McCloskey to appear carefully addresses this

fundamental yet elusive issue and concludes that subjecthood “is composite and

derivational. There is no term in the theory of grammar [derivatives of the Standard

Theory] which corresponds to ‘subject’ and, in contrast with theorizing of the 60’s, 70’s,

and early 80’s, there is no ‘subject position’”. Subjects have no primitive or privileged

status as in Relational Grammar, early versions of Lexical-Functional Grammar, or the

functional grammar tradition (see for example Li 1976). In GB and its descendants,

subjects are identified by a cluster of characteristic properties and a common derivational

history (Keenan 1976, McCloskey to appear) but they have no independent definition. In a

sense, then, the best that we can demonstrate is that INPs behave like subjects in their

syntax. The next section discusses these typical subject properties and shows that they hold

with respect to INPs. To the extent that we can make a claim about the subjecthood of some

NP within the theory, INPs merit the label.

McCloskey to appear, summarizing Keenan 1976, offers the overview of subject properties

for which (35) is the basis. While these phenomena are largely descriptive, and certainly


206

not definitional, they seem to hold to a large degree cross-linguistically. If it can be

demonstrated that some noun phrase type satisfies a majority of the properties in a

particular language, it is reasonable to hypothesize that the noun phrase is a subject.

(35) Subject Properties (after McCloskey to appear)

1) characteristic bearer of certain kinds of semantic roles, prototypically AGENT2) more prominent than any other argument

a. controls PRO in complements and adjuncts9b . may bind reflexive and reciprocal pronouns in other argument positions but

may not itself be bound by these argument positionsc . takes wider scope than an element in any other argument positiond . may license Negative Polarity Items in another argument position but cannot

itself be licensed by an appropriate element in one of these positions3) typically formally marked, positionally and/or morphologically4) overwhelmingly required in most situations5) almost always nominal6) derived by grammatical operations

a. Passiveb. Subject-to-Subject Raisingc. Unaccusative Advancementd. Tough Movemente. Floating Quantifiers

In what follows, I will consider each of these properties in turn. To the extent that we can

test them, they are confirmed for INPs. In the interest of keeping the reader’s attention, the

discussions are brief and largely demonstrational in nature.

1) Agentive nature. The imperative is perhaps the quintessential agent-oriented clause type.

In early works on the imperative, it was observed that certain predicates are felicitous as

imperatives while others, exemplified in (36)a, are not.

9McCloskey to appear does not include this as it is not only a property of subjects. It is, however, a typical property of at least subjects, so I include it here.


207

(36) a. be blonde/athletic/tallfimpatient, be written to/given something/hurt/chosen,

hiccup, understand, know, want

b. *Be blonde!

c. *Be praised by the President!

d . *Understand quantum mechanics!

The ungrammaticality of imperatives, exemplified by (36)b through d, was accounted for

with syntactic constraints (see Stockwell et. al. 1973 for a survey of previously proposals).

Semantically-oriented accounts of these restrictions (e.g. Downes 1977, Faarlund 1985,

Davies 1986a, Farkas 1988) rely on the observation that the unacceptable imperatives are

those in which such a command/request cannot reasonably be complied with by the person

to whom it is issued. One typically does not ask someone to carry out an action which is

beyond their control. By its very nature then, the imperative is restricted to situations which

can be willfully brought about—situations in which there is an agent.

2) Prominence. Prominence typically manifests itself through phenomena related to control

and binding. Syntactically, it is often encoded via the structural notion of c-command.

Subjects are prominent because they are known to asymmetrically c-command the rest of

the clause. Below I summarize several well-known phenomena that require prominence and

show that they appear in INPs as well.

Subjects play a central role in the interpretation of infinitival complements/adjuncts that lack

an overt subject nominal, illustrated in (37).

(37) a. Tim hopes to win.

b . We called the toll-free hotline to get more information.


208

In the cases of interest, the ‘closest’ overt subject controls the interpretation of these

infinitives. Zwicky 1988 surveys a wide range of cases involving subject control including

infinitival complements, infinitival w/i-complements, gerundive complements, and purpose

clauses. They are illustrated for declarative clauses. In each case, the missing subject is

interpreted as the higher, main clause subject.

(38) a. Marge tried to be more conventional.

b . The dog figured out how to open the door.

c . John works at being thoughtful.

d. Some people eat grapefruit to lose weight.

These control cases can be easily replicated with imperatives. The controller in the

examples below is an INP, just as the controller in the above examples was the matrix

subject. In this regard, INPs behave like subjects do.

(39) a. Everyone promise to be good!

b . Don’t you ask to miss class again!

c . Someone decide who to invite!

d. Don’t you ask what to wear again!

e. You work at being more professional before our next encounter!

f . (Don’t) Everyone applaud to please the comedian!

With respect to bound variable anaphora, INPs again behave like declarative clause

subjects. The prominence of traditional subjects in this domain is reflected by the fact that

they may bind a co-argument reflexive or reciprocal pronoun but may not conversely be

bound by these arguments. This sets up the contrasts in (40) for reflexives and (41) for


209

reciprocals. A reflexive/reciprocal in subject position may bind a co-argument, the a

examples, but a subject reciprocal/reflexive may not be bound in this position, the b cases.

(40) a. I expect him to vote for himself.10

b. *1 expect himself to vote for him.

(41) a. They corresponded with each other for years,

b. *Each other corresponded with them for years.

English reflexives and reciprocals may also be bound by other, non-subject, c-commanding

co-arguments, so the above examples don’t specifically illustrate a property exclusive to

subjects—they are only suggestive. Other anaphoric phenomena which are restricted to

subject control, beyond reflexives and reciprocals, will be more germane. One such case is

obligatory coindexing in certain idioms (Ross 1968). The meanings of the idioms in (42)

require that the pronoun be coindexed with the clause’s subject.

(42) a. Joe cranedhis

*your neck for me.

b. He heldhis*her*my

my

breath for Mary.

his himc. The diplomat kept- *your ► wits about • *you

*my *me

Looking at the domain of imperatives, INPs are prominent in the same way as the subjects

in (40) through (42) are. They may bind a reflexive or reciprocal. Such data were

10I use an ECM verb here to avoid placing an accusative pronoun in finite clause subject position, which is independently ungrammatical.


210

introduced earlier in the discussion of vocatives. (43) and (44) show that INPs may bind

reflexives and reciprocals but not be bound themselves, (45).

(43) a. Everyone behave himself.

b . Pablo and Vincent draw pictures of themselves and we’ll see if we can recognize

them!

c . Someone write himself a. reminder!

d . Don’t everyone withdraw himself from the contest!

(44) a. You talk to each other before fighting!

b . Don’t you ignore each otherl

(45) a. *Each other talk to you before fighting!

b . *Don’t each other ignore you!

(46) illustrates that the same behavior exists for INPs with the subject-oriented idioms. The

pronoun is obligatorily coindexed with the INP. INPs then have the same level of

prominence as finite clause subjects in this regard.

(46) a. Somebody crane

b . (Don’t) Everyone hold

his 1*their \ neck to see if the parade is coming! *my J

his *Sue's *my

breath in the tunnel!

his himc. Everybody keep- *Sue's ► wits about • *her

*my *me

A third way in which subjects show prominence is by preferring a wide scope

interpretation with respect to quantificational and indefinite co-arguments. The most natural

interpretation of the examples in (47) through (50) is with the object noun phrase within the


211

scope of the quantificational subject. The two interpretations that each sentence has are

given in b and c, with the favored reading in b. For example, (47)a, Everyone read a book,

most naturally means everyone read some book or other, the EVERY > A reading where

the quantified noun phrase everyone has scope over the indefinite. If the example is

passivized as in (48), promoting the object to subject, the second interpretation becomes the

dominant one.

(47) a. Everyone read a book.

b. Everyone read some book or other. EVERY > A (favored)

c . There is a book such that everyone read it. A > EVERY

(48) a. A book was read by everyone.

b . There is a book such that everyone read it. A > EVERY (favored)

c. Everyone read some book or other. EVERY > A

(49) a. Two people tasted every dish.

b . Two people are such that they tasted all the dishes. TWO > EVERY

c. Every dish is such that two people tasted it. EVERY > TWO

(50) a. None of us here saw every movie that was nominated.

b . Nobody here is such that he saw every movie nominated. NO > EVERY

c. Every movie is such that nobody here saw it. EVERY > NO

The same pattern applies in the interpretation of imperatives with quantificational INPs.

They too preferentially have scope over their co-arguments. Parallel to the above, the


2 1 2

imperative data in (5 l)a through (53)a favor the same interpretations in b.11 Again we see a

comparable level of prominence between subjects and INPs.

(51) a. Everyone read a book!

b. Everyone is to read some book or other. EVERY > A

(52) a. Two people taste every dish!

b . Two people are such that they should taste every dish. TWO > EVERY

(53) a. Nobody see every movie that was nominated!

b . Nobody is such that they are to see every nominated movie. NO > EVERY

A fourth area of prominence mentioned in McCloskev to appear is the licensing of Negative

Polarity Items (NPIs). NPIs such as anybody, ever, yet, etc. have a surface structure

constraint imposed upon them according to which they must appear within the c-command

domain of an appropriately negative element (Ladusaw 1992, Progovac 1994).

Subjects are prominent in this domain because they may license NPIs in lower positions

but subject NPIs are not licensed by lower elements. In (54)a and (55)a, the negative

subject nobody/no one licenses the NPI anything/ever. The reverse situation is not possible

^Schm erling 1982 goes so far as to assert that there is no scope interaction whatsoever among quantified noun phrases in imperatives. Tiiat is, none of the above examples, nor any like them, are ambiguous; they have only the scope possibilities permitted by the surface order of the quantifiers. Such a claim would give ultimate prominence to INPs within the domain of scope ambiguity. Unfortunately, it seems that the assertion is too strong. The imperatives in (i) are clearly ambiguous with the most natural reading being the one in which the subject is interpreted within the scope of the object, the reverse of the surface order.

(i) a. Someone answer ever)' phone cai»1b. A qualified supervisor oversee every experiment!c. Two people check each computation!


213

in (54)b and (55)b with the NPI in subject position and the negative element in a lower

position.

(54) a. Nobody should touch anything.

b . * Anybody should touch nothing.

(55) a. No one has ever talked back to him.

b . * John has ever talked back to no one.

The same pattern appears with INPs: INPs may license NPIs elsewhere but may not

themselves be NPIs (without a c-commanding negation). (56) shows that INPs license

NPIs but lower arguments do not, as attempted in (57). The data in (56) and (57) replicate

the pattern of the declarative sentences in (54) and (55).

(56) a. Nobody touch anything!

b. No one ever talk back to him!

(57) a. *Anybody touch nothingl

b . *You ever talk back to no one!

That INPs exhibit the same prominence as declarative clause subjects in these four cases

suggests that they too are subjects.

3) Formal marking. Although obvious perhaps, it is worth pointing out that INPs occur

largely in the places that one would expect imperative subjects to appear. English is rather

strict in its word order and its subjects can usually be identified by their position: on the left

of a sentence. This is one of the locations of INPs, as illustrated in the overview of

imperative subject position in the introduction. Examples of INPs preceding the auxiliaries

do and don’t are repeated in (58) and (59). INPs occur where one finds subject NPs.


214

(58) a. SOMEone do answer the phone!


(59) a. Those ofyou with luggage don’t leave it unattended!

b. (Girls go into the hall,) BOYS don’t move!

The post-do(n’t) INP illustrated in (60) and (61) is also a subject position not without

precedent. This inverted word order is well-known from interrogatives, negative

preposing, and other phenomena.



(61) a. Don’t anyone touch my stuff!

b . Don’t you misbehave while we’re gone!

In addition to appearing in the expected places for subjects, INPs do not appear in

unexpected locations. The positions surveyed below are ones which subjects often occupy

in other languages. (62) illustrates that they do not occur below the auxiliaries have and be.

(62) a. *Be everybody waiting for me at noon!

b . *Be someone seated next to the guest of honor!

c . *Have you read that before the exam!

They are also ungrammatical in lower positions among the main verb’s internal arguments,

(63), and right-peripherally, (64).

(63) a. *Give someone that book to me!

b . *Give that book someone to me!


215

(64) a. *Vote for that villain nobodyl12

b . *Be late nobody!

Summarily, INPs are positionally like ordinary subjects in appearing in the same string-

wise positions in a sentence.

4) Obligatory presence. In many languages, subjects have the perhaps surprising property

of being required in all clauses. As McCloskey to appear points out, this is obviously not

the case for other argument positions such as direct objects. It also does not even hold of

other typical clausal elements such as tense. It is clear that INPs are not required in

imperatives and so they are unlike subjects in this respect. Cross-linguistically, subjects are

not usually obligatory in imperatives. INPs are unusual as subjects but not unusual as

imperative subjects.

5) Nominal status. It is overwhelmingly true that subjects are largely nominal in nature

while no other position—complement, adjunct, modifier—seems so restricted. With regard

to imperatives, it seems to be the case that INPs are exclusively nominal. Although clauses

and prepositional phrases are arguably permitted as subjects of finite clauses, they do not

seem to be allowed in imperatives. Imperatives and finite clauses differ in this regard,

yielding the contrasts in (65) and (66). The pairs of examples illustrate that prepositional

phrases and clauses, respectively, are acceptable in subject position of finite clauses, the a

examples, but not as INPs, the b examples, which are ungrammatical.

I2Nobody is used here to rule out the possibility that this is a vocative, which is permitted clause finally.


216

PREPOSITIONAL PHRASES

(65) a. In the attic is a safe place to hide.

b . *In the attic be a safe place to hide!

c . Around the comer appeared a policeman.

d. *(Don’t) Around the comer appear a policeman!

CLAUSES

(66) a. For you to win is impossible.

b . *For you to win be impossible!

c. That the mess will be cleaned up is certain.

d . *That the mess will be cleaned up be certain!

The data suggest that INPs are restricted to being NPs, a result compatible with

subjecthood. This similarity between INPs and subjects cannot be given too much weight,

though, since it has already been claimed that subjects are typically agentive. These two

properties, agentivity and nominal status, would seem to be different sides of the same

coin. The ungrammaticality of the imperatives in (65) and (66) is unsurprising if agentivity

of the INP is a requirement.

6) Derived position. Subjects are unique in being implicated in a large number of syntactic

rules and grammatical relation changing operations. The five phenomena surveyed below—

passive, subject-to-subject raising, unaccusative advancement, tough movement, and

floating quantifiers—are representative. They all derive a subject from a lower argument

position, under at least some assumptions. In each case, declaratives and imperatives alike

undergo the process, suggesting that they share the same syntax. Some of the observations

and examples are taken from Schmerling 1977 and Zwicky 1988.


217

A standard analysis of passive clauses is that the subject is base-generated in object position

but becomes the subject to be assigned Case (Baker, Johnson, and Roberts 1989) or to

check features. The examples in (67) illustrate passive in declarative clauses. The base

position of the object is indicated by a trace. In (68), parallel examples with INPs show the

same behavior. INPs are targeted by the same syntactic process that derives subjects in

declarative clauses.

(67) a. Everyone was examined t by a doctor.

b . The occupants were advised t that the building was unsafe.

c. Farrah got introduced t o the movie mogul.

(68) a. Everyone be examined t by a doctor!

b. Occupants be advised t that this building is unsafe!

c. One of you get introduced t to the movie mogul!

d. Don’t you be fooled t\

e. Don’t anyone be caught t speeding!

Subject-to-Subject Raising (see Soames and Perlmutter 1979) predicates are those that take

an infinitival complement and do not assign a thematic role to an external argument. In

order to provide a main clause subject, a subject is raised from within the lower clause and

receives Case in the higher subject position. (69) gives examples for declarative clauses.

(69) a. The janitor just happened t to be here when I walked in.

b . John appeared t o be going through the private files.

c. We must stop t writing when the bell rings.

Parallel imperative examples are in (70). If these sentences have the same derivations as the

ones above then the INPs are acting syntactically like subjects.


218

(70) a. Someone just happen t to be here when he walks in!

b . John appear t to be going through the files when the boss comes in and Bill

record his reaction!

c . (Don’t) Everyone stop t writing when the bell rings!

A third source for derived subjects is unaccusative predicates. Unaccusativity, discussed

first in Perlmutter 1978 and later in detail in Levin and Rappaport Hovav 1995, accounts

for the alternation in (71).

(71) a. John melted the ice sculpture,

b. The ice sculpture melted t.

In this situation, the verb does not assign an external argument role and the subject is raised

from the direct object position. The unaccusative verbs in (72), which include a verb of

motion in b and a psych verb in c, are taken from Perlmutter and Postal 1984, McCloskey

1993 and Levin and Rappaport Hovav 1995.

(72) a. Nobody will emerge t dirty from that shower.

b . Someone is coming t.

c. Cindy amused the children t while Bob surprised the guests t.

If such subjects are in fact derived, they provide an additional diagnostic for subjecthood.

(73) illustrates that INPs are unexceptional with unaccusative verbs. Again they behave like

other subjects.


219

(73) a. Nobody emerge t dirty from that shower!

b. Someone comer quickly!

c. Cindy amuse the children t and Bob surprise the guests r!

d. Don’t anybody emerge r dirty!

e. Don’t all the children come t at the same time!

A final process that derives subjects is Tough Movement or Non-subject Raising (Soames

and Perlmutter 1979). As in Subject-to-Subject Raising, a subject is derived by movement

of a nominal from within an infinitival complement:

(74) a. Everybody should be easy for us to spot t.

b . Tourists must be tough to deal with t.

c. Noam was pleasant to talk to t.

As in the above cases, INPs may appear in the derived subject position of these predicates.

(75) a. (Don’t) Everybody be easy for us to spot r!

b . Tourists be tough to deal with t when these people bargain!

c . Someone be pleasant to talk to r!

These four processes, widely accepted as involving a derived subject position, behave

identically with respect to finite clause subjects and INPs. Again it is suggested that the two

are non-distinct.

A recent innovation in clausal organization is the VP-Internal Subject Hypothesis (ISH).

Under the ISH, all subjects, even those that do not undergo one of the above operations,

are derived in the sense that they originate in a position other than the one which they

occupy at surface structure. This additional assumption makes further probing of subjects

possible. If INPs are subjects, we expect that they too will replicate the arguments that have


220

been offered for the ISH and a lower origin position (see McCloskey to appear for

summary and discussion). I will confirm this prediction for one notable case: floating

quantifiers.13

Sportiche’s (1988) analysis of floating quantifiers, which relies on the ISH, is presented in

chapter 1. Quantifiers are stranded in various positions as the subject moves from the VP-

intemal position to its surface position. If INPs are subjects syntactically then they should

support stranded quantifiers like normal subjects. Schmerling 1977 was the first to

explicitly observe that floated quantifiers do appear in imperatives:

(76) a. (Don’t) The twins both be here for the picture!

b . Rhett, Scarlet, and Lassie all get ready for the next scene!

c. MY children all come here!

d . You be both waiting for me promptly at 3!

e. Don’t my students all take off at the bell!

These facts are surprising if INPs are not subjects as it is generally only subjects that can

support floated quantifiers (but see Maling 1976). The argument illustrates that INPs not

only undergo the typical syntactic derivation of subjects but that they also have the same

origin site. Recognizing that INPs have this particular syntactic behavior of subjects but are

not subjects would be completely anomalous in the assumed syntactic framework. Floating

quantifiers thus provide strong support for the unexceptional status of INPs as subjects.

In summary, there are numerous processes in English that derive subjects from various

base positions. All of these processes treat INPs like other subjects in that INPs may be

13McCloskey’s other arguments less-interestingly apply.


221

targeted. Each individual observation is not particularly compelling; however, taken

together they provide a substantial body of evidence supporting the subject status of INPs.

The minimal analysis is that INPs are syntactically subjects and the derivations of the

imperative examples above parallel those of the declarative counterparts in the relevant

ways. If the Internal Subject Hypothesis is correct, INPs also share this extended

derivational history. Overwhelmingly so, processes and generalizations that apply to

subjects apply to INPs also. I conclude that, syntactically, INPs are subjects.

Even if we accept that INPs are syntactically subjects, there remains the possibility that they

are yet not in the canonical subject position, specifier of IP. Two other researchers explore

this option, going in opposite directions. Beukema and Coopmans 1989 argues, based on a

restrictive semantic characterization of INPs that I present below, that all imperative

subjects have vacated the specifier of IP position by LF and have moved into an A

position. The specifier of IP is always occupied by an A-bound empty category at LF. In

other words, imperative subjects obligatorily undergo an additional movement not required

of ordinary subjects. Platzack and Rosengren 1994 takes the opposite tack. Adopting the

VP-Internal Subject Hypothesis, the work argues that imperative subjects do not move at

all, remaining in the VP-intemal position. I argue against the former position in the

following section. The latter analysis is considered, and rejected, in chapter 5. Eliminating

these two possibilities leaves unchallenged the option that INPs are subjects in spec,IP, the

expected position.

4.1.1. Beukema and Coopmans 1989

Beukema and Coopmans 1989, henceforth B&C, maintains that INPs in English are either

inherently quantificational, can be given ‘some sort of quantificational interpretation’, or are

identical to you.


222

(77) Beukema and Coopmans ’ (1989) Characterization o f INPs

imperative noun phrases are either 1) quantificational or 2) the pronoun you

If one accepts this generalization, then it suggests that there is some connection between

quantificational NPs and the imperative subject position. An assumed property of

quantified expressions is that they are required to undergo Quantifier Raising (QR) at LF,

an A movement that raises and adjoins them to a dominating maximal projection for scope-

taking purposes (May 1985). B&C capitalize on this by requiring that all INPs undergo a

QR-like movement. Quantificational NPs adjoin to IP at LF as part of QR. Non-

quantificational INPs, like you, are topicalized, a process that Baltin 1982 and Lasnik and

Saito 1992 analyze as also being adjunction to IP. All imperative subjects are thus targeted

for movement to an IP-adjoined position. To accommodate zero subject imperatives, B&C

claims that there is a null topic operator that is also topicalized from the subject position.

The null topic operator receives its referent from the discourse. The syntactic generalization

that B&C asserts regarding INPs is that they are all in an A position by LF and bind an

empty category in the specifier of IP.

(78) Beukema and Coopmans ’ (1989) Analysis o f INPs

At LF, all INPs are adjoined to IP, either by QR or topicalization, and bind

an A empty category in the subject position

Examples repeated in (1) and (2) have LF structures schematized by (79) under the

analysis.

(1) a. Tom take out the trash!

b. Everyone write to Congress about this atrocity!


223



(79) IP (LF)

DP IP

you t rI Op J . I VPeveryone •

I offer three arguments against this approach. The first two, immediately below, bring into

question the obligatory A movement hypothesized in (78). The third, developed in section

4.2, closely addresses the characterization of INPs in (77), showing that it too is

inadequate.

topicalized in all subjectless imperatives. She observes that a null subject cannot be

interpreted as third person, even when an appropriate third person discourse element is

present:

(80) a. Everybody take out their books! After that, write down their names!

b . Someone go up to the board and do this problem! Afterwards, return to his seat!

c . The boy in the comer stand up! Now recite his poem!

(80)a cannot be understood in such a way that everyone writes down their own names

rather than those of some unidentified group of people. The subject in the second

imperative must be construed as you rather than everybody. Such an interpretation would

be expected if there were a discourse-controlled topic operator since the previous subject

everybody is highly salient. Similar interpretations are also impossible in (80)b and c where

Henry 1995 raises an objection to the null discourse-bound operator that is claimed to be


224

the imperative subject is not quantificational but also cannot serve as an antecedent for the

topic operator in the subjectless imperative.

A second argument against B&C’s proposal in (78) is that it would seem to rule out overt

topicalization in imperatives. Topicalization of multiple phrases is generally prohibited,

illustrated by the ungrammaticality of the examples in (81).

(81) a. *On the table, this book, John put.

b . *To the Salvation Army, the old National Geographies, you should donate.

If there is always a syntactic, IP-adjoined topic associated with non-quantificational

subjects, a second topic should be uniformly impossible under B&C’s account. In chapter

5 ,1 demonstrate that topicalization is available in imperatives. Examples are given in (82).

If there is already a topicalized subject or null operator, these examples of imperative

topicalization should be ungrammatical, parallel to the robustly ungrammatical data in (81).

That they are not suggests that there is no other topicalized element, zero or overt.

(82) a. The teacher, you be more respectful of!

b. The tie, give to Bob; the aftershave, give to Don!

c. To yourself be true!

These two arguments indicate that the topicalization analysis in (78) is fundamentally

flawed. I conclude that INPs do not obligatorily undergo A movement to bind an empty

category in subject position. Summarizing the results, there seems to be significant

evidence against and none for the claim that INPs are syntactically exceptional and do not

overtly occupy the subject position.


225

In the following section, I turn to putative formal restrictions, such as B&C’s (77), and

show that any analysis that categorically limits the semantic properties of INPs is

undesirable since they are not so restricted. This result has two consequences. First, it

provides the third argument against B&C’s proposal: the characterization in (77) is not

correct. Second, and more importantly, it demonstrates that there is, in general, nothing

semantically idiosyncratic about INPs. Their behavior will be seen to follow from the claim

that they are the subject of imperatives.

(77) Beukema and Coopmans’ (1989) Characterization o f INPs

imperative noun phrases are either 1) quantificational or 2) the pronoun you

4.2 . Semantic Restrictions

Thus far there has not been compelling syntactic evidence for differentiating INPs from

ordinary subjects. The most robust argument against taking INPs to be subjects has

traditionally been that there are supposed, strong semantic idiosyncrasies governing their

formal realization (Schmerling 1982 and many others). Certain types of noun phrases—

names, defmites, and other referential expressions for example—are supposedly excluded.

I attempt to demonstrate that such semantic restrictions on the form of INPs are

unnecessary, in essence agreeing with Davies 1986a. In what follows, I survey the full

range of INPs and show that there are, in principle, few restrictions. In the first subsection

4.2.1, the focus is on the characterization in (77), although the point to be made is more

general: imperative subjects are not restricted in their form; they are not limited to being

quantificational or precisely the second person pronoun. Section 4.2.2 considers an

alternative analysis which restricts INPs according to the discourse role they are permitted

to have (Downing 1969). Downing argues that the imperative subject must be a subset of


226

the addressee. The section shows that such a description is also too narrow. There are also

no firm restrictions on the discourse referent of an INP. The surveys of INP options

indicate that they are semantically unexceptional as well. In assimilating INPs to subjects, I

agree with Stockwell et. al. (1973:642) when they write: “It might be convenient if in fact it

turned out that subjects of imperatives could be freely generated. There is apparently no

natural way of constraining the subjects of topmost imperatives”. This is the null

hypothesis and, under this view, we do not reasonably expect restrictions on INPs, such as

(77). As subjects, we expect INPs to show the full range of formal possibilities. At the

same time, the observations that I argue against are in some broad sense observationally

true. Quantificational and second person subjects are more common and less marked in

most contexts and in most cases the imperative subject is clearly related to the addressee.

These strong preferences require an explanation. Section 4.2.3 turns to this task. It

proposes that the apparent restrictions on imperative subjects derives from the way an

imperative is used and the role that the subject plays in its semantics.

4.2.1. Restrictions in Form

What guided B&C’s proposal is that the most typical INPs are indeed you and

quantificational elements as in (83) and (84), respectively.

YOU

(83) a. You be quiet!

b . You mind your own business, and leave this to me!

c. You show me what to do!


227

QUANTIFIED N PS

(84) a. All the children in the front row be quiet!

b. No one stop me! (Dickens, D 423)

c . Everybody hurry up!

B&C took these to be essentially the only possibilities. Such a claim is too narrow,

however. I will present data to show that almost any kind of noun phrase seems

appropriate as an INP in the right circumstances. I survey instances of indefinites (85),

partitives (86), definites (87), bare plurals (88), and even names (89), all of which are

problematic for B&C’s characterization. Many of the examples are taken from other

linguistic sources: Jespersen 1961, Visser 1963a, Downing 1969, Schmerling 1982,

Stockwell et. al. 1973, Quirk et. al. 1985, and Davies 1986a.

The indefinite INPs in (85) under some analyses (e.g. Russell (see Heim 1988 for an

overview)) would also be considered quantified noun phrases since they do not necessarily

refer. They are thus reasonably expected to occur alongside (84) and are, in fact, quite

common.

INDEFINITES

(85) a. ‘Somebody call my wife! ’ (Shakespeare, Merry Wives of Windsor IV.2.121)

b. ‘Someone arrange my shawl!’ (Meredith, R 353)

c . Whoever saw the incident come forward please!

The partitive examples in (86) might also somehow be assimilated to B&C’s

characterization, as instances of you. Second person partitives, which the examples

represent, are also prevalent INPs because they help to make explicit the identity of the

subject. They perhaps sound somewhat better than other examples because of this


2 2 8

connection to the addressee. They may be either quantificational as in (86)a and b,

indefinite, in c and d, or even definite as in e and f.

PARTITIVES

(86) a. A few o f you stay behind to help clean up!

b . Every one o f you be responsible for his own materials!

c. ‘Some three or four o f you go give him courteous conduct to this place!’

(Shakespeare, Merchant o f Venice IV. 1.147)

d. *One o f you call him forth!’ (Marlowe, Faust 1108)

e. 'The rest o f you go aloft and get the chest!’ (Stevenson, Treasure Island 38)

f . The two o f you stop fighting and get to work!

The data thus far might be made to fall within the bounds of a suitably-revised version of

B&C’s restriction on INPs. The following cases, however, are more clearly outside the

domain. The definites in (87) are quantificational under no accounts and thus are not

included in the generalization that only quantified noun phrases may be INPs. The

examples indicate that INPs may pick out a specific member of the audience.

DEFINITES

(87) a. The oldest o f the girls in this group sing a folk song!

b . The boy in the comer stand up!

c. The man with the list come here!

d . Those near the front wait until the others have left!

Bare plurals behave similarly. They are acceptable as INPs, in (88), particularly in

instruction type imperatives. Carlson 1977 argues that such phrases are non-

quantificational. Under his analysis, they refer to the kind NP.


229

BARE PLURALS

(88) a. Truck drivers keep right on grades!

b . Housewives watch out for the daily specials!

c . Visitors use the other entrance please!

d. People with questions stay behind afterwards!

Lastly, (89) indicates that even the quintessential referring expressions, names, are

acceptable as INPs under the right conditions. Although (89)a in isolation is

ungrammatical, it is acceptable in the context given in (89)b. The crucial observation is that

names are allowed as INPs and they are in no way quantificational or equivalent to the

pronoun you.

NAMES

(89) a. *Mary stand by the door!

b . Mary stand by the door, John scatter the files, and I’ll watch the front!

c. Joyce hold one end of the rope and Renee hold the other!

d . Rob take the box and Dave bring the suitcase!

The almost complete lack of actual restrictions on INPs14 demonstrated above permits us to

have the unencumbered analysis of INPs that was hinted at by Stockwell et al. 1973. They

14Pronouns, including first person pronouns in (i)a and third person pronouns in (i)b, are traditionally claimed to be categorically ungrammatical as INPs (Stockwell et. al 1973, Davies 1986a).

(i) a. *Me/UWe/Us stand up!b. * Him/He try to run faster!


230

are subjects in an ordinary sense. We do not need to impose any kind of unusual

constraints on the form of INPs. In the following subsection, I explore the possible

discourse referents of imperative subjects, showing that these too are essentially

unrestricted.

4.2.2. Discourse Roles

B&C is unique in attempting a syntactic analysis of the apparent restrictions on INPs. Most

other researchers who have approached the problem seriously have concluded that the

range of imperatives subjects is semantically determined based at least in part on the

meaning and use of the imperative itself (Downing 1969, Schmerling 1982, Davies 1986a).

Such investigators are in general agreement regarding the approach to characterizing the

imperative subject: the potential referent of an INP can be defined within the context of the

discourse roles from section 2. There is disagreement, however, as to which roles are

available. The structure of discourse roles is repeated in (3).

Such examples do seem truly terrible but it is not clear whether their ungrammatically is due the pronominal status of the INP, its referent, or something else. The observation that the pronoun you is regularly permitted suggests that it is not the syntactic status of pronouns that is problematic in (i). On an informal level, it is the very use of such examples that is dubious. It is unlikely that one would issue an imperative to oneself, making first person pronoun subjects all but useless. Third person pronouns in b might be reasonably ruled out because the pronoun does not supply enough information to identify the subject In support of this is that the examples improve if descriptive content is added to the pronoun, in(ii).

(ii) a. IHe who stole the eraser put it back immediately!b. IShe who tracked in mud take her shoes off this instant!

If this is so, it still not clear why, when the pronoun can unambiguously identify a referent the examples are still not very natural:

(iii) a. *You make the dinner and John do the washing up! No? All right then, he cook and you washup!

b. ???This next chess game, you be white, / be black, and he be the timer!


231

(3) Discourse Role Relations

PARTICIPANTS

SPEAKER HEARERS

ADDRESSEE SIDE-PARTICIPANTS

Most precise is Downing 1969, working within the performative hypothesis, which argues

that the imperative subject is restricted to standing in a subset relation to the addressee. That

is, the imperative subject may be the addressee itself or a subset of it if the addressee is a

group.15

(90) Downing's (1969) Characterization of INPs

the subject of an imperative must stand in a subset relation to the addressee

(90) can be understood as saying that the purpose of the imperative subject is to identify as

the agent of the proposed action someone other than the canonical, default addressee (this

idea is developed further in section 4.2.2). This principle is at play in the examples repeated

in (1). In (l)a, the overt subject you, although still the addressee, serves to emphasize or

contrast the addressee with respect to some other possibility (see Davies 1986a for

discussion of the use of you). In (l)b, c, and d, the subject functions to target a particular

subset of the addressee to carry out the directive and make this entity more salient. These

subject imperatives contrast with their subjectless counterparts in (2) most notably in that

I5Zhang 1990:25 holds a similar position to Downing’s, assuming that INPs are fundamentally “addressee- limited subjects”. He does recognize exceptions, which he labels “non-addressee-limited”, but states that the only possibility is a coordination of you with a third person noun phrase. Such conjunctions are still grammatically second person in nature.Downes 1977 perhaps too shares Downing’s position in stating “the fact of the matter is that the subject NP must refer to the hearer of the utterance, the addressee. Any NP whatever that does not exclude this interpretation can be the subject of an imperative.” Since hearer and addressee are distinct in the above system, it is unclear which one Downes intends.


232

the agent of the imperative action is potentially different. The interpretation of the examples

in (2) is more carefully addressed in section 5.


b. Everyone write to Congress about this atrocity!

c . Someone get me an aspirin!

d. Those in front step away from the barricade!





While Downing’s restriction on INPs holds for all the examples presented thus far, Davies

1986a: 140 argues that it is, upon closer consideration, too narrow given examples like the

ones in (91). The examples all have in common that the INP consists of the pronoun you

conjoined with another noun phrase. It was seen earlier in connection with vocatives that

such coordinations cannot pick out the addressee. The INPs in the examples do not refer to

a subset of the addressee, but a superset, although they do contain the addressee. The data

are thus counterexamples to Downing’s restriction.

(91) a. You and your men be on guard for anything suspicious!

b . You and William do the cooking and I’ll provide the wine!

c. You and them make a deal! I’m out of this.

The second person plural pronoun, you (all), argues against the proposal in a similar

manner. Section 2 determined that you (all) (pi.) identifies the addressee plus optional

others, either additional addressees or third persons. The referent of the plural pronoun is


233

then also not necessarily a subset of the addressee if its referent is the addressee plus other

people. It may nevertheless serve as an INP. (92) illustrates that the plural pronoun is an

acceptable with this interpretation.

(92) a. Your family is going camping for the week?!

Well, you all have a good time! Don’t you all spend the whole time fighting!

b . You and your wife are vacationing in Cambodia?!

You all be careful of those land mines!

Given such examples, it cannot be the case that INPs are limited to addressee reference.

Looking at the referent of these examples, we see that the INPs need not even refer to

hearers or audience members. It is possible to interpret the above examples in (91) and (92)

in such a way that the people referred to by the INP besides the addressee you (sg.) are not

in earshot of the issued imperative. They are not part of the audience in the sense defined

above. Bolinger 1967 makes this point with the example in (93)a from Poutsma 1928.

(93) a. You get the paper and pencil and the catalogue, and George write down, and I’ll

do the work!

b . Go call him, will you, and tell him we need him and what he’s supposed to do.


234

It is not necessary that George be within earshot of the utterance for the imperative to be

issued. The speaker might continue with (93)b. Further examples in which the INP

completely excludes the addressee and audience are in (94).16

(94) a. You go for help and the baby stay with me!

b . You make the dinner and John do the washing up! No? All right then, John cook

and you wash up.

c. YOUR soldiers build the bridge, General Lee!

d . Your guards be the diversion while we sneak in!

e. Those children o f yours keep out of my garden, or I’ll set the dog on them!

In (94)a and b, the first subject is the addressee so that the second subject cannot be if we

assume no shift in address, as seems possible. In (94)c, d, and e, the INP must be disjoint

from the addressee since it contains a possessive pronoun referring to the addressee itself.

I conclude that the referent of an INP is not limited to any specific discourse entity. In

particular, it is not restricted to being a subset of the addressee. This section has

demonstrated that imperative subjects are not syntactically or discoursally restricted:

(95) Imperative Subject

An imperative subject is unrestricted in form and reference

16Schmerling 1982 makes the claim that INPs are restricted to audience reference, a somewhat more permissive formulation than Downing’s. This too would seem too narrow given these examples. Schmerling’s term “audience”, however, is perhaps best paraphrased as “intended hearers”. She remarks in a footnote that her notion of “audience” must be interpreted as a subset o f the domain of discourse not necessarily limited to those who are physically present at the time of the utterance. The term has to do instead with the intentions of the speaker. She correctly observes that the imperative in (i) may be uttered with no apparent audience in sight (or earshot).

(i) Somebody help me!


235

This characterization of the subject is distinct from that of the vocative in (17):

(17) Vocative

A vocative necessarily refers exclusively to the addressee

Vocatives are truly addressee limited, while subjects are neither discourse-referent nor

formally restricted. This is source of the contrast that was illustrated in section 3.1.2,

repeated here.


b . *Your children, can the play area be kept clean?

c . * Nobody, that man just rode off with my bicycle!

(25) a. You and Fred stay out of the construction zone!

b . Your men guard the front while we creep round to the back!

c. Nobody ride off with my bicycle!

Despite this conclusion, there are strong intuitions that both of the above proposals have

some substance and represent typical states of affairs: 1) certain subjects, such as

quantificational NPs, are more prevalent and more easily accepted and 2) a connection

exists between the imperative subject and the addressee. These patterns, even if they are

only tendencies, should be accounted for. In the next subsection, I offer a new perspective

on the imperative subject that attempts to account for these observations. While agreeing

with Downing that there is a relationship between the addressee and the imperative subject,

I suggest that it is of a different nature than he envisioned. It is tied to the semantics of

imperatives and how they are used. The following section thus offers an approach to

understanding the distributional patterns of imperative subjects.


236

4.2.3. The Addressee and the Imperative Subject

Downing’s (1969) analysis of the referent of an imperative subject as necessarily being a

subset of the set of addressee(s) was highly successful in its empirical coverage, despite the

counterexamples provided in the previous subsection. Its plausibility suggests that

restrictions on the imperative subject are indeed related to the addressee and in this

subsection I profitably continue this line of investigation. I develop a new characterization

of the link between the imperative subject and the addressee which subsumes Downing’s

restriction and at the same time accounts for the newly introduced data. The range of

imperative subjects presented earlier and the examples that were problematic for Downing’s

formulation share an interesting property: even when the referent of the subject is not

identical to the addressee, the addressee is still in a control relationship over the imperative

subject. This restriction reasonably follows from what an imperative utterance is designed

to accomplish and how it is used. In developing this approach I will rely on several

concepts developed in Farkas 1988 to help formalize the discussion.

In the data presented earlier in which the imperative subject is not precisely the addressee,

what makes all the examples acceptable is that there is a specific relationship between the

addressee and the subject. I will call it a CONTROL RELATIONSHIP. As the name suggests,

in a control relationship, one member has some kind of control—authority, power, or

influence—over another, (96). Before illustrating the application of this relationship to

imperative subjects, I briefly explore the kind of associations that it is designed to capture.

(96) Control Relationship

x is in a control relationship with y if x has potential control over y in some

domain z (where z may range over social, military, political, economic,

discourse, or other situations)


237

The most transparent control relationship will be one in which there is a defined hierarchical

arrangement between two entities. One member will be in a formally inferior/superior

position with respect to the other according to some criteria. This will arise, for example, in

military, security, or punitive environments. Rules, conventions, laws, or other parameters

in these settings will create a hierarchy of control. Social structures that exist between

adults and children or various levels of workers in a business setting are also instances of

hierarchical control relationships. Control relationships do not require that there be an

inherent hierarchy between the participants, however. Relational affiliations also typically

afford one entity the opportunity to influence or control another. Such cases include family,

social, and peer groups. In these settings individual members can potentially influence

others because of shared physical situations, conventions, or goals. Most control

relationships that will be seen below are not purely hierarchical or relational but illustrate a

complex combination of these modes. I want to suggest that a control relationship of the

addressee over the imperative subject is a necessary ingredient for the felicitous use of an

imperative subject:

(97) Relationship between Addressee and Imperative Subject

The addressee must be in a control relationship over the referent of the

imperative subject

The notion of control relationship described above seems closely aligned with the

A-relation developed in Farkas 1988.17 In that work “A(x,y) names the relation which

17The context of the proposal is a discussion of the responsibility relation, to be introduced below, which relates an individual, the initiator, to a situation which s/he is seen as bringing about. Farkas 1988 suggests that, if the initiator is distinct from the actual intentional agent of the situation being brought about, then the initiator is in an A-relation with the agent and is seen as determining his/her actions.


238

exists between two individuals x and y when x determines the actions of y (Farkas

1988:37)”. I will henceforth use this as a shorthand and restate (97) as (98).

(98) Relationship between Addressee and Imperative Subject

Mflddressee, referent o f the imperative subject)

With respect to the data introduced so far, there are three cases to consider in verifying the

above claim: 1) the referent of the imperative subject is the addressee, 2) the referent of the

imperative subject is a subset of the addressee (Downing’s characterization in (90)), and 3)

the counterexamples to Downing’s analysis in which the referent of the imperative subject

is a superset of the addressee or non-intersecting with the addressee.

If the imperative subject corresponds precisely to the addressee, a trivial control

relationship obtains between the addressee and the imperative subject since there is identity.

This is clearly the most direct and strict control relationship that can obtain.

The requirement in (97) covers Downing’s analysis and predicts the acceptability of

imperative subjects when they are a subset of the addressee. In this situation, the addressee

is a plural group and the subject is itself one of this group. In such cases, the imperative

subject is in effect an addressee, although not the addressee. There will be a control

relationship in this scenario that is nearly as direct as above and the addressee as a group,

or as individuals, will have power or influence over one of its members. To the extent that

this holds, the imperative subject will be felicitous. Most of the examples above fit this

scenario.

The third case—where the imperative subject is not a subset of the addressee group—is the

most crucial, as it is where Downing’s analysis was too restrictive. The new

characterization does not categorically restrict the discourse role of the imperative subject;


239

however, it does constrain it to the extent that the addressee is able to have some control

over the imperative subject. The expectation is that the above counterexamples all contain

some type of control relationship and are not completely free as was implicitly suggested.

This is in fact borne out and I verify it for the particular examples, repeated below.

The examples in (91) and (92) are all alike in that the imperative subject corresponds to a

group including the addressee and makes explicit reference to the addressee via the pronoun

you. In such cases, the addressee is most naturally in a control relationship with the

subject, being an integral part of it. Given (97), these examples are not greatly different

from examples in which the imperative subject is a subset of the addressee rather than a

superset, as long as the addressee is a part of the imperative subject and reasonably related

to the remainder of the subject’s referent. The presence of the necessary control relationship

is clear in (91)a, in which, given the context, there exists a hierarchical relationship

between the addressee, you, and the remainder of the imperative subject, the addressee’s

men. It also obtains in the remainder of the examples.

(91) a. You and your men be on guard for anything suspicious!

b . You and William do the cooking and I’ll provide the wine!

c. You and them make a deal! I’m out of this.

(92) a. Your family is going camping for the week?!

Well, you all have a good time!

b . You and your wife are vacationing in Cambodia?!

You all be careful of those land mines!

When the addressee is not part of the imperative subject, as in the examples in (94), the

presence of and need for a control relationship can still easily be seen. In these examples,


240

there is clearly relational influence, as in (94)a and b, a hierarchical relationship, as in (94)c

and d, or both, in (94)e. The context makes it evident in these examples that the addressee

has some control in the situation—as a caretaker in (94)a and e, a dominant participant and

decision maker in (94)b, or a hierarchical superior in (94)c and d.

(94) a. You go for help and the baby stay with me!

b. You make the dinner and John do the washing up! No? All right then, John cook

and you wash up.

c. YOUR soldiers build the bridge, General Lee!

d. Your guards be the diversion while we sneak in!

e. Those children o f yours keep out of my garden or I’ll set the dog on them!

In addition to covering the above data, the analysis in (97) makes a prediction regarding the

possible cooccurrence of a vocative, which explicitly encodes the addressee, and an

imperative subject. For the most part, examples with both, in (99), do not sound

particularly felicitous unless the two are coreferential as in (99)d and e.

(99) a. #John, someone wash the dishes!

b. #You in the red chair, everyone get ready to leave!

c. #Twins, nobody touch my stuff!

d. Jack, you clean up this mess and stop blaming your sister!

e. Mia and Betsy, both of you help your mother with the preparations!

Upon hearing such examples, there is a effort to interpret them so that they make sense.

One means of accomplishing this is by forcing coreference between the vocative and

subject. When this is done in (99)a, the imperative becomes a roundabout request with the

speaker using someone as an indirect reference to John himself. An alternative is to


241

understand the examples with an implicit causative have: (99)a wants to mean something

like John, have someone wash the dishes! The latter strategy in particular suggests that

what is missing is the necessary, more explicit control relationship. When one can be

introduced, similar examples improve, (100). (100)c and d are from Hamblin 1987:52-53.

(100) a. Maitred’, someone seat these guests!

b . Counselors, everyone be packed up and ready to go in half an hour!

c. B Company deploy on the escarpment, lieutenant!

d. Nobody leave the hall, sergeant.

Of the characterizations of the imperative subject presented here, only the restriction in (97)

can account for the limited cooccurrence of a vocative and an overt imperative subject.

Further support for the claim in (97) comes from the impossibility of non-NP subjects seen

in section 4.1. The data are repeated below:

(101) a. *In the attic be a safe place to hide!

b . *(Don’t) Around the comer appear a policeman!

(102) a. *For you to win be impossible!

b . *That the mess will be cleaned up be certain!

The data are ungrammatical according to the restriction because the subject is not the type of

entity that can be controlled. Prepositional phrases and clauses realize locations, adverbial

elements, situations, or propositions. These are all elements that are not controllable. In

order to be compatible with (97), imperative subjects will have to be animate, sentient

beings, hence noun phrases.


242

The proposed connection between the addressee and the imperative subject derives further

support from its link to imperative semantics. It is a natural consequence of the meaning of

an imperative. While an analysis of imperative semantics is outside the scope of this work,

several recent proposals from other researchers converge upon the necessary ingredients to

derive the restriction in (97). Given a fairly conventional account of the core meaning of an

imperative, the need for a control relationship between the addressee and the imperative

subject reasonably follows. The central claim is that the goal in issuing an imperative is to

suggest that the addressee do something. Two clear and general proposals representing this

fundamental line are Sadock 1974 and Leech 1983 (see also Schmerling 1982, Huntley

1984, Wilson and Sperber 1988, and particularly Davies 1986a for a survey). The former

suggests that “what all of the senses of an imperative-form. . . have in common is that

they indicate that the speaker is prescribing with his sentence some future course of action

for the addressee (Sadock 1974)”. Leech comes to a similar analysis of imperatives. In

describing their role, he states “they all, in some respect or another, present the

prepositional content [of the imperative] as a candidate for fulfillment by h (hearer) (Leech

1983:117)”. A core claim of both analyses is that imperatives prescribe some course of

action for the addressee. This is equally central, although less perspicuous, in Hamblin’s

(1987) recent analysis of imperative semantics. The crux of Hamblin 1987 is the

Addressee-Action-Reduction Principle (AARP) which states that the meaning of an

imperative can be elucidated in plain predicate imperatives directed to the addressee(s).

Hamblin explains the principle: to understand the meaning of an imperative “try spelling it

out in terms of what actions it enjoins of its addressee(s) (Hamblin 1987:58)”. Even when

third-person subjects different from the addressee are present, Hamblin makes it clear that

we are to view imperatives in terms of what they require of their addressee(s) rather than

what they require of their subjects. Like Sadock and Leech’s analyses, then, a crucial part


243

of Hamblin’s account is that imperatives can be understood in light of what they require of

the addressee. This is summarized in (103).

(103) Core Meaning o f an Imperative

In an imperative, it is proposed that the addressee bring about an event

The event in (103), the IM PERATIVE EVENT, Eimp, is commonly referred to as the

prepositional content of the imperative (Searle 1969, Leech 1983, Huntley 1984). It

captures the intuition that there is a core meaning shared by an imperative, ( 104)a, and the

corresponding declarative and interrogative, (104)b and c. The imperative event is the

situation in which the subject is a participant but it is the addressee who is ultimately

responsible for bringing it about.

(104) a. You be alert!

b. You are alert.

c. Are you alert?

The central claim of (103) can be encoded using the responsibility relation from Farkas

1988. In discussing cases of obligatory control with infinitivals, to which imperatives

semantically show significant parallels, Farkas 1988 develops the RESP-RELATION, a

“two-place relation, RESP(i» that holds between an individual i and a situation s just in

case i brings s about. If RESP(/» holds, i is seen as both initiating and controlling s, i.e.

the realization of s crucially depends upon i (Farkas 1988:36)”. The work demonstrates that

the RESP-relation has linguistic relevance and wide application to a variety of phenomena.

Given this property, the core meaning of an imperative is equivalendy that the addressee is

in the RESP-relation with the imperative event:


244

(105) Core Meaning of an Imperative

RESP(addressee, Eimp)

Farkas 1988:42-43 notes that if RESP(/,s) holds, certain characteristics of s follow. In

particular, s must be compatible with being intentional and, if s has an intentional agent,

then i is either that agent or is in the A-relation with it. This is relevant for imperatives

because, given (105), it requires that Eimp is an intentional situation because someone is

being asked to bring it about. Furthermore, if it is the case that Eimp has an agent who is

the subject, the addressee is exactly this agent or is in an A-relation with it. We thus arrive

at the desired restriction put forth in (98) that the addressee must be in an A-relation with

the referent of the imperative subject.

In order for this conclusion to go through, it is necessary to assume that the imperative

event has an agent who is the subject. This is not a particularly controversial assertion and

is certainly true in the canonical case, although it does not follow from the semantic

analysis that was outlined above. I will state it as a requirement in (106) and leave it to a

more thorough semantic treatment to account for its basis.

(106) the imperative event has an agent that is the syntactic subject

To conclude, there is a strong connection between the imperative subject and the addressee,

as Downing 1969 first proposed, which can be derived from the use to which an imperative

is put. While imperative subjects are not formally restricted, they must nevertheless be in a

control relationship with the addressee to make the imperative felicitous. This claim

accounts for the intuitions from B&C’s and Downing’s proposals that motivated this

discussion: 1) certain imperative subjects are better than others and 2) there is a connection

between the addressee and the imperative subject. With regard to the second intuition, this


245

section has offered a characterization of this connection. It in turn suggests which noun

phrases will be most acceptable as imperative subjects. An imperative subject will be more

acceptable to the extent that a control relationship can be construed between it and the

addressee. A noun phrase that corresponds to the addressee, namely the pronoun you, will

clearly be most acceptable. In Downing’s scenario where the imperative subject is a subset

of the addressee, it was suggested that this control relationship is nearly as direct. Because

of the particularly transparent nature of the relationship between the addressee and the

imperative subject, these two situations are predicted to yield the most felicitous imperative

subjects. The claim, then, is that the observation that the imperative subject is more

commonly quantificational, as B&C asserted, is an epiphenomenal description that follows

from the fact that quantificational imperative subjects transparently quantify over the

addressee and fall under the larger generalization proposed above.

4.3 . Summary

This section has considered various hypotheses regarding the syntactic and semantic status

of imperative noun phrases with the goal of illustrating that INPs are unexceptional

imperative clause subjects. Both positive evidence for the claim and arguments against

alternatives have been presented. Section 4.1 demonstrated that INPs display a full range of

canonical subject properties and behave syntactically like subjects. Section 4.2

demonstrated that the null hypothesis regarding the formal realization of subjects, that they

should be unrestricted, obtains to a high degree. INPs are largely semantically conventional

in their options.

At the same time, we have eliminated numerous possibilities that would allow INPs to be

partly subject-like. Section 4.1 argued against a proposal from Beukema and Coopmans

1989 which recognized INPs as subjects but still permitted them to be syntactically


246

anomalous in their behavior. Section 4.2 presented empirical evidence against restrictions

on INPs from Beukema and Coopmans 1989 and Downing 1969. It was concluded that

such restrictive analyses, while apparently motivated, do not belong to the grammar of

imperative subjects. Instead, a restriction on the imperative subject and its relationship to

the addressee was proposed and shown to follow from a conventional analysis of

imperative semantics. Summarily, English imperative clauses may have overt subjects with

the concomitant expected behavior. In the remainder of this work, I will assume that INPs

are subjects and nothing more. To the extent that the analyses are successful, they only

serve to confirm this position.

5. The Null Imperative Subject

While great care has been taken to demonstrate that the overt noun phrases that appear in

imperatives are subjects, the fact remains that in the unmarked state of affairs, in English

and cross-linguistically, there is no such noun phrase present. Imperatives are known for

their lack of a subject, (2).





This section investigates the syntax of these null subjects. Section 5.1 first shows that

theoretical considerations drive us to recognize a subject position in the examples which

must be occupied by a null element of some kind. I conclude that it is a null pronominal

element, pro in the empty category typology of Chomsky 1981, 1982. The difficulty then

becomes to explain why a null pronominal should be permitted in English, and many other


247

languages, which do not otherwise exhibit null argument phenomena. Section 5.2 explores

one approach to this theoretical issue, which has been labeled the identification hypothesis:

null pronominal arguments must be identified by the features of a local head in order to be

licit (Jaeggli 1982, Chung 1984, Rizzi 1986, Lillo-Martin 1986, Farkas 1987, Farrell

1990, and others). Imperative pro has important implications for this line of investigation. I

argue that person features play the central role in its identification but the identification is

accomplished via semantic, and not purely syntactic, means. The null subject is identified

via the addressee nature of the subject. The analysis supports the extension of the

identification hypothesis proposed in Farkas 1987.

5.1 . Interpretation of the Null Subject

The possibility, which we have seen to be repeatedly realized, of an overt subject in

English imperatives strongly supports the availability of a subject position in imperatives

more generally. Even when there is no phonetically pronounced subject, it is suggested that

there is still a subject position which is occupied. This conclusion has both purely theory-

internal and more empirically-oriented motivation (see Jaeggli and Safir 1989a for an

overview). From a theoretical perspective, a strong D-feature of Infl ensures that there will

be a filled subject position.18 The empirically-oriented evidence for a subject position

comes from the range of subject-identifying phenomena surveyed in section 4.1. These

arguments, with the exception of quantifier float, have been well-documented elsewhere

(Chomsky 1975, Katz and Postal 1964, Stockwell et. al. 1973, Beukema and Coopmans

1989, and others) and I merely repeat the data. The arguments can easily be reconstructed.

In brief, the phenomena that require a subject noun phrase: binding, control, and

18This is also true of earlier theoretical mechanisms, Chomsky’s (1981) Extended Projection Principle in particular.


248

grammatical operations that derive subjects, show that one must be present in null subject

imperatives. (107) illustrates binding of reflexives and reciprocals; (108) demonstrates

control of infinitival complements by the null subject; and (109) shows various subjects

derived via passive, subject-to-subject raising, unaccusative advancement, and tough

movement.

(107) a. Behave yourself!

b. Don’t hurt each other!

(108) a. Try to be more conventional!

b . Figure out how to open the package!

c . Drink lots of water to lose weight!

(109) a. Don’t be fooled!

b . Appear to be reading when he walks in!

c. Come quickly!

d. Be tough to deal with!

Quantifier float (QF) also supports the claim that there is a filled subject position, although

the facts are interestingly complex. Sportiche’s (1988) analysis of quantifier float from

chapter 1 requires a subject for the presence of a floated quantifier because these quantifiers

are actually stranded by the moving subject. If floated quantifiers are permitted in

subjectless imperatives, then it is evidence for the presence of a subject. Earlier, we saw

that they may occur when the subject is overt, in (76).


249

(76) a. (Don’t) The twins both be here for the picture!

b . Rhett, Scarlet, and Lassie all get ready for the next scene!

c. MY children all come here!

d . You be both waiting for me promptly at 3!

e. Don’t my students all take off at the bell!

The analysis leads to the expectation that a phonologically zero noun phrase that can form a

constituent with all or both will also participate in quantifier stranding. That null elements in

general are able to strand quantifiers is briefly illustrated in the two situations below. (110)

demonstrates that PRO, the subject of English infinitives, supports stranded quantifiers.

The examples are from Baltin 1995 and under current assumptions the infinitivals in subject

position have the structure exemplified by (110)d.

(110) a. To all have been doing that would have been inconvenient.

b . To have all been doing that would have been inconvenient.

c . To have been all doing that would have been inconvenient.

d. CP

Q tj have tk been tk | doing that

all

The examples in (111) show that Greek, an argument drop language, permits subject pro to

support a quantifier, (111) and (112), or strand one, (113). The examples are assumed to


250

have the paired structures which are parallel to their English counterparts modulo lexical

items and raising of finite verbs to 1° (Lightfoot and Homstein 1994b:4).

“(They) all left”

b - tn > I d p Pro ] j [ p efiganj [ y p t j [y > [ t j ] ] ] ] ]

(112) a. pro Oli exunfiji.all.3pl. have leave.participle

“(They) all have left”

b • [ip [dp Pr° li [p exunj [yp tj [y. tj [yp tj [y. fiji ] ] ] ] ] ]

(113) a. pro Exun pithanos oli fiji.have probably all.3pl. leave.participle

“(They) have probably all left”

b- [ip P r 0 [ [p exunj [yp pithanos [yp tj [v > tj [yp [Qp tj oli tj ] [v . fiji ] ] ] ] ] ] ]

As expected then, floated quantifiers may appear in subjectless imperatives. The data in

(114) and (115) illustrate non-neutral examples, both negative and emphatic. The

acceptability of the negative example in (114)a with don't was first noted in Cohen 1976

and Schmerling 1977. These observations confirm that a subject element exists in

subjectless imperatives.


(114) a. Don’t Jbothl . 1 all } S°!

b. Don’t both talk at once!

c. Please don’t have all taken huge portions! I want there to be some left for

myself.

(115) a. Do BOTH come and help! We can find something for you to do.

b . Do all try the food! It was catered especially for this event.

251

The facts are more complex, and mysterious, when neutral imperatives are considered.

Schmerling 1977 and Cohen 1976 assert that floated quantifiers are not possible in positive

subjectless imperatives, citing (116)a.

b . *That’s a good idea. Both be here for the rehearsal and we can see how the

production is shaping up!

c. * All come to the feast tonight!

d. *A11 line up on the left this time!

It seems that the null imperative subject can only inconsistently support a floated quantifier.

The fact that it can do so at all, in (114) and (115), supports the claim that there is a subject

position available. Nevertheless, the data in (116) contrast with the above non-neutral cases

and very similar examples in (117) in which the quantifier is still associated with a second

person subject and the result is grammatical.

c. You both be here for the rehearsal!

It appears that floated quantifiers are ungrammatical when they appear clause initially

without an overt supporting DP. Structurally, the requirement in (118) holds. When a

floated quantifier is associated with a null element, pro or PRO, the head that most

immediately c-commands it must be overt.

(118) a specifier position whose sole phonetic content is a floated quantifier must

be properly governed by an overt head


252

This restriction correctly characterizes the contrast between (114) and (115) on the one

hand and (116) on the other. Only in the former is there an overt head c-commanding the

quantifier. It also suggests that the latter, ungrammatical type would be improved if the

quantifier could be lower in the structure, below an overt c-commanding head. Such a

situation could be realized by placing an auxiliary to the left of the quantifier and the

imperative verb phrase. When this is done, in (119), the data seem improved, yielding a

contrast between (119) and the parallel examples without the auxiliary in (120).

(119) a. ?Be both waiting for me promptly at 3!

b . Please have all made your beds before the guests arrive!

(120) a. *Both wait for me on the comer!

b . *A11 make your beds before the guests arrive!

Sportiche 1988 and Baltin 1995 observe that the same kind of restriction holds of PRO.

Baltin’s cases of PRO adjacent to a floated quantifier, (121), are bad compared to when the

quantifiers are lower, (122).

(121) a. *PRO All to do that would be inconvenient,

b . *They promised PRO all to leave.

(122) a. PRO To all do that would be inconvenient,

b . They promised PRO to all leave.

Under the assumption that the infinitival clauses are CPs with a null C° head, the data are

also covered under (118). I will not attempt to provide an analytical basis for (118).

Accepting that there is subject position in subjectless imperatives, the question then

becomes what occupies it in these imperatives? Given Chomsky’s (1981, 1982) typology


253

of empty categories, there are four choices: a trace of movement, a variable, PRO the

pronominal anaphor, and pro the null pronominal. The first two options were argued

against in section 4.1.1. PRO is also an unlikely candidate given its behavior elsewhere in

the language, a conclusion also reached in Zhang 1990. There are both syntactic and

semantic arguments against it. One compelling observation is the fact that in imperatives the

null element alternates with an overt noun phrase with no apparent syntactic modifications

elsewhere in the clause. In infinitival complements, by contrast—another context in which

such an alternation is possible—PRO in subject position correlates with the simultaneous

absence of the complementizer fo r which is otherwise obligatory. For functions as the

governor and Case assigner when there is an overt subject. Whether or not PRO is taken to

be ungovemed in all cases (Chomsky 1981, Kayne 1991), it does not generally alternate

with a lexical noun phrase without overt syntactic consequences elsewhere. In imperatives,

however, there is no apparent difference between those with and without subjects which

could signal a difference in government of the subject position.

The interpretation of the null subject also argues against PRO. Established instances of

PRO in English allow two very distinct readings: an arbitrary interpretation exemplified by

(123)a with the structure in (123)b and a controlled one that is dependent upon its

controller, (124). Only when PRO is controlled can it ever have a referential interpretation.

(123) cannot be synonymous with (124).

(123) a. To lose is always disheartening.

b. [cp [jp PRO^jj [j, to lose] ] ] is always disheartening

(124) a. You don’t want to lose.

b. YoUj don’t want [cp [jp PROj [r to lose] ] ]


254

In the case of imperatives, there is no syntactic controller yet the interpretation of the null

subject is not the arbitrary one, which is impossible. For instance, the example in (2)a,

Take out the trash! cannot mean “(some)one should take out the trash”. The subject is

interpreted as referring to the addressee. The meaning is precisely what is not expected if

the null subject were PRO. I conclude, then, that PRO is not in the subject position of

imperatives. By process of elimination, only pro remains and it is also an appropriate

analytical option. Pro, a zero pronoun, is well-known from the extensive literature on

argument drop languages (see for example Perlmutter 1971, Jaeggli 1982, Rizzi 1982,

Bouchard 1983, Burzio 1986, Safir 1986, Jaeggli and Safir 1989a, and Roberge 1990 on

subject drop which will concern us here). It has the same syntactic distribution as other

pronominal forms except that it is phonologically null.

In support of pro as the null subject in imperatives is the obvious point that, in English

imperatives at least, it freely alternates with overt pronouns and noun phrases. The

interpretation of the null imperative subject also argues for pronominal status. In the

remainder of this section, I demonstrate that the null subject is an exact counterpart to the

overt pronoun you. Given this, the section concludes that the null subject is the null

pronoun pro and the following one then turns to its specific syntactic analysis.

The empirical question regarding the interpretation of imperative pro is how the missing

subjects in (2) are interpreted.






255

The starting point in determining the interpretation of the null subject is the imprecise

observation that imperative subjects are in some sense second person. There are, a priori,

two specific instantiations of this intuition. First, pro is grammatically second person and is

restricted to the interpretations in (7) offered earlier for second person pronouns.

(7) Characterization o f Second Person Pronoun Reference

c. 2nd person, singular, you: addressee

d. 2nd person, plural, you (all): addressee plus others not the speaker

This option fundamentally claims that the null imperative subject is an exact non-overt

counterpart to second person pronouns—you and you (all) in English. Alternatively, pro is

semantically second person; that is, its referent is actually restricted to being precisely the

addressee. Given the relationship between the imperative subject and the addressee, this

possibility is a reasonable. The null imperative subject would be a syntactic realization of

the addressee participant role.

As is evident from (7)c, the two options are equivalent when the pronoun is singular

because the second person singular pronoun corresponds precisely to the addressee. The

two diverge when the pronoun is plural. A second person pronoun may include in its

referent other entities in addition to the addressee. If the interpretation of pro is done

syntactically, an imperative subject will be able to include in its referent other non

addressee entities. If the interpretation is based on the semantic entity of addressee, then

this possibility is ruled out. The empirical issue is how we understand imperative examples

as in (125) and (126). Is the interpretation of the null subject able to include non

addressees, who are made obvious by the context, or is it limited to the addressee?

Although the judgments are complicated by the fact that the subject is null so that the

intended plural interpretation is not explicitly marked, the reading in which the imperative


256

subject refers to the addressee plus others is present. The possibilities of a predicate

requiring a plural subject interpretation, in (125)c, and a reciprocal which forces a plural

subject, in (126)b, confirm this.

(125) I’m taking my wife and driving to Mexico for vacation.

a. Well, pro have a good time! (= You both have a good time!)

b . Don’t pro drink the water! (= Don’t either of you drink the water!)

c. Don’t pro fight too much!

(126) a. You and John can stop now. Have lunch and then go back to work!

(= You and John have lunch and then go back to work!)

b . You and Mary need to get moving. Help each other out and it’ll go quickly!

(= You and Mary help each other out!)

The conclusion is that imperative pro is an exact zero counterpart of the overt pronouns you

and you (all). Imperative pro is grammatically second person and corresponds to the

characterizations in (7). This is stated in (127). It is not restricted to addressee reference.

This suggests that the relevant feature in the syntactic analysis of imperative pro, which I

turn to immediately, will be related to [PERSON: 2], where the feature [PERSO N ] is

straightforwardly defined in (128) according to the relationships from (7).

(127) Interpretation of the Null Imperative Subject

The null imperative subject is second person and has the grammatical feature

[PERSON: 2]


257

(128) PERSON Feature Definitions

a. [PERSON: 1]: includes speaker

b. [PERSON: 2]: includes addressee but not the speaker

c. [PERSON: 3]: includes neither speaker nor addressee

5.2 . Identification of pro

Since Taraldsen 1978 first noted the connection between referential pro19 and the presence

of “sufficiently rich” inflection, a promising line of analysis has been that the inflectional

morphology and the associated head are what permit pro to appear in some languages but

not others. This approach has been labeled the IDENTIFICATION HYPOTHESIS (Jaeggli

1982, Chung 1984): a null pronominal argument must be identified by the features of a

local head in order to be licit. In exploring this line of analysis, I will rely on the exposition

in Rizzi because it is perhaps the most explicit treatment of the hypothesis (the reader

should refer to Y. Huang 1995 for a more complete overview of the history of the analysis

of null arguments). In Rizzi 1986, a proper use of pro requires two distinct elements:

licensing and recovery. Formal licensing is a syntactic requirement according to which pro

must be in an appropriate structural configuration with a licensing head. The recovery

condition requires the presence of suitable information which makes the identity of the null

pronoun ‘recoverable’ or ‘determinable’. Rizzi’s particular proposal is given in (129).

Licensing is performed by the head that assigns Case to pro and recovery is accomplished

through feature sharing with the licensing head.

19I will only be concerned with argument, referential pro here unless otherwise indicated. Other types of null pronominals which ultimately require analysis include quasi-argument, expletive, and arbitrary pro.


258

(129) Licensing and Recovery Conditions on pro (Rizzi 1986)a. pro is Case-marked by X°, a governing head of type y

b . let X be the licensing head of an occurrence of pro; then pro has the

grammatical specification of the features on X coindexed with it

The licensing portion of the analysis of pro will be of secondary interest here. Within

Rizzi’s formulation in (129)a, the head which licenses pro in imperative subject position

will be the same head that assigns Case to this position when an overt subject noun phrase

is present. I take this to be 1°, the traditional source of Case assignment to the subject in

English. It is not uncontroversial, however, that pro requires Case. Jaeggli 1986, Roberge

1990:39, Authier 1992, and Holmberg and Platzack 1995:105 propose that pro is

(potentially) Caseless. If this is correct, then Rizzi’s formulation of the Licensing Condition

on pro cannot be correct and the formulation, and perhaps even the existence, of a specific

licensing condition on pro suspect.20-21 For concreteness, I will adopt the position that the

licenser of imperative pro, if there is one, is 1°. 1° governs spec,IP and has the potential of

assigning Case to it as well.

The recovery requirement is the core of the analysis of imperative pro. It is what gives pro

its interpretation. The basis of Rizzi’s proposal is that pro is identified by morphosyntactic

features on the head that licenses it. These features, when they are realized in a sufficiently

transparent manner, in a sense substitute for the null argument and permit its interpretation

to be determined. The postulation of a recovery condition attempts to formally capture

Taraldsen’s intuition that rich agreement is necessary for pro to appear.

20Lillo-Martin 1986 argues with data from American Sign Language that a licensing condition formulated to require that pro be properly governed is also incorrect and that only identification is needed.21In support of the Caseless nature of pro, one might point to imperatives in the numerous languages that do not permit pro to alternate with an overt subject. This state of affairs would follow if Case was not assigned to the imperative subject position in these languages. Only pro could appear.


259

Building on Falk 1993, there are actually at least three distinct issues involved in a precise

formulation of a recovery condition: 1) the exact features that are needed to recover pro, 2)

the manner in which the relevant features reach the null argument, and 3) the extent to

which these features must be morphologically realized. Regarding the first question, it is

commonly assumed that the features needed to identify pro are (a subset of) the agreement

features person, number, and gender. These are the features that typically appear in

argument-verb agreement. Specifically, Y. Huang 1995, summarizing results of numerous

researchers, indicates that person and number marking are relevant for the identification of

referential null subjects, which includes imperative pro. I consider this issue in more detail

in section 5.2.1 and argue that the crucial feature is person.

The second issue is explicitly addressed by Rizzi’s Recovery Condition in (129)b. The

identifying features are stipulated to be coindexed between pro and the licensing head.

Other potential mechanisms for sharing the necessary features include spec-head agreement

and local feature checking. On the other hand, if pro does not require a licenser or if

recovery is not mediated through an identifying head then it is conceivable that the recovery

of pro's features is direct. The Recovery Condition that will ultimately be adopted argues

for this broader position. Identifying features may be associated with pro through

traditional syntactic means such as (129)b, coindexing, or feature checking or they may be

determined independently and associated to pro directly.

The third issue is the central difficulty surrounding the analysis of pro and, under standard

conceptions, is the problem of defining what is meant by “sufficiently rich” inflection.

How much and what range of overt morphology must there be to determine pro? Although

numerous attempts to answer this question have been made (for example Roberge 1990 and

Jaeggli and Safir 1989b), no definition has proven satisfactory. Yan Huang argues that


260

there is no tenable theory of pro recovery based on morphological agreement. He points out

that rich morphological agreement is neither a necessary nor sufficient condition on pro.

Languages such as Chinese (C.-T. J. Huang 1984, 1989) have no agreement yet permit

null arguments. This demonstrates that rich morphological agreement is not a necessary

condition on pro. That it is also not a sufficient condition is demonstrated by German

(Jaeggli and Safir 1989b) and Icelandic (Platzack 1987) which have rich systems of

subject-verb agreement but do not license referential null arguments. Y. Huang 1995 cites

additional apparent empirical difficulties for a morphological agreement view from language

change. Because agreement paradigms are inextricably linked to the possibility of pro, a

diachronic change in one leads to an expectation that the other will also be affected. Lillo-

Martin 1986 mentions the loss of distinctive second and third person subject agreement in

Cuban Spanish yet the dialect retains the possibility of null subjects. Y. Huang 1995, citing

SigurSsson 1993, claims that Old Icelandic was a full null argument language while

Modem Icelandic is not—it does not have referential subject drop. This change in null

argument options was not accompanied by a weakening in the agreement inflection.

Icelandic agreement has gone largely unchanged in the development from Old Icelandic to

Modem Icelandic. The loss of full argument drop in the language is thus surprising on a

purely agreement-based view or pro recovery.22 In summary, while there is a high

correlation between morphological richness and the possibility of null arguments, the

connection does not seem strong enough to do more than suggest cross-linguistic

^ I t is clear that both situations require further investigation before Y. Huang’s conclusion can be accepted. For the Cuban Spanish case, it could very well be that the range of distinctions necessary for “rich” agreement still obtain. With regard to the Icelandic situation, Holmberg and Platzack 1995:110 offer a different, and perhaps more faithful, interpretation of SigurSsson 1993. They take Old Icelandic to have pro subjects only in a restricted set of discourse-controlled cases. What was lost in the change to Modem Icelandic were these contextually determined instances o f pro. In essence, they are claiming that Old Icelandic was not a subject drop language in the traditional syntactic sense.


261

tendencies. The Recovery Condition to be proposed below will not require referential pro

to be recovered morphologically as in Rizzi’s analysis. The issue will remain but will take

on reduced importance.23

In the next subsection, I motivate a Recovery Condition from Farkas 1987 that will address

the three issues discussed here and will prove to be superior to Rizzi’s Condition (129)b in

empirical coverage. The cmx of the proposal is that the feature [PERSON] must somehow

be determined in order to recover pro. The discussion thus spends considerable effort

defending the claim, in agreement with other researchers, that the crucial identifying feature

for pro's recovery is in fact [PERSON].24

23A weaker proposal is to admit that the desired concept of “rich inflection” is simply a theoretically necessary construct that only imperfectly correlates with overt agreement. Lillo-Martin 1986:441 suggests that rich inflection is ambiguous between “overtly morphologically rich” and “syntactically rich enough to allow a null argument”. This latter notion she calls STRONG AGREEMENT. For example, the languages Italian, Greek, and German all have comparably morphologically-rich systems o f verbal inflection; however, only the first two have null referential arguments so that only they would be analyzed as having strong agreement which licenses null subjects. Syntactically, German has WEAK AGREEMENT which does not license pro. The way in which Lillo-Martin’s proposal presages current ideas regarding feature checking and strong/weak syntactic features (Chomsky 1993, 1995) is striking and her proposal would seem to adapt readily into a minimalist framework. Analyses in this vein are developed in Platzack 1994 and Speas 1994. A minimalist approach to the recovery o f pro would admit that identification is a morphosyntactically-based property that need not be fully determinable from looking at a language’s morphological paradigms. The possibility of identifying pro is encoded in the lexicon, in the strength of the features [PERSON] and/or [NUMBER]. Such an approach maintains the core of Taraldsen’s observation and at the same time acknowledges the overwhelming empirical difficulties that seem to face a purely morphological agreement- oriented theory of null arguments. The problematic languages mentioned above would be accounted for in a formal feature-based analysis because there are no necessary theoretical implications, in either direction, between morphological richness and the identification of pro. Furthermore, a formal feature account of pro recovery would not necessarily preclude an analysis that incorporates the concept of rich agreement. If it turns out that rich agreement could be appropriately defined then such a definition would provide a basis for asserting that a language has strong agreement features. Despite its promise, I will not pursue such an analysis here.24The proposal will be unable to account for languages such as Chinese and Korean in which pro is permitted but there is no morphology that would permit its identification. The mechanism to be proposed is therefore unlikely to be the only means o f licensing a null element in a given language.


262

5 .2 .1 . The Role o f [PERSON] in the Identification o/pro

Within the literature on null arguments, there is a significant amount of disparate evidence

that the unique feature that is relevant for the identification of pro in many languages is

[PERSON]. In this section, I survey the empirical findings and analytical results from a

range of languages—Hebrew, Arabic, Brazilian Portuguese, and Hungarian—and show

that they all converge to support a Recovery Condition for pro from Farkas 1987 which

simply requires that the feature [PERSON] be determinable.

In many null subject languages, it is the case that one cannot determine which agreement

feature(s) are responsible for identifying pro since more than one is usually present. Hence,

“rich inflection” is employed as a general cover term. However, in at least two of the

languages surveyed below—Hebrew and Arabic—it can be ascertained that the crucial

agreement feature is [PERSON]. For these languages, it can be demonstrated that [PERSON]

is the feature uniquely responsible for identifying subject pro?5

The facts regarding Hebrew null referential subjects are documented in Borer 1980, 1983

and Doron n.d., with the data in (130) through (132) taken from Borer 1983:208.

Summarily, the Hebrew data show that subject drop is permitted in the past and future

tenses in first and second persons only, (130)a, b and (13 l)a, b. It is prohibited in all of

the present tense, (130)c, (13 l)c, and (132)c, and with third person subjects, (132).

25It has been suggested to me that Finnish illustrates the same state of affairs but I have not explored the data.


263

FIRST PERSON

(130) a. (‘ani) ‘axalti ‘et ha-banana(I) eat.past. 1 .sing acc the-banana “(I) ate the banana”

b. (‘ani) ‘oxal ‘et ha-banana (I) eat.fut. l.sing. acc the-banana “(I) will eat the banana”

c. *(‘ani) ‘oxelet ‘et ha-banana(I) eat.pres.sing. acc the-banana

“I eat the banana”

PAST

FUTURE

PRESENT

SECOND PERSON

(131) a. (‘atem) ‘axaltem ‘et ha-banana(you.pl.) eat.past.2.masc.pl. acc the-banana “(You all) ate the banana”

b. (‘atem) toxlu ‘et ha-banana(you.pl.) eat.fut.2.masc.pl. acc the-banana “(You all) will eat the banana”

c. *(‘atem) ‘oxlim ‘et ha-banana(you.pl.) eat.pres.masc.pl. acc the-banana

“You all eat the banana”

PAST

FUTURE

PRESENT

THIRD PERSON

(132) a. *(Hu) ‘axal ‘et ha-banana(he) eat.past.masc.sing. acc the-banana “He ate the banana”

b. *(Hu) ‘oxal ‘et ha-banana(he) eat.fut.masc.sing. acc the-banana “He will eat the banana”

c. *(Hu) ‘oxelet ‘et ha-banana(he) eat.pres.masc.sing. acc the-banana

“He eats the banana”

PAST

FUTURE

PRESENT

Doron and Borer point out that it is precisely in the present tense that the inflectional

paradigm is defective and there is no indication of [PER SO N ], only [NU M BER] and


264

[GENDER]. In the other tenses, third person is the unmarked person. The data argue that the

crucial feature in licensing pro is [PERSON]. When it is absent, pro is not possible.

Kenstowicz 1989 documents a similar situation in the Bani-Hassan dialect of Arabic. In

looking at null subjects in various types of complement clauses, he observes that inflection

for person is a necessary condition for a null subject. In finite clauses, (133), the verb is

inflected for person and a null subject is possible. Alternatively, the subject can be realized

as a clitic pronoun on the complementizer. This situation contrasts with participial clauses

which are tenseless and used to express certain aspectual information. The participial verb

form exemplified by (134) does not inflect for person, only gender and number, and is

incompatible with a pro subject. The subject must obligatorily be realized using the clitic.26

The data thus suggest that the relevant feature in identifying pro, at least in Bani-Hassan

Arabic, is also [PERSON].

The relevance of person marking to the recovery of null arguments is not limited to pro in

subject position. Farkas 1987 and Farrell 1990 analyze null objects in Hungarian and

(133) Fariidgaal

Fariid say .past

finnuprol [ in n -h a j iStarat

.fem.sing. the-dress

“Fariid said that (she) bought the dress”

(134) Fariidgaal f*innu pro] [ inn - ha J miStarya al-libaas.

[ that pro 1 [that - shejFariid say .past buy .participle the-dress

“Fariid said that she has bought the dress”

26These are not control structures with PRO since neither example requires a coreferential interpretation as far as I can determine from Kenstowicz’ discussion.


265

Brazilian Portuguese, respectively, and argue that the person feature is crucial to the

recovery of pro in that position as well. I will present their data only as necessary to

motivate their analyses since they both formulate versions of a Recovery Condition that will

ultimately contribute to the one proposed here.

Farrell 1990, in investigating null objects in Brazilian Portuguese, modifies Rizzi’s

Recovery Condition in (129)b with the additional clause in (140)b. In the absence of

identifying features from the licensing head provided by Rizzi’s Condition, (140)a, pro is

inherently marked with particular grammatical features which allow its content to be

recovered.

(140) Recovery Condition on pro (Farrell 1990)

Let X be the licensing head of an occurrence of pro. Then:

a. pro has the grammatical specification of the features on X coindexed with

it, or

b. pro has the grammatical specifications of designated features.

Farrell shows that Brazilian Portuguese freely allows null objects but that they violate

Rizzi’s Recovery Condition because the governing verb shows no agreement. The object

occurrences of pro are limited to third person, however, as (141) illustrates.

(141) Coitado

poor

do Joao1 *de mim *de voceJoao me you

. O chefe mandou pro embora.

the boss sent pro away

‘Poor John. The boss fired him’’


266

The second clause of the modified Recovery Condition in (140) is designed to permit this

use of pro. Pro in Brazilian Portuguese is stipulated to have the feature [PERSON: 3]. Its

content is recoverable under (140)b because its person feature is fixed by the language.

Consider the consequences of Farrell’s proposal on the three issues introduced above.

First, the particular feature that is relevant for the recovery of Brazilian Portuguese object

pro is [PERSON]. This is not unexpected given what we have seen thus far and further

supports the contention that it is generally [PERSON] that is instrumental in recovering the

content of pro. Second, the feature that recovers pro is associated with pro from the

lexicon. Pro is recovered directly, without the use of an mediating licensing head. Unlike

Rizzi’s recovery clause in (130)a, Farrell’s (140)b does not invoke the licenser X. Third,

with regard to the richness of the identifying feature, the feature that identifies pro is not

morphologically realized. Under the analysis, identifying features do not need to be

morphologically overt. In short, a language-particular strategy provides another way of

recovering pro's identity by fixing it to third person.

Lastly, Farkas 1987 analyzes null direct objects in Hungarian. The work uses Hungarian

data to convincingly argue for the same conclusions: 1) recovery of pro requires the

successful determination of the feature [PERSON] and 2) overt morphology on the licensing

head is not available, and so not necessary, to recover the content of direct object pro.

Farkas proposes the Recovery Condition in (142). In Hungarian, there is an explicit,

though non-trivial, algorithm that determines the value of the feature. It is based on

syntactico-semantic properties of the direct object and the conjugation form of the verb.

(142) Recovery Condition on pro (Farkas 1987)

The feature [PERSON] must be recoverable, i.e. there must exist some

algorithm which yields its value.


267

It should be evident that the Recovery Condition in (142) can also account for the

remainder of the languages surveyed in this section. For the Hebrew and Arabic data, the

value of [PERSON] is recoverable from the overt morphology. In Brazilian Portuguese, it is

recoverable because it is fixed to being third person.27

It is worth comparing Farkas’ Recovery Condition in (142) with Rizzi’s, repeated below,

on the three issues identified above.

(129) Recovery Condition on pro (Rizzi 1986)

b . let X be the licensing head of an occurrence of pro; then pro has the

grammatical specification of the features on X coindexed with it

Farkas’ proposal is more restrictive in that it identifies the particular feature [PERSON] as

responsible recovering pro.28 It is less restrictive in the manner in which the feature is

recovered. Its value is not restricted to resulting from feature matching with the licensing

head as in Rizzi’s analysis; it need only be determinable by some method. The licensing

head is potentially not involved.

5.2.2. The Identification o f Imperative pro

We are now in a position to put together the pieces for an analysis of imperative pro. The

analytical tool will be Farkas’ Recovery Condition. It claims that imperative pro is licensed

because the value of its feature [PERSON] is available. I will show that this follows from

earlier discussions about the meaning of an imperative (section 4.2.3) and the interpretation

27Farkas 1987 mentions similar instances of pro in French, Turkish, and Palauan in which pro occurs with a fixed interpretation which is not dependent upon its licensing head but which entails a particular value for the feature [PERSON].28This same assertion is made in Holmberg and Platzack 1995.107.


268

of the null imperative subject (section 5.1). The method of recovery will clearly not be

syntactic, however, since it is apparent that no morphosyntactic realization of agreement

appears in English imperatives. I will suggest that the method of recovery is semantic.

To demonstrate that (142) accounts for imperative pro, we must show that there is an

algorithm to determine the value of its person feature. From the previous result that the null

imperative subject is exactly second person, (127), we know that the feature value that

must be recovered is [PERSON: 2], We must therefore determine an algorithm to recover

[PERSON: 2] for imperative pro.

(127) Interpretation o f the Null Imperative Subject

The null imperative subject is second person and has the grammatical feature

[PERSON: 2]

The method for accomplishing this will be via the addressee. A noun phrase whose referent

is the addressee is always second person—although the converse is not true: a noun phrase

that is second person does not always refer exactly to the addressee since second person

noun phrases may include non-addressees in the plural (see the characterization in (7)).

More accurately, then, the addressee is always a subset of the referent of second person

noun phrase. Crucially, if the referent of a noun phrase can be determined to include the

addressee, its person feature will be uniquely determined to be [PERSON: 2]. This leads

back to the core meaning proposed for an imperative, in (103), from section 4.2.3.

(103) Core Meaning o f an Imperative

In an imperative, it is proposed that the addressee bring about an event

In the canonical case, the addressee will bring about the imperative event by being its agent

or at least its initiator. The typical syntactic realization of this role is as the subject. Thus,


269

the reference of the canonical subject of an imperative includes the addressee. Combined

with the claim that a noun phrase whose referent includes the addressee is specified as

second person, we arrive at the result that the canonical null subject imperative must have a

second person subject. From this conclusion, imperative pro is licensed.

The algorithm has been developed working backwards from the required [PERSON: 2]

specification. Working forwards, the calculation is roughly as follows: The core meaning

of an imperative, (103), indicates that, in the canonical situation, the agent of the imperative

event will be the addressee. Since, in the canonical case, the agent of a clause is also

realized as the subject, the addressee will be the imperative subject. At the same time, the

addressee is grammatically realized as second person. This yields the desired conclusion

that the imperative subject is grammatically second person or [PERSON: 2], By (142) pro is

recovered in imperative subject position.

5 .3 . Consequences

Given this general line of reasoning, the result provides strong support for Farkas’ (1987)

revision of the Recovery Condition. The proposal is able to predict the appearance of pro in

imperative subject position without appeal to ad hoc syntactic mechanisms. The Recovery

Condition does not restrict the manner by which pro is recovered, thereby permitting a

semantically-oriented recovery procedure.

It is also striking confirmation of the general correctness of Farkas’ Recovery Condition

that the recoverable information actually is [PERSON: 2], exactly as claimed. Observe that

the feature [NUMBER], for example, is not recoverable in imperatives. The prediction that

accompanies this claim is that the interpretation of the null imperative subject in a given

language will co-vary with the language’s definition of grammatical second person. If there


270

were some language that defined [PERSON: 2] differently from English, we would expect

the interpretation of the null imperative subject in that language to also differ. For example,

if there were a language in which second person plural corresponded to a plural number of

addressees, rather than a plural group that includes the addressee as in English, one would

expect the imperative subject in the plural to have precisely the same interpretation, not the

interpretation of the English second person plural.

The analysis has several additional positive consequences that I make explicit here. First,

since the recovery condition makes no mention of rich agreement of morphological feature

matching as the only method for recovering pro, overt imperative agreement morphology is

not necessary to permit pro in imperatives. This is clearly a desirable result for English, and

many other languages, in which imperatives are devoid of any morphology. If pro is

indeed the subject of subjectless imperatives, rich agreement cannot be a necessary

condition on its identification and the Recovery Condition acknowledges this.

Perhaps the most significant consequence of the proposal is its cross-linguistic relevance.

The above algorithm for the recovery of imperative pro can be seen to be completely

independent of particular linguistic assumptions about English. None of the individual

computations depend upon details of English grammar. The strategy is in principle

available to all natural languages and we make the seemingly correct prediction that

imperative subjects cross-linguistically can be null. In support of this, Zhang 1990 surveys

imperatives in 46 languages from 13 families and asserts that subjects may be dropped in


271

all but two of these languages: Luo (Nilotic) and Hawaiian (Polynesian).29 Given the range

of ways in which languages syntactically encode imperatives—word order, verbal

inflection, agreement, and/or intonation—no individual strategy is likely to be the source of

the null imperative subject. That the above analysis depends upon none of these but is

fundamentally semantic instead is a positive result. The possibility of pro in imperatives

seems to be a nearly universal option and this universality must have roots in a non-

language-particular aspect of grammar since all languages seem to have the right linguistic

mechanisms. The analysis suggests that this universal property is the close connection that

can be derived between the imperative and the person feature of its subject. Although

imperatives vary in syntactic form, their use is cross-linguistically consistent. This result

cannot be ignored in the study of null arguments because of its pervasiveness. Any theory

of pro that professes any amount of generality must make the same predictions about

imperative null subjects.

In agreement with Roberge 1990, a final consequence of the Recovery Condition in (142)

and its application to imperatives is that there can be no Null Subject Parameter per se. It

cannot simply be the case that a language is either a null subject language or it is not and the

child acquiring a language sets the parameter to on or off. Pro must be a universally

29My own investigation of the languages indicates that Luo at least freely allows subjectless imperatives. None o f the examples cited in Stafford 1967, including those in (i), has subject or subject agreement morphology.

(i) a. Kel chiemo! bring food “Bring food!”

b. Go-ye! hit-him “Hit him!”

Sources on Hawaiian (Elbert 1970) indicate that the subject in imperatives is “commonly expressed”. It is unclear whether this means they are obligatorily expressed.


272

available option in language acquisition. Parametric variation is largely at the level of

whether or not the linguistic mechanisms to satisfy the Recovery Condition are in place.

Appendix. Patterns of Pronominal Agreement

In this appendix, I return to the pattern of pronominal agreement from section 3.1.3 that

was used to demonstrate that not all INPs are vocatives. I will be concerned with the

possible person attributes on pronouns coindexed with imperative subjects, in particular,

when they are grammatically third person.30 The pattern to be accounted for is in (143).

With regard to syntactically third person noun phrases, vocatives agree only with pronouns

that are second person but imperative subjects may trigger second or third person

agreement.

(143) Patterns o f Pronominal Agreement with Third Person Noun Phrases

a. vocatives agree with second person pronouns only

b . imperative subjects may agree with second or third person pronouns

As we saw earlier, vocatives can only be coindexed with second person pronouns. This is

independent of the person specification of the vocative NP itself. Second person pronouns

are used if the vocative is second person, in (144), or third person, in (26) repeated below.

30The options for number on the pronouns, singular or plural, will not be discussed because the pattern does not differ from that in declarative clauses. In particular, I ignore the possible use of third person plural pronouns in situations where a gender-neutral singular pronoun is desired:

(i) Everyone enjoyed themselves.


273

your*his

*theirtoys scattered on the floor?(144) a. Hey, you, why are

b . You with the glasses, you can sit near the front.

(26) a. Bond, do(es) >’you]*/ > have a light?

*he Jb. Those with tickets, may take j*^gZy j seats.

c. Sir, we have j * ^ J table prepared.

d. Dave, we’d better get to a doctor.


With imperative subjects, a second person pronoun is also required if the subject is second

person, in (145), which fully parallels (144). On the other hand, if the subject is a third

person noun phrase, either second person or third person was seen to be possible, in (27),

repeated here. Additional examples with subjects in negative imperatives are in (146).

(145) a. Don’t you start on j j dessert until you finish your peas!

b . Don’t you misbehave at the party!

c. You be good to ■

d. You mind j *^ "J business, and leave this to me!

e. You show me how to make that recipe like so much!


274

(27) a.

b.

c.

d.

e.

f.

g-

(146) a.

b.

c.

A way to make sense of this pattern is to claim that there are multiple agreement

mechanisms available in the grammar. Agreement features can be determined syntactically

or semantically. This idea is inherent in the proposals of Farkas and Zee 1993 and Pollard

and Sag 1994, henceforth F&Z and P&S, respectively. SYNTACTIC AGREEMENT is

agreement with the strict syntactic features of a noun phrase. This is a dominant mode in

languages that have grammatical gender (for example, the Romance languages) or extensive

noun class systems (numerous Bantu, West African, and American Indian languages). In

this situation, the form of the agreeing pronoun reflects the lexical features of the

antecedent. In contrast to this is SEMANTIC AGREEMENT in which properties of the noun

phrase’s discourse referent rather than the noun phrase itself help to determine agreement

features. These discourse-derived features may then override or fill in for morphosyntactic

features. P&S discuss several instances of semantic agreement in English. An illustrative

Somebody lend me handkerchief!

Everyone pick up | ^ ^ r | bill on the way out!

One o f the boys test while I wait!

Drivers have j ^ j y j money ready!

Nobody forget lunch for the picnic tomorrow!

Whoever knows the answer put up hand!

People interested in the project give me a way in which to contact !

Don’t everybody raise j ' <r| hand at once!

Don’t anyone trip over j ^h is \ laces^

Don’t those with cars forget that j J promised to drive tomorrow!


275

case is reference transfer, in which a salient inanimate object stands in syntactically for a

typically animate referent. Agreement, however, is mediated through the intended referent

of the noun phrase rather than the actual inanimate noun phrase. Examples are given in

(147). Semantic agreement is implicated because pronominal agreement reflects the

gender/number of the human discourse referent, not the syntactic gender/number of the

syntactic antecedent.

(147) a. The ham sandwich at table nine just made a fool of j • (said by one

waitress to another)

b. The hash browns at table six said j^ g y } cant find the men’s room.

I claim that semantic agreement also occurs when syntactically third person imperative

subjects trigger second person anaphora. This arises because the subject, although

grammatically and even referentially third person, has a referent whose discourse role is

closely aligned to the addressee and the addressee is realized morphosyntactically as second

person.

For concreteness, I follow P&S in assuming that all noun phrases have an associated index

that contains specifications for the agreement features PERSON, NUMBER, and GENDER.

The resolution of the person feature in agreement configurations is the primary concern

here and I take the possible specifications of the feature to be [PERSON: 1], [PERSON: 2], or

[PERSON: 3] as previously defined in (128).

(128) PERSON Feature Definitions

a. [PERSON: 1]: includes speaker

b . [PERSON: 2]: includes addressee but not the speaker

c. [PERSON: 3]: includes neither speaker nor addressee


276

Indices play a central role in pronoun-antecedent agreement in the P&S system. Pronouns

agree with their antecedents by requiring that indices be identical. Thus, pronoun-

antecedent agreement is fundamentally token identity of indices. Consequently, agreeing

pronouns provide a window on the features of an antecedent’s index.

A noun phrase may have its agreement features specified in one of the ways introduced

above. Both methods are in principle available and the theoretically revealing cases are

those in which they yield distinct results. This is the basis for the ambiguous pattern of

agreement with imperative subjects. The dominant or default pattern seems to be that the

values are determined morphosyntactically. Syntactic agreement is governed by the

lexically-specified agreement features of a noun phrase’s head. Lexically-headed phrases

such as the scissors, a car on the street, or blue lemons have the values of their indexical

features specified from the lexicon via the head N°. Syntactic agreement thus accounts for

the normal pattern of anaphora in which pronouns coindexed with third person subjects are

also third person. This is illustrated in (148) through (150) for both imperative and

declarative clause types. These subjects are assigned agreement features from the head

noun and trigger third person agreement.

(148) a. Everyone behave himself.

b. Everyone should behave himself.

(149) a. Somebody lend me his car to go to the hospital!

b . Won’t someone lend me his car to go to the hospital?

(150) a . Nobody forget the secret number that he chose!

b . Nobody is to forget the secret number that he chose.


277

Semantic agreement is seen when the subject’s agreement features are obtained via its

discourse referent.31 It was discussed in section 4 that the discourse role of the imperative

subject is aligned with the addressee in a particular way. Although imperative subjects are

not limited to being the addressee, it is nevertheless true that the two are, in the typical case,

closely related: the referent of the imperative subject is either identically the addressee or an

intersecting set with the addressee. This connection is the source of the second person

agreement with imperative subjects since the addressee is grammatically second person.

When the imperative subject is aligned with the addressee, second person agreement is

possible, as in the above examples. See Pollard and Sag 1994 for a system in which these

two types of agreement may be concretely implemented.

Given these two modes of agreement and the possibility that both may appear with

imperative subjects, we are led to certain expectations regarding the variation. At the very

least, the system identifies a difference between the two agreement options and predicts,

correctly, I believe, that they are not simply completely free variants. It suggests general

circumstances under which each pattern will be more felicitous. Third person agreement

will be the default, as in finite clauses. The imperative subject is viewed neutrally. Second

person agreement will be preferred when the connection between the imperative subject and

the addressee is apparent or being emphasized. The data below suggest that this division of

labor is largely borne out. Since there has little explicit investigation of the anaphora

patterns beyond the observation that both are possible (but see Stockwell et. al. 1973), the

following data makes a contribution to a deeper understanding of the phenomena.

31 F&Z argue that semantic agreement occurs when a lexical head is not available or able to supply morphosyntactic feature values. They discuss two cases in Romanian: conjoined noun phrases and deictic pronouns. The latter are not lexically headed but assumed to be heads belonging to the functional category D" (Postal 1970, Abney 1987).


278

As a preliminary, when the imperative subject is syntactically the addressee, the second

person pronoun you, only second person agreement is possible. This is illustrated in (145),

repeated below, and is unsurprising. The same pattern obtains in finite clauses in (151).

The data are compatible with either a syntactic or semantic agreement account since the

subject is simultaneously syntactically and semantically second person, being

morphologically a second person pronoun and discourse-wise the addressee.

(145) a. Don’t you start on | dessert until you finish your peas!

b . Don’t you misbehave at the party!

c. K™ be good to

d. You mind business, and leave this to me!

e. You show me how to make that recipe j * ^ j like so much!

(151) a. You can’t start on j */^j dessert until you finish your peas.

b . I hope you won’t misbehave at the party.

c. You should always be good to

d. It’s best if you mind j * / ^ J business and leave this to me.

e. Can you show me how to make that recipe J like so much?

With third person imperative subjects, either possibility is generally available. The

examples in (152) contain cases in which third person anaphora is strongly favored over

second person. Hypothesizing that third person agreement is the default and that second

person agreement becomes felicitous as the imperative subject becomes aligned with the

addressee, the preference can be understood because the italicized subjects in (152) are not


279

closely related to the addressee. Either the referent of the subject is not even present in the

discourse or because the second person pronoun you is explicitly used elsewhere.

Examples of this type, independent of whether or not anaphoric agreement is present, have

the sense of disinterested orders or instructions. They sound impersonal and this is a

consequence of the detachment from the discourse participants. The subject is referentially

independent of the addressee and thus second person agreement is largely unacceptable.

More subtle examples in which both pronouns are possible can also be easily created, in

(153) and (154). The choice of pronoun is not based on truth conditions or grammaticality

but on the intended effect of the utterance. Third person agreement suggests a distancing or

detachment, (153). When the focus of the sentence is away from the addressee and the

subject, second person agreement sounds less appropriate:


rooms!

b . You do your homework and John do

e.

d.

280

c. The bills are ready. Everyone pick up

hand at once

handkerchief!

hand at once!

e. We really need drivers for tomorrow’s rally. So, don’t those with cars forget that

As suggested earlier, when second person is used, a closer connection to the addressee is

implied. Emphasizing the addressee nature of the subject reverses the preferences. In the

examples in (154), the utterances are more blatantly directed towards the addressee. While

the contrasts between (153) and (154) are subtle, I believe they are in the indicated

direction.

(154) a. You all are so stingy. Somebody just lend me j r ^ j j handkerchief!

b . Tolls are collected on the driver’s side. Drivers have money ready!

c . Each of your bills is ready. Everyone pick up {^hi^S} on the way out!

d . You all are just so rude to each other. Don’t everybody raise j > hand at

The same distinction can be illustrated by introducing a second person partitive phrase into

the subject. This focuses attention on the addressee nature of the subject and, consequently,


promised to show up.

once!

e. Some of you volunteered for tomorrow. So, don’t those with cars forget that

promised to show up.

281

makes second person agreement preferred, in (155). When the partitive phrase is absent, as

contrasted in (156), third person agreement is dominant.

(155) a. The brightest boys among you finish homework as fast as | ?^ ^ |

can!

b. Every one o f you pick up towel!

(156) a. The brightest boys finish j homework as fast as can!

b . Everyone pick up j ‘ j ^ rj towel!

The data indicate that the choice of an agreement pattern with imperative subjects may be

either syntactic or semantic, with concomitant differences in discourse use. If these

observations are correct, the same pattern should be expected in declarative clauses. That

is, the pattern should not be unique to imperatives. The data in (157) and (158) suggest that

this is the case, although second person agreement does seem less good. The acceptability

of the data indicate that the ability of an addressee-oriented subject to trigger second person

agreement is not a property that needs to be specifically accounted for with imperatives. It

more easily appears with imperatives because, as was discussed above, the imperative

subject is often closely aligned with the addressee.

(157) a. Somebody should lend me j handkerchief if you don’t want this wine to

spill on the white carpet.

b . The manager says that everyone is to pick up bill on the way out.

c. I want one of you boys to test j while I wait.


2 8 2

d. The teacher would like for whoever knows the answer to quiedy put up j jhand.

(158) a. I’m told that the brightest boys among you have already finished

homework.

b. On the way out, every one of you may pick up j towel.

Lasdy, it remains to be accounted for why vocatives do not participate in this pattern. Third

person vocatives permit only second person agreement, never third person:

(26) a. Bond, do(es) j have a light?

b. Those with tickets, may take seats.

c. Sir, we have j * ^ J table prepared.

d. Dave, we’d better get j ^ - ^ J to a doctor.


Vocatives evidently illustrate a strict form of semantic agreement. Since a vocative is

exactly the addressee in all cases, semantic agreement dominates and makes second person

anaphora possible. The question is why it is required and third person agreement

impossible. The underlying reason appears to be because the vocative is unambiguously the

addressee. It cannot be anything else. Imperative subjects, in contrast, are discoursally

indeterminate. They are not so restricted. Part of what it means to be a vocative is that a

noun phrase’s index is specified as the addressee and second person. This overrides any

lexical specification from the head noun.


283

Chapter 4

The Status of Imperative D o ( n ’ t )

1. Introduction

The most theoretically central topic in the generative syntax literature on English

imperatives has been the status of imperative do, don't, and do not in relation to their

tensed clause counterparts (Chomsky 1975, Cohen 1976, Schmerling 1977, Lasnik 1981,

Davies 1986a, Pollock 1989, Zhang 1990, Zanuttini 1991 among others). At issue is the

extent of the syntactic similarity between the occurrences of do in the imperatives in (1) and

the corresponding declaratives in (2). The a examples illustrate an emphatic use of do while

the b and c examples illustrate its use with negation.

(1) a. DO show more compassion for the less fortunate!

b. Don’t ever walk alone!

c. Do not leave the doors unlocked!

(2) a. You always DO show compassion for the less fortunate.

b . They don’t ever walk alone.

c. We do not leave the doors unlocked.

The fundamental competition is between analyses that attempt to assimilate the imperative

markers into an analysis of finite clause Do-Support and those that assert that the


284

differences between imperatives and finite clauses are empirically and theoretically

significant and require distinct syntactic mechanisms to account for them. This work takes a

novel, stepwise approach to the controversy and the data. The imperative markers do and

don't, which are typically found initially in emphatic and negative imperatives, as in (3) and

(4), are investigated in this chapter separately from the traditionally more problematic do

not, which is the focus of chapter 6 .1 will refer to them collectively as do(n ’t) or DO.

(3) a. Do have some more tea!

b. Do be seated!

c. Do SOMEONE give him the money before he uses that gun!

d . Do at least YOU try to understand where I’m coming from!

(4) a. Don't be bullied into signing!

b. Don’t tell me to be quiet!

c . Don't anyone open this box!

d . Don’t you mention it further!

In section 2 I lay out instantiations of two general analytical approaches to such examples

which are roughly characterized according to whether they give DO a syntax that is internal

or external to the clause—as an inflectional head or an introductory imperative particle.

Both positions have been proposed and argued for in the literature. Section 3 presents

novel syntactic and semantic evidence in favor of the claim that do(n’t) is an inflectional

head base-generated in 1°, as in finite clauses. This conclusion fundamentally differs from

that of Cohen 1976 and Zhang 1990, the most thorough existing investigations of this

issue, and section 4 explicitly addresses their arguments for the introductory particle

analysis. The chapter concludes that an analysis in which DO is integrated into clausal

syntax as an inflectional head is superior in the range of data and generalizations that it can


285

capture. DO behaves like the corresponding finite forms and can be profitably analyzed

with existing mechanisms that are independently needed for finite clause structure. This

conclusion provides further support for the claim that imperatives have largely

unexceptional syntax. The elements do and don't fit expectedly into general English clause

structure.

2. Two Hypotheses

(3) and (4) above illustrate the commonly held view that do and don’t in imperatives are

found exclusively in left-peripheral position even when a subject is present. Akmajian 1984

summarizes this observation: “imperative do and don't occur sentence-initially” and it will

serve as the starting point for the two analytical hypotheses about the syntax of DO. Within

the system of phrase structure adopted here, two methods for assuring that a given element

occurs clause-initially are 1) placing it in the head position of the topmost projection if it is a

zero-level category and 2) left-adjoining it at the clausal level if it is a phrasal category. I

develop these two options as the auxiliary analysis and imperative particle analysis of DO,

respectively.

2 .1 . The Auxiliary Analysis

The auxiliary analysis of imperative DO capitalizes on the word order similarity between the

imperatives in (3) and (4) and nearly identical yes/no questions. Given the parallel, the

position of DO in non-neutral imperatives could be obtained via a subject-auxiliary-

inversion-like process as postulated for interrogatives. Under such a scenario, illustrated in

(5), DO is a zero-level category located in the head position of the topmost projection,

which I have neutrally labeled XP. It originates in a lower inflectional projection that is also

the base position of modals and finite support do.


imperative verb phrase1

The central claim being defended in the auxiliary analysis is stated in (6). DO is a head

which has lexical-categorial characteristics similar to modals and support do. (5) is one

possible realization of an analysis which incorporates this hypothesis.

(6) The Auxiliary Analysis of DO

DO is base-generated as the head of an inflectional projection

This assumption is inherent in numerous analyses of the imperative including Chomsky

1975, Emonds 1970, Schmerling 1977 for don't only, Davies 1986a, Pollock 1989 for do

only, Zanuttini 1991 in some respects, and Platzack and Rosengren 1994. It would seem

necessary for any analysis of imperatives which would claim their derivations to involve

Do-Support. The hypothesis in (6) and its instantiation in the above structure will be

defended against an alternative below. For the present, I will leave the identity of the X

projection undetermined. Likewise, I will not be concerned with the syntax—the position

and derivational history—of the imperative subject within the structure in (5). When

necessary, I will assume it to be located in the specifier of IP. Lastly, although I take the

inflectional head under which DO is inserted to be 1°, more generally, the possibility that it

*1 use the label ‘imperative verb phrase' solely to make clear that the structure under consideration is an imperative. No theoretical significance should be attached to the term.


287

is one of the heads in a complex Infl scenario (Pollock 1989) is transparently not ruled out.

These issues are all the subject of the next chapter.

2 . 2 . The Imperative Particle Analysis

The imperative particle hypothesis differs from the above in that it does not incorporate DO

into the internal structure of the clause but takes it to be external. DO is an introductory

imperative particle that is either an adjunct on the clausal node or a head which selects a

non-finite clause. Introductory particles, syntactically invariant markers of imperative

mood, are common cross-linguistically particularly for negative imperatives (Schmerling

1977, Sadock and Zwicky 1985, Zhang 1990) and so their appearance in English would be

unsurprising. A concrete proposal in this vein is illustrated in (7), based on Zhang 1990.

DO is an adjunct of category ImpP which is attached to the clausal IP. DO transparently

appears in the left peripheral position and precedes all other elements. As an adjunct, it is

not part of the internal syntax of the IP to which it is adjoined.

(7) IP

ImpP ip

DO r

f ^ ^ V P

imperative verb phrase


288

The crucial property of the particle analysis, stated in (8), is that the imperative marker is

peripheral to the inflectional projections of the clause.2

(8) The Imperative Particle Analysis o f DO

DO is base-generated external to the clause

In transformational terms, such an analysis ascribes syncategorematic properties (Pullum

1982) to DO and it is introduced directly by phrase structure rule or transformation rather

than by lexical insertion at deep structure (see the transformational analyses in Cohen 1976

and Schmerling 1977 which treat don’t and do, respectively, in this way). Zhang 1990 is a

more recent analysis of imperatives which argues for the structure in (7) for don’t, although

TP replaces the unitary IP. The imperative particle approach fundamentally differs from the

auxiliary analysis in that the imperative marker does not participate in the syntactic

derivation of the clause; it serves only as a flag to mark the imperative force of the sentence.

3 . Evidence

The data presented in this section argues for the auxiliary analysis of DO. The arguments

demonstrate that DO participates in the syntax in ways that an inflectional head would. The

data are unexpected from the position that DO is simply a peripheral adjunct. Section 3.1

reintroduces VP ellipsis facts from chapter 2 to demonstrate the inflectional head

characteristics of DO. Section 3.2 investigates similar yet interestingly different first person

2An admissible alternative to adjunction that is also found in the literature is to have DO be a head which subcategorizes for a non-finite clause. Bolinger 1967 and Pollock 1989 (for don’t only) analyze DO as a fossilized main verb which takes a VP or IP complement. Beukema and Coopmans 1989 takes DO to be a complementizer which subcategorizes for a non-finite IP. This version of the analysis is not different in essential ways from that considered in the text. The arguments carry over since they are independent of the phrasal status of the imperative particle.


289

let ’ -constructions and demonstrates that the auxiliary analysis also extends to that domain.

Analyzing let’s as an auxiliary gives an insightful account of the data while an imperative

particle approach requires largely unmotivated mechanisms and is syntactically unrevealing.

The appropriateness of an auxiliary analysis in dealing with another imperative-like

construction thus argues for its place in English syntax. The interesting differences between

the two constructions, restrictions on morphosyntactic features of the subject, are

accounted for if the central elements DO and let’s are inflectional heads, the locus of lexical

agreement specifications. Section 3.3 presents syntactic and semantic parallels between

imperatives and indicative clauses that are captured with an auxiliary analysis. First,

imperative data from Davies 1986a reveal a syntactic, word order symmetry between DO in

imperatives and do in finite clauses. Second, semantic parallels in the use of do in the two

clause types suggestively argue in favor of syntactically assimilating the two instances of

do.

3 .1 . VP Ellipsis

Chapter 2, building on Schmerling 1977, observed that ellipsis in imperatives was

grammatical just in case either do or don’t was present, in (9), but not otherwise, (10).


b . Billy didn’t tell mom what I did, so don’t you 0 either!

c. Don’t you be seen with those losers and don’t anyone else be 0 either!

d. We want everyone to come, so those who can, by all means do 0!

e. Someone needs to volunteer to help with the church bazaar.

Oh yes, please, do someone 0!

f . The movers told us to be careful with the dining table.

Yes, do be 0! It’s a priceless heirloom.


290

(10) a. *Jack started reading the poem, now Jimmy keep 0!

b . *Cinderella is doing her chores so, if you know what’s good for you, be 0 too!

c. *You should tidy up before the guests arrive. Yes, definitely have 0!

d . *We can’t have any latecomers. You should definitely be on time. Absolutely no

excuses, be 0!

The syntactic requirement on VPE from the chapter 1 analysis is repeated in (11)—a null

VP must be introduced by an overt inflectional head.




Given these observations and analysis, DO is clearly implicated as the ellipsis licenser in

(9) and it must be an inflectional head according to (11). Of the two hypotheses, only the

auxiliary analysis allows the Licensing Condition to be satisfied and provides a transparent

account of the VPE data. Under the analysis, DO is the licensing head—since it is explicitly

built into the analysis that DO is the head of an inflectional projection—and it permits the

appearance of a null VP. A typical imperative ellipsis case has the structure in (12) which is

(5) less the VP.

(12) XP

DO; (subject) V

t f ^ ^ V PI0


291

Under the particle approach, VPE in imperatives is completely unexpected. This is

explicated in (13) for the elliptical imperative Don’t! The two structures shown are

potentially available; however, neither one is actually permitted.

(13) a. * IP b. * IP

ImpP IP ImpP IPi i r . ■ .

don’t 0 don’t I’

I VPI0

(13)a is not licit because the empty category is not a VP. Furthermore, it is well-known that

adjuncts may not be stranded directly preceding an empty category (Baker 1971; Sag 1978,

1980). However this observation is captured theoretically, it is also sufficient to rule out

(13)a. (13)b is likewise not permitted because the inflectional head that is present is

apparently empty. If it contains any morphology, it is clearly not overt. Consequently, the

head cannot license the null VP in accordance with the VPE Licensing Condition. The

imperative particle analysis must assume that the data in (9) are instances of non-productive

or frozen expressions. Essentially, ‘special rules’ as proposed in Akmajian 1984:15-18 will

be required which state the use of do and don't in these contexts. Such an approach, in

addition to its inability to provide a structure for the VPE data, would seem to miss the

obvious syntactic and semantic connections between full imperatives and the elliptical

expressions above. I demonstrate this immediately below.3

First, Akmajian 1984 observes that expressions semantically similar to Do! and Don’t! are

not possible as elliptical expressions:

3 T w o of the generalizations that follow were pointed out in Akmajian 1984 and thus it is particularly unclear why that work finds it necessary to consider DO as a special set of ‘imperative particles’.


292

(14) A: Can I take the last cookie in the jar?don't

♦mustri t *shouldn't

*notB: No, -

(15) A: I wonder whether I really should have one of these appetizers? do

•! You only live once.B: Oh, *should♦can

With a particle analysis, it is accidental that the expressions that can be used in an elliptical

manner are precisely those found introducing imperatives. Under an approach like the

auxiliary analysis that can analyze these examples as simply elliptical clauses, the

correspondence is transparent.

Second, the availability of longer elliptical forms in (9)b, c, e, and f in which either a

subject and/or non-finite auxiliary is additionally present is also mysterious under a particle

analysis. In order to account for these examples, one would be forced to further duplicate

as frozen expressions all the strings or templates that might result when a VP is missing in

a non-neutral imperative. This is an obvious, and seemingly unnecessary, complication of

the grammar. The data are expected and accounted for with the auxiliary analysis. For

example, (9)c receives the structure in (16). The licenser is again the inflectional head DO

which satisfies the condition in (11).

(9) c. Don’t you be seen with those losers and don’t anyone else be 0 either!


293

(16) XP

X IP

don’tj DP r

anyone tj else

VP

V VP

be 0

Third, Akmajian claims that the elliptical expressions have precisely the same pragmatic

functions as when they are used in imperatives. He points out that the Do! in (15) is “more

like an offer or suggestion than a ‘strong’ imperative. But this seems to be true of

imperative do in all cases.” Thus, (17) (Akmajian’s (36)), the non-elliptical counterpart to

(15), has precisely the same force:

(17) Oh, do have one of the appetizers!

Under a special rule analysis of the expressions, such semantic parallels, which could be

additionally exemplified, and which are further explored in section 3.3.2, are accidental—

suspiciously so. On an ellipsis analysis, however, the pattern is again what is expected if

both examples have the same source. Although not conclusive, Akmajian’s claim is at least

suggestive.

Summarily, the elliptical imperative data behave syntactically and semantically like those

structures that would remain if a verb phrase were chopped off a negative or emphatic

imperative. The imperative particle approach is unable to capture this generalization because

full imperatives and elliptical examples are not syntactically related. Under the auxiliary


294

analysis, these facts follow straightforwardly given the analysis of VPE in chapter 1 and

the proposed inflectional head status of DO.4

3 . 2 . L efV Constructions

This section and the appendix to this chapter provide further evidence in favor of the

auxiliary analysis by showing that it is superior to the imperative particle analysis in more

easily and revealingly extending to Z.£7’S-CONSTRirCTlONS (Seppanen 1977, Davies

1986a, Clark 1993) illustrated in (18).5

(18) a. Let’s go to the park!

b . Let’s see what’s at the movies!

c. Let’s be Simpson characters for Halloween!

d . Let’s have completely cleaned the place up by the time he returns!

4The VPE data also argue against the alternative imperative particle analysis mentioned in an earlier footnote in which DO is a head, either a main verb (Bolinger 1967, Pollock 1989) or a complementizer (Beukema and Coopmans 1989), but it is not part of the inflectional complex. As is known, neither main verbs, in (i), nor complementizers, in (ii), license a following empty category.

(i) a. *People began pouring out of the building and then smoke [ y began] 0 .

b. ’"John started speaking French because Mary [ y continued] 0.

(ii) a. *Jose doesn’t believe that the fireworks will take place but I think that] 0.b. *The jury is convinced that they did the right thing but the public wonders whether] 0.

5I hypothesize that /er’s-cons tractions have one of the analyses being investigated, an auxiliary or imperative particle analysis. A third analysis, different from both of these options is not ruled out of course. The appendix to this chapter demonstrates that /ef’s-constructions are not simply imperatives with a main verb let. It discusses related but distinct /ef-imperatives, in (i), and, secondarily, /ef-constructions, in (ii), and argues that let ’r-constructions differ at least from ordinary imperatives like (i).

(i) (All of you) let your arms hang gently to the side!

(ii) ‘Let me not to the marriage of true minds admit impediments.’ (Shakespeare, Sonnet 116)


295

I focus on the syntax of /ef’s-constructions and suggest that they largely have the structure

of non-neutral imperatives. The auxiliary let’s shows similarities to DO and this is

accounted for if it is given an analysis extended from the auxiliary hypothesis: let’s in these

examples is also base-generated in 1°. A particle analysis of let’s, in contrast, makes no

predictions—correct or otherwise—about the behavior of let ’5-constructions and is thus

uninsightfiil if not incorrect.

As a preliminary, I demonstrate that let's is synchronically a lexical unit (Fries 1952, Quirk

et. al. 1985) and not simply a contraction of let us. That is, the examples in (18) are not

derived from (19) via phonological reduction.

(19) a. Let us go to the park!

b . Let us see what’s at the movies!

c. Let us be Simpson characters for Halloween!

d . Let us have completely cleaned the place up by the time he returns!

With dialectal variation, several kinds of subject-like noun phrases can co-occur with let’s,

including us itself when accompanied by contrastive stress. A range of post-let’s noun

phrases, all italicized, are illustrated in (20).

(20) a. Let’s you and me be roommates next year!

b. %Let’s US go instead!

c. % Let’s us and them challenge the winners!

d. %Let’s all o f us go!

e. %Let’s everyone try and behave!


296

f. %Let’s no one forget to turn off the lights!

g . *Let’s the children play a little while longer!

h. *Let’s those who have cars leave them here!

Such examples do not have an uncontracted source. They are robustly ungrammatical

without contraction, (21). This is surprising if ’s were simply a reduced form of us. I take

it that let's is not a contraction of let us since the clause that follows let’s may have a

subject-like element of its own. Let’s thus should be treated as a lexical unit.6 1 turn now to

consider what sort of syntactic entity this unit is.

6Clark 1993 suggests that the acceptable examples above contain appositives derived from the examples in (i) which have explicit appositive intonation.

(i) a. ?Let us, you and me, be roommates next year!b. ??Let us, US, go instead!c. ?Let us, all of us, go!

The examples in the text, however, clearly do not require the comma intonation associated with the appositives in (i). In addition, the examples in (ii) and (iii) indicate that contraction is not generally possible with appositives in this environment. These examples are particularly bad with the unbroken intonation contour that is found on the grammatical examples.

(ii) a. Let us, those who are prepared, set off at once!b. *Let’s, those who are prepared, set off at once!

(iii) a. Let us, the stragglers, make an effort to catch up!b. * Let’s, the stragglers, make an effort to catch up!

The relative acceptability of (iv) is also unexpected if the noun phrase following let's is an appositive.

(iv) ?Let’s none of us say anything to her about it!

None o f us in this example does not behave like an appositive. First, it is an unlikely if not impossible appositive because of its semantic properties: *They invited us, none o f us. Second, it licenses the negative polarity item anything in object position. I conclude that Clark 1993 is incorrect in reducing the noun phrases under investigation to appositives.


297

(21) a. *Let us you and me be roommates next year!

b . *Let us US go instead!

c . *Let us all of us go!

d . *Let us everyone try and behave!

Let '^-constructions have two interesting properties that suggest that the morpheme let's is

an auxiliary in 1°: the agreement and case reflected on the subject.7 Both characteristics are

related to and follow from the theoretical claim that, in the canonical case, Infl is the locus

of agreement features and the source of Case assignment. First, as implicitly illustrated

above, the post-Zef’s noun phrase, whether overt or not, is restricted to having a first

person plural inclusive (lpl-incl) interpretation. It must refer to a group consisting of at

least the speaker and the addressee and, syntactically, its agreement features must be

lpl-incl. In the examples in (20), the most widely accepted subjects were pronouns that are

identically lpl-incl, for example us and you and me. The quantificational subjects were

accepted by some speakers because they may have a first person inclusive interpretation.8

Unambiguously third person noun phrases are ruled out because they do not have this

referent. This restriction on agreement features can be confirmed by looking at the tags that

may accompany let’s imperatives, in (22). Only we is possible as the subject of the tag but

with the interpretation that includes both the addressee and the speaker. Particularly relevant

is Quirk et. al.’s (1985) observation that the expression shall we?, which most naturally

7I will call the post-/er'r noun phrase the subject since I later claim that it occupies the specifier o f IP. I thus use the term subject in the sense developed in chapter 3: the noun phrase that occupies spec,IP. I do not claim that the post-/ef’s noun phrase is the logical subject of let’s since I will claim below that le t’s is base-generated in Infl and is like the modals in having no external argument.8This is strikingly confirmed by examples in which a pronoun anaphoric to the quantified subject must be grammatically first person and must include the speaker in its interpretation, at least figuratively. I am told that examples as in (i) typically have a use that is most appropriate with children in a kindergarten-like setting.

(i) %Let’s everyone take out our/??his/??their crayons and coloring books!


298

tags a let’s imperative, has the curious property of necessarily referring to lpl-incl rather

than first person plural more generally.

(22) a. Let’s have dinner now,

shall we can we, please

?may we ♦shall they ♦shall you

Analytically, the construction must somehow be specified as restricted to having a lpl-incl

subject. I will suggest below that it is advantageous to locate this property in the element

let's.

The second unique property of this construction is that, as the examples repeated below

show, the subject in let’s imperatives is in the accusative case.9

(20) a. Let’s you and me be roommates next year!

b . %Let’s US/^WE go instead!

These two characteristics of the subject, its restricted agreement features and non

nominative case, motivate an auxiliary analysis over a particle analysis. Case and agreement

are precisely the properties that are typically associated with Infl. Analyzing let’s as an

element of category 1° can account for the idiosyncratic variation; the specifications

9The possibility of nominative case in (i) is clearly a result of the more general recent tendency, illustrated in (ii), for nominative case to be substituted in conjoined noun phrases independent of structural position, (i) is not a reflection of ambiguous Case assignment in let’s imperatives as (iii) shows. See Seppanen 1977 and Davies 1986a for discussion.

(i) Let’s you and I be roommates next year!

(ii) a. %This is just between you and I.b. %The superintendent needs to see Bill and I right away.

(iii) *Let’s WE take care of this for them!


299

originate in the lexicon. Under such an analysis, let’s is a modal-like element base

generated in 1°. It is restricted to first person plural inclusive agreement and assigns

accusative Case to the noun phrase in its specifier. It has the lexical entry in (23). Like

other 1° elements, let’s subcategorizes for a bare VP.

(23)

A typical example receives the analysis in (24). Case to the subject is assigned or checked

in the specifier of IP by 1°. The inverted word order is also obtained from the auxiliary

analysis. Let’s raises from 1° to the unspecified head X°.

(24) Let’s you and me be roommates!

XP

let’sj DP I’

you and me tj VP

be roommates

Such lexical properties are unexpected and largely unanalyzable with an imperative particle

analysis. In this approach, let’s would be a clause-external adjunct and the above examples

would have the structure in (25). While such a structure is certainly possible, it provides no

basis for explaining the Case and agreement patterns described above. Adjoined elements

never determine such nominal properties since they participate only peripherally in the

syntax. Adjuncts, almost by definition, are optional and not implicated in required syntactic

operations.


300

(25) IP

ImpP IP

let’s subject r

I VP

The /ef’5-construction data argue for the auxiliary analysis of DO in that the analysis is seen

to be profitably transportable to a related construction in ways that the imperative particle

uncostly manner and thus would seem to be independently needed. Extending the auxiliary

analysis to let ’5-constructions has additional desirable consequences which I point out

below.

The extension creates an interesting completeness in the imperative inflectional paradigm.

We have seen that let ’5-constructions are restricted to first person agreement. If imperatives

which may take DO are in fact also lexically restricted, in this case to second and third

person agreement, then English can be said to have a nearly complete paradigm of

imperative modals. The differences are isolated at the inflectional level, specifically, in the

contents of 1°. Parallel to the lexical entry in (23) we have (26) for the morphology in

ordinary imperatives. - 0 j m p is a zero, dependent morpheme which is restricted to second or

third person agreement and which assigns nominative Case to its specifier. When DO is

inserted into 1°, it bears this morphology.

Such an analysis provides an explanation for the observation from chapter 3 that first

person pronouns in imperatives are generally impossible, even in seemingly appropriate

contexts.

analysis cannot be. It accounts for the idiosyncrasies of these imperatives in a theoretically

(26) agreement: 2 /3 Case: nominative


301

The auxiliary analysis makes three additional correct predictions regarding the syntax of

let ’ -constructions. First, the data in (27) indicate that they are negated with not in the

expected way. The head not, which marks sentential negation, appears below the auxiliary

and subject and immediately above the verb phrase.10 The expected structure is (28).

(27) a. Let’s not fight about such trivial things!

b . Let’s you and me not show up!

c . Let’s US not be fooled like they were!

(28) XP

let’sj subject I’

tj NegP

Neg VP

not

Second, the auxiliary analysis makes exactly the right predictions regarding VPE

possibilities. The argument is precisely the one discussed in the previous subsection: the

*°The pattern with not seems to be universally accepted; however, two other patterns of negation are also acceptable for a minority of speakers:

(i) a. %Don’t let’s try that! b. %Let’s don’t try that!

The first will be discussed in section 4. An analysis of the second pattern, which I will not pursue, is that don't rather than not is the head of NegP. This proposal receives initial support from the observations that in this dialect the subject appears in the same position, preceding negation, (ii), and don’t, as the head of NegP, expectedly licenses a null VP, (iii).

(ii) a. %Let’s you and me don’t stay up too late! b. *Let’s don’t you and me worry about that!

(iii) Should we go in?No, let’s don’t 0 !


302

auxiliary analysis predicts certain patterns of ellipsis, which are realized, while the particle

analysis is silent. Ellipsis is mostly productive in let’s imperatives (Seppanen 1977). In

addition to the fully elliptical Let’s! (Quirk et. al. 1985), other examples in (29) also seem

largely felicitous. The data follow from the above structures in (24) and (28) and the

analysis of VPE in chapter 1. Let’s or not is the inflectional head that licenses the null VP.

(29) a. Should we check out the new exhibit?

Yes, let’s 0!

b . Who wants to read this dialogue out loud?

Let’s you and me 0! It’ll be fun.

c. ?Someone needs to be the clowns for the party so let’s you and me be 0!

d . Let’s not 0, and say we did.

Third, we assign a reasonable structure to let ’ -constructions with no overt subject,

repeated in (18). There is a null pronominal in subject position which is identified by the

auxiliary. In accordance with the theory of pro from chapter 3, pro is permitted because the

auxiliary in 1° is able to recover the [PERSON] feature of pro via its lexically-specified first

person inclusive specification. (30) is the structure assigned to (18)a.

(18) a. Let’s go to the park!

b . Let’s see what’s at the movies!

c. Let’s be Simpson characters for Halloween!


303

(30) XP

let’Sj DP I*

pro tj VP

go to the park

In summary, let ’ -constructions prove to be interestingly different from imperatives with

DO; yet, they can be analyzed as having largely the same syntax with a range of parametric

variation that is predicted within the analytical proposal. Let’s too shows clear connections

to its clause in ways predicted by the auxiliary analysis.

3.3 . Syntactic and Semantic Symmetry

The auxiliary analysis differs from the particle analysis in that it formally recognizes and

seeks to maximize the similarity between imperatives and finite clauses—one of its explicit

motivations. To the extent that this parallel exists, the analysis gains support at the expense

of the imperative particle analysis which denies any such connection. This section presents

two parallels between imperatives and finite clauses, one syntactic the second semantic, that

follow automatically from the auxiliary analysis and are easily analyzable within the

proposal. The imperative particle account can either not handle the data or can do so only at

the expense of missing desirable generalizations.

3.3.1. Syntactic Parallels: Imperative Subject Position

A well-known syntactic operation in finite clauses is subject-auxiliary inversion that

accounts for the alternation in word order seen in (31).


304

(31) a. You can see the planet,

b . Can you see the planet?

Analytically, it is taken to be derived by leftward head movement of the auxiliary past the

subject and is possible because of the syntax given to finite clauses (see chapter 1). By

assigning the same clause structure to imperatives, we admit the possibility of this

alternation there as well. Imperatives with DO superficially parallel yes/no questions like

(3 l)b with the auxiliary appearing to the left of the subject. We might expect, then, that an

underived form analogous to (31)a in which the auxiliary has not moved leftward might

also exist. Below I introduce data, first discussed in Davies 1986a and not previously

analyzed to my knowledge, that confirm this possibility.

(32) and (33) re-illustrate subjects following do(n ’t).


b. DoEVERYonegiveitatry!


d. For heaven’s sake, of all people, DO YOU give me some support!

(33) a. Don’t EVERYone leave!


c. Don’t anyone touch my stuff!

Subjects also may precede DO. Examples are in (34) and (35) and I refer to them as

SUBJECT-FIRST IMPERATIVES. Warner 1993 calls these cases ‘peremptory declaratives’ but

it is clear from the lack of verb agreement with singular subjects, (34)a and (35)c, that they

are not finite clauses. They have the form and meaning of imperatives.


305



(35) a. Those with luggage don’t leave it unattended!

b . Girls go into the hall, BOYS don’t move!

c. Everyone don’t expect a raise!

Given the diagnostics in chapter 3, we can confidently ascertain that the noun phrases in the

above actually are subjects. First, the characteristic vocative intonation break is absent

between the subject and DO. Second, such noun phrases can refer to non-addressees, a

possibility that rules out vocative status. (36) contains examples in which the subject cannot

refer to the addressee and therefore cannot be a vocative.

(36) a. You and Mike do take more lasagna, there’s plenty!

b . ?At least your OLDER children do come camping with us!

c. I’ll be sure to watch everything while your family is gone.

You all don’t worry about the dog, it’ll be fine!

Third, these noun phrases are compatible with third person anaphora, an impossibility for

vocatives:

(37) a. SOMEone do lend me fa's jacket, please!

b . Those without galoshes do wear their boots instead!

c. Passengers with luggage don’t leave their valuables unattended!

d. Everyone don’t forget to pick up his orientation packet!

Consider how each hypothesis would account for these data. The relevant structure for the

auxiliary hypothesis is repeated in (5).


306

(5)

subject


It should be evident that by leaving the auxiliary in its base position the desired subjectADO

order results. This is illustrated in (38). The analysis ascribes to subject-first imperatives

the same phrase structure as ordinary declarative clauses with a modal or form of do.

(38) ip

subject I’

DO imperative verb phrase

Under the auxiliary analysis, such imperatives are actually expected. It remains only to be

explained what forces DO to move to the higher head in the normal case. I address this

issue in chapters 5 and 6.

The imperative particle analysis, in contrast, assigns no structure to the subject-first

imperatives. The basic configuration is repeated in (7).

(7)

ImpP ip

DO subject I’

f ^ ^ V P



307

To obtain the desired word order, one of several moves seems possible. With the subject in

the specifier of IP, the imperative particle ImpP could be adjoined lower in the structure,

either to I’ or VP. This is workable but clearly unmotivated. Alternatively, DO could be

located in 1°. But this then would be the structure hypothesized in the auxiliary approach

and we would require both analyses, rendering the particle analysis superfluous. Lastly,

the subject could be topicalized from spec,IP to an external, IP-adjoined position, deriving

the desired subjectADO word order:

(39) IP (SS)


This account, an instantiation of a proposal from Beukema and Coopmans 1989, was

rejected in chapter 3, section 4.1.1. Here I develop an additional argument against it based

on the interaction of quantificational imperative subjects and the LF operation of Quantifier

Raising (QR) (May 1985). Schmerling 1982 first points out that the scopal interpretation of

quantified imperative subjects is restricted with respect to the negative element don't and

her observation provides the foundation for an additional argument against allowing the

syntactic operation illustrated in (39). The overt topicalization of the subject hypothesized in

(39) is, syntactically, fully parallel to LF Quantifier Raising of a subject. The interpretation

of certain negative imperatives, however, indicates that the LF version of this movement is

not available, making it mysterious why the overt version would be permitted.


308

QR is a movement operation that targets a quantified noun phrase and raises it higher in the

structure for scope purposes. The movement applies covertly at the level of LF. A typical

syntactic position for QR-ed noun phrases is adjoined to IP. This permits a quantificational

direct object, for example, to take scope over the subject as in (40)a with the interpretation

in (40)b. The example is given the LF in (40)c. Thus, as discussed in chapter 3, QR at LF

is structurally the same as topicalization at SS.

(40) a. Everyone saw two films.

b . Two films are such that everyone saw them.

c. IP (LF)

DP; IP

two films DP I’

everyone I VP

iI

saw

In imperatives, this operation should target a quantified subject and adjoin it to the topmost

IP, above the imperative particle, yielding an LF precisely identical to the surface structure

in (39). Under an imperative particle approach with optional topicalization of the subject,

the imperatives in (41) with the word order DO followed by a quantified subject should

look at LF exactly the same as subject-first imperatives look at surface structure. (41)a

would have the LF in (42) which is identical to the structure in (39). (42) also corresponds

to the surface structure of (35)c given the analysis under consideration.

(41) a. Don’t everyone expect a raise!

b . Don’t three of you leave!


309

(42) IP (LF) - (41)a / (SS) ~ (35)c

DP: IP

everyone ImpP IP

don’t tj I’

I VP

expect a raise

Permitting the structure in (42), however, makes an incorrect prediction. The interpretation

represented by the LF in (42) is a reading that is absent in the above examples. For

whatever reason, imperative subjects are not able to take wide scope with respect to the

negative element don't:

(43) Restriction on Scope Interpretation in Imperatives

subjects to the right of DO always take narrow scope with respect to DO

Further unambiguous examples to be worked through are in (44). (44)c in particular is

ungrammatical because someone is a positive polarity item which cannot be interpreted

within the scope of negation. Since it cannot take wide scope according to (43), the

example has no interpretation and is ill-formed.

(44) a. Don’t all the workers take a break now!

b . Don’t two people order the same thing!

c . *Don’t someone touch the channel!

To illustrate, (41)a, repeated below, has two potential readings depending upon the relative

scope of negation and the quantified subject. These are paraphrased in (45). Either negation

or the universal quantifier has wide scope. The example actually has only the interpretation

in (45)b which is the one determined by the surface position of negation and the subject


310

where negation takes wide scope. The meaning in (45)a where the subject has wide scope

is unavailable.

(41) a. Don’t everyone expect a raise!

(45) a. Nobody expect a raise! EVERY > NOT

b . Not everyone should expect a raise. NOT > EVERY

The non-existent reading in (45)a can be eliminated if QR of imperative subjects is not

permitted. In order to capture the observation in (43), the LF movement operation in (42)

must be illicit. Theory-intemally, we have (46).

(46) Restriction on QR in Imperatives

QR does not apply to imperative subjects

(45) is an analytical stipulation covering the observation in (43). It eliminates the possibility

of adjoining imperative subjects to IP or a higher projection. Such a move seems necessary

although a theoretical motivation remains absent.

Accepting this conclusion, the analysis proposed in (39) now becomes mysterious: why

should adjunction of the subject to IP be permitted overtly but ruled out covertly? In the

absence of a better understanding of the driving force behind QR, I conclude that the

hypothesis that imperative subjects may be overtly topicalized across DO requires

contradictory assumptions about the movement itself. We are left with no explanatory way

of deriving the subject-first imperative word order with an imperative particle analysis.

In summary, this section has claimed that there is a syntactic, word order parallel between

imperatives and finite clauses which can be expressed by the following analogy:

declarative:interrogative::subject-first imperative:DO-first imperative. The auxiliary analysis


311

of DO predicts this pattern and accounts for the full range of imperative data. The particle

analysis, on the other hand, reveals no parallels and suggests, incorrectly, that none should

exist. The data have no transparent analysis.

3.3.2. Semantic Parallels: Emphatic do

Semantic parallels with DO in imperatives and finite clauses also argue for imperative DO

being an auxiliary. In this brief section I suggest that the elements do and don't each make

largely identical semantic contributions in the two clause types. Given this generalization,

the mull hypothesis regarding the syntax-semantics interface is that they should also receive

similar syntactic analyses. They do under an auxiliary analysis. Giving them distinct

accounts, as would result if the particle analysis of DO were correct, does not transparently

capture the semantic generalizations and implicitly claims that they are either spurious or

accidental.

In order to conclusively demonstrate that the semantic contribution of do and don’t in

imperatives is the same as that in indicative clauses, an explicit account of the semantics of

imperatives that can also be compared to that of indicatives is necessary. As is well-known,

precise analyses of imperative semantics are notoriously difficult, particularly if they are to

be compatible with the truth-conditional accounts typically ascribed to declaratives (Huntley

1980, 1982, 1984; Hamblin 1987; Aarts 1989). Additionally subsuming the contributions

of negation and emphasis would be a particularly demanding task. A fully elaborated

proposal is thus outside the scope of this work and will not be presented here.

Consequently, the argumentation in this section is necessarily programmatic. It is worth

presenting, however, because of its potentially promising result. A developed semantic

analysis will determine whether the general line of reasoning stands up.


312

Don’t transparently serves as a bearer of negation in both clause types and it seems

reasonable that the two instances of negation should have the same syntactic source. On the

other hand, the lexico-semantic connection is not so complex as to require a unified

syntactic source and to rule out synchronic reanalysis away from an auxiliary scenario.

Negative imperative particles are cross-linguistically common (Schmerling 1977, Sadock

and Zwicky 1985, Zhang 1990), although in the clearest cases they do not take the same

form as sentential negation.

The use of do in imperatives, however, shows more significant parallels with its use in

finite clauses. I suggest below that imperative do is neither optional, as claimed in

Schmerling 1982, nor semantically empty. It makes a definite semantic contribution in

imperatives which mimics its contribution in finite clauses. This generalization is captured

by giving do the same auxiliary analysis in both clause types. No such generalization

follows from the imperative particle approach and there is no expectation of semantic

parallelism.

It is widely recognized that do in Modem English positive declarative sentences functions

as a carrier of emphatic affirmation (Chomsky 1957, Klima 1964, Bolinger 1977, Laka

1990, and others). It affirms the positive polarity of the clause and is obligatorily

stressed.11

1 Several researchers (Ard 1982, Stein 1990) have suggested that all uses o f do can be reduced to one of emphasis. The thesis is that the distribution of do is semantically conditioned and the common denominator in its appearance with negation, interrogatives, and affirmation is a focus on the polarity of the clause, the truth or falsity of the proposition. I will not explore this more general possibility.


313

(47) Function o f Do in Positive Declarative Clauses

do is a carrier of emphatic affirmation

1) polarity contrast

2) emotive emphasis

This affirmation is realized in two ways (Quirk et. al. 1985, Denison 1993). In the first, the

affirmation is at the level of the proposition’s truth or falsity, its polarity. Do serves to

signal contrastive polarity. A sentence with emphatic do is used to deny or contradict a

negative claim that has been stated or implied in the discourse. Examples are in (48).

(48) a. Contrary to what you think, we DID enjoy ourselves at the opera.

b . Shakespeare DID write his own plays.

c . You’d never know it but Howard DOES have a great deal of money.

d . You should listen to your mother.

But I DO listen to her.

The second use of affirmative do is to enforce an assertion. In the examples in (49), do

carries an emotive emphasis. In these contexts it affirms or emphasizes the speaker’s

attitude—his/her enthusiasm, approval, or conviction—towards the proposition.

(49) a. I DO wish you would listen.

b . He certainly DOES like that new cereal.

c. You really DO look pale this morning.

d . It DOES indeed taste sweet.

It has been claimed that do functions differently in imperatives than in declarative clauses.

(Elleg&rd 1953:148, Huddleston 1980, Denison 1993:268). For example, Ellegird

1953:148 states that “emphatic do in imperatives has not the same meaning and function as


314

emphatic do in declarative sentences: it lends intensity or insistence to the demand or

command; it does not stress the reality of the verbal action.” On the other hand, Quirk et.

al. 1985:1415 note “similar emotive connotations” and Davies 1986a explicitly argues that

do in both clause types bears emphatic affirmation, contra previous alternative accounts

which attributed a unique meaning to imperative do—for example, a contribution of

‘entreaty’. If Davies’ position is correct, it offers support to the auxiliary analysis of do

under the assumption that the parallel semantics reflect a unified syntactic analysis. Davies

is to be consulted for fuller justification that imperative do can be reduced to emphatic

affirmation. Here I simply point out that the two uses of do in positive declarative clauses

identified above reasonably coincide with the two functions that traditional grammars

typically claim for imperative do.

The most widely known use of do in imperatives is in what Quirk et. al. 1985 call the

‘persuasive imperative’, illustrated in (50).

(50) a. DO sit down!

b. DO be quiet!

c . DO be careful cleaning those glass figurines!

This use assimilates to the emotive emphasis function of do. As in (49), above, what is

affirmed in these examples is the speaker’s positive attitude or emotions towards the

realization of the imperative.

The second use of do in imperatives is contrastive and examples of this sense are given in

(51), from Davies 1986a. Do functions to mark or contrast the polarity of the imperative

with respect to an alternative. These examples semantically parallel the declaratives in (48).

The same need for a salient alternative is necessary, as is explicitly seen in (5 l)a and b.


315

(51) a. Do you think I should tell her or not?

Do tell her or don’t tell her, it makes no difference to me.

b . I know you don’t want me to invite him but he’ll be offended if I don’t.

Oh, very well. Do invite him then. If you must.

c . I’m doing my best not to make any mistakes.

Do make one, so what? It doesn’t matter that much.

The parallels between the function of do in the example pairs (48)/(51) and (49)/(50),

which illustrate the contrastive polarity function and emotive emphasis function

respectively, suggest that there is a unified meaning to the uses of do in the two clause

types. The syntactic assumption that most directly captures this claim is that the same

syntactic element in both cases contributes the meaning identified in (47). In order to

consider how this would be implemented in each of the two analyses of imperative do, we

first need an account of finite do. In declarative sentences, do is standardly taken to be

inserted into an inflectional head in order to support a tense morpheme which has been

stranded from the verb due to the intervening presence of a zero emphasis morpheme

(Chomsky 1975). The emphasis morphology which forces the insertion of do is itself

located in an inflectional projection. This follows since it is only the modals or finite

auxiliaries, elements that are in 1°, that can bear this characteristic emphasis (Quirk et. al.

1985). This is demonstrated in (52) through (54). The two functions of emphatic

affirmation, polarity contrast and emotive emphasis, are illustrated in (52) and (53),

respectively, in association with modals/finite auxiliaries.


316

(52) a. Won’t you try again?

Yes, I WILL try again.

b . Despite what she claims, Jed CAN fix the tractor.

c. You must speak to the teacher.

I HAVE spoken to her.

d. You don’t look sorry.

Well, I AM.

(53) a. I AM glad!

b. You HAVE done well!

c . Mary WILL be pleased with this development!

Examples with main verbs bearing this emphasis, (54), do not have the same interpretation.

What is emphasized or affirmed in this case is the nature of the verb itself not the polarity of

the clause. For example, (54)a does not affirm the statement that Jed can fix the tractor, as

(52)b does; rather, it states that what Jed can do with the tractor is fix it. Similarly, (54)b

asserts how the speaker has interacted with someone; it does not assert whether such an

interaction has taken place, which is what (52)c does.

(54) a. Jed can FIX the tractor.

b. I have SPOKEN to her.

c. I am GLAD.

For the sake of concreteness, assume the analysis of Do-Support in Bobaljik 1995 (an

updating of the traditional Chomsky 1975) and take the emphasis morpheme to be the head

of EP, the affirmative counterpart to NegP occupying the same structural position as NegP:

between BP and VP (Laka 1990). I will call it EMPH. Semantically, this head is the locus

of the emphatic affirmation described above. Phonetically, it is realized as heavy stress.


317

Syntactically, it blocks adjacency between morphology in 1° and the main verb, triggering

the insertion of do. The crucial observation is the following: if the emphasis morpheme is

present then an element in 1° is required. The behavior is fully parallel to the result when

Neg° not is present. The above examples have the structure in (55) in which the emphasis

morpheme has moved to associate with the head in 1°.

(55) IP

subject I’

I ZP

doImodallauxiliary Z VP+EMPH; |

Returning to imperatives, we can capture the observation that do has the same semantic

function in imperatives and finite clauses by hypothesizing that the same morpheme,

EMPH, is present in both cases. Under the auxiliary analysis, do's appearance is motivated

and accounted for in exactly the same manner as in (55): the presence of EMPH results in

stranded inflectional morphology. We can assume that either an imperative morpheme (see

(26) in section 3.2) in 1° or the emphasis morpheme itself requires a bearer. If do in

imperatives is an inflectional head, as claimed in the auxiliary hypothesis, it fulfills the

requirement that 1° be overt when EMPH is present. Do supports the stranded morphology

exactly as it does in declarative clauses and the generalization regarding the semantic

contribution of EMPH can be maintained.

If do is an imperative particle on the other hand, it is difficult to see how one could maintain

that the emphatic affirmation meaning is a consequence of this same EMPH morpheme.

Primarily, 1° is not filled as seems to be necessary. Even if this requirement could be


318

voided, it is unclear how one could obtain the implicational relationship between an internal

emphasis morpheme and the appearance of do, an IP-adjoined particle. The analytical

problem is how to trigger the occurrence of the imperative particle just in case ZP headed

by EMPH is present. The relationship, illustrated in (56), is not one of subcategorization,

spec-head agreement, government, or even being in the same checking domain. It

resembles no recently proposed syntactic relation.

(56) IP

— » ImpP ip

Z VP

EMPH imperative verb phrase

Under a particle analysis, one would instead give up on the idea that emphatic affirmation is

always syntactically encoded by EMPH and locate the semantic contribution in the

imperative particle do itself. In this case, however, the parallel with finite do becomes

accidental. Beyond a single diachronic source, there would be no immediate reason why

the semantic contributions of imperative do and finite emphatic do should converge. The

imperative particle analysis, then, can account for the meaning of imperative do only by

abandoning a desirable generalization about the syntax-semantics mapping.

Although I have illustrated the argument with a specific, structurally-based analysis of Do-

Support, namely Chomsky 1975/Bobaljik 1995, any analysis that formally recognizes this

link between do and emphatic affirmation would lead to the same conclusion. The semantic

parallel between the two clause types does not depend upon the details of the formulation of


319

Do-Support so no claims about its correct formulation have been made. This section argues

only that imperative do is an inflectional head.

To summarize, I have suggested that emphatic do makes the same semantic contributions in

imperatives and finite clauses and I have located this in a single emphasis morpheme within

the inflectional layer of the clause. The presence of this morpheme requires a filled 1° and

for imperatives this is satisfied only by the auxiliary analysis of do. An imperative particle

approach does not capture the desired generalization.

4. Closing Arguments

In the chapter thus far I have drawn on evidence from VP ellipsis, parallels and contrasts

with let ’ -constructions, and the meaning of emphatic do to support the position that the

imperative markers do and don’t are inflectional heads and not pre-clausal imperative

particles. Nevertheless, this latter position has been explicitly adopted by at least two

works, Cohen 1976 and Zhang 1990. In closing I turn to consider their arguments and

offer counterarguments.

4 .1 . Zhang 1990

Zhang 1990 gives three arguments for taking don’t to be an introductory imperative

particle. The structure of each argument is the same: interrogatives and imperatives behave

differently with respect to a given phenomena and this argues that they should receive

distinct syntactic analyses. The imperative particle analysis does provide different structures

to imperatives and interrogatives while the auxiliary analysis of do incorrectly gives them

fundamentally similar syntactic structures.


320

The first putative difference centers on the contrast in acceptability of topicalization in

imperatives versus interrogatives. It is possible only in the former

(57) a. That present, don’t you open until next week!

b . *That classic novel, can’t you read by next week?

The contrast suggests a structural distinction which Zhang asserts can be readily captured if

imperative don't is a particle. A more careful consideration of the facts in chapter 5

indicates that topicalization is in fact possible in both clause types. Specifically, it is also

available in questions, (58) (an observation made in Radford 1988 and Rochemont 1989).

The contrast thus disappears as does the argument against analyzing don’t as an auxiliary-

like element, different from interrogatives.

(58) a. Those kind of men, aren’t you tired of going out with?

b . This kind of antisocial behavior can we really tolerate in a civilized society?

Zhang’s second difference concerns the possibility of phonetically contracting you with a

preceding don 't to yield dontcha in each of the two cases. This is only possible in

interrogatives, (59), not in imperatives, (60).

(59) a. Don’t you wanna go now?

b. Dontcha wanna go now?

(60) a. Don’t you hit me!

b . ??Dontcha hit me!

Zhang takes this contrast to indicate a structural difference between (59)a and (60)a; “don't

is not in an adjacent relation to the subject (Zhang 1990:76)” in the imperative to permit

contraction but it is in the interrogative. A careful investigation of the phonetics, however,


321

indicates that this is misguided. What is impossible is not the contraction but the

destressing and concomitant phonetic reduction of you.

Both (59)a and (60)a may be pronounced in careful speech as [doOnt ju]. In casual speech,

rules of assimilation and reduction apply, changing and eliminating phonetic material. Such

low level phonetic processes are not known to be sensitive to lexical or syntactic

information; their hallmark is that they are exceptionless and apply across the board. This in

itself would make Zhang’s claim surprising. Nevertheless, consider what is happening

above. In the contraction of interest, the alveolar stop [t] at the end of don’t may assimilate

with the palatal glide [j] beginning you to yield the palato-alveolar affricate [tfl. This is

possible in both (59)a and (60)a, yielding dontchoo [doOntfu]. Crucially, the assimilation

that takes place across the word boundary is not ruled out in either case and this eliminates

any basis for claiming that there is a structural distinction between two clause types. The

difference between (59) and (60) that is reflected by the contrast in the b examples is the

possibility of reducing the vowel in you from the full form [ju] to the unstressed form [ja].

This is permitted only in the interrogative, yielding dontcha [doOnt/a] for (59)b alone.

Destressing of you when it is an imperative subject is impossible, which is all that Zhang’s

example (60)b shows. This restriction has nothing to do with don't, however, as the

neutral imperatives in (61) demonstrate. Reduction is equally impossible here. Zhang

suggests in a footnote, following a formal account in Akmajian 1984, that it reasonably has

a prosodic explanation: imperative subjects may not be stressless. I conclude that the

contrast above is not relevant.

(61) a. You [ju/*ja] eat those peas right now, got it?!

b . You [ju/*ja] get out of here! We’re not finished.


322

Lastly, Zhang 1990 draws on the putative contrast in (62) and (63) to demonstrate that

don’t is not an ordinary inflectional head. Imperative don’t has a “bizarre requirement of

obligatory accusative case marking on an ordinary subject (Zhang 1990:76)”, (62), which

does not exist in interrogatives, (63).

(62) a. Don’t you and him/them/her fight again!

b . *Don’t you and he/they/she fight again!

(63) a. Won’t you and him come to my party?

b . Won’t you and he come to my party?

Platzack and Rosengren 1994 admit that the judgments in (62) are not widely shared.12

Nevertheless, assume the facts to be as indicated—the subject of negative imperatives must

be in the accusative case. In the subject position of interrogatives, (63), there is apparently

a certain amount of freedom. Conjoined pronouns may be either accusative or nominative.

Without an analysis of the pattern in (63), it is unclear what to make of the contrast with

(62). What is clear, however, is that in the interrogatives the accusative pattern, (63)a, is

the marked case. The subject position of an interrogative is unquestionably assigned

nominative case: Did he/*him leave? Assume as a descriptive solution that conjoined

pronouns may always appear in the accusative, the default case in English,13 regardless of

the case that is structurally associated with the position that the conjunction occupies. (63)a

12Jorge Hankamer (personal communication) indicates that there are registers of certain dialects in which pronouns in conjoined NPs are always accusative, independent of clause type or structural position. I do not know whether this is the same dialect that Zhang documents.13This can be seen in left dislocation examples, where a dislocated pronoun is always accusative regardless of the case of its coindexed element:

(i) a. Him/*He, I can’t stand him. b. Me/*I, I can’t stand him.


323

instantiates this observation and (63)b is the normal pattern of nominative case on a finite

clause subject. The imperative subject in (62)a also potentially represents this pattern. (62)b

must then be ungrammatical because the imperative subject position is not assigned

nominative case. Analytically, the specifier of IP in imperatives is either a Caseless position

or is assigned accusative Case. If this is the theoretical point of the above data, it is no more

easily accounted for with an imperative particle analysis than with an auxiliary analysis.

Precisely the same situation arose with let ’ -constructions in section 3.2 and was seen to in

fact argue for an auxiliary analysis. Zhang’s patterns of case assignment thus do not

provide a convincing argument for the imperative particle status of don't.14

4 .2 . Cohen 1976

Cohen 1976 offers more challenging objections. That work and numerous other

researchers refer to the impossibility of subjects with imperative do as evidence against a

productive auxiliary analysis of the form. As I have stated earlier, however, I follow

Davies 1986a in taking such examples, repeated below in (32), to be marginal for

pragmatic reasons and not strictly ungrammatical. My position is that they are assigned a

syntactic structure but their use is governed by non-syntactic factors (see chapter 3, section

4.2.3). I thus do not consider them a serious impediment to analyzing do as an inflectional

head.15

^ O n e would like to know the judgments on similar non-negative imperatives, in (i). If they are as indicated (I have not located speakers of the dialect that Zhang documents), it is once again the case that the structural analysis of don't is not the source of the contrast.

(i) a. You and him be on time just this once!b. *You and he be on time just this once!

15For reasons developed in chapter 3, quantificational and second person subjects are best, just as in neutral imperatives.


324


b. Do EVERYone give it a try!


d. For heaven’s sake, of all people, DO YOU give me some support!

Cohen discusses two other sets of data which are indeed problematic for an auxiliary

analysis of don’t. The first centers on the existence of dialects, primarily British, that

permit the pattern in (65), the don 't-let ’ -construction.

(65) Don’t let’s eat it right away!

This instance of don’t is unlike imperatives with don’t and. First, there is no corresponding

version without contraction:

(66) a. *Do not let’s go in there!

b . *Do not let’s eat it right away!

Second, if a subject appears, it must follow both let’s and don’t:

(67) a. (?)Don’t let’s you and me worry about that!

b . *Don’t you and me let’s stay up too late!

Third, this don’t does not license ellipsis, (68), although let’s still does, (69).16

16Huddleston 1978:49 notes exactly parallel behavior for do in a do-let’s construction:

(i) Let’s go and see John.♦Oh, yes, do 0!

0 = let’s go and see John

(ii) Let’s go and see John.Oh, yes, do let’s 0!

0 = go and see John


325

(68) Should we tell him?

Oh, let’s tell him.

*No, don’t 0!

0 = let’s tell him

(69) Should we go in the haunted house?

?Don’t let’s 0!

0 = go in the haunted house

The construction behaves like a /ef’j-construction with don’t appended to the front. For

dialects in which this pattern is possible, don’t is likely an imperative particle or higher,

non-inflectional head. An auxiliary analysis of this don’t seems incorrect. In particular, I

argued in section 3.2 that let’s is an auxiliary that originates in 1° and so don’t here cannot

also do so.17 The dialects that permit don 't-Zer’s-constructions apparently admit both

analyses of don ’t: a particle analysis to account for the don’t let’s data in (65) through (69)

and an auxiliary analysis to account for the earlier negative imperative VP ellipsis facts.

Some imperatives in these dialects will potentially have multiple derivations, a proposal put

forth in Hankamer 1977. It is not clear that this conclusion should really be bothersome.

Hankamer 1977 points out other syntactic phenomena where two distinct analyses are

possible, and required, to account for the full range of facts. It is not even clear that one

must claim that multiple analyses is the marked case. The long-standing uncertainty

regarding the analysis of imperatives may be a reflection of this subtle dialectal variation

and syntactic ambiguity. For those speakers who do not accept examples like (65), an

imperative particle analysis remains unavailable.

17Seppanen 1977 suggests that don’t le t’s could be analyzed as a single auxiliary unit.


326

Cohen’s second set of problematic data are imperatives with do not. These clauses have,

almost without exception, proved troublesome for syntactic analyses of imperatives and

have prevented reasonable analyses within the larger context of a grammar of English,

since Chomsky 1975. They are the subject of chapter 6. Here, I will only introduce the

facts and indicate what is at stake.

Conventional wisdom has it that don't is a contraction of do not, at least in finite clauses. In

the uncontracted case, (70), do acts like the auxiliary, inverting in interrogatives, leaving

negation behind, (70)b. Only in exceptional instances (Zwicky 1969, Gazdar, Pullum, and

Sag 1982, Zwicky and Pullum 1983, Zwicky 1987) may they both invert, (70)c.

(70) a. They do not ever help.

b . Do they not ever help?

c . ?Do not any of you who grew up on farms remember going cow tipping?

When there is contraction, don’t acts like the auxiliary and inverts hs a unit over the subject,

(71).

(71) a. They don’t ever help,

b. Don’t they ever help?

If imperative don’t were simply the same as finite clause don’t, the same behavior in its

uncontracted form should be seen. As is well-known, however, do nor-imperatives are

extremely restricted. No subject may ever appear with do not, either before or after not,

(72). The subject must remain unexpressed and the pattern in (70) cannot be replicated.


327

(72) a. Do not leave!

b. *Do | eVy ^ ne| not leave!

c. *Do not jevey ne} leave!

Imperative behavior diverges from that of finite clauses in that (70)b and c are grammatical

while (72)b and c are not. It is this contrast more than any other single fact which has

prevented don't from being given the same analysis in the two clause types. Chapter 6

explores the data in greater detail.

4 . 3 . Summary

In this chapter, I have been concerned with the syntactic status of imperative do and don’t.

I have attempted to show that a synchronic analysis in which they are modal-like

inflectional heads is necessary to account for their behavior. An analysis of do(n’t) as an

introductory imperative particle is not adequate for Modem English. Such an analysis

would require ad hoc statements about the syntactic and semantic behavior of do(n’t)\

whereas, taking do(n’t) to be an 1° head allows its behavior to be largely subsumed under

more general English syntactic phenomena and processes.

Appendix. Imperative-Like Constructions with Let

One can identify three imperative-like constructions involving some form of let which have

potentially distinct syntactic and semantic characteristics. They are illustrated in (73).

(73) a. Let him make a few mistakes.

b . Let those who think I have it easy try living on my salary for a while.

c. Let’s go to the beach.


328

Clark 1993 labels them /er-imperatives, /er-constractions, and /ef's-constructions,

respectively. It is the lattermost which is investigated above. In this appendix, I argue that

the construction in (73)c is distinct syntactically from at least the construction in (73)a,

making it worthy of investigation on its own, as is done in section 3.2.1 discuss the syntax

and semantics of /ef-imperatives, (73)a, and demonstrate that let ’ -constructions, (73)c, are

distinct in both domains. I remain neutral on the syntax of more formal /ef-cons tractions, in

(73)b, allowing the possibility that they are more similar to either (73)a or c.

In the first case, let is a main verb with the meaning ‘allow’ or ‘permit’ which takes a small

clause complement headed by a verb in its bare form (Iveland 1993). (74) contains

instances of this let in finite clauses.

(74) a. You should let Bill explain.

b . Unbelievably, they let their children do that in public.

Examples with this let are ordinary second or third person imperatives that have been the

object of this investigation. This can be seen by the subject’s interpretation, its formal

options, and the expression of negation. When there is no overt subject, (75)a, the

understood one is the pronoun you, as discussed in chapter 3. The understood subject is

not the post-/er NP. This is confirmed by the form of the tag question, which must

obligatorily contain you in (75)b, and the required reflexive pronoun, in (75)c, which

agrees with the null second person pronominal subject. In (76), an overt subject is also

perfectly possible. It precedes let, as would be expected if let were simply the main verb of

the clause. Lastly, (77) illustrates that this main verb let is negated with don’t or do not like

normal imperatives.


329

(75) a. Let the cat go now!

b . Let John come in, willyou♦we

♦John

c. Letyourself

yourselves♦you

relax!

(76) a. You let me do this myself, you hear!

b . Nobody let her hear the message on the machine!

(77) a. jj3onot} et ^Y 0116 know about the new plans!

b . ♦Let not anyone know about the new plans!

In short, /ef-imperatives with the meaning of ‘permit’ or ‘allow’ are ordinary imperatives

with the main verb let. They have a structure exemplified by (78) for (75)a, also explicitly

proposed by Seppanen 1977.

(78)

pro I

let the cat go ‘allow’

In the second use, the ^/-construction, let is optative in nature and means roughly ‘it shall

be the case that’ or ‘may’. This construction is formal and archaic for some. Its use is

illustrated in (79).18

18Here and below, in illustrating the /ef-construction, I attempt to use examples that do not have, or disfavor, the /er-imperative interpretation. Some of the examples may, unavoidably, be ambiguous.


330

(79) a. Let no one think that a teacher’s life is easy.

b . Let each man decide for himself.

c . If anyone would like to object, let him speak now.

d . ‘Let me not to the marriage of true minds admit impediments.’

(Shakespeare, Sonnet 116)

Building on Davies 1986a, which suggests that the semantic analysis of /ef-constructions is

the same as that of imperatives, Clark 1993 summarizes the meaning of a /ef-construction:

“to utter a /ef-construction with prepositional content P is to communicate that P represents

a thought entertained as a description of a potential and desirable state of affairs”. In

Clark’s and Davies’ accounts, let makes no contribution to the logical form of an utterance

of a /ef-construction. “Let has no semantic conceptual content (Clark 1993:190)”. For

example, (79)a represents a state of affairs in which no one thinks that a teacher’s life is

easy. The semantic difference between /ef-constructions and /ef-imperatives can be isolated

in the lexical semantics of the respective forms of let. Main verb let has an external

argument, underlined in the argument structure in (80), which permits, or otherwise

somehow allows, the situation encoded by the VP internal argument.19 In a /ef-imperative,

then, there are two events corresponding to the two verbs: the event of permitting

associated with let and the second event associated with the VP small clause.

(80) let, V [ VP] ‘permit, allow’

fil 02

19Iveland 1993 claims that let also must be allowed to have an object control subcategorization frame:[ NP VP] in which the NP complement of let controls the subject of the VP small clause. I will notbe concerned with such details here. This argument structure still contains an external argument


331

In the /ef-constructions illustrated in (79), optative let does not have an external agent. This

let seems to mean the something similar to the main verb in (80) without the requirement or

possibility that there is an external causer. Consequently, there is only one obvious event,

encoded by the VP small clause.20 Semantically, main verb let differs from optative let in

the presence of this additional thematic role.21

With regard to the syntax of optative let, this semantic distinction could be reflected as

simply the lack of an external argument. Let would still be a main verb with the internal

subcategorization frame in (80). Davies 1986a and Clark 1993 in fact assume that let-

imperatives and let-constructions are syntactically identical, modulo some difference in the

external argument apparently. Optative let is a main verb. Alternatively, this semantic

distinction could be an indication of a more significant syntactic difference in which

optative let is not a main verb but a modal-like inflectional head. Seppanen 1977 argues that

optative let is precisely an auxiliary, pointing out putative syntactic differences between the

two sentence types and similarities between optative let and traditional modals. I will not

decide between these two syntactic analyses of /ef-constructions since the data are marginal

and variable in many areas and the result would not directly impact the material under

investigation. I will, however, look at the syntax of less formal let '^-constructions and

2°A close semantic parallel is the relationship between the transitive and unaccusative uses of verbs such as sink or melt. In the transitive use, (i)a, there is an external agent while in the unaccusative use, (i)b, no agent is implied.

(i) a. The navy sank the ship,b. The ship sank.

21 In practice, this difference is often subde as there is a use of the main verb let in which the subject need not be agentive and the external argument takes on a less direct role, in (i).

(i) a. The rains let our crops flourish.b. The 1984 political climate let Reagan take the White House.c. The new freeways let us get to work in record time.


332

argue that they are syntactically distinct from /ef-imperatives, in agreement with Seppanen

1977. Below I discuss four differences between the two constructions and demonstrate

how the analyses in the text account for the facts.

The first difference is the possibility of clause-initial subjects alluded to earlier in (76).

Such subjects are permitted with /ef-imperatives when the object is us, (81), and the

examples have the expected structure similar to (78). Let '^-constructions, in contrast,

absolutely do not permit such subjects, (82).22

(81) a. You let us publish this report, or else!

b . Nobody let us eat that leftover cake!

(82) a. *You let’s publish this report!

b . *Nobody let’s eat that leftover cake!

The ungrammaticality of (82) is accounted for with the text analysis by having the auxiliary

let’s identify a null pronominal in the subject position of these examples, which follows

let’s. The subjects in (82) are in an illicit syntactic position and have illicit agreement

features for the auxiliary.

The second difference between let ’ -constructions and /er-imperatives concerns the possible

form and interpretation of the first person plural pronoun us. The first person pronoun us

22The data demonstrate one unexplained difference between do(n't) and let's: le t’s obligatorily raises to the higher head X, while do(n’t) does so only optionally (see section 3.3.1). The /er’s-constructions in (i) with the subject preceding le t’s are ungrammatical. I have no explanation for this fact although the result can be mechanically achieved using the feature checking analysis developed for do(n ’t) in chapter 6.

(i) a. *Us let’s go instead!b. *You and me let’s get out of here!c. *A11 of us let’s send him a card!


333

only appears contracted in the ZefVconstruction case. The contracted form in a let-

imperative, or any other situation, is robustly ungrammatical. (83) demonstrates this for the

finite use of the main verb.

(83) a. They always let us consider several options,

b . *They always let’s consider several options.

It can also be demonstrated for /ef-imperatives. Consider the imperative in (84)a (the

presence of the second person tag ensures that this is an imperative). (84)b confirms that

contraction is impossible. The example is grammatical without the second person tag, in

(84)c. (84)c, however, does not share the same meaning as (84)a. It is the distinct let’s-

construction.

(84) a. Please let us consider all the options, will you?

b . *Let’s consider all the options, will you?

c . Let’s consider all the options!

Let’s in let ’ -constructions, then, cannot be derived from /ef-imperatives since the

contraction is not productive. Furthermore, as Davies 1986a points out, the putative

contracted and uncontracted pronouns have distinct references. In (84)a, us is first person

plural exclusive (lpl-excl). It refers to the speaker and other persons not including the

addressee. In (84)c, on the other hand, the contracted pronoun 's is first person plural

inclusive (lpl-incl). It must refer to the speaker and addressee, and others potentially. In

other words, lpl-incl and lpl-excl are in complementary distribution in the two

constructions. Semantically, there is also no source for the contraction in (84)c.

A third difference is seen in the options for question tags, pointed out in Seppanen 1977.

He observes that /er-imperatives are tagged with elliptical questions whose subject is you,


334

(85), while le t 's-constructions must have third person tags, (86). Third person tags are

impossible with /ef-imperatives and second person tags are ruled out with le t’s-

constructions (examples from Seppanen 1977:529):

c. Let’s you and John and me do it together, shall we?

The data are transparently accounted for under the assumption that the subject of a question

tag reflects the subject of the tagged clause. The two analyses being defended here assign

second and first person features to the subjects of fef-imperatives and let ’ -constructions,

respectively. A closer look at the interpretation of the tags also reveals a predicted structural

distinction between the two clause types. In the /er-imperative in (85)a, the interpretation of

the tag is ‘won’t you let us help you?’, with the missing material being let us help you,

including let. In the /er's-construction in (86)a, the tag is interpreted to be equivalent to

‘shan’t we help him?’, where the missing material is help him, not let's help him.

Assuming that tag questions are derived via VP ellipsis and the elided material is a VP, we

are led to conclude that the VPs in the two respective examples are let us help you and help

him, respectively. In the former case, let is a verb, the head of the VP, while in the latter it

is not. We thus converge on the analytical claim that main verb let is a verb but let’s is not.

The fourth and last difference between the two constructions is their interaction with

sentential negation. In the majority of English dialects, the two are negated in the manner


(85) [won’t youl [ *shall we J

b . Don’t let John keep the original, will you?

(86) a. Let’s help him, |r*

b. Let’s go now, \ „

*won’t you shan’t we

335

indicated below. In (87) and (88), /ef-imperatives are negated only with don’t and never

not.

(87) a. Don’t let him move out of your sight!

b . *(Not) let (not) him (not) move out of your sight!

(88) a. Don’t you let them get away with that!

b . *You (not) let (not) them (not) get away with that!

Let ’ -constructions, in contrast, can always be negated with not, (89).

(89) a. Let’s not be late!

b . Let’s you and me not quit just yet!

c. Let’s not invite Wanda!

The two cases receive the structures in (90)a and b, which serve to summarize the two

analyses proposed for /er-imperatives and let ’ -constructions, respectively. The structures

illustrate the parallel analyses given to don't and let's and the distinct analyses given to

main verb let and optative let's.

(90) a.

I Neg: pro tj[2sg/pl]

let DP

let’ Sj DP

don’t:

pro tj [lpl-incl]

be late

be late


336

Chapter 5

The Phrase Structure of Imperatives

1. Introduction

This chapter concludes the investigation of the syntax of non-neutral imperatives begun in

the previous chapter. The focus is on two basic questions: 1) what is the number and

identity of phrasal projections above VP in imperatives and 2) what is the position of the

two elements, the subject and do(n’t), within these projections. The chapter argues for

multiple functional projections above VP, either an exploded Infl analysis as developed in

Pollock 1989 or a unitary Infl (Chomksy 1986) dominated by a C(omplementizer)

projection. This makes imperative syntax remarkably like that of finite clauses, indicative

and interrogative sentences, and ultimately challenges claims that the imperative is an

idiosyncratic construction which merits unusual analyses and assumptions.

The results of chapter 4 serve as the departure point. There it was argued that non-neutral

imperatives with emphatic do or negative don’t, collectively DO, have the structure in (1),

with the head movement shown being optional.


337

(1)

DO:


With respect to this phrase marker, there are two specific questions that were subordinated

earlier: 1) what is the identity of the XP projection and 2) where exactly is the subject

located. In formulating potential answers to these questions I will temporarily step back

from (1) in some respects in order to be able to consider a wider array of theoretical

options. Ultimately, I accept an analysis that returns to and adopts the specifics of (1) with

XP replaced by the familiar complementizer phrase, CP.

The hypotheses to be presented all explicitly address the full range of imperative data that

has been encountered up to this point. The data are compiled in (2) through (5) where it is

seen that both orderings of subjects with do and don't must be permitted. Imperative

subjects may precede or follow do(n’t) in the right contexts. I follow Davies 1986a in not

taking any of the patterns to be strictly ungrammatical.






b. Don’t you misbehave while we’re gone!

0 0 ASUBJECT

SUBJECTADO

DOW’rASUBJECT


338

(5) a. Those with luggage don’t leave it unattended! SUBJECTA£>CWT


Section 2 presents three potential phrase structural analyses. They are alike in their ability to

account for the imperative facts above. Sections 3 and 4 present evidence to decide between

the competing hypotheses. In section 3 ,1 present four arguments against a VP-intemal

representation of the imperative subject at surface structure. The data indicate that

imperative subjects are not seen in a VP-intemal specifier at the surface. Section 4

considers the remaining two analyses in which the subject is outside VP, within some

functional projection. With this restriction, two projections above VP are required to

account for the imperative word-order facts. The data are inconclusive regarding the

identity of these projections, however. All that can be determined is that imperatives contain

a verb phrase dominated by two functional projections. Simplicity suggests adopting a

representation involving CP and IP, making imperative structure parallel to finite clause

structure in important ways. The proposed answers to the above two questions are that 1)

XP in (1) is identified with CP and 2) imperative subjects are located in the specifier of IP.

Section 5 discusses some theoretical issues and summarizes the similarities between

imperative phrase structure and standard conceptions of English clause structure.

2. The Hypotheses

This section outlines three phrase structure proposals concerning the identity of the phrasal

projections that make up the imperative clause. Given the previously justified assumptions

in (6) about the position of subjects and DO, from chapters 3 and 4 respectively, along with

the X schema in (7), the subject imperative data in (2) through (5) require the presence of

at least two specifiers and/or two heads above the verb. This is necessary to accommodate

both orderings of DO and the subject.


339

(6) 1) subjects occupy specifier positions, spec,XP

2) DO is an inflectional head, X°

(7) XP

Sp

YP

These options are explored below. The focus of the discussion at this point is on

determining the correct phrase structure, and only secondarily on providing a complete

analysis of what would motivate the various word order options. This is addressed in

section 5.

2 .1 . The VP-Internal Subject Analysis

The VP-Internal Subject Hypothesis (ISH) (Kuroda 1988, Koopman and Sportiche 1991)

claims that subjects originate in a position inside VP and raise in the syntax to a higher

position among the inflectional projection(s). They are not generated in the specifier of IP

but move there, potentially occupying various intermediate positions along the way. Taking

the VP-intemal site to be the specifier of VP permits an analysis of the English imperative

data which does not require an additional projection above IP. Structurally, this is the least

elaborate account available. The subject may remain in situ, in the VP, where it will follow

DO located in 1°. This yields the DOAsubject ordering. The alternative order subjectADO

has a structure identical to that found in finite clauses where the subject has raised to the

specifier of IP position. The structures for the two orderings of DO and an overt subject are

in (8), where the subject may appear in one of the parenthesized positions.


340

(8) IP

(subject) V

I VP

DO (subject)


Such an analysis captures the long-standing intuition that imperatives are fundamentally

verb phrases. Bolinger 1967, Downes 1977, and Jespersen with other traditional

grammarians relate the English imperative to the infinitive. The VP analysis of imperative

phrase structure also readily explains why English imperatives lack overt verbal

morphology and agreement: in the non-emphatic/non-negative cases, they are simply a VP

with no inflectional projections. The fundamental empirical claims of the VP analysis are

that 1) imperative subjects may appear overtly inside the verb phrase and 2) there is at most

one inflectional projection dominating VP.1

An alternative structure to (8) would place the subject and DO outside the verb phrase at all

times. This would necessitate two projections above VP, rather than one. I develop this

alternative in two distinct ways below.

Generative proposals claiming that imperatives are VPs or other reduced clause types include Schmerling 1977,1982 and Platzack and Rosengren 1994. Henry’s (1995) recent analysis of the imperative in a Belfast dialect of English profitably employs a VP-intemal representation for the subjectPlatzack and Rosengren 1994 also adopts a structure similar to the one above for imperatives in the Germanic languages. The work argues that these subject-like noun phrases are syntactically and semantically distinct from ordinary subjects although they are subjects in some sense. The work observes that second person pronominal subjects in imperatives in numerous languages show properties which would be unexpected if they were simply ordinary syntactic subjects. Platzack and Rosengren suggests that the argument of the imperative verb is nevertheless syntactically represented. This argument, which they characterize as an actor/addressee, is base-generated in VP and, at all times, remains there because there is no higher agreement/inflectional projection above VP in imperative clause structure for it to move into. As an actor/addressee argument rather than a subject, it does not have all the syntactic characteristics of a subject— in particular, the need to check syntactic features.


341

2 .2 . The CP Analysis

Beukema and Coopmans 1989 adopt the structure in (9) for English imperatives. Although

they are analyzing a more restricted body of data, their structure is sufficient to capture the

facts introduced here. In this analysis, there are two projections above VP: IP and CP. The

subject is always in spec,IP as in finite clauses. The free ordering of DO and the subject is

obtained by movement of DO. Do(n’t) is inserted into 1° from where it optionally raises to

the available head position C° over the subject in the specifier of IP.

(9) CPI

C

(DO) subject I’

I ^ ^ V P

(DO) imperative verb phrase

The two orderings of DO and the subject receive the analyses schematized in (10).

(10) a. [jp subject [r DO [yp. . . ] ] ] ]

b . [cp [a DOj, [jp subject [p t^ [yp. . . ] ] ] ] ]

(10)b is derived from (10)a via I°-to-C0 movement, as in interrogatives. The analysis

capitalizes on the apparent similarity between negative imperative and interrogative

structures and is fundamentally Chomsky’s (1975) original analysis of imperatives as

involving Subject-Auxiliary Inversion. Such an analysis is also assumed in Emonds 1970

and Stockwell et. al 1973. The primary claim of the CP analysis is that CP is active in the

syntax of imperatives and C° is a potential landing site for DO.


342

2 . 3 . The Articulated Infl Analysis

Like the CP analysis, the articulated Infl hypothesis also proposes two projections above

VP. It capitalizes on recent theorizing that IP is not a single projection but that it is phrase-

structurally complex (Pollock 1989). Current research suggests that there are at least two

separate X-compatible functional projections, typically labeled agreement (Agr) and tense

(T), between CP and VP which replace the unitary IP of earlier work. Although

imperatives in English, and cross-linguistically, are associated with a dearth of inflectional

morphology, the null hypothesis would be that imperatives could contain the same

functional projections that other clauses do. Not implausibly, imperatives could also make

use of these intermediate functional projections in ways that declarative sentences do not.

The structure in (11) permits the correct word order variation between the subject and DO

without invoking the VP-intemal subject position or CP. FjP and F2P are two functional

projections that replace IP.

(11) FjP


I leave the identification of the inflectional projections undetermined. They will simply be

labeled Fj and F2 to indicate two functional projections, neither of which is VP or CP. To

obtain the surface order DOAsubject in (11), DO must be in Fj and the subject must be in

the specifier of F2P. The ordering subjectADO, however, without a full analysis of the


343

identification and nature of the projections, is underdetermined. If the subject is in the

specifier of FjP then DO may be in either head position. If the subject is in the specifier of

F2P, DO may still be in F2 and give the desired word order. Pollock 1989 adopts (11) for

the structure of neutral imperatives and some emphatic imperatives with do. Although he

does not consider the option of overt subjects in imperatives, his structure can

accommodate them. Following Belletti 1990 and Chomsky 1991, we could identify FjP as

AgrP and F2P as TP.

I will in general refer to these three analyses as the VP, CP, and FP hypotheses,

respectively. I argue below that the FP or CP analysis is the preferred view of the phrase

structure of the English imperative. Specifically, the evidence indicates that the subject and

DO are always outside of VP, excluding the VP analysis.

3 . Evidence Against the VP Hypothesis

This section presents evidence against the VP analysis. The position of subjects in

imperatives with stacked auxiliaries, the appearance of subject-oriented floated quantifiers,

the placement of VP-adjoined adverbs, and the phenomenon of VP ellipsis all indicate that

the overt imperative subject cannot be inside VP.

The VP hypothesis is unusual in that it attributes to the English imperative unique syntactic

behavior not found in other areas of English syntax. In particular, the position of the

subject is claimed to be interestingly different from that in ordinary clauses. In a theory in

which movement of the subject to spec,IP is enforced by something like the Extended

Projection Principle (EPP), we would have to isolate an element in the imperative

construction that suspends EPP effects. Two characteristics of the imperative are

suggestive in this regard. First, the absence of EPP effects in imperatives could be


344

understood as a property of the imperative subject itself. Imperative subjects are often of

reduced importance because the context typically transparently indicates who the subject

is—the addressee. The subject’s pragmatic prominence overshadows and could allow for

its syntactic non-prominence. Second, the lack of EPP effects could reasonably follow

from the nature of imperative inflection, which is absent or greatly reduced. Whatever it is

about English inflectional heads that drives the subject to fill its specifier, it is not present in

the impoverished English imperative morphology.

Under this latter view, morphological properties inherent in the imperative account for its

unique behavior. Within a Minimalist Program framework (Chomsky 1993, 1994, 1995),

we could implement this by saying that the DP-feature of the inflectional projection that

requires movement of the subject is weak. Consequently, the imperative subject is able to

remain in the specifier of VP and moves into the higher inflectional specifier for checking

only in the covert part of the derivation. Henry (1995) profitably uses just such an

approach in analyzing imperatives in a Belfast dialect of English. She convincingly shows

that in the Belfast English examples in (12) the subject you is in the VP-intemal position. It

is below a shifted object pronoun, (12)a and c, and a VP-adverb, (12)b and c.

(12) a. Give it you to the teacher! BELFAST ENGLISH

b . Write carefully you that letter!

c. Tell them always you the truth!

It can remain in situ under Henry’s analysis because the DP-feature of imperative inflection

in the dialect is optionally weak.

In contrast, in a theory such as earlier versions of the Minimalist Program (Chomsky 1993)

in which movement of an XP is driven solely by its own morphological properties (the


345

principle of Greed in Chomsky 1993), it would need to be properties of the imperative

subject itself that result in its remaining inside VP. With respect to its surface position,

chapter 3 argued that there is no syntactic distinction between imperative subjects and those

of other clause types. It is initially difficult to see where the morphological difference could

be isolated. If the VP analysis of the imperative were to turn out to be correct it would thus

support more recent versions of the theory (Chomsky 1995) in which movement is largely

motivated by needs of the target projection rather than the moving object.2 I present

empirical evidence against the proposal, below.

3.1. Multiple Auxiliaries

The ISH claims that a subject originates in the VP specifier of the verb of which it is an

argument. As Koopman and Sportiche 1991 points out, this has the theoretically appealing

result of making thematic role assignment by a lexical head X° a local process that occurs

entirely within the projection of X°. Since the subject must ultimately end up in the specifier

of IP, all intervening categories—auxiliary verbs and modals in 1°—become raising

categories and the specifiers of these intermediate projections are potential stopping off

points for the subject. The claim of the VP hypothesis is that this raising does not take place

in imperatives and that we see the subject in its base position. This makes predictions about

the position of the subject when auxiliaries are present. We expect to see it in its

VP-intemal position below DO and auxiliaries and next to the main verb which thematically

licenses it. I show below that this is the wrong result given Zagona’s (1988a) analysis of

VP phrase structure which has been adopted in this work. The data indicate that the subject

2Platzack and Rosengren 1994 develops such a view, although the proposal also eliminates the inflectional projections which the subject raises up into to check its morphological features. The work can be seen as appealing to both avenues discussed above.


346

is higher than this position. The VP hypothesis, but not the CP or FP hypotheses,

incorrectly predicts the placement of subjects in imperatives with auxiliaries.

Zagona 1988a following Ross 1969 argues for a stacked VP analysis of auxiliaries in

which each one is the head of its own VP. Reiterating from chapter 1, the structure for

Zack might have been waiting is (13) with a VP-intemal subject assumed. The subject

raises from spec,VP, through the intervening specifiers, to its surface location in the

specifier of IP.

(13)

Zack I

might t;

have t:

waiting

The VP hypothesis claims that imperatives with auxiliaries have the same structure except

that the subject does not raise to spec,IP and DO occupies 1° in place of the modal. Given

imperatives with auxiliaries as in (14) through (16), the hypothesis predicts the position of

the subject to be just to the left of the main verb. This is an incorrect prediction as (17) and

(18) show.


347

(14) a. Don’t be fooled by his shoddy argumentation! PASSIVE BE

b . Don’t be hurt by what he says!

c. Don’t be misled by his flattery!

d. Be seated!

e. Be reassured by me!

f . ‘Be represented by experienced lawyers!’ (radio advertisement)

(15) a. Do be working when we get home! PROGRESSIVE BE

b . Don’t be messing around when the bell rings!

c. ‘Don’t be looking for inexpensive things in Carmel!’

c . Be waiting on the comer at six!

(16) a. Don’t have eaten everything before the guests arrive! PERFECTIVE HAVE

b . Do have tried it before you begin to criticize!

c. For heaven’s sake have prepared the thing in advance!

d . Do have checked the facts before you start accusing people!

e. Please, do have made that call by six o’clock!

f. Do have given some thought to the question!

(17) a. *Don’t be anyone fooled by his shoddy argumentation!

b . *Do have at least you tried it before you begin to criticize!

c. *Don’t be you ringing that call button!

(18) a . Don’t anyone be fooled by his shoddy argumentation!

b . Do AT LEAST YOU have tried it before you begin to criticize!

d . ‘Don’t you be ringing that call button! ’ (flight attendant on American Airlines

484, May 19, 1995)

O


348

For the other two hypotheses in section 2, the subject is outside of the highest VP (see

structures (9) and (11)), which is supported by the empirical facts in (18). The subject

precedes the topmost auxiliary. The contrast between (17) and (18) is particularly clear.

Only the CP/FP hypotheses correctly account for the position of subject in imperatives with

auxiliaries.

The data indicate that the subject is not in its base-generated spec,VP position. The

examples are compatible with the subject still being within VP, however: the specifier of

the highest auxiliary verb phrase. Although one might pursue a weaker version of the VP

hypothesis in which it is claimed only that the subject is somewhere within VP, I will not

do this because the data below indicate that this too would be inadequate.

3 . 2 . Floating Quantifiers

Sportiche’s (1988) theory of floating quantifiers laid out in chapter 1 crucially takes

advantage of the VP-intemal subject hypothesis. In association with the VP, CP, and FP

hypotheses, it makes predictions regarding the possibility of quantifier float in imperatives.

The VP hypothesis in particular makes the prediction that there should be no quantifier

stranding in imperatives because the subject does not move out of its VP-intemal position.

It never moves, so stranding is not possible. The CP and FP hypotheses, in contrast,

predict the same range of quantifier stranding as found in finite clauses. This is because the

subject in each analysis raises out of its spec,VP base position to a higher specifier. A

quantifier can then be stranded in the specifier of the main verb phrase or any intermediate

verb phrase between spec,IP and the original position. The data in (19) indicate that the

prediction of the VP hypothesis is incorrect. Stranded quantifiers are grammatical in

imperatives, an issue first discussed in Schmerling 1977. In (19), a DP preceding a


349

modifying quantifier is grammatical as a subject. The CP/FP hypotheses give them the

analysis in (20). The quantifier is stranded in the base spec,VP position. The VP

hypothesis will not assign these examples a structure.3

(19) a. The twins both be here for the pictures!

b . Rhett, Scarlet, and Lassie all get ready for their next scene!

c. MY children all come right here, I won’t tolerate such misbehavior!

d . (The girls can all leave but) don’t the boys all go yet!

(20) DPj [w [Qp all/both tj ] [v , V . . . ] ]

The data in (21) make the same point in a somewhat different way. The noun phrase and

quantifier in the examples are not string-wise adjacent indicating that they cannot both be in

their base-generated positions according to Sportiche’s analysis. The VP hypothesis

predicts the examples to be ungrammatical. The alternative hypotheses give them a structure

similar to (20), shown in (22).

3It seems clear that such examples cannot involve vocatives rather than subjects since, as observed in the discussion o f floated quantifiers in chapter 1, strings like the twins both are not constituents, which vocatives must be. Furthermore, the noun phrases without the quantifier do not pass the tests typical for vocatives discussed in chapter 3: a separate intonational phrase and an inability to trigger third-person anaphora.


350

(21) a. Don’t you be both talking at once!

b . You be both waiting for me promptly at 3 !4

c. Don’t you have all made a mess!

(22) (don t) DP [yp tj [y» have/be [yp [qp all/both tj ] [y» V . . . ] ]

Using quantifier stranding as a probe for the VP-intemal subject position indicates that

imperative subjects cannot be VP-intemal at surface structure. At the same time, if

Sportiche’s analysis of quantifier float is correct, it supports the conclusion from chapter 3

that imperative subjects are like other subjects. They originate in the specifier of the main

verb and show the same derivational history. This further supports the contention of the

CP/FP hypotheses that imperative phrase structure does not necessarily differ from that of

other clause types in English.

4For some people, including me, examples with floating quantifiers following be do not always sound fully natural. They are marked ungrammatical in Sag 1980. The preferred position for the quantifier is preceding be. This does not necessarily diminish the argument, however. What is important is the relative contrast between the imperative examples above and their finite clause counterparts. The VP and CP/FP hypotheses differ in how they relate imperatives to finite clauses. It is uncontroversial that subjects in English raise to a higher inflectional projection. The VP hypothesis, unlike the other two hypotheses, claims that the situation is different in imperatives. An alternative way of stating the argument is that the VP hypothesis predicts that the two clause types should show correspondingly different judgments on parallel examples with stranded quantifiers. The relevant pairs are in (i) and (ii). The VP hypothesis predicts that there should be a sharp contrast between the a examples and the b examples, with the a examples being fully ungrammatical.

(i) a. (?)Don’t you be both talking at once!b. (?)You shouldn’t be both talking at once.

(ii) a. (?)You be both waiting for me promptly at 3!b. (?)You should be both waiting for me promptly at 3!

To the extent that the b examples are degraded, the a examples are no more so to me.


351

3 . 3 . VP Ellipsis

VP ellipsis (VPE) can also be used to argue that the imperative subject is not VP-internal.

To do this, I first review my assumptions, from chapter 1, about how VPE works. In that

analysis, it was argued that VPE is syntactically controlled by the constraint in (23).




As was demonstrated in chapters 2 and 4, imperative do and don’t satisfy this condition

and license ellipsis. Their presence is necessary to render imperative ellipsis examples

grammatical. This yields the repeated contrast between (24) and (25).

(24) a. *The Smiths have reached a decision regarding the matter. Jane, have 0 too!

b . *We can’t have any latecomers. You should definitely be on time. Absolutely no

excuses, be 0!

c. ^Cinderella is doing her chores, so, if you know what’s good for you, be 0 too!

d . *The doctor thinks we all should cut down on caffeine, so everyone 0!

0 = cut down on caffeine


b . We want everyone to come, so those who can, by all means, do 0!

c. Mom said you shouldn’t be playing in the chimney, so don’t be 0 !

The examples in (24) are ungrammatical because the auxiliaries do not raise into the

inflectional head (Schmerling 1977, Lasnik 1981, Beukema and Coopmans 1989, Pollock

1989, Zhang 1990, chapter 2 of this work) and so don’t meet the Licensing Condition. The


352

mere presence of an auxiliary is insufficient; it must be in 1°, the appropriate structural

position for (23). The data in (25) are grammatical because, as argued in chapter 4, do(n ’t)

is structurally and lexically compatible with the requirements of (23).

Since VPE is available in imperatives, we can use it as a diagnostic for imperative phrase

structure and, in particular, the position of the subject. It makes sense to ask whether or not

imperative subjects elide in ellipsis constructions. The VP hypothesis makes a different

prediction from the CP/FP hypotheses regarding what can remain after VPE has applied in

imperatives. Given the VP hypothesis structure in (8), repeated below, VPE will leave only

the upper subject position followed by DO. The lower, post-DO, VP-intemal subject

position is obligatorily elided with the VP. The ellipsis structure DOAsubjectAelided VP (0)

is predicted to be ungrammatical.

(8) IP

(subject) I’

I VP => 0

DO (subject) V’


This is incorrect, as (26) shows. VPE in imperatives can leave the subject following DO

intact indicating that the post-DO subject is not inside VP.


353

(26) a. Rick walked out of the lecture, but don’t everyone else 0, please!

b . Don’t YOU be seen with those losers and don’t ANYONE ELSE be 0 either!

c. Billy didn’t tell mom what I did, so don’t YOU 0 either!

d. I can’t believe that no one has a story they’d like to share. Someone tell us about

an adventure that they’ve had!

?Yes, DO SOMEONE!

It is not an unreasonable prediction that the subject should be able to elide with the VP.

Such an analysis has been proposed for Irish (McCloskey 1996b). It is just not the case

that this happens in English imperatives. The alternative hypotheses easily handle the data

since they do not make this claim. In the FP hypothesis in particular, both subject specifier

positions in (11) are outside of VP. The analysis of (26) is (27).

(27)

do(n’t) subject F2’

F jf^ ^ V PI0


In this section, the position of adverbs in finite main clauses serves as a diagnostic for the

position of subjects in imperatives. By using the phrase structure positions that various

classes of adverbs must occupy in tensed clauses and assuming that their position is

constant across differing clause types, the position of imperative subjects can be

determined, or at least narrowed down. Chapter 1 developed a syntactic analysis of adverb


354

placement in English clauses. The descriptive results are in (28) and (29) for S-adverbs and

E-adverbs, respectively.



b . left adjunction to V

c . left adjunction to the topmost VP

(29) Syntactic Distribution of E-Adverbs

a. left adjunction to I’

b. left adjunction to VP or V’

The placement of adverbs is relevant to determining the location of imperative subjects in

that the position of these adverbs is fixed with respect to certain specifier positions. In

particular, we can count on two things which are germane to the discussion of imperatives:

1) S-adverbs are always at the boundary or above VP, hence, above the specifier of VP and

2) E-adverbs are always below finite clause subjects, hence, below the specifier of IP.

Now consider imperatives and their subjects. With respect to S-adverbs, the VP hypothesis

predicts that S-adverbs will appear above (to the left) of imperative subjects which are

internal to VP. This is correct as the data in (30) shows. The hypothesis also predicts,

however, that this S-adverbAsubject order should be the only possible one and this is

incorrect. In (31), the adverb may also follow the subject. Since these adverbs can appear

no lower than the left edge of the top VP, the subject cannot be VP-intemal in the

examples. A generous way to interpret the data is that the subject may be in the specifier of

VP but it need not be there. This weakens the VP hypothesis.


355

(30) a. Certainly everyone work at least the assigned problems!

b . Normally everyone wait an hour before going in the pool!

c . Certainly someone take good notes!

(31) a. Everyone certainly work at least the assigned problems!

b . Everyone normally wait an hour before going in the pool!

c . Someone certainly take good notes!

To eliminate the VP hypothesis entirely, we need an adverb that cannot appear on the left

edge of a sentence. E-adverbs provide just such a case. They appear in all the positions that

an S-adverb can except clause-initially. As the examples below clearly show, when this

adverb position is eliminated, imperative subjects following the adverb become

unacceptable.5

(32) There’s plenty of room.

a. Everyone simply move to his right a little!

b . * Simply everyone move to his right a little!

(33) a. Don’t you just stand there like a bump on a log!

b . *Don’t just you stand there like a bump on a log!

(34) Be very careful when you dust that machinery.

a. Everyone barely touch it!

b . * Barely everyone touch it!

5The acceptability of these examples given the appropriate context suggests that Stockwell et. al. 1973 are incorrect in claiming that such preverbs are incompatible with imperatives. Davies 1986a: 12 makes the same observation.


356

The ungrammaticality of the b examples confirms that the subject cannot be in spec,VP;

otherwise, they should have the structure in (35).

(35) * IP

(simply) I’

I ^ ^ V P

(simply) VP

D P ^ ^ V ’

everyone move to his right a little

It should be concluded from this that imperative surface subjects are never seen in spec,VP.

They must appear too far to the left with respect to E-adverbs to be VP-intemal.

The CP and FP hypotheses do not run into this difficulty because they posit the subject to

be outside the VP. For the CP hypothesis, we straightforwardly predict the same ordering

of adverbs and subjects as in finite clauses. This appears to be correct. To obtain this result

with the FP hypothesis, something must be said about the position of adverbs within an

articulated inflection system. I do not undertake the task here since it relies on the analysis

of finite clauses within an FP system.6

6A sketch of an analysis would run as follows. Assuming that finite clause subjects occupy spec,FjP and F} ' is the surface position o f modals, the I’-adjoined adverb position becomes adjunction to F j’. If we allow no other adjunctions within the F projections (with the possible exception of adjunction to F2 ’), then the imperative data in (i) correctly fall out

(i) a. Don’t you just stand there!b. Everyone don’t just stand there!c. Everyone just don’t stand there!d. *Don’t just you stand there!


357

Given the untenability of the VP analysis, I will not consider it further. I turn to the

contrasts between the CP and FP hypotheses and attempt to tease them apart.

4. The CP/FP Hypotheses

The evidence from the previous section indicates that there must be at least two projections

above VP in subject imperatives. This is necessary because, in order to obtain the word

order DOAsubject with both elements outside of VP, DO and the subject must be in distinct

projections. With a single projection, the subject in the specifier would necessarily precede

the head, whereas the opposite order is also required. The CP and FP hypotheses both

achieve this by proposing various functional projections above VP to house the subject and

DO.

The fundamental difference between the two analyses is what precisely the identity of these

two functional projections is. The CP hypothesis claims that the two relevant projections

are IP and CP. The FP hypothesis claims that the two functional projections are both

inflectional in nature. If CP were instantiated, it would be above FjP. Since both analyses

can account for the same range of word order facts, we cannot use this to distinguish them;

instead, diagnostics for identifying individual projections must be appealed to. In the

discussions below, I will specifically investigate evidence pertaining to the presence or

absence of CP, as this most clearly distinguishes the two hypotheses. To facilitate the

identification of CP, I will systematically compare the imperative with interrogative clauses.

Interrogatives are canonical structures involving a CP above the inflectional projection(s)

and it is the apparent similarity between negative imperatives and interrogatives which

motivates the CP hypothesis:


358

(36) a. Don’t you have a cow?

b. Don’t you have a cow!

The subject-auxiliary inversion which relates interrogatives like (36)a and their declarative

counterparts is modeled here as movement of an auxiliary or modal from I°-to-C°

(Koopman 1984, Chomsky 1986).

In this section, I explore two syntactic phenomena that can be used to identify CP:

topicalization and inversion-triggering adverbials (ITAs). The method will be to compare

the behavior of interrogatives and imperatives with respect to these two phenomena. If

imperatives do instantiate a C° projection, similar results should obtain and this would

argue for the CP analysis of imperatives. On the other hand, if interrogatives and

imperatives behave differently, the FP hypothesis might be preferred because it encodes a

structural difference. The reason for investigating imperative structure within the context of

interrogatives is that it permits a baseline, with known syntactic structure and behavior, for

comparison and helps to reduce extraneous factors which might be complicating the

imperative data.

Both topicalization and ITAs have been used elsewhere in the literature to argue that

imperative DO is not as high as CP or, alternatively, that imperatives do not overtly

manifest a C° projection. With regard to topicalization, Zhang 1990 cites the contrast in

(37) to show that imperatives and interrogatives must have different syntactic structures.

(37) a. That present, don’t you open until next week!

b . *That classic novel, did you read last week?


359

Imperatives seem to permit a fronted topic, (37)a while yes/no questions do not, (37)b.

Under the assumption that topics are to the right of C \ the data suggest that interrogatives,

but not imperatives, involve a CP. Section 4.1 evaluates this body of data more carefully.

Henry 1995 uses inversion-triggering adverbials to argue that imperative DO is not in C°.

Negative adverbial phrases such as under no circumstances, never, on no account, etc.

trigger subject-auxiliary inversion when they are clause-initial. This is typically analyzed

with the adverbial being in spec,CP and the auxiliary undergoing T-to-C" movement, as in

interrogatives. Crucially, CP is implicated. If imperatives contained a CP, IT As should be

permitted, parallel to declarative sentences. They are not, yielding Henry’s contrast in (38).

(38) a. * Under no circumstances do everybody sit down!

b . Under no circumstances do they go away.

As with the topicalization data, this result is unexpected if the CP hypothesis were correct.

ITAs are investigated in section 4.2. In both sections, closer examination of the data reveals

that imperatives and interrogatives behave the same. This suggests that imperatives are also

CP structures. The data do not rule out the FP hypothesis, however, as they are compatible

with that analysis as well. Only a theory-internal argument for the CP hypothesis can be

offered: simplicity demands that both clause types receive the same analysis.

4 .1 . Topicalization

Topicalization is the construction illustrated in (39) in which a non-w/i, maximal projection

is found in sentence-initial position and corresponds to a gap somewhere in the sentence.

(39) a. Bigotry, we will not tolerate.

b . The fuzzy pink slippers, Maxine gladly donated to Goodwill.


360

Lasnik and Saito 1992, following Baltin 1982, argue that the structural position of such

topics is adjunction to IP. This can be confirmed by considering the structure in (40). The

structural position of the topic can be determined by locating it with respect to the subject in

the specifier of IP, a complementizer in C°, and a w/t-phrase in the specifier of CP.

(40) CP

wh-phrase C’

C° IP

complementizer subject V

I VP

The data below indicate that the topic must be between the complementizer and the subject.

In the embedded topicalization example in (41) the topic occurs to the right of the

complementizer that. It cannot appear above the complementizer, (41)b, or below the

subject, (41)c.7

7A similar argument can be made using topicalization within relative clauses, although the data are less robust The topic in these examples must appear to the right of both the complementizer and w/z-phrase and to the left of the subject:

(i) a. the fake diamond which/that to Ivana we soldb. *the fake diamond to Ivana which/that we soldc. *the fake diamond which/that we to Ivana sold

Rochemont 1989 and Lasnik and Saito 1992 point out the marginal nature of examples like these. Such data and Baltin’s (1982) original example in (ii) are not accepted by all speakers. Furthermore, the data itself is not very stable; similar examples in (iii), some from Rochemont 1989, are of varying acceptability.

(ii) He’s a man to whom liberty we could never grant

(iii) a. *He’s a man who liberty we could never grant to.b. T h is is the book that under the table I saw.c. ??He’s a man from whom money we could never take.d. ??That’s the sort of joke that on Bill we could never play.e. TThis is the spot where to Janice my father proposed.

It appears that complex stylistic and pragmatic factors affect the acceptability of the examples as it improves with contextualization and focus on the topic XP.


361

(41) a. Larry believed that, his allowance, they would surely cut.

b . *Larry believed, his allowance, that they would surely cut.

c. *Larry believed that they, his allowance, would surely cut.

The topic must appear between C° and spec,IP in (40) and adjunction to IP is compatible

with the evidence.8

As Jim McCloskey points out to me, this analysis gains further support from the interaction

of topic phrases with clausal adverbials that are adjoined to IP. The analysis predicts that

the two should be freely ordered with respect to one another and, for most speakers, this is

the case. As (42) and (43) show, topics may precede or follow adverbs introduced by

when or if.

(42) a. This kind of stuff, when I graduate, I’ll never look at again,

b . When I graduate, this kind of stuff, I’ll never look at again.

(43) a. The scandalous rumor, if you talk to Horatio, he’ll deny,

b . If you talk to Horatio, the scandalous rumor, he’ll deny.

Topicalization is also possible in imperatives (Zhang 1990) as the data in (44) through (48)

illustrate. The position of the topic is such that it must always be to the left of both DO and

the subject, independent of their relative order:

(44) a. The calamari mousse, do try! It’s delicious,

b . *Do, the calamari mousse try!

8An alternative is that the topic is located in the specifier o f a projection immediately dominating IP. Culicover 1991 argues for such an analysis, positing a Pol(arity) Phrase above IP. I will not pursue this possibility here.


362

(45) You can have raisins for a snack.

a. Chocolate, don’t even think about!

b . * Don’t, chocolate, even think about!

(46) a. His lies, nobody believe!

b . *Nobody, his lies believe!

(47) a. My good wine, don’t anybody touch!

b . *Don’t, my good wine, anybody touch!

c. *Don’t anybody, my good wine, touch!

(48) a. This new beer, everybody do try!

b. *Everybody, this new beer, do try!

c. *Everybody do, this new beer, try!

Of the two hypotheses under investigation for the structure of an imperative, the CP and FP

hypotheses, only the FP hypothesis would seem to be able to account for these data.

Consider both in turn. Although the FP hypothesis with its split inflectional projections

does not immediately fit into the analysis of topicalization since there is no IP, it seems

clear that topicalization in this case has to be adjunction to the highest functional head FjP.

The hypothesis, then, correctly accounts for the data since the topic will always be

peripheral to the imperative, to the left of both the subject and DO regardless of their

locations, as illustrated in (49). The predicted orders are either topicADOAsubject or

topicAsubjectADO, seen in (47)a and (48)a, respectively. None of the ungrammatical

examples in b or c is generated.


363

(49) F l P

topic FjP

(subject) Fy

(DO) (subject) F2’


The structure of an imperative with a topic phrase under the CP hypothesis is given in (50).

DO may appear in either C° or 1°. The predicted orders are transparently DOAtopicAsubject

or topicAsubjectADO.

(50)

(DO) topic

subject


This latter order is correctly instantiated by (48)a. The former ordering, however, is a

wrong prediction. Examples such as (47)b above indicate that DO preceding the topic is

ungrammatical. Additionally, the CP hypothesis would seem to have no way of deriving

the grammatical example, (47)a, in which the topic precedes DO.

The two hypotheses diverge in the case where DO precedes the subject. The CP hypothesis

places the topic after DO, the FP hypothesis, before. The latter is correct as (47) shows.


364

The topic must precede DO and the subject. The data would then seem to argue against DO

being in C°, as Zhang concludes. In order to salvage the CP hypothesis, we would need

two apparently unmotivated stipulations: 1) a CP-adjunction position for the topic to

account for some left-peripheral, clause-external topics and 2) an account of why the

IP-adjunction site which places the topic between DO and the subject is unavailable. Before

accepting Zhang’s conclusion that imperatives do not invoke a CP, however, we should

more closely consider the behavior of topicalization in interrogatives—a construction which

instantiates a CP structure. What we will see is that the same two difficulties arise and, in

fact, receive explanations that are straightforwardly extendible to imperatives. Thus, the CP

hypothesis is not actually ruled out for imperatives.

As stated earlier, I assume that the subject-auxiliary inversion in interrogatives is a result of

I°-to-C° movement. Thus, a typical yes/no question has the structure in (51).

(51) a. Can you smell the roses?

b. CPI

C’

C ^ ^ I P

you I VP

tk smell the roses

Zhang 1990, in comparing topicalization in interrogatives and imperatives, claims that the

topic cannot appear on the left periphery in the former. The a examples in (52) through


365

(56), suggest that this is incorrect; the examples seem fine.9 (52)a is from Radford 1988.

On the other hand, reference to the structure in (40) and the analysis of topicalization as

adjunction to IP lead us to expect the topic to appear between the auxiliary and the subject.

The b examples below show that this is completely impossible.

(52) a. That kind of antisocial behavior can we really tolerate in a civilized society?

b. *Can, that kind of antisocial behavior, we really tolerate in a civilized society?

(53) a. Those kind of men aren’t you tired of going out with?

b. * Aren’t, those kind of men, you tired of going out with?

(54) a. That report shouldn’t you have read already?

b . *Shouldn’t, that report, you have read already?

(55) a. About his private life do we know anything?

b. *Do, about his private life, we know anything?

(56) a. A KANGAROO did Jim claim he had never seen?

b. *Did, a KANGAROO, Jim claim he had never seen?

The data indicate that, in a CP structure, a topic position is available to the left of C° (the

position marked by the auxiliary) but not in the expected IP-adjoined position. If it is

accepted that interrogatives are CPs, then we seem to independently need an account of the

two stipulations that are required to maintain the CP hypothesis for imperatives (see

McCloskey 1992 for similar observations and conclusions). With respect to topicalization,

imperatives and yes/no questions behave the same. This suggests a single analysis.

9Rochemont 1989 makes the same observation.


366

Rochemont 1989 refines and extends the IP-adjunction analysis, convincingly arguing that

topicalization is more generally adjunction to a propositional domain, either CP or IP.

(57) Topicalization in English (Rochemont 1989)

Topicalized XPs are left adjoined to CP or IP

Adjunction to CP straightforwardly accounts for the a examples above. They have a

structure as in (58).

(58) a. Live goldfish would you swallow?

b. [cp live goldfish^ [cp [c , wouldj [rp you tj swallow tk ] ] ]

This adjunction is also reasonably employed in left dislocation structures, illustrated in

(59). The difference between the two phenomena is that left dislocation does not involve

movement, only the base-generated, CP-adjoined topic.

(59) a. [cp This plant [cp [Ip it brightens the whole room ] ] ]

b . Women, F11 never understand them.

c. The latest Stephen King novel, Mary read it in one sitting.

d. Julia Child, the kids are dying to meet her.

In support of such an analysis is the observation (Rochemont 1989:147) that topicalization

is sometimes possible in w/i-questions, (60). This possibility can only be considered if

adjunction of the topic to CP is an option.10

10These data have a status similar to the topicalization in relative clause examples mentioned in an earlier footnote. Their acceptability is somewhat variable. Radford 1988, for example, states that such data are impossible, offering (i). Further examples in (ii) also seem marginal.

(i) a. *That kind of pen what can you use for?b. *What that kind of pen can you use for?


367

(60) a. In the living room what did Mary put?

b . To the next potluck what is Maria going to take?

c. [q , to the President [cp what [c . did [jp the butler whisper ] ] ] ]

(ii) a. *His convertible when did Bob say he drove?b. *The dog how could you toss the leftovers to?c. ?The important guests what did the waitress serve?

Within the Lasnik and Saito 1992 theory, which Rochemont initially adopts, the examples are actually Subjacency violations, thus, predicted to be ungrammatical. The definitions for Suhjacency are given in (iii) and (iv). The definition o f L-marking follows Chomsky 1986. Relevant to this discussion is that I* L-marks VP but C’ does not L-mark IP. Given the definition of barrier in (iii), IP is a barrier, as is a noncomplement CP.

(iii) y is a barrier for P ifa. y is a maximal projectionb. y is not L-marked, andc. Y dominates P

(iv) P is subjacent to a if for every Y. Y 3 barrier for P, the maximal projection immediately dominating y dominates a .

The structure that we are interested in corresponding to the above examples is in (v).

(v) 1cp2 toPicj tcPj whj fc’ AUXk [IP subJect [p tk [Vp • tj • • - tj • • 111111

There are two situations to consider, depending upon whether the topic or w/z-phrase moves first. Movement of the w/z-phrase to spec,CP crosses one barrier, IP, but the w/z-phrase is subjacent to its trace according to (iv) because the maximal projection immediately dominating this barrier, C Pj, also dominatesthe w/z-phrase. Movement o f the topic to the CP-adjoined position now violates Subjacency. The topic crosses two barriers, IP and C Pj. The maximal projection that immediately dominates IP is CP j and itdoes not dominate the topic phrase. Suppose the topic were to move first. In order to not violate Subjacency as in the above scenario, the topic cannot move in one swoop to the adjoined location. It must first stop in the spec,CPj position before moving on. This initial movement to the specifier crosses onebarrier but obeys Subjacency since the projection immediately dominating the IP barrier, CPj, alsodominates the specifier position. The second movement, from spec,CP j to the adjoined position also obeysSubjacency. One barrier is crossed, C Pj, but again the maximal projection that immediately dominates it isCP2 and this also dominates the adjoined topic. But now, movement of the w/z-phrase to the specifier ofCP j is illicit because it is occupied by a trace of the topic phrase and movement cannot cover up traces.Thus, the Lasnik and Saito system actually provides no derivation for Rochemont's examples. One could allow the data in by saying that matrix CPs are L-marked but this seems somewhat ad hoc. An alternative is to recognize the marginal status of the examples and accept that they are Subjacency violations.


368

If a CP-adjoined topic position to the left of C° is available for interrogatives, then its

analysis can be exported to imperative topicalization at no cost. The position of the topic in

the imperative data does not argue against the CP analysis.

Regarding the impossibility of adjunction to IP when T-to-C0 movement has applied, it is

worth pointing out that this, too, seems to be a more general prohibition, not simply

restricted to topicalization in interrogatives. McCloskey 1992 observes that adjunction to IP

in English is freely available, for both moved phrases and base-generated ones.

Jackendoffs S-adverbs, analyzed earlier, have as one of their permitted positions left

adjunction to IP. They appear on the left of a declarative sentence:

(61) a. •UsuallyProbablyMostly

the airliners don’t fly over this neighborhood.

b . Most likely he will regret his decision.

c. [q> Obviously [jp Mary cheated on her diet ] ]

The interrogatives corresponding to these examples are robustly ungrammatical, however.

(62) a. *Don’t -Usually

ProbablyMostly

the airliners fly over this neighborhood?11

b . *Will most likely he regret his decision?

c . *Did obviously Mary cheat on her diet?

11 It is evident that the difficulty here is not a semantic incompatibility between the adverb and posing a question (see Bellert 1977, Ernst 1984:290 and references cited therein). These questions are acceptable if the adverb is in the VP-adjoined position:

(i) a. Don’t the airliners usually fly over this neighborhood?b. Will he most likely regret his indecision?


369

The structure that must be ruled out is in (63).

(63) * CPI

e

didk AdvP IP

obviously DP I’

Mary I VP

tk cheat on her diet

Since we know that, independently, the adverb adjunction and T-to-C0 movement are

permitted, it must be the interaction that rules out (63). Rochemont proposes that the

derivation is, in fact, ruled out by Subjacency. The IP-adjunction structure creates a barrier

to movement. Within the Lasnik and Saito 1992 system, this proposal follows (see Lasnik

and Saito 1992:86 and the earlier footnote). Since Subjacency violations require only that

there be movement and are independent of interrogatives per se, such an account is

extendible to imperatives. For a parallel analysis to apply to the imperative, however, it is

necessary that the imperative auxiliary actually undergo I°-to-C° movement. This

conclusion supports claims made in chapter 4 in this work and is discussed more fully

later.

The alternative derivation in which T-to-C° occurs first followed by the topicalization

adjunction to IP is ruled out by the extension requirement on phrase structure (Chomsky

1994, 1995). The operation Move, which is responsible for the movement of elements,

applies only at the root for overt operations (modulo X° movement). Thus, at the point in a

derivation at which Topicalization must apply, there is as yet no C° position for 1° to move


370

into so I0-to-C° is not even an option. The operation of adjoining to IP is available first and

hence, if it is to occur, must occur prior to T-to-C".

It cannot be concluded based on topicalization data, then, that imperatives are not CPs. The

full range of data are ultimately compatible with either the FP or CP hypothesis. The CP

analysis cannot be ruled out, despite apparent difficulties, because the problems arise in

other situations, namely interrogative CP structures, and have independent explanations.

4 . 2 . Inversion-Triggering Adverbials

Inversion-triggering adverbials (ITAs) are a subset of adverb phrases which, when they

appear in clause-initial position, trigger inversion of the subject with the first auxiliary in

the sentence (Klima 1964, Liberman 1974, Ladusaw 1980, Rudanko 1982). Several are

illustrated in (64). They most commonly include adverbials with a implicit or explicit

negative meaning. This construction is characterized by a smooth intonation contour; there

is no pause after the fronted constituent.

b . Under no circumstances does Judy tolerate poor manners from her guests.

c. Only in emergencies would Joey share his chewing gum.

When there is no auxiliary, modal, or form of be present to undergo inversion, Do-Support

occurs, as (64)b illustrates. The examples are almost always ungrammatical without

inversion:

(64) a. ^ave * seen suc^ a disaster.

(65) a' * {Hardly} * ave seen suc a c saster- b . *Under no circumstances Judy tolerates poor manners from her guests.

c . *Only in emergencies Joey would share his chewing gum.


371

Comparing (64) to the questions in (66), we see that the process is exactly the same and

this suggests a similar analysis: the adverbial phrase is in spec,CP and the auxiliary has

moved to the local head C \

(66) a. When have I seen such a disaster?

b . Why does Judy tolerate poor manners from her guests?

c. With whom would Joey share his chewing gum?

The analysis of IT As as involving movement of the adverbial to spec,CP and the auxiliary

to C°, just as in interrogatives, has a long history (Koster 1975, Emonds 1976, Radford

1988, Rizzi 1991, Progovac 1994, Haegeman 1995). Radford 1988 observes that

movement of other phrases to spec,CP is impossible when ITAs have already been fronted,

suggesting that IT As also occupy spec,CP. (67) illustrates this claim for IT As with wh-

phrases.

(67) a. *Never ever who would he trust with such a mission?

b . *Rarely who does his share of the work around here?

The mechanism that Rizzi 1991 uses to force fronting and inversion of IT As and wh-words

is specifier-head agreement. Informally, IT As and question words are affective operators

and they must be in a spec-head configuration with a head marked with the relevant

affective feature. These heads obtain the affective feature via dynamic agreement (Rizzi

1991) once they are in the appropriate configurational relationship.

Assume that IT As are marked with a feature [+aff]. Rizzi’s proposal in (68) requires that

[+aff] operators be in a particular configuration with an X° [+aff].


372

(68) Generalized Operator Criterion (Rizzi 1991)

A. An Operator [+aff] must be in a Spec-head configuration with an X° [+aff]B. An X° [+aff] must be in a Spec-head configuration with an Operator [+aff]

An operator is taken to be something in a left-peripheral A-position.12 Another way of

representing the analysis is to say that the presence of either a head or an operator specified

as [+aff] requires the configuration in (69).

(69) CP

Op C’[+aff]

C IP[+aff]

Auxiliaries that move into C° are not inherently [+aff] but become so specified through

dynamic agreement, in (70) (See Haegeman 1995 for an extension of Rizzi’s analysis that

does not require any appeal to dynamic agreement).

(70) [+aff] Dynamic Agreement (Rizzi 1991)

l+am-Op X- =* l+affl-op ^

If a [+aff] operator appears in spec,CP, something must move into C° to satisfy the

Operator Criterion. If nothing does, clause A of the criterion is violated and the result, (65),

is ungrammatical. The crucial pieces here, in some sense independent of the particulars of

the analysis, are that IT As occupy the specifier of CP and that they require an overt element

in C°. Accepting this, IT As can be used as a diagnostic for the presence of CP. Remember

that this is the difference between the CP and FP hypotheses that we are investigating. As

l2On this view, an unmoved adverbial is not an operator and so not subject to the Operator Criterion.


373

before, the strategy will be to consider the possibility of IT As in imperatives and

interrogatives.

Most examples of IT As, like rarely, seldom, hardly ever, under no circumstances, etc., are

either explicitly negative or suggest negation in context; thus, they are often semantically

incompatible with morphologically negated sentences. Although we can figure out what the

examples in (71) should mean, they are awkward.

(71) a. ?Rarely don’11 have the recommended five servings of fruit a day.

b . ?Under no circumstances wouldn’t Bob offer to help.

Better examples can be formed with adverbial phrases that begin with only. They are strict

IT As as the contrast in (72) shows; however, they are not explicitly negative so they seem

to sound better with sentential negation, as in (73).

(72) a. Only during full moons does Herzog make offerings to the gods,

b . *Only during full moons Herzog makes offerings to the gods.

(73) a. Only when it rains doesn’t Carol go for her morning stroll.

b . Only on the weekends don’t I have to get up at this crazy hour.


374

When we test these ITAs in imperatives, the results seem uniformly bad.13

(74) a. *Only when I say don’t everyone sing!

b . ??Only next week don’t you come to class!

c. ??Only for the next thirty seconds don’t anyone talk!

d . ??Only on Fridays don’t eat meat!

e . *Only with adult supervision do someone try this!

f . *Only after Hamlet is finished do say your lines!

Henry 1995 concludes from such data that CP is not active in Standard English

imperatives; otherwise, the Operator Criterion should be satisfiable and ITAs should be

permitted in imperatives. Neither the ITA nor any form of DO can be in the C° projection

and the data would thus seem to argue against the CP hypothesis. Again, looking further to

interrogatives, however, reveals the complete picture. The examples in (75) appear to

indicate that yes/no questions also do not permit ITAs. This is perhaps surprising given that

spec,CP is not otherwise overtly occupied.

13The ungrammatically of the examples is not due to an incompatibility between sentence-initial adverbials and imperatives, as (i) shows. Initial adverb phrases are acceptable if they are not inversion triggering. It is also not due to an incompatibility between only adverbials and imperatives, as they may appear clause finally, (ii). In such cases, however, they are not in spec,CP and thus are not operators subject to Rizzi’s Operator Criterion.

(i) a. During the solo, don’t everyone sing!b. When I say the word, everyone run to the left side of the airplane!

(ii) a. Don’t everyone sing only when I say!b. Try this only with adult supervision!c. Do say your lines only when it’s your turn, please!


375

(75) a. ??Only when I say do they all sing?

b . ??Only next week shouldn’t we come to class?

c. ??Only with adult supervision are we to try this?

In general, however, the type of data in (74) and (75) is not very robust. Some of the

imperative examples seem more or less acceptable on a topicalization reading where the

only-phrase is a topic and receives some kind of focal stress:

(76) You all don’t have to be quiet for the entire class.

Only for the next thirty seconds, don’t anyone talk! I really want to hear this

announcement.

Some people also accept the question data in (75). Although I believe the examples are

good only as declarative sentences with question intonation, there is no way to test this.

The data can be taken to suggest that ITAs behave the same way in imperatives and

interrogatives but the inability to distinguish ITA fronting from topicalization makes the

whole interpretory endeavor inconclusive. The reason that we cannot distinguish ITAs

from topicalization is that there are no word order clues because inversion, normally

associated only with ITAs, has independently applied. Below I consider a set of slightly

different data introduced in Liberman 1974 which allows us to differentiate the two

phenomena and answer the original question about the possibility of ITAs in interrogatives

and imperatives.

Liberman 1974 investigates a set of ITAs in which the presence or absence of inversion

(ITA versus topicalization) in the syntactic structure corresponds to a difference in

meaning. Since the ITA and topicalization structures have different meanings, it becomes

apparent which syntactic structure we are dealing with. His well-known example is in (77).

(77) has two interpretations, paraphrased in (78)a and b, which differ in the scope that the


376

negation in the prepositional phrase takes. The two readings can be distinguished by

fronting the PP and applying inversion or not. When with no job acts as a fronted ITA with

inversion, (79)a, it has only the ‘unsatisfiable’ interpretation in (78)a. If there is only

fronting with no inversion, (79)b, the example corresponds exclusively to the

‘unemployed’ meaning in (78)b.

(77) John would be happy with no job.

(78) a. There is no job such that John would be happy with it. (‘unsatisfiable’ meaning)

b. John would be happy without a job. (‘unemployed’ meaning)

(79) a. With no job would John be happy.

b. With no job, John would be happy.

The two interpretations correspond to an ITA and topicalization, respectively. In the former

case, the PP with no job is specified as [+aff]. Consequently, it is subject to the Operator

Criterion and triggers inversion. It has the structure in (80)a. It appears in spec,CP with an

agreeing [+aff] head in C \ I will call (78)a/(79)a/(80)a the AFF reading.

(80) a. [cp [ pp With no job ]j [c - [ c would ]k [jp John tk [yp be happy tj ] ] ] ] [+aff] [+aff]

In (79)b the PP is not marked [+aff| and there is no inversion. The example is simply PP

topicalization as discussed in the previous section. The PP is adjoined to IP as in (80)b.

(78)b/(79)b/(80)b will be labeled the TOP reading.

(80) b . [jp [pp With no job ]j [jp John would [yp be happy tj ] ] ]

These data have an advantage over the examples just discussed because the meaning

distinction provides a clear indication of which structure we are dealing with: an ITA in


377

spec,CP or a topicalized phrase adjoined to IP. It is the former case that interests us

because it signals the presence of an active CP. Furthermore, other differences parallel the

semantic shift and permit further probing of the data and verification of the conclusions.

These contrasts are summarized in (81) and developed below. The reason for such

complete diagnostics is that with both interrogatives and imperatives the word order does

not distinguish between ITAs and topicalization so other means of distinguishing the two

must be relied on. In addition, the judgments on the data are rather subtle and the more

confidence we can create in them, the better.

(81) AFF reading (ITA with inversion) TOP reading (topicalization)

As Liberman observes, the two types of examples have different intonation contours.

When there is no inversion, an intonation break is present between the fronted phrase and

the remainder of the clause. When the fronted phrase is a [+aff] operator, the intonation is

smooth, with no pause afterwards. Although this is not a completely reliable diagnostic, the

intonation break in the topicalization examples will be indicated by a comma.

More interestingly, Liberman 1974 and Rochemont 1985 observe that the AFF/TOP

distinction corresponds to a difference in the syntactic status of the PP, in addition to the

presence or absence of the affective feature. Casting their observations in current terms,

the contrast is one of syntactic status as argument or adjunct. With the AFF reading, the PP


analysis:XP occupies spec,CP obligatory inversion XP marked [+aff]

XP adjoined to IP/CP inversion independent of adjunction XP not marked [+aff]

syntactic behavior: smooth intonation argument permits extraction licenses NPIsdoes not permit focus particles

comma intonation adjunctisland to extraction does not license NPIs allows focus particles

378

is an argument; in the TOP reading, the PP is an adjunct. This reflects Liberman’s

observation that with an adjective like dull that does not subcategorize for a PP argument,

only the TOP reading exists. (82)a, even with no fronting, has only the TOP reading and is

fully synonymous with (82)c in which the ITA is fronted. The AFF reading in (82)b which

I have tried to force using inversion is ungrammatical.

(82) a. John would be dull, with no job.

b . *With no job would John be dull.

c. With no job, John would be dull.

This argument/adjunct distinction provides a further contrast between the two readings.

Because extraction possibilities from within arguments and adjuncts differ, only with the

argument, AFF reading can the NP object of the preposition be fronted, stranding the

preposition. Extraction from within an adjunct, the TOP reading, is known to be

prohibited. The examples below are from Rochemont 1985:73. They differ from those in

(79) only in that the object of the preposition has been fronted rather than the entire

prepositional phrase. (83)a, movement from within the argument is grammatical and has

the AFF reading, as confirmed by the auxiliary inversion. The TOP reading in (83)b by

contrast is ungrammatical because it requires extraction from the adjunct PP.

(83) a. No job would John be happy with. APT READING

(cf. Which job would John be happy with?)

b. *No job, John would be happy with. *TOP READING


379

The examples in (84) with the predicate adjective dull are both ungrammatical because dull

does not take an argument PP. This precludes both an AFF reading of (84)a and movement

from within the adjunct, (84)b.14

(84) a. *No job would John be dull with. *AFF READING

(cf. *Which job would John be dull with?)

b . *No job, John would be dull with.15 *TOP READING

A further difference between the two structures is in the licensing o f negative polarity items

(NPIs). NPIs are phrases that must be in an affective environment such as the scope of

negation, interrogatives, antecedents of conditionals, complements of adversative

predicates, etc. The negative adverbs under discussion are suitable licensers for NPIs when

they are marked [+aff] and take scope over their clause from spec, CP. They do not

function as NPI licensers when they are simply topicalized. Consequently, the AIT and

TOP readings can be distinguished by whether or not an NPI may appear in the main

clause. An NPI will be compatible only with the AFF reading. In (85), the NPI ever is

licensed by the affective operator in (85)a with the AFF reading but not the topicalized PP

with the TOP reading, (85)b.

(85) a. With no job would John ever be happy. AFF READING

b . *With no job, John would ever be happy. *TOP READING

14While these contrasts seem real, their basis remains mysterious. I have no theoretical account of the above differences. In general, Negative Inversion cannot and should not be limited to applying to only arguments. Many ITAs are precisely that, adverbials.l5This example and the one above might be independently ungrammatical due to a general constraint against topicalizing negative quantifiers.


380

On the other hand, the AFF reading is largely incompatible with focus particles like even

(see Anderson 1972, Rooth 1985 for a discussion of the semantics of even). This accounts

for the contrast in (86).

(86) a. *Even with no job would John be happy. *AFF READING

b . Even with no job, John would be happy. TOP READING

These diagnostics, summarized back in (81) and repeated below, permit us to identify an

ITA, and hence CP, even when the word order provides no help. To reiterate, the TOP

reading can be recognized by the following cluster of properties: a comma intonation, no

extraction from within the constituent, no licensing of NPIs in the main clause, and the

acceptability of focus particles such as even. The AFF reading shows the complementary

set of properties.

(81) AFF reading (ITA with inversion) TOP reading (topicalization)

analysis:XP occupies spec,CP XP adjoined to IP/CPobligatory inversion inversion independent of adjunctionXP marked [+aff] XP not marked [+aff]

syntactic behavior:smooth intonation comma intonationargument adjunctpermits extraction island to extractionlicenses NPIs does not license NPIsdoes not permit focus particles allows focus particles

Before returning to imperatives and interrogatives, I repeat the baseline declarative data.

The facts are that fronting with inversion produces the AFF reading while no inversion

results in the TOP interpretation.


381

(87) a. Isaac is satisfied with no analysis.

b . With no analysis is Isaac satisfied.

c. With no analysis, Isaac is satisfied.

BOTH READINGS

TOP READING

AFF READING

The data in (88) establish the pattern for imperatives. Like the declarative counterpart, the

example without fronting in (88)a is ambiguous, with both the AFF and TOP

interpretations paraphrased in (89). With fronting, (88)b, the word order does not indicate

whether or not there is inversion to C° and judgments on the example are unclear. The

reader is invited to make his own assessment before continuing.

(88) a. Do be satisfied with no analysis! BOTH READINGS

b. With no analysis (,) do be satisfied!

(89) a. There is no analysis that you should be satisfied with. AFF READING

b . You should be satisfied with having no analysis. TOP READING

The consensus on (88)b is that the TOP reading paraphrased in (89)b is (marginally)

grammatical but that the AFF reading in (89)a is not. The diagnostics allow us to probe the

two readings. Extraction from within the PP, stranding the preposition, is unacceptable,

(90)a, as is an NPI in the main clause, (90)b. On the other hand, the example seems

tolerable with the topic particle even and a comma intonation, (90)c. The last example

confirms the availability of the TOP reading while the first two tests indicate the

unavailability of the AFF reading.

(90) a. *No analysis, do be satisfied with! *AFF READING

b. ??With no analysis do ever be satisfied! * AFF READING

c. Even with no analysis, do be satisfied! TOP READING


382

The syntactic consequences are that topicalization in imperatives is unsurprisingly

permitted, as was observed in the previous section, but the ITA analysis and structural

configuration—movement of the ITA to spec,CP and inversion of DO to C°—seems

unavailable. The illicit structure is

(91) CP

PPi[+aff]

with no analysis [+aff]

doL

C’

IP

DP r

VP

tk be satisfied tj

It is again tempting to conclude that the CP hypothesis for imperatives is incorrect since it

permits the structure in (91). It straightforwardly predicts that spec,CP is available for an

ITA and Dynamic Agreement could then occur with DO in C°. ITAs should be permitted

under the CP hypothesis. But again, when we turn to interrogatives, the same pattern in the

data is seen and we are forced to explain why ITAs are incompatible with them as well.

Consider the yes/no questions in (92). As before, the question in (92)a without fronting is

ambiguous. The judgments on the example with fronting, (92)b, are again subtle.

Rochemont 1985 asserts that a similar example in (92)c has only the TOP reading.

(92) a. Can you be satisfied with no analysis?

b . With no analysis (,) can you be satisfied?

c. With no job (,) would John be happy?

BOTH READINGS

TOP READING


383

The tests in (93) indicate that the AFF reading is, in fact, not available. The judgments on

the data should be compared to the corresponding examples for imperatives in (90). ITAs

appear to be not permitted in yes/no questions.

(93) a. *No analysis, can you be satisfied with?16 *AFF READING

b. Q. * With no analysis can you or Bill ever be satisfied? * AFF READING

A Bill.17

c. Even with no analysis can you be satisfied? TOP READING

ITAs are also not permitted in constituent questions. The same pattern is seen below for

w/z-questions. (94)a and b show that both reading are available when the PP is unmoved.

When it is fronted, in (94)c and d, only the TOP reading survives. The diagnostics in (95)

and (96) confirm the judgment.18 We cannot test using NPIs because they would be

licensed for independent reasons.

16The a and c examples may well be grammatical as declaratives with question (incredulity) intonation. I have found no way to exclude this possibility.17The situation is complicated by the fact that most polarity questions independently license NPIs:

(i) a. Will anyone go?b. Does he ever clean his desk?

Important for considerations here, Han and Siegel (to appear) and others (e.g. Borkin 1971 and Ladusaw 1980) observe that NPIs are licensed only in questions in which a negative answer is possible. NPIs are ungrammatical in questions in which a negative answer is not possible. One instance of this is so-called alternative questions, illustrated in (ii), in which the answer is not yes or no but must be one o f a choice of alternatives. As shown in (iii), NPIs are not licensed in this case.

(ii) Will Bob Dole or Bill Clinton win the election?

(iii) *Will Bob Dole or Bill Clinton ever win the election?

The question/answer scenario above is intended to be of this type.18See the earlier discussion regarding the marginal status of Topicalization with w/i-questions which accounts for the data marked ?/??.


384

(94) a. Who is satisfied with no analysis? BOTH READINGS

b. Why are you satisfied with no analysis? BOTH READINGS

c. ?With no analysis (,) who can be satisfied? TOP READING

d. ?With no analysis (,) why are you satisfied? TOP READING

(95) a. *No analysis, who can be satisfied with? *AFF READING

c. ??Even with no analysis, who can be satisfied? TOP READING

(96) a. *No analysis, why are you satisfied with? *AFF READING

c. ??Even with no analysis why are you satisfied? TOP READING

The data so far indicate that interrogatives and imperatives are alike in not permitting IT As

in their structure. These complex diagnostics indicate that when such examples seem more

or less acceptable, they are really only topicalized adverbial phrases that do not occupy

spec,CP or trigger inversion.

Yet a second kind of ITA that replicates the above pattern is degree-so phrases. Such

phrases, illustrated in (97), obligatorily trigger inversion. They too are uniformly

incompatible with imperatives and interrogatives. Neither imperatives, (98), yes/no

questions, (99), nor w/i-questions, (100), can contain a fronted so-phrase.

(97) a. So embarrassed was Jack that he couldn’t answer the question.

b . So lacking in self-confidence was he that it was impossible to work with him.

(98) a. *So cocky don’t you be that you miss the easy ones!

b . *So ill-mannered don’t everybody act that the governess won’t want the job!

c . *So surly don’t be that no one will talk to you!


385

(99) a. *So embarrassed was Jack that he couldn’t answer the question?19

b . *So cocky were you that you missed the easy ones?

c . *So ill-mannered were the children that the governess didn’t take the job?

(100) a . *So embarrassed which speaker was that he couldn’t answer the question?

b . *So inept to which supervisor did the worker seem that there was no chance

he would be hired?

b . *So rudely to whom did the children behave she didn’t want the job?

These data only serve to confirm the conclusion that ITAs are not compatible with

imperatives or interrogatives. Returning to an explanation, a likely account of why ITAs are

not permitted in interrogatives is that spec,CP is filled. This is obviously true for wh-

questions in which the vv/i-phrase occupies spec,CP. It is also generally accepted that

spec,CP is filled with a null operator in yes/no questions. It is this operator that triggers

subject-auxiliary inversion and licenses NPIs. Accepting this, there is an immediate

explanation for the impossibility of ITAs in all interrogatives: spec,CP is already filled.

Exporting the account to imperatives would require positing an imperative operator.

Although evidence for such a proposal is conspicuously lacking in English, it is compatible

with the data. It has also been proposed for Dutch (den Dikken 1992). As was the case

with topicalization, we cannot rule out the possibility that imperatives are CP structures.

The data are compatible with the CP hypothesis provided that an imperative operator in

spec,CP is admitted.

l9These examples may be acceptable to some. In such cases, however, it is easy enough to see that they are then not interrogatives but declaratives with question intonation. Since only questions and not declaratives license NPIs, declaratives in disguise should not permit NPIs, which is the case, (i) is ungrammatical with the NPI ever.

(i) a. *So embarrassed is Jack ever that he can’t answer a question?


386

It remains to be seen whether the FP hypothesis can also account for the ungrammaticality

of ITAs in imperatives. The structure of an imperative with DO involves only the two

projections FjP and F2P, where FjP is not CP. Although there is no evidence for CP in

imperatives, nothing in principle prevents projecting CP above F{P. If an ITA is moved

into the specifier the data could be derived provided F j0 could move to C° to satisfy the

Operator Criterion. This situation could be blocked with any of three stipulations: 1) C8

cannot project in imperatives, 2) F j8 cannot move to C8, or 3) spec,CP is already filled by

an imperative operator. Neither option is obviously motivated but one is crucially needed to

prevent the ungrammatical data with ITAs from being generated. If an imperative operator

is adopted, it becomes quite clear that the FP and CP analyses are on an equal par with

respect to the ITA data. The full range of data are ultimately compatible with either the FP

or CP hypothesis. Neither analysis is empirically preferable based on ITAs.

The conclusion from this section is that empirical differences between the CP and FP

hypotheses are not apparent. Since the accepted analysis of interrogatives is as CPs, I

tentatively propose that the imperative should also be analyzed according to the CP

hypothesis on grounds of simplicity. Constructions that behave similarly should receive

similar analyses. Recent theorizing (Pollock 1989, Culicover 1991, Chomsky 1993) offers

an FP-type of analysis for English clause structure, one involving multiple functional

projections between CP and VP, but empirical evidence for such proposals applied to

English has been slow in coming and the proposal is usually adopted for theory-internal

reasons alone. The imperative does not provide any new evidence for these analyses.

Making imperatives CPs has the desirable consequence of keeping clause structure constant

across English: declaratives, interrogatives, and now imperatives all have identical phrase

structure. Lacking empirical evidence to decide between the FP and CP hypotheses, the

preference for CP is based on a desire for uniformity. Such a theoretical goal is not


387

unreasonable in that it is, in a sense, the null hypothesis. The language learner presumably

starts out with the simplest hypothesis, namely, that all clauses have at least potentially the

same structure. Thus, the CP hypothesis is the simpler analysis in the larger picture. It

introduces no new clausal structures. I will adopt a CP analysis of imperative clause

structure, bearing in mind that the FP analysis is not actually empirically ruled out.

5 . Summary

The structure that we arrive at for some English imperatives is repeated below.


With the subject in the specifier of IP, the data in (3) and (5) are generated. DO moves from

1° to C° to obtain the alternate order in the data in (2) and (4).

(9) CP

C IP

(DO) subject I’

r^^vp



DOASUBJECT



SUBJECTAD 0



DON ’ S U B JE C T


388

(5 ) a. Those with luggage don’t leave it unattended! SUBJECTa DCW T


The first theoretical question engendered by the structure in (9) and the data above is why

the movement should be optional, particularly in light of its obligatory nature in

interrogatives. The next chapter explores this issue and suggests that the optionality of

T-to-C0 in imperatives should be viewed not with respect to interrogatives but in

comparison to finite clauses more generally, of which interrogatives are one subtype. With

regard to I°-to-C° in the domain of finite clauses, it is evident that the movement is optional

there as well. When it is absent, declarative clauses result. When it occurs, we obtain

interrogatives, which are nevertheless still finite clauses. Imperatives with and without

inversion are then also two clausal subtypes within a domain of non-finite main clauses.

The second theoretical question raised by the structure is what motivates the head

movement. In questions, the movement is widely attributed to a purely syntactic

requirement that C° be filled—for example, there is a strong feature in C° that must be

checked before PF (Chomsky 1995). Consequently, if there is an element in 1° it must

move to C°, and if there is none, one is supplied: support do. The following chapter

addresses this issue for imperatives and argues that the theoretical motivation may very well

be similar, i.e. to check a strong feature in C°, but that, unlike in interrogatives, it is not an

indiscriminate syntactic requirement. In imperatives, only a semantically contentful head

may move to C°, negative don’t or emphatic do but not dummy do without accompanying

emphasis or contraction with negation. If no appropriate head is present, there are no

consequences such as Do-Support. Summarily, I will suggest in the next chapter that

movement to C° in imperatives is a syntactic operation that is semantically driven by the

presence of negation or emphasis. Consequently, do(n’t) will move to C° as proposed

here.


389

If it turns out that there is an imperative operator in spec,CP, as hypothesized earlier, then

a deeper motivation for the head movement is available and could be subsumed under the

Operator Criterion that derives other instances of I°-to-C°. The picture that we arrive at

makes many imperatives look surprisingly like finite clauses and interrogatives, at least

from a syntactic standpoint. The structure in (9) claims that some imperatives and polarity

questions are representationally and derivationally equivalent. The movement proposed for

imperatives is not a new mechanism for the grammar since Subject-Auxiliary Inversion

(I°-to-C°) is a well-known operation in many other areas of English syntax. Thus,

imperative syntax is further brought into line with English clause structure more generally.

From the analysis in (9), the position of the subject in imperatives is always spec,IP. This

is an expected, and welcome, confirmation of the conclusions in chapter 3. Imperative

subjects occupy the same syntactic position that subjects of other clause types occupy.

They are not seen VP-intemally or in some other inflectional projection at the surface. The

evidence in English for subjects originating VP-intemally—primarily quantifier float—is

also replicated for imperatives. Imperatives subjects thus show the same syntactic history

as normal subjects. Summarily, they are syntactically unexceptional.

It is worth pointing out that these arguments concerning the structure of imperatives do not

hinge on the additional imperative data introduced by Davies (1986a): the possibility of

subjects with do, (2) and (3), and the ordering sub jection ’t, in (5). These additional

possibilities were not a crucial part of the argumentation. With more restricted data, for

example the often assumed state of affairs in which imperative subjects appear only

following don’t as in (4) and never after do or preceding either one, the structure argued for

in (9) would be valid for this more restricted set of data alone and the I°-to-C0 inversion

would be obligatory. Accepting Davies’ observations and the full range of data in (2)


390

through (5), however, permits us to generalize the structure to all imperatives, a clearly

desirable simplification. To the extent that such imperatives are less acceptable, a non-

structural account must be found since the syntax I claim does not rule them out.


391

Chapter 6

The Status of Imperative D o N o t

1 . Introduction

This chapter investigates the remaining outstanding issue in the study of English

imperatives: the analysis of imperatives introduced by do not, illustrated in (1). I will call

SUCh examples FORMAL IMPERATIVES.



c. Do not forget!

Examples like these have typically been at the center of debates over the syntactic analysis

of imperatives (Cohen 1976, Hankamer 1977, Schmerling 1977, Davies 1986, Beukema

and Coopmans 1989, Pollock 1989, Zhang 1990). They display interesting restrictions and

frustrating syntactic inertness that have defied a unified analysis. They are also a place in

the grammar where imperatives and interrogatives clearly differ. In this chapter, I

investigate their syntactic behavior and the implications for the central thesis of this work,

that imperative syntax can be assimilated to general English clause structure.

Section 2 lays out the analytical puzzle that formal imperatives pose for English syntax. The

primary difficulty centers around the universally claimed ungrammaticality of imperative


392

subjects in formal imperatives. This is surprising given the wide range of subject options in

negative imperatives with don’t. The section evaluates three proposals in the literature

which take very different approaches to the data. Schmerling 1977 proposes that do not is

an introductory imperative marker, syntactically unrelated to inflectional do not that appears

in finite clauses. As such, it does not permit a subject position in the clause. Akmajian 1984

asserts that the restrictions follow from prosodic, not syntactic, considerations. Formal

imperatives have unique intonation requirements which account for the non-appearance of

subjects. They are absent in finite clauses. Finally, Beukema and Coopmans 1989

attributes the ungrammaticality of subjects in formal imperatives to Case-theoretic reasons.

Subjects do not appear in formal imperatives because Case cannot be assigned to the

subject position. In discussing the three proposals, I offer critical commentary and

conclude that neither a prosodic account nor an imperative particle analysis can be

conclusively ruled out.

Section 3 moves on to consider a structure for formal imperatives. The development is

largely empirically driven but with an obvious bias towards assigning formal imperatives

regular clause structure to the greatest extent possible. This methodology is hopefully

supported by the results of the previous chapters which attempted to show that other

aspects of imperative syntax were unexceptional. The section addresses two specific issues:

the existence and location of a subject position in formal imperatives and the positions of do

and not. Regarding the first, section 3.1 suggests that a subject position must be available

in formal imperatives just as in other clauses. The difficulty arises in determining where

this position is with respect to the formatives do and not. The data conspire to make this

determination nearly impossible. It is nonetheless proposed that the position is potentially

either above do or below not but never between the two. This conclusion is supported by

investigation of the second issue, the analysis of do not. Section 3.2 observes that the


393

string consisting of a form of do followed by not is generally inseparable. This obtains in

declaratives but not interrogatives, where T-to-C0 movement o f do occurs. The

inseparability is also observed in formal imperatives. Since formal imperatives pattern with

declarative clauses, there must also be no inversion there. Do does not move to C° in

formal imperatives, unlike in interrogatives. These results determine the position of do,

not, and the null subject in ordinary formal imperatives. They also more clearly reveal the

similarities and differences between I0-to-C0 in imperatives and interrogatives, which can

be summarized as follows: T-to-C0 movement in imperatives is a purely formal, syntactic

device that is always required. Imperative I°-to-C0 is semantically conditioned, applying

only when the semantically contentful elements emphatic do and negation are present. As a

result, interrogatives are a canonical environment for Do-Support but imperatves are not,

the contrast between (2) and (3).

(2) a. Do you think about it much?

b. *You think about it much? (ungrammatical as an interrogative)

(3) a. *Do everyone think! (ungrammatical as a non-emphatic imperative)

b. Everyone think!

Section 4 offers a formal analysis of imperative word order within the Minimalist Program

and checking theory. Under the assumption that overt movement to C° reflects a strong

feature in C° which must be checked before PF, the contrasting behavior of the two clause

types reduces to the identity of the categorial feature that checks this strong feature in each

case. In interrogatives, the categorial feature is the V-feature of 1° (Chomsky 1995). In

imperatives it is the head Z° which has as its realizations negation and emphatic affirmation

(see chapter I and Laka 1990). This accounts for the semantic nature of T-to-C" in


394

imperatives. Section 5 closes with some general observations regarding the larger issues in

imperative syntax.

A word about the data is in order before continuing. Do nor-imperatives are highly formal

in nature and typically require substantial emotional commitment on the part of the speaker.

Consultants often find difficulty in making judgments on examples that are greatly more

complex than the standard cases in (1). This awkwardness will surface in the investigation

below, appearing as uncertainty and disagreement over several important judgments. In the

presence of this discomfort, we will be forced to proceed rather slowly, to consider

alternative sources of evidence when possible, and, in some cases, to idealize the data to a

certain extent.

2 . The Problematic Data

Chapter 4 demonstrated that emphatic do in imperatives and finite clauses showed

significant syntactic and semantic parallels and that, under fairly standard theoretical

assumptions, could be given identical analyses. Likewise, it was also concluded that

negative imperative don't too has transparent analytical parallels to corresponding tensed

forms, including its origination in an inflectional head. A maximization of this analytical

line would seek to fully assimilate all occurrences of do in the two clause types to a single

structural analysis. The remaining situation to consider is the one in which do occurs with

negation but the two are not contracted; they remain morphologically separate as a form of

do followed by not. I consider this final case here.

Conventional wisdom has it that don’t is a phonological contraction of do not. Under this

assumption, the imperatives in (4) with do not are actually expected, as Davies 1986a: 118

points out. The relationship between (4) and (5) is transparent.


395



c. Do not forget!

(5) a. Don’t pass up these deals!

b. Don’t walk on the grass!

c. Don’t forget!

No such expectation arises if don’t is an imperative particle, the analysis argued against in

chapter 4. At the same time, the full expected paradigm is not borne out, allowing the

argument to work in the other direction (Zwicky 1987). It is well-known that formal

imperatives differ from finite clauses and imperatives with don’t primarily in the

possibilities for overt subjects. Simply put, subjects are traditionally claimed to be ruled out

in formal imperatives. Thus, alongside the tensed examples in (6) and the don 7-imperative

in (7), we have neither of the formal imperatives in (8).

(6) a. (?)Do not any of you with rural childhoods like dogs?

b . Do you not like dogs?

(7) a. Don’t you open that door!

(8) a. *Do not you open that door!

b . *Do you not open that door!1

1 Given that I have recognized the possibility of subjects with emphatic imperative do, examples like this one do have a grammatical reading in which do is emphatic and not is constituent negation. Under such a reading, do is stressed and is either persuasive or contrastive. Further examples of this are given in (i). They differ from the ordinary formal imperatives under investigation in which do is neutral.

(i) a. ?Do AT LEAST SOME of you not snub our guest!b. ?For heaven’s sake, of all people, DO YOU not give me a hard time!


396

I take questions such as (6)a in which the subject follows both the auxiliary and not to be

grammatical though stilted. I will call such examples FORMAL QUESTIONS. The pattern

seems acceptable as a stylistic device used in very formal registers when the subject is

‘heavy’ (Zwicky 1969; Zwicky and Pullum 1983; Zwicky 1987). Given the marginal

nature of the interrogative, the ungrammaticality of the corresponding imperative in (8)a is

not immediately troublesome. It could reasonably be argued that the stylistic device that is

active in (6)a is not appropriate for imperatives. The ungrammaticality of (8)b, on the other

hand, is entirely unexpected. The sharp contrast between (6)b and (8)b is a primary

impediment to claiming that imperatives behave syntactically like finite clauses. In the

remainder of this section, I survey approaches that various researchers have taken to the

ungrammaticality of (8)b, which, for simplicity, I will label the DO NOT-PROBLEM. None

are entirely convincing but also not easily falsifiable. Section 4 returns to offer a new

proposal, based on the empirical-oriented claim defended in section 3 that formal

imperatives do not actually show any inversion of empty do, unlike in interrogatives. It will

be proposed that inversion in imperatives is semantically driven, by negation or affirmation

only. Under this view, the word order pattern in (8)b is ungrammatical because there is no

licit derivation for it..

2 .1 . Schmerling 1977

Schmerling 1977, working in a transformational framework, observes that taking don’t to

be a contraction of do not results in a chaotic picture. Maintaining a contraction analysis,

she claims, commits us to an unrevealing description of negative imperatives: contraction is

optional unless there is a subject, then it is obligatory. To account for the do not-problem,

Schmerling analyzes don't and do not as unrelated. While don’t receives an auxiliary

analysis much like the one espoused in chapter 4, do not is analyzed as a frozen expression


397

introduced directly by the phrase structure rule in (9). In essence, do not is given an

imperative particle analysis like the one rejected for don’t in chapter 4. Imp in (9) is an

‘initial symbol’ distinct from S. The do nor-problem is transparently solved because no

subject NP at all is introduced by the rule in (9).

(9) Schmerling 1977 Analysis o f Formal Imperatives

Imp do not VP

A modem rendering of Schmerling’s proposal might give formal imperatives the structure

in (10). Do not is a VP-adjoined introductory imperative particle.2

(10) VP

XP VPZXdo not pro V

V (complement)

One aspect of Schmerling’s analysis cannot be easily captured in current terms, however,

and it is a crucial piece: the complete absence of a syntactic subject. The lack of a subject

NP term in Schmerling’s rule (9) entails that an overt NP will never appear—the desired

result. Because of argument structure realization requirements and restrictions deriving

from the Projection Principle (Chomsky 1982, Jaeggli and Safir 1989b, and the discussion

in chapter 3), an argument corresponding to the subject must be syntactically present.

Furthermore, I offer evidence below that formal imperatives do have a syntactic subject,

like other imperatives. I will adapt Schmerling’s analysis so that one is present and

occupies the VP-intemal, specifier position, as shown in (10). We can still derive the result

2For the sake of discussion, I will ignore how to prevent inflectional projection(s) from extending this phrase marker. It might be claimed that there are no features to check and, thus, by economy, no further structure is needed, or permitted.


398

that the subject will not be overt by observing that spec,VP is not a Case position.

Assuming that pro is Caseless (Jaeggli 1986, Roberge 1990, Authier 1992), the subject

will then be restricted to non-pronounced pro. Summarily, a 1990’s translation of (9) is the

structure in (10) and the analytical claim in (11) that a subject may occupy the specifier of

VP but cannot be phonetically realized in this position because it is not assigned Case.

(11) Assumption to Accompany Particle Analysis o f Do Not Imperatives

Subjects in spec,VP are restricted to pro

Schmerling offers VP ellipsis and floating quantifier data in support of the particle analysis

of do not. She observes that neither phenomenon is permitted in formal imperatives,

although she states that most speakers are uncomfortable with the judgments. This argues

against giving formal imperatives a structure similar to finite clauses because both

phenomena are productive in that domain. Schmerling’s examples are in (12) and (13),

with her judgments given. I consider each phenomenon in turn and claim that neither one

conclusively favors a particle analysis of do not.

(12) ?*There is no reason to be nervous, so do not be.

(13) ?*Do not both talk at once.

Within the current framework, the syntactic interpretation of the ellipsis example in (12) is

that neither do nor not is an inflectional head that is capable of licensing a null VP. The

judgment is precisely as expected if do not were an imperative particle which would be inert

for VPE licensing. Although the example in (12) is indeed awkward, some speakers are

less willing to completely reject similar VPE examples. In most cases, a high degree of

emphasis and emotional commitment is required but the result is at least conceivable and

often fully acceptable, (14). At best, then, the data are inconclusive.


399

(14) a. [I’m going to open the oven and peek at the pie.]

Do not! You’ll ruin it.

b . [They’ve convinced me to get a mohawk!]

Absolutely do not! You will not be allowed back in this house!

The quantifier float data submit to the same conclusion; there is a great amount of variation

in this domain as well, both across and within speakers. In line with Schmerling’s (13),

some of the examples in (15) are judged to be degraded but many are accepted. For

example, (15)a is perfect for all my consultants.

(15) a. Please do not all talk at the same time!

b . For heavens sake, do not all be standing around when he walks in!

c. Do not both apply for the position!

d . Do not all be so eager to write him off!

Assuming the data to be ungrammatical, for Schmerling, they argued against do not being

integrated into the syntax as an auxiliary because in the formulation of quantifier float “it

seems reasonable to conclude that Aux must be mentioned (Schmerling 1977:15)”. Since

the rule introducing do not, (9), does not mention the category Aux, quantifier float was

not expected. Under current assumptions about quantifier float, this suggestive restriction

does not easily carry forward. I believe that the same result can nevertheless still be

achieved. Sportiche’s analysis described in chapter 1 analytically links floated quantifiers

with subject position(s) rather than the Aux (1°) position. Since the structure in (10) has no

derived subject position, there can be no stranded quantifiers. On the other hand, it does

not rule out the quantifier being a constituent with pro. This might be made to fall under the

independently needed restriction on pro and quantifiers from chapter 3, repeated in (16).

The subject specifier position in (10) does not satisfy this observation because there is no


400

c-commanding head, only the imperative adjunct. This results in floating quantifiers being

completely ruled out in do not imperatives.

(16) a specifier position whose sole phonetic content is a floated quantifier must

be properly governed by an overt head

The updated analysis in (10) is therefore equivalent to Schmerling’s phrase structure

proposal in its predictions regarding the absence of floated quantifiers. The imperative

particle analysis excludes quantifier float completely and predicts all of the data in (13) and

(15) to be fully ungrammatical.3

In this subsection, I have revised an imperative particle analysis of do not originally

proposed in Schmerling 1977. Her two arguments in favor of it, VP ellipsis and floating

quantifiers, I claim to be based on inconclusive evidence. In large part, the subtlety and

uncertainty of the data make the argumentation unpersuasive. On the other hand, the

analysis is empirically tenable and perhaps even makes some correct predictions, depending

upon what the data are taken to be.

3An adverbial analysis of floated quantifiers (Doetjes 1992, Baltin 1995, Bobaljik 1995) also sheds no light on the data. Under an adverbial approach, the descriptive restriction required to prevent the ungrammatical examples is somewhat unusual: adverbs must be disallowed in formal imperatives in the pre-verbal but post-do not position(s). A condition of this sort would seem to be incorrectly overly restrictive since the data in (i) indicate that all other kinds of adverbs are compatible with formal imperatives. Floated quantifiers would need to be singled out as a special kind of adverb.

(i) a. Do not always be so quick to criticize!b. ?Do not normally start cleaning until the customers have left! (S-adverb)c. Do not so rudely interrupt us next time! (VP-adverb)d. Do not simply skim the material, read it thoroughly! (E-adverb)


401

2 .2 . Akmajian 1984

Akmajian 1984 takes a very different, largely non-syntactic approach to the

do nof-problem. He argues that the ungrammaticality of examples like (8) are a result of

prosodic restrictions particular to imperatives.


b . *Do you not open that door!

In this section, I lay out his analysis and consider arguments against it from Zhang 1990.

Rejecting Zhang’s objections, I conclude that Akmajian’s prosodic approach to the

do nor-problem is not obviously incorrect. Further investigation of English prosody that is

outside the scope of this work is required to more carefully evaluate the reasonableness and

generality of the proposal.

The proposal begins with the observation that in formal imperatives either do or not must

be an intonation center, where an intonation center is characterized as follows: “a syntactic

constituent forms an intonation center when (a) its nuclear stress level is high enough to

block phonological reduction processes and (b) its phrase-final intonation contour is non

falling (Akmajian 1984:11)”. This claim is summarized in (17) and illustrated by the data in

(18) (Akmajian’s (ii) from footnote 12).

(17) Formal Imperative Intonation Restriction (Akmajian 1984)

In a formal imperative either one or the other of the formatives do and not

must be an intonation center


402

(18) a. Do not leave the room!

b. Do ndt leave the room!

c. *D5 not leave the room!4

The core of Akmajian’s analysis is the more general intonational principle for imperatives in

(19). An imperative subject must be the unique intonation center preceding the VP.

(19) Imperative Intonation Principle (Akmajian 1984)

When an imperative sentence has an overt subject, the subject must be the

only intonation center preceding the verb phrase

For formal imperatives, the tension between (17) and (19), either do or not having to be an

intonation center on the one hand and an overt subject having to be the only intonation

center on the other, results in subjects being excluded entirely. The examples in (20) and

(21), which are (8)a and b with possible stress patterns marked, will always violate the

intonation requirements. (20)a and (21)a violate (17) while the remainder of the examples

violate (19).

(20) a. *D o n o t ydu open that door!

b . *Do n o t you open that door!

c. *D o not you open that door!

d. *Do not y ou open that door!

4Akmajian surmises that unstressed do not obligatorily contracts to don 't. Akmajian’s restriction is ambiguous and silent regarding the status o f (i) in which both formatives are stressed. It seems acceptable.

(i) D<5 n6t leave the room!


403

(21) a. *D o you not open that door!

b . *Do you no t open that door!

c . *D o you not open that door!

d. *Do you not open that door!

Zhang 1990 offers two incorrect predictions of Akmajian’s prosodic account. First, it

incorrectly rules out grammatical examples of topicalization in imperatives, (22) and (23)

(Zhang’s (1990:113) (43) and (44)). Additional examples were given in chapter 5.

(22) a. You put the apples in the refrigerator!

b . The apples, y6u put in the refrigerator!

c. This pie, none of you touch!

(23) a. Don’t you put the book on the stove!

b . The book, don’t y<5u put on the stove!

c. This pie, don’t any of you touch!

The data in (22)b and c and (23 )b and c should be ungrammatical because they violate the

imperative subject intonation restriction. There are two intonation centers preceding the VP,

the topic and the subject. This difficulty with Akmajian’s proposal can be resolved if we are

more precise about the domain of the Intonation Principle than simply “preceding the verb

phrase”. Intonation peaks are typically associated with individual intonational phrases,

which are mapped from the syntactic structure (Selkirk 1986). If Akmajian’s principle

applies only in the intonational phrase that includes the subject, the examples might be

predicted. To see whether this is tenable, a means of determining intonational phrasing is

required.


404

Selkirk 1986 and Hale and Selkirk 1987 provide an explicit, end-based algorithm for

calculating prosodic phrases. They are mapped from the syntax by reference to the left or

right edge of some syntactic constituent. Intonational phrases (IntP) it turns out are often

mapped by reference to the right edge of ungovemed maximal projections (Hale and

Selkirk 1987, Aissen 1992) and this is a reasonable working hypothesis for English:

(24) English Intonational Phrasing Algorithm

the right edge of an ungovemed maximal projection corresponds to the right

edge of an intonational phrase

Related to this, Chen 1987 notes the importance of the adjunct/complement distinction in

determining intonational phrases. Adjuncts but not complements are typically ungovemed.

If topics are adjoined to IP (Baltin 1982, Lasnik and Saito 1992, and chapter 5 of this

work) as shown in the structure in (25), then they will obligatorily form a distinct

intonational phrase from the remainder of the sentence. The right edge of each of the two

ungovemed maximal projections corresponds to a distinct intonational phrase.

(25) IP

XP IP

topic

|<—IntP—>|<—IntP—>|

This permits a reformulation of Akmajian’s restriction which accounts for Zhang’s

problematic imperative data in a natural manner, (26). Intuitively, the subject does not need

to be the only intonational center in what is potentially a rather large and heterogeneous

prosodic region, it need only be the intonational center of a particular local domain.

Elements outside of this local domain may also be intonational centers.


405

(26) Imperative Intonation Principle (revised)

When an imperative sentence has an overt subject, the subject must be the

only intonation center in its own intonational phrase

Zhang’s second argument against Akmajian’s proposal centers on the putative parallel

distribution of emphatic do and please. Zhang asserts that, as with emphatic do, imperative

subjects are impossible with please, (27) (Zhang’s (1990:113) (46)).

(27) a. *Do you open the window!

b . *Please you open the window!

Akmajian correctly predicts the ungrammaticality of (27)a but Zhang claims that the

analysis is not capable of accounting for (27)b. The account of (27)a with do runs as

follows: Emphatic do obligatorily bears stress. Since the imperative subject must also, its

intonation pattern, in (28)a, violates the Intonation Principle in (26) because there are two

intonation centers. Please is unlike do, however, in that it need not bear stress. According

to Zhang, (27)b should thus be grammatical, contrary to fact, because its intonation in

(28)b does not violate (26). Therefore, (27)b is a counterexample to Akmajian’s analysis.

(28) a. *D6 you please open the window!

b . *P1 ease ydu open the window!

The logic here is certainly wrong however. The fact that (28)b satisfies the Intonation

Principle does not mean that it cannot be ungrammatical for some other reason. The

example is thus not a problem for Akmajian’s analysis, though it is in need of an

explanation. I suggest that the claimed parallel between please and do is only apparent.

Imperative subjects are not impossible with please, as the examples in (29) demonstrate.


406

The problem with the particular example in (27)b is purely pragmatic, as Zhang, in fact,

later argues.

(29) a. Please somebody open the window!

b . Please nobody forget the date!

c. Everybody please take a name tag!

d. Everyone get in line please!

Despite Zhang’s objections, there does not seem to be an empirical deficiency with

Akmajian’s analysis. It seems relatively certain that intonation plays into the judgments on

formal imperatives, witness the formal imperative ellipsis data back in (14); however, I am

ill-prepared to justify this claim and it is unlikely to be the whole story. I accept the validity

of Akmajian’s general strategy and leave the necessary further investigation of his

intonation principles for future work.

2.3 . Beukema and Coopmans 1989

Beukema and Coopmans 1989 (henceforth B&C) take a purely syntactic approach to the do

nor-problem, relying on Case Theory to rule out overt subjects in formal imperatives. The

ungrammatical data result because Case cannot be assigned to the imperative subject. For

B&C, the source of Case assignment to spec,IP in imperatives is C°. One possible Case

assigner in imperatives is do, which they claim is inserted directly into C°, bypassing 1°.

Infl is not invoked in the analysis except that not is a realization of 1°. These assumptions

provide a partial analysis of formal imperatives, in conjunction with ancillary assumptions

that I introduce below.

The account of the formal imperatives runs as follows. The structure of (8)a, a formal

imperative with the subject following do not, is given in (30). In the derivation, not has


407

moved from I8 to adjoin to C \ The example is ungrammatical because the subject cannot

now be assigned Case given the configuration of do and not shown. Case assignment is

blocked by “making structural case-assignment sensitive to the branching effect of not-

affixation (B&C 1989:431)”. In other words, head adjunction of not to C° interrupts the

ability of C° to govern the subject position. Under this assumption, a subject will never

follow do not, predicting the ungrammaticality of (8)a.


(30)

do not

case assignment

The problem with (30) can be remedied by having not cliticize and reduce onto do, yielding

Don't you open that door! Don’t on B&C’s analysis is one phonological word and does not

block Case assignment.

Regarding this head adjunction of not shown in (30), it must be obligatory for the analysis

to succeed. B&C assume that it is. If not can be raised for the purposes of affixation, then

it must. I state this in (31).

(31) Not Adjunction Requirement (Beukema and Coopmans 1989)

Not must adjoin to a head host if it can

This assumption also serves to block (8)b, in which the subject interferes between do and

not. In such examples, not has evidently not moved to C° and adjoined to do. Because it


408

has not done so, the result is ungrammatical. (8)b is not actually a Case violation; it is a

violation of (31).

(8) b. *Do you not open that door!

This approach has several undesirable aspects. First, the assumption that head adjunction

blocks government and Case assignment contradicts the widely accepted conclusions of

Baker 1988. He argues extensively that head adjunction structures created by movement are

transparent for the purposes of government. Second, the account of the ungrammaticality

of (8)a should extend, incorrectly, to finite clauses. The declarative examples in (32), with

the B&C structure in (33), have an overt subject and the sequence do/did not in 1°. The Not

Adjunction Requirement in (31) requires that not in (33) be adjoined to 1° as shown, just as

in (30). 1°, the source of nominative Case (Beukema and Coopmans 1989:427), should

then be unable to discharge its Case to the subject, just as in the comparable situation in

(30). This is contrary to fact. That Case assignment transparently does not fail indicates that

not does not interfere with government in the simple way envisioned.

(32) a. We do not approve.

b . John did not consider that possibility.

(33)

not approve

Lastly, B&C point out that (34) is problematic for their analysis. In this example, Case

assignment to the subject position, which for them is occupied by the A trace of an empty


409

operator (see Beukema and Coopmans 1989 and chapter 3, section 4.1.1) is not blocked.

B&C have no principled account of this.

(34) Do not be foolish!

The core of B&C’s approach to the do /tor-problem is a special morphosyntax of not.

Crucial to their account are the analytical claims that Y) do and not form a syntactic unit

when possible (the Not Adjunction Requirement) and 2) that not in this configuration with

do blocks Case assignment. In section 3.2,1 return to offer empirical support for the first

claim. Despite this, there does not seem to be any interaction with an ability to assign Case.

In this section 2 I have surveyed three approaches to the analysis of formal imperatives. I

have acknowledged the tenability of an imperative particle approach and recognized the

potential influence of prosody on formal imperatives. I have also tentatively rejected a

problematic Case-based account of formal imperatives. I will fully reject such an analysis

below. In the following two sections I develop my own hybrid proposal. The account

centers around two basic claims: 1) formal imperatives have a subject position and 2) as

strongly suggested by the analyses in Schmerling 1977 and Beukema and Coopmans 1989,

do not often behaves as a morphologically inseparable unit.

3 . A Structural Resolution

In this section, I propose a phrase structure for formal imperatives. The account continues

to develop the central thesis that imperative syntax is largely unexceptional and I argue for

formal imperative structure that makes them look like finite clauses in many respects. Some

remaining data will require ancillary, non-syntactic explanations which I hope to show are

plausible. Two basic sets of facts that have not previously been explored serve as the


410

foundation for the analysis. Section 3.1 considers the issue of where the subject position is

in formal imperatives. It introduces new data that determine the extent to which an overt

subject is ever permitted, a possibility completely denied by all researchers and one that is

most clearly ruled out by a Case-based analysis of formal imperatives. The section

concludes that some subjects are marginally allowed in formal imperatives, either preceding

or following do not, and it offers structures for the data which parallel those for similar

finite clauses. Section 3.2 investigates the status of the string do*not and proposes an

explanation for why subjects are completely impossible intervening between do and not,

the do nor-problem. Schmerling 1977 claimed that do not is a single frozen expression and

Beukema and Coopmans 1989 asserted that there is a particularly tight morphological link

between the two formatives which would prevent their being linearly separated. I offer

empirical support for this view and show how it provides a diagnostic for the position of

support do and whether or not it has moved to C°: only when do remains in 1° is it linearly

inseparable from not. This obtains in formal imperatives, indicating that do there does not

move to C°. This contributes to a resolution of the do nof-problem and the structure of

formal imperatives. Section 4 proposes a formal syntactic account of imperative word

order.

3 .1 . A Subject Position

The usual diagnostics, details of which were discussed in chapter 3, can be enlisted to

confirm that formal imperatives do have a syntactic subject position even if a subject noun

phrase is not overt. Under the analysis in chapter 3, this position is occupied by pro. The

data below illustrate control of infinitival complements and adjuncts, (35) and (36), control

of reflexives and reciprocals, (37) and (38), binding in idioms, (39), and derived subjects

including passives, (40), subject-to-subject raising predicates, (41), unaccusative verbs,


411

(42), and tough movement predicates, (43). Theory-internal considerations, discussed in

chapter 3, lead to the same conclusion.

(35) a. Do not promise to do things that are beyond you!

b . Do not expect to get rich immediately!

c. Do not ask what to wear again!

d. Do not decide who to invite without consulting me first!

(36) a. Do not applaud just to please the comedian!

b . Do not try those fad diets to lose weight!

(37) a. Do not demean yourself in public!

b . Do not withdraw yourself from the contest!

(38) a. Do not consult each other on these questions!

b. Do not ignore each other!

(39) a. Do not crane

b. Do not hold

your*their*my

your*Sue's*my

neck and stare, it’s rude!

breath in the tunnel!

(40) a. Do not be examined t by that quack!

b . Do not be alarmed t by the smoke!

c . Do not be caught t speeding!

(41) a. Do not appear t to be sticking your nose in others’ business!

b . Do not stop t writing when the bell rings!


412

(42) a. Do not emerge t dirty from that shower!

b . Do not surprise me t, I don’t like surprises!

e. Do not come t at the same time!

(43) a. Do not be easy for us to spot t\

b . Do not be too pleasant to talk to t or he’ll never leave!

Given that a subject position does exist in formal imperatives, we would want to know

where it is with respect to do and not. It is in trying to answer this question that there is an

impasse. I have been unable to formulate useful diagnostics5 so I pose the question to

native speakers in the following way: if a subject had to appear in a formal imperative,

where would it go? Surprisingly, intuitions on this question are in agreement. When we

consider the options of a subject before, in the middle of, or after do not, it is only the

second option, with the subject intervening between do and not, which is robustly

ungrammatical. The other two word orders are degraded but can be improved with

appropriate emphasis and intonation. In the data below, the judgments given indicate

contrasts, not absolutes, and this is what is most significant.6

5The inconclusive nature of Schmerling’s floating quantifier data indicate that they are of limited use. frTo the extent that historical data can be enlisted in a synchronic study, it supports these observations. I have found two instances of subjects in formal imperatives and both have the subject, you, following do not. I have not found any examples of the pattern cfoAsubjectAnor. The first example in (i) is from Visser 1963b:1541; the second is from Poutsma 1928:107 taken from Sweet 1891-1898:§506.

(i) a. Do not you add to the idle race! (1807)b. Do not you do that! (-1898)


413

SUB JECTA£>0AWOT





D0ASUBJECTAWOr


*Do somebody not desert me!


D0/W0rASUBJECT

(46) a. ?Do NOT ANY of you touch that cake! It’s for the wedding and if any of it is

missing heads will roll.

b. I know I’ve done wrong but I can’t survive on my own.

?Oh please, do not ALL of you desert me!

c. ?DO not YOU, of all people, insult me in this heinous and base manner!

d . ??Do not EVERYone stand off by himself in the comer, come join the party!

In finite clauses, all three word order possibilities are available. If imperatives are to be

assimilated to finite clauses, all of the above should be acceptable. Assume for now the

idealization of the data. The first case, in (44), is expected. The data parallel normal

declarative clauses and those imperatives in which subjects may precede do(n’t). The

examples in (46), with the subject following do not are parallel to formal questions, which

I suggested above are grammatical though stilted. The unexplained pattern is (45) in which

the subject absolutely cannot appear intervening between do and not. This is the normal

position for subjects in negative questions:


414

(47) a. Did somebody not tell him?

b. Do you not like the salad?

The contrast between (45) on the one hand and (44) and (46) on the other is, I believe,

worth accounting for. While prosody or other not-strictly-syntactic effects might be

responsible for the degraded status of (44) and (46), there is something deeper wrong with

(45). This is a reincarnation of the do nor-problem, which I will return to in the next

section. In the remainder of this section, I propose structures for the examples in (44) and

(46).

Given the direction that this work has taken, the null hypothesis for the data is that they

have the structure of their finite clauses counterparts. The imperative examples in (44) are

analyzed as (48), the structure of a negative declarative clause.

(48) IP

subject I’

f ^ ^ N e g P

do Neg VP

not

The null hypothesis regarding the structure of the examples in (46) with the imperative

subject following do not, is that they have the same structure as corresponding formal

questions, exemplified further in (49). I consider two analyses for formal questions, in

(50) and (51).


415

(49) a. Do not any of you who grew up in South Dakota know what a good ear of com

tastes like?

b . Will not the electorate of this country consider that they have a right to know

these facts?

c. Could not this group of sixteen energetic youngsters travel down the Colorado in

a bark canoe?

In the first proposal, (50), the auxiliary and not have together moved to C°. The subject is

in its canonical spec,IP position. This could arise by either having not adjoin to 1° before

I°-to-C° occurs, as shown, or by positing some sort of backformation de-contraction in

formal contexts.

(50)

C subject

An alternative analysis of formal questions would place the subject to the right of its normal

position rather than displacing not from its. (51) hypothesizes that the auxiliary and not

occupy their usual positions for questions but that the subject is lower, in spec,VP. The

proposal derives from a suggestion of Zwicky and Pullum 1983 who maintain that “such

sentences result from a stylistic option (quite independent of SAI) that allows heavy subject

NPs to be displaced rightward across not (Zwicky and Pullum 1983:506).”


416

(51)

auxiliaryj

not subject V

The arguments used in chapter 5 to demonstrate that imperative subjects are not a low as

spec,VP can be replicated to argue against the structure in (51). I merely present the crucial

data here; chapter 5, section 3 should be referred to for details of the argumentation. The

data indicate that it is not the position of the subject in formal questions that is exceptional.

(52) shows that stranded quantifiers are still possible in formal questions. Consequently,

the subject must be in a derived position.

(52) a. Will not the supporters of free speech all vehemently protest this outrageous

violation of our constitutional rights?

b . Do not the two conference representatives from the medical community both

support appropriation of additional research funds to this area?

(53) are examples of VP ellipsis applied to formal questions. They show that the subject

may remain after ellipsis has applied, indicating that it is not internal to VP.


417

(53) a. John has volunteered to join the search. Will not any of the rest of you 0?

b . ?They say that this variety of rose is very disease-resistant and well-suited to our

micro climate. Is not that variety over there with the dark green leathery foliage 0

as well?

c . ?We are aware that certified organic farms qualify for matching funds but do not

those which are ineligible for certification 0?

Lastly, (54) demonstrates that S-adverbs are possible in formal questions and they may

appear below the subject. This is not predicted if the subject is as low as VP.

(54) a. Will not the electorate of this country likely consider that they have a right to

know these facts?

b . Could not this group of sixteen energetic youngsters probably travel safely down

the Colorado in a bark canoe?

c. Is not that certainly the reason why you lied to the jury, Mr. Smith?

The proposal then is that the structure of both formal questions and formal imperatives with

the subject following do not is (50), with do not in C°. Ideally, one would like to probe the

imperative data directly; however, its tenuous nature makes any reliable contrasts with these

diagnostics unlikely. I am forced to appeal to a presumed parallelism with formal questions

in deciding a structure for formal imperatives. The preceding chapters have hopefully

provided some motivation for assimilating imperatives to standard clause structure in this

more murky domain.

I am claiming that structures can be assigned to the formal imperatives with subjects in (44)

and (46). At the same time, the data are degraded to a significant degree. This is not a

contradiction since other factors may conspire to rule the data out besides syntactic ones,

just as is the case with formal questions whose grammaticality is often questioned. I have


418

no concrete proposal to offer, however. One potential avenue for exploration is that,

whatever the source of the degradation in formal questions is, it carries over to formal

imperatives. As I indicate in the conclusion, this is clearly one area in need of additional

investigation.

Examples as in (44) and (46) sound best when the overt subject is quantificational or

otherwise related to the addressee—the subjects in (46)a, b, and d are quantificational,

those in (46)a and b are second person partitives, and (46)c contains a contrastive second

person subject. These same subject preferences were seen in other imperative types and the

reasons behind the apparent restrictions were explored in chapter 3, section 4. The null

hypothesis that I will put forth is that the same mechanisms that lead to a favoring of

quantificational subjects and the like in other imperatives are at work in formal imperatives;

the mechanisms are not distinct. Because of the already marginal nature of subjects in

formal imperatives, however, the need for a felicitous subject is even more crucial.

Given that a subject is possible in formal imperatives even if only marginally so, there can

be no purely Case-oriented analysis of the restrictions on formal imperatives. This makes

proposals like B&C’s in section 2.3 untenable and eliminates this family of analyses from

further consideration. In addition, if the data in (44) with the subject preceding do not are

relatively acceptable, they argue against the imperative particle approach of section 2.1. A

classic characteristic of the imperative particle is its clause-peripheral positioning.

Proposing a particle do not that occurs in exactly the same structural position that existing

syntactic mechanisms would generate the inflectional heads do and not is a clear and

unnecessary duplication in the grammar. This same line of reasoning was provided in more

detail in chapter 4 to reject a particle analysis of do(n’t).


419

To summarize, this exercise has offered data which I take to indicate that there is a subject

position available in formal imperatives which is spec,IP. Additionally, do is initially in

1° and not is in Neg° yielding the structure in (55). As in formal questions, and roughly to

the same degree of felicity, Neg° may raise to 1° and move as a unit into C° to yield the

alternative structure in (56).

(55) Base Structure o f Formal Imperatives

IP

subject V

I NegP

do Neg VPI

not

(56) A Derived Structure for Formal Imperatives

CP

Ij C subject V

I Negk tj NegP

do not tb VP

Given these structures for formal imperatives, it should be evident that my predictions

regarding Schmerling’s VP ellipsis and floating quantifier data, in (14) and (15) below, are

that they should be acceptable because the examples will potentially have the same

structures available for finite clauses.

(14) a. [I’m going to open the oven and peek at the pie.]

Do not! You’ll ruin it.


420

(15) a. Please do not all talk at the same time!

b . (?)For heavens sake, do not all be standing around when he walks in!

For some speakers, I suggested that this was the case. If so, the data cannot be taken to

support an imperative particle analysis of do not. Further investigation of the phenomena

and the imperative data itself might suggest a reason for the degradation. What is still

completely impossible and not explained by the above structural analysis is the option,

which regularly occurs in questions, of do moving to C° alone—the do nor-problem. I turn

to this directly.

3 . 2 . The Sequence do not

The approach to the do rcof-problem in this section is as follows. I demonstrate in section

3.2.1 that the sequence of do followed by not when they are in 1° and Neg°, respectively,

as in declarative clauses, exhibits a particularly tight morphological connection that does not

permit any kind of constituent to intervene. Subjects in interrogatives are able to do so

because do is not in 1° but has moved to C°. This provides a probe on the relative position

of do when sentential not is present. Section 3.2.2 demonstrates that the sequence do not in

formal imperatives shows this same inseparability. Formal imperatives pattern with

declaratives and not interrogatives with respect to the syntactic behavior of doAnot,

indicating that do is in 1° not C°. The do nor-problem thus reduces to the purely analytical

task of explaining why I°-to-C° is not available for do alone in formal imperatives. This

reduction constitutes an advance because it isolates the specific source of the do not-

problem: the head movement of do is unavailable.


421

3.2.1. A Diagnostic for the Position o f do

An important element of B&C’s account of the do nor-problem was the idea that do and not

must form a syntactic unit when possible, (31).

(31) Not Adjunction Requirement (Beukema and Coopmans 1989)

Not must adjoin to a head host if it can

On the one hand, it was seen that this requirement alone cannot provide an account of the

do nor-problem because it must be weak enough to permit I°-to-C° movement of do in

questions where a subject transparently intervenes between do and not. It cannot then block

the same movement in imperatives. On the other hand, (31) is in some sense correct when

do and not are already adjacent heads in a structure. When a form of do is followed by

sentential not, the two must indeed be strictly string adjacent as I show below. The intuition

behind (31), then, provides a diagnostic on the syntactic position of do with respect to

sentential not.

In non-inverted declarative sentences, nothing can occur between a finite form of do and

sentential not, neither floating quantifiers nor various adverbs.7 I verify this claim below

and demonstrate that the restriction is not obviously syntactic since it does not hold when

other elements, modals and finite auxiliaries, appear in 1°. The theories of adverb placement

and floating quantifiers from chapter 1 correctly allow these constituents to intervene

between 1° and Neg°. The situation must only be ruled out when 1° contains a form of do.

7ParentheticaIs could also be used to illustrate the same point.


422

The a examples in (57) and (58) show that a floating quantifier may not appear between do

and sentential not. The remaining examples in b and c indicate that they are acceptable

elsewhere in the sentence.

FLOATING QUANTIFIERS

(57) a. *The children did all not stand around.

b . All the children did not stand around.

c. The children did not all stand around.

(58) a. *The twins did both not vote.

b . Both the twins did not vote.

c. The twins did not both vote.

In contrast, in the data in (59), floated quantifiers may intervene between a modal or finite

auxiliary and sentential not.8 The examples have a structural analysis as in (60) for (59)b in

which the quantifier is stranded in spec,NegP.

**It is necessary to confirm that these examples can represent instances of sentential negation, as desired, and not solely constituent negation. Chapter 2, section 5.2.2, discussed four tests for distinguishing the two types of negation. Of use here is the observation that sentential negation, but not constituent negation, can take scope over VP adjuncts, thereby introducing scope ambiguities and licensing NPIs. (i) indicates that the negation in such examples may take scope over a quantificational adjunct. The example is ambiguous with the two readings in (ii). (ii)a in particular corresponds to the interpretation in which negation has wide scope and must be sentential, (iii) indicates that this negation may also license an NPI in a VP adjunct.

(i) My parents have both not worked for many years.(ii) a. Both my parents have worked for not many years. NOT > MANY

b. For many years, both my parents haven’t worked. MANY > NOT

(iii) The twins could both not vote during any of the elections.

An additional test not previously discussed is available since these examples may contain a modal. It is well-known that sentential negation, but not constituent negation, may take scope over some modals in I" (Zwicky and Pullum 1983, Ernst 1992). That the example in (iv) is ambiguous, with both of the readings in (v) and, in particular, the wide scope reading for negation in (v)a, indicates that we are potentially dealing with sentential negation.


423


(59) a. The children might all not stand around if they knew they’d be put to work.

b . The twins could both not vote.

c. My parents are both not working.

d . The neighbors have all not put out their recycling bins.

(60) IP

the twins I

could QPjA

tk Q’ NegA

Q tiI

both

k n o t li V’I

VI

vote

With the assumption that all three elements {do, modals, and raised auxiliaries) occupy the

same structural position, the difference between (57)a/(58)a with a form of do and (59) is

mysterious. The former examples should have the structure in (60) with did replacing the

modal. The contrast indicates that there is something unique about the sequence consisting

of a form of do followed by sentential not which does not permit an intervening floated

quantifier, even if a specifier position for the quantifier is available.

(61)/(62) and (63)/(64) show that the same pattern occurs with S-adverbs and E-adverbs,

respectively. Neither may occur between do and not, the a examples, despite being able to

(iv) The twins can both not vote.(v) a. Both the twins are not able to vote. NOT > CAN

b. Both the twins are able to not vote. CAN > NOT


424

appear above do, the b examples. The problem is not obviously semantic since the b

examples maintain the same scope of the adverb with respect to negation.

S-AD VERBS

(61) a. *The butler did apparently not have an alibi,

b . The butler apparently did not have an alibi.

(62) a. *He does obviously not understand the procedure,

b . He obviously does not understand the procedure.

E-ADVERBS

(63) a. * He did just not want to go.

b . He just did not want to go.

(64) a. *They did merely not try hard enough,

b . They merely did not try hard enough.

The situation is different when 1° is filled with a modal or raised auxiliary. In such cases,

S- and E-adverbs may grammatically occur between 1° and Neg°, (65) and (67).9 The

9We can again ascertain that the instances of not in such examples are potentially sentential negation, (i) demonstrates that similar instances of negation introduce an ambiguity with quantified adjuncts, made explicit in (ii). (ii)a is the desired reading involving wide scope of sentential negation.

(i) The union members will probably not accept the contract for many years.(ii) a. The union members will probably accept the contract for not many years. NOT > MANY

b. For many years, the union members will probably not accept the contract. MANY > NOT

VP ellipsis is clearly possible after these examples, again indicating that we are dealing with sentential negation (see chapter 2, section 5.2.2):

(ii) a. The President might be above the law but policemen are certainly not 0 .b. The Wessons are often home at night but the Smiths are usually not 0.c. Joe is pretty unpredictable. He may show up or he may just not 0 .


425

structure for (65)a is as in (66). Within the theory of adverb syntax from chapter 1, the

S-adverb is licensed by 1°, permitting it to appear adjoined to NegP, a position governed by

r .

S-AD VERBS

(65) a. He would obviously not understand the procedure.

b . The butler must apparently not have an alibi.

c. The union members will probably not accept the management’s offer.

d. Jeff has apparently not finished his homework. That’s why he’s not here.

e. Policemen are certainly not above the law.

(66) IP

D P ^ ^ :•

the butler I NegP

must AdvP NegP

apparently Neg VP

not have an alibi

E-ADVERBS

(67) a. He must just not want to go.

b . They are merely not trying hard enough.

c . You can simply not call her back.

The ready availability of an analysis for the grammatical (65) and (67) seems to again

indicate that there is something unique to the combination of do and not that results in the

ungrammaticality of the minimally different a examples of (61) through (64). In light of

these data, I offer the observation in (68).


426

(68) do not Descriptive Generalization

A form of support do and sentential negation not may not be linearly

separated except in inversion contexts

From the preceding demonstration, the source of this observation cannot be syntactic given

the following assumptions about the examples, which have the consequence of assigning

them all identical syntactic structures: 1) do, modals, and finite auxiliaries all occupy 1°, 2)

sentential negation not is always the head of NegP, and 3) there are no other projections

between IP and NegP.101 tentatively attribute (68) to a low-level head movement operation

which applies at PF and which obligatorily adjoins not to do. This rule must apply in the

10The data can be accommodated if we give up assumptions I) and 3) and adopt a structure as in (i) in which modals occupy a distinct, higher head position from do.

subject

{modals. finite auxes)

The data could then be accounted for by claiming that spec,NegP is not available as a stranding site for floated quantifiers and that I2 does not license S- or E-adverbs or, alternatively, that the Neg projection is inert for adjuncdon. Despite the tenability of such an analysis, neither the structure itself nor the individual assumptions would seem to have any support beyond these particular data. I leave it as an open alternative to be independently motivated.


427

phonological component after all syntactic operations, such as T-to-C0.11 As a

consequence of (69), nothing will intervene between do in 1° and Neg° not because of the

very local character of head adjunction.

(69) PF do not Rule (obligatory)

Adjoin Neg° not to do in 1°

3.2.2. The Position of do in Formal Imperatives

Armed with the claim in (68), we can use it to determine the structural position of do in

formal imperatives. Since the generalization is sensitive to whether or not inversion has

applied, it will indicate whether formal imperatives involve movement of do from 1° to C°

or not. The crucial data follow.

Not surprisingly, floated quantifiers cannot appear between do and not in formal

imperatives, (70). While the data might somehow be attributable to the general intolerance

for overt subjects in formal imperatives, it is worth considering the judgments in contrast to

those with the quantifier to the right of not, (71). The former are again absolutely

impossible relative to the latter. The strong ungrammaticality of (70) is expected if,

simultaneously, inversion has not applied and do and sentential not form an inseparable

unit.

11 If the source of the tight morphological connection between a form o f support do and sentential not is indeed phonological, then it might ultimately be unnecessary to assume that not actually moves to I*. The rule might better be formulated as a restriction on prosodic structure deriving from the obligatorily unstressed nature of dummy do. I leave this for further investigation. It would be preferable if the head movement above did not actually have to be posited since there is no actual evidence for it. It is one mechanism, perhaps not the correct one, to capture the observed locality requirements.


428


(70) a. *Pleasedoallnottalkatonce!

b . *1 implore you two: do both not listen to her!

(71) a. Please do not all talk at once!

b . ?I implore you two: do not both desert me!

Likewise, S- and E-adverbs may not intervene between do and not in formal imperatives,

(72) and (73).12

S-AD VERBS

(72) a. *Do definitely not tell them where you’re from!

b . *Do certainly not make claims that are unfounded!

c. *Do absolutely not mention this!

E-ADVERBS

(73) a. *Do just not think about it!

b . *Do simply not give them your address!

c. *Do merely not return the forms!

As above, we can compare these examples with corresponding ones where the adverb does

not intervene. The data in (74) and (75) are parallel to (72) and (73) except that the adverb

12These examples are potentially good on a reading where do is emphatic and not is constituent negation. The case of interest, however, is one in which do is not emphatic, but neutral.


429

is to the left of doAnot. This position maintains the scope of the adverb with respect to

negation and synonymy. There is dramatic improvement.13

(74) a. Definitely do not tell them where you’re from!

b . Certainly do not make claims that are unfounded!

c. Absolutely do not mention this!

(75) a. Just do not think about it!

b . Simply do not give them your address!

c. Merely do not return the forms and they’ll leave you alone!

13The data are once again degraded if the adverb appears below not:

(i) a. ??Do not definitely tell them where you’re from! S-ADVERBSb. *Do not certainly make claims that are unfounded!c. *Do not absolutely mention this!

(ii) a. ??Do not just think about it! E-ADVERBSb. ??Do not simply give them your address!c. ??Do not merely return the forms and they’ll leave you alone!

The account of at least (i) could be syntactic: Neg° does not license S-adverbs. This seems incorrect, however, given the declarative clauses in (iii) in which an S-adverb appears grammatically below not. These examples are analyzed as having the S-adverb licensed by Neg" in the VP-adjoined position, which is governed by Neg*, (iii)d.

(iii) a. He did not apparently have an alibi.b. Jeff has not definitely finished his homework.c. Policemen are not obviously above the law.d. The students [j» will [^egP [Neg’ not tvP obviously [yp see the problem here ] ] ] ] ]

I would suggest that the unacceptability of the data in (i) and (ii) is semantically based. The examples crucially differ from the ones in the text in that the adverb is within the scope of negation. It is not expected that the two sets o f data should be synonymous or equally felicitous, (iv) and (v) indicate that grammatical formal imperative examples with the adverb below not can be constructed.

(iv) a. Do not normally wait so long before responding!b. Do not definitely refuse until you’re sure of your plans!

(v) a. Do not just think about it, do it!b. Do not simply give them your address if you want them to call you!c. Do not just stand there!


430

This pattern is also accounted for if doAnot in formal imperatives is governed by the

observation in (68). The conclusion is that there has been no movement of do to C° in

formal imperatives. The data in (75) in fact support this claim in conjunction with the

theory of adverb licensing in chapter 1. The analysis of E-adverbs there indicated that they

are licensed by 1° but not C°. Consequently, they may adjoin to I’ but not higher. This

permits only the structure in (76)a in which do is in 1° and not the structure in (76)b in

which it is in C°.

b. * CPI

C’

I’ (E-adverb) C

[P

(76) a.

E-adverb

do Neg do, pro r

NegP

Neg

not

VP

I state the result in (77). It rules out a structure as in (78).

(77) there is no movement of do to C° in formal imperatives

(78) * CPI

C’

dOj subject V

tj NegP

Neg VP

not


4 3 1

(77) is a positive finding in that it sheds light on what exactly the do nof-problem is. A

subject is not permitted between do and not in formal imperatives because there is no

inversion to derive that word order. The do nor-problem reduces to explaining why do may

not invert by itself in imperatives. In the following section I offer a characterization of the

patterns of I0-to-C° in English imperatives and offer a formal account using the mechanisms

of checking theory.

4. Imperative I°-to-C°

This section lays out the patterns of I°-to-C°, or IN V ERSIO N , in imperatives and

interrogatives and offers a formal analysis of them. Section 4.1 explores the full paradigm

of I°-to-C° in imperatives, bringing together the results of chapter 5 and section 3 of this

chapter. It offers a characterization of inversion in imperatives: it is optional and only

occurs when there is a semantically-oriented motivation, the presence of negation or

affirmation. Section 4.2 proposes a checking-theoretic account of the inversion pattern in

relation to a formal analysis of interrogative inversion. The difference between the two

processes is claimed to be that, although the head movement itself is the same in both

cases, it is triggered by different categorial features. This accounts for the two different

patterns.

4 .1 . A Characterization of Imperative Inversion

(79) summarizes the options for T-to-C" in imperatives, what I will call IMPERATIVE

INVERSION. It also identifies the location in this work where the individual claims are

made.


432

(79) Possibility ofI°-to-C° in Imperativesformative P-to-C*

a. don’t yes (chapter 5)b. emphatic do yes (chapter 5)c. do not yes (chapter 6, section 3.1)d. unemphatic do no (chapter 6, section 3.2.2)

Chapter 5 argued that don't and emphatic do may optionally move from 1° to C° in

imperatives, (79)a and b. This accounted for the alternations in (80) through (83), repeated

from chapter 5.

(80) a. Do SOMEone help him quickly! DOASUBJECT


(81) a. SOMEone do answer the phone! SUBJECTADO


(83) a. Don’t anyone touch my stuff! DON’rASUBJECT


(83) a. Those with luggage don’t leave it unattended! SUBJECTADCW’7'


Section 3.1 of this chapter proposed that do not in formal imperatives could also optionally

appear as a unit in C°, (79)c. Such movement permitted subjects to occur, in principle,

preceding or following do not, (44) and (46).

(44) a. I know I’ve done wrong but I can’t survive on my own. SUBJECTAD0AM97'





433

(46) a. ?Do NOT ANY of you touch that cake! D0A/VO7'ASUBJECT

It’s for the wedding and if any of it is missing heads will roll.

b . I know I’ve done wrong but I can’t survive on my own.

?Oh please, do not ALL of you desert me!

c. ?DO not YOU, of all people, insult me in this heinous and base manner!

d. ??Do not EVERYone stand off by himself in the comer, come join the party!

Lastly, it was demonstrated in 3.2.2 that unemphatic do alone does not move to C° but

remains in 1°, (79)d, thereby addressing the ungrammaticality of (45) in which a subject

intervenes between do and sentential not.

(45) a. *Do somebody not desert me! *DOASUBJECTAyvor


One pattern that emerges from (79) is that I°-to-C° only occurs in imperatives when

negation or emphatic affirmation is present—that is, when there is semantic motivation.

Inversion does not occur with support do by itself because it is semantically empty but it

does occur with do not as a unit because in this case negation is attached to the semantically

empty do. A characterization of I°-to-C° in imperatives that I will put forth is that it is a

syntactic operation that has a semantic driving force:

(84) imperative inversion is semantically driven

In support of the characterization is the fact that inversion does not occur when negation or

emphasis is absent in the representation. In a neutral imperative such as Give me a hand!

there is no movement of the main verb to C° (as argued in chapter 2, section 4), ruling out

the representation in (85), and neither is a dummy element, do, inserted to provide an

inversion target, (86). (86) in particular indicates that imperatives in and of themselves do


434

not trigger Do-Support, unlike interrogatives in (87) which show the opposite pattern of

grammaticality.

(85) *[cp [c - Give [[p me a hand! ] ] ]

(86) a. *Do give me a hand! (ungrammatical as a non-emphatic imperative)

b . Give me a hand!

(87) a. Do you think much about it?

b. *You think much about it? (ungrammatical as an interrogative)

Section 4.2 presents an analysis of imperative inversion which gives theoretical substance

to the characterization in (84) and at the same time highlights the similarities and differences

with finite interrrogative clauses.

4 .2 . An Analysis

In analyzing the pattern of imperative inversion in (79), I take as the starting point the

account of interrogative I°-to-C° in the current Minimalist program. Within checking theory

(see chapter 1, section 2), interrogative T-to-C0 is a consequence of a strong Q feature

resident in C° (Chomsky 1995:289-294). In order to eliminate the strong feature by Spell-

Out, it must be checked by some other feature Fq in the checking domain of C°. This

feature is taken to be the V-feature of 1°; hence, SAI is analyzed as checking of the Q

feature in C° by head adjunction of 1° to C°. This is in keeping with the restrictive

assumption that if a feature, in this case Q, is strong then it is a feature of a functional

category and it is checked by a categorial feature (Chomsky 1995:232). SAI in questions is

thus the reflex of a purely formal syntactic requirement: the strong Q feature must be


435

checked. Movement is thus always required since the feature must always be eliminated,

even at the expense of using a semantically empty element, do.

If the pattern for imperatives in (79) is correct, they clearly differ from interrogatives in

crucial respects and it would be incorrect to precisely mimic the analysis of interrogatives

for imperatives. First, the optionality of imperative inversion indicates that the feature in

imperative C° that is to be checked, call it IMP, is not always present in the derivation.

Second, the semantic nature of the inversion trigger in imperatives indicates that the

categorial feature that checks IMP, call it F^p , cannot be simply a V-feature of 1°. It must

be a feature associated exclusively with negation and affirmation. Given the generalization

that only semantically contentful inflectional heads undergo imperative inversion, we can

hypothesize that this category which must check IMP is Laka’s (1990) 1°, the syntactic

realization of negation and affirmation. This accounts for the fact that it is precisely these

two which yield imperative inversion because these are the two instantiations of 1°. This

also accounts for the observation that inversion does not take place when IP is not present,

as illustrated by (86) above. Anaytically, imperative inversion is checking of a strong IMP

feature in C° by the categorial feature E \

Because both a strong IMP feature in C° and a X projection are optional elements in a

structure, there are four cases to consider according to whether or not each one is present in

the derivation. These cases are defined in the table below.


436

(88) C° [IMP] I-FEATURE

yes yes Case I: inverted imperatives

no Case 2: crash

no yes Case 3: subject-first imperatives

no Case 4: neutral imperatives

The presence or absence of IMP in the numeration, the first column, correlates with

inversion or its absence. The E-feature, the second column, indicates whether the

imperative is a neutral imperative or not. If E° is present then the imperative will be

negative or emphatic. The four cases cover the full range of grammatical imperative data

surveyed as follows. Case 1 yields inverted negative and emphatic imperatives exemplified

by the schema in (89) and the data in (80), (83), and (46). Both C° [IMP] and Z° are

present, requiring I°-to-C0.

(89) Case I

a. [q) do(n’t)j [j-p subject [j. tj [^p [yp VP ] ] ] ] ] (= (80), (83))

b. [qj [do not]j [jp subject [j. tj [j-p [yp VP ] ] ] ] ] (= (46)

Case 2, in which C° [IMP] is present but there is no E° projection, does not yield a

convergent structure because there will be an unchecked strong feature at PF. Cases 3 and

4, in which C° [IMP] is absent, will have no overt inversion because the inversion trigger

is not present. Case 3, schematized in (90), is the non-inverted counterparts of Case 1 and

corresponds to non-neutral subject-first imperatives, the data in (81), (83), and (44).

(90) Case 3

a. [[p subject [j< do(n t) [j-p [yp VP ] ] ] ] (= (81), (83))

b . [[P subject [r do not [sp [vp VP ] ] ] ] (= (44))


437

Lastly, Case 4, in which neither C° [IMP] nor Z° is in the numeration, corresponds to

neutral imperatives schematized in (91) and exemplified by the data in (86)b, where the

inflectional layer contains only null imperative inflection.

The do nor-problem, with the putative structure in (92), does not have a derivation in the

analysis because the movement of do does not satisfy any feature checking requirements; it

is not Z°. Consequently, the principle of Last Resort (Chomsky 1995:280) rules out the

movement.

It is worth considering this analysis of imperative inversion in comparison to the analysis

of interrogative inversion. In theoretical terms, the two are alike in having strong features in

C° and in having them checked by adjunction of an inflectional head. Both movements,

which are ultimately I°-to-C\ are realized as inverted word order. A fundamental difference

between interrogatives and imperatives is the identity of the categorial feature that checks

the strong feature of C°. In the interrogative case, it is a V-feature in 1°. The feature is

purely syntactic in nature and can be present in a structure with no semantic contribution.

Consequently, interrogative inversion is seen as an exemplary syntactically-driven process.

(91) Case 4

[jp subject [r -0 [yp VP ] ] ] (= (86)b)

(92) CP

not


438

In imperatives, by contrast, the checking feature is Z° and it necessarily has semantic

content. It thus makes imperative inversion appear semantically driven.

We can also make sense of the optionality of imperative inversion in contrast to the

apparent obligatory nature of interrogative inversion. To do so, however, interrogative

inversion needs to be considered in the larger picture of finite clauses. In the domain of

finite clauses, interrogatives are clearly just one subtype. While inversion is obligatory in

interrogatives, it is not required in all finite clauses more generally. In particular, declarative

sentences do not exhibit I°-to-C°. Viewed in this light, inversion in finite clauses is optional

in the same way that it is in imperative clauses. In theoretical terms, the Q feature of C° is

optional when finite clauses as a whole are considered. This can be summarized by the

table in (93) for finite clauses which parallels the one in (88) for imperative clauses.

(93) C° [Q] V-FEATURE

yes yes Case 1: interrogatives

no Case 2: crash

no yes Case 3: declaratives with filled 1°

no Case 4: declaratives with empty 1°

For all finite clauses, there are again four cases to consider depending upon whether C° [Q]

and/or a V-feature in 1° are present in the derivation. As before, the presence of the strong

feature in C° corresponds to inversion. The availability of a V-feature in 1° has no necessary

semantic consequences but merely corresponds to a syntactically filled 1°. The four cases

are the following. When C° [Q] is present, Case 1, an interrogative results, with Do-

Support if necessary to provide the appropriate checking feature. If this feature is absent,

the derivation will crash because the strong Q feature will remain unchecked, Case 2. If, on


439

the other hand, the Q feature in C° is not present, Cases 3 and 4, we obtain finite

declarative clauses, either with or without a filled 1°.

Viewed in this light, imperatives and finite clauses once again appear remarkably similar

syntactically. They have the same phrase structures and appeal to the same syntactic

operations, in particular I°-to-C0 movement and feature checking. They differ at the lexical

level, in the content of the heads 1° and C°. I have claimed that both imperative and finite

clause 1° contain dependent inflectional morphology. C° in each case may also contain a

strong feature which drives movement. Differences have been reduced to the nature of this

strong feature and what its checking needs are—a lexical specification. To the extent that

there is parallelism in the above analyses, there is theoretical support for the overarching

claim that imperative syntax is not fundamentally anomalous, but highly regular and largely

unexceptional.

5. Conclusions

In this chapter I have used empirical observations and the success of the previous chapters

to motivate assimilating the syntax of formal imperatives to that of finite declarative and

interrogative clauses. An analysis within Minimalist Program assumptions was then

proposed for imperative word order which highlighted the similarities and differences

between the syntax of imperatives and finite clauses. The analysis raises several important

issues and questions which I present in this closing section.

The analysis sheds light on what elements contribute to the notion of syntactic clause type.

From the analysis and the two tables in (88) and (93), two major syntactic clause types are

identified: finite clauses and imperatives. They are individually characterized and

distinguished by the content of 1°. All finite clauses share the same paradigm of inflectional


440

morphology which is distinct from the inflectional morphology in imperatives. Subtypes of

these clauses are indentified by the content of C° and, under this view, interrogatives are a

subtype of finite clause. They obligatorily contains finite morphology and, additionally, C°

contains a Q feature. The analysis suggests that a typology of clauses is built around the

contents of Infl and secondarily the contents of C°.

From the analysis and facts presented here it is easy to see why a particle analysis of formal

imperatives has always been seriously entertained (Cohen 1976, Schmerling 1977, Zhang

1990). Section 2.1 was not actually able to eliminate the imperative particle analysis on

empirical grounds. The decision not to pursue that line was based largely on theoretical

biases. The primary reason for this failure is that two of the major claims of this section—

that do and not form a tight morphological unit and that there is no movement of do to

C° in formal imperatives—conspire to make it always the case that do and not are adjacent

in the linear string of words. Both the imperative particle analysis and the one adopted are

equivalent in their empirical predictions at the level of complexity to which the language

learner is exposed and both would seem to be permitted by principles of Universal

Grammar. This leaves a certain amount of indeterminacy in this particular comer of the

grammar. It is unclear what language learners do in such a situation and it is not evident

that they would all arrive at the same internal analysis. Hankamer 1977, in discussing just

this topic in relation to imperatives, suggested that the native speaker may retain and appeal

to both analyses. In light of this, it is perhaps not surprising that the data that Schmerling

used in order to support the imperative particle approach and other data that might

distinguish the two analysis are so uncomfortable for consultants.

It should be conspicuous that nothing has been said about the precise analysis of Do-

Support in this investigation. The reason should now be clear. In assimilating imperatives


441

to finite clauses, whatever analysis one ultimately adopts in the latter case will automatically

extend to the former. The two have been given identical syntax in the relevant respects:

dependent morphology in 1°, a general inability of verbs to raise to 1°, and a NegP

dominating VP. The analysis of finite clause periphrastic do has been extensively discussed

in generative linguistic works (e.g. Chomsky 1957, 1975, 1991; Klima 1964; Emonds

1970; Pollock 1989; Laka 1990; Wilder and Cavar 1994; Bobaljik 1995; Grimshaw 1995)

and is still very much a subject of inquiry and interest but I am suggesting that imperative

syntax has little to contribute to its formulation. Imperatives are not inherently an

environment where Do-Support applies, unlike interrogatives, and it only occurs in

imperatives because of the independent presence of negation and affirmation, which are

environments for Do-Support.

At least three outstanding issues remain and are highlighted by the formal analysis. First,

the analysis assumed a certain idealization of the data and thus there remains the fact that

formal imperatives with subjects preceding or following doAnot are highly degraded in

many cases despite the acceptable examples offered. It is certainly the case that they are rare

and this is the first work to my knowledge to claim that they are acceptable. It was

concluded that prosodic factors could not be ruled out as a source of the degradation,

following Akmajian 1984. Stylistic influences were also suggested. I leave these and other

possibilities for future investigation. Second, the analysis raises the question of what, if

anything distinguishes imperatives with and without inversion. It has been suggested that

pairs such as (94) are not simply be free variants. The analysis claims that they have at least

a syntactic difference: the presence or absence of the feature IMP in C°. One expects that

such a difference would correspond to a semantic distinction as well. While the two cases

do seem to have different uses and subtle differences in subject preferences, a full

exploration of the data remains to be done.


442

(94) a. Don’t everybody leave!

b . Girls can go, but boys don’t leave!

Lastly, the claim that imperatives and finite sentences are distinct clause types distinguished

by the content of 1° raises the question of what exactly characterizes these inflectional

elements in each case. What is the content of the respective heads, not syntactically but

semantically, that characterizes the two clause types? I leave these interesting larger issues

for future work.


443

References

Aarts, Flor. 1989. “Imperative Sentences in English: Semantics and Pragmatics”. Studia Linguistica 43, 119-134.

Abney, Steven. 1987. “The English Noun Phrase in its Sentential Aspect”. MIT Ph.D dissertation.

Aissen, Judith L. 1992. ‘Topic and Focus in Mayan”. Language 68,43-80.

Aissen, Judith L. 1995. “Tzotzil Auxiliaries”. Linguistics 32, 657-690.

Akmajian, Adrian. 1984. “Sentence Types and the Form-Function Fit”. Natural Language and Linguistic Theory 2, 1-23.

Akmajian, Adrian, Susan Steele, and Thomas Wasow. 1979. “The Category AUX in Universal Grammar”. Linguistic Inquiry 10, 1-64.

Akmajian, Adrian and Thomas Wasow. 1975. “The Constituent Structure of VP and AUX and the Position of the Verb Be”. Linguistic Analysis 1, 205-245.

Anderson, Stephen. 1972. “How to get Even”. Language 48, 893-906.

Ard, Josh. 1982. “Auxiliary Do: Support or Emphasis?” Linguistics 20,445-466.

Austin, J. L. 1962. How To Do Things with Words. Cambridge, Ma.: Harvard University Press.

Authier, J.-Marc. 1992. “A Parametric Account of V-Govemed Arbitrary Null Arguments”. Natural Language and Linguistic Theory 10, 345-374.

Baker, C. L. 1971. “Stress Level and Auxiliary Behavior in English”. Linguistic Inquiry 2, 167-181.

Baker, C. L. 1991. “The Syntax of English Not: The Limits of Core Grammar”. Linguistic Inquiry 22, 387-429.

Baker, Mark. 1988. Incorporation: A Theory of Grammatical Function Changing. Chicago: University of Chicago Press.

Baker, Mark, Kyle Johnson, and Ian Roberts. 1989. “Passive Arguments Raised”. Linguistic Inquiry 20, 219-251.


444

Baltin, Mark. 1982. “A Landing Site Theory of Movement Rules”. Linguistic Inquiry 13, 1-38.

Baltin, Mark. 1995. “Floating Quantifiers, PRO, and Predication”. Linguistic Inquiry 26, 199-248.

Bellert, Irena. 1977. “On Semantic and Distributional Properties of Sentential Adverbs”. Linguistic Inquiry 8, 337-351.

Belletti, Adriana. 1990. Generalized Verb Movement. Turin: Rosenberg and Sellier.

Belletti, Adriana. 1994. “Verb Positions: Evidence from Italian”, in Lightfoot, David and Norbert Homstein (eds.). Verb Movement. Cambridge: Cambridge University Press, 19-40.

Benveniste, Emile. 1966. Problemes de Linguistique Generate. Paris: Gallimard.

Beukema, Frits and Peter Coopmans. 1989. “A Government-Binding Perspective on the Imperative in English”. Journal of Linguistics 25,417-436.

Bobaljik, Jonathan David. 1995. “Morphosyntax: The Syntax of Verbal Inflection”. MIT Ph.D dissertation.

Bobaljik, Jonathan David and Dianne Jonas. 1996. “Subject Positions and the Roles of TP”. Linguistic Inquiry 27, 195-236.

Bolinger, Dwight. 1967. “The Imperative in English”, in To Honor Roman Jakobson, Vol. 1. The Hague: Mouton.

Bolinger, Dwight. 1977. Meaning and Form. London: Longman Press.

Borer, Hagit. 1980. “Empty Subjects in Modem Hebrew and Constraints on Thematic Relations”, in Proceedings of the Tenth Annual Meeting of the North Eastern Linguistics Society. Ottawa, Ontario.

Borer, Hagit. 1983. Parametric Sytnax: Case Studies in Semitic and Romance Languages. Dordrecht: Foris Publications.

Borer, Hagit. 1995. “The Ups and Downs of Hebrew Verb Movement”. Natural Language and Linguistic Theory 13, 527-606.

Borkin, Ann. 1971. “Polarity Items in Questions”, in Proceedings o f the Seventh Regional Meeting o f the Chicago Linguistic Society, University of Chicago, Chicago, Illinois.

Bouchard, Denis. 1983. On the Content of Empty Categories. Dordrecht: Foris.

Bowers, John. 1993. “The Syntax of Predication”. Linguistic Inquiry 24, 591-656.

Bresnan, Joan. 1976. “The Form and Functioning of Transformations”. Linguistic Inquiry 7, 3-40.


445

Broadie, Alexander. 1972. “Imperatives”. Mind 81, 179-190.

Browning, Marguerite. 1987. “Null Operator Constructions”. MIT Ph.D dissertation.

Burzio, Luigi. 1986. Italian Syntax. Reidel.

Carlson, Greg N. 1977. “Reference to Kinds in English”. University of Massachusetts at Amherst Ph.D dissertation.

Chao, Wynn. 1988. On Ellipsis. New York: Garland Publishing, Inc.

Chen, Matthew Y. 1987. “The Syntax of Xiamen Tone Sandhi”, in Phonology Yearbook4.

Chomsky, Noam. 1957. Syntactic Structures. The Hague: Mouton.

Chomsky, Noam. 1975[1955]. The Logical Structure o f Linguistic Theory. New York: Plenum.

Chomsky, Noam. 1981. Lectures on Government and Binding. Dordrecht: Foris.

Chomsky, Noam. 1982. Some Concepts and Consequences o f the Theory o f Government and Binding. Cambridge, Ma.: MIT Press.

Chomsky, Noam. 1986. Barriers. Cambridge, Ma.: MIT Press.

Chomsky, Noam. 1991. “Some Notes on Economy of Derivation and Representation”, in Freidin, Robert (ed.). Principles and Parameters in Comparative Grammar. Cambridge, Ma.: MIT Press.

Chomsky, Noam. 1993. “A Minimalist Program for Linguistic Theory”, in Hale, Kenneth and Samuel J. Keyser (eds.). The View from Building 20. Cambridge, Ma.: MIT Press, 1-52.

Chomsky, Noam. 1994. “Bare Phrase Structure”. MIT Occasional Papers in Linguistics 5.

Chomsky, Noam. 1995. The Minimalist Program. Cambridge, Ma.: MIT Press.

Chung, Sandra. 1984. “Identifiability and Null Objects in Chamorro”, in The Proceedings of the Tenth Annual Meeting of the Berkeley Linguistics Society, 116-130.

Cinque, Guglielmo. 1990. Types of A-Dependencies. Cambridge, Ma.: MIT Press.

Clark, Billy. 1993. “Let and Let’s: Procedural Encoding and Explicature”. Lingua 90, 173-200.

Clark, Herbert H. and Thomas B. Carlson. 1982. “Hearers and Speech Acts”. Language 58, 332-373.


446

Cohen, Ava-Robin. 1976. “Don’t You Dare!”, in Hankamer, Jorge and Judith Aissen (eds.). Harvard Studies in Syntax and Semantics, Volume 2. Cambridge, Ma.: Harvard University Press.

Culicover, Peter W. 1976. Syntax. New York: Academic Press.

Culicover, Peter W. 1991. “Topicalization, Inversion, and Complementizers in English”, in Delfitto, Denis, Martin Everaert, Arnold Evers, and Frits Stuurman (eds.). Going Romance and Beyond. OTS Working Papers. University of Utrecht.

Culicover, Peter W. 1996. “On Distinguishing A-Movements”. Linguistic Inquiry 27, 445-463.

Dalrymple, Mary. 1991. “Against Reconstruction in Ellipsis”, ms. Stanford University.

Davies, Eirlys E. 1986a. The English Imperative. London: Croom Helm.

Davies, Eirlys E. 1986b. “English Vocatives: A Look at their Function and Form”. Studia Anglica Posnaniensia 19, 91-106.

Denison, David. 1993. English Historical Syntax: Verbal Constructions. London: Longman.

den Dikken, Marcel. 1992. “Empty Operator Movement in Dutch Imperatives”, in Gilters, D. and S. Looyenga (eds.). Language and Cognition 2. Groningen.

Doetjes, Jenny. 1992. “Rightward Floating Quantifiers Float to the Left”. The Linguistic Review 9, 313-332.

Doron, Edit. n.d. “On the Complementarity of Subject and Subject-Verb Agreement”, ms. Stanford University and the Hebrew University of Jerusalem.

Doron, Edit. 1990. “VP Ellipsis in Hebrew”, ms. The Hebrew University, Jerusalem.

Downes, William. 1977. “The Imperative and Pragmatics”. Journal of Linguistics 13, 77- 97.

Downing, Bruce T. 1969. “Vocatives and Third-Person Imperatives in English”. Papers in Linguistics 1, 570-592.

Dowty, David and Belinda Brodie. 1984. “A Semantic Analysis of ‘Floated’ Quantifiers in a Transformationless Grammar”, in Cobler, M., S. MacKaye, and Michael Westcoat (eds.). Proceedings of the Third West Coast Conference on Formal Linguistics. Stanford: Stanford Linguistics Association, 75-90.

Elbert, Samuel H. 1970. Spoken Hawaiian. Honolulu: University of Hawaii Press.

Ellegard, Alvar. 1953. The Auxiliary Do: The Establishment and Regulation o f its Use in English. Stockholm: Almqvist and Wiksell.


447

Emonds, Joseph E. 1970. Root and Structure-Preserving Transformations. MIT Ph.D dissertation.

Emonds, Joseph E. 1976. A Transformational Approach to English Syntax. New York: Academic Press.

Emonds, Joseph E. 1978. “The Verbal Complex V’-V in French”. Linguistic Inquiry 9, 151-175.

Emonds, Joseph E. 1985. A Unified Approach to Syntactic Categories. Dordrecht: Foris Publications.

Emonds, Joseph E. 1987. ‘The Invisible Category Principle”. Linguistic Inquiry 18, 613- 632.

Ernst, Thomas. 1984. Towards an Integrated Theory o f Adverb Position in English. Bloomington, Indiana: Indiana University Linguistics Club.

Ernst, Thomas. 1992. ‘The Phrase Structure of English Negation”. Linguistic Review 9, 109-144.

Faarlund, Jan Terje. 1985. “Imperative and Control: First Person Imperatives in Norwegian”. Nordic Journal o f Linguistics 8, 149-160.

Falk, Cecilia. 1993. “Non-Referential Subjects and Agreement in the History of Swedish”. Lingua 89, 143-180.

Farkas, Donka F. 1987. “DO pro in Hungarian”, in Kenesei, Istvan (ed.). Approaches to Hungarian, Volume 2: Theories and Analyses. JATE: Szeged, 191-211.

Farkas, Donka. 1988. “On Obligatory Control”. Linguistics and Philosophy 11,27-58.

Farkas, Donka and Draga Zee. 1993. “Agreement and Pronominal Reference”. UCSC Linguistics Research Center Report No. 93-01.

Farrell, Patrick. 1990. “Null Objects in Brazilian Portuguese”. Natural Language and Linguistic Theory 8, 325-346.

Fiengo, Robert and Robert May. 1994. Indices and Identity. Cambridge, Ma.: MIT Press.

Fries, C. C. 1952. The Structure o f English: An Introduction to the Construction o f English Sentences. New York: Harcourt Brace.

Gazdar, Gerald, Ewan Klein, Geoffrey Pullum, and Ivan Sag. 1985. Generalized Phrase Structure Grammar. Cambridge, Ma.: Harvard University Press.

Gazdar, Gerald, Geoffrey K. Pullum, and Ivan A. Sag. 1982. “Auxiliaries and Related Phenomena in a Restrictive Theory of Grammar”. Language 58. 591-638.

Goffman, Erving. 1975. “Replies and Responses”. Language in Society 5, 257-313.


448

Goldberg, Lotus. 1995. “VP Ellipsis and Clause Structure in Modem Hebrew”, ms. University of California, Santa Cruz.

Goodwin, C. 1981. Conversational Organization: Interaction Between Speakers and Hearers. New York: Academic Press.

Grice, H. P. 1975. “Logic and Conversation”, in Cole, Peter and Jerry Morgan (eds.). Syntax and Semantics 3: Speech Acts. New York: Academic Press.

Grimshaw, Jane. 1991 “Extended Projection”, ms. Brandeis University.

Grimshaw, Jane. 1995. “Projections, Heads, and Optimality”, ms. Rutgers University.

Gundel, Jeanette K. 1988. The Role of Topic and Comment in Linguistic Theory. New York: Garland Publishing, Inc.

Haegeman, Liliane. 1995. The Syntax o f Negation. Cambridge: Cambridge University Press.

Hale, Kenneth and Elisabeth Selkirk. 1987. “Government and Tonal Phrasing in Papago”, in Phonology Yearbook 4, 151-184.

Hamblin, C. L. 1987. Imperatives. Oxford: Basil Blackwell Ltd.

Han, Chung-hye and Laura Siegel, to appear. “Syntactic and Semantic Conditions on NPI- Licensing in Questions”, in Proceedings of the Fifteenth West Coast Conference on Formal Linguistics. Stanford, Ca.: Stanford Linguistics Association.

Hankamer, Jorge. 1977. “Multiple Analyses”, in Li, Charles (ed.). Mechanisms of Syntactic Change. Austin: University of Texas Press.

Hankamer, Jorge and Ivan Sag. 1976. “Deep and Surface Anaphora”. Linguistic Inquiry 7, 391-426.

Haverkate, W. H. 1978. ‘The Vocative Phrase in Modem Spanish. A Contribution to the Study of Elocutionary Functions”. Linguistics in the Netherlands 1974-1976. Lisse: University of Amsterdam, 46-62.

Heim, Irene. 1988[1982]. The Semantics of Definite and Indefinite Noun Phrases. New York: Garland Publishing, Inc.

Heine, Bemd. 1993. Auxiliaries: Cognitive Forces and Grammaticalization. Oxford: Oxford University Press.

Henry, Alison. 1995. Belfast English and Standard English: Dialect Variation and Parameter Setting. Oxford: Oxford University Press.

Hoekstra, Teun. 1994. “HAVE as BE plus or minus”, in Cinque, Guglielmo, Jan Koster, Jean-Yves Pollock, Luigi Rizzi, and Raffaella Zanuttini (eds.). Paths Towards Universal Grammar: Studies in Honor of Richard S. Kayne. Washington D.C.: Georgetown University Press, 199-215.


449

Holmberg, Anders and Christer Platzack. 1995. The Role o f Inflection in Scandinavian Syntax. Oxford: Oxford University Press.

Huang, C.-T. James. 1984. “On the Distribution and Reference of Empty Pronouns”. Linguistic Inquiry 15, 531-544.

Huang, C.-T. James. 1989. “Pro-Drop in Chinese: A Generalized Control Theory”, in Jaeggli, Osvaldo and Ken Safir (eds.). The Null Subject Parameter. Dordrecht: Kluwer, 185-214.

Huang, Yan. 1995. “On Null Subjects and Null Objects in Generative Grammar”. Linguistics 33, 1081-1123.

Huddleston, Rodney. 1978. “On the Constituent Structure of VP and Aux”. Linguistic Analysis 4, 31-59.

Huddleston, Rodney. 1980. “Criteria for Auxiliaries and Modals”, in Greenbaum, Sidney, Geoffrey Leech, and Jan Svartvik (eds.). Studies in English Linguistics for Randolph Quirk. London: Longman, 65-78.

Huntley, Martin. 1980. “Propositions and the Imperative”. Synthese 45, 281-310.

Huntley, Martin. 1982. “Imperatives and Infinitival Embedded Questions”, in Schneider, Robinson, Kevin Tuttle, and Robert Chametzky (eds.). Papers from the Parasession on Nondeclaratives. Chicago Linguistics Society, 93-106.

Huntley, Martin. 1984. “The Semantics of English Imperatives”. Linguistics and Philosophy7, 103-133.

Iatridou, Sabine. 1990. About Agr(P). Linguistic Inquiry 21, 551-577.

Iveland, Paula. 1993. “VP Small Clauses”. LRC-93-03. Linguistics Research Center, University of California, Santa Cruz.

Iwakura, Kunihiro. 1977. “The Auxiliary System in English”. Linguistic Analysis 3, 101- 136.

Jackendoff, Ray. 1972. Semantic Interpretation in Generative Grammar. Cambridge: MIT Press.

Jaeggli, Osvaldo. 1982. Topics in Romance Syntax. Dordrecht: Foris Publications.

Jaeggli, Osvaldo. 1986. “Three Issues in the Theory of Clitics: Case, Doubled NPs, and Extraction”, in Borer, Hagit (ed.). Syntax and Semantics 19: The Syntax o f Pronominal Clitics. Orlando, Fla.: Academic Press, Inc.

Jaeggli, Osvaldo and Ken Safir. 1989a. The Null Subject Parameter. Dordrecht: Kluwer.

Jaeggli, Osvaldo and Ken Safir. 1989b. “The Null Subject Parameter and Parametric Theory”, in Jaeggli, Osvaldo and Ken Safir (eds.). The Null Subject Parameter. Dordrecht: Kluwer, 1-44.


450

Jespersen, Otto. 1961. A Modem English Grammar on Historical Principles, Part V: Syntax (Fourth Volume). Copenhagen: Ejnar Munksgaard.

Johnson, Kyle. 1988. “Verb Raising and Have”, in McGill Working Papers in Linguistics: Special Issue on Comparative German Syntax. Montreal: McGill University, 156- 167.

Johnson, Kyle. 1991. “Object Positions”. Natural Language and Linguistic Theory 9,577- 636.

Johnston, Michael. 1994. “Because Clauses and Negative Polarity Licensing”, in Proceedings o f the Eastern States Conference on Linguistics. Ithaca, NY: DMLL Publications, Cornell University.

Kaisse, Ellen. 1983. “The Syntax of Auxiliary Reduction in English”. Language 59, 93- 122.

Katz, Jerrold. 1977. Propositional and Elocutionary Force. New York: Crowell.

Katz, Jerrold and Paul Postal. 1964. An Integrated Theory o f Linguistic Descriptions. Cambridge, Ma.: MIT Press.

Kayne, Richard S. 1984. Connectedness and Binary Branching. Dordrecht: Foris Publications.

Kayne, Richard S. 1989. “Null Subjects and Clitic Climbing”, in Jaeggli, Osvaldo and Ken Safir (eds.). The Null Subject Parameter. Dordrecht: Kluwer, 239-262.

Kayne, Richard S. 1991. “Romance Clitics, Verb Movement, and PRO”. Linguistic Inquiry 22, 647-686.

Kayne, Richard S. 1994. The Antisymmetry o f Syntax. Cambridge, Ma.: MIT Press.

Keenan, Edward. 1976. “Towards a Universal Definiton of Subject”, in Li, Charles N. (ed.). Subject and Topic. New York: Academic Press, 303-334.

Kenstowicz, Michael. 1989. “The Null Subject Parameter in Modem Arabic Dialects”, in Jaeggli, Osvaldo and Ken Safir (eds.). The Null Subject Parameter. Dordrecht: Kluwer, 263-276.

Kim, Jong-Bok. 1995. “English Negation from a Non-Derivational Perspective”, in Ahlers, Jocelyn; Leela Bilmes; Joshua S. Guenter; Barbara A. Kaiser; and Ju Namkung (eds.). Proceedings o f the 21st Annual Meeting o f the Berkeley Linguistics Society. Berkeley: Berkeley Linguistics Society, 186-196.

Kim, Jong-Bok and Ivan A. Sag. 1995. “French and English Negation: A Lexicalist Alternative to Head Movement”, ms. Stanford University.


451

Klein, S. 1976. “A Base Analysis of the Floating Quantifier in French”, in Kegl, Judy, D. Nash, and Annie Zaenen (eds.). Proceedings o f the Seventh Annual Meeting o f the North Eastern Linguistics Society. Amherst, Ma.: GLSA, Department of Linguistics, University of Massachusetts, Amherst.

Klima, Edward S. 1964. “Negation in English”, in Fodor, Jerry A. and Jerrold J. Katz (eds.). The Structure of Language: Readings in the Philosophy o f Language. New Jersey: Prentice-Hall, Inc.

Koizumi, Masatoshi. 1993. “Object Agreement Phrases and the Split VP Hypothesis”, in MIT Working Papers in Linguistics 18: Papers on Case and Agreement I. Cambridge, Ma.: M1TWPL, Department of Linguistics and Philosophy, MIT, 99- 148.

Koopman, Hilda. 1984. The Syntax o f Verbs: From Verb-Movement Rules in the Kru Languages to Universal Grammar. Dordrecht: Foris.

Kooman, Hilda. 1992. “On the Absence of Case Chains in Bambara”. Natural Language and Linguistic Theory 10, 55-594.

Koopman, Hilda and Dominique Sportiche. 1991. ‘The Position of Subjects”. Lingua 85, 211-258.

Koster, Jan. 1975. “Dutch as an SOV Language”. Linguistic Analysis 1, 111-136.

Kratzer, Angelika. 1993. “The Event Argument”, ms. University of Massachusetts at Amherst.

Kuroda, Shige-Yuki. 1988. “Whether We Agree or Not: A Comparative Syntax of English and Japanese”, in Poser, William (ed.). Papers from the Second International Workshop on Japanese Syntax. Chicago: University of Chicago Press.

Ladusaw, William. 1980[1979]. Polarity Sensitivity as Inherent Scope Relations. New York: Garland Publishing Co.

Ladusaw, William. 1988. “Adverbs, Negation, and QR”, in The Linguistic Society of Korea (ed.). Linguistics in the Morning Calm 2. Seoul: Hanshin Publishing Co.

Ladusaw, 1992. “Expressing Negation”, in Barker, Chris and David Dowty (eds.). Proceedings o f the Second Conference on Semantics and Linguistic Theory. Ohio State Working Papers in Linguistics, vol. 40. Columbus: The Ohio State University.

Laka, Itziar. 1990. “Negation in Syntax: On the Nature of Functional Categories and Projections”. MIT Ph.D dissertation.

Lakoff, George. 1987. Women, Fire, and Dangerous Things: What Categories Reveal about the Mind. Chicago: University of Chicago Press.

Lakoff, George and Mark Johnson. 1980. Metaphors We Live By. Chicago: University of Chicago Press.


452

Langacker, Ronald. 1969. “On Pronominalization and the Chain of Command”, in Reibel, D. and S. Schane (eds.). Modem Studies in English. New Jersey: Prentice-Hall, 160-186.

Larson, Richard K. 1985. “Bare-NP Adverbs”. Linguistic Inquiry 16, 595-621.

Lasnik, Howard. 1976. “Remarks on Coreference”. Linguistic Analysis 2, 1-22.

Lasnik, Howard. 1981. “Restricting the Theory of Transformations: A Case Study”, in Homstein, Norbert and David Lightfoot (eds.). Explanations in Linguistics: The Logical Problem o f Language Acquisition. London: Longman.

Lasnik, Howard. 1995a. “A Note on Pseudogapping”, in MIT Working Papers in Linguistics 27: Papers on Minimalist Syntax. Cambridge, Ma.: MITWPL, Department of Linguistics and Philosophy, MIT.

Lasnik, Howard. 1995b. “Verbal Morphology: Syntactic Structures Meets the Minimalist Program” in Kempchinsky, Paula and Hector Campos (eds.). Evolution and Revolution in Linguistic Theory: Essays in Honor o f Carlos Otero. Georgetown: Georgetown University Press.

Lasnik, Howard and Mamoru Saito. 1992. Move-a. Cambridge, Ma.: MIT Press.

Leech, Geoffrey N. 1983. Principles o f Pragmatics. London: Longman Press.

Levin, Beth and Rappaport Hovav Malka. 1995. Unaccusativity: At the Syntax-Lexical Semantics Interface. Cambridge, Ma.: MIT Press.

Levinson, Stephen. 1983. Pragmatics. Cambridge: Cambridge University Press.

Li, Charles N. 1976. Subject and Topic. New York: Academic Press.

Liberman, Mark. 1974. “On Conditioning the Rule of Subject-Auxiliary Inversion”, in Kaisse, Ellen and Jorge Hankamer (eds.). Papers from the Fifth Annual Meeting of the North East Linguistic Society. Cambridge: Harvard University Press.

Lightfoot, David and Norbert Homstein (eds.). 1994a. Verb Movement. Cambridge: Cambridge University Press.

Lightfoot, David and Norbert Homstein. 1994b. “Verb Movement: An Introduction”, in Lightfoot, David and Norbert Homstein (eds.). Verb Movement. Cambridge: Cambridge University Press, 1-18.

Lillo-Martin, Diane. 1986. “Two Kinds of Null Arguments in American Sign Language”. Natural Language and Linguistic Theory 4,415-444.

Linebarger, Marcia. 1987. The Grammar of Negative Polarity. MIT Ph.D dissertation.

Lobeck, Anne. 1987. Syntactic Constraints on VP Ellipsis. University of Washington Ph.D dissertation.


453

Lobeck, Anne. 1991. “Phrase Structure of Ellipsis in English”, in Rothstein, Susan (ed.). Syntax and Semantics, Volume 25. New York: Academic Press.

Lobeck, Anne. 1992. “Licensing and Identification of Ellipted Categories in English”, in Berman, Steve and Arild Hestvik (eds.). Proceedings of the Stuttgart Ellipsis Workshop.

Lobeck, Anne. 1995. Ellipsis: Functional Heads, Licensing, and Identification. Oxford: Oxford University Press.

Lopez, Luis. 1994. ‘The Syntactic Licensing of VP-Ellipsis: A Comparative Study of Spanish and English”, in Mazzola, Michael L. (ed.). Issues and Theory in Romance Linguistics: Selected Papers from the Linguistic Symposium on Romance Languages XXIII. Washington D.C.: Georgetown University Press, 333-354.

Maling, Joan M. 1976. “Notes on Quantifier-Postposing”. Linguistic Inquiry 7, 708-718.

Marantz, Alec. 1995. ‘The Minimalist Program”, in Webelhuth, Gert (ed.). Government and Binding Theory and the Minimalist Program: Principles and Parameters in Syntactic Theory. Cambridge, Ma.: Blackwell.

May, Robert. 1985. Logical Form: Its Structure and Derivation. Cambridge, Ma.: MIT Press.

McConnell-Ginet. Sally. 1982. “Adverbs and Logical Form: A Linguistically Realistic Theory”. Language 58, 144-184.

McCawley, James D. 1988. The Syntactic Phenomena o f English. Chicago: University of Chicago Press.

McCloskey, James. 1991. “Clause Structure, Ellipsis and Proper Government in Irish”, in McCloskey, James (ed.). The Syntax of Verb-Initial Languages, Lingua Special Edition, 259-302.

McCloskey, James. 1992. “Adjunction, Selection, and Embedded Verb Second”. LRC-92-07. Linguistics Research Center, University of California, Santa Cruz.

McCloskey, James. 1993. “A Crude Test for Unaccusativity in English”, in Pullum, Geoffrey K. and Eric Potsdam (eds.). Syntax at Santa Cruz, Volume 2. Santa Cruz: Linguistic Research Center.

McCloskey, James. 1996a. “On the Scope of Verb Movement in Irish”. Natural Language and Linguistic Theory 14, 47-104.

McCloskey, James. 1996b. “Subjects and Subject Positions in Irish”, in Borsley, Robert and Ian Roberts (eds.). The Syntax o f the Celtic Languages, A Comparative Perspective. Cambridge: Cambridge University Press, 241-283.

McCloskey, James, to appear. “Subjecthood and Subject Positions”, in Haegeman, Liliane (ed.). A Handbook of Formal Syntax. Dordrecht: Kluwer Academic Publishers.


454

McCloskey, James, in preparation. Irish Syntax: Clauses and Clause-Types.

Merchant, Jason. 1996. “Object Scrambling and Quantifier Float in German”, in The Proceedings of the Twenty-Sixth Annual Meeting o f the Northeastern Linguistic Society. Amherst, Ma.: GLSA, Department of Linguistics, University of Massachusetts, Amherst.

Perlmutter, David M. 1971. Deep and Surface Constraints in Syntax. New York: Holt, Reinhart, Winston.

Perlmutter, David M. 1978. “Impersonal Passives and the Unaccusative Hypothesis”, in Proceedings of the Fourth Annual Meeting o f the Berkeley Linguistics Society. Berkeley: Berkeley Linguistics Society, 157-189.

Perlmutter, David M. and Paul Postal. 1984. “The 1-Advancement Exclusiveness Law”, in Perlmutter, David M. and Carol Rosen (eds.). Studies in Relational Grammar 2. Chicago: University of Chicago Press, 81-125.

Pesetsky, David. 1987. “W?i-in-situ. Movement and Unselective Binding”, in Reuland, Eric and Alice ter Meulen (eds.). The Representation o f (In)definiteness. Cambridge, Ma.: MIT Press.

Platzack, Christer. 1987. “The Scandinavian Languages and the Null Subject Parameter”. Natural Language and Linguistic Theory 5,377-402.

Platzack, Christer. 1994. “Null Subjects, Weak Agr, and Syntactic Differences in Scandinavian”. Working Papers in Scandinavian Syntax 53, 85-106.

Platzack, Christer and Inger Rosengren. 1994. “On the Subject of Imperatives: A Minimalist Account of the Imperative Pronoun and Negated Imperatives”, ms. Lund University.

Pollard, Carl and Ivan Sag. 1994. Head-Driven Phrase Structure Grammar. Chicago: University of Chicago Press.

Pollock, Jean-Yves. 1989. “Verb Movement, Universal Grammar, and the Structure of IP”. Linguistic Inquiry 20, 365-424.

Postal, Paul M. 1970. “On So-Called Pronouns in English”, in Jacobs, Roderick and Peter Rosenbaum (eds.). Readings in English Transformational Grammar. Waltham, Ma.: Ginn and Company.

Postal, Paul M. 1974. On Raising. Cambridge, Ma.: MIT Press.

Potsdam, Eric, in press. “NegP and Subjunctive Complements in English”. Linguistic Inquiry.

Potsdam, Eric, to appear. “English Verbal Morphology and VP Ellipsis”, in The Proceedings of the Twenty-Seventh Annual Meeting o f the North East Linguistic Society. Amherst, Ma.: GLSA, University of Massachusetts at Amherst.


455

Poutsma, H. 1928. Grammar of Late Modem English Part I: The Sentence, First Half: The Elements o f a Sentence. 2nd Edition. Groningen: P. Noordhoff.

Progovac, Ljiljana. 1994. Negative and Positive Polarity: A Binding Approach. Cambridge: Cambridge University Press.

Pullum, Geoffrey K. 1982. “Syncategorematicity and English Infinitival to". Glossa 16, 181-215.

Pullum, Geoffrey and Deirdre Wilson. 1977. “Autonomous Syntax and the Analysis of Auxiliaries”. Language 53, 741-788.

Quirk, Randolph, Sidney Greenbaum, Geoffrey Leech, and Jan Svartvik. 1985. A Comprehensive Grammar o f Contemporary English. London: Longman. Press.

Radford, Andrew. 1988. Transformational Grammar: A First Course. Cambridge: Cambridge University Press.

Rizzi, Luigi. 1982. Issues in Italian Syntax. Dordrecht: Foris.

Rizzi, Luigi. 1986. “Null Objects in Italian and the Theory of pro". Linguistic Inquiry 17, 501-557.

Rizzi, Luigi. 1990. Relativized Minimality. Cambridge, Ma.: MIT Press.

Rizzi, Luigi. 1991. “Residual Verb Second and the Wh-Criterion”, ms. University of Geneva.

Rizzi, Luigi and Ian Roberts. 1989. “Complex Inversion in French”. Probus 1, 1-30.

Roberge, Yves. 1990. The Syntactic Recoverability o f Null Arguments. Montreal: McGill- Queen’s University Press.

Roberts, Ian G. 1985. “Agreement Parameters and the Development of English Modal Auxiliaries”. Natural Language and Linguistic Theory 3, 21-58.

Roberts, Ian. 1990. “Some Notes on VP-Fronting and Head Government”, in Mascaro, Joan and Marina Nespor (eds.). Grammar in Progress: GLOW Essays for Henk van Riemsdijk. Dordrecht: Foris Publications, 387-396.

Rochemont, Michael. 1985. A Theory o f Stylistic Rules in English. New York: Garland Publishing, Inc.

Rochemont, Michael. 1989. “Topic Islands and the Subjacency Parameter”. Canadian Journal o f Linguistics 34, 145-170.

Rohrbacher, Bernhard W. 1994. “The Germanic VO Languages and the Full Paradigm: A Theory of V to I Raising”. University of Massachusetts Ph.D dissertation.

Rooth, Mats. 1985. “Association with Focus”. University of Massachusetts at Amherst Ph.D dissertation.


456

Ross, John Robert. 1967. “Constraints on Variables in Syntax”. MIT Ph.D dissertation.

Ross, John Robert. 1968. “On Declarative Sentences”, in Jacobs, Roderick A. and PeterS. Rosenbaum (eds.). Readings in English Transformational Grammar. Waltham: Ginn and Company.

Ross, John Robert. 1969. “Auxiliaries as Main Verbs”. Studies in Philosophical Linguistics 1, 77-102.

Rouveret, Alain. 1989. “Cliticisation et Temps en Portugais Europeen”. Revue de Langues Romanes 93, 337-371.

Rudanko, Juhani. 1982. “Towards a Description of Negatively Conditioned Subject Operator Inversion in English”. English Studies 63, 348-359.

Runner, Jeffrey T. 1994. “Noun Phrase Licensing and Interpretation”. University of Massachusetts at Amherst Ph.D dissertation.

Runner, Jeffrey T. 1995. “Overt and LF Object Positions in English”, in Gabriele, Leslie and Robert Westmoreland (eds.). Proceedings of the Sixth Annual Meeting of the Formal Linguistics Society o f Midamerica. Bloomington: Indiana University Linguistics Club Publications, 35-46.

Sadock, Jerrold. 1974. Towards a Linguistic Theory o f Speech Acts. New York: Academic Press.

Sadock, Jerrold M. and Arnold M. Zwicky. 1985. “Speech Act Distinctions in Syntax”, in Shopen, Timothy (ed.). Language Typology and Syntactic Description. Volume I: Clause Structure. Cambridge: Cambridge University Press.

Safir, Ken. 1986. “Relative Clauses in a Theory of Binding and Levels”. Linguistic Inquiry 17, 663-689.

Safir, Ken. 1994. “Comments on Quicoli: PRO and pro”, ms. Rutgers University.

Sag, Ivan A. 1978. “Floated Quantifiers, Adverbs, and Extraction Sites”. Linguistic Inquiry 9, 146-150.

Sag, Ivan A. 1980[ 1976]. Deletion and Logical Form. New York: Garland Publishing, Inc.

Sag, Ivan A. and Janet Dean Fodor. 1995. “Extraction Without Traces”, in Aranovich, Raul, William Byrne, Susanne Preuss, and Martha Senturia (eds.). The Proceedings o f the Thirteenth West Coast Conference on Formal Linguistics. Stanford, Ca.: Stanford Linguistics Association.

Schachter, Paul. 1977. “Does She or Doesn’t She?” Linguistic Inquiry 8, 763-767.

Schachter, Paul. 1978. “English Propredicates”. Linguistic Analysis 4, 187-224.


457

Schmerling, Susan. 1975. “Imperative Deletion and Some Related Matters”. Linguistic Inquiry 6, 501-511.

Schmerling, Susan. 1976. Aspects o f English Sentence Stress. Austin: University of Texas Press.

Schmerling, Susan. 1977. “The Syntax of English Imperatives”, ms. University of Texas at Austin.

Schmerling, Susan. 1982. “How Imperatives are Special, and How They Aren’t”, in Schneider, Robinson, Kevin Tuttle, and Robert Chametzky (eds.). Papers from the Parasession on Nondeclaratives. Chicago Linguistics Society, 202-218.

Searle, John. 1969. Speech Acts: An Essay in the Philosophy of Language. Cambridge: Cambridge University Press.

Selkirk, Elisabeth. 1984. Phonology and Syntax: The Relation Between Sound and Structure. Cambridge, Ma.: MIT Press.

Selkirk, Elisabeth. 1986. “On Derived Domains in Sentence Phonology”, in Phonology Yearbook 3, 371-405.

Sells, Peter. 1983. “Juncture and the Phonology of Auxiliary Reduction in English”, in Finer, D. and T. Borowsky (eds.). University of Massachusetts Occasional Papers in Linguistics 8. Amherst, Ma.: Graduate Linguistics Students Association.

Seppanen, Aimo. 1977. “The Position of Let in the English Auxiliary System.” English Studies 58, 515-529.

Shlonsky, Ur. 1991. “Quantifiers as Functional Heads: A Study of Quantifier Float in Hebrew”. Lingua 84, 159-180.

Sigurbsson, Halldor Armann. 1993. “Argument-Drop in Old Icelandic”. Lingua 89, 247- 280.

Soames, Scott and David M. Perlmutter. 1979. Syntactic Argumentation and the Structure o f English. Berkeley: University of California Press.

Speas, Margaret. 1994. “Null Arguments in a Theory of Economy of Projection”, in University o f Massachusetts Occasional Papers in Linguistics 17, 179-208.

Sportiche, Dominique. 1988. “A Theory of Floating Quantifiers and Its Corollaries for Constituent Structure”. Linguistic Inquiry 19,425-449.

Stafford, R. L. 1967. An Elementary Luo Grammar. Nairobi: University of Oxford Press.

Stalnaker, Robert. 1974. “Pragmatic Presuppositions” in Munitz, M. and P. Unger (eds.). Semantics and Philosophy. New York: New York University Press, 197-213.

Stalnaker, Robert. 1979. “Assertion”, in Cole, Peter (ed.). Syntax and Semantics 9: Pragmatics. New York: Academic Press, 315-332.


458

Steele, Susan; Adrian Akmajian; Richard Demers; Eloise Jelinek; Chisato Kitagawa; Richard Oehrle; and Thomas Wasow. 1981. An Encyclopedia o f AUX: A Study in Cross-Linguistic Equivalence. Cambridge, MA: MTT Press.

Stein, Dieter. 1990. The Semantics o f Syntactic Change: Aspects o f the Evolution o f Do in English. Berlin: Mouton de Gruyter.

Stockwell, Robert P., Paul Schacter, and Barbara Hall Partee. 1973. Major Syntactic Structures o f English. New York: Holt, Rinehart and Winston, Inc.

Sweet, Henry. 1891-1898. A New English Grammar, 2 Vols. Oxford: Oxford University Press.

Szabolcsi, Anna and Frans Zwarts. 1993. “Weak Islands and an Algebraic Semantics for Scope-Taking”. Natural Language Semantics 1, 235-284.

Takezawa, Koichi. 1984. Perfective Have and the Bar Notation”. Linguistic Inquiry 15, 675-687.

Taraldsen, Tarald. 1978. On the NIC, Vacuous Application, and the That-Trace Filter. Bloomington: Indiana University Linguistics Club.

Thome, J. P. 1966. “English Imperative Sentences”. Journal o f Linguistics 2, 69-78.

Travis, Lisa. 1984. “Parameters and Effects of Word Order Variation”. MIT Ph.D dissertation.

Travis, Lisa. 1988. "The Syntax of Adverbs”, in McGill Working Papers in Linguistics: Special Issue on Comparative German Syntax. Montreal: McGill University, 280- 310.

Vikner, Sten. 1995. Verb Movement and Expletive Subjects in the Germanic Languages. Oxford: Oxford University Press.

Visser, F. Th. 1963a. An Historical Syntax o f the English Language, Part One: Syntactical Units with One Verb. Leiden: E.J. Brill.

Visser, F. Th. 1963b. An Historical Syntax o f the English Language, Part Three: First Half Syntactical Units with Two Verbs. Leiden: E.J. Brill.

Warner, Anthony R. 1985. The Structure of English Auxiliaries: A Phrase Structure Grammar. Bloomington, Ind.: Indiana University Linguistics Club.

Warner, Anthony R. 1993. English Auxiliaries: Structure and History. Cambridge: Cambridge University Press.

Wasow, Thomas. 1972. “Anaphoric Relations in English”. MIT Ph.D dissertation.

Wilder, Chris and Damir Cavar. 1994. “Word Order Variation, Verb Movement, and Economy Principles”. Studia Linguistica 48,46-86.


459

Williams, Edwin S. 1984. “77zere-Insertion”. Linguistic Inquiry 15, 131-153.

Williams, Edwin. 1994. “A Reinterpretation of Evidence for Verb Movement in French”, in Lightfoot, David and Norbert Homstein (eds.). Verb Movement. Cambridge: Cambridge University Press, 189-206.

Wilson, Deirdre and Dan Sperber. 1988. “Mood and the Analysis of Non-Declarative Sentences”, in Dancy, Jonathan, J. M. E. Moravcsik, and C. C. W. Taylor (eds.). Human Agency: Language, Duty, and Value. Stanford, Ca.: Stanford University Press.

Zagona, Karen. 1982. “Government and Proper Government of Verbal Projections”. University of Washington Ph.D dissertation.

Zagona, Karen. 1988a. Verb Phrase Syntax: A Parametric Study o f English and Spanish. Dordrecht: Kluwer Academic Publishers.

Zagona, Karen. 1988b. “Proper Government of Antecedentless VP in English and Spanish”. Natural Language and Linguistic Theory 6, 95-128.

Zanuttini, Raffaella. 1991. “Syntactic Properties of Sentential Negation: A Comparative Study of Romance Languages”. University of Pennsylvania Ph.D dissertation.

Zhang, Shi. 1990. “The Status of Imperatives in Theories of Grammar”. University of Arizona Ph.D dissertation.

Zwicky, Arnold. 1969. “Phonological Constraints in Syntactic Description”. Papers in Linguistics 1, 411-463.

Zwicky, Arnold. 1970. “Auxiliary Reduction in English”. Linguistic Inquiry 1, 323-336.

Zwicky, Arnold. 1974. “Hey, Whatsyoumame!”, in La Galy, Michael W., Robert Fox, and Anthony Bruck (eds.). Papers from the Tenth Regional Meeting o f the Chicago Linguistic Society. Chicago: Chicago Linguistic Society, 787-801.

Zwicky, Arnold. 1987. book review of Davies 1986a in Language 25, 805-810.

Zwicky, Arnold. 1988. “On the Subject of Bare Imperatives in English”, in Duncan-Rose, Caroline (ed.). Rhetorica, Phonoligica, Syntactica: A Festschrift fo r Robert P. Stoclcwell from his friends and Colleagues. London: Routledge. 437-450.

Zwicky, Arnold M. and Geoffrey K. Pullum. 1983. “Cliticization vs. Inflection: English n ’f \ Language 59, 502-513.


syntactic issues in the english imperative

Documents