MINIMIZE EXPONENCE: ECONOMY EFFECTS ON A MODEL OF THE MORPHOSYNTACTIC

MINIMIZE EXPONENCE: ECONOMY EFFECTS ON A MODEL OF THE MORPHOSYNTACTIC
MINIMIZE EXPONENCE:
ECONOMY EFFECTS ON A MODEL OF THE MORPHOSYNTACTIC
COMPONENT OF THE GRAMMAR
by
Daniel Siddiqi
_________________________
A Dissertation to the Faculty of the
DEPARTMENT OF LINGUISTICS
In Partial Fulfillment of the Requirements
For the Degree of
DOCTOR OF PHILOSOPHY
In the Graduate College
THE UNIVERSITY OF ARIZONA
2006
2
THE UNIVERSITY OF ARIZONA
GRADUATE COLLEGE
As members of the Dissertation Committee, we certify that we have read the dissertation
prepared by Daniel Siddiqi
entitled Minimize Exponence: Economy Effects on a Model of the Morphosyntactic
Component of the Grammar
and recommend that it be accepted as fulfilling the dissertation requirement for the
Degree of Doctor of Philosophy.
_______________________________________________________________________
Date: 5/10/06
Andrew Carnie, Committee Chair
_______________________________________________________________________
Date: 5/10/06
Michael Hammond
_______________________________________________________________________
Date: 5/10/06
Heidi Harley
_______________________________________________________________________
Date: 5/10/06
Simin Karimi
Final approval and acceptance of this dissertation is contingent upon the candidate’s
submission of the final copies of the dissertation to the Graduate College.
I hereby certify that I have read this dissertation prepared under my direction and
recommend that it be accepted as fulfilling the dissertation requirement.
________________________________________________ Date: 5/10/06
Dissertation Director: Andrew Carnie
3
STATEMENT BY AUTHOR
This dissertation has been submitted in partial fulfillment of requirements for an
advanced degree at the University of Arizona and is deposited in the University Library
to be made available to borrowers under rules of the Library.
Brief quotations from this dissertation are allowable without special permission, provided
that accurate acknowledgment of source is made. Requests for permission for extended
quotation from or reproduction of this manuscript in whole or in part may be granted by
the head of the major department or the Dean of the Graduate College when in his or her
judgment the proposed use of the material is in the interests of scholarship. In all other
instances, however, permission must be obtained from the author.
SIGNED: Daniel Siddiqi
4
ACKNOWLEDGEMENTS
This dissertation could not have been written without the support of a host of different
people throughout my graduate years at the University of Arizona. First among this list is
my wife, Julianna, who followed me out to the desert and endured four long years of
seclusion from her family, constant submersion in a field she couldn’t care less about,
and hundred plus degree summers just so that I could fulfill my dreams. The last year
and a half must have been especially hard for her given the sign I’ve worn across my
forehead reading: WE’RE SORRY, THIS MARRIAGE HAS BEEN INTERUPTED SO
WE CAN BRING YOU THIS LONG PERIOD OF DISSERTATING. Julie has been
with me through all of the good times and the hard times of my graduate career.
Everything herein is made both possible and worthwhile by her constant love and
support.
What finally ended up on the page here, in particular, has come about through the efforts
of several different groups of people (with significant overlap of course). The
germination for the ideas presented here came in the form of a term paper I submitted for
Heidi Harley’s Linguistics 522. I want to thank Heidi for encouraging me to take this
particular project on and turn it into my dissertation. Were it not for her input, I likely
would have produced a much more boring dissertation (Old English inflection was my
original idea...shudder). That paper was developed into my preliminaries paper and then
was part of my orals examination. I want to thank all the members of those committees
including Andy Barss, Andrew Carnie, Dick Demers, Heidi Harley, Simin Karimi, and
Andy Wedel whose comments and feedback on earlier versions of these ideas made it
possible for me to expand them to the work here. I also want to thank Adam Ussishkin,
who, while not on any of these committees, taught the prelim course, Ling 697, and had
his hands in the mix as well.
The people most responsible for this work are, of course, my dissertation committee:
Andrew Carnie, Heidi Harley, Mike Hammond, and Simin Karimi. I could not have
asked for more support, feedback, and inspiration than I received from these great people.
Any student would be lucky to have a committee such as mine—I could be quoted as
saying, “I have the perfect committee”, and I meant it. If it weren’t for the constant
efforts of these four people, what ended up on the following pages would be largely
unreadable and, honestly, embarrassing.
In particular, I want to single out Andrew Carnie, my advisor and chair of all the
committees listed above. Andrew has been like a rock for me and has guided me in my
evolution from a disenfranchised English major to a real linguist. This, and all work I do
for the rest of my career in linguistics, is due in large part to Andrew’s efforts.
I also want to thank all the people who have worked in the framework of Distributed
Morphology, whose work in and development of DM made this dissertation possible. I
5
want to acknowledge David Basilico, David Embick, Morris Halle, Heidi Harley (again),
Alec Marantz, and Rolf Noyer.
It is my firm belief that the single most important part of a graduate education is the
interaction and support of your contemporaries. There aren’t words to describe my debt
to my peers in the graduate program at Arizona, so I won’t even try. Some people I want
to thank in particular are my graduate student elders Cathy Hicks Kennard and Meghan
O’Donnell who served as mentors throughout my career here. Also among the list of
elders whose aid was invaluable to me are Jason Haugen, Erin O’Brien, Tania Granadillo,
Jianyuan Wang, and Scott Jackson. There are two more groups of my colleagues that I
feel I need to acknowledge, as they were crucial to my development here (again, there is
overlap). First, I want to acknowledge my incoming class: Shannon Bischoff, Erin
Good, Emily Kidder, Naomi Ogasawara, and Lynnika Butler. These are the fellows that
were by my side each step of the way and I am indebted to them for their support.
Second, I want to thank the Geek Group (the activities of which will be omitted here due
to high school fears of derision) whose membership over the years has included Scott
Jackson, Dave Medeiros, Peter Richtsmeier, Jaime Parchment, Emily Kidder, Paul Saito,
Heather Jackson, Jeff Punske, Michael Anderson, Hannah Soreng, Erin Good, and Erin
O’Brien. Paradoxically, the Geek Group was at once crucially important to my finishing
and the biggest impediment. Plenty of other friends have come along in the four years
I’ve been in Tucson and among them I want to thank Benjamin V. Tucker, Jordan
Brewer, Polly O’Rourke, Bob Kennedy, Lika Chtareva, Stacy Oberly, Yosuke Sato,
Azita Taleghani, and Alina Twist.
The department here at the UA has been spectacular. I want to thank them for all their
efforts and for their financial support. Faculty I haven’t mentioned yet but that I want to
thank include Diana Archangeli, D. Terrence Langendoen, and Massimo PiattelliPalmarini. I want to give a special thanks to the great staff of the department including
Marian Wisely, Jennie Bradley, Jennifer Columbus, and Rosemary Emery.
I also want to thank my family for everything they have done for me and for tolerating
my (now ten-year) absence from home. Thanks to my mother, Ann Jaline, and my
brother Jamal Siddiqi. Also thanks to my new family, Kerry Motts, Cathy Hays, Don
Hays, Brad Loomis, Jeremy Motts, and Max Loomis.
There are two people whose contributions to my life were before my coming to Tucson
but are nonetheless critical. First, Elmaz Abinader, one of my advisors in a previous life
who noticed I was miserable as an English major and encouraged me to pursue
linguistics, a field I had never heard of.
Finally, I want to thank Jennifer Farley Scharff. My life and all my accomplishments are
a result of her inspiration.
6
DEDICATION
To my wife, Julianna. I’m sorry I’ve been an absentee husband while working on this
and that you had to join the “Spouses of Dissertators” support group. It’s done now.
This dissertation is as much yours as it is mine.
7
TABLE OF CONTENTS
LIST OF FIGURES................................................................................................... 10
ABSTRACT............................................................................................................... 11
CHAPTER 1: INTRODUCTION............................................................................. 12
CHAPTER 2: DISTRIBUTED MORPHOLOGY.................................................... 17
Section 1: The Structure of the Distributed Morphology Grammar................... 17
Section 1.1:
Late Insertion......................................................................... 18
Section 1.2:
Morphosyntactic Decomposition........................................... 19
Section 1.3:
Underspecification................................................................. 21
Section 1.4:
Why reject Lexicalism?......................................................... 22
Section 1.5:
Construction of an Utterance................................................ 26
Section 1.6:
Spellout, Competition, and the Subset Principle................... 28
Section 1.7:
The Distinction between Functional and Contentful............. 35
Section 1.8:
Lexical Categories in DM...................................................... 37
Section 1.9:
Special Meanings and the Encyclopedia................................38
Section 2: Morphological Operations in DM: From Spellout to PF.................. 41
Section 2.1:
Morphological Merger........................................................... 41
Section 2.2:
Fission and fusion.................................................................. 42
Section 2.3:
Readjustment rules................................................................. 44
CHAPTER 3: ON A THEORY OF ROOT ALLOMORPHY................................. 46
Section 0.1:
Licensing and Readjustment in DM...................................... 47
Section 1: Roots in the Grammar........................................................................ 49
8
TABLE OF CONTENTS—Continued
Section 1.1:
Licensing Insertion.................................................................51
Section 1.2:
Alternative Analysis of Root allomorphy.............................. 56
Section 1.3:
Competition of Roots Revisited............................................. 59
Section 2: Differences between This and the Traditional DM Grammar........... 63
Section 2.1:
Readjustment Rules............................................................... 63
Section 2.2:
Licensing................................................................................ 67
Section 2.3
Null Heads............................................................................. 68
Section 3: Expansion of the Fusion Analysis......................................................71
Section 3.1:
The ¬ Specification............................................................... 74
Section 3.2:
Enter MINIMIZE EXPONENCE.................................................. 77
Section 4:
Inflection in Compounds....................................................................83
Section 4.1:
MINIMIZE EXPONENCE Analysis............................................. 89
Section 5: Summary............................................................................................ 96
CHAPTER 4: SUBCATEGORIZATION WITH MINIMIZE EXPONENCE........ 100
Section 1: Subcategorization............................................................................... 100
Section 1.1:
Subcategorization within DM................................................ 105
Section 2: MINIMIZE EXPONENCE-based account of Subcategorization..............107
Section 2.1:
Determining the Feature Specification of a VI...................... 111
Section 2.2:
Differences among Subcategorization Models...................... 118
Section 2.3:
Evidence for the Feature [v] in English................................ 120
Section 2.4:
Blocking “extra” arguments................................................. 121
9
TABLE OF CONTENTS—Continued
Section 3: Rampant Polysemy........................................................................... 125
Section 4: Structural Coercion............................................................................ 130
Section 5: Dative Alternations........................................................................... 133
Section 6: C-selection......................................................................................... 137
Section 7: Summary........................................................................................... 139
CHAPTER 5: REMAINING ISSUES...................................................................... 141
Section 1: Event Semantics................................................................................. 142
Section 1.1:
Parsons (1990)....................................................................... 144
Section 1.2:
Neo-Davidsonianism Meets Syntax...................................... 148
Section 1.3:
Remaining Questions............................................................ 150
Section 2: Typological Predictions.................................................................... 153
Section 3: Verb Classes..................................................................................... 156
Section 4: On Feature Specification and the Elsewhere Principle...................... 158
Section 5: Summary............................................................................................ 160
CHAPTER 6: CONCLUSIONS............................................................................... 162
Section 1: Some Final Thoughts………………………………………………. 166
REFERENCES.......................................................................................................... 168
10
LIST OF FIGURES
Figure 1:
The structure of the DM Grammar.................................................... 28
11
ABSTRACT
Working within the morphosyntactic framework of Distributed Morphology (DM, Halle
and Marantz 1993, 1994) within the Minimalist Program (Chomsky 1995), this
dissertation proposes a new economy constraint on the grammar, MINIMIZE EXPONENCE,
which selects the derivation that realizes all its interpretable features with the fewest
morphemes. The purpose of this proposal is to capture the conflicting needs of the
grammar to be both maximally contrastive and maximally efficient.
I show that the constraint MINIMIZE EXPONENCE has a number of effects on
analyses of morphosyntactic phenomena. I propose that, in order to satisfy MINIMIZE
EXPONENCE, the roots in a derivation fuse with the functional heads projected above
them, resulting in a simplex head that contains both a root and interpretable features.
Following the tenets of DM, this head is now a target for the process of Vocabulary
insertion. Since the target node contains both content and functional information, so too
can Vocabulary Items (VIs) be specified for both types of information. This allows VIs
such as eat and ate to compete with each other. This competition of forms linked to the
same root allows for a new model of root allomorphy within the framework of DM. In
this model of root allomorphy, following proposals by Pfau (2000), VIs that realize roots
participate in competition in the same was as do VIs that realize abstract morphemes.
Since root VIs are participating in competition and are specified for both content and
formal features, the need for licensing through secondary exponence as proposed by
Harley and Noyer (2000) is removed from the framework. Further, since eat and ate in
this model are different VIs with different specifications that compete with each other for
insertion, this model of root allomorphy also eliminates the need for readjustment rules as
proposed by Halle and Marantz (1993, 1994) and elaborated on by Marantz (1997). This
new model of root allomorphy allows for an account of the blocking of regular inflection
in English nominal compounds (e.g. *rats-catcher), which was problematic for theorists
working with DM, given the tenets of the framework.
I also show that the fusion of roots and functional elements driven by MINIMIZE
EXPONENCE allows for a new account of subcategorization. The model of
subcategorization presented here falls out of the following facts: 1) arguments are
introduced by functional heads; 2) those heads fuse with the root they are projected
above, resulting in the node containing both the root and the features of the functional
heads; 3) since the root now contains both the root and the formal features, the
corresponding VI can be specified for both; 4) VIs that realize roots can also be specified
for compatibility or incompatibility of the features of the functional heads that license
argument structure. The result here is an underspecification model of subcategorization
that predicts a number of behaviors of verbs with respect to their argument structure that
it is difficult for a full specification model to account for. Those include polysemy (I ran
the ball to Mary) and structural coercion (I thought the book to Mary).
12
CHAPTER 1: INTRODUCTION
The study of the language faculty must address a central conflict about the grammar: On
one hand, we have the need to convey a message and the need for that message to be as
clear as possible. On the other hand, we need our message to be as efficient as possible.
These opposing forces, contrast and efficiency, are the driving force behind a host of
phenomena we see in language. For example, a number of phonological processes such
as place or voice assimilation make utterances easier to say while processes such as
dissimilation make the contrasts more transparent.
This conflict is prevalent in much of the linguistic literature. For example, in the
realm of phonology and prosodic morphology, Optimality Theory (Prince and Smolensky
1993) captures this conflict using two classes of constraints on the grammar: faithfulness
constraints ensure that contrastive meanings of an utterance aren’t destroyed by the need
to be efficient while markedness constraints capture that need for efficiency. In such a
model, a grammar with all faithfulness constraints ranked above markedness constraints
would result in a language that is maximally contrastive but is a mouthful to use, while
the opposite ranking would result in a language that is maximally efficient but has
eliminated the vast majority of its contrasts—making it unable to actually convey any
meaning.
In historical linguistics, these forces are seen in cases where a language changes
to make itself more efficient or to make itself more contrastive. For example, the loss of
a sound in a language means a more efficient (easier) system at the cost of losing a
contrast (e.g. the loss of the E sound in some dialects of English results in the loss of the
13
contrast between the words pin and pen). Similarly, adding a contrast (e.g., some dialects
have added a new form for the second person plural, you guys, y’all, or youse, to contrast
against the second person singular you) makes a system more contrastive at a cost of
efficiency.
In many ways, language is shaped by this central struggle between contrast and
efficiency. However, in the realm of syntax (and by extension morphosyntax), this
struggle is largely ignored by theoreticians. A grammar that was maximally contrastive
would not have structural ambiguity (e.g. I saw the man with the telescope) yet, would
result in pronouncing every single functional head and every single formal feature in a
maximally contrastive manner (e.g. one feature = one morpheme). On the other hand,
maximum efficiency would entail summing all the features of an utterance into one word
(e.g. Yup.).
Since the Minimalist tradition is composed of models of language competence
rather than performance, it’s not at all surprising that these competing forces are
marginalized. The study of syntax is largely a question of what the system is and isn’t
capable of, not how that system is put to use. However, this results in a certain loss of
predictive power. Limiting the discussion to the maximal pronunciation of formal
features, there are immediately two concerns relevant to a model of UG that need to be
discussed. The first is the more obvious question: why don’t languages pronounce more
functional morphemes (thus realizing more formal features)? Interpretable features,
especially in languages like English, are largely unpronounced, despite the fact that,
according to the Minimalist tradition, they must be present in each derivation. Second,
14
why do languages differ in the amount of the interpretable features that are realized by
morphology and how they are realized?
Minimalist tradition has included a number of economy constraints on the
grammar whose purpose is to select the most economical derivation measured in energy
used to create a derivation (e.g. greed, procrastinate, shortest move, etc, see Chomsky
1989, Rizzi 1990, Adger 1994). The purpose of this dissertation is to propose a new
economy constraint: one that selects the most economical derivation measured in energy
used to produce it.
We can create an economy constraint that captures the competing forces on the
grammar—in particular the balance necessary in pronouncing all the interpretable
features of a given derivation in the most efficient way possible. I propose the following
constraint:
(1.1)
MINIMIZE EXPONENCE
The most economical derivation will be the one that maximally realizes all the
formal features of the derivation with the fewest morphemes.
The gist of this constraint is that the best utterance is the one that conveys the most
amount of information with the least effort (measured in number of morphemes that have
to be pronounced). In terms of the production of an utterance, this constraint captures the
struggle between the need to be maximally contrastive and the need to be maximally
efficient.
15
This dissertation focuses on exploring the role that this type of constraint would
have on the model of the grammar by specifically looking at the effects of MINIMIZE
EXPONENCE on analyses of familiar morphosyntactic phenomena.
This dissertation is couched within the morphosyntactic framework of Distributed
Morphology (DM), proposed in the early 1990’s by Halle and Marantz. Chapter 2 is
intended to provide the reader with the requisite understanding of DM for following the
major claims of this dissertation. Section 1 provides a survey of DM, specifically
focusing on how it is different from Lexicalist Minimalism. Section 2 outlines those of
the morphological processes available to DM that this dissertation employs.
Chapter 3 proposes an analysis of root allomorphy (e.g. mouse/mice) within the
framework of DM that showcases the economy constraint MINIMIZE EXPONENCE.
Chapter 3 also accomplishes two other things: First, it proposes revisions to how the
framework of DM explains root allomorphy. Second, it provides an analysis of nominal
compounds in English (e.g. mousetrap). Nominal compounds in
English uniquely disallow certain kinds of inflectional morphology (e.g. *rats-catcher).
While this phenomenon has been treated in the Lexicalist tradition, a proper treatment of
it in DM is problematic. Section 4 of Chapter 3 uses MINIMIZE EXPONENCE to solve this
theory internal problem.
Chapter 4 extends the analysis of the effects of MINIMIZE EXPONENCE beyond the
realm of morphology and into the realm of syntax. Chapter 4 provides an analysis of
subcategorization within the tenets of DM and characterizes the effects that MINIMIZE
EXPONENCE has on a model of subcategorization within DM.
16
Finally, Chapter 5 is a remaining issues chapter, discussing several topics that,
due to the focus of this dissertation, are not large enough to warrant their own chapter but
still deserve discussion. First, MINIMIZE EXPONENCE has an interesting potential effect
on the syntax’s interface with event semantics. Since the rest if this dissertation focuses
mainly on English data, Section 2 of Chapter 5 details some of the typological predictions
of the inclusion of MINIMIZE EXPONENCE in UG. Section 3 considers a model of verb
classes in a MINIMIZE EXPONENCE model of the grammar. Section 4 provides a
discussion on the nature of features specification and the Elsewhere Condition in light of
MINIMIZE EXPONENCE and the feature blocking system proposed in chapters 3 and 4.
17
CHAPTER 2 DISTRIBUTED MORPHOLOGY
This chapter provides a brief overview of the framework of Distributed Morphology.
Section 1 describes the structure of the grammar in DM as opposed to that Lexicalist
Minimalism. Section 2 describes all the morphological operations at work in DM.
2.1.
The Structure of the Distributed Morphology Grammar
Distributed Morphology (henceforth DM, Halle and Marantz 1993, 1994) provides a
proposal concerning the structure of the grammar—in particular, the lexicon, the
interface with syntax, and constraints on semantics—as an alternative to other current
models of Universal Grammar (UG). In recent models of the grammar within the
tradition of Government and Binding Theory (GB, Chomsky 1981), UG is structured in
such a way that the lexical module precedes the syntax in the derivation and feeds the
syntax words for manipulation. In particular, the GB approach is similar to many other
models of the grammar (such the Lexical Functional Grammar (Bresnan 2001, Falk
2001) and the Head-driven Phrase Structure Grammar (Pollard and Sag 1994)) in
assuming that the lexicon is a generative component of the grammar, independent from
any generative syntactic component. The lexicon in GB, following Chomsky (1970), is
formulated as a word-building component that is structured independently of the syntax.
The Lexical Hypothesis (Chomsky 1970) assumes that the words that are fed to the
syntax come out of the lexicon fully formed, regardless of whether or not they are
multimorphemic or simplex.
18
Distributed Morphology is a framework within the Minimalist program which
rejects the Lexicalist hypothesis and the notion of a generative lexicon. This position is
discussed below in Section 1.1 of this chapter. In DM, there is only one generative
component of the grammar, the syntax, whereas in Lexicalist Minimalism, there are two:
the syntax and the lexicon. The three key differences between DM and Lexicalist
Minimalism to be discussed here are late-insertion, morphosyntactic decomposition, and
underspecification.
2.1.1
Late-insertion
In DM, unlike in GB and its Lexicalist derivatives, rather than manipulating fully
formed words, the syntax only manipulates abstract formal features to generate syntactic
structures. These morphosyntactic features (such as [plural] and [past]) are selected from
a fixed list of abstract features (or feature bundles) rather than being selected from the
output of a generative lexicon. The late insertion hypothesis (Halle and Marantz 1994)
holds that the phonology which represents the morphological features manipulated by the
syntax is provided at PF rather than being present throughout the derivation. At spellout,
syntactic terminals in DM are entirely comprised of interpretable features (including
roots). Only once all syntactic processes are finished with the structure is phonological
content added.
This phonology is provided by a component of the grammar called the
Vocabulary. The Vocabulary is a static list of items whose function in the grammar is to
provide phonology to realize the interpretable features contained in the terminal nodes of
19
a derivation so that that derivation can be pronounced. Individual items within this list
are called Vocabulary Items (or VIs for short). These VIs represent the basic
sound/meaning correspondences of a language. Thus, the Vocabulary is the inventory of
signs available to the language, which can be used to represent the featural nature of the
syntax.
For example, imagine that syntactic operations result in a tree containing a
terminal node comprised of the features [present], [singular], and [3rd person] (Each
terminal node must be spelled out by some VI or other, including this 3psgPres one). In
English, those three features are realized by the affix –s. This means there is an item in
the Vocabulary, -s, which is inserted into that node at spellout to realize those features
with the overt phonology /z/ (2.1).
(2.1)
[present]
[singular]
[3rd person]
↔
-s
/-z/
Since the phonology of any given derivation is added after spellout, Distributed
Morphology is considered a late-insertion model. This crucial difference between DM
and Lexicalism reduces to the point at which the phonological material is added to the
derivation.
2.1.2
Morphosyntactic Decomposition
One of the strengths of the Distributed Morphology framework is the parallel
between syntactic structure and morphological structure. Since the grammar of DM
manipulates only syntactic features, the complex structure of a word is created in the
20
same way as is the complex structure of a sentence. Spelling out a complex constituent
of the syntax as a “phrase” or a “word” depends on the nature of VIs in the structure. For
example, a complex NP such as dogs of war has obvious syntactic structure, but, DM
claims, so does wardogs. To show a more complex example, consider the word
grammaticalization. According to the tenets of DM, grammaticalization is a complex
noun phrase (notated in DM as nP, since the dominating constituent is a projection of the
functional nominalizer n—read ‘little-n’) composed of a nominalizing head and the verb
phrase grammaticalize. That verb phrase is itself composed of a verbalizing head, -ize,
and an adjective phrase grammatical, and so on (2.2).
(2.2)
grammaticalization
nP
3
vP
3
AP
-ize
3
nP
-ical
3
grammatø
-ion
The reason that the complex verb phase grammaticalization is spelled out as one complex
word rather than a phrase containing isolated functional morphemes (such as of) is
entirely a result of the available inventory of the English Vocabulary and the application
of morphological processes (such as head movement and adjunction, see below). The
VIs realizing the functional heads of A, v, and n happen to be affixes, which adjoin to
their complements. The same structure in another language (such as an isolating
21
language like Mandarin where the word grammaticalization1` is instead three isolated
morphemes resulting in yu fa hua—literally “language” “law” “ization” 1 ) could very
well use free morphemes rather than bound morphemes to realize the same syntactic
structure.
Since morphological structure derives from syntactic structure, lexical
decomposition and phrase structure are identical—i.e. both are the result of the
application of Merge in the syntactic component of the grammar. In the literature of DM,
this transparency of morphosyntactic structure has been called “syntactic hierarchal
structure all the way down” (Harley and Noyer 1999) or the “Pervasive Syntax
Perspective” (Haugen 2004).
2.1.3
Underspecification
Distributed Morphology uses underspecification in the insertion of Vocabulary
Items into a terminal node of the syntax. The insertion of a VI is governed by the subset
principle (see below), which allows for a VI with certain specifications to be inserted into
any node that satisfies those specifications, regardless of whether or not it exceeds those
specifications. For example, consider the English coplula are, an example I will return to
throughout this chapter. Are can appear in 1st person plural present tense, 2nd singular
present, 2nd plural present, and 3rd plural present. The distribution of the VI, are, is
attributable to the fact that its specification—just the feature [present]—is a subset of all
four environments.
1
Thank you to Jian Wang for this Mandarin example.
22
Many of the incarnations of the Lexicalist hypothesis, including Lexicalist
Minimalism, HPSG, and LFG, are models that require full specification of all lexical
entries. For example, in Lexicalist Minimalism, a word in the numeration is fully
inflected and always contains the exact set of features that the syntax needs to check—
otherwise the derivation crashes. This set is always the same and represents all and only
the features that that word realizes.
Since DM uses underspecification, it makes very different predictions from
Lexicalist models as to the behavior of syntactic elements (e.g. polysemy and
syncretism). It also predicts a vastly smaller number of required lexical items since one
item can fill many different roles.
2.1.4
Why reject Lexicalism?
The most cogent defense of Lexicalism is Chomsky’s (1970) “Remarks on
Nominalization.” Chomsky argues that there must be a generative lexicon to account for
the two different types of nominalizations that occur in English: gerunds and derived
nominals.
Gerunds can be formed productively from a subject-predicate form whereas
derived nominals can’t always be (e.g. *John’s growth of tomatoes is not allowed though
John’s growing tomatoes is). Gerunds exhibit a regular relationship between the
meaning of the verb and the meaning of the gerund whereas the same relationship
between the derived nominal and its corresponding verb is not always as regular (e.g.
laughter, marriage, construction, actions, activities, revolution, etc.). Finally, gerunds do
23
not have the same structure as an NP (for example, the possessive cannot be replaced by
a determiner such as that, the, or no).
Chomsky argues that there are two different possible ways for the grammar to
account for this data: either we could suppose that there is a generative lexicon that
creates these nominals before the syntactic component or we could build these
components in the syntax—for example, by allowing the syntax to build verb phrases and
then subjecting them to a transformation such as "Nominalization", thereby making the
syntax more complex. Ultimately he decides that the Lexicalist position is correct for a
number of reasons: a) derived nominals are not wholly productive and the syntax is, so
the syntax must not be responsible for creating nominals; b) since the relationship of the
meaning of the verb and the meaning of the derived nominal is often idiosyncratic, it is
better to claim that this relationship is stored rather than created by the syntax; and c)
since derived nominals behave like nouns not verbs (i.e. they cannot carry aspect) it is
better to claim that they are not derived from VPs but are rather just nouns with verb-like
selectional restrictions.
Marantz (1997b) approaches the conclusions Chomsky makes in “Remarks”
(1970) through the eyes of contemporary grammatical theory, specifically the Minimalist
Program (Chomsky 1995). Marantz uses nominals like growth and laughter to show that
the lexical transformations of causitivization and nominalization actually occur in the
syntax. For example—if grow is causativized in the Lexicon, then there is no way of
ruling out Nominalization being applied to the causativized grow as well as the
24
inchoative grow. If causitivization and nominalization are in the syntax instead of the
lexicon, however, the ungrammaticality of John's growth of tomatoes arrises because:
a) √grow does not inherently take an agent argument
b) it can get an agent argument only through becoming a verb, via the agentintroducing causative head (v)
c) becoming a verb and becoming a (root-derived) noun are incompatible
d) ergo growth can't have an agent.
There need not be a relationship between the structure of a derived nominal and that of
the corresponding verb. Marantz claims that the relationship between words like receive
and reception is not that of a transformation from one to the other, but actually of a
linking to a common root, √RECEIVE, which is unspecified for a category. Rather, it
becomes a noun or a verb based on the environment it appears in. In a verbal
environment, the root is realized as receive, but in a nominal environment the root is
realized as recept- (see section 2.1.4 below for details). Marantz shows that the
apparently idiosyncratic and somewhat contrary behavior of some nominalizations
relative to the corresponding verbs is an effect of the combination of the functional
projections above the root (for example, the variety of little-v projected above the root)
and the class of that root (see Marantz 1997 for more details). In short, Marantz shows
that Chomsky did not need to create a generative lexicon to capture the data. The
syntactic mechanisms necessary to capture the differences between gerunds and derived
nominals have already independently developed. With the syntactic mechanisms already
in place, there is no need to propose a second grammatical engine for the composition of
words.
25
Part of the Lexicalist Hypothesis is that the word has a special status in the
grammar. In particular, the word is the locus for special sound processes, special
structural process, and idiosyncratic non-compositional meaning. Since the word has all
these special attributes, it follows that the language faculty contains a component
dedicated to the construction and storage of just words. Marantz (1997b) refutes each of
these “special” qualities attributed to the word.
The special status of the word derives from the fact that in phonology the prosodic
word is one of the crucial units or levels of representation (Marantz 1997b). Lexicalism
entails that the prosodic word correlates to the basic units needed for the syntax (i.e.
lexical items). However, this claim is only unilateral. While syntacticians have assumed
that the prosodic word and the “lexical item” overlap, phonologists make no such claim.
Within the realm of prosodic phonology and morphology (cf. Kiparsky 1982, Prince &
Smolensky 1993), in general there is no empirical evidence to show that syntactic
structure is identical to prosodic structure. Marantz claims that zero-level units are often
too small a unit for lexical phonology and at other times those same zero-level syntactic
units contain complex prosodic structure. Marantz further argues that, even if the
prosodic word and the zero-level unit did happen to be identical, there is no a priori
reason for that to show that the phonological formation of the prosodic word must happen
before the syntax unless LF was in some way sensitive to that level of the phonology.
Marantz also disputes that the lexicon is the location of idiosyncratic
sound/meaning correspondences. Chomsky’s assumption is that the words are the largest
unit of structure that contain idiosyncratic meaning, whereas all structures larger than the
26
word are constructed by the syntax and have wholly predictable meanings as a
consequence. Marantz points to Jackendoff (1996) who reiterates the evidence that there
is crucially no empirical difference between the idiosyncratic meaning of words and the
idiosyncratic meanings of idiom phrases (such as kick the bucket). DM recognizes that
there can be idiosyncratic meaning at any level 2 , using the term idiom to refer not only to
complex phrases and to complex words with non-compositional meanings, but also
simple words such as cat (see below).
2.1.5
Construction of an Utterance
The structure of the Distributed Morphology grammar expands upon the familiar Ymodel of Principles and Parameters, but differs in interesting and important ways (see
Figure 1 in 2.3). The numeration in DM includes only morphosyntactic features. As
discussed above, no phonological words are produced by any generative lexicon.
Syntactic operations (i.e Merge and Move) operate on the abstract features to create
semantically interpretable syntactic structures just as in Lexicalist Minimalism. These
syntactic operations discharge uninterpretable features from the derivation.
Spellout works a little differently in DM than in Lexicalist Minimalism. The two
models are similar in that there is a division where the derivation proceeds along two
different paths. Down one path, the derivation continues with some syntactic operations
that eventually create the Logical Form of the utterance. The other path leads to
2
The upward boundary of idiomatic meaning is often assumed by the literature on idioms to be somewhere
between vP and CP. Ultimately, this distinction is not relevant to the presentation of DM given in this
dissertation.
27
Phonological Form. However, along the path to PF, DM differs in proposing that there
are some purely morphological operations (such as fusion, fission, morphological
alignment, morphological merger, etc) which can alter the derivation in certain restricted
ways such as (but not limited to): a) changing the featural content of a given terminal
node; b) adjoining adjacent terminal nodes into a complex terminal node; or c) realigning
or even adding morphemes (terminal nodes) to the structure, in accordance with the
language’s morphological rules. These operations do not affect the featural content of the
interpreted path (LF). Thus, they are LF neutral.
At PF, Vocabulary Items are drawn from the Vocabulary and inserted into the
terminal nodes of the structure produced by the derivation. The insertion of these VIs
discharges all the interpretable features—i.e. it removes unpronounceable formal features
from the derivation and replaces them with pronounceable “words”. Following insertion,
any applicable readjustment rules (see below) are implemented as well as are any
phonological rules that may change the phonological form. Different from the traditional
Lexicalist model, however, is the fact that the two arms of the Y-model join again at the
conceptual interface. Here is where the features in LF and the semantics of the root VIs
provided at spellout are interpreted by the Encyclopedia.
28
(2.3)
Figure 1
MORPHOSYNTACTIC
FEATURES:
[+N]
[+singular]
[3rd person]
Syntactic
Operations
Morphological
Operations
VOCABULARY
INSERTION:
/kæt/
/-s/
Phonological Form
Logical Form
Conceptual
Interface
(Meaning)
ENCYCLOPEDIA:
Non-linguistic
Knowledge-Little furry thing,
likes to sleep on
my face.
The Structure of the DM Grammar
2.1.6
Spellout, Competition, and the Subset Principle
The primary operation through which “words” are constructed in DM is the syntax. Each
terminal node in a given derivation is composed of one or more interpretable features.
The terminal nodes now need phonology in order to be pronounceable. In DM, after
spellout is when the “words” (i.e sound-meaning correspondences) enter the derivation—
through Vocabulary Insertion.
29
The Vocabulary contains entries linking a formal feature (or features) to a series
of sounds that realize that feature. These entries are called Vocabulary Items (VIs). For
example, the familiar verbal inflectional morpheme, -s, in English realizes three features.
Its VI might look like this (2.1 repeated here as 2.4):
(2.4)
[present]
[singular]
[3rd person]
↔
-s
/-z/
Vocabulary Items are specified for the features they realize. For example, the Spanish
determiners los and el are both specified for the formal feature [definite] and the phi
feature [masculine]. However, los will also be specified for [plural] whereas el is not
specified for number at all. The spelling out of functional morphemic material with these
phonological items is done through a process of competition where different VIs compete
with each other to be inserted into a syntactic node. In the Spanish example above, when
el and los compete for insertion into a D° terminal node specified for plural, los will win
because it is better specified for the features present in the node. Both are eligible for
insertion, since neither contains any features which conflict with the fully specified D°,
but in a competition for insertion, the VI which is specified for the largest number of
features without being specified for any features that are not in the target node will win
the competition. This is called the Subset Principle.
(2.5) Subset Principle:
The phonological exponent of a Vocabulary item is inserted into a
morpheme... if the item matches all or a subset of the grammatical features
specified in the terminal [node]. Insertion does not take place if the
Vocabulary item contains features not present in the morpheme. Where
several Vocabulary items meet the conditions for insertion, the item
30
matching the greatest number of features specified in the terminal
morpheme must be chosen (Halle 1997).
Competition for insertion in DM is governed by the following constraints:
a)
The VI that has the most features matching the node features is inserted.
b)
The node may contain more features than the VI is specified for, but the VI may
not be more highly specified than the node. (The VI will still discharge all the
features in the node, even if it is underspecified)
c)
Since DM is an underspecification model, the “elsewhere condition” (or default)
is the VI in the appropriate competition that is specified for the fewest features.
d)
The Vocabulary Item may be sensitive to features in surrounding nodes, as well as
to features on their loci of insertion.
e)
The operation of insertion is usually taken to be cyclic (Bobaljik 2002), allowing
stems to be inserted before affixes. To use an English example, this allows
affixes to 'see' the stems they will be attaching to, for instance, Latinate affixes
may be sensitive to Latinate stems.
As a quick exercise in how insertion works, let’s take the case of the English copula.
Let’s assume that the different forms of the copula are specified as follows (simplifying
the data for ease of presentation):
(2.6)
were
are
be
am
is
was
is specified for the feature
is specified for the feature
is the elsewhere condition
is specified for
is specified for
is specified for
[past]
[present]
[1st] [present] [singular]
[3rd] [present] [singular]
[singular] [past]
31
(2.7.a) Competition into I am happy.
conflicting feature
were
[past]
are
[present]
be
unspecified
inserted candidate
am!!
[1st][present] [singular]
overspecified
is
[3rd] [present] [singular]
conflicting feature
was
[singular] [past]
were
[past]
are
[present]
be
unspecified
conflicting feature
am
[1st][present] [singular]
inserted candidate
is!!!
[3rd] [present] [singular]
conflicting feature
was
[singular] [past]
less specified
Syntactic
Node:
[present]
[singular]
[1st]
less specified.
(2.7.b) Competition into He is happy.
conflicting feature
less specified
Syntactic
Node:
[present]
[singular]
[3rd]
less specified
32
(2.7.b) Competition into They are happy.
conflicting feature
were
[past]
are
[present]
be
unspecified
conflicting feature
am!!
[1st][present] [singular]
conflicting feature
is
[3rd] [present] [singular]
conflicting feature
was
[singular] [past]
inserted candidate
Syntactic
Node:
[present]
[plural]
[3rd]
less specified.
In the schema of competition as described above, it is possible for two
morphemes to arrive at a tie. The method for breaking ties proposed by Halle and
Marantz (1993, 1994) is that competition was simply (but crucially) extrinsically ordered.
The winning candidate was that the VI to come first in this crucial ordering. However,
Noyer (1997) alternatively proposes that such ties are resolved by the Universal
Hierarchy of Features: VIs that realize features higher on the hierarchy are preferred for
insertion. For example, 1st person and 2nd person outrank 3rd person in the hierarchy.
Should a tie occur between a VI that was specified [3rd] and one that was specified for
[1st], the VI specified for [1st] would win because the contrastive feature in it is higher
ranking. Similarly, since [1st] outranks [2nd], a VI specified for [1st] would win in a tie
with a VI specified for [2nd].
Another possible result from competition is that a VI will be inserted into a node
that contains more features than does the VI. Insertion of a VI can be understood as the
replacement of the formal semantic features in a terminal node with something that is
33
pronounceable (i.e. phonology). It follows that the formal features are removed from the
derivation by insertion (so as not to be realized repeatedly). This removal of features is
called feature discharge, or 'exponence'. An important aspect of insertion in DM is that
the insertion of a VI into a node discharges all of the features contained in that node, not
just the ones that the VI realizes. For example, in the competition given above in (2.7),
the VI are, which is specified for the feature [present], is inserted into a node containing
[present], [3rd], and [plural]. Even though are only realizes [present], it discharges all
three of the features from node. In this way, every terminal node need only be spelled
out by one VI.
A simple way of thinking of it is this: the phonology is presented with this
complex syntactic object and instructed, 'pronounce this!' There are two simple
conditions on this pronunciation: a) only terminal nodes must be pronounced, since they
are the only linearizable parts and b) every terminal node must be pronounced. Providing
a terminal node with phonological content—any phonological content—satisfies these
two conditions. Therefore, it doesn't matter if a VI doesn't realize all the features of a
node, as long as it provides it with phonological content. Competition will ensure that
whatever phonology a node gets will be the best available representation of its content.
Noyer (1997) showed that there are times where the insertion of a VI is also
dependent on features present in other nodes. To account for this type of data, Noyer
(1997) shows that VIs may be specified both for things that they are the primary
exponents of (i.e the features in the node into which they are inserted), as well as for
things that they are secondary exponents of (i.e. features in nodes other than the node into
34
which they are being inserted). This is called secondary exponence. For example, the
allomorph of the A head /-t/ (which is not susceptible voice assimilation) is selected via
secondary exponence in forms like burnt (cf. the tensed verb form burned)
Another good example of the usage of secondary exponence is the example of
Spanish determiners seen above. The competition between el and la is not settled by the
featural content of the node they are being inserted into (by hypothesis, the nodes are
identical), rather it is settled because el is a secondary exponent of the feature [masc],
present on the stem, while la is a secondary exponent of the feature [fem]. The gender
feature is not in the determiner node and is not discharged by the insertion of a
determiner VI, but rather is likely in the noun and is discharged by the noun. Similarly,
the difference between el and los is that one is a secondary exponent of [sing] and the
other of [plural], again not a feature that is in D°, but rather is located in Num°
One of the usages of secondary exponence is using it to license VIs that realize
roots (proposed by Harley and Noyer 2000). For example, the VI thrash realizes a root
but also can only be a verb (a verb is a root dominated by v, a noun is a root dominated
by n). Harley and Noyer propose using secondary exponence to require that thrash only
be inserted into a node dominated by the little-v head, even though the little-v head is
realized by another VI. This use of secondary exponence for licensing root VIs is called
licensing and is elaborated below and addressed in detail in Chapter 3.
35
2.1.7
The Distinction between Functional and Contentful
In DM, the definitional distinction between functional morphemes and content
morphemes is the presence of a root. Roots are “abstract morphemes” linked to a basic
concept (the root for cat is √CAT). They are formal elements of the grammar that, unlike
other syntactic featutres, are linked to extragrammatical information (such as reference or
encyclopedic knowledge). The major effect that this has on the grammar in DM is that
the VIs that realize functional features participate in competition but those that realize
roots don’t.
The original proposal for DM did not include different “varieties” of roots, such
as √CAT or √DOG. Rather, there was only one syntactic element “√”, which was just
called a root. This √ meant that the terminal node containing it needed to be realized by
some sort of content morpheme. The syntax was indifferent to which content morpheme
realized that √. Thus, the syntax component of the grammar was not sensitive to the
difference between cat and dog. All that was relevant to the syntax was the formal
syntactic features. This division of form from referential or encyclopedic meaning was
originally considered one of the strengths of DM. However, Pfau (2000), in an account
of speech error phenomena, proposed that the roots manipulated by the syntax were
specific to the particular concept they were linked to (such as √DOG). This change has
largely been adopted by the DM community, and I assume it in this dissertation 3
3
I have not attempted the analyses presented here in a model of DM with “general” roots (√) other than in
the most rudimentary form and do not know how rejecting that assumption would affect the analyses
presented here other than to say that certain parts, particularly chapter 3 would need to be reviewed
36
Since cat and dog were both specified for √, they could not compete with each
other (they have identical specification). Thus, root VIs could not participate in
competition. Since the addition of Pfau’s specific roots, the groundwork has been laid for
the competition of root VIs – each VI is linked to a different root.
For example, Harley and Noyer’s l-morpheme hypothesis 4 (1999, 2000)
recapitulates the difference between roots and functional morphemes. They claim that
VIs which realize roots (since they do not participate in competition—they all have
identical primary exponence) are licensed for insertion through secondary exponence (see
above). Given that dog and cat could compete with each other in Pfau’s model, the
licensing proposed by Harley and Noyer reduces to specific secondary exponence
requirements on particular VIs. However, as of yet, no one has proposed a model of the
grammar where roots participate in competition.
Within the older (pre-Pfau) model of DM, Marantz (1997) points out that, since
root VIs aren’t in competition with each other, irregular forms of the same root, such as
rise and raise or eat and ate, which look to be different forms of the same verb, also
cannot be competing with each other even though they appear to be (being conditioned
by different structural configurations). That means that the change from rise to raise (or
from eat to ate) needs to be some type of syntactically-conditioned allomorphy. In prePfau DM, there had to be two types of allomorphy that are conditioned by morphosyntax
(excluding phonologically conditioned allomorphy, such as voicing assimilation in
4
Harley and Noyer name roots “l-morphemes” and call fuctional morphemes “f-morphemes”. Given the
preponderance of different names for this crucial division, in this dissertation, I will try to maintain the
names “functional morphemes” vs. “content morphemes/roots”.
37
inflectional suffixes in English). The allomorphy that results from competition of
different VIs (thus limited to functional morphemes) is suppletion. However, content
morphemes have to be subject to some other kind of allomorphy. In this case, (pre-Pfau)
DM proposes that there is one VI, which undergoes readjustment (discussed below in
Section 2 of this chapter).
This division between the processes that condition the insertion of functional VIs
(competition) and contentful VIs (licensing and readjustment) largely means that the prePfau model of Distributed Morphology had two separate grammars, one dedicated to the
insertion of functional morphemes and the other to content morphemes. As discussed
above, Pfau’s proposal of contentful roots allows a model of DM where root VIs compete
with each other (in this model cat and dog would be specified for different roots,
therefore their specifications aren’t identical—allowing them to compete). Again as
above, licensing then is just a matter of the secondary exponence requirements of the VI.
As of yet, nobody has proposed how to incorporate Pfau’s hypothesis with the
conditioning of the insertion of root VIs. Such a proposal is one of the purposes of
Chapter 3 of this dissertation.
2.1.8
Lexical Categories in DM
Traditional parts of speech receive a special treatment in DM, which recognizes that
some roots may surface as multiple different lexical categories (for example, book as a
noun or book as a verb in I booked the appointment; similarly I saddled the horse or
Verbing wierds language). DM asserts that the familiar classifications for content
38
morphemes—noun, verb, adjective, etc—are epiphenomenal. More precisely, the
grammar does not store a content item labeled as a noun or a verb. Rather, the properties
of lexical classes are derived from the syntactic complex of functional syntactic heads
and a root. A root “becomes” a verb by being immediately c-commanded a verbalizing
functional head, such as little-v, Aspect, Tense, and perhaps Trans. So a verb is not a
simplex word in the traditional notion, but rather it is a complex of a functional head
dominating a root. Similarly, nouns are roots c-commanded by a functional nominal
head (n). According to Harley and Noyer’s L-morpheme Hypothesis (as described
above), the immediate c-commanding head often licenses the insertion of a contentful VI
into the root position, if that VI is restricted in its categorical behavior, and thus is called
the licenser.
2.1.9
Special meanings and the Encyclopedia
In Lexicalist syntax, one of the important functions of the Lexicon is to be the locus of
special meanings. In particular, theoretical linguistics has always recognized that some
complex structures need to be stored whole because the whole structure has an
idiosyncratic meaning. Some familiar idioms are kick the bucket, buy the farm, rain cats
and dogs. These meanings are not composed from the meanings of the individual
constituents of the structure. Often the syntax itself does not function “as it should”. For
example, the idiom a lot (of) is a DP, but almost always functions as an adjective or an
adverb. The idiom the hell out of is not a traditionally recognized constituent.
39
DM as I described here seems to have a problem with such structures since the
syntax is built up before the words are inserted. In that case, there is no way for a
complex idiomatic structure to be stored unless there was a mechanism in spellout that
allowed one VI to discharge complex structure. Rather that take this route, DM
recognizes that idioms may be simpler than originally assumed. The following is a list of
structures that have meanings that are not predictable from composition.
(2.8)
a)
b)
c)
d)
e)
let the cat out of the bag
the whole nine yards
terrific
scissors
duck
The examples in a) and b) are widely accepted as idiomatic, because the meaning of the
whole phrase is not compositional. However, from the viewpoint on which syntax and
morphology are the same mechanism, c) and d) must also be idioms as their meanings are
not a sum of their morphemes. Terrific and scissors just have to be memorized as not
being compositional—terrific does not have the compositional meaning that terror-ify-ic
would have (roughly equivalent to horrific) but rather means something analogous to
great or swell; similarly, scissors refers to just one thing despite apparent plural
morphology. It seems then that words by themselves can be idioms. The logical
extension of this is to recognize that we already memorize the meaning of arbitrary
collections of phonemes when we memorize monomorphemic words such as duck. There
really is no difference between memorizing the meaning of a large chunk like a) and a
small chunk like e). Through this view, we can define an idiom as any grammatical
expression whose meaning is not predictable. The grammar then needs not to have a
40
place to store large chunks that are idiosyncratic, because everything is idiosyncratic.
Rather, the grammar has to have some way of interpreting idiosyncratic meaning.
Crucially, this part of the grammar must come after insertion, as in the original DM
model, the words cat and bag aren’t present in a derivation until after spellout. It must
also be sensitive to the final LF form as well since some idioms include structures
sensitive to processes that occur between spellout and LF (such as QR. C.f. Every dog
has its day).
In DM, the mechanism through which these interpretations are made is called the
conceptual interface or, more often, the Encyclopedia. The Encyclopedia is the locus for
idiosyncratic knowledge of roots (such as the fact that kick and buy have special
meanings when their objects are the bucket and the farm respectively) and our real world
knowledge of the referents of words (for example, the fact that a cat is a furry quadruped
and thinking is something that occurs only in your head). As I will discuss below, this
knowledge often clashes with otherwise well-formed grammatical structures (Mary
thought the book to John; Colorless green ideas sleep furiously) Because of this, much
of the information that is stored in the Encyclopedia is considered extra-linguistic. This
raises the question as to whether or not the Encyclopedia is itself a grammatical
component. I have assumed it to be in my sketch of the grammar, but this assumption is
debatable.
41
2.2.
Morphological Operations in DM: From Spellout to PF
In Section 1 of this chapter, I detailed the basics of the framework of DM. Since DM is
both a theory of syntax and a theory of morphology, there are a number of operations in
DM to account for morphological behavior. The majority of these operations occur at
spellout or between spellout and PF. For the sake of simplicity, I will use the
terminology “at spellout” as I did in the previous section, though many of these
operations are crucially ordered.
As shown above, much of the morphological structure is a result of syntactic
processes, but some things have to occur in a morphological component that occurs after
the syntax proper. In this section, I will discuss some of those processes that are
available to DM that are relevant to the work at hand: morphological merger 5 , fusion /
fission, and readjustment rules.
2.2.1
Morphological Merger
Marantz (1984) first proposed morphological merger as a way of realizing (or replacing)
syntactic structure with morphological structure. Marantz (1984) proposes that specific
relations between two syntactic nodes can be replaced by affixing those two nodes to
each other. Later, Marantz (1988) defined merger to be:
(2.9)
At any level of syntactic analysis…, a relation between X and Y may be
replaced by (expressed by) the affixation of the lexical head of X to the
lexical head of Y (Marantz 1988 261).
5
The name for this process is frustratingly similar to the syntactic operation MERGE. They are, however,
unrelated.
42
Fundamentally, merger is the process whereby an abstract morpheme realized by an affix
is attached to a stem, creating the difference at linearization between two elements that
form one compositional unit (a complex word) and elements that just happen to be next
to each other (a difference that is often realized orthographically by a space). Also,
merger has been argued by Bobalijk (1994 cal1ed “merger under adjacency”) to be one
process by which (zero-level) elements adjacent in the syntactic structure come to occupy
the same syntactic node (for example, the attachment of the tense node to the verb head
in English).
Another recent use of merger has been what has been historically called head
movement. Chomsky (1994) has argued that head movement is not a function of syntax
but rather something that occurs after spellout. Using morphological merger, head
movement can be argued to the merger of two zero-level elements (Baker 1988).
Essentially, merger is the name of the process whereby two independent zerolevel nodes are reorganized into one complex zero-level node be it by affixing to each
other, re-linearizing, or conjoining.
2.2.2
Fission and Fusion
The process of fusion exists within DM to account for situations where merger is not
quite adequate to account for the distribution of vocabulary items in a string. This is for
cases where the formal features from several different terminal nodes are all realized by
one Vocabulary Item. For example, it is common for phi-features and tense to be
43
realized by the same morpheme (as is the case in English and Spanish. What the process
of fusion does is reduce all the features of a several distinct terminal nodes (perhaps
dominated by a single X°-level element, such as that created by the process of merger) to
one simple head containing all the features of the complex head. For example, in
English, the features [present], [3rd person], and [singular] are all realized by the affix -s.
By hypothesis, the phi-features and the tense features are in separate syntactic nodes,
AgrS° and T°. The process of fusion applies to the two nodes to combine all the features
of the two former nodes into a new, single node that can then be targeted for insertion by
–s.
Complementary to the process of fusion is ‘fission’, as proposed by Noyer (1997)
and Marantz (1997). Whereas fusion takes two positions of exponence (i.e. terminal
nodes) and reduces them to one node, fission takes one position of exponence and splits it
into many. The most common usage of fission in the literature has been to separate phifeatures from one another, since by hypothesis they all are base-generated in a single
feature bundle, which appears as one terminal node at the end of the syntactic derivation.
Both Marantz (1997) and Noyer (1997) employed this mechanism to spilt AGR into
separate heads, each containing a phi feature that is realized independently. Thus, a
language, such as Noyer (1997)’s example of Tamazight Berber, can have separate
morphemes for person, gender, and number (see 2.10, taken from Harley and Noyer
1999)
44
(2.10.a)
Tamazight Berber Prefix Conjugation. dawa ‘cure’
1
2m
2f
3m
3f
(2.10.b)
plural
n-dawa
t-dawa-m
t-dawa-n-t
dawa-n
dawa-n-t
Vocabulary Items proposed by Noyer (1997)
/n-/
/-γ/
/t-/
/t-/
/-m/
/i-/
/-d/
/-n/
/-t/
2.2.3
singular
dawa-γ
t-dawa-d
t-dawa-d
i-dawa
t-dawa
↔
↔
↔
↔
↔
↔
↔
↔
↔
[1st] [plural]
[1st]
[2nd]
[3rd] [sing] [fem]
[plural] [masc] ([2nd])
[sing] [masc]
[sing] ([2nd])
[plur]
[fem]
Readjustment Rules
Finally, there are types of allomorphy that cannot be captured using the processes
described above: Allomorphy whose conditioning environment is not phonological 6 , but
rather is morphological. In the case of functional morphemes, this morphological
conditioning is captured through some combination of competition and secondary
exponence. However, since DM assumes root morphemes are not in competition with
each other, there is a process whereby the phonology of the VI is changed as a result of
morphologically conditioning. For example, -ceive and -cept are both realizations of the
same VI where the change from -ceive to -cept is conditioned by the nominalizer –ion.
6
such as nasal place assimilation, vowel harmony, voicing assimilation, etc.
45
DM proposes that there is one basic allomorph (above –cieve) and all the other
allomorphs are the result of the application of a phonological Readjustment Rule.
In the case of the morpheme, -ceive (receive, deceive, conceive, perceive), DM
hypothesizes a readjustment rule that changes v to p, and shortens i to E in just this
vocabulary item, when the nominalizing suffix -tion is attached to it. Similarly, DM
hypothesizes a readjusment rule that changes foot to feet when dominated by the feature
[plural] (cf goose/geese, mouse/mice). Additionally, there has been some suggestion that
readjustment rules can target classes of VIs instead of being idiosyncratic to just one (see
Embick and Noyer 2004, Embick and Halle 2004 and others). For example, there could
be a readjustment rule in English that changes /aj/ to /ow/ in a VI when it is
c-commanded by [past] (drive/drove, ride/rode, and dive/dove). This readjustment rule
would apply to an entire class of VIs instead of just one.
Readjustment rules are different from the syntactic and morphological
transformations of older work (see Chomsky and Halle 1968, Kiparsky 1982 and many
others). Readjustment rules are taken to be idiosyncratic rules that apply only an
individual VI or class of VIs, they are not general phonological rules in the grammar.
46
CHAPTER 3 ON A THEORY OF ROOT ALLOMORPHY
Root allomorphy is a subset of relationships traditionally called irregular morphology.
Root allomorphy comes in two varieties 7 . The first is suppletive allomorphy where the
two forms cannot be derived from each other by some sort of phonological process.
Some examples of suppletive allomorphy are in (3.1).
(3.1)
go/went
good/better/best
bad/worse
person/people
The other type of allomorphy is what I call irregular allomorphy, in which there is some
common phonology between the two forms. This commonality is usually attributable to
some type of historically regular phenomena (such as i/j umlaut) which has since fallen
out of the language.
(3.2)
eat/ate
mouse/mice
receive/reception
sleep/slept
In DM, as discussed in chapter 2, since root allomorphy always involves a root, these
relationships are not captured the same way that suppletion of functional morphemes is
(i.e. competition). Rather, these relationships are always considered the application of a
readjustment rule.
7
I will use the terms suppletive and irregular as contrastive. In actuality, suppletion is usually considered a
subset of irregular morphology. I am using the term irregular to mean irregular morphology minus
suppletive morphology.
47
The ultimate purpose of this chapter to show an analysis of root allomorphy that
uses the same mechanisms used for allomorphy of functional morphemes. This chapter
will also aim to do four other things: 1) offer an explanation for data that is otherwise
unexplained in DM (i.e. the blocking of inflection in the non-head position of nominal
compounds); 2) show that DM does not need two different “grammars” for morphology
(one set of operations for roots and one for functional morphemes) as described in
Chapter 2, reducing the number of operations proposed by DM and make for a more
economic model of the grammar; 3) greatly reduce the number of null morphemes that
are predicted by DM—a prediction which is a potential criticism; and finally, 4) show a
functional application of MINIMIZE EXPONENCE, the ultimate purpose of this dissertation.
To those ends, I propose that, in order to satisfy MINIMIZE EXPONENCE, the
functional heads projected above the root fuse with that root. This results in the root and
the formal features being in the same node. Because of this fusion, VIs can be specified
for both a root and formal features, allowing eat and ate to be different VIs that compete
with each other for insertion.
3.0.1
Licensing and readjustment in DM
For the purposes of refreshing the reader’s memory, in this section I quickly review some
of the aspects of DM presented in Chapter 2. The purpose of this section is to show root
allomorphy in action in DM and then to showcase some of the concerns about the
traditional DM analysis that will be addressed in this chapter.
48
Consider the typical derivation of root allomorphy in DM. As an example,
consider the derivation for mice, a form showing what I have called irregular root
allomorphy. The syntax results in the complex head found in (3.3):
(3.3)
Num
3
[plural]
nP
3
n
√MOUSE
Were mice a regular form, the node containing the feature [plural] would be realized by
the VI –s. Since the root, √MOUSE, is a noun, its VI would be licensed for insertion by
the feature little-n, which is itself realized by a null morpheme. The root itself would be
realized as mouse. Linearization of the morphemes would result in mouses. However,
since mice is irregular, there are a number of key differences to this derivation. First,
instead of the affix –s, the null allomorph of [plural] is conditioned by the presence of the
root √Mouse. Mouse is inserted, again having been licensed by little-n (still realized by a
null morpheme). Then, a readjustment rule is conditioned by mouse being c-commanded
by [plural]. This readjustment rule changes the phonology of mouse to mice.
In addition to the marked complexity of the derivation of a relatively innocuous
word like mice, a strange interdependence occurs in the derivation. The null plural
morpheme is licensed by the presence of mouse and the readjustment of mouse to mice is
licensed by the presence of [plural]. 8 What follows will be an alternative analysis of root
allomorphy that is ultimately less complex.
8
This interdependence is the effect of both the application of readjustment rule and the secondarily licensed
affix and is actually a prediction of the DM grammar. Since both readjustment of a root and a secondarily
49
3.1.0
Roots in the grammar
Recall from chapter 2, VIs realizing abstract morphemes are specified for formal features
such as [past] or [1st]. Thus, an example of a typical specification of a VI is (3.4).
(3.4)
Vocabulary Entry for -ed
[past]
→
-ed
/-d/
VIs must be specified for the morphosyntactic features that they realize. These features
encode the meaning of the syntactic node in the semantic computation. The VIs that
realize roots are also so specified. Thus, the VI for cat is likely specified for realizing the
core meaning of cat, which according to DM would be a root (√CAT). Earlier work in
DM (Halle and Marantz 1993, 1994, Halle 1997, Marantz 1997, Harley and Noyer 1999,
2000) suggests that roots came in only one variety and the syntax was not sensitive to the
different ways that root could be realized. Pfau (2000), on the other hand, suggests that
roots are in fact individually distinguished from the beginning of the syntactic
computation in order to account for specific types of speech errors (word substitution,
etc) within the framework of DM. This change to DM has largely been accepted and in
general, in relevant analyses, roots are treated as being specific to the concept they refer
to (see Embick and Marantz 2006 for example). That is, the numeration not only
licensed affix are possible the grammar predicts any combination of one or both of those to happen, seen in
the table below.
regular root and regular affix
regular root and irregular affix
irregular root and regular affix
irregular root and irregular affix
walk
hit
slep
mice
-ed
-ø
-t
-ø
50
includes the formal features to be manipulated by the syntax, but the formal concepts that
will later be interpreted by the Encyclopedia as well. Following that assumption, cat is
likely specified to realize a formal instantiation of the concept of cat-ness, which can then
manipulated by the syntax. Thus, a VI for cat might look like (3.5.a)
(3.5.a.) Vocabulary Entry for cat
√CAT
→
cat
/kQt/
The VI seen in (3.2) can only be inserted into a terminal node containing the very specific
root √CAT. Just as the -s that realizes [present] can’t be inserted into a node containing
[past], cat can’t be inserted into a node containing √DOG. However, a VI such as (3.5.b)
would be perfectly able to be inserted into such a node:
(3.5.b.) Vocabulary Entry for dog
√DOG
→
dog
/dag/
Thus, the VIs in (3.5.a) and (3.5.b) compete against each other for insertion, deciding the
winner based on which better matches the contents of the target node. (3.5.a) wins a
competition into the node containing √CAT while (3.5.b) wins the competition for
insertion into a node containing √DOG. Neither would win a competition to be inserted
into a node containing √DUCK.
51
3.1.1
Licensing Insertion
While in English zero derivation from one “grammatical category” such as verb or noun
is not uncommon (meaning the words like pen, ink, and table move easily between verb
and noun without overt morphology), we are all at least intuitively aware that some
words must be verbs while others must be nouns. For example, for the most part, thrash
must be a verb and technique must be a noun.
(3.6)
a) *Dave techniqued his writing skills. 9
b) *Dave also awaited the thrash.
Thus, there has to be some way in which VIs restrict their insertion to nodes where the
root receives the appropriate category. Currently in DM (Harley and Noyer 2000), the
hypothesis is that VIs are licensed by the immediate syntactic environment through
secondary exponence—for example, the immediately C-commanding functional head.
Thus, thrash might have an entry such as (3.7).
(3.7)
Vocabulary Entry for thrash
√THRASH*
→
thrash
/TrQS /
*must be c-commanded by [v]
This type of specification allows the insertion of thrash into (3.8) where the root is
dominated by vP, but blocks it from insertion into (3.9) where the root is dominated by
nP.
9
Given that zero derivation of nouns from verbs is so productive in English, this sentence becomes
grammatical given enough context. Assume no context for now.
52
(3.8)
Sharks thrash.
TP
3
Sharksi 10
T’
3
T
vP
3
ti
v’
2
v
√THRASH
(3.9)
*The thrash stopped. 11
TP
qp
DPi
T’
3
3
the
nP
T
vP
3
[past] 3
n √THRASH
ti
v’
2
v
√STOP
This formulation of licensing (through secondary exponence), whereby the VI checks the
functional element c-commanding the target node rather than checking the rather than the
target node itself is a side-effect of having VIs that realize roots not participate in
competition. However, in the Pfau (2000) model of DM where the roots can compete,
there is another way to capture the “noun”-ness or “verb”-ness of a given VI.
10
As a simplification tool, I will be using italicized words (rather than triangles) to indicate summarized
sections of trees.
11
The tree shown here assumes stop to be unergative, just for ease of presentation. It was the only verb
that even began to make sense with thrash as a noun.
53
I propose that this is possible because of the applications of morphological merger
and fusion to roots and the functional heads c-commanding them. In particular, I propose
that the functional verbalizing element—little v (Kratzer 1996)—carries an interpretable
feature 12 , [v], whose syntactic content is something along the lines of “I am a verb”.
Similarly, the nominalizing head—little n (Harley and Noyer 2000)—carries the feature
[n]. Furthermore, I propose that the root nodes themselves acquire these features through
several applications of morphological processes. First, the root undergoes “head
movement” (i.e. morphological merger) to adjoin to the functional heads above it. The
resulting complex head then undergoes the process of fusion to incorporate all the
features of the complex head (including the root) into one simplex head.
If we assume an application of morphological merger (called merger under
adjacency) to the tense head and the verb as proposed by Bobaljik (1994), then the
resulting head after fusion contains a root, a functional verbal element, and a tense
feature.
12
or more likely a bundle of features—I will refer to it as one feature until the distinction becomes
important in chapter 4.
54
(3.10) The dog ran.
TP
3
The dogi
T’
3
T
vP
Application of
[past] 3
Spelled out form
merger
ti
v’
3
v
√RUN
[v]
Application of
merger
Complex result of head movement
T
3
v
T
3 [past]
v
√RUN
[v]
(3.11)
application
of fusion
√RUN
[past]
[v]
Resulting form
TP
3
T’
The dogi
3
vP
3
ti
target node
v’
3
√RUN
[v]
[past]
Ready for Insertion
55
As seen in (3.10), the applications of head movement and fusion to the complex
verbal structure results in a single simple node containing the formal features of the entire
structure. The VI instead looks directly at the target node in the usual way to discover
whether that node can satisfy the “functional” or grammatical requirements of that VI.
For example, the VI in (3.12) can be inserted into the node created in (3.10, copied as
3.13) because the features it is specified for are a subset of those appearing in the node.
(3.12) Vocabulary Entry for ran
√RUN
[v]
[past]
→
ran
/rQn/
compatible for insertion
(3.13) Resulting form
TP
3
The dogi
T’
3
vP
3
ti
v’
3
√RUN
[v]
[past]
56
The example VI for ran requires two different functional features, [past] and [v]. While
the specification for [v] identifies it as a verb, the specification for [past] sets it apart
from run. Thus, similar to how dog will compete with cat for insertion into a node, run
will compete with ran for insertion into (3.13), with ran winning the competition due to
its more complete specification. Thus, the competition of roots enabled by Pfau’s
proposal of specific roots allows for another analysis of root allomorphy within DM.
3.1.2
Alternative Analysis of Root Allomorphy
The processes of morphological merger and fusion described above apply to an account
of root allomorphy. As an example, I show a derivation and insertion of the word mice
using a fusion-based analysis. Recall the structure that results from the syntax above
(3.1, repeated here as 3.14):
(3.14)
NumP
3
[plural]
nP
3
n
√MOUSE
If we assume head-movement, all those heads move to one terminal node where they
make up a complex head.
57
(3.15) Complex head resulting from head movement
(features contained in that head shown in square brackets)
Num
3
n
Num
3 [plural]
√MOUSE
n
[n]
The process of fusion is applied to complex heads such as that in (3.15) resulting in a
simplex head. After the application of fusion, the resulting head is simplex and contains
all the features previously in the complex head.
(3.16) [plural]
[n]
√MOUSE
This node, containing the root and several grammatical features, in now the target node
for insertion. As we saw above with the VI for ran being specified for the feature [v]
(indicating that it is a verb) and the feature [past], there is a specific VI for mice which is
specified for the feature [past] and differs from the one for mouse in that it is more
specified for number.
(3.17.a)
Vocabulary entry for mouse.
√MOUSE
[n] 13
13
Æ
mouse
/maws/
This specification means that mouse may only appear as a noun. Mouse may be underspecified for this
feature if it can be used as a zero-derived verb. I use this specification for ease of presentation.
58
(3.17.b)
Vocabulary entry for mice
√MOUSE
[plural]
[n]
Æ
mice
/majs/
With the VI specifications in (3.17) we can show the following distributional patterns of
the mouse/mice pairs: a) Since mice is the best specified of the two for the feature
[plural] (mouse being underspecified for number), it will win competition into a node
containing the [plural] feature (3.18.a). b) Mice would not win competition into a node
containing [singular] since it is specified for the conflicting feature [plural], thus, mouse,
being unspecified for number will be inserted into the singular environment (3.18.b). c)
Mice would not win competition into one that contains neither feature (which hypothesis
is the case in “general” constructions such as compounds like mouse-trap) since it is
overspecified, thus again the underspecified mouse will win the competition (3.18.c)
(3.18.a)
Competition for insertion into two mice
Target
Node:
inserted candidate
mice: √MOUSE [n] [plural]
underspecified
mouse: √MOUSE [n]
√MOUSE
[plural]
[n]
(b)
Competition for insertion into a mouse
Target
Node:
overspecified
mice: √MOUSE [n] [plural]
inserted candidate
√MOUSE
[singular]
[n]
mouse: √MOUSE [n]
59
(c)
Competition for insertion into mousetrap
Target
Node:
overspecified
mice: √MOUSE [n] [plural]
inserted candidate
√MOUSE
[n]
3.1.3
mouse: √MOUSE [n]
Competition of Roots Revisited
Suppose now that we had another word like cat, which is specified as a noun. Its
entry (3.19) would be specified for the feature [n] rather than the feature [v], which
would make it incompatible with the feature bundle found in (3.10, repeated as 3.20)
because of both the specific root and the [n] feature.
(3.19) Vocabulary Entry for cat
√CAT
[n]
→
cat
/kQt/
(3.20) Resulting form
incompatible for insertion
TP
3
The dogi
T’
3
vP
3
ti
v’
3
√RUN
[v]
[past]
60
Consider now the set of words speech, speak, and spoke. In previous incarnations of
DM, the fact that these words were all linked to the same core meaning would be
captured by postulating one VI that had a series of readjustment rules stipulated for it.
Once we propose that root VIs are competing, there are three different VIs, one for each
form. The feature specifications of all three VIs mention the same root, √SPEAK, but
contain different functional material (3.21-3.23). Thus the common meaning shared
among the three VIs is attributed to the fact that they are all specified for the same root
rather than being phonological derivatives of one VI.
(3.21) Vocabulary Entry for speak
√SPEAK
[v]
→
speak
/spik/
(3.22) Vocabulary Entry for spoke
√SPEAK
[v]
[past]
→
spoke
/spowk/
(3.23) Vocabulary Entry for speech
√SPEAK
[n]
→
speech
/spitS/
Since there are three different VIs, they can each participate in competition separately. In
particular, they can all compete against each other. The VIs compete with each other for
insertion into a target node, with the best-specified form winning the competition. For
example, as seen in the sentence John spoke (3.24 and 3.25), spoke will win the
competition into a past tense environment because it is better specified than speak. In the
sentence The speech began (3.26 and 3.27), speech wins competition for insertion
61
because both speak and spoke are specified for conflicting features ([v] and in the case of
spoke [past]).
(3.24) John spoke.
TP
3
T’
3
Johni
vP
3
ti
v’
3
√SPEAK
[v]
[past]
(3.25) Competition
conflicting feature
Target
Node:
speech: √SPEAK [n]
not well-enough spec.
speak: √SPEAK [v]
√SPEAK
[past]
[v]
inserted candidate
spoke: √SPEAK [v] [past]
62
(3.26) The speech began.
TP
wo
DPi
T’
3
3
the
np
vP
3
3
√SPEAK
ti
v’
[n]
3
began
(3.27) Competition
inserted candidate
Target
Node:
speech: √SPEAK [n]
conflicting feature
speak: √SPEAK [v]
√SPEAK
[n]
conflicting feature
spoke: √SPEAK [v] [past]
As the example with speak, spoke, and speech shows, proposing that a) roots are
contentful rather than general (Pfau 2000), b) functional nominalizing and verbalizing
heads carry formal features, c) that those features come to adjoin to the root through head
movement, and d) that those features fuse with the root to result in a simplex head with
both content material and functional material allows us to argue that content morphemes
such as mouse and mice compete with each other.
63
3.2.
Differences between This and the Traditional DM Grammar.
The changes I have proposed above in section 3.1 make for a different model of DM than
the one originally proposed by Halle and Marantz (1993, 1994). While the core
fundamentals—i.e. syntactic structure of morphology, late insertion, the majority of
morphological processes—remain unchanged, there are two key differences to this model
that merit discussion. Proposing that mouse and mice are separate VIs eliminates the
function of readjustment rules, whose job in traditional DM it is to change the unmarked
form of a root VI into its marked form. In the model of DM sketched here, mouse and
mice are separate VIs coincidentally linked to the same root. In addition, licensing a VI
for insertion based on its “category” has historically been considered a part of that VIs
secondary exponence (Harley and Noyer 2000). In the analysis here, that grammatical
material is now part of a VI’s primary exponence. In this section, I will detail the effects
that this has on the model of the grammar and suggest that the exclusion of readjustment
rules and the limitation of the scope of secondary exponence proposed here are positive
adjustments to the model of the grammar.
3.2.1
Readjustment Rules
As discussed above, readjustment rules are no longer necessary in this model of the
grammar. Since mouse and mice are separate VIs that compete with each other for
insertion, there is now no need for there to be a rule that changes mouse into mice. An
immediate upside to this treatment of root allomorphy is that it unifies the treatment of
functional morphemes and content morphemes. In traditional DM, there are in essence
two different grammars to account for allomorphy. Allomorphy of functional items (such
64
as el, la, los, and las in Spanish) is captured through competition, the underlying
assumption that a language’s inventory of functional morphemes is entirely suppletive.
On the other hand, root allomorphy is captured through the insertion of an unmarked VI
whose phonology is readjusted given a particular conditioning environment. In the
model of the grammar proposed here, all morphologically conditioned allomorphy is
essentially suppletive and all VIs, even those realizing roots, compete in the same fashion
for insertion. At the simplest level, the root allomorphy grammar proposed here and the
one assumed by traditional DM differ in one key way: the grammar proposed here
eliminates the need for the ubiquitous use of readjustment rules and relies only on
competition, making it a simpler model of the grammar—one that contains fewer
mechanisms and reduces the computational load.
As a side benefit, the particular aspect of the grammar that is discarded—
readjustment rules—was already suspect. Though DM is a minimalist model of the
grammar, and as such is a derivative of the GB tradition, readjustment rules are
transformational. Further, despite the claim of some researchers such as Marantz (1997)
and Halle (1997) (see also Embick and Noyer 2006) who claim that one rule can apply to
a whole class of verbs, they are largely idiosyncratic and language specific. This means
that a grammar has to have a major component that is composed of a long list of
memorized readjustments. This adds to both the computational load of the grammar and
to the memory load on the grammar. By proposing that each allomorphic pair is a set of
memorized words, we don’t alleviate the load on memory—each VI - readjustment rule
pair is replaced with a VI-VI pair—but we do relieve the computational load of the
65
grammar. In this model, there isn’t the computational load of readjusting roots; instead,
there is the already extant load of choosing the VI to be inserted.
However, this reintroduces the problem that Marantz (1997) appealed to in his
justification of readjustment rules: learnability. Marantz proposed that readjustment was
a solution to a central learnability concern with root allomorphy: how a learner of a
language can learn that two VIs are linked to the same root. To put it into context, how
does a learner ever learn that mouse and mice are linked to the same root? The mutual
exclusivity constraint (proposed by Markman and Wachtel 1988) on the acquisition of
language mandates associating new words with new concepts. A learner’s purposefully
ignoring that constraint in the case of mice/mouse but not in the case of car/truck is
difficult to explain. Marantz argued that suppletion was limited to functional
morphemes, whose fundamental presence in the grammar is mandated by the language
acquisition device itself—they do not have to be 'acquired' in the same way as roots. The
learner can learn that in one context, 'be' is spelled out by the VI 'is' while in another it's
the VI 'are' because the underlying grammatical structure has already informed them that
this functional morpheme is present. This way the learner only has to memorize the
formal specifications of a given functional morpheme. Never do they have to learn that
two separate VIs are linked to the same root, avoiding that particular problem. Rather,
the learner learns one VI corresponding to one root and then later acquires a readjustment
rule to change that VI to its allomorphs.
This proposal immediately made the somewhat strange prediction, which Marantz
acknowledges, that all clearly suppletive alternations must be of functional allomorphs—
66
they must not be linked to roots. Thus, it predicts that pairs like go/went and bad/worse
are actually functional morphemes (not linked to a root) because they are instances of
true suppletion and suppletion is limited to functional morphemes. Marantz claimed that
all suppletive pairs that appear to be linked to roots would be limited to pairs such as
go/went which aren’t really “full verbs” but are better treated as light verbs, making them
functional.
The existence of content morphemes that are undeniably suppletive would be
counter evidence to Marantz’s claim. One such form in English is person/people (cf. mea
‘kill.sgO’ ~ sua ‘kill.plO’ in Yaqui 14 ; in Hopi, tiyo ‘boy' (sg.)’ ~ tootim 'boy' (pl.) and
wùuti ‘woman (sg.)’ ~ momoyam ‘woman (pl)’ 15 ). Thus, Marantz’s claim that suppletion
is only for functional morphemes, while ideal for addressing the learnability problem, is
falsified by the data. Therefore, while the model of the grammar I propose here
essentially removes Marantz’s solution to that problem without proposing a new one,
Marantz’s solution already ran into the problem that there is in fact suppletion of content
morphemes. This proposal, however, does have the effect of discarding the strange
prediction that forms like bad/worse and go/went are not linked to roots. Rather, I
propose here that suppletion is all that is necessary for all allomorphy, which reduces the
number of mechanisms necessary to capture allomorphy. 16
14
Yaqui example provided by Heidi Harley
Hopi example provided by Jason Haugen
16
Another “function” readjustment rules is to capture the historic relationship between related forms. For
example, the phonological process of i/j umlaut is what created the pairs foot/feet and goose/geese.
Readjustment rules capture this relationship. However, a historical change such as i/j umlaut does not
entail a synchronic change, removing that justification for the existence of readjustment rules.
15
67
3.2.2
Licensing
Recall from chapter 2 that in the pre-Pfau model of DM, VIs which belong to only one
“class” (such as noun) are licensed by the c-commanding functional head according to
the proposal by Harley and Noyer (2000). My proposal here evacuates the need for such
licensing 17 as the licensing functional head fuses with the root and the root VI can select
for its features. One of the inherent problems in using secondary exponence to license a
root morpheme in an environment is that is strictly less local than it could be.
An example of words that have “lexical class” requirements that is the pair
speak/speech as discussed above. Recall that if we hypothesize the following VIs for
speech and speak licensing falls out naturally without requiring that the VI be able to
look at the node above the target node for licensing.
(3.28)
Vocabulary Entry for speak
√SPEAK
[v]
→
speak
/spik/
Vocabulary Entry for speech
√SPEAK
[n]
→
speech
/spitS/
Since the licensing of speech as a noun or speak as a verb in the model I propose here
requires only looking at the node targeted for insertion rather than the part or all of the
entire derivation, the model proposed here requires a significantly lighter computational
17
Licensing itself is argued to be a part of secondary exponence, a feature of DM I allude to here (the
selection of the zero morpheme for [plural] due to its adjoining to mice is an application of secondary
exponence) and discuss in detail in Chapter 2. In suggesting that licensing of nouns, verbs, etc is no longer
necessary, I do not mean to suggest that secondary exponence as a whole is not necessary. Rather, it is
necessary for many reasons as discussed in Chapter 2.
68
load than secondary exponence model. For each content morpheme that has a
distribution limited to its “lexical class”, this model proposes a lighter computational load
than a model of DM that requires licensing. Given the sheer magnitude of a language’s
inventory of content morphemes, the savings of computational load is drastic.
3.2.3
Null heads
A side effect of removing secondary exponence as the means for licensing class specific
root morphemes is the drastic reduction of the number of null morphemes that DM is
forced to propose. The original model of DM (Halle and Marantz 1993, 1994) proposes a
large amount of null heads. Recall the derivation for mice (3.1) (repeated here as 3.29)
(3.29)
NumP
3
[plural]
nP
3
n
√MOUSE
In DM, the structure after insertion is that in (3.30) wherein, √MOUSE is realized by mice
(conditioned by the c-commanding elements), [plural] is realized by ø (an unpronounced
head), and n is also realized by ø.
(3.30)
NumP
3
ø
nP
3
ø
mice
In just this one little complex, for one overt VI, there are two null VIs. Such would be
the case for all irregular forms similar to mice where there is no “overt” realization of an
69
element that conditions a readjustment rule. This by itself is a large number of null
morphemes. However, if you employ the traditional licensing through secondary
exponence, those functional licensers are also overwhelmingly realized as null heads.
Since the vast majority of “free” content morphemes would involve these null licensers
(regardless of whether they are required by the VI or not) the number of these null
morphemes in a given derivation is substantial and in the grammar as a whole would be
enormous. DM is not unique in proposing null morphemes. What makes DM different is
that it predicts a huge number of null morphemes in two cases where other theories of the
grammar (such as Lexicalist models) don’t have to – licensers (such as n) and morphemes
such as [past] or [plural] that are realized as zero when a readjustment rule changes the
root (as is the case with mice or drove). 18 While many models of grammar predict null
heads, these particular null heads are unique to DM and create a perceived fault in the
predictions of the grammar.
DM is susceptible to criticism because it predicts so many more null morphemes
in a given derivation than do other models of grammar. The fusion analysis given here
removes that particular criticism of DM as licensing heads and heads that condition
readjustment rules are not realized by zero in this model but are instead fused with the
root and are realized by the same VI that realizes the root. Recall (3.30) (repeated here)
18
Note that DM does not always change a functional head to zero when a readjustment rule is applied. In
forms like slept, better, went, houses, etc, it is proposed that the root morpheme changes but the functional
morpheme is still present and in its default form.
70
(3.30)
NumP
3
ø
nP
3
ø
mice
In the pre-Pfau model of DM, the licenser (n) and the head that conditions the application
of the readjustment rule (num) are realized as zero. However, in the fusion account of
root allomorphy proposed here, the root and [n] both move to [plural] and fuse (3.31)
(3.31)
NumP
3
[plural]
nP
[n]
3
√MOUSE
Therefore, whereas the traditional DM account contains one overt head and two null
morphemes, the fusion account sketched here contains only one overt head and two
traces.
(3.32)
NumP
3
nP
miceij
3
ti
tj
Thus, the fusion analysis simplifies the grammar by reducing the number of null
morphemes in the inventory of a given language that has to be predicted by DM. Taken
as a whole, the three central changes to DM—the rejection of readjustment rules, not
71
using secondary exponence to license root VIs, and the reduction of zero morphemes—
all drastically simplify the DM grammar. The reduction of both the computational load
and the memory load 19 entailed by the grammar makes the grammar more economical.
Further, the removal of readjustment rules discharges the last of the “transformational”
rules in DM, bringing it more in line with recent GB tradition. Overall, the new model of
DM proposed here is simpler and more attractive to the Minimalist community as a
whole.
3.3
Expansion of the fusion analysis.
The combination of head movement, Bobaljik-style merger under adjacency, and
fusion results in a tidy description of how the [past] feature fuses with the root √RUN to
be realized by the word ran. However, even a cursory look at the details of the
mechanics proposed in section 3.1.2 will show the prediction that formal features should
always fuse with the root, ultimately resulting in only suppletion as a morphological tool.
It seems that the current proposition actually removes the great strength of DM—that it
predicts lexical decomposition as a function of the syntax.
For example, recall the different VIs presented for speech, speak, and spoke
(repeated here as 3.33.a-c).
19
The memory load is reduced in two ways: a) fewer null morphemes and b) two different types of
memorization—readjustment rules and Vocabulary inventory—are reduced to just one—Vocabulary
inventory.
72
(3.33.a)
Vocabulary Entry for speak
√SPEAK
[v]
(3.33.b)
speak
/spik/
Vocabulary Entry for spoke
√SPEAK
[v]
[past]
(3.33.c)
→
→
spoke
/spowk/
Vocabulary Entry for speech
√SPEAK
[n]
→
speech
/spitS/
These three VIs very well predict that speech will appear in a nominal
environment and spoke will appear in a past tense environment. However, it also predicts
that speak will appear in all other verbal environments. In English, this is fine for most
cases, except third person singular. For example, in the derivation of He speaks, the
present tense and the phi-features are overtly spelled out by the affix –s. However, in an
account where fusion applies to all complex heads, specifically fusing tense with the root,
the tense should never be spelled out by its own head. The sentence should be *He
speak.
73
(3.34.a)
John speaks.
TP
3
Johni
T’
3
vP
3
ti
(3.34.b)
v’
3
√SPEAK
[v]
[present]
Competition results in insertion of speak.
conflicting feature
Target
Node:
speech: √SPEAK [n]
inserted candidate
speak: √SPEAK [v]
√SPEAK
[present]
[v]
conflicting feature
spoke: √SPEAK [v] [past]
In order to maintain the decomposability of the form speaks—and for that matter all other
concatenative morphology—the derivation must result in a complex head that has not
been fused to a simplex form. Thus, the derivation for John speaks must actually be a
tree such as seen in (3.35).
74
(3.35) John speaks.
TP
3
Johni
T’
3
vP
3
ti
v’
3
T
3
√SPEAK [present]
[v]
[φ]
insertion
T
3
speak
-s
If we make the reasonable assumption that there is always fusion, then we cannot have
affixed forms such as speaks. Thus, we have to assume that complete fusion fails most of
the time, which gives us complex forms. However, if we say that √SPEAK can fail to fuse
with [present] in order to result in speaks we must also predict that it can fail to fuse with
[past] and result in *speaked. Thus, the application of fusion has to be blocked in order
to trigger regular concatenative morphology, but must be allowed in order to trigger
suppletion.
3.3.1
The ¬ specification.
The immediate problem described above is that, as outlined above, regular forms
like thrash are going to be inserted into a node where the [past] feature had fused with the
75
root, resulting in no overt realization of the [past] morpheme. I propose that the solution
lies in specifying the VI for thrash for an incompatibility with the feature [past], ensuring
that the VI will not be inserted into a node containing that feature.
(3.36) Vocabulary Entry for thrash
√THRASH
[v]
→
thrash
/TrQS/
¬ [past]
I specified thrash with the specification ¬ [past] (read: “not past”). I use this
notation to indicate that the Vocabulary Item thrash is not compatible with the feature
[past]. What this means is that the VI cannot be inserted into a node containing the
feature [past] This specification is in many ways the inverse of the normal specification
used by DM. For example, thrash as specified above for [v]. In terms of DM, this
specification means that thrash must realize the feature [v]. Thrash is also specified as ¬
[past]. That means that thrash must not be inserted into a node containing the feature
[past]. With the addition of the ¬ specification, a VI can now lose a competition for three
reasons: 1) not being well enough specified (i.e. there is a better specified candidate), 2)
containing a conflicting feature (i.e. the VI is specified for a feature that is not present in
the node), and 3) being specified for incompatibility with a feature present in the target
node. So this type of competition would occur as in (3.37), imagining a word *thras that
was the nominalization of thrash.
76
(3.37) Competition
Target
Node:
conflicting feature
*thras: √THRASH [n]
√THRASH
[v]
[past]
blocked feature
thrash: √THRASH [v] ¬ [past]
With this type of notation, the insertion of a regular word like thrash or walk
cannot occur into a node that has fused with the past tense.
(3.38.a)
John thrashed.
TP
3
Johni
T’
3
vP
3
ti
(3.38.b)
Target
Node:
√THRASH
[past]
[v]
v’
3
√THRASH
[v]
[past]
Competition
blocked feature
thrash: √THRASH [v] ¬ [past]
77
The only possible VI in the inventory of English that can realize the target node in
(3.38.b) is the past tense affix –ed, which is specified only for the feature [past].
Crucially, in addition to discharging but not realizing the feature [v], inserting –ed also
discharges the root without realizing it. What sets content VIs apart from functional VIs
is that they realize a root. The “special”-ness of roots is that they that contribute the
extra-grammatical meaning to an utterance. In other words, a sentence doesn’t properly
convey a message if all the roots are discharged by functional morphemes without overt
realization. Thus this “special”-ness of roots is captured by a constraint on insertion that
a root must be realized by the VI which discharges it.
Since there is no VI that is specified for the root √THRASH and is not blocked
from the feature [past], no VI can be inserted into the derivation and thus the derivation
crashes. The only derivation containing both the feature [past] and the root √THRASH that
will converge will be one where the root and the [past] feature have failed to fuse, thus
resulting in insertion of both thrash and –ed.
3.3.2
Enter MINIMIZE EXPONENCE
The constraint MINIMIZE EXPONENCE proposed in Chapter 1 enters play at this point in
the analysis of root allomorphy. Up until now, fusion of the root with the functional
material accounts nicely for root allomorphy and proposes a simpler model of DM.
However, to this point I have not shown what motivates this fusion. Whereas the model
of grammar proposed here is more economical in the sense that it reduces the
computational load entailed by readjustment rules and secondary licensing, I have
78
proposed the ubiquitous application of the morphological process of fusion, which makes
the computational load of this grammar heavier—i.e. now a given derivation is littered
with applications of fusion, which entail more computation for every derivation. I will
show that this addition of the process of fusion to the derivation, while more
computationally intensive actually satisfies a larger economy constraint at the cost of
extra computation. That constraint, of course, is MINIMIZE EXPONENCE.
Returning to the speak trio, speak, like all verbs in English, is likely specified for
incompatibility with some feature in the complex of [3rd person], [singular] and [present],
since all verbs in most standard English dialects have third singular present overtly
spelled out with the affix –s (i.e. there are no strong verbs where there is an irregular
form for third person singular present). For ease of presentation, I will indicate that this
incompatibility with the notation ¬ [3sg], despite the fact that this notation simplifies a
more complex phenomenon. Thus, the three words are specified as follows:
(3.39.a)
Vocabulary Entry for speak
√SPEAK
[v]
¬ [3sg]
(3.39.b)
→
speak
/spik/
Vocabulary Entry for spoke
√SPEAK
[v]
[past]
→
spoke
/spowk/
79
(3.39.c)
Vocabulary Entry for speech
√SPEAK
[n]
→
speech
/spitS/
These specifications will result in exactly the distribution that we see with the three
forms. However, this analysis relies entirely on there being two different derivations
competing for convergence—the one where the [3sg] feature has fused and the one where
it has failed to fuse. It follows then that there must be those two derivations, one fused
and one unfused, for all utterances.
Imagine the two following possible derivations for John ate. In the derivation in
(3.48), [past] has fused with the √EAT, meaning that the VI candidate that will win
competition will be ate, which is specified for [past]. However, in (3.40), the node
containing √EAT has not fused with the feature [past]. That means that even if eat is
specified for ¬ [past], that specification will not stop eat from being inserted because the
past tense is in another node. Both eat and –ed would be inserted, meaning that the same
set of formal features could result in two possible utterances, *John eated and John ate.
80
(3.40) John ate.
TP
3
Johni
Application of
merger
T’
3
T
[past]
vP
3
ti
Spelled out form
v’
3
v
[v]
√EAT
Application of
merger
Complex result of head movement
T
3
v
T
3 [past]
v
√EAT
[v]
application
of fusion
√EAT
[past]
[v]
81
(3.41) *John eated.
TP
3
Johni
Application of
merger
T’
3
T
[past]
vP
3
ti
Spelled out form
v’
3
v
[v]
√EAT
Application of
merger
Complex result of head movement
T
3
v
T
3 [past]
v
√EAT
[v]
application
of fusion
T
3
√EAT [past]
[v]
Since *John eated is ungrammatical, it cannot be the case that two derivations for the
same feature set exist in parallel. Since a derivation with √EAT fusing with [3sg] will
crash because there is no VI compatible with the derivation, the trouble is that there is no
as-yet-formulated way that explains why *John eated crashes.
I propose that this is where MINIMIZE EXPONENCE, an economy constraint on the
grammar that prescribes that the most economical derivation is the one that is realized by
the fewest Vocabulary Items, is used to choose the most economical derivation (and thus
the one that will converge) between the two possible derivations of the same set of formal
features. In other words, the most economical utterance is the one that realizes all the
82
formal features that need to be realized using the fewest words possible. Recall from
Chapter 1, the definition of MINIMIZE EXPONENCE (repeated here as 3.42)
(3.42) MINIMIZE EXPONENCE: The most economical derivation will be
the one that maximally realizes all the formal features of the
derivation with the fewest morphemes.
The need to use as few words as possible and the need to express all the formal
features are both satisfied by using fusion. In that way, more than one formal feature is
realized by just VI, reducing the time and energy needed to utter one sentence. The
logical extension of this is that the ideal language would maximize the number of forms
like ate which capture both roots and formal features. However, that would mean a much
larger inventory of stored words, which is also inefficient. Thus the compromise is to
have fused forms for the most frequently used roots while leaving less frequent forms to
regular morphological processes. We see this effect of MINIMIZE EXPONENCE on the
Vocabulary cross-linguistically, lending credence to my proposal of its existence.
Thus, fusion is motivated by an effort to reduce an utterance to the fewest
pronounced morphemes (null morphemes and morphemes with overt morphology are
indistinguishable to this constraint). What blocks complete fusion of all functional
material with the content material it c-commands is the limitations of the Vocabulary
inventory of a given language as seen above and elaborated below.
To show Minimize Exponence in action, we return to the ungrammaticality of
*eated. The two derivations, John ate and *John eated, can be evaluated for economy
based on the constraint MINIMIZE EXPONENCE. In John ate, √EAT and [past] are realized
by only one VI. In *John eated, √EAT and [past] are realized by two VIs, eat and –ed.
83
Thus, as far as MINIMIZE EXPONENCE is concerned, John ate is the more economical
derivation and the one that converges. On the other hand, consider the word climb.
Since there is no *clomb, a word that is specified for √CLIMB and [past], and climb is
presumably specified for ¬ [past], climbed is the most economic derivation possible since
one where √CLIMB and [past] are realized by fewer VIs (i.e. only one) is not possible.
In this way, MINIMIZE EXPONENCE will force fusion to occur wherever it can
without resulting in a crash. Thus, the fusion we see in this chapter is actually driven by
the need to make the utterance contain as few morphemes as possible. Though fusing the
complex nodes that result from head movement involves more of a computational load,
the utterances ultimately created by a grammar containing MINIMIZE EXPONENCE are
more economical measured in the amount of energy consumed producing them rather
than in deriving them.
3.4
Inflection in Compounds
The MINIMIZE EXPONENCE analysis presented here offers a new analysis of a classic
observation about irregular root allomorphy. Jesperson (1909) observes that in addition
to traditional synthetic and analytic compounds, English contains a variety of compounds
that exhibit a unique behavior. Since the non-head member of these compounds is
always a noun, these compounds are usually called nominal compounds.
The unique behavior of these nominal compounds is that, normally, inflection is
not allowed in the non-head (left) member of the compound, even if the interpretation of
the compound would require that inflection. For example in (3.43, adapted from Sproat
84
1985), a coat-rack is specifically designed to hold more than one coat, yet it cannot be
called a *coats-rack.
(3.43) dog-lover
rat-chaser
log-cutter
hand-made
finger-bowl
coat-rack
*dogs-lover
*rats-chaser
*logs-cutter
*hands-made
*fingers-bowl
*coats-rack
This seems to be the effect of some constraint on the grammar of English that disallows
inflection in the non-head member of a nominal compound. However, a number of
researchers have observed that when the non-head member of the compound is an
irregular (i.e. it undergoes some sort of root allomorphy), the compounding of an
inflected form is allowed. (see 3.44, adapted from discussion in Thomas-Flinders 1981
via Sproat 1985 and from Kiparsky 1982). For example, while a head infested with fleas
must be flea-infested and never *fleas-infested, the same head infested with lice can
either be louse-infested or lice-infested. Similarly, a group of people jumping into a pool
leading with their heads cannot be jumping *heads-first, but those same people could
jump feet-first (example adapted from Sproat 1985).
(3.44) feet-first
lice-infested
teeth-marks
alumni club
dice pool
people eater
Kiparsky’s (1982) analysis for this data was couched within his model of Lexical
Phonology and Morphology (LPM). He proposed that morphological operations happen
85
at several levels. The first incarnation of his proposal included a Lexical level and a
grammatical level. His recent revisiting of LPM made it compatible with Optimality
Theory (LPM-OT) and proposed three different levels: the stem, the word, and the
phrase (Kiparsky 2003).
In both of his theories, this pattern where irregulars behave one way and regulars
behave another is captured by level-ordering. In LPM, Kiparsky proposed that the rule
that inflected irregular forms occurred before the compounding rule, which itself
happened before the rule inflecting regular forms.
(3.45) Ordering of operations:
1. Inflect irregular forms
2. Compound
3. Inflect regular forms.
Since louse is inflected to lice before compounding, lice-infested is permissible.
However, since compounding happens before regular forms are inflected, compounding
bleeds the regular inflection. Once rat has compounded with infested to form ratinfested, rat is no longer a valid target for the operation that adds –s. This means that
*rats-infested is blocked but lice-infested is permitted.
In LPM’s modern incarnation, LPM-OT, which doesn’t use rules but rather uses
constraint hierarchies, the prohibition of regular inflection in compound but the
permission of irregular inflection arises because the processes occur at different levels of
the grammar, which have different constraint rankings.
There have been many objections to the use of level ordering analysis of this
phenomenon, especially given that many models of the grammar reject either the idea of
86
level ordering or the particular levels that Kiparsky proposed. One such objection is that
regular inflection is not always prohibited from these structures. Hammond (1984)
observed that the inflection is allowed in compounds where the plural form is interpreted
as a group meaning (3.46, adapted from Hammond 1984). For example, an admissions
committee isn’t necessarily in charge of several admissions, but rather the process of
admissions.
(3.46) systems analyst
parks department
admissions committee
numbers racket
reservations desk
Similarly, pluria-tantum forms permit inflection of the non-head member of the
compound (3.47, adapted from discussion in Sproat 1985). For example, in addition to
pant-pocket being a legal compound, so is pants-pocket.
(3.47) pants-pocket
alms-giver
odds taking
To complete the picture of the data, it is important to show that while regular inflectional
morphology is blocked in English nominal compounds, regular derivational morphology
is allowed.
(3.48) grammaticality judgment
grading session
participation grade
marketing suggestion
cooler unit
shifter knob
copier service
unhappiness factor
87
An analysis of this pattern within the framework of DM is difficult for a number of
reasons. The first is that DM contains no generative lexicon. Thus, the analysis where
by compounds are created in a level-ordered lexicon is unavailable. By DM principles, if
one structure (regular inflection) is disallowed and another (irregular inflection) is
allowed, that must be indicative of different syntactic structure in the two forms.
However, in DM, the structures for *rats-infested and for lice-infested are largely
identical. The first difference between the two is that in *rats-infested the plural is
realized by an overt VI while in lice-infested the plural morpheme is realized by a null
morpheme. According to the tenets of DM, a VI without phonology is no different from
one with overt phonology. Thus, this cannot contribute to a difference in grammaticality.
The only other difference is that, in lice-infested, there has been the application of a
readjustment rule. Again, the application of a readjustment rule should not affect the
grammaticality of a syntactic structure.
(3.49.a)
*Rats-infested
T
3
V
-ed
3
Num
infest
3
n
-s
3
ø
rat
88
(3.49.b)
Lice-infested
T
3
V
-ed
3
Num
infest
3
n
ø
3
ø
lice
The next important problem for the framework of DM is that it makes no distinction
between “inflectional” affixes and “derivational” ones. From the point of view of DM,
both are just the overt realization of terminal syntactic nodes with VIs that happen to be
bound rather than free. Since DM doesn’t recognize a difference between these two
different types of affixation, referring to this difference is not a possible way to explain
the grammaticality difference between derivational morphology being allowed in these
structures but inflectional morphology being disallowed. Again, with the exception of
the specific functional heads involved, the structure of cooler-unit and *rats-trap should
be identical.
The inclusion of the economy constraint MINIMIZE EXPONENCE to the framework
of DM makes possible an analysis of the banning of inflection within English nominal
compounds. In section 3.4.1, below, I detail that analysis.
89
3.4.1
MINIMIZE EXPONENCE analysis
Compounding, in all of its forms, is an application of Merge in the syntax that creates a
phrase that is later adjoined into one “word” by an application of morphological merger.
In particular, compounding is an application of morphological merger to a pair of nodes α
and β, where α is a phrase (Xn, n>0) and β is a root, dominated by the phrase (or √P).
The √P resulting from Merge can then be sister to functional heads such as little-n or
little-v in order for the compound as a whole to participate in the syntax and receive
argument structure or undergo affixation.
The difference between the phrases that undergo compounding and those that
undergo affixation is the difference of labeling in the syntax proper (before spellout). In
particular, what will become compounds after spellout are dominated by a root phrase
whereas affixation is dominated by the functional head. Thus the application of Merge
that generates a compound merges Xn (n>0) to any √ and projects √ whereas affixation
merges √n to X0 and projects X where X is a functional head.
I argue that nominal compounds are an application of morphological merger in
English that adjoins a noun (nn, n>0) to a root under a projection of that root. More
specifically, nominal compounds are the joining of the feature [n] to a root. Since the
feature [n] is imbedded in the case of a regularly inflected form but, as a result of fusion,
is not imbedded in irregular forms, morphological merger can target an irregularly
inflected form but not a regular.
90
If nominal compounds are the result of an application of morphological merger to
two adjacent heads where one is a root and the other bears the feature [n], we can see
why regular inflection is blocked in forms like *rats-infested (3.50).
(3.50) *Rats-infested
A
3
compounding
v
-ed
3
√P
v
3
Num
√INFEST
3
rat 20
-s
[n]
In such forms, even after fusion, the feature [n] is imbedded below the inflection (in this
case Num). Since feature [n] is not adjacent to the root, compounding cannot target the
two nodes and merge them into one form (the root is adjacent to the feature [plural] in
(3.50)).
However, in forms with irregular inflection such as lice-infested, the root √LICE,
the n-head, and the num-head have all fused to make one node. This means that the [n]
feature is no longer embedded and is now adjacent to the root √INFEST. This adjacency
of the [n] feature to the root allows compounding to target that structure.
20
Shown after fusion.
91
(3.51) Lice -infested
compounding
A
3
v
-ed
3
√P
v
3
lice
√INFEST
Node contains:
√LICE
[n]
[plural]
Regular forms such as rats are two VIs composing a phrase dominated by –s, whereas
irregular forms like lice are only one VI (with no complex structure after fusion). The VI
for lice realizes both the root and the feature [n]. Thus, before insertion, the feature [n] is
adjacent to the dominating root.
Similarly, pluria-tantum cases and cases like admissions committee and systems
analyst are the effect of a root that has fused with an [n] feature compounding with
another root. For example, on this analysis, pants, scissors, odds, and alms are each
really just one Vocabulary Item (they refer to one thing) despite what appears to be
surface morphology. That is the VI for scissors is (3.52).
92
(3.52) Vocabulary Entry for scissors
√CUTTING APPARATUS 21
[n]
[plural] 22
→
scissors
/sIzr`z/
Similarly, an admissions committee is in charge of the process of admissions not several
admissions, the numbers racket involves the process of gambling, not dealing with
several numbers. Nouns like admissions and numbers are really simplex forms, not
complex ones containing inflection. Some may not even be linked to the same root as the
apparent stem (such as is the case with numbers racket). Rather, they are just one node
with just one corresponding VI.
(3.53.a)
Vocabulary Entry for numbers
√GAMBLING
[n]
[plural]
(3.53.b)
→
Vocabulary Entry for scissors
√CUTTING APPARATUS
[n]
[plural]
21
numbers
/n√mbr`z/
→
scissors
/sIzr`z/
I use this root to illustrate that the root is really about the referent not about the word. An unfortunate
effect of the language being studied and the metalanguage being the same is that it is sometime not clear
that the concept and the word are not the same thing.
22
we can assume that all these forms and pluria tantum forms are specified for [plural] because they cause
plural agreement.
93
(3.53.c)
Vocabulary Entry for admissions
√ADMISSIONS
[n]
[plural]
→
admissions
/QdmISn`z/
Since pluria-tantum forms such as scissors and “group” forms such as admissions are one
VI, they are inserted into a node where the root has fused with all the dominating
functional material, including [n] and [plural], above them just as lice is inserted into a
node where the root has fused with those features. That being the case, again in forms
like pants-pocket and numbers-racket the feature [n] is adjacent to the dominant root in
the compound so the application of morphological merger is legal (3.54).
(3.54) Numbers racket
compounding
n
3
√P
n
3
numbers
√INFEST
Node contains:
√GAMBLING
[n]
[plural]
Having addressed the banning of regular inflection in nominal compounds and the
admission of irregular inflection, pluria-tantum forms, and “group” nouns, we can
address why forms with derivational morphology are allowed despite the ban of having
94
morphology dominating the root. Recall that the targeting restriction of nominal
compounds is that the application of compounding must target the feature [n]. Since all
nominalizing heads contain the feature [n], we expect the derivational morphology should
be allowed in compounding exactly when the most dominant derivational morpheme is a
nominalizing head. When the non-head member of a nominal compound is dominated by
a nominalizing head, the maximal projection of that head, and thus the feature [n] is
directly adjacent to the root and thus can be the target of morphological merger.
This is exactly what we see in the data (3.48 repeated here as 3.55)
(3.55) grammaticality judgment
grading session
participation grade
marketing suggestion
cooler unit
shifter knob
copier service
unhappiness factor
As shown in (3.55), the derivational morphology allowed in the non-head position is not
unbounded. Rather, the dominant derivational morpheme is always a nominalizing
element, as seen in (3.56).
(3.56)
-ity
-ing
-ion
-er
-ness
Since each affix is a nominalizer, each realizes the feature [n]. Thus, each of the derived
words in the non-head position of a nominal compound is dominated by the feature [n].
95
This allows them to be the target of nominal compounding (since the feature [n] is not
embedded).
(3.57) cooler unit
compounding
nP
3
√P
n
3
nP
√UNIT
3
v 23
-er
3
[n]
v
√COOL
(3.66) acceptability judgment
compounding
nP
3
√P
ment
3
n
√JUDGE
3
a
-ity
3
[n]
-able
√ACCEPT
To summarize, nominal compounding is an application of morphological merger
that targets a root and the feature [n] and adjoins them. In the case of regular inflection in
the non-head member (such as *rats-catcher), that feature [n] is inaccessible to the
application of merger because the [n] feature is imbedded. However, as a result of the
fusion driven by the economy constraint MINIMIZE EXPONENCE, in irregular forms such
23
This head here probably fuses. I have shown it unfused so the derivation is clear.
96
as teeth-marks, pluria-tantum forms such as pants-pocket, and “group noun” forms such
as parks service, the [n] feature is no longer embedded and is now adjacent to the
dominating root. Compounding can now apply, resulting in grammaticality. Finally, in
the case of derivational morphology, the compounding is grammatical because, even
without fusion, the [n] feature is already adjacent to the root targeted for compounding.
Recall the two reasons that the blocking of regular inflection in nominal
compounds was problematic for Distributed Morphology: 1) the structures of regular
inflection and irregular inflection were identical and 2) there is no difference between
derivational morphology and inflectional, so there was no good way to block one and not
the other. The analysis here solves both those problems and provides an analysis for this
phenomenon within DM. 1) I propose here that the structures of *rats-infested and liceinfested are in fact not identical. Where rats has failed to fully fuse (thus the overt
realization of [plural]), lice has fully fused. 2) The blocking of *rats-catcher versus the
grammaticality of cooler unit is not due the difference between derivational morphology
and inflectional morphology, but rather is due to the embeddedness of the feature [n].
Thus, DM can maintain that there is no difference between derivational and inflectional
affixes.
3.5
Summary
The purpose of this chapter was to present an alternate analysis to lexical categories and
root allomorphy within the framework of Distributed Morphology. The traditional
analysis of lexical categories in DM is to have the content VI licensed for insertion
97
through secondary licensing (targeting the immediately c-commanding functional head).
The traditional DM analysis of root allomorphy employs the use of readjustment rules
(again conditioned by c-commanding functional heads) to change the phonology of on VI
to another VI. In the analysis presented in this chapter, I argued that licensing is a local
process of targeting features actually in the node targeted for insertion. I argued that the
root and the formal features come to occupy the same terminal node through the
application of the process of fusion. I extended this analysis to an analysis of root
allomorphy. This extension allows words like eat and ate to be separate VIs that compete
with each other for insertion rather than one being the result of the application of a
readjustment rules.
This proposal has a number of effects on the DM model of grammar. The first is
that it limits application of secondary licensing, removing lexical categorization from the
list of responsibilities of secondary exponence. Since secondary licensing is inherently
less local and thus less efficient than primary licensing, this is taken to be a strength of
this proposal. The other important effect of this proposal was to evacuate the need for
readjustment rules in the DM grammar. Since VIs linked to the same root now compete
with each other for insertion, readjustment rules are no longer needed to alter the
phonology of irregular forms. As readjustment rules only serve the purpose of changing
the phonology of irregular morphemes, since they aren’t needed for that function any
longer, the DM grammar doesn’t need them at all. Since readjustment rules entail both
long distance relationships and extra computational load, another strength of this
dissertation is that it proposes a model of DM without readjustment rules.
98
In the current model of DM, functional morphemes and content morphemes in
many ways participate in two different grammars. Whereas functional VIs participate in
competition, content VIs did not, relying on readjustment rules and secondary exponence
(never primary exponence) to license their insertion. The proposal here, in abolishing the
need for both secondary licensing and readjustment rules and in elaborating on a theory
of competition of roots, shows that it is possible for the model of the grammar to use just
one process for insertion of both functional and content VIs. As this simplifies the model
of the grammar, this is again taken as a strength.
This chapter also served to showcase MINIMIZE EXPONENCE. As proposed in
chapter 1, the purpose of MINIMIZE EXPONENCE is to capture the conflict between the
need to be maximally contrastive and maximally efficient at the same time. The
realization of MINIMIZE EXPONENCE here is the ubiquitous fusion of functional heads in
complex head arrangements. The effect of this fusion is that the features of those heads
are realized without each needing to be realized by its own VI. The efficiency in the
grammar comes from there being fewer VI items to be pronounced without loss of
contrast (any ambiguity is then only a result of a given language’s inventory). Thus,
fusion is one possible tactic available to the grammar to satisfy this central conflict. As
an added effect, a large portion functional heads that are realized as null morphemes in
DM instead fuse with other heads to be realized by overt morphemes. This drastically
reduces the number of null heads predicted by DM.
Finally, I proposed an analysis for the blocking of regular inflection in nominal
compounds in English (e.g. *rats-infested). For DM, this phenomenon has been difficult
99
to analyze due to the tenets of the framework. There were two central problems for DM:
1) in DM the structures lice-infested and *rats-infested are identical with the exception of
the application of a readjustment rule; and 2) DM recognizes no difference between
inflectional and derivational morphology (and derivational morphology is allowed in
such structures). The analysis shows here that the structures of *rats-infested and miceinfested are not identical: one involves more fusion driven by MINIMIZE EXPONENCE.
The grammaticality of the constructions reduces to the embeddedness of the feature [n]
(which meant that derivational morphology is allowed as long as a nominalizer was the
dominant morpheme in the non-head member).
Taken as a whole, this chapter does three things. It showed an alternative analysis
for root allomorphy. It proposed refinements to DM. It showed that that an analysis of
nominal compounds is possible in DM.
100
CHAPTER 4: SUBCATEGORIZATION WITH MINIMIZE EXPONENCE
Chapter 3 discusses the effects the constraint MINIMIZE EXPONENCE has on an analysis of
root allomorphy and offers a novel analysis for the unique behavior of inflection in
English nominal compounds. Chapter 4 extends the survey of the effects of MINIMIZE
EXPONENCE from the realm of morphological phenomena to the realm of syntax. In this
chapter, I explore an analysis of subcategorization that results from the claim that roots
fuse with functional heads merged above them in order to best satisfy MINIMIZE
EXPONENCE.
Chapter 4 is organized as follows: Section 1 offers a brief survey of the
phenomenon of subcategorization (or argument selection) and a description of the only
extant treatment of subcategorization within the framework of DM—Harley and Noyer
(2000). Section 2 provides a sketch of the effects fusion has on an analysis of the
argument selection of a verb. Sections 3 through 6 detail some of the predictions about
the behavior of verbs that a model based on the MINIMIZE EXPONENCE makes including
the polysemy of verbs, structural coercion, and dative alternations.
4.1.
Subcategorization
A tremendous amount of work has been done in syntactic theory on the very simple
observation that many verbs tend to appear with a variety of different argument
structures. (Grimshaw 1979, Chomsky 1981, Borer 1994, 2004, Pesetsky 1995,
Ramchand 1998, Baker 1988, and others). The aim of this work is to capture a number of
101
different behaviors such as polysemy (a single verb is found in different syntactic and
semantic environments).
(4.1). a.
b.
c.
d.
e.
f.
g.
h.
i.
Harry kept the bird in the cage (keep=detain)
Susan kept the money (keep=retain)
Sam kept the crowd happy (keep=maintain)
.
Let’s keep the trip on a Saturday (keep=maintain)
(Jackendoff 1996, definitions for keep are mine)
I ran.
I ran the dog.
I ran the dog to Julie
The water ran.
The water ran into the sink
Above, in examples (4.1.a-d), the verb keep appears in the same syntactic configuration
in all four examples, but its meaning changes to account for the different meanings of its
object. In older lexical syntactic theory, what meanings that a verb selected for and how
that selection affected the meaning of the verb was called S-selection (semantic selection,
Grimshaw 1979). In DM, this specification is for the root √KEEP and is analyzed as
extra-linguistic information found in the Encyclopedia.
However, in examples (e-i) we see the same verb appear in different syntactic
configurations. The verb run appears with only an agent (e), an agent and a patient (f), an
agent, a patient, and a location (g), only a patient (h), and a patient and a location (f).
Explaining this variability in argument structure has been a theoretical goal of several
different proposals (see Woolford 1984, Pesetsky 1995 for example).
A straightforward solution to the type of behavior seen in run and keep (4.1) is to
stipulate that the argument-selecting behavior of verbs is variable or there are multiple
102
instances of one verb, each with different selectional requirements. However, for many
verbs, certain semantic or syntactic frames are not permitted.
(4.2)
a.
b.
c.
d.
John put the paper on the shelf.
*John put on the shelf
*John put the paper.
*John put. (Pesetsky 1995)
e.
f.
The ship arrived.
*The captain arrived the ship.
g.
h.
I wondered whether the fate of the world was in my hands.
*I wondered the fate of the world.
In (4.2) we see a number of verbs that block certain structures. Put strictly requires an
agent, a patient, and a location. Any syntactic configuration with fewer than those
arguments is ungrammatical. Opposite to put, which requires a minimum number of
arguments, arrive blocks an “extra” argument. (4.2.f) is ungrammatical because of the
presence of an “extra” agent argument.
Some verbs seem to be sensitive to the category of their arguments, not only their
presence. (4.2.g, h) show that wonder is sensitive to the syntactic category of its
objects—it requires a CP. That some verbs are sensitive to the category of their
arguments is the part of subcategorization called C-selection (category selection,
Grimshaw 1979). This phenomenon is loosely called subcategorization (Grimshaw
1979), which is said to contain S-Selection and C-selection as well as idiosyncratic
argument structure.
103
To capture this behavior, work in syntactic theory has claimed that a verb is
lexically specified in some way for the environments that it is licensed to appear with
(Grimshaw 1979 and following). The scope of this type of analysis is not limited to a
simple licensing of a verb in some contexts but not in others. Theories of licensing also
must capture frame alternations such as the double object alternation (or dative shift).
(4.3)
a.
b.
Julie gave Ripley a bone.
Julie gave a bone to Ripley.
c.
d.
e.
f.
Julie delivered the scratching post to Gimli.
*Julie delivered Gimli the scratching post.
Jack asked Julie a question
*Jack asked a question to Julie.
Compounding studies of argument requirements of particular verbs is the
phenomenon of structural coercion (Gleitman 1990, Lidz 1998). Structural coercion is
the phenomenon whereby the specific argument requirements (and the corresponding
semantic and categorical restrictions) of a verb are violated, but the violation of those
requirements does not result in ungrammaticality. Rather, we are able to interpret the
utterance.
(4.4)
a.
b.
c.
d.
e.
The giant ham sandwich ate the dog (cf. The ham sandwich over-tipped 24 )
John thought the book to Mary
Fast-forward eats the tape.
The aliens glassed the planet
The bridge exploded the engineers.
The sentences in 4.4 are all grammatical. However, they are “odd” in one of two
different senses. Examples (a, c, and e) are semantically odd. All of them involve a
24
Notice that the interpretation of the giant ham sandwich ate the dog literally involves a sandwich capable
of eating a dog while the ham sandwich over-tipped only involves a person who ordered a sandwich.
104
violation of the s-selection of the verb. In (a), we are forced to imagine a giant dogeating ham sandwich in order to make it compatible with its role as the agent of eat. On
the other hand, in (c) we change the meaning of eat to something like “ruin through
catching in gears” in order to accommodate fast-forward as its agent. In the realm of
Minimalist syntax, these remain a problem. For DM however, these effects are extralinguistic and are “coerced” in the encyclopedia.
Examples (b and d) are crucially different. Both examples involve a verb
appearing with extra arguments (in the case of glass, appearing with any arguments at
all). Despite the extra arguments, both are interpretable. John thought the book to Mary
seems to mean that either he used telekinesis to move the book to Mary or he memorized
the book and used telepathy to transfer its contents. The aliens glassed the plant means
the aliens bombarded the planet so thoroughly that the surface of the planet was turned to
glass.
In the realm of construction grammar (Lakoff 1977, Langacker 1987, Goldberg
1995) this effect is attributed to the fact that syntactic constructions themselves contain
meaning and that structural coercion is just a matter of concatenation of the meaning of
the verb, the arguments, and the structure. In Minimalist syntax, the problem is more
perplexing because the structure itself carries no meaning. The coercion must result from
the presence of interpretable features. In Lexicalist Minimalism, the phenomenon is truly
troubling since a violation of a Lexical entry’s requirements should result in a crash of
the derivation due to the incompatibility of features in the syntax with those that need to
be checked in the verb.
105
Most of the work that has been done in this area of syntax and lexical semantics
has been done within Lexicalist models of syntax (Grimshaw 1979, Woolford 1984, and
others). As a result, much of this behavior has been captured by specifying a verb for the
types and number of arguments that it takes in its lexical entry and then allowing for
productive alterations of these specifications in a generative lexical component of the
grammar. In a model of syntax without a generative lexical component that precedes the
syntax, such as DM, this has to be captured differently.
4.1.1
Subcategorization within DM
The Lexicalist type of analysis is incompatible with late-insertion models such as
DM because the concept of the lexicon in DM is different from that of Lexicalism. In
DM, the lexicon as a generative mechanism does not exist, per se. Rather, the work of
the lexicon is distributed among many different modules within the syntax. Crucially, in
Lexicalist theories a verb can specify what types of elements it can be merged with and
then project the structure it appears in, since fully formed words are the atoms that are
manipulated by the syntax. However, since words are inserted into a fully formed
structure in DM, the words themselves cannot dictate the structure of the sentence.
Rather the word must be licensed for insertion into some derivations but banned from
insertion into others.
Harley and Noyer (2000) provide an analysis to capture the licensing of a VI in a
set of possible structures. In their analysis, the insertion of a VI is licensed via the
secondary exponence features of that VI. Harley and Noyer (2000) propose licensing
106
(discussed above in chapters 2 and 3) whereby particular verbs are licensed for insertion
by the features of elements that c-command them or that they c-command.
(4.5)
Phonology
a. sink
b. big
c. open
d. destroy
e. arrive
f. grow
Licensing environment
[±v], [+DP], [±cause]
[–v], [+DP]
[±v], [+DP], [±cause]
[+v], [+DP], [+cause]
[+v], [+DP], [–cause]
[+v], [+DP], [±cause]
For example, the VI for sink is specified for the following: a) it may or may not be ccommanded by a little-v head (the sink vs. The boat sank) and that little-v head may or
may not be causal (The boat sank vs. I sank the boat). Its object must be a DP. On the
other hand, destroy must be c-commanded by a causal little-v (*The city destroyed vs.
The barbarians destroyed the city) and its object must be a DP, while arrive must be ccommanded by a non-causal little-v (*The captain arrived the ship vs. The ship arrived).
Harley and Noyer’s analysis crucially relies upon secondary licensing in order to account
for subcategorization.
However, as shown above in chapter 3, the model of DM I propose uses fusion
and the MINIMIZE EXPONENCE constraint to eliminate the need for secondary licensing to
account for irregular morphology. Those same features of the proposed grammar
eliminate the possibility of using secondary licensing in this case as well since MINIMIZE
EXPONENCE will trigger the fusion of c-commanding elements with the root. Thus,
another explanation of subcategorization within DM is necessary. Section 4.2 details my
107
proposal for capturing subcategorization within DM making use of the fusion that is
driven by the economy constraint MINIMIZE EXPONENCE.
4.2
MINIMIZE EXPONENCE Based Account of Subcategorization.
As we saw in Chapter 3, employing MINIMIZE EXPONENCE eliminates the use of
secondary licensing in order to capture a VI’s lexical class because the root fuses with the
c-commanding functional heads in order to best satisfy the constraint. Since in any given
derivation, fusion of roots with the functional heads projected above them allows the
roots to gain the features of the functional those heads, corresponding Vocabulary Items
may be specified for those features gained through fusion.
Following the claim that every argument is introduced to the syntax by a
particular functional head, the fusion analysis easily extends to include subcategorization,
allowing for an analysis that is not dependent on secondary licensing. This new analysis
shows another application of the fusion driven by MINIMIZE EXPONENCE.
Marantz (1984), Larson (1995), Kratzer (1994, 1996) and many others have
argued that agent arguments are not selected for by verbs. As I have assumed in my trees
to this point, the agent argument is selected by a functional head called “light verb”
(little-v) or VOICE that is generated above VP (Kratzer 1994). Further work in event
structure has argued the light verb head to be the locus of the meaning CAUSE (see
Kratzer 1994, 1996, Ramchand 1997, Harley 1995, and related).
Following the “severing” of the external argument from the verb, Borer (1994),
Jelinek (1998), Ramchand (1997, 1998) and others have also endeavored to sever the
108
theme argument from the verb as well. They have argued that the theme argument is also
projected by a functional head. The name of this head varies from author to author. For
the sake of clarity and ease of presentation, I will adopt Jelinek’s (1998) name for that
head, TRANS, so that it will easily be differentiated from the other verbal head, little-v.
Since this functional head takes the root as its sister, the theme argument must be
projected in specifier position. Under such an analysis, the structure of a transitive
sentence looks like (4.6) (shown before movement):
(4.6)
The dog ate the bone.
TP
3
T’
3
T
vP
3
the dog
v’
2
v
TransP
3
the bone
Trans’
2
Trans √EAT
As a matter of parsimony, we can also assume that functional heads that are
merged above the root (Pesetsky 1995) project all arguments including instruments,
goals, experiencers, locatives, etc. I assume this throughout this dissertation. For the
purposes of this dissertation, I give simple names to the argument heads that I show. The
head that projects a dative object I call G and the head that projects locatives I call L.
109
In Chapter 3, I propose that all functional heads carry formal features that are
identifiable for competition. Here, I extend that proposal to include the functional heads
that project arguments. The features (or feature bundles) carried by these heads bear a
meaning that roughly corresponds with the theta role of the argument that the head
licenses. For example, the type of “v” that licenses agent arguments carries the feature
[v], which has the meaning CAUSE (Kratzer 1994, Ramchand 1997, 1998). The head
that projects themes, Trans, carries the feature [Trans]. 25
Like all other features, these “argument features” move up the tree through
normal head movement. Recall that head movement (an instantiation of morphological
merger) will cause the root to move up the tree and collect the c-commanding formal
features into a complex head. In the example above (4.6), √EAT moves to v through
Trans. As it moves, the features of each head √EAT attaches to are added to the complex
head structure through head adjunction. Again, we assume a Bobaljik-style “merger
under adjacency” analysis of the marking of tense on verbs in English (see Bobaljik
1994). The combined applications of merger under adjacency and head adjunction result
in the complex head show in 4.7.
25
At this point in this dissertation, I am treating [Trans] as one feature. However, it is likely that these are
actually a bundle of features. For now, the difference is moot. I will return to this distinction below when
it becomes important.
110
(4.7)
Complex head created by derivation in (4.6).
T
3
v
T
3
[past]
Trans
v
3
[v]
Trans
√EAT
[Trans]
Again, as shown in Chapter 3, the process of fusion applies to the complex head resulting
in the simplex head in (4.8), which carries the formal features [v], [Trans], and [past] as
well as the root.
(4.8)
√EAT
[v]
[V]
[past]
This simplex node is now a candidate for vocabulary insertion and the discharge
of its features by just one VI rather than the four that would have been necessary to
discharge the features of the four different heads in the complex structure. Thus, root VIs
can be specified for these features just as they can be specified for any other features.
The features of a root VI can be determined by examining the possible and impossible
environments for that word and finding feature specification that is compatible with the
possible environments and incompatible with the impossible ones. As an example, below
I show how we can arrive at the feature specifications for ate.
111
4.2.1
Determining the Feature Specification of a Vocabulary Item
To show how to determine the specifications of a VI, I continue with the example
of ate. Since the VI ate can be inserted into the node created in (4.6) above, The dog ate
the bone, we can conclude from ate's insertion in this context that it is specified for some
subset of the features contained in that node, up to and including all of them. If it had
any more features, insertion would be blocked. In (4.9.a), I show a derivation of The dog
ate the bone that results in a fused node containing the features [v], [past], and [Trans].
In (4.9.b), I show all the possible specifications of ate and how we can determine the
specifications for ate since we know it is allowed in the node created in (4.9.a). Since we
know that ate is allowed in such a derivation, we know that the specifications in (4.9.c)
are the maximum specifications for a VI for ate (an overspecified candidate would be
disallowed).
(4.9)
a.
The dog ate the bone.
TP
3
The dogi
T’
3
vP
3
v’
3
√EAT
TransP
[v]
3
[past]
the bone
Trans’
[Trans]
3
ti
112
b.
Possible specifications for ate
Node that
ate is
allowed in:
not a possible specification
(overspecified)
ate: √EAT [v] [past] [Trans] [G]
maximum possible specification
ate: √EAT [v] [past] [Trans]
√EAT
[past]
[v]
[Trans]
possible specification
ate: √EAT [v] [past]
not a possible specification
(conflicting feature)
eat: √EAT [v] ¬ [past]
c.
Maximum specification for Vocabulary Item for ate
√EAT
[v]
[Trans]
[past]
→
ate
/ejt/
Just examining the maximum features necessary for insertion into The dog ate the
bone is not sufficient to determine the specification of ate. Since this is an
underspecification model of subcategorization, a VI that is underspecified for one or both
of [Trans] and [v] are still possible VIs (we know that [past] is required to distinguish ate
from eat).
We can precisely determine the specification for ate by examining other
environments it can appear in, as well as environments in which it loses the competition
113
to another VI. Ate can appear in a derivation that does not have an object such as in
sentences like Julie ate. If we assume a structure for Julie ate where the object is not
base-generated rather than an elision account of the missing object, its structure is (4.10).
(4.10) a.
Julie ate.
TP
3
Juliei
T’
3
T
[past]
vP
3
ti
Spelled out form
v’
3
v
[v]
Complex result of head movement
T
3
v
T
3 [past]
v
√EAT
[v]
√EAT
application
of fusion
√EAT
[past]
[v]
114
b.
Resulting form
TP
3
Juliei
T’
3
vP
3
ti
target node
Ready for Insertion
v’
3
√EAT
[v]
[past]
In the structure in (4.10), there is no [Trans] feature to be fused to √EAT. It is clear then,
since ate can still be inserted into the node that the VI for ate must not be specified for
[Trans].
(4.11) Possible specifications for ate
Node that
ate is
allowed in:
not a possible specification
(overspecified)
ate: √EAT [v] [past] [Trans]
maximum possible specification
ate: √EAT [v] [past]
√EAT
[past]
[v]
possible specification
eat: √EAT [past]
However, since ate cannot appear as an unaccusative or a zero-derived nominal, the VI
for ate must be specified with the feature [v].
115
(4.12) a.
*The sandwich ate. (where the sandwich is devoured)
TP
3
The sandwichi
T’
3
T
TransP
[past] 3
ti
Trans’
3
Trans
√EAT
[Trans]
Complex result of head movement
T
3
v
T
3 [past]
Trans √EAT
[Trans]
b.
application
of fusion
Spelled out form
√EAT
[past]
[Trans]
Resulting form
TP
3
Juliei
T’
3
vP
3
ti
target node
v’
3
√EAT
[Trans]
[past]
Ready for Insertion
116
c.
Competition (no winning candidate...both eat and ate require [v])
Target
Node:
overspecification
√EAT
[past]
[Trans]
(4.13) a.
ate: √EAT [v] [past]
conflicting specification
eat: √EAT[v] ¬ [past]
*The dog did the ate..
TP
3
The dogi
T’
3
tk
vP
3
ti
v’
3
didkj
TransP
qp
DP
tj
3
the
nP
3
√EAT
[n]
117
b.
Competition
(no winning candidate)
Target
Node:
overspecification
√EAT
[n]
overspecification
ate: √EAT [v] [past]
eat: √EAT[v] ¬ [past]
The competitions in (4.12.c) and (4.13.b) show us that in order to capture the
ungrammaticality of *The sandwich ate and *The dog did the ate, ate must be specified
for the [v] feature. Thus the relevant features that the VI ate is minimally specified for
are [past] and [v], which can be interpreted as the VI requiring that it a) realize the root
√EAT, b) have an agent, and c) be past tense. Since this is the same as the maximum
specification we determined to be possible in (4.1), Julie ate, we can conclude that (4.14)
is the specification for ate. Conveniently, this means that we arrive at the same VI we
assumed in Chapter 3.
(4.14) Vocabulary Item for ate
√Eat →
[v]
[past]
ate
/ejt/
The VI in (4.14) predicts that ate will be able to appear in any environment as
long as the utterance is in the past tense and as long as there is an agent argument. The
VI is indifferent to whether there is a theme present.
118
4.2.2 Differences among Subcategorization Models
The predictions above are different from the predictions made by Lexicalist
theories of subcategorization such as those proposed by Grimshaw (1979), Chomsky
(1981), Pesetsky (1995), Ramchand (1997,1998), Baker (1988), and others in Lexicalist
models. This model proposes that there is only one instantiation of a VI and that that VI
will appear in a multitude of different environments as long as those environments
include at least the minimum number of arguments. In this way, this is an
underspecification analysis. Lexicalist analyses tend to include lexical entries that are
fully specified for their arguments or that contain optional elements (see Woolford’s 1984
analysis).
However, within the realm of DM, the Harley & Noyer (2000) analysis (repeated
as 4.15) can easily be made to make underspecification predictions in the same way that
this analysis does if the optional elements (such as [±v]) are instead treated as unspecified
(4.16).
(4.15) Harley and Noyer (2000) licensing
Phonology
a. sink
b. big
c. open
d. destroy
e. arrive
f. grow
Licensing environment
[±v] [+DP] [±cause]
[–v] [+DP]
[±v] [+DP] [±cause]
[+v] [+DP] [+cause]
[+v] [+DP] [–cause]
[+v] [+DP] [±cause]
119
(4.16) Harley and Noyer (2000) licensing adjusted to an underspecification analysis.
Phonology
a. sink
b. big
c. open
d. destroy
e. arrive
f. grow
Licensing environment
[+DP]
[–v] [+DP] N/A
[+DP]
[+v] [+DP] [+cause]
[+v] [+DP] [–cause]
[+v] [+DP]
The availability of an underspecification analysis is a benefit of the competition
aspect of DM. The subset principle dictates that VIs are often inserted into environments
with more features than the VIs are specified for. Thus, VIs are already underspecified
for the features they realize in the sense that they discharge all the features in a node, not
only the ones they are specified for.
The aspect of this analysis that sets it apart from the Harley and Noyer (2000)
analysis is not the predictions that it makes, but rather the locality of the effect that
licenses insertion of the VI. Since the Harley and Noyer (2000) analysis uses secondary
licensing, the same concerns about licensing we discussed in Chapter 3 apply to the
Harley and Noyer (2000) analysis. That is, the constraints on insertion are inherently
non-local: The VI must not only check the node for compatibility but also the rest of the
VP shell. In the fusion analysis, the VI only checks features in the node it is inserted
into, a process that is more local and is thus more economical.
Having now sketched the fundamentals of the fusion-based account of thematic
licensing of verbs, I dedicate the rest of Chapter 4 to the finer aspects of this analysis
including what behaviors are predicted by this model.
120
4.2.3
Evidence for the Feature [v] in English
The proposal here rests on the assumption that functional heads that select for arguments
contain the features such as [Trans] and [G] that I propose above. Since the complex
heads containing these features undergo fusion, the features themselves are realized by
the same VI that realizes the root. Therefore, the features that I propose here are often
not realized by overt morphology. It can be argued that verbalizing morphemes such as –
ize and -ify, realize that feature.
However, if the features such as [Trans] can be selected by Vocabulary Items, it
follows that there should be minimal pairs that are differentiated by those features. Just
as eat and ate are different only in that ate selects for [past] and speech and speak differ
only in that speech selects for [n], there should be pairs that are different only in that they
select for [v]. An example of such an alternation is the rise/raise alternation.
(4.17) a.
b.
My hand rose from the table.
I raised my hand off the table.
The form rise appears as an unergative while raise appears as a transitive verb.
Under this analysis, the VI for raise must be specified for the [v] feature, requiring it to
be inserted into a derivation with an agent argument. Since there is an alternation for
transitivity here, this means that the VI for rise must not be specified for [v] (also
evidenced by the fact that rise appears without an agent).
121
(4.18) Vocabulary Item for rise 26
√RISE →
rise
/rajz/
(4.19) Vocabulary Item for raise
√RISE →
[v]
[Trans]
raise
/rejz/
Both VIs are linked to √RISE. In this way, the alternation of rise and raise is
treated as parallel to the alternation of eat and ate. Thus, the alternation of rise and raise
can be taken to be evidence that the features I propose here do exist in the grammar of
English—the difference between the two words is the presence of the feature [v] (and,
coincidentally, [Trans]).
4.2.4
Blocking “Extra” Arguments
As I have set this proposal out so far, it is apparently too strong. It entails that any verb
can appear with any number of arguments beyond its minimum, as shown by run in
Section 2.2 of this chapter. However, there are verbs that are ungrammatical in the
specific constructions where they appear with more arguments than its “default”
conditions (4.20).
(4.20) a.
b.
c.
d.
26
The ship arrived.
*The captain arrived the ship.
Ripley fell down the stairs.
*Julie fell Ripley down the stairs.
Note that rise is not actually specified for [Trans]. Rise not only appears as a zero-derived nominal, but
it can also be argued that the intransitive forms it surfaces in can either be agentive (I rise early from bed)
or non-agentive (The X-wing rose from the swamp).
122
In (4.20b), the “extra” agent argument is blocked for arrive. What is necessary, then, is
to design a mechanism within the framework of Distributed Morphology to block the
insertion of VIs into nodes that they otherwise qualify for. We return then to the ¬
specification proposed in chapter 3. By specifying a Vocabulary Item for incompatibility
with a feature carried by one of the functional heads that licenses an argument, we ensure
that that VI cannot be inserted into a node that has that particular root and that particular
feature together. This feature blocking will prevent a derivation containing that root and
that feature from being grammatical unless that feature is otherwise realized by another
VI. Ultimately, this means that there is an incompatibility of the root the VI realizes and
the blocked feature. Since the head is what licenses the merging of the argument, the
argument and the VI will not co-occur unless there is another overt morpheme, such as a
transitivizer, to license the argument.
To see how this works, let’s return to the example of arrive. Arrive is
incompatible with an agent argument, as we see with the ungrammaticality of sentences
like *The engineer arrived the train. Such behavior is indicative of the VI for arrive
being incompatible with [v], the feature from the little-v head which licenses the agent
argument. The VI would then be (4.21, ignoring specifically morphological specification
like ¬ [past]).
(4.21) Vocabulary Item for arrive
√ARRIVE
¬ [v]
[Trans]
→
arrive
/´rajv/
123
A VI specified in this way is blocked from insertion into the derivation of *The
captain arrived the ship because the ¬ [v] it is specified for is not compatible with the [v]
in the fused head it is inserted into in (4.22).
(4.22) a.
*The captain arrived the ship.
TP
3
The captaini
T’
3
T
vP
3
ti
v’
2
v
TransP
3
the ship
Trans’
2
V
√ARRIVE
T
3
v
T
3
[past]
Trans
v
3
[v]
√ARRIVE
Trans
[Trans]
b.
T
3
√ARRIVE
[v]
[Trans]
[past]
Competition (no winning candidate)
Target
Node:
√ARRIVE
[Trans]
[v]
conflicting specification
arrive: √ARRIVE ¬ [v] [Trans]
124
Of course, the VI for arrive in (4.21) is compatible with a derivation that does not include
a little v. Thus, arrive is easily inserted into a derivation such as The ship arrived (4.23).
(4.23)
The ship arrived.
TP
3
the ship
T’
3
T
TransP
3
ti
Trans’
2
Trans √ARRIVE
T
3
Trans
T
3
[past]
√ARRIVE
Trans
[Trans]
b.
T
3
√ARRIVE
[past]
[Trans]
Competition
Target
Node:
inserted candidate
arrive: √ARRIVE ¬ [v] [Trans]
√ARRIVE
[Trans]
A VI specified in this way captures the grammatical incompatibility of arrive with
the light verb. Crucially, there is no claim in this formalization that the meaning of arrive
is incompatible with the meaning CAUSE. We know that the meaning of arrive is
compatible with the meaning of CAUSE because things in the real world can certainly
125
arrive with some sort of external causation. This type of blocking shows the illformedness of the otherwise meaningful *The captain arrived the ship to be purely
grammatical. In this way, the model proposed here presents an elegant account for the
blocking of a verb into particular environments where there is an “extra” argument.
4.3
Rampant Polysemy
A strong prediction of this analysis is that a verb will be able to appear in any
environment that has at least the minimum number and types of arguments that it is
specified for and none that it bears a ¬ feature for. Since a VI such as ate is only
specified for the features [v] and [past] it must be able to appear in any environment as
long as an agent argument is present. This is the exact behavior this analysis sets forth to
capture. In the paradigm in (4.24), we see the same root appear in a number of different
semantic and thematic environments.
(4.24) a.
b.
c.
d.
I run.
I run the race.
I run the dog.
The water runs.
e.
f.
g.
h.
I run Mary the paper
I run the paper to Mary.
John went on the run
John built a run for the dog.
A lexicalist approach to this behavior would need to generate at least five different lexical
entries for run based on syntactic structure alone. Acknowledging the fact that (4.24.b)
and (4.24.c) have the same structure but the interpretations are slightly different, the
number of lexical entries grows substantially. Excluding the possibility of a
transformation within the lexicon for the double object construction or a Woolford (1984)
style optional theta-grid, a strong lexicalist theory needs to generate eight different lexical
entries to capture the above behavior, each with its own s-selection and linking.
126
However, an underspecified analysis of this behavior as proposed here captures
this behavior with just one VI.
(4.25) Vocabulary Item for run
√RUN
¬ [3sg]
→
/r√n/
run
This VI can be inserted into any of the structures in (4.24) as its requirements—a root
only—are a subset of the features of the fused nodes in all eight derivations, including the
zero derived nominals, as illustrated in (4.26-28).
(4.26)
I run.
TP
3
Ii
T’
3
T
vP
3
ti
T
3
v
T
3
[3sg]
v
√RUN
[v]
v’
2
v
√Run
T
3
√RUN
[3sg]
[v]
127
b.
Competition
Target
Node:
inserted candidate
run: √RUN ¬ [3sg]
√RUN
[v]
(4.27)
The water runs.
TP
3
the wateri
T’
3
T
TransP
3
ti
Trans’
2
Trans √RUN
T
3
Trans
T
3
[3sg]
√RUN
Trans
[Trans]
b.
T
3
√RUN [3sg]
[Trans]
Competition
Target
Node:
√RUN
[Trans]
inserted candidate
run: √RUN ¬ [3sg]
128
(4.28)
I run Mary the paper.
TP
3
Ii
T’
3
T
vP
3
ti
v’
3
v
GP
3
Mary
G’
3
G
TransP
3
The paper
Trans’
3
Trans
√RUN
T
3
v
T
3 [present]
G
v
3
[v]
Trans
G
3
[G]
√RUN
Trans
[Trans]
T
3
√RUN [3sg]
[Trans]
[v]
[G]
129
b.
Competition
Target
Node:
inserted candidate
run: √RUN ¬ [3sg]
√RUN
[Trans]
[v]
[G]
(4.29)
John built a run for the dog.
TP
3
√RUN
[n]
b.
T’
ei
T
vP
wo
John
v’
qp
builtj
TransP
qp
DP
Trans’
3
3
a
nP
Trans
tj
ei
n’
PP
3
4
n
√RUN
for the dog
Competition
Target
Node:
inserted candidate
run: √RUN ¬ [3sg]
√RUN
[n]
Lexicalist theories are often dependent on rigid s-selectional or thematic
properties of lexical entries. When a verb behaves the way that run does above, a lexical
130
theory of this sort is forced into the precarious prediction: They are committed to the
existence of a vast number of lexical entries which are mostly synonymous (i.e. linked to
the same concept) and completely homophonous. These “homonyms” differ only by the
syntactic environments they are licensed for. The underspecification analysis of a verb’s
licensing conditions that is presented in this paper captures why verbs appear to be
polysemous without asserting a lexicon full of homonyms.
4.4
Structural Coercion
The crux of this proposal is that roots are underspecified for the environment in which
they can appear. It follows that the grammar will produced structures in which a root
appears with more arguments than it normally does. A root may even appear in an
environment that our real world knowledge is not compatible with. For example, the
utterance #The ham sandwich ate the dog is produced grammatically by the syntax. (see
syntactic bootstrapping literature: Lidz 1998, Naigles 1993, Gleitman 1990; see also Lidz
and Gleitman 2004 for rebuttal) The Encyclopedia marks this sentence as ill-formed
because ham sandwiches do not make good eaters. However, we can adjust our real
world knowledge to fit the sentence by interpreting ham sandwich in such a way that a
ham sandwich is somehow capable of eating a dog. We can also adjust the meaning of
eat rather than the meaning of its arguments. This is of course how we get sentences
such as Fast forward eats the tape
A verb that appears with “extra” arguments that are “incompatible” with our real
world knowledge forces us to coerce our real world knowledge to fit the utterance,
131
especially if, as listeners, we are to assume that the speaker strictly adheres to Grice’s
maxims. In this way, when we are presented a sentence such as John thought the book to
Mary, we are able to interpret it.
Not only does the framework proposed here account for structural coercion, it
predicts it. For example, since thought typically appears in unergative structures (or in
CP complement transitive structures) we can preliminarily assume that the VI for thought
is specified only for [v] (and the past tense).
(4.30) Vocabulary Item for thought
√THINK
[v]
[past]
→
thought
/Tat/
Since thought is so underspecified, it easily appears in an environment that is specified
more fully for arguments, such as the environment in #John thought the book to Mary. In
that environment, the fused node that the VI is inserted into (4.30) is a superset of the
features the VI contains.
132
(4.31)
John thought the book to Mary
TP
3
Johni
T’
3
T
vP
3
ti
v’
3
v
TransP
3
the book
Trans’
3
Trans
LP
3
to Mary
L’
3
L
√THINK
T
3
v
T
3 [past]
Trans
v
3
[v]
L
Trans
3
[Trans]
√THINK
L
[L]
√THINK
[past]
[Trans]
[v]
[L]
133
Competition
Target
Node:
inserted candidate
thought: √THINK [past] [v]
√THINK
[past]
[Trans]
[v]
[L]
4.5
Dative Alternations
One of the more difficult alternations that subcategorization theories have tried to capture
is the dative alternation. Historically, the pattern seen in (4.32) was a classic example of
a transformation where the structure with an oblique indirect argument was transformed
into a double object construction with the same meaning (Chomsky 1965). More recent
work (Oehrle 1976 and following) has shown the two different structures to be different
derivations without a transformational connection. In the realm of work within
subcategorization, the onus then fell on explaining why, especially if the sentences
conveyed the same information, some verbs allow both structures while others prohibit
one structure and allow the other (4.32).
(4.32) a.
b.
c.
d.
e.
f.
Julie gave Ripley a bone.
Julie gave a bone to Ripley.
Julie delivered the scratching post to Gimli.
*Julie delivered Gimli the scratching post.
Jack asked Julie a question
*Jack asked a question to Julie.
Oehrle (1976) observed that the two structures do not actually entail the same
relationship between the arguments to the verb. Rather than a goal argument being
134
present in both the “to dative” construction and the double object construction, the “to
dative” contains a locative argument (instead of a goal argument) and a thus a locative
head that projects it.
The fact that the “to dative” corresponds to location rather than goal arguments is
exemplified by the word sent (4.33).
(4.33) a.
b.
Jack sent Julie a message.
Jack sent a message to Julie.
When a person is the DP in the locatum/goal alternation such as in (4.33), the
interpretations of both sentences are so close that they give rise to the intuition that the
sentences are related. However, if the locatum/goal DP in the alternation is a place rather
than a person as seen in (4.34), it becomes clear that there are two different roles being
associated with the positions because only one of the structures is permitted. This shows
that they are not the same argument: a person is a “good” goal and a “good” location. A
place, however, is not a good goal, only a good location.
(4.34) a.
b.
Julie sent the package to France
#Julie sent France a package.
In the derivations above, I have been calling the head that projects the goal argument G
and its corresponding feature [G]. I will continue that practice here, differentiating it
from the locative, which I call L. With the assumption of two different heads comes two
different derivations for the locative (4.35) and the dative (4.36). Applying the fusion
analysis described above results in two different fused heads that are subject to VI
insertion as illustrated below.
135
(4.35) Julie gave Ripley a bone.
TP
3
Juliei
T’
3
T
vP
3
v’
ti
3
v
GP
3
Ripley
G’
3
G
TransP
3
a bone
Trans’
3
Trans √GIVE
T
3
v
T
3 [past]
√GIVE
G
v
[past]
3
[v]
[Trans]
Trans
[G]
[v]
3
[G]
Trans
√GIVE
[Trans]
Competition
Target
Node:
√GIVE
[past]
[Trans]
[v]
[G]
inserted candidate
gave: √GIVE [past] [Trans] [v]
136
(4.36) Julie gave a bone to Ripley.
TP
3
Juliei
T’
3
T
vP
3
ti
v’
3
v
TransP
3
the bone
Trans’
3
Trans
LP
3
to Ripley
L’
3
L
√GIVE
T
3
v
T
3 [past]
Trans
v
3
[v]
L
Trans
3
[Trans]
L
√GIVE
[L]
√GIVE
[past]
[Trans]
[v]
[L]
Competition
Target
Node:
√GIVE
[past]
[Trans]
[v]
[L]
inserted candidate
gave: √GIVE [past] [Trans] [v]
137
This type of analysis of dative constructions allows us to generalize about the
specifications for VIs of different classes of verbs. For example, words like deliver and
donate allow locatives and themes optionally but strictly prohibit goals. Such a class of
verbs would have a VI such as in (4.37).
(4.37) Vocabulary Item for deliver
√DELIVER
[v]
¬ [G]
→
deliver
/d´lIvr/
Such a VI allows insertion into a derivation with or without a locative argument and with
or without a theme, but expressly bans its insertion into a derivation with a goal
argument.
(4.38) a.
b.
c.
d.
John delivered the pizza.
John delivered the pizza to Mary.
*John delivered Mary the pizza.
John hasn’t delivered to Mary yet.
Similarly, the class of verbs that behaves like ask would be underspecified for most
arguments but blocked in the case of [L]. Verbs such as give and run would be truly
underspecified in terms of L and G. In such a way, a MINIMIZE EXPONENCE-based fusion
approach to dative alterations that captures “dative shift” as a difference in a verb’s
selecting for or blocking locative arguments and goal arguments.
4.6
C-selection
Recall from (4.2, repeated here as 4.39) that there are words like wonder which select for
the category that arguments must have.
138
(4.39) a.
b.
I wondered whether the fate of the world was in my hands.
*I wondered the fate of the world.
Wonder requires that its theme must be a CP. As discussed above, this aspect of
subcategorization has historically been called category selection. This appears to be a
problem for the analysis shown here because the specification of wonder—which by
hypothesis would require the feature [Trans]—mandates only that the root be
c-commanded by the Trans head, which introduces theme arguments. That specification
has no way of determining that the specifier of Trans be a CP.
Harley and Noyer (2000) proposed that the little-v head comes in different
varieties. If we extend that claim, all the heads that license arguments have different
varieties. Thus, there is a Trans head that licenses a DP in its specifier, another that
licenses a CP, and yet another that licenses a TP. Similarly, the claim that some vPs
license the meaning CAUSE and others don’t can be extended to Trans to apply to the
difference between affected and non-affected objects: one Trans head licenses the
meaning BECOME, another doesn’t. The one that does not license BECOME would be
the Trans head present in sentences with verbs like push, wipe, hit etc.
Throughout this dissertation, I have claimed that little-v carries one feature [v]
and similarly Trans carries one feature [Trans]. Given that there is a variety of those
heads, the heads are not composed of just one feature, but rather a bundle of features and
that the different types of these heads are a result of subtle differences of the contents of
those bundles. Words like wonder which specifically select for the Trans head that
allows for CP specifiers and words like push that specifically select for the Trans head
that does not carry a feature corresponding to BECOME can be assumed to be better
139
specified than VIs that select for just [Trans]. That is, all VIs that require themes have
been marked here as selecting for the feature [Trans], which is shorthand for the bundle
of features contained in any given Trans head. Those VIs that are specific about which
Trans heads they are compatible with must be even better specified for the features from
the [Trans] head they must realize. They will be specified for a feature in a particular
Trans head that sets it apart from the other Trans heads. Thus, wonder is only compatible
with CP objects because the VI is only compatible with the features contained in the
specific Trans head that licenses CP arguments.
4.7
Summary
The ultimate purpose of this chapter was to extend the predictions of the ubiquitous
fusion motivated by MINIMIZE EXPONENCE beyond the realm of morphological
phenomena to the realm of syntactic phenomena. To this effect, I showed that a fusion
analysis can explain subcategorization. I argue here that the selectional behavior of a
given verb is an effect of the functional features that the VI is specified for. Arguments
are selected for by functional heads. Those functional heads fuse with the root, giving
the target node both the root and the relevant formal features. This allows the root VI to
select for those features.
Since DM is an underspecification model, this led to the prediction that roots
should be compatible with arguments in excess of their core requirements, which can be
seen in normal verbal behavior and extends to structural coercion. I also extend the
feature blocking system described in chapter 3 to account for the blocking of arguments
140
(*The engineer arrived the train). The end result is to show that the fusion motivated by
MINIMIZE EXPONENCE also drives the selectional behavior of verbs.
141
CHAPTER 5: REMAINING ISSUES
Chapters 1, 3, and 4 represent the central proposals of this dissertation. The purpose of
this chapter is to examine some other issues that result from MINIMIZE EXPONENCE. By
its very nature, this chapter comes in the form of a list of several unconnected discussions
on avenues of further research.
Section 1 considers the possibility of an interface between syntax and neoDavidsonian event semantics that arises as a result of both the ubiquitous fusion driven
by MINIMIZE EXPONENCE and the subcategorization analysis proposed in Chapter 4.
Section 1 first details the model of semantics based on Parsons 1990, then its apparent
incompatibility with modern syntactic theory. Then I examine how the proposals in this
dissertation point to a possible interface between Minimalist syntax and neo-Davidsonian
event semantics.
Since the majority of this dissertation focuses on English data, section 2 of this
chapter is included to present some of the typological predictions that this model makes.
In a similar vein, Section 3 details the predictions made by this model about the nature of
verb classes in light of the specifications detailed in Chapter 4. Section 4 provides a
discussion on the nature of specification given the proposal of the ¬ specification
proposed in Chapter 3. In particular, Section 4 discusses the nature of unmarked forms
and the Elsewhere Principle.
142
5.1
Event Semantics
In Chapter 4, I present a model of syntax based upon the proposal by Jelinek (1988). She
proposes that there is a functional head Trans, which (parallel to little-v) is the head that
is responsible for licensing the presence of a theme argument in the syntax. She argues
that this head is often realized cross-linguistically as a transitivizing morpheme. In
adopting her analysis, I essentially rejected the view that the theme argument is the
complement of the verb (or the root). This assumption was crucial to my model of
subcategorization. In order for the root to fuse with it and gain the [Trans] feature, the
Trans head must be present in the syntax. Otherwise, no verb could be specified for
requiring a theme argument.
Had this dissertation come before Jelinek’s proposal, it would predict that model.
In particular, I claim that VIs are specified for both content and formal features. Those
formal features determine the argument structure of the verb. That means that there must
be a formal feature that corresponds to the presence of the theme argument. It follows
then that there must be a functional head associated with the theme argument. Thus,
Jelinek’s model of the syntax is now confirmed by an analysis that predicts the existence
of that head.
Further, I claim that roots themselves have no inherent argument structure.
Rather, the specifications of corresponding Vocabulary Items determine what argument
structures a particular root is compatible with. This effectively reduces ungrammaticality
of sentences such as *The engineer arrived the train to a simple problem: The
Vocabulary does not have a word with which it can realize the features generated by the
143
syntax. There is nothing inherently ungrammatical with combining the root √ARRIVE
with the features [Trans] and [v]; we just don’t have a word to realize that particular
combination of features.
The logical inference from this is that √ARRIVE itself does not have any
requirements that the combinatorial system of syntax needs to satisfy—its only job is to
provide content to what would otherwise be a derivation composed entirely of formal
features. Roots enter the syntax already saturated (or more precisely, there is nothing to
saturate). All the thematic relations of a particular root are structural, rather than
inherent. This is consistent with the observation that, especially in English, roots such as
hit, table, and chair so easily appear as both verbs and nouns. It also predicts the extreme
productivity we find in English where verbs are zero-derived into nouns and vice-versa.
Finally, it predicts the variety of grammatical argument structures and structural coercion.
Parsons (1990) proposes a theory formal semantics which is often today called
Compositional neo-Davidsonian Event Semantics (Bayer 1997). In this model, sentences
are treated as quantification over events. Specifically, Parsons treats every argument of a
verb, even the verb itself, not as a function that needs to be saturated, but rather as an
element that is conjoined to an event. In neo-Davidsonianism, all “words” entering into
the semantics are themselves already saturated and all arguments are introduced by an
independent function on the event.
The model of syntax proposed in Chapter 4 predicts both that verbs enter the
syntax already saturated and that arguments are a function of the combinatorial system
144
rather than the verb themselves. In this way, the model of syntax proposed here mirrors
the Parsonian model of semantics.
5.1.1
Parsons (1990)
Parsons (1990) defended an observation first made by Davidson (1967)—hence
the name, neo-Davidsonianism—that individual verbs can appear with a variable set of
arguments and modifiers.
(5.1)
a.
b.
c.
d.
Brutus stabbed Caesar in the back with a knife.
Brutus stabbed Caesar in the back.
Brutus stabbed Caesar with a knife.
Brutus stabbed Caesar.
For the sake of completeness, I’ll add the following sentences:
(5.1)
e.
f.
g.
h.
i.
Brutus stabbed.
Caesar was stabbed.
Caesar was stabbed with a knife.
Caesar was stabbed in the back.
Caesar was stabbed in the back with a knife.
There are a number of important observations about the sets of sentences in (5.1) that
Parsons (1990) makes.
1. (5.1.a.) entails all the other sentences in the two sets. In particular, the meaning
of any one of the other sentences is present in (5.1.a) along with additional
meaning. Importantly, this entailment is (often) unidirectional. (a) entails (b) and
(c), but neither (b) nor (c) entail (a). (b) and (c) both entail (d) which entails
neither
2. (5.1.a) entails the conjunction of any two of the other sentences (allowing for
overlap). So Brutus stabbed Caesar in the back with a knife entails the
conjunction of Brutus stabbed Caesar in the back and Brutus stabbed Caesar with
a knife.
145
Extending those observations to the discussion in Chapter 4, it is important to add one
more crucial observation.
3. There is nothing about stabbed that requires any of the modifiers or arguments in
any of the sentence. Stabbed appears grammatically with all of them. 27
The modifiers, including the arguments, in (5.1) are called intersective modifiers.
Kadmon and Landman (1993) identify the following the following two important
principles of these types of modifiers: a) they allow permutation, and b) they allow
dropping. The example provided by Bayer (1997) is quite good at showing these
processes. Examine the sentence in (5.2) from Bayer (1997).
(5.2)
John is a forty-year-old, blond, blue-eyed American dressed in a suit, with a
beard, in his midlife crisis.
The modifiers within (5.2) can be reordered (within the restrictions of the syntax) without
changing the truth conditions of the sentence.
(5.3)
John is a blond, blue-eyed forty-year-old, American with a beard, dressed in a
suit, in his midlife crisis.
In addition, any of these modifiers can be dropped without changing the core meaning of
the utterance.
Since these types of intersective modifiers can occur over both verbs (as seen in
5.1) and nouns (5.2), they must be quantifying over some other element. Parson,
following Davidson (1967), concludes that they are (independently) quantifying over
events. This leads Parsons to propose the formal semantic structure of an utterance such
27
Stabbed cannot appear without any arguments, however, a fact attributable to the presence of some
verbal head (and, perhaps, the EPP) in any given derivation.
146
as Brutus stabbed Caesar in the back with a knife as being a conjunction of several
functions on one event.
(5.4)
Brutus stabbed Caesar in the back with a knife
(∃e)[Stabbing (e) & Subj (e, Brutus) & Object (e, Caesar) & In (e, back) & With
(e, knife)] 28
This structure perfectly captures the ability of the modifiers to be dropped to result in
smaller sentences.
(5.5)
Brutus stabbed Caesar in the back with a knife
(∃e)[Stabbing (e) & Subj (e, Brutus) & Object (e, Caesar) & With (e, knife)]
(5.6)
Brutus stabbed Caesar in the back
(∃e)[Stabbing (e) & Subj (e, Brutus) & Object (e, Caesar) & In (e, back)]
Parson’s (1990) proposal is crucially different from classical semantic theory in two
ways: (a) there is only one operation: conjunction and (b) none of those operations are
on stab. Rather stab is a function on the event and is not a function of Brutus on Caesar.
Crucially, stab is indifferent to the presence of the arguments and other modifiers,
contrary to classical formal semantics, where stab is a function whose variables (object
and subject) need to be realized.
Thus the crucial features of neo-Davidsonian event semantics are the following:
(a) individual words are not functions over their arguments, rather they are functions over
an event; (b) the arguments of a verb are not functions on the verb, nor do they saturate
28
This notation should be read as follows: there exists an event; that event is an event of stabbing, the
subject of that event is Brutus, the object of that event is Caesar, that event was in such a manner that it was
in the back, and that event used the instrument a knife.
147
the variables of the verb; they are functions on the event; and (c) there is only one
function on the event: conjunction.
The combination of all the features of Parsonian event semantics has some
interesting effects on the model of syntax. Assuming that the syntax’s “job” is to
introduce all the formal features necessary for LF to be interpretable, a model of the
syntax that will be compatible with Parsonian event semantics will have to have several
features.
Minimalist syntax shows a sentence to be built in some way or another around the
verb (i.e. the verb is a member of the initial application of Merge to begin a derivation).
In a model compatible with Parsons, the derivation would begin with the merger of the
event with some other element, not the merger of the verb and its complement. Some
functional head must introduce this event so that it can be interpreted at LF (and so all the
syntax can be build on it).
The event is the element of the syntax that is consistently operated upon
(functional heads would now serve the job of conjoining modifiers (including arguments)
to the event head). Content items themselves must not be providing any syntactic
information. Rather, since each modification is something like Subj (e, Brutus), there
must be a functional head for each of those that conjoins the modifier to the event.
The next section will elaborate on the model of syntax that is compatible with
Parsons’ proposal.
148
5.1.2
Neo-Davidsonianism Meets Syntax
In the last decade, there have been a number of proposals that have mirrored neoDavidsonian event semantics in the syntax. Some of note are Borer (1994, 2004),
Marantz (1997), Ramchand (1997, 2002), Ritter and Rosen (1998), and Travis (1994).
Ramchand (1997) is one of the more important of the proposals in that it proposes that
Aspect is the locus for the introduction of the event element in the syntax. Ramchand
argues that, in many cases, verbs themselves are complex—being composed of a content
morpheme and a functional head. She argues that the Asp head introduces the event and
also introduces the object.
Ramchand (2002) elaborates that schema to show that three functional heads
compose a verb and its argument structure: one introducing the causer, one introducing
the “process” meaning, and one introducing the “result”. Borer (1994, 2004) also
proposes a similar breakdown where all semantic relations with the event are introduced
with a corresponding functional head, though hers are different from Ramchand’s.
Especially in regard to arguments and theta roles, the Parsonian syntactic analyses all
tend to have one thing in common: most, if not all, arguments are introduced by a
functional head rather than by the verb.
Ramchand (1997) is one of the exceptions to this rule. For Ramchand, unlike
Borer’s (1994, 2004) and Marantz’s (1997) proposal, the verb itself is still responsible for
some arguments—in particular, those that get weak case (i.e. genitive). For Ramchand,
the arguments projected by the root (she calls it the verbal noun) are different from
arguments projected by functional heads in two ways: a) in the semantics, they are
149
interpreted as modifying the verb, not the event, and b) their interpretation is imperfective
(rather than perfect, which Asp introduces).
Ultimately, models of the argument structure that are based on Parsonian event
semantics are split into two camps. The difference between the two camps is whether the
verb itself licenses an argument. Models such as Borer’s and Marantz’s show all
argument structure to be syntactic while the models that follow Ramchand assume some
arguments licensed by the verb itself. For Ramchand, the difference is motivated by the
overt presence of an Aspect head, which triggers different argument structure and case.
In order to account for Ramchand’s data, the model I have espoused here (Borer’s and
Marantz’s) predicts an unpronounced head that licenses what Ramchand calls “weak
case”.
The syntax that I propose in Chapter 4 contributes to this debate. As I mention
above, the MINIMIZE EXPONENCE-driven fusion analysis of morphology and
subcategorization, which strives to capture the same data the analyses I have discussed
here do, predicts (and requires) a functional head for each argument in order for root to
gain the relevant interpretable features. Further, there is an account for the “missing
head” that Ramchand observes. An analysis of Ramchand’s data that is compatible with
my proposal would show the imperfective head to have fused with the root while the
perfective head failed to fuse, causing overt spellout. In this way, the predictions I have
made here act as a kind of tiebreaker between Borer and Ramchand in that for other
reasons (specifically, in order to license the insertion of the Vocabulary Item), I predict
that all arguments are licensed by functional heads.
150
Ramchand’s model of event semantics is clearly a hybrid of Parsonian semantics
and more classical semantics (which is why she calls it Post-Davidsonian rather than neoDavidsonian) in that not all modifiers modify the event—some modify the verb. Borer’s
model is still consistent with the observation that argument structure is independent of the
root (which was one of the motivations behind the proposals made by both Davidson
1967 and Parsons 1990).
5.1.3
Remaining Questions:
In Section 1.2 above, I describe some of the ways that the predictions of the neoDavidsonian model of semantics are transferred to the realm of syntax. This transference
serves both to show a mirror between the syntax and the semantics and provide a means
for the semantics and the syntax to interface. To be more specific, if the logical form of a
sentence is such as we saw in (5.4, repeated here as 5.7), we want the syntax to send to
LF something that looks like that.
(5.7)
Brutus stabbed Caesar in the back with a knife
(∃e)[Stabbing (e) & Subj (e, Brutus) & Object (e, Caesar) & In (e, back) & With
(e, knife)
Section 1.2 discusses a few of the things that would have to be true of the syntax in order
for it to produce such a logical form. The first is that the syntax would have to have
functional projections for each modifier whose job was to conjoin the argument and the
event in a meaningful way. We see that realized in the proposal here and those by Borer,
Ramchand, and others. The next is that the individual roots would have to be completely
151
saturated (i.e. they themselves are not functions with empty slots, but the targets of the
conjunction introduced by the functional heads). In the proposal in Chapter 4, we see
that to be predicted as well.
However, a number of issues impede a total mirroring of the logical form
predicted by Parsonian event semantics and the LF that would result from even the syntax
predicted here. The first is the event variable. According to the LF predicted by Parsons,
each modifier is a conjunction of the argument with the event variable. Ramchand
proposes that the event variable is introduced by the Aspect head, but that accounts for
only one of the event variables. The simplest solution is to argue, expanding upon
Ramchand, that in fact, each functional head introduces the event variable that it conjoins
to its specifier. This immediately accounts for the fact that for each argument, there is a
corresponding event variable.
This however leads inexorably to the next problem with a matching of syntax to
the LF predicted by Parsonian semantics: hierarchical structure versus flat structure.
Parson’s model has a flat structure. Each argument function such as Subj (e, Brutus) is
just conjoined one after the next in a series of conjunctions. This property is one of the
key reasons that this type of LF so easily allows for permutation and dropping. There is
no hierarchical structure at all (except in the case of complex phrases, which have their
own corresponding event variable).
One tenet of the Minimalist Program is that syntax has complex hierarchical
structure. In particular, Merge is construed as recursive set formation (Chomsky 1995).
The application of Merge (except at the very bottom of the derivation) takes one or more
152
extant sets and constructs a set that contains both sets. This results in complex
hierarchical structure. That structure is completely absent from the Parsonian model of
event semantics – it is just a series of conjunctions of an event and an individual.
Therefore, while Parsons’ mechanisms create sets, they are often sets of individuals.
Thus, Parsons’ mechanisms don’t show the embedding assumed in Minimalist syntax.
This is complicated by the fact that hierarchical sentence structure is crucially ordered.
Parsons’ mechanism, as it employs only conjunction, results in an unordered LF.
There are two possible ways to account for this apparent incompatibility. The
first is to stipulate that the functional heads in the core syntax have selectional
requirements of their complements. While the core job of the functional head is a
function between its specifier and the event, it also selects for where it can appear in the
derivation. Specifically, it selects what it can be merged with to get it into the derivation.
This stipulation would mean that the ordering (and some of the hierarchical structure) is a
property only of the core syntax and that it has no effect on the LF. Its function would be
linearization. The selection of the heads would permit the pronunciation of the
derivation, since we speak one dimensionally (excluding time) and an LF representation
is two-dimensional. Another, solution to the problem is to show that Merge is itself not
actually a function that makes a set out of other sets, but is instead an operation that only
targets individuals. 29
Ultimately, while the purpose of this section is only to show that the remaining
apparent incompatibilities between syntax and Parsons’ model are not insoluble, the
29
See forthcoming work done by David Medieros.
153
argument that syntax is hierarchical while LF is not is more satisfactory. The basics of
the recursivity mandated by the Minimalist model of syntax do not entail that LF need be
recursive. Rather, just as there as a linearization process at PF, there could be a similar
flattening at LF for interpretation or the hierarchical nature of syntax could only serve the
purpose of linearization.
5.2
Typological Predictions
MINIMIZE EXPONENCE as proposed here is intended to be a PF constraint on UG, not an
English-specific constraint. Since the majority of this dissertation revolves around
English data, the purpose of this section is to elaborate on some of the typological
predictions of this model.
I argue the fusion that is driven by MINIMIZE EXPONENCE is mandatory. In
order to realize the largest amount of formal features with the fewest morphemes, the
solution proposed here is to fuse functional material whenever possible (fusion being
only applicable to complex head structure, thus it will never reduce a sentence to just one
word, nor will it ever cause two roots to fuse together 30 ). This is not the only way to
satisfy MINIMIZE EXPONENCE. If a language had access to a process to discharge features
from a derivation other than insertion (one possible explanation of pro-drop, I suppose)
that would also satisfy MINIMIZE EXPONENCE.
30
The only possible time two roots could fuse together is if we hypothesized root-root compounding to be
the result of head movement or some other process that creates a complex head rather than a syntactic
phrase. In my description of compounding in chapter 3, I assumed that compounding results in a syntactic
phrase, not a complex head, so in the model I assume the issue is moot. In a syntactic theory where
compounds are complex roots, the constraint on insertion mentioned earlier that roots must be realized by a
VI, would disallow fusion of two roots.
154
Furthermore, whether or not a node fuses in the system sketched above is entirely
dependent upon the specification of the VI 31 that will be inserted into that node. For
example, since walk is specified a ¬ [past], then the fusion of the root √WALK with the
[past] feature is effectively blocked. Since walk is indifferent to [3sg], the fusion is not
blocked. The amount of fusion in a given language is entirely dependant on Vocabulary
inventory of that language. In that way, many of the morphological and syntactic
differences of a language can arise entirely from the specifications of that language’s VIs.
For example, in English, nominative and accusative cases always fuse with the root
(which is why the only overt realization of these cases is in suppletive forms like
pronouns and whom). This means that no words in English contain ¬ specifications for
nominative or accusative. Conversely, in a language like Latin, where nominative and
accusative are regularly spelled out with an affix (5.8), the “default” for a VI would be to
have a ¬ specification for nominative and accusative.
(5.8)
Latin nominal inflection
stem
puella
deumanure-
nominative
‘thing’
‘god’
‘hand’
‘thing’
-ø
-s
-s
-s
accusative
-m
-m
-m
-m
On a larger scale, this difference in specifications could give rise to the difference
between fusional and agglutinative/isolating languages. Agglutinative/isolating
languages such as English or Mandarin have an inventory of VIs with a proportionally
31
In a roundabout way... what is actually happening is that there is a relationship between the amount of
fusion in the derivation selected to converge and the VIs inserted into that derivation.
155
high number of ¬ specifications. These ¬ specifications impede fusion more often,
leading to more features being realized as independent morphemes (either bound in the
case of English or free in the case of Mandarin). On the other hand, the inventory of
more fusional languages (such as Spanish) contain proportionally fewer ¬ specifications
(and correspondingly, VIs contain more features that they realize) leading to morphemes
which realize comparatively more features. For example, in Spanish the verbal affix –o,
in hablo ‘I speak’, realizes a large set of features—1st person, indicative mood, singular,
present tense, and perfect aspect.
Similarly, the amount of syncretism in a given language is also function of that
language’s ¬ specifications. A language whose VIs contain a large number of ¬
specifications will undergo less fusion, requiring an greater inventory of functional
morphemes while a language with fewer ¬ specifications will undergo more fusion
allowing it to employ fewer functional morphemes to realize the same amount of
interpretable features.
These differences among languages is not parametric but is rather an effect of the
specifications in their VIs. So-called fusional languages are languages where the
inventory of the Vocabulary contains a large amount of VIs that are highly specified for
features they realize. Conversely, a non-fusional language contains a large amount of
VIs that are specified for what they must not realize, forcing those features to be
otherwise discharged. The difference between an agglutinating and an isolating language
is a function of how much the functional morphemes in the language’s inventory are
free—an agglutinating language contains more bound functional morphemes.
It follows
156
then that polysynthetic languages may not be a matter of parametric variation (Baker
1988) but rather just a matter of VI content.
5.3
Verb Classes
So much work has been done on verb classes throughout the field of linguistics (see
Levin 1993 for a detailed approach) that they are often assumed, especially in the realms
of inflectional morphology and subcategorization, realms that this dissertation addresses
in detail. The basic hypothesis is that verbs (and nouns for that matter) are organized in
the Lexicon and that the morphology and syntax are sensitive to that organization. For
example, the general assumption in the study of classical Latin is that its verbs belong to
five classes (or conjugations). Membership to these classes determines the inflectional
paradigm that the verb employs. Similarly, for argument structure, the general argument
(Levin 1993) is that verbs belong to certain classes and that those classes determine that
verb’s argument structure.
The first concern about the idea of verb classes comes from the very structure of
the Distributed Morphology grammar. Since there are no paradigms in DM, class
membership would have to be a realization of some interpretable feature in the syntax.
Returning to Latin conjugations, according to DM, if there is an affix that is somehow
sensitive to the class membership (e.g. –amus, -imus, -emus for first person plural present
tense—1st conj., 2nd conj, and 3rd conj. respectively), that must mean that there is
something in the syntax that contributes that meaning. That syntactic feature would be
analogous to gender. That feature would be realized by its own affix—in the Latin case –
157
a-, -i-, or –e-, since those sounds appear in almost every member of their respective
paradigms. 32
Since in DM, paradigmatic morphology is reduced to a set of Vocabulary Items
that happen to have shared features in their specifications—which explains cross
paradigmatic syncretisms (-mus in the example above)—the existence of verb classes as
targets of paradigmatic morphology is disallowed by the mechanics of the framework. In
Chapter 4, I propose that the selectional behavior of a verb is also just a matter of its
feature specification. Thus the fact that throw and fax (example from Bresnan and
Nikitina 2003) have the same selectional behavior is not due to their belonging to any
particular verb class, but rather is attributable to their having identical (or near identical)
specifications for abstract (non-root) formal features.
In order to account for the sum of verbal behavior in English, Levin had to
propose a large number of verb classes, many of which have a very small membership.
This large number of verb classes is a prediction of the analysis proposed in Chapter 4.
The apparent large number of verb classes is attributable to the fact that VIs can be
specified for any combination of the relevant features. Since there are several of the
argument features proposed in Chapter 4, and a VI can select for not only any set of those
features but also for the corresponding ¬ specification, there is a large but finite set of
possible specifications. The magnitude of the set of possible specifications gives rise to
the large number of verb classes proposed by Levin. The finiteness of that set is what
32
Alternatively , the VI for the verb could be in someway specified for class membership. Then that class
membership could be a secondary exponent of the affix. Then however it is curious as to what feature is
being realized by –a-, -i-, and –e-.
158
gives rise to the appearance of verb classes in general. Thus, verb classes are an artifact
of the finite set of functional heads that project arguments and the finite set of possible
feature specifications. Verb classes reduce to the sets of VIs that have identical feature
specification.
This argument helps to account for why speakers can disagree on the selectional
behavior of a verb (e.g. whether I pondered whether the world is round is grammatical).
In the model developed here, the difference across speakers is attributable only to the
presence or absence of a feature specification for that verb. In a model with verb classes,
that difference would have to be attributable to the verb being a member of a different
class. Since class membership is in part driven by semantics, it is not entirely clear how
speakers who agree on the meaning of a word could differ on the class to which it
belongs.
5.4
On Feature Specification and the Elsewhere Principle
There is a curious result of the model predicted here as to the structure of the
Vocabulary (or lexicon). Recall from Chapter 3 that regular morphology is triggered by
the ¬ specification. So the past tense of thrash is thrashed because thrash is specified as
¬ [past] (3.43, repeated here as 5.8)
(5.8)
Vocabulary Entry for thrash
√THRASH
[v]
¬ [past]
→
thrash
/TrQS/
159
It follows then that all words which are regularly inflected and any stem that requires
derivational morphology (i.e. does not zero derive) must be contain a ¬ specification for
each feature realized by an affix it is compatible with. This means that unmarked forms
such as thrash, walk, black, arrive, etc which participate in the all normal regular
morphological processes of a language, while they are relatively unspecified for the
formal features they must realize (cf. ran, ate, mice etc which are relatively more
specified) are heavily specified for the features the must not realize. I.e. while they have
a small amount of traditional feature specifications, they have a large amount of ¬
specifications. Returning to thrash: thrash is relatively unspecified in traditional terms
(realizing only a root and the feature [v]). However, since it gets regular inflection for
past tense, progressive aspect, third person singular present, etc., it is heavily specified
with ¬ specifications.
One of the fundamentals of Distributed Morphology is the Elsewhere Principle.
The Elsewhere Principle mandates that the VI with the smallest feature specification will
appear in the most environments (i.e. is the “default”). However, in this model the
“default” forms in a language are likely to be heavily specified with ¬ specifications. In
terms of formal morphological theory, this doesn’t have any drastic effect on the model
of the grammar. Rather, it is just counterintuitive that the most underspecified forms tend
to be highly specified on another axis.
Where this might have some effect on the model of the grammar is in learnability
theory. A child overgeneralizes regular morphology (e.g. *goed). In the model proposed
here, that indicates that when a learner of the language hears a word, they assume that it
160
is heavily specified with ¬ specifications in addition to its traditional specifications. For
example, a learner might assume the following VIs for go and went, giving rise to *goed.
(5.9.a)
Speculative Vocabulary Item for go
√GO
[v]
¬ [3sg]
¬ [past]
(5.9.b)
go
/gow/
Speculative Vocabulary Item for went
√WEN
[v]
[past]
5.5
→
→
went
/wEnt/
Summary
The purpose of this chapter was to provide a quick list of future avenues of research in
light of the MINIMIZE EXPONENCE-driven fusion that I proposed in Chapter 3. One of
those avenues I discussed here was the interface of the syntax outlined in Chapters 3 and
4 with the neo-Davidsonian model of event semantics. As I discussed here, it seems that
the next obvious step in that direction is to outline the reasons that syntax is hierarchical
while neo-Davidsonian event semantics is not.
I also described some of the typological predictions of this model. Obviously, the
next step down that avenue of research is to describe the behavior of irregular forms and
inflectional patterns in other languages, such as Spanish, Latin, or Mandarin as surveyed
above, in light of the fusion predicted here. A further step is to describe the argument
structures of other languages as a function of fusion. One interesting avenue to pursue in
particular are languages that have overt heads that introduce arguments such as the Uto-
161
Aztecan language family (Jelinek 1988, Haugen 2004) or the Salishan language family
(Bischoff 2004). In that case, the argument would be that the root has failed to fuse with
Trans (since Trans is realized by an overt morpheme), so the prediction from this model
is that verbs could not select for themes.
Another avenue of research stemming from the discussion in this chapter would
be to compile a list of the specifications of verb VIs. From the predictions here, the sets
of VIs with identical specifications should somewhat align to the verb classes predicted
by Levin (1993). Finally, the final obvious avenue of research is examining the
ramifications the ¬ specification has on the Elsewhere Principle and analyses that have
relied upon it. The corollary to this is to see if there is a way to test the abstract claims of
this model against child acquisition data to see if the realization of the Elsewhere
Principle is realized as the positing of ¬ specifications in language acquisition.
162
CHAPTER 6: CONCLUSIONS
The purpose of this dissertation is to address conflict in the grammar whereby on one
hand, we have the need to convey a message and the need for that message to be as clear
as possible but on the other, we need our message to be as efficient as possible. While
this conflict is prevalent in much of the linguistic literature, in the realm of
morphosyntax, this conflict is largely ignored by theoreticians. In syntax, one such
realization of this conflict is the following: a maximally contrastive grammar involves
pronouncing every single formal feature in a maximally contrastive manner whereas a
maximally efficient grammar sums all the features of an utterance into fewest words.
The Minimalist tradition is largely a study of what the system is and isn’t capable
of, not how that system is put to use. However, this results in a certain loss of predictive
power. I proposed an economy constraint that captures these competing forces on the
grammar—in particular the balance necessary in pronouncing all the interpretable
features of a given derivation in the most efficient way possible (1.1, repeated here as
6.1).
(6.1)
MINIMIZE EXPONENCE
The most economical derivation will be the one that maximally realizes all the
formal features of the derivation with the fewest morphemes.
The gist of this constraint is that the best utterance is the one that conveys the most
amount of information with the least effort (measured in number of morphemes that have
to be pronounced). In terms of the production of an utterance, this constraint captures the
163
struggle between the need to be maximally contrastive and the need to be maximally
efficient.
This dissertation focused on exploring the role that this type of constraint would
have on the model of the grammar by specifically looking at the effects of MINIMIZE
EXPONENCE on analyses of familiar morphosyntactic phenomena.
To this end, I presented an alternate analysis to lexical categories and root
allomorphy within the framework of Distributed Morphology. The traditional analysis of
lexical categories in DM is to have the content VI licensed for insertion through
secondary licensing (targeting the immediately c-commanding functional head). The
traditional DM analysis of root allomorphy employs the use of readjustment rules (again
conditioned by c-commanding functional heads) to change the phonology of on VI to
another VI. In the analysis presented in Chapter3, I argued that licensing is a local
process of targeting features actually in the node targeted for insertion. I argued that the
root and the formal features come to occupy the same terminal node through the
application of the process of fusion. I extended this analysis to an analysis of root
allomorphy. This extension allows words like eat and ate to be separate VIs that compete
with each other for insertion rather than one being the result of the application of a
readjustment rules.
This proposal has a number of effects on the DM model of grammar. The first is
that it limits application of secondary licensing, removing lexical categorization from the
list of responsibilities of secondary exponence. Since secondary licensing is inherently
less local and thus less efficient than primary licensing, this is taken to be a strength of
164
this proposal. The other important effect of this proposal was to remove the need for
readjustment rules in the DM grammar. Since VIs linked to the same root now compete
with each other for insertion, readjustment rules are no longer need to alter the phonology
of irregular forms. As readjustment rules only serve the purpose of changing the
phonology of irregular morphemes, since they aren’t needed for that function any longer,
the DM grammar doesn’t need them at all. Since readjustment rules entail both long
distance relationships and extra computational load, another strength of this dissertation
is that it proposes a model of DM without readjustment rules.
In the current model of DM, functional morphemes and content morphemes in
many ways participate in two different grammars. Whereas functional VIs participate in
competition, content VIs did not, relying on readjustment rules and secondary exponence
(never primary exponence) to license their insertion. The proposal here, in abolishing the
need for both secondary licensing and readjustment rules and in elaborating on a theory
of competition of roots, shows that it is possible for the model of the grammar to use just
one process for insertion of both functional and content VIs. As this simplifies the model
of the grammar, this is again taken as a strength.
The realization of MINIMIZE EXPONENCE here is the ubiquitous fusion of
functional heads in complex head arrangements. The effect of this fusion is that the
features of those heads are realized without each needing to be realized by its own VI.
The efficiency in the grammar comes from there being fewer VI items to be pronounced
without loss of contrast (any ambiguity is then only a result of a given language’s
inventory). Thus, fusion is one possible tactic available to the grammar to satisfy this
165
central conflict. As an added effect, a large portion functional heads that are realized as
null morphemes in DM instead fuse with other heads to be realized by overt morphemes.
This drastically reduces the number of null heads predicted by DM.
I proposed an analysis for the blocking of regular inflection in nominal
compounds in English (e.g. *rats-infested). For DM, this phenomenon has been difficult
to analyze due to the tenets of the framework. There were two central problems for DM:
1) in DM the structures lice-infested and *rats-infested are identical with the exception of
the application of a readjustment rule; and 2) DM recognizes no difference between
inflectional and derivational morphology (and derivational morphology is allowed in
such structures). The analysis shows here that the structures of *rats-infested and miceinfested are not identical: one involves more fusion driven by MINIMIZE EXPONENCE.
The grammaticality of the constructions reduces to the embeddedness of the feature [n]
(which meant that derivational morphology is allowed as long as a nominalizer was the
dominant morpheme in the non-head member).
In Chapter 4, I extended the predictions of the ubiquitous fusion motivated by
MINIMIZE EXPONENCE beyond the realm of morphological phenomena to the realm of
syntactic phenomena. I showed that a fusion analysis explains subcategorization effects
in the grammar. I argue here that the selectional behavior of a given verb is an effect of
the functional features that the VI is specified for. Arguments are selected for by
functional heads. Those functional heads fuse with the root, giving the target node both
the root and the relevant formal features. This allows the root VI to select for those
features. Since DM is an underspecification model, this led to the prediction that roots
166
should be compatible with arguments in excess of their core requirements, which can be
seen in normal verbal behavior and extends to structural coercion. The end result is to
show that the fusion motivated by MINIMIZE EXPONENCE also drives the selectional
behavior of verbs.
In Chapter 3, I proposed a system of ¬ specifications that allows VIs to block
their insertion into nodes containing features they are incompatible with. This set up a
system that accomplished several things: first and foremost, it provides DM with a
feature blocking mechanism. Second, it triggers regular morphology. I also extended the
feature blocking system to account for the blocking of a verb’s arguments (*The engineer
arrived the train), since in the proposal here, a verb’s argument structure is an artifact of
the specification of its VI.
In Chapter 5, I provided a quick list of future avenues of research in light of the
MINIMIZE EXPONENCE-driven fusion that I propose. One of those avenues I discussed
here was the interface of the syntax outlined in Chapters 3 and 4 with the neoDavidsonian model of event semantics. I also described some of the typological
predictions of this model, discussed a possible theory of verb classes within DM, and
provided an overview of the effects of my proposal on the Elsewhere Principle.
6.1
Some Final Thoughts
MINIMIZE EXPONENCE is PF constraint on the grammar which essentially instructs
the grammar to produce utterances that perfectly balance the need to be contrastive with
the need to be energy efficient. I wonder, then, what other kinds of PF constraints are on
167
the grammar that affect the morphosyntax. For example, linearization might be the effect
of a constraint that instructs the grammar to produce utterances that are pronounceable,
given the constraints on using language.
One such area that I think these types of PF constraints might extend into is
homophony-avoidance. Since homophony-avoidance in the realm of morphosyntax is
gradient (in that we try to try to create maximally contrastive utterances but still end up
with structural ambiguity) it follows that this is again a balance struck by the grammar
(please see a survey of morphosyntactic homophony avoidance in Haugen and Siddiqi
2006) . This balance might arise from a violable constraint that prohibits syntactically
derived ambiguity (Flack 2006) or morphologically derived ambiguity (Urbanczyk 1999,
Siddiqi 2004), given a model of the morphosyntax driven by constraint hierarchies (such
as OT-LFG, Bresnan 1996, 2000, Johnson 1998). This balance could also arise from a
PF economy constraint similar to MINIMIZE EXPONENCE that requires the grammar to
produce utterances with the minimal amount of morphosyntactic ambiguity.
I think that MINIMIZE EXPONENCE is likely one of a family of economy constraints
on the PF branch of a derivation whose purposes could range from maximizing efficiency
of the utterance to maximizing its contrastiveness and even to maximizing its
interpretability. I believe this realm of morphosyntax (constraints and operations that act
upon a derivation between spellout and phonological processes) is an exciting avenue of
future research and I believe that the framework of DM is a perfect framework to pursue
those questions.
168
REFERENCES
Adger, D. 1994 The Licensing of Quasi-Arguments. Proceedings of CONSOLE 1 ed. by P.
Ackema & M. Schoorlemmer, 1-18. The Hague: Holland Academic Graphics.
Aronoff, M. 1994. Morphology by itself. MIT Press: Cambridge.
Baker, M. 1988. Incorporation: a Theory of Grammatical Function Changing. University of
Chicago Press: Chicago.
Bayer, J. 1997. CP-Extraposition as Argument Shift. Rightward Movement, ed. by D.
Beerman, D. LeBlanc, and H. C. van Riemsdijk, 37–58. John Benjamins: Amsterdam.
Bischoff, S. 2006. Basic Clause Structure in Coeur d’Alene: A Preliminary Working Paper.
MIT Working Papers in Linguistics. Working Papers on Endangered and Less
Familiar Languages 7, Studies in Salishan ed. by S. Bischoff, L. Butler, P. Norquist,
and D. Siddiqi, 159-190 Cambridge: MIT Working Papers in Linguistics.
Bobaljik, J. 1994. What does adjacency do? MIT Working Papers in Linguistics 22: The
Morphology-Syntax Connection, ed by H. Harley and C. Phillips, 1-32.
Bobaljik, J. 2002. A-chains at the PF-interface: Copies and ‘Covert’ Movement. Natural
Language and Linguistic Theory 20.2:197-267.
Borer, H. 1994 The Projection of Arguments. Functional Projections, University of
Massachusetts Occasional Papers 17, ed. by E. Benedicto and J. Runner. Amherst
Massachusetts: GLSA, University of Massachusetts.
Borer, H. 2004 Structuring Sense (two book manuscript), Oxford University Press
Bresnan, J. 1996. Morphology Competes with Syntax: Explaining Typological Variation in
Weak Crossover Effects Is the Best GoodEnough?, edited by Barbosa et al. MIT Press.
Bresnan, J. 2000. Pidgin Genesis and Optimality Theory. Processes of Language Contact:
Case Studies from Australia and the Pacific, ed. by J. Siegel. Montreal: Les Editions
Fides, 145--173.
Bresnan, J. 2001. Lexical-Functional Syntax. Oxford: Blackwell
Bresnan, J. and T. Nikitina. 2003. On the Gradience of the Dative Alternation. [ms.]
Chomsky, N. 1957. Syntactic Structures. The Hague: Mouton.
Chomsky, N. 1965. Aspects of the Theory of Syntax. Cambridge: MIT Press.
Chomsky, N. 1970. Remarks on Nominalization. Reprinted in Davidson and Harman. pp. 262
Chomsky, N. 1981. Lectures on Government and Binding. Foris, Dordrecht.
Chomsky, N. 1989. Some Notes on Economy of Derivation and Representation. MIT
Working Papers in Linguistics 10 ed. by I. Laka and A. Mahajan. Cambridge: MIT
Working Papers in Linguistics
Chomsky, N. 1994 Bare Phrase Structure. Occasional Paper #5, Cambridge: MIT Working
Papers in Linguistics.
Chomsky, N. 1995. The Minimalist Program. Cambridge: MIT Press.
Chomsky, N, and M. Halle. 1968.The Sound Pattern of English. New York: Harper and Row.
Davidson, D. 1967. Truth and Meaning. Synthese 17. 304-323
Embick, D. 2003. Locality, Listedness, And Morphological Identity Studia Linguistica,
169
Volume 57, Number 3 (December 01, 2003), pp. 143-169
Embick, D. and A. Marantz. 2006. Architecture and blocking. University of Pennsylvaniaand
MIT. [ms.]
Embick, D. and R. Noyer. 2001. Movement Operations after Syntax. Linguistic Inquiry 32:4,
555–595.
Embick, D. and M. Halle. 2005. On the status of stems in morphological theory. Proceedings
of Going Romance 2003 ed. by T. Geerts and H. Jacobs. John Benjamins,
Amsterdam/Philadelphia, 59–88.
Falk, Y. 2001. Lexical-Functional Grammar: An Introduction to Parallel Constraint-Based
Syntax. Standford: CSLI Publications
Flack, K. 2006. OT contrast preservation and disambiguation in Japanese scrambling. Paper
presented at the Annual Meeting of the Linguistic Society of America. Albuquerque,
NM: January 6, 2006.
Fodor, J. 1998. Concepts. Oxford: Clarendon
Gleitman, L. 1990. The Structural Sources of Verb Meaning. Language Acquisition 1. 3-55.
Gleitman, L., H. Gleitman, C. Miller, and R. Ostrin. 1996. “‘Similar’ and Similar Concepts”.
Cognition 55. 321-376.
Goldberg, A.. 1995. A Construction Grammar Approach to Argument Structure. Chicago:
University of Chicago Press.
Grimshaw, J. 1979. Complement Selection and the Lexicon. Linguistic Inquiry Cambridge,
MA 1979; 10: 279-326
Grimshaw, J. 1990. Argument Structure MIT Press Cambridge MA
Gropen, J. 1989. The Learnability and Acquisition of the Dative Alternation in English
Language: Journal of the Linguistic Society of America, June; 65 (2): 203-57.
Hale, K., and S. Keyser. 1993. On Argument Structure and the Lexical Representation of
Syntactic Relations. The View from Building 20 ed. by Samuel J. Keyser & Kenneth
Hale. Cambridge: MIT Press. 53-109.
Halle, M. 1997. Distributed Morphology: Impoverishment and Fission. MIT Working Papers
in Linguistics: PF: Papers at the Interface. B. Bruening, Y. Kang, M. McGinnis, 42549. Cambridge: Department of Linguistics, MIT;
Halle, M. and A. Marantz. 1993. Distributed Morphology and the Pieces of Inflection. The
View from Building 20 ed. by S. Keyser and K. Hale,111-176.. Cambridge: MIT
Press.
Halle, M. and A. Marantz. 1994. Some Key Features of Distributed Morphology. MIT
Working Papers in Linguistics 21: Papers on Phonology and Morphology. Ed. by A.
Carnie, H. Harley, and T. Bures, 275-288. Cambridge: MIT Working Papers in
Linguistics.
Hammond, M. 1984. Level Ordering, Inflection and the Righthand Head Rule. MIT Working
Papers in Linguistics vol. 7
Harley, H. 1995. Subjects, Events and Licensing. PhD dissertation, MIT.
Harley, H. and R. Noyer. 1999. State-of-the-Article: Distributed Morphology. Glot
International 4.4, 3-9
Harley, H. and R. Noyer. 2000. Formal versus Encyclopedic Properties of Vocabulary:
170
Evidence from Nominalisations. The Lexicon-Encyclopedia Interface, edited by Bert
Peeters, Elsevier Press.
Haugen, J. 2004. Issues in Comparative Uto-Aztecan Morphosyntax. Ph.D. dissertation, U.
Arizona
Haugen J. and D. Siddiqi. 2006. From Semantic Contrast to Phonological Contrast:
Homophony Avoidance in Morphology. Paper presented at the Annual Meeting of the
Linguistic Society of America. Albuquerque, NM: January 7, 2006.
Hayes, B. 1980. Aklan stress: disjunctive ordering or metrical feet? Cahiers linguistiques
d’Ottawa 9,in Papers from the Tenth Meeting of the North East Linguistic Society,
University of Ottawa.
Hornstein, N. 1999 Movement and Control. Linguistic Inquiry, 69-96.
Jackendoff, R. 1996 “Semantics and Cognition” in S. Lappin (ed.), Handbook of Semantics,
Oxford: Basil Blackwell Inc.
Jelinek, E. 1998 “Voice and Transitivity as Functional Projections. The Projection of
Arguments: Lexical and Compositional Factors in Yaqui ed. by M. Butt and W.
Geuder. Stanford, CA: Center for the Study of Language and Information viii
Jesperson, O. 1909-1949. A Modern English Grammar on Historical Principles. London:
George, Allen and Unwin.
Johnson, M. 1998. Optimality-theoretic Lexical Functional Grammar. Presented to the 11th
Annual CUNY Conference on Human Sentence Processing, Special Session on the
Lexical Basis of Syntactic Processing: Formal and Computational Issues, Rutgers
University, March 20, 1998.
Kadmon, N. and F. Landman 1993. Any. Linguistics and Philosophy, 16: 353-422.
Kiparsky, P. 1982. Lexical phonology and morphology. In Linguistics in the Morning Calm,
ed. I. S. Yang, pp. 3-91. Seoul: Hanshin.
Kiparsky, P. 1997. Remarks on denominal verbs. Complex Predicates ed. by A. Alsina, J.
Bresnan and P. Sells, 473-499. Palo Alto, CA: CSLI Publications.
Kiparsky, P. 2003. Syllables and moras in Arabic. The Syllable in Optimality Theory, ed. by
C. Féry and R. van de Vijver, 147-82. Cambridge: Cambridge University Press.
Kratzer, A. 2002. Telecity and the Meaning of Objective Case The Syntax of Time ed. by J.
Guéron, and J. Lecarme, 389-423. Cambridge, MA: MIT Press
Kratzer, A. 1996. Severing the External Argument from its Verb. Phrase Structure and the
Lexicon ed. by J. Rooryck and L. Zaring, 109-37. Dordrecht: Kluwer Acad.
Lakoff, G. 1977. Linguistic gestalts. CLS 13:236-87.
Langacker, R. W. (1987) Foundations of cognitive grammar: Theoretical Prerequisites.
Stanford, CA: Stanford University Press.
Larson, R. 1995. Semantics. Invitation to Cognitive Science Vol. 1 Language ed. by L.
Gleitman. Cambridge: MIT Press
Levin, B. 1993.English verb classes and alternations: A preliminary investigation. Chicago:
University of Chicago Press
Lidz, J. 1998. Constraints on the Syntactic Bootstrapping Procedure for Verb Learning.
Proceedings of the 22nd Annual Boston University Conference on Language
Development, I-II ed. by A. Greenhill, M. Hughes, H. Littlefield, and H. Walsh, 488-
171
98. Somerville, MA: Cascadilla
Lidz, J. & Gleitman, L.R. 2004. Yes, we still need universal grammar; reply to Goldberg, A.
(2004). Cognition.
Marantz, A. 1984. On the Nature of Grammatical Relations. Cambridge: MIT Press
Marantz, A. 1988. Clitics, morphological merger, and the mapping to phonological structure.
In Theoretical morphology, ed. by M. Hammond and M. Noonan, 253–270. New
York: Academic Press.
Marantz, A. 1997. 'Stem suppletion, or the arbitrariness of the sign.' Talk given at the
Universite' de Paris VIII
Marantz, A. 1997b. No Escape from Syntax. Don't Try Morphological Analysis in the Privacy
of Your Own Lexicon Proceedings of the 21st Annual Penn Linguistics Colloquium
(U. Penn Working Papers in Linguistics 4:2) ed. by A. Dimitriadis, L. Siegel, C.
Surek-Clark, and A. Williams, 201-225.
Naigles, L., H. Gleitman, L. Gleitman. 1993. Syntactic Bootstrapping and Verb Acquisition.
Cognition and Language. A Developmental Perspective ed. by Esther Dromi.
Norwood: Ablex.
Noyer, R. 1997. Features, Positions and Affixes in Autonomous Morphological Structure.
New York: Garland.
Oehrle, R. 1976. The Grammatical Status of the English Dative Alternation, Ph.D dissertation,
MIT.
Parsons, T. 1990. Events in the Semantics of English. Cambridge, MA: MIT Press.
Pesetsky, D. 1995. Zero syntax: experiencers and cascades. MIT Press, Cambridge.
Pfau, R. 2000. Features and Categories in Language Production. Ph.D dissertation, University
of Frankfurt
Pinker, S. 1989. Learnability and cognition: the acquisition of argument structure. Cambridge,
Mass. : MIT Press
Prince, A. and P. Smolensky 1993. Optimality Theory: Constraint Interaction in Generative
Grammar. Rutgers University Center for Cognitive Science Technical Report
Pollard, C. and I. Sag. 1994. Head-driven phrase structure grammar. Chicago: University of
Chicago Press
Ramchand, G. 1997. Aspect and Predication Oxford Press
Ramchand, G. 1998. Deconstructing the Lexicon The Projection of Arguments: Lexical and
Compositional Factors ed. by M. Butt, and W. Geuder. Stanford, CA: Center for the
Study of Language and Information viii
Ramchand, G. 2003. First Phase Syntax. Ms. University of Oxford.
Rizzi, L. 1990. Speculations on verb second. In Mascar'o, J. and Nespor, M., editors,
Grammar in Progress, pages 375--386. Foris, Dordrecht
Ritter, E. and S. Rosen. 1998. Delimiting events in syntax. The Projection of Arguments:
Lexical and Compositional Factors ed. by M. Butt, and W. Geuder. Stanford, CA:
Center for the Study of Language and Information viii
Sproat, R. 1985. On Deriving the Lexicon. Ph.D Dissertation MIT
Siddiqi, D. (2004). Obtaining lexical contrast in Optimality Theory. Ms, University of
Arizona.
172
Travis, L. 1994. Event Phrase and a Theory of Functional Categories. Paper presented at the
Canadian Linguistics Association, University of Calgary.
Urbanczyk, S. 1999. Reduplicative Size-segmentism Correlations as Root-affix Asymmetries.
ms. Rutgers University, New Bruswick.
Woolford, E. 1984 “Dative Verbs with Unspecified Objects” Ling. Rev., 1984 June; 3 (4):
389-409
Was this manual useful for you? yes no
Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Download PDF

advertisement