RESTRUCTURING AND INFINITIVES: THE VIEW FROM APPALACHIA
By
Gregory Johnson II

A DISSERTATION
Submitted
to Michigan State University
in partial fulfillment of the requirements
for the degree of
Linguistics – Doctor of Philosophy
2014

ABSTRACT
RESTRUCTURING AND INFINITIVES: THE VIEW FROM APPALACHIA
By
Gregory Johnson II
This dissertation explores four instances of restructuring (clause union) in verb forms found
in the Appalachian English (AppE) of eastern Kentucky. I compare novel restructuring
analyses of causative and experiencer have, perception verbs, and the verb liketa. In addition I examine restructuring as reanalysis in the non-finite wh-adjunct how come. In each
case, syntactic and semantic diagnostics for clause structure are used to identify types of
restructuring and to develop a syntax for these understudied forms. One main goal of the
dissertation is to apply Wurmbrand’s mono-clausal approach to restructuring (Wurmbrand
2001) to cases of micro-variation between standard and Appalachian English. This reveals
the difference between morphosyntactic variation present in bundles of features manipulated
by the syntax and variation which ultimately stems not from the syntax, but from differences in the inventory of post-syntactic morphological rules in standard and non-standard
grammars. For instance, I argue that variation in the form of causative and experiencer have
complements in AppE is not the result of a different syntax but rather the result of variation
in post-syntactic morphological rules. Investigation of the boundary between variation in
syntactic objects and morphological operations performed on those objects will allow us a
cleaner view of micro-variation in restructuring phenomena. On the other hand, the study of
perception complements shows that differences in complements of this dialect are not always
only superficially different. For example, in contrast with Standard English, perception complements in the dialect allow matrix passivization. Following major strands in the literature
which argue that passive participle -en blocks some relations between matrix and embedded
clauses, this is predicted in a language where the perfect and passive syntactic heads do not
license participle morphology on the verb. I argue that Appalachian English is just such a

dialect. Irregular forms have leveled to the extent that that distinction between preterite and
participle forms is unclear. I argue that matrix passivization in perception verbs is the result
of grammatical reanalysis of the perfect and participle heads; they are no longer interveners.
This means that any distinctions in past and participle forms present in adult grammars
of AppE speakers are superficial and occur by way of what I have termed ‘morphological
enrichment’ or unleveling; a process that most likely occurs in language/dialect contact situations with the standard. Essentially, morphological enrichment is the opposite of leveling
by impoverishment. Secondly, comparison of perception complements with the restructuring
have complements reveals that a unified account of the behavior of both sets of complements
requires independent mechanisms such as agreement or head-movement; mechanisms previously used to explain restructuring in some bi-clausal accounts. Analysis of the AppE verb
form liketa reveals a syntax similar to wanna contraction found in Standard English. Again,
I apply a Wurmbrand style mono-clausal approach to restructuring to the form and make a
theoretical comparison with a previous bi-clausal approach to wanna contraction. The major finding is that the mono-clausal approach is more favorable as it requires less stipulation
with respect to restructuring. However, in the end, a proper analysis and comparison of
the forms requires that they be differentiated by agree relations between the restructuring
verb and the embedded auxiliaries. Finally, I analyze the syntax of non-finite how come
complements found in AppE as a type of matrix clause restructuring and reanalysis.

To my friends and family.

iv

ACKNOWLEDGEMENTS

I am sure that I will not be able to mention everyone who has helped me finish this dissertation in the space provided. I will give it my best shot. To everyone that helped me over
the past few years, thank you.
I would of course like to thank my committee members for bearing with me through all of
those times I just didn’t get something and encouraging me/lighting a fire under my ass along
the way. A special thanks goes out to my committee chair Alan Munn for being a mentor.
I hope to emulate him when it comes to thinking about problems, analyses, and making
sound arguments. Similarly, I would like to thank Cristina Schmitt for her encouragement,
perspective, knowledge, and ability to see connections between things when others do not. I
could never have done this without either of them. Any failure in argumentation or lack of
vision in this dissertation is entirely mine.
Aside from my committee, I owe a great deal to my friends and colleagues here in the
Linguistics Program and beyond. They really have made life bearable. Looking back grad
school seems divisible into two parts. There is the early part, where you have no idea what
is going on and just hope that what you say makes sense and the latter part in which you
have some idea of what is going on and during which almost nothing anyone says makes
sense. That said my early years in grad school were entirely too much fun. I would like to
thank Matthew ‘Elvis’ Husband for all of his friendship, encouragement, and willingness to
engage with almost any insane statement that came out of my mouth. He is my linguistics
older brother. In addition to Elvis, I could also never forget Olga Eremina, Ai Kubota, or
Jan Anderssen who always had insightful comments about my work. I really feel like these
four showed me the ropes in grad school. Outside of the program, I have to thank Micalee
Sullivan for being a great friend and convincing me that we should start a punk band in the
middle of our grad careers. This was a truly brilliant idea. The house shows were epic. I

v

would also like to recognize the Greater Lansing Hash House Harriers for showing me how to
make running enjoyable. The trick is to hide beer along the way. You always have to have
a hobbies, right? Later, when all of those awesome people were gone, new awesome people
appear. The list of awesome people includes but is not limited to Joe Jalbert my syntax
buddy, Jess Gamache my work buddy, Curt Anderson my semantics buddy, Kali Byble my
syntax little sister, Karthik Durvasula my argument buddy, Tuuli Morrill my brunch buddy,
and Kim Fenn my psych buddy. Thank you all for all of your help. Y’all also helped make
this dissertation possible.
Outside of academia entirely, I must thank a few more people. Ann, *rock*. Everyone
needs an evil twin. Lilly, thanks for being one of the most amazing people I have ever
met. You truly are weird. Dad, thanks for being the first person who encouraged me to
ask questions about the world. Sorry if it didn’t have quite the effect you intended. Mom,
thanks for listening and checking in on me. You are the best. Last but not least, I have to
thank Jason Zalack for being there while I completed this dissertation. You make everything
ok.

vi

TABLE OF CONTENTS

LIST OF FIGURES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

x

CHAPTER 1 INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . .
1.1 Goals of this work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.2 Appalachian English . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.2.1 Demographics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.2.2 On speaking and working with stigmatized dialects and dialect speakers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.3 Analytical frameworks . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.3.1 Generative assumptions . . . . . . . . . . . . . . . . . . . . . . .
1.3.2 Distributed Morphology . . . . . . . . . . . . . . . . . . . . . . .
1.4 Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1
4
4
5
7
9
9
11
13

CHAPTER 2 HAVE + TO CONSTRUCTIONS . . . . . . . . . . . . . . . 15
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.2 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.2.1 Have + to . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.2.2 General properties of causative and experiencer have . . . . . 19
2.2.2.1 Causative have . . . . . . . . . . . . . . . . . . . . . . . . 19
2.2.2.2 Experiencer have . . . . . . . . . . . . . . . . . . . . . . 24
2.2.3 Two haves, one analysis . . . . . . . . . . . . . . . . . . . . . . . 27
2.2.4 Causative and experiencer have as restructuring verbs . . . . 31
2.3 Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.3.1 Have is in v . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.3.2 Have selects an AspP complement . . . . . . . . . . . . . . . . 43
2.3.2.1 The status of to under restructuring have in AppE . 45
2.3.3 Default infinitival morphology . . . . . . . . . . . . . . . . . . . 47
2.3.3.1 The assignment of English verbal morphology . . . . 48
2.3.3.2 The assignment of verbal morphology in restructuring complements . . . . . . . . . . . . . . . . . . . . . 49
2.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
CHAPTER 3 PERCEPTION VERBS AND THEIR COMPLEMENTS . . 55
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
3.2 Properties of perception complements . . . . . . . . . . . . . . . . . . 59
3.2.1 Types of perception . . . . . . . . . . . . . . . . . . . . . . . . . . 59
3.2.1.1 Direct perception . . . . . . . . . . . . . . . . . . . . . . 60
3.2.1.2 Indirect perception . . . . . . . . . . . . . . . . . . . . . 62
3.2.2 Direct and indirect perception complements in AppE . . . . . 63
3.2.2.1 Direct perception in AppE . . . . . . . . . . . . . . . . 63
3.2.2.2 Indirect perception in AppE . . . . . . . . . . . . . . . 65
vii

3.3

3.4

3.5
3.6

3.2.3 Interim Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
Previous Analyses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
3.3.1 Bennis and Hoekstra (1989/2004) . . . . . . . . . . . . . . . . . 68
3.3.1.1 T-linking . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
3.3.1.2 Perception complements and t-linking in Dutch . . . 70
3.3.1.3 Implication for perception complements in AppE . . 73
3.3.2 Felser (1999) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
3.3.2.1 Felser’s assumptions . . . . . . . . . . . . . . . . . . . . 74
3.3.2.2 Event-control and matrix passivization . . . . . . . . . 77
3.3.3 Hornstein et al. (2008) . . . . . . . . . . . . . . . . . . . . . . . . 81
3.3.3.1 The active and passive as derivationally related . . . 82
3.3.3.2 The role of case bearing elements . . . . . . . . . . . . 83
3.3.3.3 The Case of the constraint against the passive . . . . 85
3.3.4 Summary of previous analyses . . . . . . . . . . . . . . . . . . . 87
Analysis of AppE perception complements . . . . . . . . . . . . . . . . 88
3.4.1 Reiteration of theoretical assumptions . . . . . . . . . . . . . . 89
3.4.1.1 The classification of perception complements in AppE 89
3.4.2 The restructuring verb and its complement . . . . . . . . . . . 90
3.4.3 Derivation of lexical restructuring perception complements
in AppE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
3.4.4 The status of participle morphology in AppE . . . . . . . . . . 93
3.4.5 Examples of leveling in AppE . . . . . . . . . . . . . . . . . . . . 93
3.4.5.1 Participle forms in AppE are not distinct . . . . . . . 95
3.4.6 The features of participles in AppE . . . . . . . . . . . . . . . . 96
3.4.7 Impoverishment and presumptions about the grammatical
baseline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
3.4.7.1 Leveling via Impoverishment . . . . . . . . . . . . . . . 101
What’s the difference? Causative have and perception complements 103
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105

CHAPTER 4 INFINITIVAL CONTRACTION AS RESTRUCTURING
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.2 Previous observations . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.2.1 Observations about liketa in Appalachian English . . . . . .
4.2.2 Liketa in Alabama English . . . . . . . . . . . . . . . . . . . .
4.2.3 Liketa in the historical record . . . . . . . . . . . . . . . . . .
4.3 Properties of liketa . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.3.1 Semantic properties . . . . . . . . . . . . . . . . . . . . . . . . .
4.3.1.1 Comparison with almost . . . . . . . . . . . . . . . . .
4.3.1.2 Semantic interpretations of liketa . . . . . . . . . . .
4.3.2 Syntactic properties . . . . . . . . . . . . . . . . . . . . . . . . .
4.3.3 Uncontracted liketa as a non-restructuring predicate . . . .
4.4 Contractions as restructuring complements . . . . . . . . . . . . . .
4.4.1 Liketa contraction as restructuring . . . . . . . . . . . . . . .
4.4.2 A head-movement analysis of contraction as restructuring .
viii

. 107
. 107
. 109
. 109
. 111
. 112
. 113
. 114
. 114
. 119
. 123
. 123
. 125
. 129
. 131

4.4.3 A bi-clausal head-movement analysis of liketa .
4.5 Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.5.1 Liketa is a restructuring verb generated in v . .
4.5.2 Complements of liketa are headed by PerfP . . .
4.5.3 Deriving the ban on the perfect auxiliary . . . .
4.5.4 Accounting for variation in liketa complements .
4.5.5 Mono vs. bi-clausal approaches to restructuring
4.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . .

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

CHAPTER 5 NON-FINITE HOW COME . . . . . . . . . . . . . . . . . . .
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.2 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.2.1 Non-finite how come in context . . . . . . . . . . . . . . . . . .
5.3 Previous analyses of how come . . . . . . . . . . . . . . . . . . . . . . .
5.3.1 How come and what for . . . . . . . . . . . . . . . . . . . . . . .
5.3.1.1 How come bars subject/aux inversion . . . . . . . . .
5.3.1.2 How come blocks long distance contstrual . . . . . .
5.3.2 Why and how come . . . . . . . . . . . . . . . . . . . . . . . . . .
5.3.2.1 How come as a syntactic unit merged in C . . . . . .
5.3.2.2 How come is illicit in multiple wh-questions . . . . .
5.3.2.3 How come does not interact with quantifier scope .
5.3.2.4 Against how come as a reduced clause . . . . . . . . .
5.3.3 How come as a secondary wh-adjunct . . . . . . . . . . . . . .
5.3.3.1 Why the hell as a secondary wh-adjunct . . . . . . . .
5.3.3.2 What as a secondary wh-adjunct across languages .
5.3.3.3 The syntax of secondary wh-adjuncts . . . . . . . . . .
5.3.3.4 Accounting for the lack of auxiliary inversion . . . . .
5.3.3.5 The issue of finiteness . . . . . . . . . . . . . . . . . . .
5.4 Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.4.1 Properties of non-finite how come . . . . . . . . . . . . . . . . .
5.4.1.1 Non-finite how come behaves like finite how come .
5.4.1.2 Non-finite how come selects non-restructuring complements . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.4.2 The syntax of non-finite how come . . . . . . . . . . . . . . . .
5.4.2.1 Reviving the reduced matrix clause analysis . . . . .
5.4.2.2 Case assignment and the embedded subject . . . . . .
5.4.2.3 Properties of come . . . . . . . . . . . . . . . . . . . . .
5.4.2.4 Verb second and null expletives . . . . . . . . . . . . .
5.4.2.5 The location of causal how : evidence from Chinese .
5.4.3 Deriving the core properties . . . . . . . . . . . . . . . . . . . . .
5.4.3.1 Evidence for the reduced root clause analysis . . . . .
5.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

134
138
139
144
145
147
149
150
151
151
152
152
153
153
154
154
156
156
158
159
160
161
161
162
165
170
171
173
173
174
175
176
178
178
179
181
184
185
188
189

BIBLIOGRAPHY . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190

ix

LIST OF FIGURES

Figure 1.1

Figure 1.2

Figure 1.3

Location of dialect area under discussion relative to the eastern portion
of the United States. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

5

Jonancy and communities within two hours walking distance along Highway 610. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

6

Jonancy in relation to Lexington, KY, Charleston, WV, and Kingsport,
TN. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

7

x

CHAPTER 1
INTRODUCTION

This dissertation is about the syntax of complementation in Appalachian English. It focuses
on points of syntactic and morphosyntactic variation found in infinitival complements in the
dialect. Put simply, this dissertation looks at how clauses are put together for a subset of
verbs and how those clauses differ in Appalachian English (AppE) from their counterparts
in Standard American English (SAE). Consider for example, the following verbs and the
variation in the complement clauses between dialects.
(1) Experiencer have
a. They had glasses [to break on them]
b. They had glasses [break on them]

AppE
SAE

(2) Causative have
a. I had Mary [to wash the dishes]
b. I had Mary [wash the dishes]

AppE
SAE

The Appalachian English versions of these sentences both contain an infinitival-to, which is
not licit in their Standard English counterparts. What accounts for this difference? Causative
and experiencer have are not the only instances in AppE where verbs select infinitival complements which differ from those found in Standard American English. Consider now, the
case of perception complements.
(3)

a.
b.

John saw Mary to leave.
* John saw Mary to leave.

AppE
SAE

This example shows that infinitival-to is banned in perception complements in Standard
American English but not in Appalachian English. However, infinitival-to isn’t totally ruled
out in SAE. It is acceptable in passivized perception complements (4a), and it is traditionally
assumed to be the passive counterparts of (4b).

1

(4)

a. Mary was seen to leave
b. John saw Mary leave

SAE
AppE/SAE

This assumption probably arose because SAE does not allow the bare passivized version
of (4b) below as in (5).
(5) Mary was seen leave.

*SAE

However, if we look at the data from Appalachian English, the idea that the active bare
perception complement in Standard English has the infinitival-to complement as its passive,
no longer makes sense. It seems that AppE has the full set of these complements.
(7) Passive AppE
a. Mary was seen leave.
b. Mary was seen to leave.

(6) Active AppE
a. John saw Mary leave.
b. John saw Mary to leave.

The fact that these verbs exhibit variation is not surprising. Causation and perception
complements belong to a very special set of verbs found in many languages called restructuring verbs. For example, it is still common practice for many linguists to assume that clauses
generally come in two types. They are thought to either appear as complementizer phrases
(CPs) as in example (8a) and (8b), tense phrases (TPs) in (8c).
(8)

a. I expected [CP that John would leave]
b. I expected [CP for John to leave]
c. I expected [TP PRO to leave]

That a majority of verbs select either CP or TP complements is not contested. However,
there exists a small subset of verbs which do not easily fit neatly into these categories. The
verbs in example (9) are all restructuring verbs.
(9)

a. I had John wash the dishes.
b. I had John burn my dinner on me.
c. I saw Mary leave.

causative have
experiencer have
perception verb

These verbs select complements which have different properties from the CP and TP complements selected by other verbs. Specifically, these complements are smaller than TP. It
2

is also commonly assumed that this smaller than TP status is the result of clause union;
either the product of some external mechanism such as head movement or that these verbs
simply select smaller than TP clauses directly. In the first part of this dissertation, I provide
a syntactic analysis of the variation in complements of causative and experiencer have and
perception complements.
Cases of restructuring can also be found and compared within the same dialect. AppE
has a lexical item liketa whose meaning is similar to the English adverb almost.
(10) John had liketa punched Bill before you arrived.
John had almost punched Bill before you arrived.
Liketa itself varies in terms of its morphosyntax in the dialect. For example, it may also
appear in an uncontracted form.
(11) John had liked to have punched Bill before you arrived.
Following recent work on infinitival contraction as restructuring, I provide an analysis of
this understudied form and propose that the relationship between the contracted and uncontracted forms is one of restructuring. The analysis of this form adds to an ongoing
discussion of choosing between two well known theories of restructuring found in the first
two case studies.
Lastly, it also seems to be the case that restructuring-like effects are not only found in
embedded complements. In the final case study I examine what appears to be a matrix complement which has undergone clausal changes that appear to be leading to clause unification.
Consider the case of how come and its non-finite counterpart found in Appalachian English.
(12)

a. How come he left the party?
b. How come him
to
leave the party?
WH
3s.acc inf leave the party
“How come he left the party?”

SAE

AppE

I argue that because non-finite how come in Appalachian English selects a TP infinitive it
is best analyzed as a reduced matrix clause.
3

1.1

Goals of this work

The goals of this work are twofold. First, there is a theoretical agenda which seeks to show
that rigorous inquiry into stigmatized and understudied dialects is worthwhile. It shows us
just how little we actually know about syntactic variation and it can inform larger theories
of language. The second goal of this work is more social. I believe that by subjecting
understudied and stigmatized dialects to rigorous linguistic inquiry, more linguists will come
to view them as interesting and worthy of study. This work seeks to build upon existing work
done by sociolinguists, most notably Wolfram and Christian (1976). Moreover, by having
linguists researching, working, and communicating with these communities the speakers also
come to see the dialect as one which is interesting, worthwhile, and valid as a way of speaking
for themselves and each other.
This dissertation highlights the importance of identifying morphosyntactic variation and
disentangling it from syntactic variation through these 4 case studies of forms involving
restructuring. In terms of determining clause size, failure to disentangle these two levels of
representation leads to misleading conclusions about both syntactic and morphosyntactic
variation. The major finding that this work has for theories of restructuring is that a hybrid
approach between the two main types under discussion is necessary in order to account for all
of the facts at hand. Such hybrid approaches must in turn pay special attention to divisions
between syntactic and morphosyntactic variation.

1.2

Appalachian English

My interest in this dialect has deep roots in my own life. The data contained in this dissertation is the result of grammatical acceptability judgements from myself, my family members
and other consultants from Southeastern Kentucky (including Pike and Letcher Counties)
and Northeastern Tennessee, in figure 1.1.
Since, I can not list the demographic and family information for all of my informants,

4

ArcGIS - appe

http://www.arcgis.com/home/webmap/p

appe
stuff

NHD Plus, NPS, Esri, DeLorme, FAO, USGS, EPA, NOAA

Figure 1.1: Location of dialect area under discussion relative to the eastern portion of the
United States.

I will use my own history as an example of the family and community histories commonly
found throughout this part of Appalachia.

1.2.1

Demographics

I was born in Whitesburg, KY (population 21391 ). It was the closest hospital to where I
grew up 30 miles away. It might not seem like a long drive but the geography of the mountain
roads at the time turn a 30 mile trip into about an hour’s drive. I grew up off Old US 23 in
a place called Jonancy, KY (population 97) on the banks of Shelby Creek.
Jonancy, shown in figure 1.2 below, is a small unincorporated community centered around
a post office. My dad was born and raised 6 miles away in a holler called Beefhide Creek
(population ∼ 732 ). My mother was born about a mile up the road in the other direction
1 All
1 ofcensus
1
data gleaned from www.census.gov
2 This is actually the approximate population of

5

12/2/13,

closest identifiable statistical entity in

11/21/13

ArcGIS - My Map

My Map

National Geographic, Esri, DeLorme, NAVTEQ, UNEP­WCMC, USGS, NASA, ESA, METI, NRCAN, GEBCO, NOAA,
iPC

Figure 1.2: Jonancy and communities within two hours walking distance along Highway
610.

in a place called Elwood which, though adjacent to Jonancy belongs to a larger statistical
area called Virgie (population 279). Virgie is the remnant of a mining community founded
in the one of the many coal booms of the earlier part of the 20th century. My grandparents
and great-grandparents all have similar stories. This patchwork of small isolated communities common to rural Appalachia and is pervasive throughout this part Eastern Kentucky.
Speakers from other areas of eastern Kentucky were consulted as well, including McRoberts
in Letcher County (population 784), Elkhorn (population 445), as well as Caney and Chloe
Creek (included within the statistical area of Pikeville).
The area is not itself home to any major businesses or sources of employment other
than coal mining, coal cleaning, and coal related manufacturing operations which are found
throughout www.arcgis.com/home/webmap/print.html
the county. As you might imagine local schools, groceries, and health-care clinics
the 2000 census.

6

1/1

My Map

Esri, DeLorme, FAO, USGS, EPA, NPS

Figure 1.3: Jonancy in relation to Lexington, KY, Charleston, WV, and Kingsport, TN.

make up the remainder of the jobs in this rural part of southeastern KY. Pikeville, KY
(population 6903) is now the nearest ‘city’ as well as the county seat. It is now located
approximately 20 minutes north of Jonancy on US 23. Before the four-lane portion of US 23
was completed around 1995 it took about an hour to get from Jonancy to Pikeville. To give
an idea of the relative isolation of this area to the rest of the world, the closest interstate
lies to the south and is now (again due to new roads) approximately 1 hour and 40 minutes
away in Kingport, TN. This is shown in figure 1.3.

1.2.2

On speaking and working with stigmatized dialects and dialect speakers

One of my earliest childhood memories involves a situation where a woman in a department
store in Pikeville made a derogatory comment about my English to me and my mother.
In some ways, this scene would continue to play itself out over and over again into my
undergraduate years at the University of Kentucky. I understand first hand that people
7

experience unnecessary dialect bigotry and exactly why some speakers of stigmatized dialects
might be suspicious of researchers and research goals. Needless to say, I take the social goal
of this dissertation very seriously.
Working with speakers of Appalachian English is both challenging and rewarding. Of
course the experience varies by individual speaker, but a lot of times when you ask a stigmatized dialect speaker for an acceptability judgment, even after you have heard them use
the form in question, what you get back is not entirely a straight answer. Instead the responses range from affirmative and negatives to things like “Well, I’ve heard people say
that but I don’t say that” or responses injected with more social commentary like “Yeah,
that sounds really ‘hickish’ and I’ve heard it before but I would never use it.” As is well
established in the sociolinguistic literature, this is because the average person is not entirely
aware of what people say or how they say it (see Labov, 1996). Normally, this fact does
not pose a serious problem for eliciting acceptability judgments, but many speakers have
internalized the stigma associated with speaking Appalachian English. To those speakers,
speaking Appalachian English is synonymous with ‘Hillybilly English’, ‘the way hicks talk’,
or just a general lack of education or lower socio-economic status. Thus the quite natural
phenomenon of lacking metalinguistic awareness is compounded by observer effects which
arise when speakers are asked about certain forms. For example, I first became interested
in the have + to forms of causative and experiencer have because my mother, family, and
acquaintances use them quite frequently in their speech. However, when asked about the
forms they either recognize it as non-standard and deny having said it or they assume simply
because you are interested in the form that it is ‘bad’ non-standard English and deny saying
it. Once, I asked a college educated childhood friend about the acceptability of the non-finite
how come form analyzed in this work. He instantly cringed, “That sounds so hick, it makes
my skin crawl” and followed this up with “But yeah, its totally acceptable”. This was quite
an informative statement for me as my mother, father, and myself all use the form. In all
likelihood, my friend does as well.

8

There are however speakers who are quite proud of the way they talk as Appalachians and
they provide rather different answers. I have found it necessary to include them in the mix
of speakers even though these speakers may have a tendency to be more permissive in their
acceptability judgments simply because they are proud of fact that the dialect is different.
In both cases, I have used my own acceptability judgments as an waypoint of sorts. As a
linguist, I know that ‘bad English’ is socially determined and does not reflect the state of the
grammar that produces it. As a syntactician, I know that dialects are more similar than they
are different. The key to working with stigmatized dialects lies in a diversity of informants.
Though it never hurts to be able to weigh in with your own acceptability judgments and
navigate the sea of social attitudes with insider knowledge, I firmly do not believe that it is
a necessary condition for doing linguistic fieldwork.

1.3

Analytical frameworks

This dissertation assumes the basic assumptions of generative syntax as found in The Minimalist Program (Chomsky, 1995; 2001). Though I assume common knowledge of minimalist
syntax, I will nevertheless lay out some basic assumptions of the framework here for the sake
of accessibility.

1.3.1

Generative assumptions

The generative approach to human language assumes that the faculty of language operates
relatively independent of other general cognitive systems and is itself part of the larger set of
cognitive systems such as the visual or auditory systems. The linguistic system reflects the
basic facets that we commonly associate with human language and is commonly conceived
of as having three subparts. For instance, perceiving and producing language is handled by
the perceptual/articulatory system. In generative traditions, instructions for this system are
the result of processes which occur at the level of phonological form or PF. Similarly, the

9

semantic or meaning component of language is handled by a conceptual-intentional system.
Thus processes involved in the composition of meaning happen at the level of logical form
or (LF). Finally, the structure associated of linguistic expressions is generated by the syntax
which must then interface with both PF and LF.
Though in the past it was assumed that the syntax manipulated words or lexical items
which were pulled from a speaker’s Lexicon, I will assume that this is not the case. Rather,
I will adopt the framework of Distributed Morphology which assumes that words and word
level processes are ‘distributed’ throughout the three part linguistic system just described.
Under this account, the syntax is only ever responsible for the manipulation of abstract
morphemes or bundles of formal features. I will return to Distributed morphology below.
The linguistic system is relatively invariant. In particular, I assume that syntax is universal and part of the biological endowment common to all human beings. In fact it is this
endowment which allows them to acquire and speak their respective languages. Thus we
posit a Universal Grammar to which all human languages conform.
In terms of syntactic analysis, I adopt the framework of the Minimalist Program (Chomsky, 1995) which among other things has sought to reduce the process by which linguistic
expressions are created to their bare conceptually necessary minimum. In the syntax, this is
achieved by assuming that all linguistic structure is the result of the operation Merge. Put
simply, Merge takes a syntactic object {a} and an object {b} and creates a new object {a, b}.
Other operations like Agree are tolerated only to the extent that they are necessary for making the structural descriptions produced by the syntax legible for the articulatory/perceptual
system (PF) and the conceptual/intentional system (LF).
Broadly, this means that the morphological correlates of inflection, words, and phrases
are assembled and combined together by the syntax in such way that they adhere to or meet
interface conditions which are imposed by requirements of mapping to adjacent linguistic
subsystems PF and LF.
Syntactic tree structures are then descriptions of operations performed on morphemes

10

by the syntax. The structures are constructed via merge. The features and possibly surface
distribution of the morphemes that make up those structures may be further altered by
agreement processes between features of syntactic objects in the derivation of the structure.
For example perfect verbal inflection in “Mary had kissed John” is the result of the
merger of the perfect head with the verbal phrase containing the verb kiss.
(13)

PerfP
VP

Perf
perf

V
kiss
uinflection
Features contained on the perfect head are thought to undergo agreement with features in
the verb phrase. The effect of this agreement is that the verb phrase now contains a PF
interpretable feature which corresponds to the past participle.
(14)

PerfP
VP

Perf
perf

V
kiss
uinflection :perf
Only after the derivation is complete and the relevant chunk has been sent to PF will the
actual item kiss and the appropriate participle morphology be inserted into the syntactic
head resulting in kissed.

1.3.2

Distributed Morphology

In the next few paragraphs, I will provide a brief overview of Distributed Morphology (DM)
(Halle and Marantz, 1993; Embick and Noyer, 2007) drawing specifically on Embick and
Noyer (2007) and Siddiqi (2010).

11

DM is a framework within the MP which assumes that there is no separate morphological
component in the language faculty. Specifically, DM rejects the idea of a separate Lexicon.
In previous theories of morphology the lexicon was thought of as the ‘mental dictionary’ or
word storage component as well as the locus of many word-level or morphological operations
like suffixation, affixation, and even inflection. Under these theories the syntax manipulates
fully formed complex lexical items
In DM, morphological operations are assumed to be similar in form from syntactic operations like merge and agree. This “syntax all the way down” approach, insists that words and
any necessary derivational or inflectional morphology are composed by operations which occur after any syntactic operations are completed. As such the old conception of a lexeme can
no longer apply. The syntax must have access to and operate only on morphemes that are
abstract bundles of features. Morphemes manipulated by the syntax are thus divorced from
their phonological information. The phonological components of traditional morphemes are
thought of as Vocabulary Items (VIs) in DM and exist in a ‘storage only’ component of the
grammar referred to as the Vocabulary. Any real world knowledge that we have about what
specific VIs mean is stored in the Encyclopedia and is not available during the derivation.
Specifically morphemes are now only those syntactic units subject to syntactic operations
which are contained in the terminal nodes of a given syntactic tree. They are divisible into
abstract morphemes, which are bundles of syntactic and semantic features not including
phonetic features, and roots which correspond to morphemes which bear lexical content
√
and meaning such as whiskey, represented as whiskey. Roots are subject to categorizing
requirements which means that they must receive categorizing information from categorizing
functional heads like n to, in the case of whiskey, become a noun.
Abstract morphemes and roots are joined with their corresponding phonological features
post-syntactically in the PF process of vocabulary insertion. Vocabulary insertion proceeds
by way of feature matching constrained by the subset principle. A morpheme’s corresponding
phonology or VI is inserted if that VI contains all or a subset of the features specified on

12

the terminal node. A vocabulary item which has features not specified on a given terminal
node will not be inserted. Further, in cases of competition for insertion among phonological
exponents, the exponent with the greatest number of features matching those on the terminal
node will be selected for insertion.
To see how this all works, consider vocabulary insertion of inflectional morphology in the
following example adapted from Siddiqi (2010). Assume the syntactic node below.
(15)

XP
X



+present
 +singular 
+3rd person
For example this particular set of features is realized as -s in English on T after syntactic
agreement between the subject DP and T has occurred. This is the feature specification of
the verbal affix -s in English. Thus (16) is an example of a vocabulary item listed in the
Vocabulary of English. Barring other conditions on its insertion (omitted here for clarity) it
states that there is a vocabulary item -s that will be inserted onto nodes like the one above.
Crucially, this happens after syntactic spellout at PF with the phonology /z/.
(16)

1.4

[present, singular, 3rd]

↔

-s /-z/

Outline

This dissertation is structured as follows. Chapter 2 is an analysis of experience and causative
have + to constructions in Appalachian English. I propose, based on syntactic diagnostics,
that the causative and experiencer have in Standard and Appalachian English have an identical syntactic structure even though infinitival-to is present in Appalachian English. The
infinitival-to in this particular case in AppE is the result of a default insertion rule in the
dialect. Thus causative and experiencer have are both analyzed identically as restructuring

13

verbs and the discovery of the default insertion rule is evidence that some surface form variation is actually the product of post-syntactic processes. Moreover, such morpho-syntactic
variation must be accounted for in any comparative or micro-variationist syntactic analyses.
Chapter 3 examines the syntax of perception complements in Appalachian English. They also
exhibit variation in Appalachian English and are often considered close syntactic relatives of
causative structures. I propose that differences in the behavior of perception complements
in the dialect is due to a deeper syntactic difference in the grammar of Appalachian English.
Previous accounts of perception complements pin the seemingly odd behavior of infinitive
form in the active and passive on the behavior of the passive participle. They posit the
passive participle as an intervener between two heads in a probe/goal relationship. I suggest that Appalachian English does not have a passive participle which is distinct from past
participles and thus intervention effects do not arise. This is supported by evidence from
participle leveling. In chapter 4, I examine the behavior of infinitives in contractions via the
dialect form liketa, a verbal counterpart to almost. The examination of the form and its
analysis as a restructuring predicate allows us to tip the scales in the comparison of two theories of restructuring. Then in chapter 5, I examine the behavior of non-finite how come in
Appalachian English as compared to its finite counterpart in Standard English. This chapter
shifts the focus on restructuring from embedded complements to matrix complements. It
raises several questions about the intersection of restructuring, grammaticalization, and the
minimal requirements for features of root clauses.

14

CHAPTER 2
HAVE + TO CONSTRUCTIONS

2.1

Introduction

In Standard American English (SAE), the clausal complements of causative and experiencer
have appear as bare infinitives. This means that they appear with uninflected verbal forms
and without the standard infinitival-to which overtly marks the absence of any other tense or
person agreement. This is illustrated for causative have in example (1a) and for experiencer
have in example (1b).
(1) SAE
a. I had Mary wash the dishes.
b. They had glasses break on them.
Appalachian English (AppE) allows an infinitival-to in these same have constructions. Note
the acceptability of infinitival-to in Appalachian English for the causative in example (2a)
as well as the experiencer have in example (2b).
(2) AppE
a. I had Mary to wash the dishes.
b. They had glasses to break (on them).
Crucially, in SAE infinitival-to is illicit in both causative and experiencer have constructions.
A bare infinitive is required.
(3) SAE
a. * I had Mary to wash the dishes.
b. * They had glasses to break on them.
Interestingly, the presence of infinitival-to in the AppE cases does not alter the meaning.
For instance, the meaning of the causative have in Standard English and Appalachian English is the same. Consider causative have in examples (4a) and (4b) which both have the
15

interpretation in example (4c). Thus the bare-infinitive in Standard English is equivalent to
the infinitival-to form found in Appalachian English.
(4)

a. I had Mary wash the dishes.
b. I had Mary to wash the dishes.
c. I am the instigator or cause of Mary washing the dishes.

SAE
AppE

The experiencer have construction behaves in the same way, modulo the requirement of the
ethical dative on X in Standard English. Consider the situation below. In both dialects
experiencer have in examples (5a) and (5b) corresponds to the interpretation in (5c).
(5)

a. They had glasses break on them.
b. They had glasses to break (on them).
c. There was an event of glasses breaking and they experienced this event.

SAE
AppE

The central claim of this chapter is that the variation in distribution of the bare infinitive
in Standard English and infinitival-to in Appalachian English is best captured by assuming
an identical syntax for both causative and experiencer have in both dialects. This means
that the infinitival-to found in the AppE examples (2a) and (2b) is the result of a postsyntactic default insertion rule. Likewise, if the properties of these structures are the same
across dialects, then infinitival-to in AppE is indicative of a null syntactic head in Standard
English. Investigation into the nature of the syntactic head on which infinitival-to in these
particular have constructions reveals that it is not T but rather AspP. In simpler terms,
AppE has two infinitival-tos: The familiar one which is inserted on T in the tense layer of
the functional structure of a propositional clause and one which is inserted as the result of a
default insertion rule within the auxiliary layer of smaller than TP clauses, arguably AspP.
This chapter has the following structure. Section 2.2 is a general overview of what
is known about have + to constructions in Appalachian English and more generally for
Standard English. In this section, I review the properties of both causative and experiencer
have in SAE, as well as one general analysis of have in SAE. Crucially, I propose that we
think of causative and experiencer have as restructuring verbs. Then in section 2.3 I lay out

16

my analysis which argues that these two haves are located in v and select for complements
that are maximally headed by an AspP. I also argue that the to in the AppE is not where we
usually assume infinitival-to is. After I argue for a particular articulation of the clauses of
causative and experiencer have, I will lay out my assumptions about English verbal inflection,
and show how slight alterations to the account given in Ritter & Rosen (1997) allow us to
account for the variation that exists between Standard and Appalachian English with regards
to the presence or absence of to in the complements of these verbs. Section 2.4 concludes
the chapter.

2.2

Background

This section provides a context for the analysis presented in this chapter. First, I touch on
a few specific references to the relevant cases of have + to constructions, before moving on
to a non-dialect specific review of the syntax of causative and experiencer have.

2.2.1

Have + to

Precedent for claims that Appalachian English has extended the use of to beyond that of
Standard American English can be found in Wolfram and Christian (1976) who note that it
is quite common for verbs like go, start, and get to appear with to and be suffixed with -ing.
(6)

a. He went to runnin’ after her.
b. She started to eatin’.
c. I got to readin’.

In a later communication Wolfram notes the following.
(7) “One thing to be kept in mind about to in [West Virginia] dialect: there seems to be
a general extension of the form beyond its limits in other varieties. For example, I
think some speakers are inserting it as a kind of filler functioning something like the
a-prefix (e.g. I just heard a tape with . . . you go to checking... ). ”
(Butters and Stettler, 1986:p.c. Walt Wolfram)

17

Anecdotally, it seems that AppE makes use of to + verb + ing construction more than
Standard varieties. These cases must be set aside because they are prepositional and not
instances of infinitival-to. For instance, a great deal of prepositions can be modified by right.
(8)

a. They set the whiskey shots right in front of us.
b. I stood the hoe right inside the door.
c. He ran right to the edge.

Like other prepositions the ones in question repeated in example (9) are acceptable when
modified by right. Wolfram’s examples of the gratuitous to in AppE are therefore prepositional.
(9)

a. He went right to runnin’ after her.
b. She started right to eatin’.
c. I got right to readin’.

Crucially though, note that example (10) shows that infinitival-to can not be modified by
right.
(10)

a.
b.
c.

* I wanted right to run.
* She remembered right to drink the bourbon.
* They tried right to fix the problem.

Interestingly, right modification is unacceptable in both experiencer have and causative have
constructions in Appalachian English. This indicates that these versions of infinitival-to are
not prepositional, but not that they are themselves instances of the presence of a Tense layer
in a clause.
(11)

a.
b.

* They had glasses right to break on them.
* John had Mary right to wash the dishes.

Have + to causative and experiencer forms are explicitly mentioned in the literature as
a marker of Southern speech. The form is attested for the entirety of central and southern
Appalachian states including: West Virginia, Kentucky, Virginia, Tennessee, North Carolina, and Georgia. Beyond this, Butters and Stettler (1986) mention the prevalence of the
18

causative and experiencer have+to construction across the Southern United States and note
that it is so prevalent that it in fact may be used to identify Southern United States English
speakers. Anecdotally, I have never heard have + to structures used beyond the confines
of central Appalachia. They speculate that the use of the to-infinitive complement might
be subject to syntactic constraints (Butters and Stettler, 1986:pg. 189). As we will see in
this dissertation, these data support two different syntactic distributions for infinitival-to in
AppE.

2.2.2

General properties of causative and experiencer have

Zooming out from the specific context of Appalachian and Southern Englishes, it is necessary
to review some general properties of causative and experiencer have that have been observed
for SAE in the literature. Causative and experiencer have share a core set of properties that
distinguish them from other verbs which take bare infinitive complements. Aside from this,
experiencer have has an extra syntactic requirement where its complement must contain an
anaphoric referential element, the commonly termed ethical dative on X.

2.2.2.1

Causative have

Since causative have and other causative verbs often appear with bare infinitive complements
in English, they are often described along with other verbs which take bare infinitive complements, like the perception verbs see, hear, and feel as in example (12). Thus causatives
are often described in terms of how they differ from perception complements.
(12)

a. John had Mary wash the dishes.
b. John saw Mary wash the dishes.

Mittwoch (1990) notes that the complement of periphrastic causatives can not be independently referenced by a pronoun in example (13a), whereas the complement of perception
verbs can be referenced by a pronoun in example (13b).

19

(13)

a.
b.

* John had Mary wash the dishes, and Bill had it too.
John saw Mary wash the dishes, and Bill saw it too.

Many authors point to this data as evidence that the event composition of causatives and
perception complements differs in terms of how many events they actually denote. Because
of data like that in example (13a) where Mary wash the dishes is not a possible referent
of the pronoun it, they argue that causatives are thought to denote single complex events.
Thus causatives may be thought of as an event e which both precedes a second dependent
event e and acts as the cause or instigation point of e , rather than two independent events
which happen to overlap temporally but are not otherwise related.
Perception complements on the other hand are thought to have independent reference and
denote two separable eventualities which overlap temporally but are otherwise independent.
They are commonly assumed to be composed of a perceiving event and an event perceived
which are required to overlap temporally. In simpler terms, this reflects an intuitive difference
between perception and causation complements. Events denoted by perception complements
are not causally related to their matrix perception events, while the events expressed in
causation complements are in essence dependent on the truth of their instigating event.
This is illustrated in the following example where perception complements are semantically
infelicitous (marked with #) as the continuation of a because phrase where causal relations
are made explicit.
(14)

a.
Mary washed the dishes because John had her wash the dishes.
b.
Mary washed the dishes because John made her wash the dishes.
c. # Mary washed the dishes because John saw her wash the dishes.

Without getting further into the semantics of causatives, I will assume that this is essentially
correct and that causatives are best represented as single complex events while perception
complements denote two events. We will return to perception complements in the next
chapter.

20

Following Felser (1999), causative have may not take a derived event nominal as an object
though perception verbs may. We see in example (15a) that causative have allows verbal
predicates in its complements while example (15b) shows that derived event nominals are not
licensed. Example (15c) shows that the perception verb see does license derived event nominals. Crucially, example (15b) can not have the causative interpretation of example (15a).
(15)

a.
b.
c.

We had Mary depart.
* We had Mary’s departure.
We saw Mary’s departure.

This shows us that causative have has strict selectional requirements and it can not select a
nominal object as its complement.
Causative have has certain semantic restrictions on its predicate as well. Ritter and
Rosen (1993) note that although causative make in example (16a) allows individual level
predicate complements, causative have in example (16b) does not. This distinction has
to do with whether or not the predicate is individual-level and would hold generally for
that individual or if that predicate is stage-level and holds only for some temporal stage.
Following Kratzer (1995), they assume that this distinction is due to the fact that individuallevel predicates denote properties while stage-level predicates denote events and that this
distinction is projected into the syntax from the level of argument structure. Ritter and
Rosen note that, although these facts may be further altered by the aspectual composition
of the clause, the basic distinction suffices for their purposes. Causative have only selects
stage-level predicates.
(16)

a.
b.

John made Bill like French cooking.
* John had Bill like French cooking.

Similar to Ritter and Rosen, Felser (1999) also invokes Kratzer (1995) and argues that
the restriction on individual-level predicates in closely related perception complements is
a result of the fact that individual-level predicates fail to project an event argument. She
argues that individual-level interpretation requires a clause minimally headed by TP. This

21

tracks proposals in Diesing (1992a;b) who argues that subjects of individual-level predicates
are restricted to appearing in the specifier of TP at the level of logical form, while subjects of
stage-level predicates may either appear in the specifier of vP or TP. For Felser, perception
complements are maximally headed by an AspP and not a TP and so the constraint against
individual level predicates falls out from a difference in clause structure. More recently, the
notion that eventivity is associated with structural configurations of AspP and vP is echoed
in Embick (2004) where he accounts for the differences between verbal and adjectival passives
in terms of the merge site of Asp0 : either below or above v. In his account, AspP is the
locus of participle morphology, while the eventivity of the construction is determined by v.
Though I believe a satisfactorily thorough explanation of the fact in (16) is beyond the scope
of this dissertation and possibly unrelated to an explanation of the behavior of these verbs
with regards to dialectal variation, I will follow the general trend in the literature and assume
that the ban on individual level predicates in perception complements and some periphrastic
causatives is a result of the internal aspectual composition of the complement clauses with
regards to the properties of v and AspP. I leave the specifics of this configuration to further
research.
Felser (1999) also reports that causative have licenses what she calls an aspectual be in
example (17a). Based on the remainder of her observations and arguments aspectual be is, as
far as I can tell, supposed to be the auxiliary associated with progressive aspect in English.
(17)

a.
b.
c.

? Mary had the frogs be entertaining the dwarfs.
Mary had the frogs entertaining the dwarfs.
Mary saw the frogs entertaining the dwarfs.

I have found however that a majority of informants find example (17a) to be unacceptable.
This is especially true in comparison to the same sentence without be in example (17b). I
have labeled the data in example (17a) with the symbol ? to show decreased acceptability
by my informants. That said, several informants noted that inclusion of this be is alleviated
by adverbial modification as in example (18).

22

(18) During the play, Mary had the frogs be entertaining the dwarfs.
To the extent that sentences like example (17a) and example (18) are acceptable they are
only acceptable on an act be interpretation. Following Schmitt (1996), this be is not
the progressive auxiliary but rather a dummy verbal element which must be licensed by a
temporal operator because it has no intrinsic temporal specification. Thus to the extent
that these are acceptable they must always occur in the scope of a temporal operator.
Note that the alleviation fact is left unexplained if we assume that the be in example (17a)
and example (18) is the progressive aspectual be rather than act be. This means that what
Felser has really shown is that complements of causative have allow progressive -ing and to
some extent act be, but not the progressive aspectual be.
Aside from these c-selectional and semantic restrictions on the predicate, Ritter and
Rosen (1993) also note that causative have can not take expletives as the subject of its
complement clause. The same is not true for causative make. Again, some speakers of
Standard American English find expletive subjects acceptable in the complement of causative
have. They are labeled with ? to reflect degraded acceptability.
(19)

a.
b.
c.
d.

John
? John
John
? John

made it seem to be a problem.
had it seem to be a problem.
made there be computers available for the students.
had there be computers available for the students.

I would like to add that expletive it in commonly termed weather-it constructions also seems
to be marginally acceptable.
(20)

? Do you seriously believe God had it rain today?

Beyond these thematic issues, the actual argument structure of causative have also differs
from other verbs. In example (21a), we see that causative have does not allow matrix
passivization, while example (21b) shows that passivization of the embedded complement is
possible.

23

(21)

a.
b.

* John was had to wash the dishes.
John had the dishes washed.

Note that the behavior of have in these cases differs from the behavior of other periphrastic
causatives which undergo matrix passivization.
(22)

a. John was made to wash the dishes.
b. John was let go by the police.
c. John was helped to wash the dishes.

In minimalist terms, I take the fact in example (21) to reflect an underlying licensing selectional restriction which I will mention briefly here and revisit later in the analysis. I
will argue that example (21a) shows that the syntactic head which is responsible for passivization never selects for the syntactic head which contains causative have. As we see in
example (21b) however, it must be the case that causative have or some head embedded
under it licenses the head which is responsible for passivization.
Finally, example (23b) shows that the complement of causative have does not license
PRO.
(23)

a.
b.

John had himself wash the dishes.
* John had PRO wash the dishes.

In summary, the evidence that we have seen so far indicates that causative have constructions denote one complex event and not two separate eventualities. Causative have
only selects for stage-level predicates and never nominal objects. In terms of auxiliaries,
causative have never allows aspectual auxiliaries in its complements and only marginally
allows act be. As for subjects, some speakers also allow expletive subjects in causative
have complements but PRO is never licensed.

2.2.2.2

Experiencer have

Experiencer have shares the majority of its syntactic properties with causative have. However, experiencer have forms require that the matrix subject of have be associated with some
24

anaphoric element in their complement. This does not change the basic nature of the complement, rather it is an extra restriction on constructing the meaning of experiencer have.
Based on the shared core set of syntactic properties, I argue that causative and experiencer
have should be treated as a structurally identical pair even though experiencer have is subject to this extra restriction. I briefly review the core set of syntactic properties shared with
causative have before moving on to examine the specific restrictions on anaphoric elements
in the complements of experiencer have.
Experiencer have constructions also seem to denote one complex eventuality in example (24). This reflects the fact that the complement is not independently referenceable by a
pronominal form; unlike other event denoting complements.
(24)

* John had a pet die on him and Bill had it too.

Like causative have, experiencer have disallows derived event nominal complements, example (25a). Events described in the complement of experiencer have can typically be
thought of as results, example (25b), they are also typically stage-level, example (25c).
(25)

a.
b.
c.

* John had Mary’s departure.
John had his wine sour on him.
* John had Mary like French Cooking on him.

The presence of aspectual be seems to be banned in experiencer have, shown in example (26).
Informants who liked aspectual be in causative have report that be is banned in experiencer
constructions.
(26)

* John had the frogs be entertaining the dwarfs on him.

Similarly, expletives which were variably acceptable in causative have constructions are
banned in experiencer have constructions, example (27).
(27)

* John had it seem to be a lie on him.

However, it seems like weather expletives as in We had it rain on us are acceptable. Beyond these restrictions, the complements of experiencer have may be passivized but the
25

constructions can not undergo matrix passivization, shown in examples (28a) and (28b).
(28)

a.
b.

John had his room dirtied on him.
* His room was had dirtied on him.

Finally, PRO subjects are banned in experiencer have complements like the one in (29)
just as they are in causative have complements.
(29)

* John had PRO wash the dishes on him.

One big difference between causative and experiencer have for Standard English speakers
is that the experiencer interpretation of have is aided a great deal by the presence of the
ethical dative “on X” (Harley, 1997). This fact does not hold for Appalachian English and in
fact Harley shows that this fact is not always true of Standard English either as the example
in 30 shows. What does seem to be required however is that some pronominal element which
is anaphoric to the matrix subject.
(30)

a. Luke had a storm trooper blow up his x-wing fighter.
b. Han had Leia kick him (on the shin).
c. Yoda had Luke make a mess in his hut.

She maintains that given a sufficiently strong context this pronominal element may be
excluded entirely.
(31)

a. Poor Mr. Chips had students walk out of class.
b. Do you think we’ll have that loudmouth Rush drop in tonight?

Harley argues that this anaphoric relationship must be syntactic because it is not the case
that any relationship will do. For instance, in example (31a) she argues that the experiencer
have reading is only available if we take the class in question to be Mr. Chip’s class.
Similarly, example (31b) is only felicitous with a experiencer have interpretation if we take
it to be us that Rush might drop in on.
Harley further rules out a semantic/pragmatic account of the anaphoric requirements
on experiencer have by showing that SELF anaphors (Reinhart and Reuland, 1993) are

26

sufficiently discourse salient but yet are not sufficient to license an experiencer have interpretation. She highlights the following asymmetry in experiencer have complements.
(32)

a.
b.

Asterix had goat’s milk spilled on him.
Asterix had goat’s milk spilled on himself.

*experiencer

In example (32a) both an experiencer and a causative reading are available, while the reflexive
pronoun in example (32b) blocks the experiencer reading. Again, a semantic/pragmatic
context analysis leaves this distinction unexplained because the relationship in question
between the reflexive and the matrix subjects is sufficiently salient and yet the experiencer
reading is unavailable. Harley argues that anaphoric pronominal elements that are referential
are always required and that SELF anaphors are logophoric rather than referential. I will
assume the general distinctions made in Harley (1997) and in the next section I review an
analysis which argues a single treatment for all versions of have.
The findings so far about SAE causative and experiencer have are summarized below.
(33)
independent reference
derived event nominals
only stage-level
aspectual be
embedded expletives
passivized complements
matrix passivization
PRO embedded subj

2.2.3

causative

experiencer

✗
✗
✓
✗
?
✓
✗
✗

✗
✗
✓
✗
✓
✓
✗
✗

Two haves, one analysis

In this section, I discuss an analysis of causative have which fits well with the properties
that were just summarized before moving on to my own analysis of causative and experiencer
have. Ritter & Rosen (1997) provide a uniform analysis not only of causative and experiencer
have but of other uses of have as well. This stems from a thread in the wider literature on
have and be where the two are thought to be share a common derivation (see Freeze, 1992;
Kayne, 1993). Ritter & Rosen argue however, that have is not reducible to be and that the
27

two do not share a common derivation. On the other hand they argue that all instances of
have do share a common derivation. This means that have is lexically different from the
copula. This split between have and be is not as drastic as it might seem at first. They posit
a syntax which is similar to structures proposed in earlier work. Since I argue for an identical
syntax of at least two haves, I will review their analysis in some detail. They propose that
different senses of have below are all derived from an identical syntax.
(34)

a.
b.
c.
d.
e.
f.

John
John
John
John
John
John

had the students write three letters.
had his car stolen
had a party.
has a hat on today.
has a sister.
has read the New York Times.

causative
experiencer
nominal event
locational
possessive
auxiliary

Though the analysis is proposed for all meanings of have, I will focus on the analysis of
the relevant experiencer and causative have. They propose that have is synthetic and is a
complex predicate composed of more than one syntactic head. Ritter & Rosen argue for the
structure example (35) where have is both heads F1 and F2.
(35)

F1P
YP

F1
F2P

F1
±instigator
ZP

F2
F2
±delimit

...

They argue that these two heads are specified for the features ±instigator and ±delimiter
respectively. These featural specifications determine what semantic or thematic roles will
be assigned to the elements in their respective specifiers. For instance, +instigator is associated with a causative semantic role which is assigned to a DP in its specifier. Conversely
−instigator might be associated with a experiencer role in the case of a non-causation predicate. The lower projection provides the measure or delimitation of the embedded eventuality.
28

Ritter & Rosen argue that these features and corresponding semantic relationships are determined by properties of an embedded VP small clause. Consider their derivation of causative
have in John had Mary wash the dishes.
(36)

F1P
John

F1
F2P

F1
Maryi

F2
VP

F2

ti wash the dishes.
Assuming a uniform syntax, experiencer have is generated by an identical structure whose
feature content is determined by its embedded complement. Like Harley (1997), they argue
the difference between causative and experiencer have is that the complement of experiencer
have must contain an element that is anaphoric to the matrix argument. Specifically for
Ritter & Rosen it must be anaphoric to the argument merged into the specifier of F1.
Consider experiencer have in John had students walk out of his class below.
(37)

F1P
Johnk

F1
F2P

F1
Studentsi

F2
VP

F2

ti walk out of of [hisk class]
Though I will argue for an identical syntax for experiencer and causative have, I do not
assume that their syntax can be extended to other instances of have for reasons which will
be made apparent below. I will however follow Ritter & Rosen (1997) and Harley (1997) in

29

assuming that causative and experiencer have differ minimally in the type of thematic role
that they assign to the subjects in their specifiers and that this is reflected in the feature
content of that head. Under the restructuring analysis that I will propose there is no reason
to assume that have is synthetic. In fact, the shared behaviors that causative and experiencer
have exhibit indicate quite explicitly that a uniform analysis of have as proposed by (Ritter &
Rosen, 1997) is not entirely possible, a conclusion that Ritter & Rosen discuss in some detail.
Consider the case of auxiliary or perfect have and T to C movement in example (38). In
order make a yes/no question out of the sentence in example (38a) perfect have obligatorily
undergoes T to C movement as in example (38b), when no other auxiliaries preceede it in a
clause.
(38)

a. John has drunk all my whiskey.
b. Hasi John ti drunk all my whiskey?

Neither causative or experiencer have undergo T to C movement in yes/no questions. This is
shown in examples (39a) to (39d). Examples (39e) and (39f) show us that forming a yes/no
question with causative and experiencer have is done with do-support.
(39)

a.
b.
c.
d.
e.
f.

John had Mary wash the dishes.
John had glasses break on him.
* Had John Mary wash the dishes?
* Had John glasses break on him?
Did John have Mary wash the dishes?
Did John have glasses break on him?

This reflects a fundamental structural difference between auxiliary have on one hand and
causative and experiencer have on the other. That is, auxiliary have eventually ends up in
T and is a candidate for T to C movement while causative and experiencer have do not.
Further, Ritter & Rosen note that auxiliary have does not add an extra argument, fails to
assign case, contracts with a subject, and takes negation. In the next section, I argue for
the placement of causative and experiencer have in the larger class of restructuring verbs.

30

2.2.4

Causative and experiencer have as restructuring verbs

So far, we have covered what little there is to say about have + to constructions in Appalachian English, the general properties of both forms which just so happen to be the same
across dialects, and one analysis which attempts to provide a unified treatment. The final
question to be answered before I begin my analysis of causative and experiencer have is how
these two verbs and their properties square with a larger class of verbs. Where do they fit
in with respect to other types of verbs?
Causative and experiencer have are restructuring verbs. That is to say, being restructuring verbs the infinitives that they select are transparent for particular syntactic operations that are otherwise thought to be subject to locality constraints or clause boundedness
(Wurmbrand, 2001). One classic example of inter-clausal transparency effects is commonly
referred to as auxiliary switch. Consider the following examples taken from Roberts (1997),
in Italian the choice of auxiliary is determined locally. The transitive non-restructuring use
of the verb voluto ‘want’ requires the auxiliary avere ‘have’ as shown in example (40).
(40) Piero ha/*`e voluto questo libro.
Piero has/is wanted this book
‘Pierre wanted this book.’
However, Roberts notes that this restriction disappears when the complement of voluto
contains an infinitival unaccusative verb like venire ‘to come’ which usually takes essere ‘to
be’ rather than avere.
(41) Piero ha/`e voluto venire con noi.
piero has/is wanted to-come with us
‘Pierre wanted to come with us.’
Thus Roberts claims that in the infinitival construction there is a restructuring effect when
the embedded infinitive determines the auxiliary in the matrix clause. In simpler terms,
when the `e of essere appears as the auxiliary of the verb voluto in example (41) it appears
as though the clause boundary has become transparent and the auxiliary type is no longer
determined by the matrix verb. Transparency effects are not limited to instances of auxiliary
31

switch but cover a range of phenomena in the Romance and Germanic language families
much to broad to cover explicitly here. In addition to auxiliary switch, transparency effects
in Romance languages includes clitic climbing and object preposing. In Germanic languages
transparency effects include long distance scrambling, the long passive, and verb raising.
In the rest of this study and throughout the rest of this work, I will assume an analysis
along the lines of Wurmbrand (2001) to account for similar transparency effects in causative
and experiencer have. Her overarching argument concerns the complement structure of verbs
in general but is formulated specifically to deal with restructuring verbs. She argues that infinitival complements are not uniform in their clause structure; some infinitival complements
are even as small as VPs. Specifically, she argues that infinitival complements like those in
example (42a) are ‘true’ lexical restructuring configurations which have the syntax in (42b).
(42)

a. weil Hans den Traktor
zu reparieren versuchte.
since John the tractor-ACC to repair
tried
‘since John tried to repair the tractor’
b.
TP
John

T
T

vP
tsubj

v
v

VP
V
tried

VP
obj
V
the tractor to repair

Thus versuchen ‘to try’ is considered a restructuring verb. It is a lexical verb which selects
another lexical VP as its complement. The object of the embedded verb will always receive
accusative case from the ‘matrix’ v, while the subject is generated normally in spec,vP where
it is assigned a theta-role and will be interpreted as the agent of both verbs by virtue of the
close relationship between the two VPs. The subject subsequently moves to spec,TP.

32

She argues that this syntactic analysis accounts for a specific set of properties and that the
proposed clause structure leads to semantic and syntactic differences in the behavior of these
verbs and their complements. For instance, lexical restructuring predicates do not license
embedded PRO subjects or assign structural case. They do not exhibit tense or sentential
negation. They are the counterparts of what are traditionally called obligatory control complements and they frequently occur with strong implicatives, aspectuals, irrealis predicates.
They allow long object movement (matrix passivization), scrambling, and pronoun fronting.
They prohibit extraposition of the infinitive and relative clause pied-piping.
Wurmbrand notes that while this list of properties may serve as a guide to the classification of restructuring complements, not all languages can be expected to have exactly
this configuration of properties or prohibitions associated with restructuring. Her analysis
of lexical restructuring complements is specifically built around German.
Further, she notes that although there is some variation in which specific verbs exhibit
restructuring properties across languages, there is a general core set which seems to be
repeated.

(43)

Verb type

example verbs

Modal
Motion
Aspectual
Causative

must, may, can, want
come, go, return
begin, continue, finish
let, make

Wurmbrand argues that restructuring predicates are determined both by their semantic or
thematic properties which may delineate them as a coherent group and by lexical properties
assigned to specific sets of infinitives in certain languages. For example, the modal verbs
must, may and can fall into the class of functional restructuring predicates. This means
that they are generated only in the functional structure of the clause; in the auxiliary space
under tense and above the lexical structure of the verb phrase. As such, they do not assign
thematic roles. This particular configuration of syntax and semantics leads to the properties
that we generally associate with modal auxiliaries.
33

For Wurmbrand, restructuring predicates are essentially mono-clausal, even though they
contain two verbs. This assumption separates the complements of non-restructuring predicates which are clausal and propositional (CP/TP) from those of restructuring predicates
which are generated as complements smaller than those headed by tense. This assumption is
less than surprising for modal auxiliaries because we do not commonly assume that clauses
which contain modal auxiliaries are bi-clausal. However, this is a somewhat surprising claim
to be made about main verbs. Main or lexical verbs which exhibit matrix passivization
among other things, fall under the class of lexical restructuring predicates generated in
V. Like other verbs, lexical restructuring predicates have normal theta-assigning properties
but select VPs as complements. Thus lexical restructuring predicates are the true set of
restructuring predicates in Wurmbrand’s account. In general, Wurmbrand’s account and
classification of restructuring verbs is a pertinent guide for the investigation of misbehaving
clauses.
She notes several advantages to a mono-clausal approach to restructuring infinitives. The
first advantage is that no mechanism of restructuring is required. This contrasts with biclausal approaches which require (i) some way of reducing the infinitival clause through a
structure altering operation, or (ii) a movement operation which allows syntactic objects to
evade the clausal boundaries. It would seem that any approach which assumes uniformity of
clause structure across infinitival complements requires a mechanism to alter the structure
of those predicates which exhibit restructuring effects. Beyond this, Wurmbrand notes that
under a mono-clausal approach to restructuring no specific assumptions need to be made to
account for transparency effects. For example, no mechanisms need posited to account for
the ban on infinitival-to in causative and experiencer have in SAE if we just assume that
the clause is smaller than T. Basically, under a mono-clausal approach the boundaries which
are thought to prevent inter-clausal syntactic operations simply don’t exist. Finally, the biclausal approaches may appeal in terms of uniformity of all complement structure but they
are burdened with providing evidence for the initial fully clausal state of the complement of

34

the restructuring predicate.
With regards to what motivates restructuring, proponents of the mono-clausal approach
such as Cinque (2004) or Wurmbrand (2001) assume that restructuring verbs are more like
auxiliaries than main verbs (for an extensive list see Wurmbrand, 2001:pg. 11). They
are sometimes thought to have less thematic content than main verbs or assumed to be
less like lexical verbs and share more properties with auxiliaries and functional structure.
Wurmbrand notes that proponents of the bi-clausal approach must frame the question in
a different way. They must ask why restructuring mechanisms apply to only a subset of
infinitival constructions.
In my analysis, I will show that causative and experiencer have are best analyzed as
what Wurmbrand calls semi-functional restructuring predicates. Wurmbrand argues that
semi-functional restructuring predicates are syntactically functional but thematically lexical.
Generally, they are bare events or actions which do not license PRO and whose complements
lack any manifestations of tense, sentential negation, or complementizers. While the semifunctional restructuring verb itself is able to assign thematic roles to its arguments they
generally do not allow matrix passivization. In short, they are projections of a matrix vP
which sits atop some FP. We have already seen evidence that causative and experiencer
have assign a cause and experiencer role to their external arguments (Ritter & Rosen,
1997; Harley, 1997). I present further evidence that causative and experiencer have assign
thematic roles to their subjects and argue that the verbs themselves are inserted into vP and
show that their complements are maximally headed by an aspectual auxiliary phrase since
their complements lack properties associated with T.

2.3

Analysis

With the basic properties of causative and experiencer have in mind we can more forward
with the analysis of the variation in the bare infinitive found between Appalachian English
and Standard American English. I will assume for the time being that causative and expe35

riencer have fit with Wurmbrand’s analysis of semi-functional restructuring predicates and I
will show how this classification falls out from my analysis. In the introduction to the chapter, I suggested that the structures of the Appalachian English versions of causative and
experiencer have in examples (44a) and (45a), are syntactically identical to each other and
to their bare infinitive Standard American English counterparts in examples (44b) and (45b).
(44) experiencer have
a. They had glasses to break (on them).
b. They had glasses break on them.

AppE
SAE

(45) causative have
a. I had Mary to wash the dishes.
b. I had Mary wash the dishes.

AppE
SAE

Under the assumption that to in the AppE cases is a vocabulary item with a corresponding
insertion site in the syntactic derivation, comparison of the data across dialects reveals a
syntactic head position above the embedded verb into which infinitival-to is inserted. I
assume that this structure is phonologically null in SAE. Consider this shown with causative
have in examples (46a) and (46b).
(46)

a. I had Mary to wash the dishes.
b. I had Mary ∅ wash the dishes.

AppE
SAE

In what follows, I will argue that the syntactic position of to and ∅ is an aspectual head
and that it is the maximal projection of the embedded complement. I will argue that the
underlying structure of causative and experiencer have across dialects is shown below in the
phrase John had Mary to/∅ wash dishes.

36

(47)

vP
John

v
v
had

AspP
Maryi

Asp
vP

Asp
to/∅
ti

v
v
washk

VP
tk dishes

I will show that have is generated in v and selects for complements headed by AspP. This in
turn means that the infinitival-to which appears in the Appalachian examples (44a) and (45a)
and is present in Asp0 of example (47) is not the familiar infinitival-to which is associated
with T0 in either dialect. This infinitival-to is the result of the lexical insertion rule which
is present in the lexicon of AppE speakers, but not in the lexicon of SAE speakers.
In the rest of this section, I will provide evidence which shows that have is generated in
v, and assigns a theta-role to its specifier. Then I will argue for a complement maximally
headed by an aspect phrase. Next, I will say a bit about the nature of the covert aspectual
head and argue that in general aspectual heads in English only exhibit auxiliary insertion
in the presence of morphology assigned by a higher head. This suggests a separation in the
morphological licensing features of a given aspectual head and its candidacy for vocabulary
insertion in a string of auxiliaries. Finally, I will briefly present a cross linguistic instance
of a causative restructuring verb in Salentino Italian which appears with overt infinitival
morphology in an almost identical situation. This provides precedent for asserting a default
insertion rule for infinitival-to in these restructuring contexts.

37

2.3.1

Have is in v

The position of causative and experiencer have can be inferred by examining their distribution relative to fully articulated auxiliary structure, the lack of obligatory T to C movement
in question formation, the lack of matrix passivization, and a ban on expletive subjects in
the matrix clause.
First, I assume the following basic structure of the English clause.
(48)

TP
T
Aux1P

T
nom
tense

Aux2P

Aux1
perf

vP

Aux2
prog
subj

v
v 
θ
 voice 
acc


VP
V
verb
θ

DP
D
object

Following common assumptions of the Minimalist Program (Chomsky, 1995), I assume that
the verb assigns a θ-role or thematic-role to the object or internal argument in its complement. Similarly, θ-roles generally associated with agentive or active voice subjects are
assigned by v via a spec-head relationship. Thus external arguments are merged in spec,vP.
Common assumptions about case-assignment tell us that v is responsible for assigning accusative case to DPs in its domain. I will also assume that nominative case is similarly
assigned by T0 .
A fully articulated string of the modal/aux domain is found in example (49a) with the
structure in example (49b).

38

(49)

a. John might have been stealing my whiskey.
b.

TP
Johnj

T
T
might

perfP
progP

perf
have
prog
been

vP
tj

v
v
stealingk

VP
V
ik

DP
my whiskey

It stands to reason that any lexical item that appears below a fully articulated set of
auxiliaries in English must be in a syntactic position below the lowest auxiliary, namely it
must be in vP or lower. Causative and experiencer have appear below a full set of auxiliaries.
Again, since I am arguing for an identical syntax across dialects, I will include both the AppE
to and its SAE counterpart ∅ in the derivations. Thus to/∅ represents the same head in
both dialects.
(50)

a. John might [perfP have [progP been [vP stealing my whisky.
b. John might [perfP have [progP been [vP having Mary to/∅ wash dishes.

Notice that have in example (50b) appears to be in the same position as the verb stealing
in the previous derivation. They even both express the progressive -ing morphology of the
preceding progressive head. I will argue below that causative have in example (50b) is a
semi-functional verb base generated in v, while stealing is a lexical verb generated in V,
eventually undergoing movement to v.
Another piece of evidence that have is not in an auxiliary position comes from the fact
that both causative and experiencer have require do-support for yes/no question formation

39

and do not move from T to C themselves. Many types of non-subject questions trigger T
to C movement in English and as we previously saw for yes/no questions in example (39)
repeated below in example (51), causative and experiencer have do not undergo T to C
movement but they require do-support.
(51)

a.
b.
c.
d.
e.
f.

John had Mary wash the dishes.
John had glasses break on him.
* Had John Mary wash the dishes?
* Had John glasses break on him?
Did John have Mary wash the dishes?
Did John have glasses break on him?

This shows that causative and experiencer have do not behave like auxiliaries in English. If
they were auxiliaries generated above vP in some aspectual head then we would expect them
to pattern with the aspectual auxiliaries have and be which, as we can see from example (52),
obligatorily undergo T to C movement.
(52)

a.
b.
c.
d.

John has drank all my whiskey.
Hasi John ti drank all my whiskey?
John is drinking all my whiskey.
Isi John ti drinking all my whiskey?

Beyond distributional evidence, there are several theta-role related factors which bear
on the distribution of causative and experiencer have with v. Remember that according to
the clause structure of English assumed here, external arguments are assigned theta-roles
by v. Traditional raising verbs like seem are not thought to assign theta-roles to external
arguments and this is reflected in the fact that they license expletive subjects like there and
it in example (53). The exact mechanics of where and how these dummy subjects enter the
derivation may be set aside for our purposes here. Suffice it to say that such subjects are
barred from appearing with verbs which assign theta-roles.
(53)

a.
b.

There seemed to be someone washing the dishes.
It seemed that Mary was washing the dishes.

40

Like traditional raising verbs, auxiliaries in English also do not assign theta-roles. This is
evident in that they impose no restrictions on the licensing of expletives in raising predicates.
(54)

a.
b.

There had seemed to be someone washing the dishes.
It was seeming more and more like Mary was washing the dishes.

It would follow that if causative and experiencer have are more auxiliary-like then they would
also fail to impose restrictions on the presence of expletive subjects. However, causative and
experiencer have do not allow expletive subjects presumably because they must assign agent
and experiencer theta-roles.
(55)

a.
b.
c.

John had Mary to/∅ seem to be washing the dishes.
* There had Mary to/∅ seem to be washing the dishes.
* It had Mary to/∅ seem to be washing the dishes.

Finally, I would like to suggest following Wurmbrand (2001) that causative and experiencer have in example (56) do not allow matrix passivization because they are in competition
with passive v.
(56)

a.
b.

* Mary was had to/∅ wash dishes.
* The dishes were had to/∅ break on themselves.

Assuming the passive and unaccusative locus of the verbal domain is v means that passivization and merger of either experiencer or causative have are in direct competition for
merger.
(58) passive

(57) causative/experiencer have

vP

vP

v

v

v
passive

v
have

Certainly, they are opposites in terms of what effect they have on argument structure.
Passivization arguably reduces the number of arguments of a predicate while causative and
41

experiencer have add an extra argument to an already saturated predicate. Another possible
explanation can be gleaned directly from this fact. Simply put, the passive vP is not a semifunctional restructuring predicate. Rather it is functional structure which either suppresses
or simply does not assign a theta-role and it is always associated with lexical verb. There
is also no evidence to suggest that passive vP in the verbal domain ever selects anything
other than a VP in English. In passivized English sentences passive -en morphology always
appears on the verb, presumably licensed by v.
(59) John’s whiskeyi [perfP had [progP been [passP being [vP:passive stolen ti .]]]]
This means that passive vP will never select a vP complement containing causative or experiencer have as in example (60).
(60)

*

vP
v
v
passive

vP

...
v
have
Yet another possibility is that causative and experiencer have are actually lexical verbs
in V0 . On purely syntactic grounds, this possibility resurrects the question of why these
verbs do not passivize. Setting semantic considerations aside, if causative and experiencer
have were fully lexical verbs then one could argue that they should passivize but they don’t
for some non-syntactic reason. The fact that they do not would have to be argued on other
grounds. Assuming that causative and experiencer have are in V only pushes the problem
back a step. Under the analysis forwarded here, the v of causative and experiencer have is
in competition with the passive and the passive can not select have because the passive only
ever selects for a VP. If have is in V then we lose Wurmbrand’s competition generalization
and require stipulation as to why in this very specific case, the passive v can not select this

42

small set of lexical verbs.
Lastly passivization facts show us that although matrix passivization is disallowed, the
embedded passivization in example (61) is acceptable and we should therefore minimally
posit vP layer in the complement clause.
(61)

John had the dishes washed.

In summary, we can know the syntactic position of causative and experiencer have based
on their position with respect to the English auxiliary system. More specific evidence for
the position of these verbs comes from common assumptions about thematic role assignment
in English. Causative and experiencer have behave like lexical versions of v which assign
causative and experiencer theta-roles to their specifiers. We assume they are in v as opposed
to V on theoretical grounds. Assuming that have is lexical and in V deprives us of the
competition account of the ban on matrix passivization and requires unnecessary stipulation
to account for the facts. This again is evidence that they occur below the auxiliary system
and above the lexical verb phrase.

2.3.2

Have selects an AspP complement

While the types of morphology that appear in complements of causative and experiencer
have are few, this constraint is itself informative. Part of what it means for have to be a
semi-functional restructuring verb in Wurmbrand’s account, is that the complement that it
selects for is larger than the bare VP found with full lexical restructuring verbs. Ultimately
it must be a complement without any TP or CP structure. In the previous section, we briefly
saw evidence that the complements of causative and experiencer have can be passivized and
this indeed shows that the complements of these verbs must be larger than a bare VP. On the
other end of the clausal scale though, there is some reason to believe that not all instances
causative and experiencer have are always restructuring. For instance, Butters and Stettler
(1986) note the acceptability of causative have followed by a for-to complement for some

43

speakers in North Carolina, this configuration is also acceptable in the Appalachian English
of Eastern Kentucky.
(62) John had for Mary to wash the dishes.
However, such configurations fall outside the class of restructuring predicates as a whole
and therefore will not be dealt with in this analysis, except to show that they are nonrestructuring predicates. The non-restructuring causative and experiencer have predicates
must be set aside in order to provide coherent and convincing arguments for a causative and
experiencer have restructuring configuration. Subsequently, we will discuss the makeup of
the complements under have restructuring predicates along with their respective properties
in detail.
As shown in example (63a), the clausal for-to versions of these predicates in AppE license perfect morphology on the verb along with perfect have. The restructuring predicates
in examples (63b) and (63c) do not license perfect morphology either with or without the
auxiliary in a bare infinitival form. For the sake of clarity, I will contrast the clausal version of these verbs with the restructuring have in Standard English. Keep in mind that
examples (63b) and (63c) are restructuring complements.
(63)

a.
b.
c.

John had for Mary to have washed the dishes by the time...
* John had Mary washed the dishes by the time...
* John had Mary have washed the dishes by the time...

This shows that the perfect is banned in restructuring versions of causative and experiencer
have. If perfect have is present then complementizer for is also available and the constructions must be non-restructuring.
Furthermore, act be is acceptable in for-to clauses and is noticeably worse in the restructuring versions of these predicates. Recall that the instances of be in these examples only
receive act be interpretations and they are not instances of an overt progressive auxiliary.
Again consider the contrast with the SAE restructuring verb in example (64b).

44

(64)

a.
b.

John had for Mary to be washing dishes.
? John had Mary be washing dishes.

What we see is that overt perfect auxiliary co-occurs with a TP and a higher CP and that
act be likewise is more acceptable in a full for-to clause. Thus the infinitival-to which
appears with overt auxiliaries is actually located in T. Making this division we can move on,
setting aside instances of causative and experiencer have which contain tense and a complementizer layer as non-restructuring verbs. This sets the stage to show that the to present in
restructuring cases in AppE is not located in the tense layer of the restructuring complement.

2.3.2.1

The status of to under restructuring have in AppE

As is the case with other semi-functional restructuring predicates most things associated
with tense are non-existent in the complement of the restructuring varieties of causative
and experiencer have. First, as we have previously seen, expletive-there is banned in the
complements of causative and experiencer have, repeated below in example (65). Remember
the sentences in this section correspond to the SAE version without to.
(65)

a.
b.

* John had there to seem to be a problem.
* John had there to seem to be a problem on him.

PRO is also banned as the subject of complements embedded under causative and experiencer
have. The causative in example (66a) is unacceptable on a causative reading though an
obligational reading available. Although I speculate obligational have may be reducible to
the same syntax as the non-restructuring version of causative and experiencer have, this is
beyond the scope of this work. It is sufficient that causative have does not license PRO as
the subject of its complement.
(66)

a.
b.

* John had PRO to wash dishes.
* John had PRO to wash dishes on him.

45

This fact falls out from a basic theory of PRO in which it requires null case. Under the
assumptions that I have made here about the articulation of causative and experiencer have
clauses, PRO is only in the position to receive accusative case from the v above it. So, even
if we were to assume that T assigns null case per Martin (2001) it would not be assigned
in the context of these restructuring complements as they are not maximally headed by T.
Beyond this, I remain agnostic as to a theory of PRO, especially considering that AppE is
at least a partial for-to dialect.
Finally, auxiliary have and be associated with the progressive and the perfect are degraded
in restructuring complements.
(67)

a.
b.

* John had Mary to have washed dishes.
* John had Mary to be washing dishes.

In Appalachian English this argument is a bit trickier to make. With the SAE bare-infinitive
counterparts we said that any instance of auxiliary have requires overt infinitival-to in T and
that this entails the possibility of inserting a for complementizer. This argument does not
hold so clearly here. Fortunately, insertion of complementizer for alleviates the unacceptability and reveals a distinction between the unacceptability of example (67) and example (68)
below.
(68)

a. John had for Mary to have washed dishes.
b. John had for Mary to be washing dishes.

Crucially, only the morphological markers shown in example (69) are acceptable in the
restructuring instances of have. Since, I am assuming a uniform treatment and causative
and experiencer have don’t differ on these facts, only causative have is shown for clarity.
(69)

a. John had Mary ∅ wash the dishes.
b. John had Mary washing dishes.
c. John had the dishes washed.

bare-infinitive
progressive
passive

This leaves the Appalachian English to in example (70) as a dialectal variant of the SAE
bare-infinitive in example (69a).
46

(70) John had Mary to wash the dishes.

default to

The evidence that T and overt auxiliaries are banned in restructuring have complements
combined with the fact that only the progressive -ing, the passive -en, and the pre-verbal
to are allowed in the restructuring cases indicates that these complements are maximally
headed by and aspectual phrase which licenses the morphology on the verb and that same
head acts as the insertion site for to in AppE. This leads to the conclusion that the AppE
to is the result of a default insertion rule which is triggered in the absence of inflectional
morphology on the embedded AspP head.

2.3.3

Default infinitival morphology

In this section, I discuss the precedent for the notion that there is a default insertion rule
in Appalachian English assuming that syntax that I have just argued for. Then I will show
how it is implemented using the framework of Distributed Morphology. First, I will briefly
present some restructuring data from a dialect of Italian where an identical default insertion
rule has been proposed for just these situations.
Consider the case of one causative restructuring verb in Salentino Italian shown in example (71) from Calabrese (1997).
(71) ntSi
addZu fattu kkattare lu milune
to.him have-1s made buy-inf the melon
‘I made him buy the melon.’
Notice that the causative fattu or ‘make’ is followed by a verb which bears infinitival morphology. Calabrese (1997:48) argues that “modal/aspectual verb + infinitive sequences should be
treated as mono-sentential verbal complexes where the infinitive is responsible for determining the argument structure of the sentence and its basic sentential configuration.” A notion
that I repeat and one that is reflected in many of the works mentioned so far (see Ritter &
Rosen, 1997; Harley, 1997; Wurmbrand, 2001). To capture this for Italian, Calabrese posits
the rule in example (72) following a similar set of facts found in Catalan (Picallo, 1990). The
47

rule is applied when the verbs are embedded in structure where they receive no inflectional
agreement.
(72) [∅] ↔ infinitive

2.3.3.1

The assignment of English verbal morphology

In this section, I will briefly discuss Halle and Marantz (1993) which proposes an account
of the morphology of English verbal inflection which is relevant to the morphosyntax of
causative and experiencer have. Using this base to work from I will then present and argue
for the vocabulary insertion rules for the verbal morphology found in the complements of
causative and experiencer have in Standard and Appalachian English.
English verbal inflection is the result of the combination of morphosyntactic features
that may be generated either in the terminal node of the tense layer or at some terminal
inflectional node in the IP constituent, Halle and Marantz (1993). I will assume this is the
aspectual auxiliary system in English located between vP and T as this must be the case for
our smaller than TP restructuring complements.
They divide the verbal inflection system into five parts.

(73)

past participle
beat-en put
past finite
beat
put
nonpast finite 3rd sg beat-s
put-s
nonpast participle
beat-ing putt-ing
nonpast finite
beat
put

dwel-t
play-ed
dwel-t
play-ed
dwell-s
play-s
dwell-ing play-ing
dwell
play

Thus assuming that vocabulary items will be inserted onto a fused head bearing both
tense and agreement features Halle and Marantz posit the following tense/aspect related
features to account for the distribution of verbal morphology: [±past] and [+participle].
This five feature system is cashed out in the following vocabulary insertion rules for T head.

48

(74) I = fused tense and agr
[+participle, +past]

↔

[+past]

↔

[+past]

↔

[+past]
[+participle]
[3sg]

↔
↔
↔
↔

/-n/ / X +
where X = hew, go, beat, . . .
∅/Y+
where Y = beat, drive, bind, sing, . . .
/-t/ / Z +
where Z = dwell, buy, send, . . .
/-d/
/-ing/
/-z/
∅

The list is ordered from complex to least complex feature sets. Verbs ending in participle
-en have features comprised of [+participle, +past] while verbs with other forms are licensed
by [+past] morphology which is differentiated by virtue of disjoint lists of specific verbs in
the lexicon. Of interest for our analysis is the fact that the past and progressive participle
differ by only [+past]. Presumably under this analysis the [+past] feature is licensed in the
syntax by T, but in our restructuring clauses which are maximally headed by AspP there
is no evidence that such a feature is ever present. Hence, to maintain the distinctions that
a two feature system grants us we would have to make some slight readjustments to any
vocabulary insertion rules that are applicable in the smaller than TP restructuring contexts.
However, I will simply state that the feature [+past] is a cover term for whatever the real
feature might be and continue to use it for ease of exposition and translation between this
system and the list of vocabulary items that I will propose for Appalachian English.

2.3.3.2

The assignment of verbal morphology in restructuring complements

Here I will argue for a distributed morphology account of the verbal morphology in restructuring contexts. First, let’s reconsider the structure in question, repeated below.

49

(75)

vP
John

v
v
had

AspP
Maryi

Asp
vP

Asp
to/∅
ti

v
v
washk

VP
tk dishes

Following Halle and Marantz (1993), we will consider the spell-out of verbal morphology on
the verb in these cases to be the result of merger of Asp0 and v under structural adjacency
Marantz (1988; 1989). Further, we must assume that actual tense is not present but that
we are using [+past] as a feature which will let us maintain the distinctions in the featural
content of the auxiliary system of full TP clauses as well as those found in restructuring
complements. For Standard English, we saw that these complements can only be passive,
progressive, or bare. The assumption that [+past] is simply a cover term for the actual
feature involved allows us to implement exactly the list of rules for vocabulary insertion
even though T is absent. Further, assuming person and number agreement is specified on
T means that they play no part the spell out of morphology in restructuring contexts as we
expect.
(76) v = fused Asp0 and v
[+participle, +past]

↔

[+past]

↔

[+past]

↔

[+past]
[+participle]

↔
↔
↔

/-n/ / X +
where X = hew, go, beat, . . .
∅/Y+
where Y = beat, drive, bind, sing, . . .
/-t/ / Z +
where Z = dwell, buy, send, . . .
/-d/
/-ing/
∅
50

Lets consider the morphological merger in the context of the passive in example (77) for
the sentence John had the dishes washed where we will assume following Halle and Marantz
(1993) that merger is represented by downward movement.
(77)

a.

vP
John

v
v
had

AspP
the dishesi
Asp
vP

Asp

+participle


+pass
+past


v
v
VP
washk ti

b.

vP
John

v
v
had

AspP
the dishesi

Asp
vP

Asp
tj

v
vm

VP

V DP
Aspj

tk ti
+participle


+pass
+past

washk 

At this point, I will assume that features below the level of v constitute a morphological word
and as such are an insertion point for phonological exponents. This is represented by m in the
derivation. The modified rules in example (76) will now account for the passive, progressive,

51

and bare morphology that we see in SAE in the complements of causative and experiencer
have. We now need only to account for the AppE cases which include to. Note that I will
also assume that the bare infinitive is also possible in AppE for ease of transposition, but I
do not wish to make any claims about the state of the grammar for AppE speakers who find
both to and bare variants acceptable.
Under the system presented here, I take the variation that occurs in AppE to mean
that for Appalachian speakers the process of morphological merger is optional. In the case
where speakers morphologically merge AspP and v the results will be just like those in SAE,
complete with a bare infinitive form. The aspectual-to case then falls out of the sytem as an
the result of a lack of merger of the requisite heads. In such cases, the verb will remain in v
where it will receive no features via merger and appear in the verb’s bare form according to
example (76). This is explicitly repeated in a rule specifically for v below.
(78) v = no morphological merger
[ ]

↔ ∅

Similarly, if there is no morphological merger then Asp0 will also be evaluated in the absence
of external morphological features, and to will be inserted.
(79) Asp0 = no morphological merger
[ ]

↔ /to/

Example (78) and example (79) are illustrated in on tree below, where the empty brackets [ ]
represent the site of vocabulary insertion for to in AppE versions of causative and experiencer
have. It serves as an example of default infinitival morphology similar to cases in Salentino
Italian and Catalan.

52

(80)

vP
John

v
v
had

AspP
Maryi

Asp
vP

Asp
[ ]
ti

v
v
washk

VP
tk dishes

The question remains as to why AppE speakers have this infinitival-to insertion rule to begin
with. I speculate that it might serve to satisfy a prosodic constraint but this is of course
beyond the scope of this dissertation. I leave this question to further research.

2.4

Summary

Though the surface forms differ and speakers of SAE often find the AppE version of causative
and experiencer have unintelligable, I have argued that the syntax for causative and experiencer have in SAE and AppE is identical. This uniform treatment of causative and experiencer have also supports a mono-clausal approach to restructuring verbs and this has
implications for theories of non-finite complementation in language. I argued that causative
and experiencer have are an excellent fit for the class of semi-functional restructuring predicates as outlined in Wurmbrand (2001). These verbs generally have some thematic role
assigning properties but behave somewhat like heads which are more functional and less like
true lexical verbs. Specifically, these two versions of have are shown to be in v and select
complements maximally headed by AspP. As we would expect, the causative and experiencer
have in AppE which allow infinitival-to were shown to have none of the usual properties associated with a true infinitival-to. This in turn means that the presence of the infinitival-to
53

in AppE (as opposed to the bare infinitive in SAE) is the result of a morphosyntactic rule,
Asp0 :[ ] ↔ to which inserts infinitival-to onto aspectual heads which fail to receive feature
content from higher up the tense/auxiliary chain or via morphological merger. This situation in AppE is similar to those described in Catalan (Picallo, 1990) and Salentino Italian
(Calabrese, 1997). Finally, This examination of micro-syntactic variation has shown implications for how we understand features of the tense and auxiliary system in English more
broadly. It requires that the features assumed by Halle and Marantz (1993) to account for
verbal inflection in English be slightly readjusted to account for the lack of tense features in
the context of restructuring complements.

54

CHAPTER 3
PERCEPTION VERBS AND THEIR COMPLEMENTS

3.1

Introduction

In Standard American English (SAE), infinitival-to is absent in active perception complements and required in passive perception complements, as shown in example (1). The passive
is shown in example (2). This asymmetry in English complement clauses is peculiar to perception complements and some causation complements (Hornstein et al., 2008).
(2) Passive SAE
a. * Mary was seen leave.
b.
Mary was seen to leave.

(1) Active SAE
a.

John saw Mary leave.

b.

* John saw Mary to leave.

Perception complements in Appalachian English exhibit a different pattern. They show us
that they are not subject to the same restrictions as those in Standard American English.
As shown in (3), AppE allows infinitival-to along with the bare infinitive in the active.
Similarly in (4), AppE allows both bare complements and those headed by infinitival-to
in passive contexts. In comparing these two dialects, we find yet another point at which
Appalachian English varies from the standard in infinitival complementation.
(3) Active AppE

(4) Passive AppE
a. Mary was seen leave.
b. Mary was seen to leave.

a. John saw Mary leave.
b. John saw Mary to leave.

While the requirement of the bare infinitive in active perception complements in SAE
as shown in (1) has been described in terms of c-selectional constraints on the verb, the
constraint on the passive in SAE shown in (2) is commonly assumed to be a product of
special properties of the passive and the passive participle.

55

Following the notion that the passive participle is involved in the ban on matrix passivization, we find that many linguists have argued that the passive participle ‘disallows’ the
establishment of some relation between the matrix and embedded clauses. For illustrative
purposes suppose, following Bennis and Hoekstra (1989/2004), that this relation has to do
with the establishment of a single tense domain for both clauses. In example (5a), both the
matrix and the embedded tense layer have identical indices. Assume that this denotes the
establishment of a single tense domain. Now suppose that in example (5b) the generation of
the passive participle has interrupted the process of establishing a single tense domain; the
embedded T remains un-indexed.
(5)

a. [Ti John saw [Ti Mary leave]]
b. [Ti Mary was se-en [T t leave]]

*SAE

Interestingly, a majority of the analyses of perception complements include some version
of the stipulation that the passive morpheme, -en blocks the establishment of some syntactic
relationship. For Hornstein et al. (2008), perception complements in English fail to passivize
because the passive -en participle bears phi-features which prevent the successful probe and
consequent agreement with various embedded elements. Bennis and Hoekstra (1989/2004)
argue that the passive participle blocks proper establishment of tense relations between
the matrix and embedded clauses; the situation illustrated by example (5). Finally, Felser
(1999) argues that passivized verb itself blocks the establishment of relations between event
variables in the matrix and the embedded clause. These three analyses all share the common
assumption that the passive morphology is the culprit responsible for the unacceptability
of example (5b). Recall that Wurmbrand’s analysis of the lack of matrix passivization in
perception complements is quite different. She assumes that some verbs are more a part of
the functional structure of the clause and generated in v while others are purely lexical and
are generated in V. Additionally, she posits a host of embedded complement sizes. As we saw
in the previous chapter, she predicts the unacceptability of example (5b) by placing the set
of semi-functional restructuring verbs (motion verbs, causatives, perception complements) in

56

a type of complementary distribution for merge with passive vP. For the rest of this chapter,
I will assume Wurmbrand’s approach to the ban on matrix passivization in semi-functional
restructuring complements to hold. This does not however preclude the possibility that
lexical restructuring verbs in V are not subject to further constraints against passivization
induced by the passive participle as many authors have claimed. For that reason, I will also
provide a solution to the problem of matrix passivization of perception complements which
takes into account the commonly observed effect of the passive participle.
I will argue that the variation which exists in perception complements in AppE in examples (3) and (4) is the result of two properties of the language. First, matrix passives
of bare PCs are possible in AppE because syntactic properties of passive participle are different. To put it simply, participles are transparent and allow for the proper establishment
of relations between clauses. As such, the passive morphology no longer makes the embedded clause opaque for syntactic operations. Such claims are consistent with the fact that
Appalachian English is a dialect whose verbal paradigms have undergone severe leveling.
Paradigm leveling in AppE irregular forms is ubiquitous and is occurs in both the passive
and perfect participle contexts. For instance in example (6a), the verb broke appears in
a verbal passive without the passive -en morphology. Likewise, in example (6b), the verb
also appears in a perfect construction with an overt perfect auxiliary without the requisite
participle morphology.
(6)

a. It was broke by somebody.
b. I had broke it before they arrived.

This configuration of verbal morphology and the restructuring analysis that I propose here
have implications for theories other than those of complementation. The facts require that we
question basic assumptions that we have about the composition of non-standard grammars
as well as the acquisition such purportedly ‘variably leveled’ forms for speakers of those
non-standard grammars. Specifically, in terms of leveling of certain grammatical paradigms
it reveals certain egocentrics views about how similar non-standard dialects are to standard

57

dialects.
Second, since the focus of this work is on non-finite complementation in the context of
restructuring predicates, I have less to say about the why AppE allows infinitival-to in the active versions of perception complements. I will however show that they are non-restructuring
predicates and that the infinitival-to that appears there is an actual manifestation of T. I
have even less to say about this should be the case beyond c-selectional restrictions and I will
show that even SAE allows these to occur with with expletive subjects in a very restricted
set of cases.
Finally, I will compare the analysis of causative and experiencer have to the syntax of
perception complements in AppE and argue that the Wurmbrandian approach to restructuring verbs alone is insufficient to account for the differntial behavior of these two classes
of verbs.
This chapter is structured as follows. In the next section, I discuss the main types and
properties of perception complements as found in Standard English before discussing those
types in light of the Appalachian English data. Then in section section 3.3, I review three
previous analyses in which the passive participle is the culprit behind the constraint against
matrix passivization of perception complements. In section 3.4, I propose my analysis. It is
a synthesis of Wurmbrandian style mono-clausal restructuring approaches and approaches
which assume that the passive participle still prevents matrix passivization. Successful analysis of the facts requires an extensive detour into the morphology of participles in Appalachian
English. Section 3.5 is a comparison of the analyses of causative and experiencer have and
perception complements in AppE which leads the to conclusion that external mechanisms
such as head-movement and semantic constraints are also relevant in micro-variationist analyses of restructuring complements. Section 3.6 is a brief summary and conclusion.

58

3.2
3.2.1

Properties of perception complements
Types of perception

Perception complements have two basic types of interpretation. They come in direct perception form which seem to denote bare events and correspond to what we have been calling
restructuring predicates and in a indirect perception form which are propositional and correspond to what we have called non-restructuring predicates. The propositional perception
complements are realized as tensed complements in English and are exemplary of indirect
perception
(7) Indirect or epistemic perception complements
a. John saw that Mary left.
b. John heard that John drank all my whiskey.
c. John felt that Mary was being mean.
In indirect perception complements, the perception event and the event perceived are not
required to overlap in the time. For instance, in example (7a), The utterance is felicitous
with a situation where the seeing event that holds of John is located after the point in time at
which Mary leaving event is completed. More colloquially, Imagine that John arrived home
after Mary had left. He was aware that she had reason to leave and he saw that she left
because all of her things were missing. Because of this interpretation fact, these propositional
complements are considered to be indirect perception. Further, indirect perception see often
reflect the notion of “discovery”. Consider the comparison below where see and discover
share an interpretation.
(8)

a. John saw that Mary left.
b. John discovered that Mary left.

On the other hand, direct perception complements pattern with restructuring verbs.
They require that the two events that comprise such utterances overlap in the time course
of events.

59

(9) Direct perception complements
a. John saw Mary leave.
b. Mary heard the bone break.
c. John felt the ground shake.
For instance, the seeing event in example (9a) and the leaving event of its complement must
overlap in a time course of events. The sentence is simply infelicitous if John’s seeing and
Mary’s leaving do not overlap in time. Felser (1999) refers to this fact as the constraint on
temporal simultaneity. These bare perception complements never license a complementizer
or elements associated with tense.

3.2.1.1

Direct perception

Given Wurmbrand’s diagnostics for the properties which divide predicates into restructuring
and non-restructuring predicates, it seems that direct perception verbs simply take different complements than indirect perception verbs. Following Felser (1999) and Wurmbrand
(2001), I assume that these two types of complements are not related derivationally. More
specifically, we could say that direct perception complements in SAE are semi-functional
restructuring complements just like causative and experiencer have and set indirect perception complements aside as non-restructuring complements. In what follows, I describe the
behavior of direct perception complements. True to form with restructuring verbs, direct
perception complements ban most things associated with Tense. The list of properties in
this section have been adapted from Felser (1999) unless otherwise noted.
First, note that direct perception complements ban sentential negation.
(10)

* John saw Mary not hit Bill.

Example (10) is unacceptable under a sentential negation interpretation which is associated
with tense. It does not mean John saw Mary didn’t hit Bill. The only possible interpretation
is one of constituent negation which can appear regardless of tense. For instance, with a
continuation phrase John saw Mary not hit Bill but kiss him instead is quite acceptable

60

but not on a sentential negation interpretation. It only allows a reading where the verbal
constituent in the complement is negated and contrasted with another constituent of the
same type.
Since direct perception complements are restructuring verbs they pattern with other
restructuring verbs in that they do not license modal auxiliaries. Example (11) below shows
that epistemic modal might is not licensed.
(11)

* John saw them might hit Bill.

Similarly, aspectual have and be which correspond to the perfect and the progressive
auxiliaries are also banned in direct perception complements just at they were banned in the
complements of causative and experiencer have.
(12)

a.
b.

* We heard Bill be talking to himself.
* We saw Mary have finished her breakfast.

Note that just as we saw with causative and experiencer have, the only possible interpretation
of example (12a) is the act be reading which requires agentivity on the part of the embedded
subject. This is not the same reading that we find with the progressive. To my knowledge
there are no felicitous interpretations of the sentence in example (12b).
Not surprisingly, since direct perception verbs pattern with restructuring verbs in terms
of their properties, they do not license infinitival-to associated with T or the default to
outlined in the last chapter. I will return to a discussion of this fact towards the end of the
analysis section.
(13)

* John saw Mary to hit Bill.

Finally, direct perception complements do not license expletive subjects.
(14)

* John saw there be an accident Bill.

In summary, direct perception complements seem to lack a tense layer and should therefore be thought of as restructuring predicates. They do not allow elements associated with

61

tense such as sentential negation interpretations, modal auxiliaries, infinitival-to of any sort,
or expletive subjects. They also do not allow any aspectual auxiliaries. All in all, the
diagnostics so far reveal a configuration that is similar to causative and experiencer have.

3.2.1.2

Indirect perception

Indirect perception complements are quite different from direct perception complements and
may be characterized by an epistemic reading along with the fact that they do not require
temporal event overlap. For instance in example (15), John’s seeing event need not overlap
in a time course of events with and event of Mary smoking.
(15) John saw that Mary smoked.
Beyond this, indirect perception complements correspond to non-restructuring predicates
because they commonly allow disjoint tense interpretations between matrix and embedded
clauses, sentential negation in the embedded clause, complementizers, and other phenomenon
associated with the presence of an embedded tense layer. Again the list of properties which
follows is adapted from Felser (1999).
Given that these complements are non-restructuring complements we expect that interpretations of sentential negation are allowed in indirect perception complements.
(16)

John heard that Mary didn’t like sushi.

We can see that the most natural interpretation of example (16) is one of sentential negation.
In fact, special prosodic focus is required on the verb in order to force a different contrastive
interpretation, John heard that Mary didn’t LIKE sushi, he heard she loved it.
Modal auxiliaries and aspectual have and be are licit in indirect perception complements.
This is indicative of T/Aux structure which is present in non-restructuring predicates.
(17)

a.
b.
c.

John saw that Mary might hit Bill.
We heard that Bill was talking to himself.
We saw that Mary had finished her breakfast.

62

Example (17a) shows that modal auxiliaries are possible in indirect perception complements
but they are indicative of a tense layer and may co-occur with complementizers. In example (17b) and example (17c), we see that the progressive and perfect auxiliaries are also
allowed.
Finally, tense marking and expletive there subjects are also allowed in the complements of
indirect perception complements. Again, these are indicative of a tense layer in the embedded
clause.
(18)

3.2.2

a.
b.

John saw that Mary slapped Bill.
We saw that there was a need for security guards.

Direct and indirect perception complements in AppE

We have seen some of the properties that distinguish direct perception complements from
indirect perception complements in Standard English and in this section I discuss those
differences in terms of the Appalachian English data. This means that the bare infinitivals
constitute restructuring predicates and the infinitival-to predicates are non-restructuring
predicates, regardless of whether they appear in active or passive voice.

3.2.2.1

Direct perception in AppE

Direct perception complements in AppE differ from their SAE counterparts only in that
they are allowed in both the active and the passive. Consider the bare forms in both dialects
repeated below.
(19) AppE
a.

John saw Mary leave.

b.

Mary was seen leave.

(20) SAE
a.
John saw Mary leave.
b. * Mary was seen leave.

The most intuitive analysis of these facts and the one I adopt here, is that the set of perception complements in example (19) are true lexical restructuring predicates while the set of

63

perception complements in example (20a) are semi-functional complements. For Wurmbrand
this explains why the sentence in example (20b) resists passivization. This analysis fits with
the previous restructuring analysis of causative and experiencer have. Recall that neither
causative nor experiencer have allow matrix passivization because they are semi-functional
restructuring predicates. Under this analysis they pattern almost exactly like bare perception
complements in Standard English. Such mono-clausal approaches to restructuring also automatically require us to assume that the two types of clauses are not derivationally related.
Rather, in SAE we must say that the semi-functional restructuring verb in example (21a) is
generated in v, while the non-restructuring predicate in example (21b) is verb which selects
a non-restructuring complement with a full TP.
(21) SAE
a. John saw Mary leave.
b. Mary was seen to leave.

semi-functional restructuring PC
non-restructuring PC

In fact, if we take Wurmbrands’s analysis of German restructuring predicates seriously and
apply it to English, then matrix passivization may be the only visible feature which defines
lexical restructuring predicates as a subset of restructuring predicate for English. Under
Wurmbrand’s analysis the assumption that Standard English perception complements are
related and somehow defective between the active and passive thereby causing alternations
in the presence or absence of infinitival-to disappears. Instead, Standard English perception
complements are simply not representative of a full set as compared to those in other dialects
such as Appalachian English.
Meanwhile, Appalachian English has a lexical restructuring version of see because they
may undergo matrix passivization as shown in example (22) and a non-restructuring see
shown in example (23) which allows both active and passive non-restructuring complements.
(22) Lexical restructuring PCs in AppE

(23) Non-restructuring PCs in AppE
a. John saw Mary to leave.
b. Mary was seen to leave.

a. John saw Mary leave.
b. Mary was seen leave.

64

The question of why SAE seems to be missing an active non-restructuring complement
with infinitival-to is beyond the scope of this work, and I will set it aside for further research.

3.2.2.2

Indirect perception in AppE

Indirect perception in Appalachian English proceeds for the most part as it does in SAE as
we saw in section 3.2.1.2. However, it seems that AppE allows indirect perception for both
the active and passive versions of perception complements via infinitival-to. In this section,
I show that active indirect perception complements with infinitival-to in AppE are actual
instances of to in T as opposed to the default infinitival-to discussed in previous chapter.
(24)

a. John saw Mary to have drank all the whiskey.
b. Mary was seen to have drank all the whiskey.

active
passive

Thus, Appalachian English appears to have an extra way of signifying indirect perception
via infinitival-to. Unlike the infinitival-to that we have previously seen with causative and
experiencer have, the presence of infinitival-to in perception complements corresponds to
a difference in meaning. Specifically, it corresponds to indirect perception and as such
relegates these forms to being non-restructuring complements which minimally contain a
tense layer. Evidence for the notion that this infinitival-to is actually representative of a TP
layer in the embedded clause can be seen by the fact that aspectual have is only licensed
by an overt infinitival-to. Example (25a) shows that have is licit with the infinitival-to and
example (25b) shows that aspectual have is illicit without an infinitival-to. Example (25c)
and example (25d) show that the same fact about infinitival-to in the passive.
(25) AppE
a.
John saw Mary to have left.
b. * John saw Mary have left.
c.
Mary was seen to have left.
d. * Mary was seen have left.
Further evidence that this infinitival-to is in fact representative of T comes from the related
set of facts for causative and experiencer have. Only non-restructuring causative and ex65

periencer have allow aspectual have in their complements. Recall in example (26a) that in
restructuring causative and experiencer have, infinitival-to is optional for AppE speakers.
However, example (26b) shows that the optionality of infinitival-to disappears when have is
is licensed. Furthermore, in all of the cases where have is licensed a for complementizer is
optionally licensed.
(26)

a. Mary had John (to) leave.
b. Mary had (for) John *(to) have left.

restructuring
non-restructuring

Returning to the case at hand, if the infinitival-to in active indirect perception complements
were the default infinitival-to of causative and experiencer have then we would expect it to
be optional and we might expect it to license the for complementizer.
This caveat aside, those elements which are were not admissible under direct perception
complements are admissible in Appalachian English active indirect perception complements.
For example, sentential negation interpretations are licensed in indirect perception complements.
(27) John heard Mary to not like sushi.
Example (27) may be interpreted as having a sentential negation reading as in John heard
that Mary didn’t like sushi.
Beyond this, these non-restructuring, indirect perception complements also license both
the progressive and the perfect auxiliaries in Appalachian English when infinitival-to is
present. As we saw above, the infinitival-to is required.
(28)

a.
b.
c.
d.

We
* We
We
* We

heard Bill to be talking to himself.
heard Bill be talking to himself
saw Mary to have finished her breakfast.
saw Mary have finished her breakfast.

Further, we can apply the analysis of auxiliary insertion from the previous chapter which
says that auxiliary insertion is only licit as the result of agreement with some higher auxiliary

66

head or T. This explains why T is required in these cases. If there was not a tense layer then
have would not be licensed.
Finally, we see that expletive there is not banned in indirect perception complements.
(29)

We saw there to arise a great haze over the valley.

(30)

We heard there to be a great commotion in the next room.

The findings thus far are summarized below.

3.2.3

Interim Summary

According to the diagnostics presented here, it seems clear that there is a set of perception
verbs in both dialects which have direct perception interpretations and require temporal
overlap for the events described in each clause. These verbs can be thought of as restructuring predicates because neither set’s complements exhibit properties associated with a
tense layer; they disallow complementizers, auxiliary and nominal elements associated with
tense, and sentential negation. Then we can split this set of restructuring predicates along
dialect boundaries for whether or not they allow passivization. I take this to be evidence
for Wurmbrand’s split in the classification of restructuring verbs into semi-functional restructuring predicates and ‘true’ lexical restructuring predicates which allow passivization.
With regards to indirect perception complements which have epistemic and ‘discovery’ interpretations, they fit neatly into classification as non-restructuring predicates as they allow
elements which are banned under their direct perception counterparts. Because they are
propositional in nature and are non-restructuring predicates they must be set aside for the
current discussion.

3.3

Previous Analyses

In this section, I will lay out three previous analyses of perception verbs and their complements which, among other things, seek to explain the constraint against the passive for

67

Standard English.

3.3.1

Bennis and Hoekstra (1989/2004)

The patterning of Dutch perception complements with respect to the active/passive asymmetry is similar to Standard English. Example (31) repeats the familiar English data showing
the ban on matrix passivization along with the same data in Dutch. Dutch patterns with
English in disallowing the matrix passivization of perception complements.
(31)

a.
b.

* Kaatjei was heard [AgrP ti [VP sing a song]]
* Kaatjei werd [AgrP ti [VP een liedje tk ]] gehoord zingenk

It should be said here that without explicitly naming it as such, Bennis and Hoekstra assume
a type of mono-clausal restructuring analysis for perception and causation complements.
Specifically, they argue the now familiar point that these complements are maximally headed
by a phrase smaller than T. They argue that these clauses are maximally headed by an agreement phrase AgrP and crucially not headed by T. As mentioned this structure constitutes a
type of restructuring clause under the Wurmbrand framework. The main argument however
is based on the idea that all verbs must be identified by tense, either in their own clause in
the case of full TP/CP clauses, or by the matrix tense in the case of embedded restructuring
clauses which lack a local TP. The second piece of the argument is that this identification
relationship between matrix and embedded verbs in restructuring contexts is interrupted by
the passive participle in both Dutch and English. In the remainder of this section, I will
review the basic mechanics of the T-linking analysis and then review the evidence for the
status of the passive participle as a non-t-linking element.

3.3.1.1

T-linking

As mentioned earlier, this account is based on the principle of T-linking which requires that
all verbs in clause be identified by tense in a tense chain or t-chain. This is formulated as
the condition in example (32).
68

(32) T-linking: A verb must be identified by tense.
Any T-chain must have the form in (33) where each P is locally connected to the next and
no non-t-linking elements may intervene.
(33) (tense, P1 , . . . , Pn , V)
Because the base position of tense itself varies across languages, they argue that the base
of T-chains vary across languages. For instance, there is good reason to believe that tense
is situated in T for languages like English while it seems to be based in C for verb-second
languages like Dutch. Languages may also vary on how they implement the establishment of
t-chains. They may either be established by overt verb movement as evidenced by languages
like Dutch, or in languages without overt verb movement they are established by feature
percolation down the tree. In general, the chain ranges from the left edge of a clause (originating in T in English) and through any heads between tense and the verb. They assume
the following structure for a fully articulated clause.
(34) [CP C [TP T [AgrP Agr [VP V ]]]]
T-linking as formulated, is a condition on the output of a derivation which the authors
claim to be the underlying factor in determining verb movement rules cross-linguistically.
Take for instance one possibility for verb movement in French (Pollock, 1989) as illustrated
by Bennis and Hoekstra.
(35)

a. (Jean a d´ecid´e de) ne pas souvent [V lire ] un livre.
John has decided to not
often
read a book
b. (Jean a decide de) ne pas [Agr0 lirei ] souvent [V ti ] un livre.
John has decided to not
read often
a book

This example shows us that in French, the verb in example (35a) may move from its base
position in V to a higher position Agr0 in example (35b). AgrP is an agreement head which
was commonly thought to be located between T and V in early versions of the Minimalist
Program. It was responsible for handling various types of agreement. In English however,

69

that same movement of the verb is not licensed. Consider a simplified version of the French
sentence below.
(36)

a.
b.

John has often [V decided ] to read books.
* John has [?P decided ] often [V ti ] to read books.

This shows that whatever mechanism allows for such movement in the French examples is
not active in English.

3.3.1.2

Perception complements and t-linking in Dutch

The idea with perception complements then is that they do not contain their own clause
internal tense layer and given the condition on t-linking in example (32) which requires
verbs to be identified by tense, they must be tense-linked to the higher clause. In Dutch,
this requirement is usually satisfied by overt verb movement. In the next few paragraphs, I
will lay out the supporting evidence for the T-linking account in Dutch.
Perception and causation verbs in Dutch require verb raising (VR) of the embedded verb
to that matrix tense phrase. In example (37) notice that the verb eten ‘eat’ has moved from
the embedded clause to the right of the matrix verb hoort/ziet/laat ‘hears/sees/lets’.
(37) dat Jan [Marie een appel ti ] hoort/ziet/laat eteni
that Jan Marie an apple ti hears/sees/lets eati
Crucially in perception and causation complements in Dutch, the embedded verb obligatorily undergoes verb movement as shown in example (37). Failure to undergo VR results
in ungrammaticality for these constructions. Below in example (38a), we see that the verb
can not remain in situ nor can the entire embedded clause be extraposed as shown in example (38b).
(38)

a.
b.

* dat Jan [Marie een appel eten] hoort/ziet/laat
* dat Jan hoort/ziet/laat [Marie een appel eten]

in situ
extraposition

Given the requirements for overt verb movement as opposed to percolation, the Dutch data
is a bit more telling in terms of the cause of the unacceptability of such constructions.
70

Taken alone this data is not too revealing. However, once we place this observation in
the context verbal participles we start to see interesting effects. Though VR is required in all
complements of perception and causation in Dutch, when participle assigning auxiliaries are
involved in matrix clause, everything changes. In the case of past participle morphology, the
verb still raises as per the requirement of VR, but only if the matrix verb is infinitival form.
This is called the infinitivus pro participio ‘infinitive for participle’ or IPP effect. This means
that in cases where we expect the past participle form of the verb, we see the infinitival form
of the verb appearing instead.
(39)

a. dat Jan mij heeft gehoord/*horen
that Jan me has heard/hears
b. dat Jan [ mij een liedje ti ] heeft *gehoord/horen zingeni
singi
that Jan [ me a song ti ] has *heard/hear

Example (39a) shows that in the derivation of a single clause, heeft ‘has’ triggers perfect morphology on the verb requiring the participle form gehoord and effectively bans the
infinitive horen. In the case of the restructuring verbs in example (39b) we see the opposite
effect with respect to what morphology is required on the matrix verb. The participial form
gehoord is unacceptable and the infinitive form is required instead. From this Bennis and
Hoekstra conclude that infinitives are possible links in a t-chain. The passive participle as
you might recall simply renders the derivation unacceptable. Thus, Bennis and Hoekstra
argue that VR can not trigger verb movement to a verb which bears participle morphology.
They argue explicitly from the requirement of IPP effects that participles may not be links
in a T-chain.
Now, consider again the introductory data for Dutch and English which shows the ban
on matrix passivization in (40b). If we assume that participles are not possible links in a
t-chain, then the reason for the ungrammaticality of example (40b) is now parallel with the
ungrammaticality of example (39b).

71

(40)

a.
b.
c.

* Kaatjei was heard [AgrP ti [VP sing a song]]
* Kaatjei werd [AgrP ti [VP een liedje tk ]] gehoord zingenk
* Kaatjei werd [AgrP ti [VP een liedje tk ]] horen zingenk

Remember that all verbs are required to be identified by tense. Perception complements
in Dutch and English correspond roughly to what we have been calling restructuring complements which crucially lack a tense layer. This means that the embedded verb must be
identified by matrix T via t-chain formation. The process of T-linking in Dutch perception
complements is arguably established via movement where the embedded verb must move to
the matrix clause in order to be identified. However, the data so far shows that VR while
obligatory in perception complements, may never proceed if the matrix verb is in participle
form. In the case of the perfective participle, VR may proceed if the matrix verb appears as
an infinitive; as shown by IPP constructions. However, in the case of the passive participle
not even IPP can save the derivation as shown by example (39b).
As for an explanation of why there should be such a split in the behavior of perfective
and passive participle morphology, Bennis and Hoekstra (1989/2004:pg.140) speculate that
IPP effects are the result of a ‘marked rule of phonological deletion of participle morphology’.
Further they argue that, in the case of past participles no unrecoverable material is deleted
when the past participle morphology is left out. The passive participle, they argue, does
contain unrecoverable material which must not be deleted. They argue that it is the bearer
of the external theta-role following Roberts (1987) and Hoekstra (1986). Based on this,
deletion of the passive participle morphology would violate the principle of recoverability
of deletion. In turn, this requires that the explanation of the unacceptability of matrix
passivization in perception complements is explained along the same lines. However, the
notion that participles in general are not possible links in a t-chain comes solely from Dutch
IPP and VR data.

72

3.3.1.3

Implication for perception complements in AppE

This analysis is important because it highlights the relationship between obligatory verb
movement in restructuring complements in Dutch which requires overt verb movement and
its interaction with participial morphology; technical details of the mechanism aside. As for
the finer details of the mechanism of percolation in the case of English, there seems to be
no reason that one could could not update this system to a more modern syntax involving
probe/goal agreement relations originating with T. In fact, all three analysis of perception
complements that I discuss here might be updated in this way. Even so, the question would
remain as to what exactly those agreeing ‘features’ might be. Again, the most important
point raised by this analysis is the notion of the participle as an element which intervenes
between intra-clausal relations, even in the mono-clausal context of restructuring verbs.
That said, regardless of the findings for Dutch, it can not be the case that the perfect
morphology on the verb in English results in the verb being non-t-linking element.
(41)

a.
b.
c.

Mary had seen John drink whiskey before we arrived.
* Mary had see John drink whiskey before we arrived.
* Mary had to see John drink whiskey before we arrived.

In this exact situation we might expect some correlate of the Dutch IPP effects with English
sentences however, there is no such correlation. Of course, it could be the case that the mechanism responsible for IPP effects only operates in the context of verb raising and not feature
percolation. Regardless, it seems that the perfect participle is not only a t-linking element in
English, it is absolutely required. This fact, coupled with the fact that matrix passivization is
banned in English seems to indicate that the two participles do behave differently in English
perceptions complements, if we take Bennis and Hoekstra seriously. This also means that
since neither participle leads to ungrammaticality in AppE perception complements that
neither participle is a non-t-linking element. The participles in Appalachian English must
be different. The discussion of the status of participle morphology in Appalachian English
will play a major role in my analysis and I will return to it in my analysis.

73

3.3.2

Felser (1999)

Claudia Felser’s Event Control-Hypothesis is a slight variation on the theme that we just
saw. Aside from being an amazingly complete descriptive work on the semantic and syntactic
properties of perception complements in Standard American English, Felser offers another
account of the constraint against matrix passivization in perception complements in English
and across languages. Felser repeats the data that we are by this point so very familiar with.
(42)

a.
b.

We saw John draw a circle
* John was seen draw a circle

The approach is a bit more technical in terms of syntactic details but the essentials of the
argument, as compared with Bennis and Hoekstra (1989/2004), remain more or less the
same. She posits a sort of linking requirement for both clauses and makes the now familiar
claim that the passive verb interferes with the establishment of that requirement.
Felser argues following Kratzer (1995) that all stage-level predicates project an event
variable. This event variable must be identified by tense locally, or by some other event
variable which has been identified by T. She then argues that passivized perception complements are ruled out because the matrix passive does not project an event variable and
therefore can not serve as a controller for the event variable of the embedded clause. Again
the locus of the problem of deriving the constraint against the passive is located in some
facet of the matrix passive.

3.3.2.1

Felser’s assumptions

In order to unpack her analysis, I will now review her relevant assumptions. She assumes
the basics of a mono-clausal approach to restructuring complements by arguing that the
complement of perception verbs lack a TP. Instead she claims that they are maximally
headed by an aspectual phrase. She bases this claim on the fact that direct perception
complements maximally allow the -ing progressive participle to appear on the embedded

74

verb. Compare example (43a) and example (43b) below and note that they both allow a
direct perception interpretation, the latter just happens to also be in the progressive.
(43) SAE
a. We saw John draw a circle.
b. We saw John drawing a circle.
Rather than tying the requisite intra-clausal relationship directly to tense as in the previous account, Felser links the relationship to an event variable. Evidence that all perception
complements project an event variable is argued from the fact that perception complements
only license stage-level predicates and never license individual-level predicates. This is purportedly shown by the contrast in example (44) where the stage-level predicate John be
obnoxious is fine as the complement of a perception verb but the individual level predicate
John be tall is banned.
(44)

a.
b.

We saw John be obnoxious.
* We saw John be tall

Unfortunately, this claim does not entirely hold. Schmitt (1996) has shown that some individual level predicates are licensed under perception complements, as in We saw John be
cruel. However, for the sake of exposition I will simply note this fact and return to Felser’s
Analysis.
So if all stage-level predicates project event variables (Davidson, 1967; Higginbotham,
1985; Kratzer, 1995) as is commonly assumed and all perception complements are stagelevel, then all perception complements project an event variable. Felser locates this event
variable as being external to the external argument of the verbal complex.
(45) [e [Argexternal [Verb Arginternal ]]]
Specifically, she argues that this external event argument is located in the head of an aspectual phrase. Consider the reduced tree for We saw John draw a circle, where e represents
Felser’s event variable.

75

(46)

VP
AspP

V
see
John

Asp
VP

Asp
e

V
draw

DP
a circle

Per Felser’s analysis, the perception complement structure in example (46), is projecting the
event variable required of a stage-level predicate. However, since there is no tense in the
embedded clause, it must undergo identification via entering into a control relationship with
another event variable. Identification is indicated by subscript i.
(47)

AspP
subj

Asp
V

Asp
e
see

AspP
John

Asp
VP

Asp
ei

V
draw

DP
a circle

Notice that in (47) the embedded event variable has entered into an identification relationship
with the matrix event variable e which must now be identified by T. Recall that, similar
to the requirement that verbs be linked to tense in a T-chain, Felser argues that an event
variable otherwise unidentified, must be identified by tense. The indexing of the matrix
event variable with T is shown below.

76

(48)

TP
Wei

T
AspP

T
tensei
subj

Asp
VP

Asp
ei
V
see

AspP
John ei draw a circle

With her basic assumptions about the derivation of perception complements and the mechanism of event-control out of the way, we are in a position to describe how Felser derives the
constraint against the passive.

3.3.2.2

Event-control and matrix passivization

To illustrate how Event-Control can account for the constraint against the passive, Felser
makes several assumptions about the passive in Standard English with regards to the eventivity and the architecture of the passive vP/VP shell. In this section, We see how her
assumptions about the status of passives with regards to eventivity are founded on shaky
theoretical and empirical ground and how her assumptions about the vP/VP shell are in
opposition to current minimalist assumptions concerning the structure of passives. In what
follows, I ultimately reject Felser’s argument that passivized PCs lack the appropriate controller, in favor of the idea that the passive interferes with temporal indexing of the embedded
event argument by the matrix event argument.
The claim is that passivized direct PCs are ungrammatical because they do not contain the right controller. Notice the lack of an event variable in the matrix clause of the
unacceptable passive in (49).
(49)

a.
b.

We ei saw [AspP ei John draw a circle.]
* Johnk was seen [AspP tk e draw a circle.]
77

In example (49a) the embedded event variable has been identified with the matrix event
variable which has been identified with T in the matrix clause. This allows the derivation
to converge. On the other hand, example (49b) depicts a situation where the embedded
event variable is unidentified by the matrix variable because the matrix clause simply lacks
an event variable.
Felser notes that passive constructions either do not have an external argument or suppress their external argument, and following Burzio’s Generalization, they do not check
objective case (Burzio, 1986). Further, she takes the lack of an external theta or argument
role to mean that passive VPs do not project a vP shell. Felser also suggests, that passive
VPs are individual level predicates and therefore they do not inherently project an AspP
or an event argument. She states that “This claim is consistent with the fact that passive
clauses are often considered to be stative in character” (Felser, 1999:pg.187) Felser ties the
claim that passives do not inherently project an aspect phrase to the fact that if a “predicate projected from a past participle or from an adjective is to be used in the progressive, an
auxiliary must be inserted as the carrier of progressive morphology,” (Felser, 1999:pg.187).
(50)

a. John was being obnoxious.
b. Mary was being questioned.

She claims that when the auxiliary is “inserted” it brings an event argument with it, but
since the auxiliary introduces the event argument, it is not a “designated argument of the
participle or participial VP” (Felser, 1999:pg.188) and therefore it is not able to act as a
controller for other event arguments.
First, I will address the idea that an event variable is “inserted” by the auxiliary before
moving on to address the status of passives with regards to their eventivity. The auxiliary
that Felser assumes to be inserted as a carrier of the progressive morphology, is not “inserted”
in the way she claims. The auxiliary in italics (at least in 50b) must be in passP and not
AspP and therefore is not evidence of an “inserted” event variable. Under the common
assumptions about English auxiliaries the head that being appears on must always be there.

78

So it is not the case that a head is inserted. Further, if we take my proposal that auxiliary
“insertion” is only the result of agreement with a higher auxiliary head or T then this is
reduced to an agreement phenomenon. Specifically it is an agreement phenomenon which
only has implications for morphological component post spellout. Minimally, her claim is
merely indicative of the hierarchy of projections for English. Consider the following examples
with both passive and progressive morphology.
(51)

a. [T P The ball [T wasj [passP j [vP thrown ]]]]
b. [T P The ball [T wasj [aspP j [vP rolling ]]]]

passive
progressive

The morphological marking of both the passive and aspectual heads appears on the verbs
that they have respectively selected for, while the auxiliary has presumably moved to T to
check tense. If we assume this basic ordering of functional heads then a derivation involving
both the passive and the progressive should reveal their ordering with regards to each other,
(52).
(52) [T P The ball [T wasj [aspP j [passP being [vP thrown ]]]]]

passive progressive

The verb throw is inflected with the participial -en associated with the passive. Likewise,
the passive auxiliary be has been inflected for the progressive with -ing and the auxiliary
in the aspectual phrase has presumably moved to T for tense reasons. If morphology is
any indication, then the heads that auxiliaries appear on are not ‘inserted’ as carriers of
morphology, but rather computed derivationaly in a manner consistent with the C-selectional
features of the heads higher in the derivation.
Furthermore, I propose that if in fact event variables are found in AspP then any eventive
predicate would project one, and in turn I propose that Felser’s own diagnostics for assuming
the presence of an event variable can be applied to passivized direct perception complements
to show that they do indeed have an event variable.
This leads us to the second problematic assumption that Felser makes about the eventivity
of passive clauses. She claims that passive clauses are often thought of as being stative in
nature. One problem with this claim is that they easily admit the progressive as we saw
79

earlier and they do so in a normal, derivationaly consistent way. Felser herself also claims
that certain types of aspectual adverbs like frequently and gradually are licensed by the
aspectual phrase. Assuming this is the case, then the following sentences are indicative of
passive clauses with adverbs either in the specifier of, or adjoined to, AspP.
(53)

a. Mary was frequently seen in the office.
b. John was gradually seen in the fog.

It is also commonly noted that stage level predicates exhibit prepositional phrase ambiguities, while individual level predicates generally do not. Example (54) is ambiguous with
regards to which event happened from school. Either the seeing took place from the school,
or the shooting event took place from the school.
(54) Mary saw Bill shoot John from school.
Now contrast this with example (55), in which the main clause is stative and does not exhibit
an event/prepositional phrase ambiguity, instead from school can only modify the shooting
event or the DP John.
(55) Mary knew Bill shot John from school.
If passive clauses were less eventive than their active counterparts then we might expect that
passivization would disambiguate sentences like the one in (56) in AppE.
(56) Bill was seen shoot John from school.
However, both events can still be modified by the PP from school exactly as in (54). Both
interpretations are repeated below.
(57)

a. Someone watching from the school saw Bill shoot John.
b. Bill was at the school and someone saw him shoot John.

This means that passive clauses do not behave like individual level predicates in terms of
event and prepositional phrase ambiguities.
I have shown Felser’s key assumptions about the passive to be unfounded. Passive clauses
pass eventivity tests, including the ability to appear in the progressive and the ability to
80

admit adverbs that by Felser’s own work must originate in the specifier of AspP. Furthermore, passivized perception complements pattern with eventive predicates with regards to
event/PP ambiguities and not with stative predicates which do not exhibit the ambiguities.
So if the structure is not missing as Felser suggests, then the ban on matrix passivization
of direct PCs must be accounted for in another way. Fortunately evidence that the passive
interferes with temporal indexing can be gleaned from other research and it seems at this
point that we are left with the t-linking hypothesis of Bennis and Hoekstra. However, there
may be some reason to believe that the relationship that is required between the clauses
has to do with case checking. Suppose that the passive is an argument that receives a theta
role and supporesses the external argument as in (Bennis and Hoekstra, 1989/2004) but also
gets checked for case, (Baker et al., 1989; Hornstein et al., 2008). Further if we assume
that case is the realization of an unvalued tense feature on D (Pesetsky and Torrego, 2004),
then perhaps -en is receiving a t-index and interfering with the control relationship between
the event arguments. The catch would be that all of the arguments involving temporal
identification would have to be reduced to case checking. This clarifies how we might build
a connection between passive -en and the possibility that it alters the way that temporal
indexing of arguments proceeds in passive derivations. In fact, this line of inquiry leads to
the final account of the constraint on passivization in English perception complements.

3.3.3

Hornstein et al. (2008)

Hornstein, Martins, and Nunes (or henceforth HMN) make use of yet another mechanism
for deriving the syntax of perception complements and the constraint against the passive. In
a nutshell they argue that we can account for the syntax of perception and causation verbs
and their unique set of passivization constraints by assuming a multiple agree approach.
The concept of multiple agree means that a probe may agree with elements other than
the primary goal. As per normal agree, uninterpretable features may probe down into a
structure in search of a goal which contains one or a set of interpretable features. However,

81

a probe which exhibits multiple agree will agree with one or more sub-goals which contain a
non-exhaustive set of features contained in the probe. Or in English, multiple agree assumes
that one probe may trigger agreement with various items on the way to its goal until all
features on the probe head are satisfied. HMN assume that under multiple agree, a probe is
active until the maximal feature person or the full set of features are checked and the probe
is deactivated.
They combine this syntactic mechanism with the notion that perception complements
contain an embedded but extremely deficient T. Strictly speaking this is not a mono-clausal
approach to perception complements. As noted by Wurmbrand, any non-restructuring approach bears the burden of providing evidence that structures like the purported deficient T
exists in the first place. This final account of the syntax of perception complements varies
slightly from the last two in that the authors argue that the main relationship that is established between the two clauses is reducible to requirements on case checking. Temporal
requirements and information about the two clauses simply do not figure into the account and
it largely ignores the rich set of properties that perception complements have. In this sense,
HMN’s analysis is a purely syntactic account. However, their account shares the general
spirit of the previous accounts in that they propose that the passive morpheme is the culprit
responsible for the impossibility of establishing intra-clausal case checking relationship.
In the next few sections, I will discuss and review the basic assumptions that HMN make
about the nature of perception complements in Standard English, before moving on to a
discussion of their analysis.

3.3.3.1

The active and passive as derivationally related

HMN assume that the derivation of an active bare perception complement like the one in
(58a) is derivationally related to the form in example (58b) where *(to) indicates that to is
not optional.

82

(58) SAE
a. John saw Mary leave.
b. Mary was seen *(to) leave.
HMN assume that active bare complements and the infinitival-to passive complements
are related. They argue that the infinitival-to in (58b) appears as a repair strategy in
order to ‘rescue’ the derivation. They set aside other types of perception complements
assuming that they are full propositional CP complements. Keep in mind that this is a
departure from the analysis presented by Felser, Wurmbrand, and the analysis presented
in this chapter. I argued that Standard English has a set of non-restructuring perception
complements which contain both CP and TP layers and which allow matrix passivization
and a set of semi-functional restructuring complements which correspond to what we have
been calling direct perception complements. Hornstein et al. assume that all other non-bare
instances of perception complements are propositional and set them aside.

3.3.3.2

The role of case bearing elements

HMN’s argument reduces to the claim that passive -en morphology blocks the checking of
case on a severely deficient non-finite T in the embedded clause. In order for this to work
of course, both the passive -en and non-finite T must bear an uninterpretable case feature
which must be checked before spellout. The evidence for both of these claims and their
plausibility for English is scarce to say the least.
HMN, assume -en has uninterpretable case because of the association made between case
and agreement made in Kayne (1989). However, Kayne’s analysis is concerned with cases
where perfect participles in French agree with cliticized objects.
(59)

a.
b.
c.

Paul a repeint les chaises.
Paul has repainted the chairs
* Paul a repeintes les chaises.
Paul les a repeintes.

83

Example (59a) shows the required agreement facts for a non-cliticized object in French.
As you can see there is no agreement between the object and the verb in the context of
the auxiliary a ‘has’. However, when the object undergoes cliticization there is agreement
between the cliticized object les and the verb which is now inflected for feminine plural. The
assumption of phi and case features on the participle in English is supposed to follow from
agreement facts where the object clitic and verb agree in the context of the auxiliary and
participle form of the verb.
Ignoring the proposed evidence, there is precedent in the literature for the nominal nature
of the passive morphology. The reasoning is simple. Nominal arguments receive theta-roles
from the verb. So you can either assume that passive verbs are different and do not discharge
an external theta-role or, you can assume that all verbs are the same in terms of their event
structure and the passive morphology absorbs the theta-role of the external argument. If
the passive morphology can receive a theta-role then it must be nominal in nature. This is
similar to the line of reasoning put forward in Bennis and Hoekstra (1989/2004) to explain
why passive morphology seems to be different. HMN attempt to strengthen their argument
with phi-feature data from languages like Brazilian Portuguese in which the passive participle
inflects for number and gender. There is unfortunately no evidence like this in Standard
English as the passive morphology does not show any overt agreement morphology.
The claim that the non-finite T bears a case feature comes from Raposo (1987) which
shows that certain infinitive structures in Romance languages seem to inflect for various
agreement patterns. Specifically, he shows that inflected infinitives in Portuguese behave
like nominals and may only appear in positions where case is licensed. Nunes (1995) takes
the infinitive as nominal projection idea a step further and argues the same for English
infinitives from the historical record. Again, there is no overt evidence that Modern English
infinitives are case bearing projections.

84

3.3.3.3

The Case of the constraint against the passive

Keeping these assumptions in mind, consider the derivation of *The girls were seen leave
under HMN’s account step by step. Example (60) shows movement of the embedded DP
The girls to the specifier of the embedded T. Uninterpretable number on embedded T has
probed down the tree to the DP and agrees with it for plural. The EPP feature on embedded
T then requires movement of the DP to its specifier.
(60)

TP
DPi
T
The
girls


3
vP
T


 pl 
unum:pl
 epp  t
v
ucase
i
ucase
leave

These two operations are shown to be complete by strike-through and the appearance of the
corresponding feature where applicable. Since embedded T is non-finite and does not have
interpretable case, both case features remain unvalued.
Next we add the verb see and the passive participle -en to the derivation. I have relabeled
HMN’s passP as vP and omitted the passive auxiliary head for simplicity.
(61)

vP
v
-en
unum:pl
ucase

VP
V
see

TP
DPi
T
The girls
3
vP
T


 pl 
unum:pl
 epp  t
v
ucase
i
ucase
leave

The structure in example (61) shows the addition of the verb see along with the participle
-en. Note that at this point the uninterpretable number feature on -en has probed down the
tree and has been valued for plural by the interpretable number feature on the DP. Crucially,
85

all three case bearing elements remain unvalued for case. Below in example (62), the matrix
T is merged.
(62)

TP
T
T
upers:3
 unum:pl

 epp
nom


vP






v
-en
unum:pl
ucase:nom

VP
V
see

TP
DPi
T
The
girls


3
vP
T




unum:pl
pl
 epp  t
v
ucase:nom
i
ucase
leave

At this point in the derivation, uninterpretable person and number on T probe down the tree
passing over the participle to receive a value from the DP goal. Now, HMN argue that the
participle is valued for nominative case by virtue of having all of its phi-features previously
valued when the probe bearing interpretable nominative case passes through it. The crux of
the ban on passivized perception complements is that both the DP and non-finite T require
case checking via a probe/goal relationship in example (62). They are also both equidistant
from the matrix T probe.
HMN argue that though they are equidistant, non-finite T is passed over in terms of case
checking because it shares a valued number feature with the participle and this essentially
renders it invisible as a goal for the probe from matrix T. Because of this, the person feature
on the probe is valued and the probe values the goal for nominative case. Finally, the EPP
feature will trigger movement of the DP to spec,TP. The non-finite T remains unvalued for
case and the derivation crashes.
The derivation of the acceptable perception complement passive in Standard English
which includes the infinitival-to, for HMN, is the result of a repair strategy which is triggered
86

as last resort to save the derivation from crashing. They argue that to insertion is enough to
satisfy the unvalued case feature on non-finite T. This means that derivations like The girls
were seen to leave all have as their basic form the unacceptable *The girls were seen leave.
However, to-insertion is a viable last resort option and it rescues the derivation. Reconsider
the AppE data and note that if we take this account as it is, it makes no clear predictions
about the AppE data.
(63) AppE
a. John saw Mary hit Bill.
b. John saw Mary to hit Bill.
c. Mary was seen hit Bill.
d. Mary was seen to hit bill
The idea that Appalachian English would make use of this repair strategy in both active and
passive complements in this way is incoherent and raises more questions than it answers. For
example, what do we make of an optional repair strategy? Moreover, if the repair strategy is
truly optional then what triggers it? Crucially, this problem disappears if we do not assume
the two are derivationally related. This is just another reason to treat the complements of
perception complements separately and use Wurmbrand’s classification in conjunction with
an HMN-style analysis which solves the seemingly special behavior of the passive participle.

3.3.4

Summary of previous analyses

In this section, I have discussed three approaches to understanding the ban on matrix passivization in perception complements. Though all three approaches assume different syntactic mechanisms underlying the behavior of perception complements, they share a common
thread. They all assume that the passive is in some way responsible for the ban on matrix
passivization. For Bennis and Hoekstra (1989/2004) the ban on matrix passivization is tied
to the passive participle and its status as a non-t-linking element. Thus the passive participle
interrupts the establishment of a tense domain inclusive of the untensed embedded clause.
The primary problem with this analysis lies in finding some principled reason that the par87

ticiple should interupt t-linking. Beyong this consideration, it might simply be updated to
current agree based, probe/goal syntax much like the one found in Hornstein et al.. Felser
(1999) argues that passive participles lack an event variable and hence interrupt control
relations between matrix and complement clauses in perception verbs. Finally, Hornstein
et al. (2008) express the special status of the passive participle and its relationship to perception complements in a multiple agree case based approach to establishing what is essentially
clause union. Under this approach, the participle is argued to be a nominal element which
contains features such that it is an intervener in the probe/goal relationships between T and
other nominal elements in the clause.

3.4

Analysis of AppE perception complements

I will argue for a syntax of perception complements using the syntactic framework of Hornstein et al. (2008). Passivization of direct PCs in AppE are admissible because -en does
not block case checking in AppE. Specifically, passive -en lacks phi-features. Ultimately, I
remain agnostic on the nature of the features that need to be established between the matrix
and embedded clauses.
I will however assume that perception complements come in restructuring and nonrestructuring flavors and may be classified as in Felser (1999) and Wurmbrand (2001). That
is, I will not assume that the bare and non-bare perception complements are derivationally related or related by the last resort process of to-insertion. I intend a hybrid analysis
where we assume Wurmbrandian style complementation but also show that the passive participle behaves differently in Appalachian English. This would reveal how passivization of
perception complements proceed in a language like AppE while thoroughly illuminating the
particular structures involved in these complements. It will also require a major detour into
the morphosyntax of participles in AppE as well as the mechanisms and consequences of
participle leveling. Finally, once I have argued for a particular syntax of these clauses, I will
compare it to the syntax that I proposed for causative and experiencer have in order to show
88

that mechanisms of restructuring might not only be helpful but necessary to distinguish
between types of restructuring clauses. In fact we will see that according to the diagnostics
used here, there are virtually no discernible differences in the structure of the previous have
complements and perception complements in AppE. Thus, I will posit head movement as a
restructuring mechanism.

3.4.1

Reiteration of theoretical assumptions

3.4.1.1

The classification of perception complements in AppE

Bare perception complements in Appalachian English are lexical restructuring predicates.
Following Wurmbrand’s classification, lexical restructuring predicates are those predicates
which allow matrix passivization. They are base generated as lexical verbs in V. Semantically,
they always allow a direct perception interpretation.
(64)

a. Mary saw John drink all the whiskey.
b. John was seen drink all the whiskey.

active
passive

Being lexical restructuring complements, these bare perception complements lack any properties associated with Tense. They are shown in the active for simplicity.
(65)

a.
b.
c.
d.
e.

* Mary
* Mary
* Mary
* Mary
* Mary

saw
saw
saw
saw
saw

John not drink whiskey.
John might hit Bill.
John have drank the whiskey.
John be drinking the whiskey.
there be a problem.

sentential negation
modal auxiliaries
aspectual have
aspectual be
expletive subjects

We can contrast these findings with the fact that Appalachian English also allows the nonrestructuring forms to appear with infinitival-to in the active.
(66) Mary saw John to have drank all the whiskey.
I would like remind the reader that SAE allows these in the passive in (67a) and in highly
restricted cases of certain expletive-there complements in (67b). However since the main

89

goal of this work is the syntax of restructuring, I will set these non-restructuring cases aside
in both dialects.
(67) SAE
a. Mary was seen to leave to party around midnight.
b. We saw there to be a need for more bourbon.
We know that perception complements with infinitival-to are structurally distinct because
properties normally associated with T which are banned in direct PCs are admissible in
indirect complements in AppE. Thus, infinitival-to in these cases is actually representative
of a tense layer and are therefore non-restructuring.

3.4.2

The restructuring verb and its complement

It should be relatively clear by now that because bare perception complements in AppE may
undergo matrix passivization they are lexical restructuring predicates and are generated in
V.
I will also assume that complements of direct perception verbs are maximally headed by
an aspectual phrase, AspP as indicated by the presence of progressive morphology.
(68) SAE
a. We saw [AspP John draw a circle.]
b. We saw [AspP John drawing a circle.]

–progressive
+progressive

Keep in mind that according to the analysis of auxiliaries laid out in the last chapter, the
auxiliary head responsible for assigning progressive morphology is only overt in the case that
it has agreed with some higher head. Further agreement with that higher head has nothing
to do with its ability to assign morphology to lower heads.

3.4.3

Derivation of lexical restructuring perception complements in AppE

As predicted by HMN’s analysis, if passive -en does not bear the same (or any) φ-features
as embedded T, then the derivation can converge. Below, I argue just that. I will provide
90

an analysis which assumes that the passive participle in AppE has no φ-features that might
interrupt probe/goal relations. In the section immediately following, I discuss evidence from
Appalachian English which is consistent with such assumptions. Consider the derivation of
The girls were seen leave.
I have implemented the assumption that these complements are headed by AspP below
and will continue to use agreement between case features as the driving mechanism although I
am agnostic about the exact nature of the features involved. The main line of my argument
may be implemented using the probe/goal machinery of Hornstein et al., or in terms of
Bennis and Hoekstra’s t-linking. I have chosen to use the probe/goal system Hornstein et al.
in spite of the weak claims about the nominal nature of the passive -en morpheme and the
embedded T, mainly because it is more widely recognized and more technically fleshed out.
As before the uninterpretable feature on Asp0 will probe down and find a value on the DP
before the DP moves to spec,Asp to satisfy the EPP feature.
(69)

AspP
DPi
Asp
The girls
3
vP
 Asp 
 pl 
unum:pl
 epp  t
v
ucase
i
leave
ucase

Next we will add verb see, the passive vP, and the passive auxiliary head to the derivation.
I have abstracted away from the inflectional features found on the passive auxiliary and v in
order to highlight the case relations.

91

(70)

PassP
vP

Pass
passive
v
∅

VP
AspP

V
see

DPi
Asp
The
girls


3
vP
 Asp 
 pl 
unum:pl
 epp  t
v
ucase
i
leave
ucase
The structure in example (70) shows the addition of the verb see along with the participleless passive vP. Crucially for AppE, the passivized vP has none of the relevant features and
both case bearing elements remain unvalued. Below in example (71), matrix T is merged.
(71)

TP
T
T
upers:3
 unum:pl

 epp
nom


PassP






vP

Pass
passive
v
∅

VP
V
see

AspP
DPi
Asp
The girls
3
vP
 Asp 
 pl 
unum:pl
 epp  t
v
ucase
i
leave
ucase

Upon merger of matrix T, the uninterpretable features will probe down the tree and since
both the case-bearing DP and AspP are equidistant, the probe will value both uninterpretable
case features there while simultaneously receiving a value for person and number. The EPP

92

feature on T will then trigger movement of the DP to spec,TP. The derivation converges.

3.4.4

The status of participle morphology in AppE

Since all three of the major works reviewed here and my own analysis of perception complements rely on the passive participle either having or not having a particular set of features,
it is again necessary to delve into the morphosyntax of verbal participles in Appalachian
English. What is the evidence that passive participles in AppE are different from those
in SAE? Recall that for Bennis and Hoekstra the passive participle had special status in
English and Dutch because it was responsible for bearing the external argument of the verb.
In Dutch the passive participle is required for felicitous interpretation of passive contexts in
a way that the past participle, as evidenced by IPP effects, is not. They argue that this
difference between the participles is the result of some ‘principle of recoverability of deleted
material’. Next, recall that Felser argued that the passive participle prevented matrix passivization in perception complements because they are individual-level predicates and failed
to project an event variable to act as a controller for the embedded event variable. A claim
that, as far as I can tell, simply does not hold. Finally in Hornstein et al., we saw an account
where the participle morphology was argued to be nominal enough in nature that it bears
an uninterpretable case and phi-feature which interferes with the case checking of elements
further down the tree. Based solely on these three analyses it is clear that, if there were
differences in the properties of the participle in a given dialect then we might expect that
passivized perception complements would be acceptable.

3.4.5

Examples of leveling in AppE

It just so happens that the participle in irregular verbs in Appalachian English has undergone
severe paradigm leveling (see Wolfram and Christian, 1976). Morphological forms of the bare,
past, and participle that were once distinct in the Standard Dialect have come to be indistinct
in AppE. Specifically, there no longer seems to be a three way distinction; irregular forms
93

now only maximally exhibit a two way distinction. In the next few paragraphs, I provide
examples of the types of leveling which can be found in Appalachian English in order to show
just why there is no evidence to support a syntactic distinction between past and participle
form of the verb. Such distinctions are only made periphrastically with auxiliaries. Thus
passivization is only signified overtly by the passive auxiliary head, though it is most likely
still partially semantically encoded on v.
In Standard English the verb eat has the following forms in the present (bare form), past,
and past participle.
(72)

inflection

SAE verb

bare
past
participle

eat
ate
eaten

Much like the verb hit has only one form in Standard English, an extended set of verbs in
Appalachian English have leveled in this manner (Wolfram and Christian, 1976), losing the
distinction between bare, past, and participle forms. They are all felicitously represented as
the same form. This is what is meant by paradigm leveling.
(73)

inflection

AppE verb

bare
past
participle

eat
eat
eat

Just to be clear, it is possible to use one form for the verb eat in Appalachian English
regardless of the normal inflection pattern in more standard varieties.
(74)

a. We eat cornbread and soupbeans every Thursday.
b. We eat cornbread and soupbeans two hours ago.
c. We had eat all the cornbread and soupbeans before y’all got here.

present
past
participle

Leveling of this type, where the base form of the word appears unchanged is not the only
type that is found in the dialect. There are forms which have been leveled from participle
to past in example (75a) and from past to participle as in example (75b). It is these forms
that are most relevant to the discussion at hand.
94

(75)

a. That kind of bottle has broke on more than one occasion.
b. They seen each other yesterday at the party.

It should also be noted that the participle can be leveled in both passive and perfect contexts
in (76a) and (76b), as well as adjectival contexts in (76c).
(76)

3.4.5.1

a. The bottle was broke over my head.
b. The bottle had broke before it fell.
c. The broke bottle was on the floor.

Participle forms in AppE are not distinct

Thus for Appalachian English, the participle morphology on irregular verbs is not distinct.
Moreover, it seems to be optional in all of the cases that it should not be, if it was as important
as a ‘principle of recoverability of deleted material’ would have us believe. In fact, AppE
blatantly violates such a principle in a way that SAE does not. That is, on irregular verbs
neither participle seems to have any special status and neither is required for the derivation
and proper interpretation of the sentences to converge. This has serious implications for the
accounts of Bennis and Hoekstra and Hornstein et al.. Although SAE certainly has verbs
like hit which exemplify a similar lack of morphological forms, it differs from AppE in that
it is not the case that all irregular verbs may be leveled. For instance in SAE, it seems as
thought a majority of irregular forms maintain the distinction of the participle, especially in
passive contexts. The fact that there are irregular verbs in SAE which do not level, requires
that distinct lists of vocabulary items, with distinct vocabulary insertion rules, which map
to distinct feature sets on syntactic heads are maintained. As we will see below, with no
such requirements in AppE, there is no reason to posit distinguishing features associated
with the passive morpheme in the syntax or in any corresponding vocabulary lists in AppE.
Assuming an account like Bennis and Hoekstra (1989/2004) for AppE, we are forced into
saying that passive morphology has no adverse effect on T-linking because the participles
have no special status. As for what the specific status of the participle forms in AppE might

95

be, I argue that they are either minimally optional, resulting from variable leveling or they
simply do not exist in the dialect and any appearance is the result of second language contact
with the standard dialect. Either way, I argue that this is evidence that we can not assume
that the participle morpheme plays a special role in the absorption of case or bearing of
external theta-roles in AppE.
In terms of HMN, participle morphology on irregular verbs in AppE is not distinct form
past morphology and has no phi-features which require a distinct participle form, something
which the analysis requires if it is to hold. The participles should have no uninterpretable
features associated with case and thus passive morphology in AppE never interferes with case
agreement between a probe and its goal. This falls out of Hornstein et al.’s own assumptions
about the tight co-occurrence of case and phi-features found in Chomsky (2001).
In the next section, I describe in more detail the participle system of Appalachian English
in from the point of view of Distributed Morphology in order to clarify just how Appalachian
verbal morphology differs from the Standard Dialect.

3.4.6

The features of participles in AppE

With the basic analysis in place we are now in a position to inquire more closely into to
the nature of the participle morphology in Appalachian English. Lets return to the basic
paradigm for participle morphology in the English auxiliary system (Halle and Marantz,
1993) and be explicit about the nature of the participle system in AppE. Again, it is crucial
to keep in mind that [+past] here is not meant in the sense of tense but is used as a cover
term for the relevant morphological feature in the syntax that provides the distinction.

96

(77) v = fused Auxiliary/T and v
[+participle, +past]

↔

[+past]

↔

[+past]

↔

[+past]
[+participle]

↔
↔
↔

/-n/ / X +
where X = hew, go, beat, . . .
∅/Y+
where Y = beat, drive, bind, sing, . . .
/-t/ / Z +
where Z = dwell, buy, send, . . .
/-d/
/-ing/
∅

In terms of phonological features, notice that the only distinguishing feature between
any of the past morphology only exists for the set of -en participles. For Halle and Marantz
this distinguishing feature is [+participle] in addition to [+past]. The other distinctions are
handled by separate lists. Recall from the last chapter our assumptions about how auxiliary
morphology is assigned. After all syntactic operations are complete the relevant chunk is sent
to PF for morpho-phonological operations. This assumes that the verb has moved to v in the
syntax and at the level of PF we assumed that the morphology assigning head immediately
above it is merged with v in a process of morphological merger. Here specifically, I again
assume it is a type of morphological lowering.
The past and participle form of a verb like broke in Standard English has the following
configuration for vocabulary insertion post syntax.
(79) broken

(78) broke

v

v
v
break

v
break

T
+past

Pass
+past
+participle

Example (78) shows that only the [+past] feature is necessary for the insertion of broke as
opposed to break and example (79) shows that the feature bundle [+past, +participle] allows
for the insertion of broken.

97

There are several different possibilities for the locus of variation in Appalachian English
at this point. I argue that passive and perfect heads in Appalachian English syntax only
contribute a [+past] feature to the derivation. This assumption is minimal and conservative considering the leveling data, which indicates the lack of a distinct passive or perfect
morpheme. Consider the morphemes of the passive and perfect heads below.
(80) Passive Head

(81) Perfect head

PassP

PerfP

Pass

Perf
+past

+past

This forces the notion that any -en form participle morphology in the speech of Appalachian
speakers is non-native and the result of dialect contact with the Standard. Again, such distinctions are not available from the syntax of AppE. Put simply, I would argue that speakers
who have the participle forms have acquired morphological doublets of the passive and perfect heads which contain the [+participle] feature of the standard variety. These doublets
would contain the extra feature and undergo morphological merger as in example (79). I
want to be clear however, that such acquisitions need only be superficial in nature.
While this appears to be a straightforward process for a verb like break which has undergone leveling to the preterite form, we need to understand what has happened to verbs
which level to the participle. Note that see in Appalachian English is just such a verb and
has undergone leveling to the participle form as in I seen him yesterday. To handle this case
we just need to assume that the form seen and others like it belong to a different list for
Appalachian speakers. It is entirely possible that they have replace the preterite form saw.
This is not as bold a move as one might first think. If all participle forms of irregular lexical
verbs exhibit leveling, this questions the need for a specific participle list in AppE at all.
Consider the revised set of vocabulary insertion rules.

98

(82) Revised Insertion Rules for AppE
↔ ∅/Y+
where Y = beat, drive, bind, sing, . . .
[+past]
↔ /-t/ / Z +
where Z = dwell, buy, send, . . .
[+past]
↔ /-d/
[+participle] ↔ /-ing/
↔ ∅
[+past]

Notice that have removed the following rule from the list.
(83) [+participle, +past] ↔ /-n/ / X +
where X = hew, go, beat, . . .
We can remove the rule because it is completely unnecessary. All verbs like hew, go, and
beat have either leveled or have been regularized. Because of this fact it is plausible to
assume that Appalachian speakers have redistributed the items on this list to other lists.
For example, hew has been regularized while verbs like go have been leveled to the participle
and verbs like beat have been leveled to the preterit.
(84)

a. His head was hewed clear off.
b. He had went into town.
c. He was beat within an inch of his life.

In, this section I have argued for a particular structure of the Appalachian English lexicon
in terms of what heads the syntax has access to and one possibility for what the feature
content of those heads might actually be. I took a strict view of the AppE lexicon which
said that speakers of AppE simply do not have separate vocabulary insertion rules for the past
participle and that the perfect and past heads which assign such morphology are different.
It is possible that they are morphological doublets with those in Standard English and this
difference is grammaticalized in the lexicon of AppE speakers. Thus, morphologically distinct
past participle forms in Appalachian English do not exist. The expression of semantic notions
associated with the perfect and passive participles are only expressed on the perfect or passive
auxiliary head. This means that neither the Appalachian English leveled past/participle
forms nor any standard morphological doublets acquired by AppE speakers need be any
99

more than superficially related to participle forms in the grammar of Standard English in
terms of underlying syntaco-semantic features like case. Though the standard participle
forms in English may have some special status in the form of specific features, this need not
be the case for Appalachian English. Thus apparent participle forms in AppE are, depending
on your favorite analysis, not non-t-linking elements or bearers of phi-features that might
interrupt case checking.
Before moving on to compare the final structure for causative and experiencer have to the
structure proposed for perception complements in Appalachian English, I would like to take
a moment to comment on the assumed directionality of leveling in non-standard dialects.

3.4.7

Impoverishment and presumptions about the grammatical baseline

The presence of variably leveled forms in the grammar requires variation in the input. I
speculate that the creation of morphological doublets like the ones described above begins
with a process similar to variable leveling. Variable leveling is said to apply in cases where
speakers variably level from standard forms to dialect forms. In fact, most linguists assume
variable leveling from standard to dialect as the basic state of affairs in the grammar of
non-standard dialects like Appalachian English. The point of this section is to question and
raise awareness about the directionality of leveling. The main problem with this assumption presupposes that children acquiring Appalachian English are aware of the standard or
that standard forms are salient enough in the input to be encoded. However, If we assume
that child language acquisition is a more conservative process, then children might discard
information in the input that is too variable or not salient enough. Awareness and integration of standard dialect forms into the grammar may come later rather than earlier in the
acquisition process.
In order to illustrate the directionality problem, I will describe one mechanism which
has been proposed to handle variable leveling and illustrate just how it makes this standard
dialect centric assumption. I will then provide a rough sketch of an alternative mechanism

100

which assumes the dialect form itself as a baseline. It also assumes that any standard forms
are added to the grammar later through an opposite process that I have termed morphological
enrichment.

3.4.7.1

Leveling via Impoverishment

Variable leveling has been modeled in the framework of Distributed Morphology by variable
impoverishment rules (Nevins and Parrott, 2007; 2010). These rules are post-syntactic morphological rules which apply with variable probability to delete morphological features on
heads. This allows the insertion of less specific forms. If we take the variation in AppE past
participles as an example, an impoverishment theory of leveling would proceed as follows.
The syntactic spell out of the sentence The vase was broke by the children is identical to
Standard English. Remember that we are interested in how the standard broken becomes
broke. Under impoverishment, everything proceeds as usual and at the level of phonological
form we are left with the output of a standard syntax of english shown in example (85).
(85) broken
v
v
break

Pass
+past
+participle

However, the morphology of the dialect also contains a variable rule shown in example (88).
The rule states that the feature [+participle] is deleted in the case where it appear in the
specific context of the feature bundle [+participle, +past], (Munn, 2013).
(86) Impoverishment rule for participle leveling in AppE
[+participle] → ∅ / [+participle, +past]
Applied variably, this rule would replicate the production of variably leveled forms in the
dialect. When applied, the only remaining feature would be [+past]. So, in the case of a
verb like break, broke and not broken would be selected from the list of vocabulary items and

101

inserted. While this system works if we assume that speakers have native speaker knowledge
of standard paradigms, there is no reason to assume that they do. Ultimately, it is an
empirical question.
On the other hand, one could just as easily imagine a process of variable enrichment for
speakers of non-standard dialects. It would handle the same cases as variable impoverishment
but assume a different set of ‘baseline’ features in the syntax. If a sentence like The vase
was broken by the children was produced by a speaker of a severely leveled dialect, we could
assume that morphological enrichment is at work. It could even apply variably. Post syntax,
the insertion site for the verb would have the features in example (87).
(87) broke
v
v
break

T
+past

Assuming a conservative representation of their grammar we might assume that dialect
speakers might represent such ‘special’ standard forms with a cover feature φ that refers to
a special standard vocabulary list and ensures the more feature specific form broken.
(88) Enrichment rule for participles in AppE
[+past] → [φ, +past] / [+past]
The vocabulary insertion rule for more standard -en forms like broken would be as follows.
(89) [φ, +past] ↔ /-n/ / X +
where X = hew, go, beat, . . .
In summary, I argued that assumptions about the nature of the grammar and how grammars vary has led to some standard-centric assumptions about how variation in grammars is
represented and derived. Depending on the dialect specific condition, children might acquire
morphological doublets as per my analysis, they might acquire all of the features of the
standard only to delete them via impoverishment, or they might only acquire the features
they can depend on in their dialect and tack on the standard later.
102

3.5

What’s the difference? Causative have and perception complements

The complements of causative and experiencer have on one hand and perception complements
on the other share more syntactic properties than not. We saw that both have complements
and perception complements are maximally headed by an AspP and both sets of complements
pass similar tests for showing that they lack elements associated with embedded T. They
differ in two ways. First, causative and experiencer have are generated in v and their
complements are classifiable as semi-functional restructuring complements, while perception
verbs are generated in V and their complements are classifiable as lexical restructuring
complements. To my knowledge this distinction in English hinges solely on whether or not
matrix passivization is allowed. Secondly, the have complements may optionally contain a
default to and the perception complements can not.
(90)

a.
b.
c.
d.

John had Mary to wash the dishes.
John had Mary to leave on him.
John saw Mary (*to) leave the party.
Mary was seen (*to) leave the party.

If the complement clauses of these verbs have identical syntaxes then why should there be a
difference in the presence or absence of the default to. After all, I argued that the insertion of
the to is a post syntactic morphological rule which inserts to on an auxiliary head which does
not enter into agreement with higher auxiliary heads. The real question is, what prevents
insertion of default to in the perception complement cases in examples (90c) and (90d).
I argue that this is evidence the classification system proposed by Wurmbrand for restructuring complements breaks down for English. Specifically, this is direct evidence that
external mechanisms such as head movement or agree are necessary components of restructuring in addition to the differential clause size approach of Wurmbrand.
I would argue that the different temporal and semantic constraints on these verbs coupled
with their respective syntactic categories is the cause of this difference. In bare perception

103

complements default to does not appear because Asp0 head moves into the verb to ensure
temporal simultaneity between clauses.
(91)
VP
V
AspP

V
Aspk see

Asp
tk vP
...

While the same fact does not hold for the Appalachian English version of causative and
experiencer have. Causation complements differ minimally in that they do not require event
overlap.
(92) Last Wednesday, John had Bill to buy him a ticket this Thursday.
Thus, head movement of Asp0 to v is not necessary and to may be inserted as the default
on the unmoved aspectual head.
(93)

vP
v
v
had

AspP
Asp
Asp
to

...

104

3.6

Conclusion

In this chapter, I showed that Appalachian English perception complements are best analyzed according to Wurmbrand’s lexical restructuring predicate classification. Thus the
lexical restructuring verb itself is base generated in V. Their complements ban most things
associated with T and are maximally headed by an aspectual phrase. This classification also
separates them from the class of semi-functional restructuring predicates like causative and
experiencer have which do not undergo matrix passivization. We reviewed three different
accounts of the syntax of perception complements which sought to explain the ban on matrix
passivization of perception complements in Standard American English. While one suffered
from some empirical issues, two remain viable analyses. Though the ban on matrix passivization of perception complements falls out of Wurmbrand’s classification for ‘free’, it is
not particularly explanatory. For that reason we focused on an analysis of the morphosyntax
of the passive in Appalachian English. It is the one element crucial to all three previous analyses of the ban on passivization. I showed that the heads which license passive and perfect
in Appalachian English are different from those of Standard English based on the fact that
the dialect has undergone extreme leveling. As such, we would predict that Appalachian
English would behave in exactly the way that is does by allowing passivization of perception
complements under those two analyses which remained viable. With the analysis in place, I
explicitly outlined the morphosyntax of the AppE past participle (or lack thereof) and raised
the question of ‘standard dialect centrism’. Given that at least one recent analysis of leveling
via variable impoverishment rules assumes that leveling is leveling from the standard form
to the dialect form (Nevins and Parrott, 2007; 2010), it is important to realize that such
assumed baselines are just that, assumptions. The answer to such questions of course lies
in empirical work done in the field. Finally, I compared the proposed syntax of causative
and experiencer have to that of perception complements in this chapter. We saw that the
two types of complement clauses appear virtually identical except in the application of a
default insertion rule and this requires that we rethink Wurmbrand’s differential clause size
105

only approach. I proposed that head-movement motivated by semantic constraints as one
way we might differentiate these clauses.

106

CHAPTER 4
INFINITIVAL CONTRACTION AS RESTRUCTURING

4.1

Introduction

Appalachian English makes use of restructuring complements not only in perception and
causation complements but also in describing approximative events, or events which come
close to happening but are never instantiated. Example (1a) shows that Standard English
(as well as AppE) expresses a similar meaning with adverbial almost. The Appalachian
English approximative liketa is listed below this in example (1b).
(1)

a. John had almost punched Bill before you arrived.
b. John had liketa punched Bill before you arrived.

Keep in mind that the meaning of liketa is similar to that of almost in Standard English
and example (1b) can be paraphrased as John came close to punching Bill last night but the
did not punch Bill last night.
Now consider the non-contracted form of liketa which has the uncontracted structure in
(2).
(2) John had liked to have punched Bill before you arrived.
For all intents and purposes the meaning is the same in the contracted and uncontracted
form. I take the difference in normally paced speech to be in the reduction of the high back
vowel [u] of infinitival-to or [tu] in the non-contracted form to schwa in the contracted form
[lAkt@]. Both forms require that the embedded clause is marked for perfect aspect. In the
uncontracted form, perfect have and the past participle morphology on the verb are required.
(3)

a.
b.
c.

John had liked to have punched Bill before you arrived.
* John had liked to have punch Bill before you arrived.
* John had liked to punched Bill before you arrived.

107

Example (3b) shows us that the past participle is required, while example (3c) shows us that
the auxiliary have is required.
In the contracted form only the participle morphology is present on the embedded verb
and it is required. Example (4b) shows that omission of the past morphology results in
unacceptability.
(4)

a.
b.

John had liketa punched Bill before you arrived.
* John had liketa punch Bill before you arrived.

The perfect auxiliary is not licensed as a standalone lexical item in the embedded clause of
the contracted form.
(5)

* John had liketa have punched Bill before you arrived.

This is especially interesting considering that no such restriction holds of want contraction
where want to is contracted to wanna.
(6) I wanna have eaten before they arrive.
In this chapter, I will present a syntax for the contracted and uncontracted forms of
liketa. Contracted liketa is a restructuring verb and the constraint on the embedded have
can be derived from common assumptions about restructuring properties in conjunction with
aspectual constraints on the forms which possibly stem from the semantics of the past participle. The analysis provides evidence for the importance of taking semantic requirements
into account when trying to determine the nature of restructuring predicates. I would argue in general that this is important whether they are the unique semantic constraints of
perception, causation, or approximativity and whether or not the verbs are formed from
contractions or not.
This chapter has the following structure. In what follows, I explore the syntax of both
the contracted and uncontracted forms of liketa via observations made primarily in the sociolinguistic literature. I will supplement these observations with native speaker acceptability
judgments. Since, liketa is an understudied form with extremely complex semantic interpre108

tations, I will describe those interpretations as clearly as possible and in a bit more detail
than the verbs in previous chapters. This is done by comparing interpretations about liketa
to existing research on the interpretations of almost. As for the syntax of liketa, I argue that
it is a verb rather than an adverb based on the fact that liketa seems to have more verb-like
properties than adverbial properties. In addition, I provide evidence that liketa is a raising
verb. I briefly discuss the structure of uncontracted liketa and propose a syntax for it. With
these basics covered, I review a head movement analysis of contractions as restructuring
predicates and extend it to contracted liketa. Finally, I treat liketa in the now familiar style
of Wurmbrand and advance a new analysis that speaks to the assumptions made in previous
chapters about the morphosyntax of verbal morphology in restructuring complements. I
show that we can make sense of the relationship between the two forms and the ban on overt
auxiliaries in terms of either of the two restructuring analyses. Finally, I compare the two
analyses with each other in an attempt to make broader theoretical headway for theories of
restructuring.

4.2

Previous observations

In this section, I discuss the few observations that have been made about liketa beginning
with Walt Wolfram and Donna Christian’s observations from Appalachian English in West
Virginia and moving outwards to observations about liketa in African American Vernacular
English of New York from Labov, and Corky Feagin’s observations from Southern English in
Alabama. Lastly, I present some observations gleaned from various written language corpora.

4.2.1

Observations about liketa in Appalachian English

Wolfram and Christian (1976) discuss some usages and constraints on liketa as it occurred in
sociolinguistic interviews carried out in Mercer and Monroe counties in southwestern West
Virginia. They note that liketa occurs in both positive and negative contexts.

109

(7) (Wolfram and Christian, 1976)
a. And I knew what I’d done and boy it liketa scared me to death.
b. That thing looked exactly like a real mouse and I liketa went through the roof.
c. When we got there, we liketa never got waited on.
d. I liketa never went to sleep that night.
Wolfram and Christian note that the only negative element that occurs under the scope of
liketa is never as shown in examples (7c) and (7d). I take this as evidence that the truth
value of a proposition under liketa can be negated.
They also note that the past form of the verb is required, as I mentioned in the introduction, but they assume that the verb embedded under liketa is the only tense bearing element
in the clause. This can only mean that they assume that it is receiving tense information
from T. They also note that there were no instances of liketa appearing in question form or
in embedded clauses.
Similarly, liketa in African American English of New York does not take tense marking
or undergo inversion in yes/no questions (Labov, 1972). For these reasons, the Wolfram and
Christian assume that liketa is an adverb. Consider the following evidence.
(8)

a.
b.

* John had liketa-ed punched Bill before you arrived.
* Liketa John had punched Bill before you arrived.

Finally, Wolfram and Christian note that liketa is said to resemble almost only in the
cases where almost scopes over the entire proposition. Example (9a) is an example in which
almost scopes over the proposition of John building a chair.
(9)

a. John almost built a chair but he didn’t even start.
b. John liketa built a chair but he didn’t even start.

Liketa shares this interpretation, and it can be paraphrased as John came close to starting
to build a chair but he didn’t start. However, the interpretation available below in example (10a) is one where John has begun but not finished building a chair; an interpretation
where almost is not scoping over the entire proposition. This interpretation is not available
in example (10b) with liketa. The symbol # expresses an infelicitous interpretation.
110

(10)

a.
John had almost built a chair but he hadn’t finished it before we arrived.
b. # John had liketa built a chair but he hadn’t finished it before we arrived.

In summary, the only negation allowed under liketa is the negative time adverbial never.
The verb under liketa must be marked for past and there are no observed instances of liketa
appearing in questions or embedded contexts. Labov noted that liketa itself also does not
show overt tense marking or invert in questions, ruling it out as an auxiliary verb form.
Finally, liketa only means almost in cases where almost scopes over the entire proposition.

4.2.2

Liketa in Alabama English

Feagin (1979) offers and extremely detailed sketch of the occurrence of liketa in the Southern English of Anniston, Alabama. Feagin suggests that liketa is derived either from the
transitive verb liken ‘to see, mention, or show as like or similar ’ or from the adjective like.
As noted for Appalachian English, Feagin’s liketa forms must also always appear with past
participle morphology on the embedded verb. Feagin also repeats the notion that liketa
never occurs in questions but adds that liketa never occurs in commands. Feagin suggests
that the form in example (11) is the underlying form of liketa in general.
(11) like to have V-ed
Though she makes no mention of contracted liketa versus uncontracted liked to themselves, she does explicitly mention fluctuation in the appearance of perfect have in the
embedded clause. She notes that the perfect auxiliary either appeared in the data in its full
form in (12a), a contracted form as in (12b), or deleted as in (12c).
(12)

a. She liketa have died!
b. An’ that just liketa’ve killed him.
c. She liketa killed me! She was so mad!

(Diane B. W15:38.II.120)
(Virginia L. U60)
(Diane B. W15:38.II.120)

As far as the observation that liketa in Alabama English occurs in both positive and
negative contexts, Feagin observes that liketa occurred with the negative not one time (13a)
and with the negative adverbial never a total of four times in examples (13b) to (13d).
111

(13)

a.
b.
c.
d.

They liketa not got any food or anything to ’em.
They liketa never git them needles up!
And I like never to have found her
You like never got material to fix it.

(Sam C. W70:34.I.236)
(Myrtice J. W62)
(Frances B. U59)
(Ella B. W:37.I.361)

Ultimately, I find the presence of full have in the complement of liketa in Alabama English
to be evidence of different trajectories of grammaticalization. We have to keep in mind that
the dialect spoken in Anniston, Alabama and the dialect spoken in Pike County, Kentucky
are not the same dialect even if they happen to share cognate lexical items. We should
expect them to behave differently. Aside from this, I believe an explanation of this form
might be found in Feagin’s data. Out of 70 total tokens of liketa only 7 are found with the
full auxiliary have. If we look at the social distribution of those 7 we see that all but two of
them were produced by teenage females. Of the remaining 2 one was produced by a female
in the range of 20 to 60 years old and the final token occurred in the speech of a female in the
60+ group. Thus it is possible that liketa with full auxiliary have was reintroduced into the
community as an innovation. I believe similar arguments could be made for the appearance
of the negative adverbial not. Recall that Wolfram and Christian found no occurrences of
not under liketa in Appalachian English.

4.2.3

Liketa in the historical record

There is a long historical record of the usage of liketa in various British and American
English corpora Kyt¨o and Romaine (2005). Kyt¨o and Romaine find that uncontracted liketa
constructions like have/had liked to + V, are found as far back as the mid-fifteenth century
in conditional if clauses.

112

(14) (Kyt¨o and Romaine, 2005)
a. by on of the mynysteris of the said Cathedrall Churche was sette afire, and began
to brenne, and yf hit hadde had his course lyke to have sette a fyre and
brende the cheif and grete parte of the citee (1447, Helsinki ME4, John
Shillingford, Letters and Papers of John Shillingford, p. 87).
b. ‘One of the minsters of the said Cathedral church was set afire and began to
burn, and if it had had its course, would have come close to setting a
fire and burning the chief and great part of the city’.
Though the main concern of their paper is one of determining the trajectory of grammaticalization of ‘constructions’ the Kyt¨o and Romaine take liketa to have its roots in adjectival like
with a meaning of imminent likelihood and probability. They also point out that throughout
the corpora under consideration uncontracted forms of liketa also appear alongside a nonperfect be like to + V which only variably has a counterfactual interpretation. All instances
of liketa which occur with perfect auxiliaries are counterfactual in nature. They have the
interpretation of almost.

4.3

Properties of liketa

Here I will discuss the semantic and syntactic properties of liketa. Because of the semantic
similarities that liketa shares with almost, the relative wealth of semantic work done on almost, and the lack of semantic work done on liketa, I will begin by comparing interpretations
of liketa to semantic interpretations of almost noted in the literature. Since it is probably the
best paraphrase for liketa it lends itself to being a logical starting point for understanding
the semantics of liketa. The result of this comparison is the finding that liketa only has a
subset of the interpretations available to almost. Then I will turn to syntactic properties of
the constructions where I will provide evidence that liketa is in fact a restructuring verb.
Liketa’s location in the main clause and the architecture of its complement will be compared
to the restructuring haves and perception verbs of the two previous chapters.

113

4.3.1
4.3.1.1

Semantic properties
Comparison with almost

Because there are basically no analyses of the semantics of liketa this section is largely an
endeavor to compare and contrast liketa and almost. In doing so, I hope to thoroughly
outline the ways in which interpretations of liketa differ from those of almost, in an exact a
manner as possible. In the formal semantic literature, lexical items like liketa and almost are
commonly termed approximatives. Approximatives are commonly assumed to have both a
proximal and polar component to their meanings (see Sadock, 1981; Rapp and von Stechow,
1999; Morzycki, 2001; Penka, 2006; Nouwen, 2006). The proximal or ‘closeness’ component
and the polar ‘negation’ component for the sentence in example (15) are given below.
(15) John liketa/almost punched Bill.
a. John came close to punching Bill.
b. John did not punch Bill.

proximal
polar

As you can see, the proximal component implies closeness or imminence of the eventuality
located under the approximative, while the polar component negates the proposition. Liketa
shares this two part conventional meaning with almost. However the possible interpretations
do not stop there. Its widely known that interpretations of approximatives involving these
two components give rise to various interpretations depending on what type of verb almost
modifies. For example, the verb kill is thought to be decomposable into ‘atomic’ components.
Rapp and von Stechow (1999) assume that an approximative like almost has at least the
following three interpretations based on scope possibilities in a decomposed verb following
McCawley (1971).
(16) Johns almost killed Harry
a. almost(ACT(John) CAUSE BECOME dead(Harry))
b. ACT(John) almost(CAUSE BECOME dead(Harry))
c. ACT(John) CAUSE BECOME almost(dead(Harry))

114

counterfactual
scalar
resultative

Each configuration of almost scoping between particular elements is thought to have a different interpretation. The verb is composed of has an agentive component or act, a causative
component or cause, and a change of state component labeled become. The last component was considered to be the core lexical content of the verb. The verb kill was thought have
the lexical semantic core meaning of die combined with the other atomic pieces mentioned
above. The interpretations which correspond to the counterfactual, scalar, and resultative
configurations shown above are paraphrased below for the phrase John almost killed Harry.
(17)

counterfactual: John almost does something that causes Harry to die.
scalar: John does something that almost causes Harry to die.
resultative: John does something that causes Harry to almost die.

If almost has wide scope with respect to act then it is said to have a counterfactual interpretation. Under the counterfactual interpretation it is generally the case that no action
is initiated by the agent of the sentence. When the approximative scopes just under act
it is said to give rise to scalar interpretation. Scalar interpretations are those where some
action was initiated by the agent of the clause. Finally, resultative interpretations arise when
the approximative scopes just under become. Resultative interpretations are those which
involve an agent which has initiated a change of state which fails to be manifested.
The atomic components of the decomposed verb kill, represented in (16), translate into
the following structure.
(18)

vP
act
cause
become

dead

Given common assumptions about theta role assignment the maximal projection of a decomposed agentive verb would roughly correspond to vP, where the external argument is
merged.

115

Rapp and von Stechow (1999) argue that the observations made in example (16) translate directly into the syntax of decomposed verbs which we represented in (18). Based on
evidence from German fast ‘almost’, Rapp and von Stechow (1999) find evidence for two
of the three interpretations mentioned by McCawley. They argue that the counterfactual
interpretation arises in German and English when fast or almost scopes above act and when
it scopes just under act, it gives rise to a scalar interpretation. The position of almost in example (19) corresponds to the counterfactual interpretation and the structural configuration
corresponding to the scalar interpretation is in example (20).
(19) counterfactual

AP
almost

vP
act
cause
become dead

(20) scalar
vP
act
AP
almost
cause
become dead
They argue that the resultative is not discernible from scope relations and verbal decomposition in their particular dialect of German, even though it is available in other dialects of
German and with English almost. Though it is not represented above, the position corresponding to the resultative reading is argued to be even lower in such structures. Thus for
Rapp and von Stechow, the difference between the counterfactual and scalar interpretations
comes down to whether fast scopes under or over act.
116

This finding allows us to make a theoretical prediction about liketa which might scale
up to the set of approximatives in general. If liketa is an adverb then there is no principled
reason that it should not give rise to the same set of readings as almost. However, if liketa
is a verb in a higher clause, as I will argue in this chapter, then we would expect the set
of readings that arise with liketa to be different from the interpretations of almost. I take
the determination of syntactic category of an element like liketa to be as straightforward
as comparing a list of properties associated with that element to properties we know are
associated with a given syntactic category.
For example, it is well known that adverbs are generally less restricted in their distribution
than verbs are. Compare the possible distribution of almost below to liketa.
(21)

a.
b.
c.
d.

John
John
John
* John

almost died.
died almost.
liketa died.
died liketa.

This data shows that liketa is not able to appear in different positions in the sentence.
Though this is certainly not a property of all adverbs in English it is certainly a hallmark
property of the majority of them.
Similarly, another commonly assumed property of adverbs is that they are licensed as
standalone responses to questions. Liketa is not a possible response to the question, while
almost is a possible response to the question.
(22) Did you finish your work?
a.
Almost
b. * Liketa
Although liketa is more constrained in the types of auxiliaries that are licensed on either
side of it, we can still learn about liketa’s syntactic category by looking at its distribution in
a hierarchy of projections, much like we did for causative and experiencer have. Notice that
in its uncontracted form, liketa can appear between two perfect auxiliaries.

117

(23)

a.
b.

John had like to have finished his work.
* John had almost have finished his work.

Almost is unacceptable in the same position.
The situation is only slightly different for the contracted form. Though the perfect is not
licensed as a stand alone lexical item, a pro-clitic corresponding to the perfect is licensed on
the verb.
(24)

a.
b.

John had liketa a’finished his work.
* John had almost a’finished his work.

Though a proper analysis of pro-cliticization of the perfect is beyond the scope of the
current work, it is worth mentioning that it actually supports rather than poses a problem
for the larger analysis proposed here. Recall the claim from the previous chapters, that
auxiliary insertion in English is the result of the morphology assigning head entering into an
agreement relationship with a higher inflection assigning head. I will argue in this case that
there is no such higher head which the embedded perfect head may enter into an agreement
relationship with. The head therefore would not undergo auxiliary insertion. However, it is
still possible for the features to become associated with the embedded verb and expressed
there as a bound morpheme if the appropriate phonological form pro-clitic form is available
in the lexicon.
Finally, liketa allows expletive subjects in its matrix clause and this indicates that it is a
raising verb.
(25)

a. There had liketa been a man killed in the mines today.
b. It had liketa rained.

Based on these observations, I propose that liketa is verbal and not adverbial. It can
not be moved like many adverbs and it is not a felicitous response to a yes/no question.
Further, liketa appears in a verbal position between two perfect auxiliary heads and it
licenses expletive subjects like a raising predicate.

118

Liketa as a verb is depicted in example (26). This structure constrains the distribution
of liketa to that of a matrix verb and assuming Rapp and von Stechow, it predicts that
liketa should only ever get a counterfactual interpretation. I have included an auxiliary
phrase above the verb shell as the assigner of participle morphology to more accurately
represent the distance between liketa and the embedded verbal shell. Explicit arguments for
the embedded perfect are presented later in the analysis.
(26)

VP
V
liketa

AuxP
vP

Aux
act

cause
become

X

Specifically, this is because there is no evidence that verbal elements in English may scope
inside the verbal shell of the verb in an embedded clause. In other words, under the assumptions presented here, there is no way for liketa to be interpreted as if it were located under
act with a scalar interpretation.

4.3.1.2

Semantic interpretations of liketa

Now that we have discussed at least one of the accounts of the structure and semantic interpretations of almost we can see just how liketa measures up to its more standard counterpart.
It is clear from the data in the introduction and from the observations made by Wolfram
and Christian that liketa is minimally interpretable as having a counterfactual interpretation.
The introductory data also indicates that liketa does not receive a resultative interpretation.
Examples (27a) and (27b) repeated from the introduction below show that a counterfactual
interpretation is possible while a resultative interpretation is not possible. Recall that the

119

symbol # marks an unavailable interpretation.
(27) John liketa built a chair.
a.
John came close to starting to build the chair but he did not start.
b. # John started building a chair but he didn’t finish it.
The truth or falsity of the prediction that I highlighted from Rapp and von Stechow’s
analysis hinges on acceptability of the scalar interpretation with liketa. Liketa resists their
analysis because it is acceptable with both the counterfactual and the scalar interpretation,
yet does not originate inside of the embedded verb. A liketa phrase and the acceptability of
all three interpretations with the incremental theme verb build are repeated below.
(28) John liketa built a chair.
a.
John came close to starting to build a chair but he did not start.
b.
John did something that came close to building a chair but he didn’t.
c. # John started building a chair but he didn’t finish it.

CF
SC
RS

This data shows that the verbal decomposition approach is not feasible with liketa. Liketa,
which does not scope under act in its complement, receives a counterfactual and a scalar
interpretation. This is not predicted by Rapp and von Stechow’s analysis. I also propose that
the verbal decomposition analysis in conjunction with a verbal analysis of liketa are evidence
that the counterfactual and the scalar reading are actually very similar if not stronger and
weaker versions of the same interpretation. With liketa at least, it is the case for both
readings that with a verb like build, the building eventuality can not have been initiated.
Thus it seems only that the difference between these interpretations is found in whether or
not the agent has undertaken some action which is felicitous with ‘coming close to’ initiating
the event in question. If the counterfactual and scalar interpretations were separate and
independent of each other then, we might expect to find a verb which only exhibits the scalar
interpretation and not the counterfactual interpretation with liketa (Marcin Morzycki, p.c.).
To my knowledge this is not the case.
Moreover, there is evidence from psychological states and activities that the verbal decomposition approach is again insufficient for liketa. Rapp and von Stechow argue that
120

psychological states and activities also receive scalar interpretations when German fast or
English almost scope under the holder (the bearer of an emotional state) or act relations
respectively. This of course assumes that holder is structurally analogous to act for these
verbs and that the interpretation that they receive is actually the same scalar interpretation
that we see with telic verbs. The following data shows that both liketa and almost phrases
have the same interpretation with psychological states and activities.
(29) John liketa/almost hated his teacher.
John felt something that was close to hating his teacher, but he didn’t hate his teacher.
The authors follow Kratzer (1994) and argue that verbs like hate have the decomposition
below.
(30)
holder

VP
AP
almost

VP
V
hate

From (29) we know that liketa also gets a scalar interpretation with such verbs, yet presumably can not scope under holder as in (30) any more than it can scope under act. Again,
the verbal decomposition account, as it stands, can not account for the liketa data.
I want to make one further point about scalar interpretations. For the purposes clarity
in description, I will make a point of differentiating between the interpretation of the scalar
reading with atelic psychological states and activities from other scalar readings for the
following reason. There is a difference between scalar interpretations with telic phrases
or phrases which depict events with natural endpoints like build a chair and with atelic
psychological states and activities like hate his teacher and gallop which do not seem to have
naturally terminating endpoints.
Notice that the scalar interpretation in (29) is an interpretation of resemblance. That
is to say, whatever John feels, it must be ‘close to’ or resemble hating. In a sense, this
121

presupposes that John is in a state which resembles hating to begin with, specifically for
that salient instant in time that the liketa phrase refers to. On the other hand, there is no
requirement of resemblance with telic verbs like build. In fact, if we look back to the scalar
interpretations with build the only requirement seems to be that the eventuality has not been
initiated and that the agent has done something which does not preclude initiation of the
eventuality in question. To my knowledge, this distinction is not mentioned in the literature.
I speculate that a proper analysis of the interpretations of liketa and other approximatives
should reflect this difference via a semantics of imprecision similar to Morzycki (2011).
Finally, a minor modification to the introductory data poses another problem for the
verbal decomposition approach. Why should it be the case that the resultative interpretation
is infelicitous except in cases which involve number? I would speculate that almost can
actually scope inside of the verb in order to operate over those individual bits of events that
are meted out by incremental theme verbs like build, while liketa cannot. Remember that
the resultative interpretation in (28c) is not allowed with liketa in (28). However, when
objects are modified by number then the resultative interpretation is possible.
(31) John liketa built 5 chairs
a. John came close to but didn’t start building 5 chairs.
b. John started building 5 chairs and though he came close he didn’t finish.
Under a decomposition analysis its not clear what the addition of number to an object DP
could be doing to license a particular interpretation, especially when those interpretations
are supposedly determined by the scope of the approximative relative to decomposed heads
of the verb.
In brief, liketa sentences license both a counterfactual and scalar reading with incremental theme verbs like build and if the object of an incremental theme verb happens to
include a numeral then it will license a resultative interpretation as well. With regards to
psychological state and activity verb liketa patterns more or less like almost in that two interpretations are available. One is the familiar counterfactual interpretation while the other

122

is an interpretations of resemblance. I differentiated between Rapp and von Stechow’s scalar
reading in the context of psychological states and activities a resemblance reading because
the so called ‘scalar’ interpretation with these verbs has an extra requirement that the action or emotion instantiated at the time of utterance must resemble the eventuality in the
complement clause. No such resemblance restrictions hold of the scalar interpretation with
building for instance.

4.3.2

Syntactic properties

Though liketa has been deemed an adverb in previous sociolinguistic research, I have argued
that liketa is a verb.

4.3.3

Uncontracted liketa as a non-restructuring predicate

In its uncontracted form liketa is not a restructuring predicate in the sense that its complement contains an instance of infinitival-to. Of course, recalling my earlier proposal that AppE
has a default to insertion rule in some restructuring contexts, it is necessary to point out
that this instance of infinitival-to is actually associated with T and not some lower aspectual
head. It is not an instance of default to insertion because it is not optional. Example (32a)
shows that uncontracted liketa with infinitival-to is acceptable, while example (32b) shows
that the uncontracted form is unacceptable without infinitival-to.
(32)

a.
b.

They had liked to have froze to death last night.
* They had liked have froze to death last night.

That said, liketa in either form is the most restricted verb that we have looked at thus far
in terms of semantic requirements placed upon its complements. For example without the
past morphology in both clauses, the approximative reading disappears.
(33)

a. They like to slap each other.
b. They liketa slap each other.

123

The sentence in example (33a) is basically Standard English and has the paraphrase, they
enjoy slapping each other. Even in the case of the contracted form in example (33b), no
counterfactual interpretation is available.
As we would expect the non-restructuring forms of liketa licenses sentential negation.
(34)

a. It had liked to have not thawed out in time for supper.
b. It had liked to not have thawed out in time for supper.
c. It is not the case that, It had liked to have thawed out in time for supper.

Since the complements of uncontracted liketa are always infinitive, it is worth mentioning
that they never license for complementizers.
(35)

* They had liked for to have froze to death last night.

This indicates that they are maximally headed by TP, which is in itself enough to grant
non-restructuring status.
The uncontracted form in example (36) shows that liketa is also a raising verb and licenses
expletive subjects in much the same way that the contracted form does.
(36) There had liked to have been 3 men killed in the mines.
I take the structure of non-restructuring liketa to be uncontroversial. I will assume for the
rest of this work that it is as shown below.

124

(37)

TP
Theyi

T
T

PerfP
VP

Perf
have

TP

V
liked
ti

T
T
to

PerfP
have froze to death last night

4.4

Contractions as restructuring complements

Goodall (1991; 2006) argues that infinitival contractions like wanna are restructuring predicates a notion that he attributes to Frantz (1977) and Postal and Pullum (1982). He notes
that the class of verbs that exhibit restructuring properties are commonly modal or aspectual
in nature. Consider the list of ‘core’ restructuring verbs and their corresponding semantic
classes adapted here from Wurmbrand (2001).
(38)

Verb type

Example verbs

modal
motion
aspectual
causative

must may can want
come go return
begin continue finish
let make

With this in mind, the contraction wanna is surely a semantic fit with the set of restructuring
verbs on the whole. Goodall goes on to note that plenty of verbal contractions which are
arguably contractions of V + to are all either aspectual or modal in terms of their semantics.

125

(39)

verb

contracted form

go
gonna
used
useta
have
hafta
got
gotta
supposed suposta

example
I’m gonna dance.
I useta swim here.
I hafta eat something.
I gotta drink a beer.
I’m suposta meet her today.

However, Wurmbrand and Goodall both note in their own ways that semantic class alone
does not a restructuring verb make. They assert that a restructuring verb must also exhibit
certain syntactic properties to be considered as such, a point that has surface and resurfaced
throughout my own work here.
Goodall himself notes specifically that though verb class may be a coincidence there are
other factors which make contractions and restructuring more likely to be the same thing
or at least the result of similar mechanisms. Consider the following similarities that he
proposes. The following data is adapted from Goodall (1991; 2006) via Postal and Pullum
(1982).
Wanna, contraction is only possible when to is in the complement clause.
(40)

a.
b.
c.
d.
e.
f.

I don’t want to flagellate oneself in public to become standard practice in this
monastery.
* I don’t wanna flagellate oneself in public to become standard practice in this
monastery.
I don’t want anyone who continues to want to stop wanting.
* I don’t want anyone who continues to wanna stop wanting.
One must want (in order) to become an over-effective consumer.
* One must wanna (in order) to become an over-effective consumer.

In example (40) to is not part of the complement clause and contraction is not possible.
Goodall also notes that contraction is blocked when to belongs to the main verb of a conjoined
clause as in (41a) or when to belongs to the complement of a conjoined verb like need or
want as in example (41c).
(41)

a.
b.
c.
d.

I
*I
I
*I

want to dance and to sing.
wanna dance and to sing.
don’t need or want to hear about it.
don’t need or wanna hear about it.
126

Goodall’s point is that these constraints on wanna contraction are reflected in the constraints on restructuring phenomena. Assuming this mirroring of constraints, it should follow
that restructuring is only possible between a verb and its complement clause.
He begins with examples of clitic climbing, a common restructuring phenomena in Romance languages where a cliticized argument of an embedded verb may raise and attach to
a matrix verb. Consider Goodall’s examples from Italian and Spanish. The remained of the
examples in this section were taken as is from Goodall (2006)
(42)

a. Lo voglio leggere.
it want read
‘I want to read it.’
b. Lo quiero leer.
it want read
‘I want to read it.’

(Italian)

(Spanish)

The crucial point for both examples is that the clitic form lo or ‘it’ is an argument of the
embedded verb read in both sentences. However, in both examples it has moved to the left
of even the matrix verb.
Goodall shows that clitic climbing of this sort is impossible if the matrix verb is contained
in an adjunct clause. Example (43) shows that clitic climbing of las or ‘them’, which is an
argument of the verb comer, to a spot adjacent to quisiera or ‘want’ in the adjunct clause is
impossible. A situation that is parallels the same constraint on wanna contraction.
(43)

a.
b.

Asun si quisiera, comer las no ser´ıa muy buena idea.
* Asun si las quisiera, comer no ser´ıa muy buena idea.
Even if I wanted, to eat them would not be a very good idea.

(Spanish)

Both require that the embedded subject is co-referential with the matrix subject and thus
they also commonly occur in control or raising structures. The constraint on co-referentiality
is present for overt and null subjects. Consider the following data from Spanish.

127

(44)

a.
b.

Juan quiere que Mar´ıa las compre.
* Juan las quiere que Mar´ıa compre.
Juan wants Maria to buy them.

(Spanish)

(45)

a.
b.

¿Qui´en quieres que las compre?
* ¿Qui´en las quieres que compre?
Who do you want to buy them?

(Spanish)

Notice that in example (44) the subject of the embedded clause is Maria and not Juan, which
violates the co-referentiality constraint on clitic climbing. The same is true in example (45)
where the subject of the embedded clause has undergone wh-movement and presumably left
behind a wh-trace, regardless of this covertness the constraint on co-referentiality of the
subject is violated and clitic climbing is disallowed. Goodall argues that this is exactly the
same problem seen with basic wanna contraction data below.
(46)

a.
b.

Who do you wanna see?
* Who do you wanna see Bill?

Restructuring as a phenomenon also seems only to affect verbs. Goodall observes that
infinitival complements of nouns might also exhibit restructuring effects but they do not.
The Spanish preposition de is always required in such complements.
(47) [NP deseos de salir ]
desire to leave
(Spanish)
The same is true of wanna contraction.
(48)

a.
b.

The want to eat is felt by all.
* The wannna eat is felt by all.

Goodall notes one final parallel between the two phenomena concerning conjunction.
Restructuring is not possible when two restructuring verbs are conjoined and one restructures
with the complement that follows them. Again consider the clitic evidence from Spanish.

128

(49)

a.
b.

Quiero y tengo que hacerlo.
* Quiero y lo tengo que hacer.
I want and have to do it.

(Spanish)

In example (49a), there is no restructuring and the sentence is acceptable. However, if we
try to move lo up and create a restructuring situation with the verb tengo the result is unacceptable. Similarly, if we try to conjoin to complements under a restructuring verb, then
restructuring from one complement to the other is unacceptable.
(50)

a.
b.
c.

Quiero comprarlo y cocinarlo.
* Lo quiero comprarlo y cocinar.
* Lo quiero comprar y cocinarlo.
I want to buy it and cook it.

(Spanish)

Again example (50a) shows that conjoined complements are acceptable under restructuring
verbs. However, examples (50b) and (50c) show that restructuring between quiero and either
of the complements is impossible. According to Goodall, these conjunction facts mirrors the
wanna contraction facts in example (41).

4.4.1

Liketa contraction as restructuring

The question now becomes: does liketa pattern with wanna contraction it terms of sharing
these four parallels with restructuring processes like clitic climbing? Here I will compare and
contrast liketa with the list of parallels in (51) developed by Goodall in the last section.
(51) Goodall’s restructuring and contraction parallels
a. The infinitival contraction is modal or aspectual in nature.
b. The infinitival must be a complement of the verbal host of the contraction.
c. The subject of the matrix and embedded clauses must be co-referential.
d. The host of the contraction must be a syntactic verb.
e. Infinitival contraction is not possible with conjoined verbs.
f. Infinitival contraction is not possible with conjoined complements.
I will argue that contracted liketa indeed patterns with wanna contraction and restructuring verbs.
129

First, we have already seen that the meaning of liketa is purely irrealis in nature and
inextricably tied up with perfect aspect. Thus liketa follows the pattern of other infinitival
contractions and restructuring verbs.
Likewise, in the case of liketa, the contraction is not possible if the infinitive is not a
complement of the verb.
(52)

a.
b.
c.
d.

Even if I had liked, to have drank all the whiskey in town would have been a
bad thing.
* Even if I had liketa drank all the whiskey in town would have been a bad
thing.
Even if I had liked to, to have drank all the whiskey in town would have been
a bad thing.
* Even if I had liketa to a’drank all the whiskey in town would have been a bad
thing.

Subjects must also be co-referential with liketa as they must with wanna contraction and
clitic climbing.
(53)

a.
b.

Who had you liketa seen?
* Who had you liketa seen Bill?

Liketa contraction is only possible if the syntactic category of like is verbal. This seems
trivially true because only verbal like actually sub-categorizes for an infinitive. The problem
is that not all contractions involving like and to are instances of approximative liketa.
(54)

a.
b.
c.
d.
e.

He likes bourbon now.
He likes to break things now.
He likesta break things now.
I liketa break things now.
I liketa broke things yesterday.

transitive verb
transitive verb + infinitive
transitive verb + infinitive
transitive verb + infinitive
restructuring liketa

Examples (54a) and (54b) show that normal transitive like in English may either take an
object DP or a non-finite clause. Example (54c) shows that in quicker speech the infinitive
can adjoin to the verb. Finally, examples (54d) and (54e) differ only in that approximative
liketa requires past morphology in the verb. Needless to say, this is evidence that liketa

130

contraction is only possible with a verb. Moreover, liketa contraction resulting in the approximative is only possible when like has a perfect infinitive as its complement.
As in the wanna and restructuring cases, liketa is unacceptable when conjoined with
another verb.
(55)

a.
b.

I needed and liked to have heard those things.
* I needed and liketa heard those things.

Finally, conjoined complements are unacceptable under liketa.
(56)

a.
b.

I had liked to have danced and to have sung.
* I had liketa danced and to have sung.

Thus liketa patterns exactly as wanna contraction does with respect to its parallels with
respect to clitic climbing. As such liketa fits the class of verbs which exhibit contraction
and restructuring effects; raising/control verbs of modal or aspectual nature, which take
infinitival complements.

4.4.2

A head-movement analysis of contraction as restructuring

Roberts (1997) argues for a purely head-movement driven account of wanna contraction in
English which is directly extendable to liketa and an account for the ban on the embedded
perfect auxiliary in contracted liketa forms. In the next few paragraphs I will sketch out
the relevant pieces of Roberts’s analysis. A proper treatment of all facets and properties of
wanna are beyond the scope of this work focusing on liketa but, see (Pullum, 1997; Roberts,
1997; Goodall, 2006) for an overview of the issues.
Similar to the ban on embedded have with contracted liketa, Roberts is interested in
accounting for the presence or absence of embedded T and infinitival-to. His primary claim
is that wanna and want are separate lexical items and that wanna is sub-categorized for
a bare infinitive. He argues that the bare infinitive contains a restructuring feature which
triggers head movement of the embedded T to the matrix verb. A sentence like They wanna
leave has the following structure and I have taken the liberty of adapting Roberts’ features
131

to a probe/goal feature checking system. Consider the derivation of the embedded clause
below.
(57)

TP
Theyi

T
T
rstrct

VP
leave ti

In example (57), the embedded TP has been created. The subject has been merged and no
head movement has occurred. The resulting non-finite T has an interpretable restructuring
feature and is phonologically null.
In the next structure we see that the matrix V is added.
(58)

VP
V
wanna
u:rstrct

TP
ti

T
T
rstrct

VP
leave

At this point in the derivation the uninterpretable restructuring feature on matrix V will
probe down the tree seeking satisfaction by its interpretable counterpart on T. Assuming
the feature is strong and movement is required the embedded T will then head move and
left adjoin to matrix V, following Kayne (1994), in order to satisfy the checking of a strong
feature. Matrix T is added and the embedded subject must move to satisfy the EPP feature
on matrix T. The resulting structure is below in example (59).

132

(59)

TP
Theyi

T
VP

T

TP

V
Tk
rstrct

V
wanna
u:rstrct

ti

T
tk

VP
leave

Finally, Roberts must stipulate a constraint on the matching of vocabulary to heads to
ensure that infinitival-to is disallowed from appearing to the left of the V. This is achieved
by the statement of an output condition on the spell out of morphological words. This
statement is shown example (60) and in plain English it says that two lexical items may not
be spelled out on the same syntactic head.
(60)

* [X W1 , W2 ]

Mechanism aside, Pullum (1997) notes that the resulting structure proposed by Roberts
and adapted in example (59) makes a prediction. It predicts that wanna contraction in
VP fronting and ellipsis contexts would be unacceptable given those contexts require the
presence of overt T and infinitival-to.
(61)

a.
b.

I said I’d feel like climbing the mountain, and climb it I want to.
* I said I’d feel like climbing the mountain, and climb it I wanna.

It should be mentioned that both Pullum and Goodall mention that the VP fronting and
ellipsis data is vague and even questionable. However, among Standard English informants
that I have asked about this, there does seem to be a noticeable distinction in terms of
acceptability of the two forms. In short, the prediction holds.
In summary, wanna contraction may be modeled syntactically using a head movement
only approach to restructuring. Under such a model, the verb want and the contraction
133

wanna are separate lexical items with distinct properties of complementation; a notion that is
definitely familiar given the forms and analyses proposed in this work. Restructuring wanna
bears an uninterpretable restructuring feature and is sub-categorized for a bare infinitive
TP which must contain an interpretable counterpart of the restructuring feature on the
matrix verb. The uninterpretable feature on the matrix V then triggers head movement
of the embedded T to left adjoin on matrix V via probe/goal feature checking. Infelicitous
insertion of infinitival-to onto the adjoined T is barred by conditions on the spell out of more
than one word per syntactic head.

4.4.3

A bi-clausal head-movement analysis of liketa

In this section, I break from the pattern of the last few sections where I have assumed Wurmbrandian style mono-clausal restructuring to explicitly consider liketa in terms of Roberts’
head movement analysis of wanna before moving on to the primary analysis in the next
section. Such analytical comparisons can only strengthen my claims that liketa behaves as
a restructuring verb if it is amenable to analysis in not one but two different frameworks
which account for restructuring. We can extend Roberts’ analysis of wanna to liketa with a
minimal modifications.
Recall that with liketa is it not only the case that infinitival-to does not show up in
the complement of contracted liketa but the perfect auxiliary is also banned in Appalachian
English.
(62)

a.
b.

* John had liketa have punched Bill before you arrived.
John had liketa punched Bill before you arrived.

This is accounted for if the perfect and T head move into the matrix verb. We can now
apply the same prediction to liketa that Pullum (1997) asserted for Robert’s treatment of
head movement under wanna contraction. Specifically, in a system where we assume head
movement is the mechanism by which restructuring proceeds then, restructuring predicates
in VP fronting and ellipsis contexts should be unacceptable.
134

(63)

a.
b.
c.
d.

They swore they’d liketa died and died they’d liked to(’ave).
* They swore they’d liketa died and died they’d liketa.
Mary’d liked to have slapped John and Sue had liked to(’ave) too.
* Mary’d liked to have slapped John and Sue had *liketa too.

This means that liketa truly patterns with wanna contraction in terms of presence and
absence of T in embedded clauses. Borrowing Robert’s analysis, those unacceptable examples
of the contractions in VP fronting and ellipsis contexts are unacceptable as compared to the
uncontracted form because they do not actually contain an embedded T.
Now recall the fact that liketa complements must minimally contain past participle morphology.
(64)

a.
b.
c.
d.

John
* John
John
* John

had
had
had
had

liked to have punched Bill before you arrived.
liked to punch Bill before you arrived.
liketa punched Bill before you arrived.
liketa punch Bill before you arrived.

This fact coupled with the evidence from Pullum’s VP fronting and ellipsis prediction
which says that non-finite T is absent in infinitival contractions now explains the ban on
overt have in the embedded complement of restructuring liketa. If restructuring complements
do not contain an embedded T then we would expect auxiliary have to be banned in the
complement of restructuring complements. To clarify, a mono-clausal restructuring approach
also predicts the ban on embedded auxiliary have. This is especially interesting given the
condition on liketa that the embedded clause must be perfect in combination with the lack of
a non-finite T and in light of my claim that auxiliary insertion is the result of the morphology
assigning head entering into an agreement relationship with a higher head.
Thus, we can argue that liketa selects for a perfect infinitive. This is operationalized by
positing that liketa has both a restructuring feature and an uninterpretable perfect feature.
This requires that the T selected by liketa must have both features. This is because the
strong uninterpretable restructuring feature on liketa will trigger head movement as soon as
its probe values its features against its interpretable counterpart. The probe, match, and

135

valuation will occur regardless of whether or not the perfect feature is available on T. We
know descriptively that these complements are always perfect and ban auxiliary have. The
only way to ensure that the T you are selecting for is restructuring and perfect is to have
both features visible on the embedded T prior to being probed.
Consider the derivation of the embedded clause in John liketa died where Perf has already
undergone head movement and adjoined to T.
(65)

TP
John

T
T
perfk

PerfP
T
rstrct

Perf
tk

VP
died ti

This partial derivation shows the head movement of the perfect head to embedded T where
the perfect and restructuring feature will be simultaneously visible to the probe from matrix
V in order for the derivation to converge. This is achieved via a previous head movement of
Perf to T.
Next, liketa is added. Notice that it contains strong uninterpretable features which are
required to be satisfied in the next operation lest they crash the derivation. Having undergone
a previous head movement, all necessary features on non-finite T are in a position to be
checked.

136

(66)

VP
V
liketa
u:rstrct*
u:perf*

TP
Johni

T
T
perfk

PerfP
T
rstrct

VP

Perf
tk

died ti
The strong uninterpretable features on the verb will probe down the tree and be satisfied
simultaneously by T this is shown in example (67) .
(67)

VP
TP

V
V

Tl
perfk

T
rstrct

Johni

T
tl

liketa
u:rstrct*
u:perf*

PerfP
Perf
tk

VP
died ti

This is how liketa may be analyzed as having an almost identical derivation to wanna
contraction assuming a head movement style analysis of restructuring. The final derivation
shows how we might simultaneously get the required past morphology on the embedded verb,
ban the insertion of the perfect auxiliary, and account for the fact that non-finite T seems
to be non-existent in liketa complements according to VP fronting and ellipsis tests. It also
shows why wanna contraction does not bar auxiliary have in example (68a). Presumably, the
perfect auxiliary head is not ‘rolled up’ with the T via a previous head movement operation.
Moreover, example (68b) shows that wanna definitively does not select for a perfect infinitive.
If it did, we would expect the appearance of participle morphology on the verb in the absence
of the overt auxiliary in (68b) to be acceptable.

137

(68)

4.5

a.
b.

I wanna have eaten before you arrive.
* I wanna eaten before you arrived.

Analysis

Given liketa’s similarities to wanna contraction and its general list of properties, the contracted form is is best analyzed as a restructuring verb. Given the success of Roberts’ head
movement account any further account of the properties of liketa should attempt to cover
more theoretical ground with less theoretical machinery and/or stipulated conditions. So
any account of liketa must now minimally and naturally account for the ban on perfect auxiliary in the embedded clause. In this section, I will continue with the general spirit of the
previous analyses of restructuring predicates and make use of Wurmbrand’s mono-clausal
approach to the classification of restructuring verbs in order to provide another analysis of
contracted liketa. Once, We have the primary analysis in place, I will compare and contrast
the two ways of approaching restructuring predicates. In what follows, I will argue for the
following structure for contracted liketa.
(69)

TP
Subji

T
T

PerfP
vP

Perf
(has)
v

PerfP

v
Perf
liketa ∅

vP
died ti

Though we have seen that liketa patterns with wanna contraction and shares properties
with other restructuring phenomena, I will provide further evidence from the perspective of
Wurmbrand (2001) that liketa is a restructuring verb. Then in line with previous chapters I

138

will argue that the embedded clause of liketa complements are maximally headed by PerfP.
Next, I will derive the ban on the perfect auxiliary. In the last section of the analysis, I will
compare this approach to the head-movement approach.

4.5.1

Liketa is a restructuring verb generated in v

Aside from the parallels that I have drawn from Goodall’s observations, Liketa contractions
have a majority of the other properties that we have come to associate with restructuring
predicates. They do not contain sentential negation associated with T, they never license
complementizers, and though they might show embedded aspectual marking, they never
show overt tense marking.
Let us first locate liketa itself in the matrix clause. Recall that one main argument of
Wurmbrand (2001) is that ability of the verb to undergo matrix passivization is an indicator
of belonging to the class of verbs which she terms true lexical restructuring predicates. Liketa
does not undergo matrix passivization.
(70)

a.
b.

* The whiskey was liketa drank.
* The whiskey had been liketa drank.

Thus liketa patterns with causative and experiencer have in that it is not a true restructuring
predicate in terms of matrix passivization. I take this as evidence that liketa is not a lexical
verb in V but, must either be in v or an auxiliary head since it is not an adverb.
The question of whether or not liketa is an auxiliary can be answered by once again
looking at its distribution relative to other auxiliary heads. We know from Kyt¨o and Romaine
(2005) that liketa is historically found preceded by the perfect have and this is still the case.
Although in AppE the overtness of the matrix auxiliary seems to be optional.
(71)

a. John (had) liketa drank the whiskey by the time we arrived.

Further evidence of the presence of a spot for the matrix perfect auxiliary is that liketa only
exists as a question where a matrix perfect has undergone T to C wh-movement.

139

(72) Had you liketa drank all that whiskey before we got there?
I take these two facts to indicate the presence of a perfect head in the auxiliary space
located above contracted liketa. If we take the passivization facts into consideration as well,
it seems liketa must be located above V and below PerfP. In a fully articulated English clause
the only two heads that appear between these positions are ProgP which assigns progressive
morphology and v which licenses external arguments and theta-roles. For reference, I assume
the basic functional structure in (73) for Standard English and AppE clauses.
(73)

TP
T
T

PerfP
Perf

ProgP
Prog vP
...

Crucially, liketa never occurs with progressive aspect. Example (74) illustrates this fact
for both matrix and embedded clauses.
(74)

a.
b.

* John had been liketa drank my whiskey.
* John had liketa been drinking my whiskey.

Give the basic clause structure in (73) and the evidence that liketa may never appear with
the progressive aspect in (74), we can conclude that liketa is located in v.
(75)

PerfP
vP

Perf
v

v
embedded clause
liketa

140

If liketa were an auxiliary then we would expect it to undergo auxiliary inversion in yes/no
questions. Again, this is not the case.
(76)

* Liketa John drank the whiskey?

Of the standard auxiliaries it seems that only modal auxiliary might resists T to C movement in matrix yes/no questions. However, might and liketa can co-occur in an utterance
and they are generated at different points in a hierarchy of projections. The ordering is
might < have < liketa.
(77) He might’a liketa been killed.
Further evidence for a liketa in v configuration is the fact that liketa is only available
in questions with the perfect auxiliary. To be clear example (78a) shows that liketa is
unacceptable in yes/no questions assuming do-support as noted by Labov (1972); Wolfram
and Christian (1976); Feagin (1979). However as shown in example (78b), it is possible for
liketa to appear in questions if they involve the perfect auxiliary.
(78)

a.
b.

* Did John liketa drank the whiskey?
Had John liketa drank the whiskey by the time you arrived?

The verb liketa patterns with causative and experiencer have. However, we face contradictory evidence in deciding whether liketa is base generated in v or V. If we follow
Wurmbrand’s classification system then semi-functional verbs located in v do not undergo
passivization because they are in a type of complementary distribution with passive v. On
the other hand traditional raising structures are commonly assumed to be missing a caseassigning v in the matrix clause because embedded subjects appear in the matrix clause
with nominative case. They never appear marked for accusative case. Combine this observation with the fact that raising verbs do not assign an external argument theta-role and
the evidence seems to support a vP-less analysis of raising structures. This is why expletive
subjects are licensed as the subjects of raising verbs.
Recall that liketa licenses expletive subjects.
141

(79) There liketa been 3 men killed in the mines.
From here the evidence for a v vs. V generation point for liketa only becomes more
complicated. Wurmbrand suggests that traditional raising verbs like German scheinen ‘to
seem’ are generated as low auxiliaries. At first glance this proves enticing for English as
well but the problem remains that auxiliaries in English usually undergo subject auxiliary
inversion and liketa clearly does not. I believe the best solution is to assume that liketa is
base generated in an unaccusative v.
This captures the fact that liketa resists matrix passivization and allows the fact to
fall naturally out of the familiar claim about semi-functional restructuring verbs being base
generated in a v. The only difference is that semi-functional restructuring verbs generated in
unaccusative v do not assign a semantic role to their external arguments and license expletive
subjects.
Aside from the properties common to restructuring complements, the major pieces of
evidence for the architecture of clauses embedded under liketa come from the scant appearance of past participle morphology. For clarity, I will reiterate those properties of liketa
that Wurmbrand claims are common to all restructuring complements before discussing the
perfect aspectual requirements of the clause.
Liketa does not license sentential negation associated with T. Example (80) show that
neither stand alone not nor didn’t forms are licensed.
(80)

a.
b.

* John had liketa not drank my whiskey.
* John liketa didn’t drink all my whiskey.

The simplest explanation of this fact is that both forms are parasitic on the presence of T and
they are unacceptable in these examples because T is not present. Liketa only ever admits
the negative adverbial never. which I take to be a negative constituent which crucially does
not rely on the presence or absence of T in a clause. Specifically, adverbial never does not
require attachment to TP, (Zanuttini, 1996).

142

Further, liketa phrases containing adverbial never do not pass the not even a little continuation test for sentential negation, (Klima, 1964). Consider the test operating on a normal
sentence with a sentential negation interpretation.
(81) John didn’t drink any whiskey, not even a little.
Now consider, this test in a liketa phrase containing adverbial never.
(82) John liketa never drank any whiskey...
a. # not even a little.
b.
except once every long while.
The continuation in example (82a) is infelicitous here. To unpack this we must remind
ourselves of what liketa phrases actually means. The positive version of this phrase John
liketa drank some whiskey is only true if he came close to drinking some whiskey but did
not actually drink any whiskey. Now, for the phrase in (82) to be judged true, it must be
the case that John did drink some whiskey if only rarely. Hence the second continuation in
example (82b) is felicitous with example (82). Thus it seems that the inclusion of never in
liketa constructions only reverses the negation found inherently in the polar component of
liketa’s conventional meaning. Though this type of negation seems to negate the proposition
that liketa operates on, it is not the negation associated with T. Since it is not negation
associated with T, the appearance of adverbial never poses no problems for the claim that
liketa complements are not maximally headed by T.
The eventuality described by liketa and the eventuality described in the complement
clause are temporally inseparable.
(83)

* Last Monday, John had liketa drank my whiskey the next day.

This is not too surprising since liketa has a proximal or closeness component as part of its
conventional meaning. The eventualities described must be sufficiently close enough in a
time course of events, if not contiguous.

143

Finally, as I mentioned earlier complementizers are never licensed by liketa. Even though
liketa has a peculiar form and strict semantic requirements it passes the basic diagnostics
required of restructuring complements.

4.5.2

Complements of liketa are headed by PerfP

The past participle is the only morphology that appears in liketa’s complement. So, I contend
that liketa clauses are headed by perfect phrases which license participle morphology. In
example (84a), past morphology is licensed on the verb but progressive morphology is not
licensed.
(84)

a.
b.

Mary had liketa slapped John.
* Mary had liketa slapping John

Further evidence that PerfP heads the embedded clause is shown in passivized liketa
complements. Example (85a) shows a normal perfect passive complement under liketa,
while example (85b) shows that the perfect complement is required. Lastly, example (85c)
shows that the be of the infinitival passive is also unacceptable.
(85)

a.
b.
c.

John had liketa been slapped.
* John had liketa was slapped.
* John had liketa be slapped.

The acceptable case shows that the passive head been is marked for the perfect. One
could argue that this is simply showing that the matrix perfect had in all cases is required
to mark the embedded auxiliary be for perfect. However if this were the case then it is not
clear why removing auxiliary had does not result in acceptability.
(86)

a.
b.
c.

John liketa been slapped.
* John liketa was slapped.
* John liketa be slapped.

In sum, liketa complements lack a tense phrase and are maximally headed by a perfect
phrase which we see the effects of even though it is banned as a stand alone lexical item.
144

4.5.3

Deriving the ban on the perfect auxiliary

Reconsider this data from the introduction which illustrates the ban on over auxiliary have in
the complement of liketa clauses; an interesting and unexpected fact given the acceptability
of an over perfect auxiliary in the uncontracted form.
(87)
(88)

* John had liketa have punched Joe, by the time...
John had liketa punched Joe, by the time...

We are finally in a position to analyze this constraint as the result of unique aspectual
properties of liketa in conjunction with the structural requirements of complements of restructuring verbs. As argued for in the previous sections of this analysis, contracted liketa
has the structure repeated in example (89). It is a semi-functional restructuring verb base
generated in v which selects a complement maximally headed by a perfect phrase.
(89)

TP
Subji

T
T

PerfP
vP

Perf
(has)
v

PerfP

v
Perf
liketa ∅

vP
died ti

Given our previous assumptions about auxiliary insertion, the ban on the insertion of an
embedded perfect is expected from this structure. In both previous analyses, I argued
that auxiliary heads may do double duty. They are capable of assigning morphology to a
lower auxiliary or verbal element and they themselves may bear the morphology of some
higher morphology assigning head. The overtness of the head itself is the result of having
entered into an agreement relation with a higher morphology assigning head. Taking this
into account, the overt auxiliary have is banned because its insertion would require the
145

perfect head to enter into an agreement relationship with a higher head, presumably T. In
short, since only T is ever above PerfP in the hierarchy of projections in the English T/Aux
system, the absence of T in a restructuring verb predicts the absence of overt auxiliary have.
However, not all restructuring verbs exhibit the same restrictions on auxiliary licensing.
Keep in mind that other restructuring contracted forms do not show this constraint.
(90) I wanna have eaten before you arrive.
I argue that this must be an example of ‘exceptional’ auxiliary licensing. The derivation
in example (91) shows that matrix T must satisfy the uninterpretable inflection feature on
the embedded auxiliary head.
(91)

TP
DP
I

T
VP

T
prs
V
wanna

PerfP
Perf
have
uinfl:prs
perf

VP
V
eaten
uinfl:perf

Evidence for such a relation can be seen in the fact that wanna never exhibits and agreement
and is incompatible with morphology assigning heads.
(92)

a.
b.
c.
d.
e.

* I wanna-ed eat.
* I had wanna eat.
* I am wanna eat.
* I was wanna eat.
* He wanna eat.

past
perfect
progressive
passive
3.sg

Further, these agreement facts are not just semantic or c-selectional restrictions in disguise
because wanna is compatible with them if agreement is handled elsewhere.

146

(93)

a.
b.
c.
d.

Did I wanna eat?
Does he wanna eat?
He will wanna eat.
He might wanna eat.

past
3.sg
future
modal

Finally, perfect have under liketa is not licensed in this way because of locality conditions
on Agree and intervention effects (Chomsky, 2001).
(94) Intervention:
If probe P matches inactive K that is closer to P than matching M, this bars Agree
(P,M).
Intervention effects arise in the case of liketa then because unlike wanna, liketa is compatible
with morphology assigning auxiliaries like the perfect. I take this as evidence that liketa
takes part in the exchange of inflectional features which are commonly assumed to occur
between T, auxiliaries, and verbs. Consider the derivation below in which the embedded
auxiliary head remains unvalued by matrix T because liketa has the requisite feature which
forces it to act as an intervener.
(95)

TP
Subji

T
T
prs

PerfP
Perf
(has)
uinfl:prs
perf

vP
v
v
liketa
uinfl:perf

4.5.4

PerfP
Perf
uinfl
perf

vP
died ti

Accounting for variation in liketa complements

Having outlined in depth the properties of restructuring liketa and having accounted for its
behavior with via two separate restructuring frameworks, I would now like to account for
147

variation in the form of the complements that we have encountered along the way. Consider
the following instances of variation.
(96)

a.
b.
c.
d.

He liketa a’died.
I liketa didn’t win.
She liketa have died!
They liketa not got any food or anything to ’em.

(Diane B. W15:38.II.120)
(Sam C. W70:34.I.236)

Under the analysis forwarded here, the prefixed a form in example (96a) is not the result
of auxiliary insertion. Given there is no embedded T, it is reasonable to assume that a’ might
be the result of a default insertion rule on the perfect head. Assuming [+past,+participle]
are the standard features on a perfect head following Halle and Marantz (1993).
(97) P erf = default insertion rule
[+past, +part]

↔ a’

This is not to be confused with the actual auxiliary insertion rule which requires the presence
of T, a higher agreeing head. Assuming T agrees with embedded perfect heads for a feature
φ then the feature set resulting in auxiliary insertion is as follows. Again [+past] in both
cases is simply a cover term for the distinguishing feature which separates past and past
participle morphology; it is not tense.
(98) P erf = auxiliary insertion rule
[+phi, +past, +part] ↔ have
As for examples (96b) and (96c), I would argue that they are not restructuring predicates.
The presence of didn’t and have are indicative of an embedded tense layer. It is possible that
in these dialects, liketa has been grammaticalized as an adverb. Example (96b) is reported
by a speaker who has liketa in their dialect in south eastern Ohio, and (96c) was reported
by Feagin. Again, There is no reason to believe that for every speaker who has the lexical
item liketa, it should be grammaticalized in exactly the same manner. I am not a speaker
of Alabama English or the English of southeastern Ohio and so I can make no judgments
about these forms aside from saying that I find them unacceptable.
148

Finally, I would argue that the negative not in example (96d) is possibly constituent
negation. Again this example is from Alabama English and I do not believe it is appropriate
to subject it to analysis by my intuitions. I can only offer speculation. However, example (99)
it is the case in more standard varieties of English that negation using not requires an over
auxiliary.
(99)

a.
b.
c.
d.

I
*I
I
*I

could not see.
not see.
did not eat.
not eat.

However, constituent negation is not subject to this constraint.
(100)

a. I saw not one beer but two.
b. I got not food but water.

Though these variations in the complement of liketa are indeed interesting and require
further study, They do not pose a problem for my analysis. In fact, if we assuming a strong
version of the Chomsky-Borer Conjecture (Borer, 1984; Chomsky, 1995), whereby syntactic
variation is essentially reducible to differences in the lexicon, we should expect just such
slight micro-variations in nearby dialects. Ultimately, I leave these forms to further study.

4.5.5

Mono vs. bi-clausal approaches to restructuring

Restructuring liketa and its unique set of properties may be captured by both the bi-clausal
head movement approach to restructuring or by the mono-clausal variation in complement
size approach. Comparing the two approaches using the same form, highlights the strengths
and weaknesses of both approaches. The mono-clausal approach to restructuring suffers
from a general explosion in the size, properties, and types of complement clauses. This is
mirrored by a loss in uniformity of those same clauses. Moreover, this seems to require a
general explosion in the lexicon as well. Suppose a speaker has not just the contracted and
uncontracted forms of liketa but also a contracted form which seems to be non-restructuring.

149

The mono-clausal approach, having zero syntactic mechanisms by which to derive those
3 forms requires at minimum three possible entries for the syntax of liketa alone. This
is a disconcerting thought to those who worry about issues of learnability and an overly
syntax-ified lexicon. This is a natural extension of Wurmbrand’s own criticism of the loss of
uniformity of complement clauses under her approach.
Beyond this, we saw in the last chapter that the mono-clausal approach to restructuring
only has a certain level of resolution and it must be the case that post syntactic morphological
processes must bear at least part of the burden as I have argued in the past two chapters.
On the other hand, the head-movement approach of Roberts (1997) requires extra machinery but it provides us with a way to effect smaller changes and operations in the syntax
before derivations are sent off to be interpreted at spell out. The grand finding in terms of
restructuring thus far is that we must pay attention to minute morphological processes in the
context of restructuring clauses but even then and even under the operational simplicity of
a mono-clausal approach, some external processes such as head movement are inescapable.

4.6

Conclusion

This chapter provided an analysis of the Appalachian English approximative form liketa. It
have shown that liketa is best analyzed as a verb in both its contracted and uncontracted
forms. Such an analysis allows us to understand the relationship between the two forms
and their respective syntactic properties in terms of restructuring. For example, the ban on
the embedded auxiliary in contracted liketa is argued to fall out from this a restructuring
analysis. Following arguments which show that wanna contractions are actually examples
of restructuring in English, I argued that liketa should be included in this class. Infinitival
contractions are fertile ground for restructuring effects. Finally, I analyzed liketa according
to two different frameworks which capture restructuring effects.

150

CHAPTER 5
NON-FINITE HOW COME

5.1

Introduction

In this chapter, I explore one final infinitival form found in Appalachian English. Consider
the standard finite form of the wh-word how come in example (1a) and the non-finite form
in example (1b) available in Appalachian English.
(1)

a. How come he left the party?
b. How come him to leave the party?
WH
3s.acc inf leave the party
‘How come he left the party?’

This chapter is an exploration of the syntax of what I have termed non-finite how come.
Current analyses of the syntax of finite how come indicate that it is best analyzed as being
a single syntactic unit merged directly into the left periphery of the clause. I will argue
that non-finite how come is slightly different. I will revive and apply an analysis in which
non-finite how come is not a single syntactic unit but rather a reduced clause. I propose that
non-finite how come as it exists today in AppE originally occurred with null expletives in
a verb second configuration in many English dialects. Interestingly, none of the diagnostics
used to differentiate finite how come from its counterpart why reveal any contrasts between
the two forms of how come. Further, the properties of how come which have been claimed
as evidence of its status as a single lexical item are actually compatible with the possibility
that the form is actually phrasal. This supports an analysis in which the matrix clause
is the result of clause reduction. In this sense then, this chapter is a continuation of the
investigation of restructuring found throughout this work. However, this final investigation
of restructuring and clause reduction switches focus from the embedded clause to the matrix
clause.

151

This chapter has the following structure: First, I lay out what little information there
is on non-finite how come before moving on to discuss the literature on finite how come.
How come is predominantly talked about in comparison to other forms from what for to
why, to other wh-adjunct phrases like why the hell. In the analysis, I show that non-finite
how come behaves almost exactly like finite how come and that the clause embedded under
how come in non-finite how come is a full CP clause. Then, I revive the reduced clause
analysis of non-finite how come by examining how case is assigned, the behavior of come in
modern English, and the assumption that non-finite how come is a relic verb second form
which involves null expletives, and evidence from the behavior of how when it is used for
asking about cause in Chinese. Finally, I show how the proposed structure derives all of the
properties associated with how come as well as accounting for the ability of non-finite how
come to allow for some lexical material to intervene between how and come.

5.2

Background

In this section, I will provide a general context for non-finite how come as well as discuss
properties of non-finite how come as compared with properties of how come. In the final few
sections, I will discuss previous analyses of finite how come.

5.2.1

Non-finite how come in context

Non-finite how come was present in literary works from at least the turn of the 20th century
until the late 1930’s.
(2)

a. She heard Miss Euphemia wonderin’ the other day how come the right shoulder
of her black silk dress to wear out.
(Stuart, 1899)
b. That’s how come him to have appetite for a nip o’ pore old Betsy.
(Rawlings, 1938/1985)

152

It also appears throughout the south in general as show in the transcribed interviews
with former slaves (PBS, 2013).
(3)

5.3

a. JHF: And he’d round up the votes, and that’s how come them to kill him.
b. JHF: Well what I was, what I’m trying to, to find out is, how come him to kill
your husband. Was it over politics?

Previous analyses of how come

To my knowledge there are no previous analyses of non-finite how come. In fact, it is
commonly assumed that how come only exists with finite complements (Culicover, 1999;
Collins, 1991). For these reasons, it is necessary to discuss in detail the previous syntactic
analyses of how come. The next few subsections are a discussion of the properties of how
come as they are mentioned in the literature along with the analyses they were used to
support.

5.3.1

How come and what for

Zwicky and Zwicky (1971) make the following observation that how come is used solely for
questioning reasons and not purpose. The question and response pairs in example (4) shows
us a split between questions about cause on one hand and purpose on the other.
(4)

a. Question:
i. Why don’t I read you this letter?
ii. Wouldn’t you like to have me read you this letter?
iii. Will you let me read you this letter?
b. Answer:
i.
What for?
ii. # How come?

The argument is that the wh-word what for is used to question purpose and it is licensed in
situations where the relation between the actions of an agent and an intended state of affairs
are questioned. The wh-word how come on the other hand is not licensed in such situations.

153

Zwicky and Zwicky argue that cause interpretations are the relationship between one state
of affairs and another state of affairs. Thus how come is not felicitous when asking about
the intentions of an agent, it is only felicitous when asking about reasons.
In the remaining portion of this section, I will lay out the syntactic properties of how
come as noted by Zwicky and Zwicky.

5.3.1.1

How come bars subject/aux inversion

Turning to more syntactic observations Zwicky and Zwicky also note that how come fails to
trigger subject auxiliary inversion as compared to what for, why, and the lone how.
(5)

a.
b.
c.

What has she read the book for?
Why has she read the book?
How has she read the book?

Example (5) shows that the wh-words what, why, and how trigger the normal subject auxiliary inversion that we expect in matrix questions in English. The example below in (6a)
show that how come is unacceptable with and does not trigger subject auxiliary inversion.
Example (6b) shows that why questions are unacceptable without subject auxiliary inversion.
(6)

a.
b.

* How come has she read the book?
* Why she has read the book?

Zwicky and Zwicky take this to be evidence that clauses embedded under how come
appear to be more like embedded clauses than interrogative matrix clauses. Specifically,
they argue that how come is diachronically derived from how has it came about that.

5.3.1.2

How come blocks long distance contstrual

Moreover, Zwicky and Zwicky suggest that a bi-clausal origin in the phrase how has it came
about that accounts for the fact that how come is only allowed to be interpreted relative to
the proposition in its own clause. Why and what for do not behave this way, since they
allow embedded clause construals.
154

(7)

a. Why did Herman say Gwen ate the goldfish?
i. Why did Herman say this?
ii. Why did Gwen eat the goldfish?
b. What did Herman say Gwen ate the goldfish for?
i. What did Herman say that for?
ii. What did Gwen eat the goldfish for?

For example, example (7a) shows that why may be construed as questioning the reason
that Herman made the statement in (7a-i) or the reason that Gwen ate the fish in (7a-ii).
Similarly, (7b) also shows that what for may also be construed as questioning the matrix or
embedded clauses represented as questions in examples (7b-i) and (7b-ii).
In spite of the behavior of the other two wh-words, how come may not be construed with
the embedded clause. Specifically, how come in example (8) is only able to be paraphrased
as What is the reason that Herman said that Gwen ate the goldfish.
(8) How come Herman said Gwen ate the goldfish?
a.
How come Herman said that?
b. * How come Gwen ate the goldfish?
Relatedly, they observe that the lack of long distance construal is not a quirk of main clause
phenomena as how come clauses may be embedded.
(9)

a. Margaret wondered how come Herbert said Bill grew piranhas.
b. I realized that Margaret knew how come Herbert said Bill grew piranhas.

Finally, they note that colloquial variants of how come do in fact sometimes occur with
an overt complementizer that, a fact which I would argue is evidence of a C.
(10) How come that she has read the book?
In summary, Zwicky and Zwicky argue that how come is used to question cause as
opposed to purpose. It is a syntactic unit derived from how has it came about that, which
suggests a retention of its original clausal nature in the sense that it never moves. Thus
it does not undergo subject auxiliary inversion or allow for long distance construal with
embedded clauses yet may be an embedded clause itself. Given that Zwicky and Zwicky
155

claim that how come may occur with an overt complementizer, it would then be located
above C in a normally articulated clause, either as a single unanalyzed constituent or as a
phrase.

5.3.2

Why and how come

In comparing, why and how come Collins (1991) argues the how come is a syntactic unit
merged in C. Specifically, he argues that how come is best analyzed as a wh-complementizer.
He also adds to the list of properties unique to how come. In addition to Zwicky and
Zwicky’s observations that how come does not exhibit subject auxiliary inversion, he also
notes a lack of quantifier scope ambiguities and the inability of how come to appear in
multiple wh-questions.

5.3.2.1

How come as a syntactic unit merged in C

Collins (1991) claims that the differences between why and how come can be captured with
two basic assumptions. First, he assumes that why is base generated adjoined below C to
either VP or TP and ends up undergoing wh-movement to end up in spec,CP as in example (11). Second, Collins assume that how come is base generated in C as in example (12)
and never moves or binds a trace.
(11)

(12)

CP
whyi

C

C
TP

C
ti

CP

TP
C
how come
...

TP
T
T

VP
ti VP
...
156

Crucially, then in English why is in spec,CP while how come is in C. The primary reason
for this claim comes from the fact that how come does not undergo subject auxiliary inversion,
repeated below.
(13)

a. Why did John leave?
b. How come John left?

Notice that, if how come is actually in C then it would block any further movement to
that head. Collins also argues that the how come in C analysis explains a myriad of other
differences between why and how come. These are presented in the next few sections.
Recall the constraint against long distance construal or LDC for how come. The why
sentence in example (14) has two possible interpretations. You can either intend to ask
about the John’s reasons for saying or Mary’s reasons for leaving. The same is not true for
the how come question. The only interpretation available is one in where you inquire about
John’s reasons for saying.
(14)

a. Why did John say Mary left?
b. How come John said Mary left?

Collins argues that this falls out directly of the assumption that why has the ability to
merge low and bind a wh-trace. Consider the following bracketed representation adapted
from Collin’s work which shows the points in the derivation at which why might be merged
in the lower clause to get the LDC interpretation.
(15) [CP Whyi [ did [TP John say [CP ti [TP ti [TP Mary left ]]]
Notice that in example (15), the indices show the possible merge points for why the lower
clause, according to Collins. Similarly, if how come were to garner a LDC interpretation it
would require that it be merged low as indicated by the trace below.
(16) CP How comei [CP John say [CP [C ti [TP Mary left ]]
Collins claims that the structure is illicit because the movement depicted is a violation of the
head-movement constraint (Travis, 1984) which states that such movement may only occur
157

if no heads intervene.
Collins argues that the complementizer/wh-word distinction between how come and why
respectively, also accounts for the fact that tenseless why clauses are licit while tenseless how
come clauses are not.
(17)

a.
b.

[VP Why [VP go to the store, when there is juice at home?]]
* [ VP How come [VP go to the store?]]

He claims that if we assume that the phrase go to the store in (17a) is a bare VP then why
and not how come has a viable merge point. The claim is that how come is unacceptable in
the same situation because it is a complementizer in C.

5.3.2.2

How come is illicit in multiple wh-questions

Collins also points out that why and not how come may occur in multiple wh questions.
(18)

a.
b.

Why did John eat what?
* How come John ate what?

He assumes Chomsky’s condition on question interpretation which requires that all wh-words
not in CP must be associated with CP and be interpreted as uniformly binding a wh-trace
(Chomsky, 1973).
(19) Condition on Question Interpretation
Assign a wh-phrase not in COMP to some higher structure [COMP +wh] and interpret as in [(20)] where the interpretation is uniform in this COMP node.
(20) All wh-quantifiers interpreted by a given C must bind a trace.
Accordingly, both why and what in example (18a) are thought to end up in COMP and
from this position they will both uniformly bind their traces. Example (18b) on the other
hand is different because how come occupies C and does not bind a trace, therefore it simply
can not be uniformly interpreted at the level of logical form per the condition on question
interpretation. Consider this difference in example (21)

158

(21)

5.3.2.3

a.
b.

[CP Whyj whati [C did] John tj eat ti ]
* [CP whati [C How-come] John ate ti ]

How come does not interact with quantifier scope

Finally, as further evidence that why binds a trace lower in the structure and how come is
base generated in C, Collins notes that why exhibits quantifier scope ambiguities while how
come does not. Consider the following sentence.
(22) Why did everybody hate John?
Example (22) has two interpretations. They are shown in (23).
(23)

a. For each person x, Why does x hate John?
b. Why does the whole group of people hate John?

∀ > WH
WH > ∀

For clarity, I will refer to the interpretation in (23a) as the distributed reading and the
interpretation in (23b) as the collective reading. The overarching assumption that Collins
makes about scope ambiguities is that they are indicative of movement and may be useful
in determining the origins of elements that interact with quantifiers. Specifically he assumes
that why is base generated as adjoined to VP. When it moves to CP it will leave a wh-trace.
Thus a quantifier may scope over the trace of the wh-word. This would only require that the
quantifier is associated with an element which c-commands the wh-trace. This is represented
by the scope relation ∀ > WH associated with example (23a). Likewise, if the quantifier
everybody does not c-command the wh-word at any point then, this gives rise to the collective
reading. This scope relation, represented by WH > ∀, is associated with example (23b).
Now consider example (24), where it is not possible for the quantifier to scope over the
wh-word even with the wh-word why.
(24) Why did John hate everybody?
This sentence can only have the collective interpretation in example (23b). Because the
object everybody never has the possibility of scoping over a wh-trace, the distributed config159

uration is impossible.
Now keeping these facts in mind and turning our attention back to how come we see that
it does not show quantifier/scope ambiguities.
(25)

a. How come everybody hates John?
b. How come John hates everybody?

WH > ∀
*∀ > WH
WH > ∀

For either sentence in (25), there is only the possibility for the collective interpretation. This
is exactly what we would expect if we assume that how come is base generated in C. The
quantifier everybody may be associated with either the subject or object position. Since C
is higher in the structure than both of these positions, how come and quantifiers may never
interact.

5.3.2.4

Against how come as a reduced clause

Collins also argues explicitly against the ‘reduced clause’ analysis of how come suggested
by Zwicky and Zwicky (1971). He argues that speakers who accept instances of how come
occurring with an overt complementizer such as how come that have further grammaticalized
the complementizer as part of the how come constituent. Thus the complementizer that along
with the wh-word is in C. Secondly, he observes that the how portion of how come is not
able to be extracted. This is contrary to what we might think, if how come were truly a
reduced clause. Finally, he argues that though how may be modified by else, it may not
intervene in how come as in the unacceptable *How else come John left.
To sum up the discussion, Collins argues that how come is a syntactic unit base generated
in C. As a wh-complementizer, it never binds a trace and thus never interacts with quantifiers
to create ambiguities which arise from scope ambiguities. In addition, a how come in C
analysis also predicts that subject/auxiliary inversion and multiple wh questions will also be
unacceptable.

160

5.3.3

How come as a secondary wh-adjunct

Ochi (2004) argues that how come in English shares a majority of its properties with whadjuncts like why the hell forming a group that of what he calls secondary wh-adjuncts.
Ochi also attempts to carve out a set of cross-linguistic counterparts to how come which also
share these properties but appear as the lexical item analagous to what. Ochi notes that the
term secondary wh-adjunct is meant to carve out a class for how come, why the hell, and
what used for asking reasons in other languages. Secondary wh-adjuncts as a class do not
behave like normal why adjuncts.
Ochi’s analysis is based closely on the findings of Collins (1991). He adopts the majority
of Collins’ observations repeated below.
(26) How come . . .
a. does not allow long distance construal.
b. does not occur in multiple wh-questions.
c. does not interact with scope to create ambiguities.
However, Ochi rejects the hypothesis that how come is an interrogative complementizer
in C. This is because the properties in example (26) are shared by other objects which are
used for asking reasons like why the hell and its cross-linguistic counterparts, none of which
are possible complementizers in C.

5.3.3.1

Why the hell as a secondary wh-adjunct

Before jumping into the cross-linguistic data, consider the properties of how come as applied
to the aggressively non-d-linked phrase why the hell (see Pesetsky, 1987; Lasnik and Saito,
1994:for relevant discussion). Example (27) shows that though why may be construed long
distance, the phrase why the hell may not be and it patterns with how come.
(27)

a.
b.

Why did you say that John is mad?
Why the hell did you say that John is mad?

161

Ochi notes that the embedded reading Why the hell is John mad? may not be entirely
unacceptable but is definitely not preferred in (27b).
Like how come, the phrase why the hell is unacceptable with multiple wh questions.
(28)

a.
b.

Why did you eat what?
* Why the hell did you eat what?

Finally, why the hell does not yield scope ambiguities with quantifiers.
(29)

a.
b.

Why does everyone hate John?
Why the hell does everyone hate John?

(every > why, why > every)
(*every > why, why > every)

As is indicated by Ochi’s in-line description, example (29b) may only be interpreted as
asking about the collective reason that everyone hates John. The distributed interpretation
is not permitted. Finally, Ochi points out that the phrase why the hell also triggers subject
auxiliary inversion and is therefore not a complementizer. Why the hell must be merged
above C.

5.3.3.2

What as a secondary wh-adjunct across languages

Now I turn to the cross-linguistic data in which the wh-word corresponding to English what
is used to ask, rather emphatically, about the cause of a particular state of affairs. Following
Ochi, I will refer to this cross-linguistic counterpart as what. In German, Hungarian, and
Serbo-Croatian what or was/mit/ˇsta respectively, may be used to ask about cause.
(30) Was schl¨afst du so lange?
what sleeps you so long
‘Why are you sleeping so long?’

(German)

(31)

a. Mit/miert
ulsz itt?
what-acc/why sit-2sg here
‘Why are you sitting here?’
b. Mit/miert
fenyegeted a gyerekeket?
what-acc/why threaten-2sg the kids-acc
‘Why are you threatening the kids?’
ˇ
(32) Zaˇsto/Sta
si
ustao tako rano?
why/what have get up so early
‘Why did you get up so early?’
162

(Hungarian)

(Serbo-Croatian)

Ochi points out that these secondary wh-adjunct what questions are true questions and
not rhetorical questions because they can appear as the complements of verbs that select for
interrogative clauses.
(33)

a. Ich frage mich, was Hans so gestresst ist.
I ask myself what Hans that stressed is
‘I wonder why Hans is so stressed.’
b. Nem tudtuk hogy mit ulsz itt.
not knew-1pl that what sit-2sg here
‘We didn’t know why you’re sitting here.’
c. Zanima
me zaˇsto/ˇsta se Ivan pokunjio.
it-interests me why/what self Ivan got-depressed
‘I would like to know why Ivan got depressed.’

(German)

(Hungarian)

(Serbo-Croatian)

Further, neither the what or why the hell adjuncts are identical in meaning to why.
In use they both carry what Ochi calls an ‘emotional connotation’ in which the speaker is
emotionally affected in some way.
The next relevant properties of what are their abilities to be construed long distance,
occur with multiple wh-questions, and their ability to induce quantifier scope ambiguities.
The acceptability of long distance construal with cross-linguistic what adjuncts in shown
in examples (34) to (36). Each example in the data shows an ambiguous why questions and
a non-ambiguous what question.
(34) German
a. Warum glaubst du, daB er so lange schl¨aft?
why
believe you that he so long sleeps
‘Why do you believe that he sleeps so long?’
b. Was glaubst du, daB er so lange schl¨aft?
what believe you that he so long sleeps
‘Why do you believe that he sleeps so long?’
(35) Hungarian
a. Kati miert gondolta hogy fenyegeted a gyerekeket?
Cathy why thought that threaten-2sg the kids-acc
‘Why did Cathy think that you are threatening the kids?’
b. Kati mit
gondolta hogy fenyegeted a yerekeket?
Cathy what-Acc thought that threaten-2sg the kids-acc
‘Why did Cathy think that you are threatening the kids ?’

163

(ambiguous)

(matrix only)

(ambiguous)

(matrix only)

(36) Serbo-Croatian
a. Zaˇsto Petar tvrdi da se Ivan pokunjio?
why Peter claims that self Ivan got-depressed
‘Why does Peter claim that Ivan is depressed ?’
ˇ
b. Sta
Petar tvrdi da se Ivan pokunjio?
what Peter claims that self Ivan got-depressed
‘Why does Peter claim that Ivan is depressed ?’

(ambiguous)

(matrix only)

The unacceptability of what interrogatives in multiple wh-questions is shown in examples (37) to (39). For these examples I have distilled Ochi’s data down to the minimum
relevant examples, for the full set of data which represents other possible word orderings see
Ochi (2004:37).
(37) Wer schl¨aft warum so lange?
who sleeps why so long
‘*Who is sleeping why so long?’
(38)

(German)

a. Ki
miert fenyegeti a gyerekeket?
who-Nom why threatens the kids
‘Who is threatening the kids why?’
b. *Ki
mit
fenyegeti a gyerekeket?
who-Nom what-Acc threatens the kids

(39) Ko se zaˇsto pokunjio?
who self why get-depressed
‘*Who is depressed why?’

(Hungarian)

(Serbo-Croatian)

what in German and Serbo-Croatian also shows a lack of quantifier interactions. Ochi
notes that this property of secondary wh-adjuncts is not found in Hungarian for independent
reasons.
(40) German
a. Warum seid
ihr
alle gestresst?
Why be-2nd.pl you-2nd.pl all stressed
‘Why are you all stressed?’
b. Was seid
ihr
alle gestresst?
what be-2nd.pl you-2nd.pl all stressed

164

(ok pair-list)

(*pair-list)

(41) Serbo-Croatian
a. Zaˇsto je svako
toliko nervozan danas?
why is everyone so
nervous today
‘Why is everyone depressed today?’
ˇ
b. Sta
je svako
toliko nervozan danas?
what is everyone so
nervous today?

(why > every, every > why)

(why > every, ??every > why)

Finally, Ochi argues that what is a phrasal element in CP because anything to the left
of the verb in V2 languages like German are assumed to be in CP. Moreover, the Hungarian
version of what or mit always appears bearing case, and complementizers do not bear case.

5.3.3.3

The syntax of secondary wh-adjuncts

According to Ochi (2004), how come along with what and what the hell each behave as a
unit and are merged in spec CP. How come differs from why and other wh-words in that its
strength is argued to co-vary with the strength of the interrogative head in a given language.
That is, if C in a given language has strong uninterpretable features then how come and what
should also have strong uninterpretable features. For instance, in English the interrogative
head C is argued to have a strong uninterpretable wh-feature, in turn how come also is
argued to have a strong uninterpretable wh-feature. All aspects of how come’s behavior are
captured via the requirement that it also checks the strong V feature of C. These mechanisms
also capture the behavior of secondary wh-adjuncts as well. In the partial derivation below,
I show how how come is merged into spec-CP along with the requisite feature specifications.
(42)

CP
how-come
V*
u:wh*

C
C
u:V*
wh*

...

The strong feature on the wh-phrase how come ensures that the distribution of how
come is constrained as opposed to why. Ochi finds that these patterns are replicated with
aggressively non-d-linked wh-phrases like why the hell and with what he terms secondary

165

adjunct phrases like HC cross-linguistically. In the next few paragraphs, I walk through the
necessary theoretical assumptions which lead to the structure in example (42).
Ochi assumes that the familiar probe/goal system of feature valuation and the standard
Chomskyan assumption that wh-phrases have an uninterpretable wh-feature which must be
checked off against the interrogative head specifically C in English (Chomsky, 2000). Under
such a system, agree establishes a long distance relation between the wh-phrase and C. This
relationship values the uninterpretable wh-feature of the wh-phrase and the uninterpretable
Q or question feature on C. The movement of the wh-phrase to the spec,CP is commonly
attributed to the strong feature strength of the interrogative head. This is illustrated below
for the embedded clause in the sentence I wonder who Mary shot.
(43)

(44)

CP

who
u:wh

C
TP

C
u:Q*

CP

Mary

C

VP
V
shot

TP

C
u:Q*
Mary
DP
whoi
u:wh

VP
V DP
shot ti

Ochi amends this basic configuration with the following proposal.
(45) Amendment to feature strength
a. A strong feature must be checked off...
i. by the next operation and...
ii. in a local configuration
(46) Amendment to the nature of wh-phrases
a. Regular adjunct and arguement wh-phrases
i. The strength of the wh-feature is [-strong].
b. Secondary adjunct wh-phrases
i. The strength of the wh-feature is [α strong] and co-varies with the feature
strength of the interrogative head for a given language.

166

Ochi argues that secondary wh-adjuncts are specified as [+strong]. As such, the feature
must be checked off in the next operation and in a local configuration. This essentially bars
how come from being merged anywhere but in the specifier of CP. Merger of how come in
any position in the derivation which is not a local configuration of CP would lead to an
unacceptable derivation.
From these assumptions, Ochi derives all of the constraints on the syntax of how come
and the other secondary wh-adjuncts. Consider the standard assumptions about the features
and their strength along with Ochi’s amendments to those assumptions below.
(47)
Interrogative H
primary wh-phrases/adjuncts
secondary wh-adjuncts

Chomsky (2000)

Ochi (2004)

C[u:Q*]
who/why[u:wh]

C[u:Q*]
who/why[u:wh]
how come[u:wh*]

Thus we are left again with the structure from the beginning of this section with a very
similar featural configuration. The lack of long distance construal is ruled out straightforwardly by the fact that how come can only be merged directly into spec,CP.
(48)

a.

b.

CP

how-come
u:wh*

C
C
u:Q*

CP

...

C
C
u:Q*

...

If how come were to be merged in a lower argument position then, by Ochi’s amendment
to feature strength and the assumption that secondary wh-adjuncts have a strong uninterpretable wh-feature, the derivation would crash directly after merge. Under either Chomsky’s
or Ochi’s system, wh-phrases originating outside of CP must first agree with a higher interrogative head and then move into a spec-head configuration as a result of the strong feature
on that head. This option is not available in the case of secondary wh-adjuncts. How come
requires that its strong features be satisfied upon merge, and the only way to achieve this is
by direct merger into the specifier of CP.
167

The lack of interaction with quantifiers falls out from similar assumptions. Because how
come is only mergable in the specifier of CP, it has no trace which quantifiers may scope
over.
The ban on how come in multiple wh-questions falls out from the strict constraint on
possible merge sites imposed by the strong features on both C and the secondary wh-adjunct.
I turn to Ochi’s explanation. Take the phrase I wonder who left. The embedded portion of
this sentence has the following structure.
(49)

(50)

CP

CP
whoi
u:wh

C
TP

C
u:Q*
who
u:wh

C
TP

C
u:Q*
ti

T

T

T VP

T VP

left

left

This illustrates how normal wh-movement works under Ochi’s system. Crucially, the
wh-word who has a weak uninterpretable wh-feature and the derivation proceeds as normal.
Regular multiple wh-word phrases work in a similar fashion. Consider the derivation of the
phrase Who bought what?.
(51)

C
TP

C
u:Q*
DP
Who
u:wh

VP
V
bought

DP
what
u:wh

At this point in the derivation the uninterpretable question feature on C will probe into the
derivation and agree with both wh-words in (52). Next, the strong uninterpretable question
168

feature on C requires that some wh-word satisfies its uninterpretable features directly in a
local spec-head relationship shown in (53).
(52)

C
TP

C
u:Q*
DP
Who
u:wh

(53)

VP
DP
what
u:wh

V
bought

CP
DPi
Who
u:wh

C
TP

C
u:Q*
ti

VP
V
bought

DP
what
u:wh

All features are checked and the derivation may converge.
The crucial case for us is found in the unacceptable derivation of *How come John bought
what. I will adopt Ochi’s examples directly. At the point where C is merged two things
might happen, (i) either C agrees with the embedded wh, in which case how come can not
be merged without causing the strong feature of C to not be checked of in the next step of
the derivation, or (ii) how come is merged satisfying the strong features both on how come
and C, leaving the wh-features on the embedded wh-phrase unvalued.
Following Ochi’s reasoning, let’s consider the first case. Consider the derivation of *How
come John bought what at the point where C is merged with TP. In example (54), C has
agreed with the wh-feature of the embedded wh-phrase as indicated by strike-out on the
feature.

169

(54)

*

C
TP

C
u:Q*

VP

John

V
bought

DP
what
u:wh

With regards to the constraints imposed by the strong Q feature on C, if the next operation does not involve movement of the embedded wh-word to spec,CP then the derivation
will crash. As for the second case, if we were to merge how come as in example (55), the
wh-feature on the embedded wh-word remains unvalued and crash the derivation.
(55)

*

CP
How come
u:wh*

C
TP

C
u:Q*
John

VP
V
bought

5.3.3.4

DP
what
u:wh

Accounting for the lack of auxiliary inversion

Ochi’s assumptions thus far do not account for the fact that how come fails to trigger subject
auxiliary inversion. Ochi theorizes that subject/auxiliary inversion is the result of a strong
uninterpretable V feature on C. Thus in the case of normal cases of subject/aux inversion,
C has attracted the closest V feature in the derivation; specifically, those contained on T.
This is illustrated below.

170

(56)

a.

b.

C

C

TP

C
u:V*
subj

TP

C
shouldi
i:V

T
T
should
i:V

...

C
u:V*

subj

T
T
ti

...

With regards to how come, he speculates that the combined word how come has properties
not only of wh-phrases but of verbs as well. Thus if how come is in the specifier of CP the
local spec-head relationship would suffice to satisfy the strong V feature of C. How come
checks off two features of the interrogative head: a strong V feature as well as the strong Q
feature. Thus we can update the features on the structural representation of the insertion
of how come in example (48) below in example (57).
(57)

a.

b.

CP

How-Come
u:wh*
i:V

C
C
u:Q*
u:V*

CP

...

C
C
u:Q*
u:V*

...

Thus Ochi’s analysis accounts for all of the properties of finite how come in English and
sets them equal in behavior to what phrases in other languages and why the hell phrases in
English.

5.3.3.5

The issue of finiteness

An extension of Ochi’s analysis to Appalachian English is unsatisfactory for two main reasons.
First, such an extension does not provide any further explanation about how the syntax
differentiates between the finite and non-finite forms of how come. Second, it fails to predict
the unacceptabiligy of non-finite how come in Standard English. To be fair, no one could
171

have expected Ochi’s analysis to account for non-finite how come to without knowledge of
it.
For the sake of argument, consider a straightforward extension of Ochi’s analysis to nonfinite how come through the addition of the necessary features to come. Such an analysis
requires that there are two lexical items how come and SAE speakers have the finite version while AppE speakers have the non-finite version. Moreover, the secondary wh-adjunct
analysis fails to explain why non-finite how come is not available in Standard English in
any principled way. Nothing about the Ochi’s analysis constrains how come to being base
generated solely in the specifier of finite C as opposed to non-finite C.
Ochi suggested that the ban on subject-auxiliary inversion is a result of the retention of
a verbal feature on come in how come. This verbal feature would satisfy strong V features
on T itself rather than have them satisfied by auxiliary inversion. Since we don’t usually
assume that specifiers are in the business of selecting the complements for phrases that
they occupy, one might imagine that non-finite how come is a single syntactic unit and has
whatever features are associated with root clauses in addition to this verbal feature. Such
a configuration would only require that the unit how come has retained some c-selectional
feature of the matrix clause which requires agreement with a non-finite C. At this point we
should ask ourselves, what is the difference in stipulating a feature that grants clause selection
properties and simply positing a type of reduced clause. I would argue that such a feature
may only be justified by showing unequivocally that non-finite how come is a unanalyzable
syntactic unit and is in fact not bi-clausal. Support for this claim even requires more evidence
that we have previously seen, because as I will argue below, none hallmark properties of how
come actually preclude a bi-clausal analysis.
This is exactly how this line of questioning is similar to the set of questions raised by
restructuring predicates. Restructuring is concerned with clauses that were traditionally
analyzed as two clauses behaving like one. Minimally, non-finite how come forces us to
assume that the phrase how come contains features that license root phenomena and select for

172

embedded clauses. The case of non-finite how come seems to indicate that a phrase in what
has been analyzed as a single clause seems to lend it the properties of two clauses. The main
piece of evidence for this is that there is no evidence that English ever licenses independent
matrix infinitives which are not root infinitives. Again framed by restructuring the question
returns, what is the difference between a phrase which undoubtedly has matrix features
and a syntactic unit which contains the necessary features to license root phenomena. For
these reasons, I will argue that there is only ground to be gained by rejecting the secondary
wh-adjunct analysis of non-finite how come and reviving a reduced clause analysis.

5.4

Analysis

The analysis has the following structure. First, I will discuss the basic properties of non-finite
how come and show that it behaves like finite how come in addition to providing evidence
that non-finite how come selects for a non-restructuring complement. Following discussion
of the properties of non-finite how come, I will argue that a proper syntactic analysis of nonfinite how come requires that we revive the bi-clausal or reduced clause analysis; a proposal
with origins in Zwicky and Zwicky (1971). Revival of the bi-clausal analysis means that nonfinite how come is not analyzable as a syntactic unit. Thus I will investigate the relevant
properties of the matrix clause, including properties of come, the location of how how, and
the general structure of the matrix clause. I will then use the proposed syntax to derive all
of the core properties shared between finite and non-finite how come.

5.4.1

Properties of non-finite how come

Though we have seen analyses for finite how come, there are no syntactic analyses of nonfinite how come. Oddly enough, we will see that the basic properties of these clauses seem
to be the same; a fact which has its own implications for existing analyses of how come.
Non-finite how come differs in behavior from how come only in finiteness. Both forms share

173

all of the core properties that have been described thus far in the previous literature on how
come. With respect to restructuring, the infinitival-to complement of non-finite how come
is just what it seems; a non-finite complement maximally headed by CP. This distinction
is important for several reasons. It shows us that the complement is not a reduced complement and hence not a restructuring predicate under the mono-clausal approach followed
throughout this work. It is also a necessary point to make in terms of distinguishing what
appears to be a matrix infinitive and the host of properties and phenomena associated with
root infinitives. With regard to the restructuring theme of this work, non-finite how come
is of interest because I will show that it appears to be a reduced matrix clause. Given the
fundamental assumption that restructuring is about reduced clauses, non-finite how come
allows us to examine one possible configuration of a reduced matrix clause. This is important for theories of restructuring because they almost entirely focus on embedded clauses.
In this analysis, I will revive the reduced clause approach based on the fact that non-finite
how come appears to have clausal properties in that it can select for a non-finite clause.

5.4.1.1

Non-finite how come behaves like finite how come

Non-finite how come, being comprised of a non-finite complement, does not exhibit subject
auxiliary inversion. This is expected as infinitival-to never exhibits subject/aux inversion in
English.
(58)

a.
b.

How come them to leave?
* How come to them leave?

The sentences in example (58) show that it is unacceptable for infinitival-to to appear in a
position consistent with the result T to C movement.
Like how come, non-finite how come does not exhibit long distance construal.
(59) How come John to say Mary left.
a.
What is the reason for John’s saying that Mary left.
b. * What is the reason that Mary left.

174

Notice that in example (59) non-finite how come can only be associated with the matrix
clause and never the embedded clause.
Non-finite how come is not licensed in multiple wh-questions, while other wh-words may
appear in multiple wh-questions.
(60)

a.
b.

* How come John to eat what?
Who bought what?

Finally, non-finite how come does not exhibit scope ambiguities with quantifiers.
(61) How come everybody to hate John?
What is the sole reason that everyone hates John?
What are all the reasons that each person hates John?

WH > ∀
*∀ > WH

Crucially, only the collective interpretation is available under non-finite how come. The
distributed interpretation is unavailable. In conclusion, non-finite how come shares these
core properties with finite how come.

5.4.1.2

Non-finite how come selects non-restructuring complements

Since I intend to revive what is essentially a bi-clausal analysis of non-finite how come, it
is necessary to show that the embedded clause is actually a full non-restructuring clause.
Showing that the embedded clause of non-finite how come is virtually identical to finite how
come clauses, means that any possible explanation of the differences between how come and
non-finite how come must be a difference in the syntactic structures represented by the phrase
how come in both cases. The infinitival-to in the complements of non-finite how come clauses
is not optional. Recall from the first chapter that in some restructuring complements AppE
allows the insertion of an infinitival-to form which is the result of a default morphological rule
and not the result of normal to-insertion on T. Keep in mind that the default infinitival-to
is always optional.
(62)

a.
b.

How come them to leave?
* How come them leave?
175

Thus verbs embedded under non-finite how come are not reduced in a restructuring sense.
At this point, evidence suggests that they are at least headed by T.
As we would expect, sentential negation is available and is licensed by T. Application of
the previously used not even a little continuation as a test for sentential negation shows us
that sentential negation is available.
(63)

a. How come them not to leave?
How come they didn’t leave?
b. How come her not to drink this whiskey?
. . . not even a little.
c. How come she didn’t drink this whiskey?
. . . not even a little.

AppE
SAE
AppE
SAE

For the sake of clarity of distinction between sentential negation and constituent negation,
compare the judgments above with non-sentential negation presented below.
(64)

How come her to not drink one shot but two?
# . . . not even a little.

Further evidence of a tense layer is found in the presence of expletive subjects. These
subjects are in fact possible as the subject of non-finite how come clauses.
(65) How come there to seem to be a problem?
Finally, complementizers are marginally acceptable with non-finite how come in AppE.
This suggests that non-finite how come occurs to the left of a complementizer located in C.
(66)

? How come for him to drink all my bourbon?

Based on the data in this section it is rather clear that the complement of non-finite how
come is a non-restructuring complement in the sense of Wurmbrand (2001).

5.4.2

The syntax of non-finite how come

Combining the findings of the previous two sections leaves us with the following tentative
structure in example (67). It reflects the notion that non-finite how come clauses are non176

restructuring. They are headed by a CP. This indicates that the phrase how come appears
in some capacity at the left edge.
(67)

?
How come

C
TP

C
(?for)
themi

T
vP

T
to
ti

v
v

VP
V
leave

Moving beyond this tentative structure, I will argue piece by piece for one possible structure for the matrix clause of non-finite how come sentences. Non-finite how come is analyzed
as a reduced matrix clause which licenses a non-finite CP complement. This analysis is primarily supported by the fact that when how come takes a non-finite clause in Appalachian
English it must have root properties in addition to verbal properties. I propose a vestigial
V2 syntax as a base for the possibly reduced structure that non-finite how come has now.
This will account for its peculiar form. Such an analysis is compatible with all properties
shared between finite and non-finite how come while simultaneously predicting the lack of
intervening auxiliaries. I ultimately believe that non-finite how come as it exists today in
Appalachian English is indicative of an earlier stage of grammaticalization on a trajectory
somewhere between a full phrase and a single lexical item. As such, it provides unique insight
into the properties of reduced matrix clauses. It also accounts for preliminary evidence that
non-finite how come is in fact separable as compared with finite how come.

177

5.4.2.1

Reviving the reduced matrix clause analysis

Before we can successfully begin to posit a syntax for non-finite how come, it is necessary
to make some assumption about the phrase that it is reduced from. I will assume, based on
intuitions of native speakers who were asked to elaborate on non-finite how come, that the
non-finite how come phrase in example (68) is reduced from something like the sentences in
examples (68a) and (68b).
(68) How come them to leave.
a. How did it come for them to leave.
b. How came it for them to leave.
I have included example (68a) as a more modern paraphrase of the sentence in example (68b)
which I believe is most probable origin phrase for non-finite how come clauses. For the rest
of this section, I will assume that non-finite how come is a reduction of example (68b) which
I take to be an example of verb second come. This assumption will allow us to account for
the general covertness of the complementizer for, as well as the lack of auxiliaries. Further,
a V2 syntax provides the possibility that the lack of overt subjects in the matrix clause is
linked to null expletives in older Englishes.

5.4.2.2

Case assignment and the embedded subject

Recall the introductory data in example (69).
(69)

a.

b.

How come him to leave the party?
WH
3s.acc inf leave the party
‘How come he left the party?’
* How come he to leave the party?

Note that accusative case on the matrix subject is required.
(70)

* How come he to leave the party?

From the requirement of accusative case on the subject of non-finite how come clauses
we can infer case assignment from a higher head. Under common minimalist assumptions

178

accusative case is either assigned clause internally by C0 or by matrix v. Thus we begin our
investigation with two possible options for accusative case assignment in non-finite how come
clauses. The case assigning head has implications for the structure of the matrix clause.
(71)

a.

CP
C
TP

C
(for)
acc

him
ucase:acc

T
to leave the party?

b.

vP
v
v
comei
acc

VP
V
TP

V
ti
him
ucase:acc

T
to leave the party?

In example (71a), case is assigned from C to the embedded DP in subject position, while
example (71b) illustrates ECM or exceptional case marking from matrix v. In what follows,
I show that subjects of non-finite how come are not receiving accusative case from matrix v
but rather from C which contains null for as in Martin (2001)

5.4.2.3

Properties of come

Come appears in various syntactic distributions and has many peculiar properties. It appears
as a raising verb, it has vestigial verb second forms, an unaccusative usage, and it may appear
as an semi-functional restructuring verb. However, come never assigns accusative case. If

179

we make the reasonable assumption that the lexical item come in non-finite how come is
related to at least one of the other types of come then accusative case assignment must be
somewhat independent of the verb itself. Thus, example (71a) where case is assigned by C,
is the only possible case assignment configuration for non-finite how come.
The example below shows that come behaves identically to canonical raising verbs seem.
As a raising predicate come does not assign accusative case.
(72)

a.
b.
c.

There seemed to be a problem around the house.
He seemed to be a problem around the house.
* Him seemed to be a problem around the house.

(73)

a.
b.
c.

There came to be a problem around the house.
He came to be a problem around the house.
* It came him to be a problem around the house.

Notice that just like the raising verb in example (72), the verb come in example (73) licenses
expletive there and only licenses nominative case. This indicates that it does not assign a
theta-role to its external argument and that the internal argument receives nominative case
from matrix T0 .
Vestigial verb second come never appears with pronouns regardless of their case in either
Standard American English or Appalachian English.
(74) Here comes the bourbon/*him/*he/*it!
Moreover, only the nominative pronouns are possible with the non-V2 variants of these
sentences. This is what we would expect of an unaccusative verb.
(75)

a. Here he/*him comes!
b. There she/*her goes!

Finally, aspectual come never seems to take internal arguments (see Wurmbrand, 2001),
though the verb directly under it may.
(76)

a.
b.
c.

You should come drink with us!
* You should come this time drink with us!
You should come drink it with us!
180

Assuming there is a similarity between come in non-finite how come and other types
of the verb come, accusative case licensing is not a property of the reduced matrix clause
in non-finite how come constructions. It is a property of the embedded clause itself. This
is consistent with the body of data which shows come is now a raising predicate in both
dialects and that accusative case is not assigned by the matrix verb. This however does not
preclude different instances of come from selecting a case assigning C head or not.
Taking the case assigning properties of come in English into account, the verbal shell of
non-finite how come with its complement is as follows. Note that since there is no evidence
for a case assigning vP in these clauses, I will omit it completely for simplicity.
(77)

VP
CP

V
come

C
(for)
acc

TP
him
ucase:acc

T
to leave the party?

5.4.2.4

Verb second and null expletives

There is one particular state of affairs found in the history of English that simultaneously
explains the lack of auxiliaries and the apparent lack of a matrix subject in non-finite how
come sentences. Verb second phenomena seemed to be intrinsically tied to the ability of the
grammar to license null expletives (Haeberli, 2002). Haeberli also notes that according to
Hulk and van Kemenade (1995; 1997) the loss of empty expletives coincides with the loss of
Old and Middle English type V2. Thus it is not so far a reach to claim that if a language
retains some relic V2 structures, we might expect them to also retain null expletives in those
structures as well. Finally, the converse of Haeberli’s proposal that the loss of V2 in English
is dependent upon the loss of null expletives would mean that the presence of a null expletive
subject in so far as it was retained in any modern English dialect should tend to require V2
181

ordering. I argue that this is exactly what we see with non-finite how come.
This claim is tentative and given the fact that this is not a work of historical linguistics,
I do not intend for it to be the definitive structural analysis of non-finite how come. Rather,
I intend to make some connections about the possible nature of the subjects of non-finite
how come to historical work which supports the analysis provided here, especially given the
mounting evidence which supports a bi-clausal analysis and the confounding opacity of how
come forms.
This aside, assuming null expletives, the next logical step is to add a TP to our structure
which will provide matrix tense and house the expletive subject.
(78)

TP
proexpletive

T
VP

T
comei
V
ti

CP
C
TP

C
(for)
acc

him
ucase:acc

T
to leave the party?

Assuming this null subject designating this structure as one which exhibits relic verb raising accounts for the fact that no auxiliaries intervene between how and come. Next, the
interrogative head C is merged.

182

(79)

C
TP

C
uQ*

proexpletive

T
VP

T
comei

CP

V
ti

C
C
(for)
acc

TP
him
ucase:acc

T
to leave the party?

Finally, the wh-word how is merged in the spec,CP to satisfy the Q feature on C and the
wh-feature on how.
(80)

CP
How
uwh

C
C
comei
uQ*

TP
proexpletive

T
VP

T
ti V

CP

ti

C
C
(for)
acc

TP
him
ucase:acc

T
to leave the party?

As I mentioned, it is entirely possible that the actual structure of non-finite how come is a
reduced form of this clause. One could imagine that whatever its structural interpretation
183

some or all of the derivationally opaque movements, heads, or features found in this structure
have undergone reanalysis and no long exist in such a highly articulated form.

5.4.2.5

The location of causal how : evidence from Chinese

Tsai (1999) shows that use of Chinese zenme ‘how’ in a sentence only results in a question
about causation when it is base generated above T. Since, how of non-finite how come is
not analyzed as a single lexical item under my analysis, it is necessary to provide some
evidence that it should only be merged in CP given that it always results in a question
about causation.
Consider Tsai’s data from Chinese. In example (81), zenme is interpreted as asking about
the manner or way in which the eventuality proceeded.
(81)

a. Akiu zenme zou?
Akiu how leave
’By what means/way will Akiu leave?’
(Answer: by bus, by interstate 15, etc.)
b. Akiu (pingchang) zenme shuijiao?
Akiu usually
how sleep
’With what style does Akiu (usually) sleep?’
(Answer: faceup, facedown, on one side, etc.)

However, when perfective aspect is added to the predicate, the manner/style reading dissappears and zenme may only be interpreted as causal. This is consistent with Tsai’s eventual
analysis where he argues that when zenme unambiguously appears to the left of elements
which modify T they are consistently interpreted as being causal.
(82)

a. Akiu zenme zou-le?
Akiu how leave-Prf/Inc
’Why has Akiu left?’
b. Akiu zenme shuijiao-le?
Akiu how sleep-Prf/Inc
’Why has Akiu slept?’

This basic difference is also supported by the fact that causal zenme scopes over modal verbs
and sentential adverbs and manner zenme scopes under those same sentential elements. For
184

the sake of clarity and succinctness, I will only show the distinction using modal verbs but
Tsai (see 1999:pp 4-10) for an exhaustive list and explanation of the data.
(83) Akiu zenme hui/bixu/neng/keyi/yinggai zou?
Akiu how will/must/can/may/should leave
a. # ‘By what means will/must/can/may/should Akiu leave?’
b. # ‘With what style will/must/can/may/should Akiu leave?’
c.
‘Why would/must/can/may/should Akiu leave?’
(84) Akiu hui/bixu/neng/keyi/yinggai zenme zou?
Akiu will/must/can/may/should how leave
a.
‘By what means will/must/can/may/should Akiu leave?’
b. # ‘With what style will/must/can/may/should Akiu leave’
c. # ‘Why will/must/can/may/should Akiu leave?’
Tsai also notes that these distinctions also hold for the Chinese lexical item weishenme
‘why’ as well. Therefore, if non-finite how come is truly bi-clausal it seems likely that how is
merged in spec,CP. This is also supported by the fact that how in non-finite how come only
ever exhibits a causal interpretation.

5.4.3

Deriving the core properties

A clausal analysis of non-finite how come captures all of the core properties that have been
mentioned in the previous literature on how come. Interestingly, this indicates that we need
some principled way by which we can differentiate the two forms and the fact that some
dialects simply do not allow non-finite how come. Recall that finite and non-finite how
come share all properties except finiteness, this includes the ban on long distance construal,
multiple wh-questions, and quantifier scope ambiguities.
First, consider the case of long distance construal. Neither non-finite how come nor
how come exhibit ambiguity in terms of whether they may be construed with the matrix or
embedded clause, while why does exhibit this ambiguity.

185

(85)

a. How come John to say Mary left the party.
b. How come John said Mary left the party.
c. Why did John say Mary left the party.
i. What are the reasons for John’s saying?
ii. What are the reasons for Mary’s leaving the party?

why/how come
why/*how come

As I will show below, this property is instantly explained for non-finite how come under the
reduced clause approach. Under this assumption, non-finite how come is a tensed matrix
clause which selects for an infinitival complement.
(86) [CP How come [CP John to say [CP Mary left the party]]].
This accounts for the fact that it may be base generated along with its infinitival complement
under another finite clause or appear as a root clause, just as a normal tensed clause would.
(87)

a.
b.
c.
d.

How come Mary to say Bill left.
John wondered how come Mary to say Bill left.
Mary left.
John said Mary left.

Once we conceive of non-finite how come as constituting a separate clause, the reason
for the ban on multiple wh-questions becomes apparent. Although the matrix clause in
non-finite how come is interrogative it never selects for an interrogative complement. In
effect, the same logic that was used to account for the ban on multiple-wh questions with
finite how come is applicable to the ban on multiple wh-questions with non-finite how come.
Reconsider the case of the unacceptable How come John bought what, in light of the bi-clausal
approach. In the following structure, I have omitted the functional structure of the matrix
clause between the two CPs for the sake of space.

186

(88)

CP
C
C
come
u:Q*

...
CP
C
C
(for)

TP
John to buy what u:wh

Since only the matrix clause is interrogative the only possibility for the embedded whword to satisfy its wh-feature lies with the matrix CP. However, just as Ochi explained for
finite how come there are two possible options for proceeding and both of them crash the
derivation. If the matrix C probes down the tree to establish an agree relation with the
embedded wh-feature and triggers wh-movement in order to satisfy said features in a local
relationship, the result is a wh-argument question and not a wh-adjunct question because
how come is left in the numeration. On the other hand, if we arrive at the stage in the
derivation depicted in example (88) and merge the adjunct how directly in spec,CP then
the wh-feature on the embedded wh-word is left unsatisfied and crashes the derivation.
Keep in mind though that for normal multiple wh-word questions like Who bought what,
I follow Ochi in assuming that probe from C will simultaneously match all uninterpretable
wh-features which exist in the derivation. The crucial piece of timing in Ochi’s analysis must
still apply.
Quantifier scope ambiguities also fall out of a bi-clausal analyis. Recall that sentences
such as example (89) exhibit quantifier scope ambiguities with why. Example (90) shows
that non-finite how come does not exhibit those ambiguities.
(89) Why does everyone hate John.
What is the sole reason that everyone hates John?
What are all the reasons that each person hates John?

187

WH > ∀
∀ > WH

(90) How come everybody to hate John?
WH > ∀
*∀ > WH
This fact is explained because the quantified argument originates in the embedded clause
and how of non-finite how come is base generated in the specifier of the interrogative matrix
CP. There is simply no way for the quantifier to scope over any position in the matrix clause.
To sum up, a reduced clause approach to non-finite how come allows us to account for all
of the core properties that how come and non-finite how come share under the assumption
that come in non-finite how come selects for a for-to complement.

5.4.3.1

Evidence for the reduced root clause analysis

One nagging question remains. If non-finite how come truly constitutes its own clause why
are intervening elements generally disallowed? To rephrase the critique of the reduced clause
analysis mentioned in Collins (1991) why can the modifier else which is generally allowed
to modify how not modify it in the case of how come? I argued above that the unique
requirement of verb raising was enough to preclude any modal or aspectual auxiliaries in
non-finite how come cases but what about how else? It is worth noting that examples (91)
and (92) illustrate that how else is not used for questioning causation and as such it is
possible, following Tsai (1999), that else only modifies manner how which originates below
TP.
(91) How else did they drink the whiskey?
a. * Because John made them.
b.
...Like a bunch of jerks.
(92) How else did they leave?
a. * Because the room was to smoky.
b.
...Through the back door as well.
Thus we would not expect for else to be able to intervene between either finite or non-finite
how come.

188

I would like to offer one piece of evidence in support of the reduced matrix clause analysis.
There seems to be a distinction between finite and non-finite how come in terms of whether
they allow secondary adjunct-like forms.
(93)

a.
b.
c.
d.

* How
How
* How
How

the
the
the
the

hell come they left?
hell come them to leave?
fuck come they left?
fuck come them to leave?

Assuming this distinction between the two forms would indicate that finite how come is a
single syntactic unit but that non-finite how come is not. As such the secondary wh-adjunct
analysis would still apply to finite how come, and the reduced clause analysis would apply
to non-finite how come especially since non-finite how come may itself occur with secondary
wh-adjuncts.

5.5

Conclusion

In this chapter, I have analyzed the syntax of non-finite how come. The analysis indicates
that non-finite how come is a bi-clausal structure because it exhibits root properties and
licenses a non-finite embedded clause. I have argued that the matrix clause portion of nonfinite how come sentences is a V2 relic which contains an expletive pro subject. This analysis
not only accounts for the constrained form of the matrix clause but also accounts for the fact
that finite and non-finite how come share all properties except finiteness. Finally, I offered
some evidence from secondary wh-adjuncts to support my claim that the matrix clause of
non-finite how come clauses is truly a clause.

189

BIBLIOGRAPHY

190

BIBLIOGRAPHY

Baker, M., K. Johnson, and I. Roberts. 1989. Passive arguments raised. Linguistic Inquiry
20:219–251.
Bennis, H, and T Hoekstra. 1989/2004. Why kaatje was heard not sing a song. In Arguments
and structure: studies on the architecture of the sentence, ed. Harry van der Hulst Henk van
Riemsdijk, Jan Koster, chapter 2, 125–144. Mouton de Gruyter.
Borer, Hagit. 1984. Parametric syntax . Foris Publ.
Burzio, Luigi. 1986. Italian Syntax: A government-binding approach, volume 1. D. Reidel
Publishing Company.
Butters, Ronald R, and Kristin Stettler. 1986. Existential and Causative have... to. American
Speech 61:184–190.
Calabrese, Andrea. 1997. Distributed morphology and sentential complementation in the
Salentino dialect of Italian. Harvard Working Papers in Linguistics 6:47–85.
Chomsky, Noam. 1973. Conditions on transformations. In A festschrift for Morris Halle,
ed. Stephen Anderson & Paul Kiparsky, 232–286. New York: Holt, Rinehart & Winston.
Chomsky, Noam. 1995. The minimalist program. Cambridge, Mass: The MIT Press.
Chomsky, Noam. 2000. Minimalist inquiries: the framework. Step by step. Essays on minimalist syntax in honor of Howard Lasnik . Cambridge, Mass.: MIT Press.
Chomsky, Noam. 2001. Derivation by phase. In Ken Hale: A life in language, ed. Michael
Kenstowicz, 1–52. Cambridge, MA: MIT Press.
Cinque, Guglielmo. 2004. Restructuring and functional structure. In Structures and Beyond.
The cartography of syntactic structures, ed. Adriana Belletti, volume 3, 132–191. Oxford
University Press.
Collins, Chris. 1991. Why and how come. MIT working papers in linguistics 15.
Culicover, Peter W. 1999. Syntactic nuts: Hard cases, syntactic theory, and language acquisition, volume 1. Oxford University Press New York.
Davidson, D. 1967. The logical form of action sentences. In The Logic of Descision and
Action, ed. N. Rescher. University of Pittsburgh Press.
Diesing, Molly. 1992a. Bare plural subjects and the derivation of logical representations.
Linguistic Inquiry 23:353–380.
Diesing, Molly. 1992b. Indefinites, volume 20 of Linguistic Inquiry Monograph. MIT Press,
Cambridge, Mass.
191

Embick, David. 2004. On the structure of resultative participles in English. Linguistic
Inquiry 35:355–392.
Embick, David, and Rolf Noyer. 2007. Distributed morphology and the syntax/morphology
interface. In The oxford handbook of linguistic interfaces, ed. Gillian Ramchand & Charles
Reiss, 290–322. Oxford: Oxford University Press.
Feagin, Crawford. 1979. Variation and change in alabama english: A sociolinguistic study of
the white community. Georgetown University Press.
Felser, Claudia. 1999. Verbal complement clauses: A minimalist study of direct perception
constructions. Amsterdam and Philadelphia: John Benjamins.
Frantz, D. 1977. A new view of to-contraction. volume 21 of Work Papers of the Summer
Institute of Linguistics, 71–76. University of North Dakota.
Freeze, Ray. 1992. Existentials and other locatives. Language 68:553–595.
Goodall, Grant. 1991. Wanna-contraction as restructuring. In Interdisciplinary approaches
to language: Essays in honor of S.-Y. Kuroda, ed. S. Kuroda, Carol P. Georgopoulos, and
Roberta L. Ishihara, 239–254. Kluwer Academic.
Goodall, Grant. 2006. Contraction. In The blackwell companion to syntax , ed. Martin
Everaert and Henk C. v. Riemsdijk, volume 19, 688–703. Malden, Mass.: Wiley-Blackwell.
Haeberli, Eric. 2002. Inflectional morphology and the loss of verb-second in English. In
Syntactic effects of morphological change, ed. David Lightfoot, 88–106. Oxford University
Press.
Halle, Morris, and Alec Marantz. 1993. Distributed morphology and the pieces of inflection.
In The view from building 20: Essays in linguistics in honor of Sylvain Bromberger , ed.
Ken Hale and Samuel Keyser, 111–176. Cambridge, MA: MIT Press.
Harley, Heidi. 1997. Logophors, variable binding and the interpretation of have. Lingua
103:75–84.
Higginbotham, James. 1985. On semantics. Linguistic Inquiry 16:547–93.
Hoekstra, Teun. 1986. Passives and participles. In Linguistics in the Netherlands, ed. Frits
Beukema and Aafke Hulk, Algemene Vereniging voor Taalwetenschap (Netherlands), 95–
104. AVT publications.
Hornstein, N., A.M. Martins, and J. Nunes. 2008. Perception and causative structures in
English and European Portuguese: Feature agreement and the distribution of bare and
prepositional infinitives. Syntax 11:198–222.
Hulk, Aafke, and Ans van Kemenade. 1995. Verb second, pro-drop, functional projections and
language change. In Clause structure and language change: Oxford studies in comparative
syntax , ed. Ian G. Roberts Adrian Battye, 227–256. Oxford University Press.

192

Hulk, Aafke, and Ans van Kemenade. 1997. Negation as a reflex of clause structure. Amsterdam Studies in the Theory and History of Linguistic Science Series 4 183–208.
Kayne, Richard. 1989. Facets of Romance past participle agreement. In Dialect variation
and the theory of grammar , ed. P. Beninca, 85–103. Dodrecht: Foris.
Kayne, Richard. 1993. Toward a modular theory of auxiliary selection. Studia Linguistica
47:3–31.
Kayne, Richard. 1994. The antisymmetry of syntax , volume 25 of Cambridge, MA. The MIT
Press.
Klima, Edward. 1964. Negation in english. In The structure of language: Readings in the
philosophy of language, ed. Jerry Fodor and Jerrold Katz. Englewood Cliffs, NJ: Prentice
Hall.
Kratzer, Angelika. 1994. The event argument and the semantics of voice. Manuscript.
Kratzer, Angelika. 1995. Stage-level and individual-level predicates. In The generic book ,
ed. Gregory N. Carlson and Francis Jeffry Pelletier, 125. Chicago: University Of Chicago
Press.
Kyt¨o, M., and S. Romaine. 2005. “We had like to have been killed by thunder & lightning”:
The semantic and pragmatic history of a construction that like to disappeared. Journal
of Historical Pragmatics 6:1–35.
Labov, William. 1972. Language in the Inner City: Studies in the black english vernacular .
Philadelphia, Pennsylvania: Univ. of Pennsylvania Press.
Labov, William. 1996. When intuitions fail. In 32nd regional meeting of the Chicago Linguistics Society, volume 32, 76–106.
Lasnik, Howard, and Mamoru Saito. 1994. Move [alpha]: conditions on its applications and
output. The MIT Press.
Marantz, Alec. 1988. Clitics, morphological merger, and the mapping to phonological structure. In Theoretical morphology, ed. M. Hammond and M. Noonan, 253–270. San Diego:
Academic Press.
Marantz, Alec. 1989. Clitics and phrase structure. In Alternative conceptions of phrase
structure, ed. Mark R. Baltin and Anthony S. Kroch, 99–116. Chicago: University of
Chicago Press.
Martin, Roger. 2001. Null case and the distribution of PRO. Linguistic Inquiry 32:141–166.
McCawley, J.D. 1971. Prelexical syntax. In Report of the 22nd Roundtable Meeting on
Linguistics and Language Studies, ed. Richard J. O’Brian, S.J., volume 24 of Monograph
Series on Language and Linguistics, 19–33. Washington, D.C.: Georgetown University
School of Language and Linguistics.

193

Mittwoch, Anita. 1990. On the distribution of bare infinitive complements in English. Journal
of Linguistics 26:103–131.
Morzycki, M. 2001. Almost and its kin, across categories. In Proceedings of Semantics and
Linguistic Theory, SALT 11 .
Morzycki, Marcin. 2011. Metalinguistic comparison in an alternative semantics for imprecision. Natural Language Semantics 19:39–86.
Munn, Alan. 2013. Some observations on participle levelling. Talk given at University of
Chicago, June 2013.
Nevins, Andrew, and Jeffrey Parrott. 2007. A Variable impoverishment analysis of leveling
variation in English be and -n’t. University of York: Formal Approaches to Variation in
Syntax conference.
Nevins, Andrew, and Jeffrey K Parrott. 2010. Variable rules meet impoverishment theory:
Patterns of agreement leveling in english varieties. Lingua 120:1135–1159.
Nouwen, R. 2006. Remarks on the polar orientation of almost. In Linguistics in the Netherlands, volume 23, 162–173. Dordrecht: John Benjamins Publishing Company.
Nunes, J. 1995. The diachronic distribution of bare and prepositional infinitives in English.
In Historical Linguistics 1993: Selected papers from the 11th international conference on
historical linguistics, 357. John Benjamins Publishing Company.
Ochi, M. 2004. How come and other adjunct wh-phrases: A cross-linguistic perspective.
Language and Linguistics 5:29–57.
PBS. 2013. Slavery and the Making of America. The slave experience: Legal rights and
government. John Henry Faulk, circa 1930.
Penka, Doris. 2006. Almost there: The meaning of Almost. In Proceedings of Sinn und
Bedeutung 10 , ed. Christian Ebert and Cornelia Endriss, 275–286. Berlin: Zentrum fur
Allgemeine Sprachwissenschaft (ZAS).
Pesetsky, D., and E. Torrego. 2004. Tense, case, and the nature of syntactic categories. In
The syntax of time, ed. Gu´eron, J. & J. Lecarme. MIT Press, Cambridge, Mass.
Pesetsky, David. 1987. Wh-in-situ: Movement and unselective binding. In The representation
of (in)definiteness, ed. Eric J. Reuland and Alice G. B. ter Meulen, volume 98, 98–129.
Cambridge, MA: MIT press.
Picallo, M Carme. 1990. Modal verbs in Catalan. Natural Language & Linguistic Theory
8:285–312.
Pollock, J.Y. 1989. Verb movement, universal grammar, and the structure of IP. Linguistic
Inquiry 20:365–424.

194

Postal, Paul M, and Geoffrey K Pullum. 1982. The contraction debate. Linguistic Inquiry
13:122–138.
Pullum, G.K. 1997. The morpholexical nature of english to-contraction. Language 79–102.
Raposo, E.P. 1987. Romance infinitival clauses and case theory. In Studies in romance
languages, ed. Neidle & R. A. Nunes Cedeno, 237–249. Dordrecht: Foris.
Rapp, I., and A. von Stechow. 1999. Fast ‘almost’ and the visibility parameter for functional
adverbs. Journal of Semantics 16:149.
Rawlings, Marjorie Kinnan. 1938/1985. The yearling. 3. Atheneum Books.
Reinhart, Tanya, and Eric Reuland. 1993. Reflexivity. Linguistic Inquiry 24:657–720.
Ritter, E., and S.T. Rosen. 1993. Deriving causation. Natural Language & Linguistic Theory
11:519–555.
Ritter & Rosen. 1997. The function of have. Lingua 101:295–321.
Roberts, I. 1997. Restructuring, head movement, and locality. Linguistic Inquiry 423–460.
Roberts, Ian. 1987. The representation of implicit and dethematized subjects, volume 10 of
Linguistic Models. Mouton De Gruyter.
Sadock, Jerrold. 1981. Almost. In Radical pragmatics, ed. Peter Cole, 257–271. New York:
Academic Press.
Schmitt, Cristina J. 1996. Aspect and the syntax of noun phrases. Doctoral Dissertation,
University of Maryland.
Siddiqi, Daniel. 2010. Distributed morphology 1. Language and Linguistics Compass 4:524–
542.
Stuart, Ruth McEnery. 1899. In simpkinsville.... Character Tales. New York: Harper &
Bros.
Travis, Lisa deMena. 1984. Parameters and effects of word order variation. Doctoral Dissertation, Massachusetts Institute of Technology, Cambridge, MA.
Tsai, W-T Dylan. 1999. The hows of why and the whys of how. UCI Working Papers in
Linguistics 5:155–184.
Wolfram, W., and D. Christian. 1976. Appalachian Speech. Center for Applied Linguistics,
Arlington, Virginia.
Wurmbrand, Susanne. 2001. Infinitives: Restructuring and clause structure, volume 55 of
Studies in Generative Grammar . Berlin and New York: Mouton de Gruyter.
Zanuttini, Raffaella. 1996. On the relevance of tense for sentential negation. In Parameters
and functional heads: Essays in comparative syntax , ed. Adriana Belletti and Luigi Rizzi,
Oxford Studies in Comparative Syntax, 181–207. New York: Oxford University Press.
195

Zwicky, A.M., and A.D. Zwicky. 1971. How come and what for. In Working papers in
linguistics, ed. Elliot, D et. al, volume 8, 173–185. Ohio State University.

196