The Morpheme - A Theoretical Introduction 2015
The Morpheme - A Theoretical Introduction 2015
Interface Explorations 31
Editors
Artemis Alexiadou
T. Alan Hall
De Gruyter Mouton
The Morpheme
A Theoretical Introduction
by
David Embick
De Gruyter Mouton
ISBN 978-1-5015-1078-6
ISBN (PDF) 978-1-5015-0256-9
ISBN (EPUB) 978-1-5015-0258-3
ISSN 1861-4167
Chapter Dependencies ix
Preface xi
Notes 221
Bibliography 233
Index 244
Chapter Dependencies
1!
2!
3! 4!
5!
6!
7!
8!
Preface
“I’m not as dumb as you think...I’ve seen how it worked out. I guess I can put
two and two together.”
“Sometimes the answer’s four,” I said, “and sometimes it’s twenty-two.”
As far as its contents and title are concerned, this book could have been
called The Morphemes, since the distinction between Roots and functional
morphemes is central to the theory that is developed here. But because my
main focus is on the central importance of the discrete piece—that is, the
morpheme—I have opted for the singular version.
There is also something to be said about the book’s subtitle (A Theoreti-
cal Introduction). This classification reflects my primary goals for this work.
There are three of these.
The first goal is that it will function as a component in a course (or course
sequence) on morphology, at the introductory graduate (or advanced under-
graduate) level; this is the kind of course that I have been teaching at the
University of Pennsylvania over the past several years, and a fair amount of
the material that has found its way into the work, especially in chapters 4-6,
is drawn from those classes.
The second goal is that it will serve as a clear statement of some of the
central themes in Distributed Morphology, and thus prove useful to advanced
researchers from other subareas (or adjacent areas of language research) who
are interested in the general theoretical issues that are addressed here, and in
the particular view of the morpheme that has been developed in this frame-
work.
The third goal is that it will connect a number of ideas that have been in-
vestigated within Distributed Morphology and related frameworks—notably,
the relations between the theory of the morpheme and the Vocabulary Inser-
tion operation on the one hand, with the phenomena of syncretism, block-
ing, and allomorphy on the other—in a way that will be of interest to re-
searchers working in a variety of areas. My impression from the literature
is that, while substantial progress has been made in understanding the phe-
xii Preface
nomena just mentioned, there is a need for a systematic overview of how the
different parts of the theory might fit together.
Along with the three goals just mentioned, it is worth mentioning what
this book is not. For one, it is not intended to be a textbook per se, even if
it is a book that could be used in a class. While it offers an introduction to
the morpheme, its concerns are often directed at theoretical points that are
complex and open to many directions of resolution. It does not start with a
slow introduction to the phenomena that are considered to be part of “mor-
phology”; instead, it starts with a core notion—the morpheme—and proceeds
from there to two of the central theoretical questions for the morpheme, syn-
cretism and allomorphy. Moreover, there are no exercises of the type that
are central to texts. This might mean that it is somewhat closer to being a
handbook. In a limited sense, this is probably true. However, a handbook
is typically regarded as a comprehensive overview of the literature up to a
point, and I have made no attempt to give a detailed overview of the literature
here. Instead, I have tried to connect the essential components of one view
of the morpheme into a coherent system. My hope is that the overarching
system will allow specific points of theoretical interest to be understood and
developed further; i.e., that those who use the book will be in a position to
understand the primary literature, and to make contributions to it.
**
Regarding what is covered in the book, and the scope of the background
literature that is cited—
As far as the time-period of the work covered here goes, I think it is fair to
say that there is no good time to finish a work like this one. Looking back into
the work I see several things that could have been done differently, connected
in a completely different way, and so on. Looking at the present, it is, of
course, impossible to keep updating a book in a way that reflects a field that
is developing continuously. In practical terms, this means that the book omits
some of the most recent research in this area; a large component of this book
took its current form in the summer and fall of 2010, and in making revisions
and additions over the past few years I have not included works published
after that time.
As far as the scope of the work is concerned, I have tried to concentrate
on the theory of the morpheme (and on the attendant theory of Vocabulary
Insertion) as much as possible. The focus on Vocabulary Insertion (and on
Preface xiii
****
*****
1.1 Introduction
logical theory because the objects of study often consist of more than one
morpheme, and it must be asked how morphemes are composed into these
larger objects. In terms that are inherited from traditional grammar, the rules
governing the construction of phrases are in the domain of syntax, while the
rules that are responsible for the derivation of complex words are the purview
of morphology or word formation. Whether or not these are truly distinct
domains—i.e., whether or not the grammar of human language employs one
set of computations for the derivation of words and phrases, or two—is an
empirical question. On the assumption that there is some kind of derivational
system at play in the generation of both words and phrases, every model of
the grammar must specify whether words and phrases are derived by the same
rule system, or by different systems.
The questions outlined immediately above can be stated as Q1 and Q2:
Question 2 (Q2): What is the nature of the rule system that assembles mor-
phemes into words? How does this system relate to the syntax?
The theory of the morpheme that is developed in this work assumes and
articulates some specific architectural assumptions that are drawn from the
framework of Distributed Morphology, which was initially presented in Halle
and Marantz (1993) (building on Halle 1990); see Embick and Noyer (2007)
for a short overview. With respect to Q2, Distributed Morphology takes the
position that the syntax is the generative system responsible for the derivation
of all complex objects; this assumption was introduced in the first paragraph
of this chapter. At a number of points in this chapter and in later chapters,
some additional assumptions will be made about the types of structures that
are relevant for morphological concerns. However, it is not my intention in
this book to consider the arguments for a syntactic approach to morphology;
nor will I look in detail at the syntactic processes and structures that are impli-
cated in the affixation of morphemes to each other. Rather, assumptions about
structure are introduced only as a means of advancing the primary focus of
the work, which is directed at Q1, the question of the morpheme.
To look at the morpheme in detail, it is necessary to begin with some
assumptions about the organization of the grammar. Because Q1 and Q2
The Architecture of the Grammar 3
The model of grammar that is assumed here is shown in (1). The syntactic
part of the grammar generates syntactic structures, which are “spelled out”
and subjected to further operations relevant to sound and meaning: these are
the PF and LF interfaces respectively. For concreteness, I will speak of the
derivations in (1) in terms adapted from the approach of Chomsky (2000,
2001) and related work; in principle, though, the theory of the morpheme
that is advanced here is compatible with a number of different assumptions
about syntactic theory.
4 Morphemes in the Grammar
Syntactic Derivation
(Spell Out)
Morphology
PF! LF!
Above, I introduced the idea that the primitive elements of syntactic deriva-
tions are morphemes. In terms of (1), this means that morphemes are the ter-
minal nodes in the syntactic derivations that appear at the top of the diagram.
The status of morphemes as basic objects—that is, in what form(s) they are
represented as primitives in memory—is a matter of central importance in the
theory of morphology. The basic representation of morphemes does not, how-
ever, exhaust what there is to be said about morphology and its connections
with other components of the grammar, as can be seen from the presence of
the Morphology box in (1). This box stands proxy for a set of computations
that apply to the output of syntactic derivations, in the PF component of the
grammar. With reference to (1), then, the primary goals of this book are to
present (i) a theory of the basic representation of morphemes, along with (ii)
a theory of the computations and representations in the Morphology box that
are most relevant to the theory of the morpheme.
Some further comments are required with respect to the second of these
goals. PF serves as an interface between syntax and sound-related cognitive
systems.1 Among the computations that apply to serve this purpose, there are
some that are primarily morphological in nature, while there are others that
serve a more general purpose. For example, one operation that applies in the
PF branch of the grammar adds phonological content to syntactic nodes. It is
called Vocabulary Insertion, and it is examined in detail in chapter 4. This op-
eration is specifically “morphological” in the sense that I intend here: its sole
function is to supply a sound form to (certain) morphemes. On the other hand,
other PF operations have a much more general character. For example, the
Syntactic Terminals 5
Morphemes must ultimately relate sound and meaning; more precisely, they
connect a particular type of phonological representation with a particular type
of semantic representation. Exactly how they do this is a matter of central
theoretical importance. A standard (or, at the very least, familiar) view of this
relation is that morphemes are specified from the beginning (i.e., in memory)
with features of each of these types. The theory presented in this book rejects
this view. As will be seen below, it is not the case that all morphemes are
represented underlyingly with both a meaning component and a sound com-
ponent. Instead, some morphemes possess only a representation of meaning-
(and syntax-) related features, and acquire phonological properties only after
they have been combined into complex structures.
An important component of the theory of the morpheme is the idea that
the grammar contains representations of the sound and meaning facets of
language: that is, representations of sound and meaning in terms of features.
There are two kinds of features to be considered initially: phonological fea-
tures and syntacticosemantic (henceforth, synsem) features. The former are
featural representations of the type familiar from phonological theory. The
latter are features like the [past] feature that is responsible for past tense in-
terpretation, or the [def] feature that is found in definite determiner phrases,
and so on.
Depending on further hypotheses, it could very well be the case that (certain)
synsem features are binary, like the phonological features referred to in (2a),
so that (2b) would have [±past] and [±def]. Some questions along these lines
Syntactic Terminals 7
An important aspect of (3) is that not all morphemes are specified for each
of the feature types in (2). In particular, the functional morphemes contain
no phonological features as part of their basic representations; instead, they
receive phonological content after the syntactic structure is spelled out, in the
PF component of the grammar. On the other hand, Roots do not possess (or
consist of) synsem features; and they (unlike functional morphemes) have an
underlying phonological representation, at least in the default case.
1.3.2 Roots
p p
The Roots—items like C AT or S IT—are typically spoken of as mem-
bers of the traditional lexical categories; while this description is commonly
adopted, it is perhaps more accurate to say that they are members of the
open class vocabulary.2 They are language-specific combinations of sound
and meaning—crucially, though, they relate to lexical or conceptual mean-
ing, not the type of (grammatical) meaning encoded in synsem features. For
8 Morphemes in the Grammar
p
example, the fact that the Root C AT in English has the phonological rep-
resentation /kæt/, and that it has something to do with felis catus (and not
something else), is particular to English. This aspect of the Root—the arbi-
trary connection between sound and meaning—must simply be memorized.
Crucially, this type of sound/meaning connection p does not involve synsem
features; another way of putting this is that C AT cannot be decomposed
(that is, constitutively broken down) into synsem features.
While the way in which a Root connects a particular phonological repre-
sentation with a particular meaning is arbitrary, it is nevertheless quite likely
that there are significant universal constraints on what could be a possible
Root in human language; particularly with respect to what kinds of meanings
could be associated with a Root. This is a substantial topic in its own right,
one which must take into account both the nature of conceptual representa-
tions and their relationships with objects in the grammar, and the “differen-
tial” contributions to meaning of the type associated with Saussure’s theory of
the sign. As the details of Root semantics and representation are beyond the
scope of the present work, I will focus primarily on matters of morphosyn-
tactic and morphophonological representation in the discussion to come.
In this book, it is assumed as a working hypothesis that Roots possess a
phonological representation
p as part of their primitive make-up. In this sense,
then, the Root C AT in English is specified with the phonological matrix
/kæt/ from the beginning. While in many cases a phonological representation
might suffice to uniquely identify a Root in a language, this is not always true.
In particular, it is sometimes necessary for a Root to possess an index that
uniquely identifies it. The need for this type of index is clearest when there is
homophony between Roots. For instance, the Roots underlying bank ‘finan-
cial institution’ and bank ‘shore of a river, etc.’ have phonologically identical
underlying forms, but are nevertheless distinctpRoots. Using numbers
p for the
indices, these Roots could be specified as e.g. BANK254 and BANK879 , in
addition to possessing a phonological form. In representations of this type,
each Root has a unique non-phonological identifier in addition to its phono-
logical features. Beyond the indices just mentioned, it is also conceivable that
Roots in some languages are specified with “morphological” features of the
type associated with systems of conjugation or declension (or grammatical
gender); some questions along these lines are addressed in chapter 2.
The assumption that Roots can have a phonological representation under-
lyingly has important consequences in a number of domains. It means, for
example, that in the default case Roots are not subject to “late insertion” of
Syntactic Terminals 9
[ ab g ] $ /X/
|{z}
| {z }
synsem features (phonological) exponent
Collectively, the set of Vocabulary Items like the one schematized in (4) com-
prise the Vocabulary of a language.
10 Morphemes in the Grammar
#
H
�� H
n [+pl]
p ��HH
C AT n
When the item in (5) applies to the node [+pl], the effect is that the phono-
logical exponent /-z/ is added to that node. Informally, it is said that [+pl] is
spelled out or realized as /-z/; correspondingly, the Vocabulary Insertion op-
eration is sometimes referred to informally as spell out.4 Terminologically,
theories that allow for morphemes to receive phonological form after they
are combined in this way are said to have late insertion; sometimes the term
realizational is applied to theories of this type.
The results of Vocabulary Insertion applying to (6) are shown in (7), which
also shows the n head with a null exponent (i.e. -Ø):
(7) Structure for cats, after Insertion
#
�H
� �� HH H
n [+pl, /z/]
�H
� HH
p �
C AT [n, /Ø/]
Lists and Decomposition 11
All Vocabulary Items have the form shown in (4). The only modification
to this picture is found in cases in which morphemes in the context of the
morpheme undergoing Vocabulary Insertion are referred to in a Vocabulary
Item. This results in a phenomenon called contextual allomorphy. For exam-
ple, while the [+pl] node is given its form by the Vocabulary Item in (5) when
it is attached to cat, this is not the case when [+pl] is attached to ox. In this
latter case, the [+pl] morpheme is realized as (orthographic) -en. In this sit-
uation, it can be said that the [+pl] morpheme has two distinct (contextual)
allomorphs, -s and -en.
Vocabulary Insertion is central to the view of the morpheme that is ad-
vanced in this book. Chapter 4 provides a much closer look at this operation,
and the motivations for it are examined in chapters 5 and 6. Contextual allo-
morphy is treated in detail in chapter 7.
A basic tenet of the theory developed here is that morphemes (whether Roots
or functional heads) are pieces, and that morphological phenomena can only
be understood in a theory that is piece-based. In light of this claim, it might
look as if the approach fails to consider some other issues that are traditionally
regarded as part of morphology. This is particularly true of various types of
non-affixal morphological changes; for example, the change seen in English
sing/sang, where it is the vowel seen in the Root (and not a segmented af-
fix) that—informally speaking—distinguishes the past tense from the present
tense form. This kind of effect is put to the side for the moment, because sev-
eral more issues have to be addressed before it can be seen how non-affixal
processes can be understood in terms of functional morphemes and Roots.
Ultimately, it will be seen that the Distributed Morphology framework treats
piece-based morphology and non-affixal morphology with distinct mecha-
nisms. The motivation for distinguishing affixal and non-affixal morphology
in this way is discussed in chapter 7.
Both computations and lists play an important role in the theory of grammar.
In this section I will briefly comment on the computational part of the theory,
12 Morphemes in the Grammar
before looking in more detail at some specific assumptions about lists that are
part of the theory of Distributed Morphology.
Distributed Morphology holds that all complex linguistic objects are de-
rived by the syntax. A consequence of this view is that there is no architectural
difference between the manner in which words and phrases are derived. Or,
put another way, this means that the grammar does not contain two distinct
generative systems, one responsible for the creation of complex words, the
other for the creation of phrases. Rather, the theory answers (Q2) above by
holding that there is a single generative component in the grammar.
With this architectural position in focus, it is important to emphasize that it
is the basic derivation of complex objects that is syntactic in this theory. This
leaves open the possibility that structures created by the syntax may be further
operated on by computations of the PF branch of the grammar, in ways that
are relevant to morphology and phonology. For example, in some contempo-
rary views of the mechanics of affixation, the creation of the complex heads
that are at the center of the theory of morphology (see chapter 3) may occur
either in the syntax, or in terms of PF relations that are derived from syntac-
tic representations. The second disjunct here refers to postsyntactic affixation
operations, which create complex heads in the PF component of the grammar.
These PF affixation operations apply to the output of a syntactic derivation;
from this, it follows that the syntax plays a defining role even when complex
heads are created at PF. Because the syntactic structure is the input to PF, and
because linear relations are derived from syntactic constituent structures, it is
the syntax that determines which nodes are in local PF relationships. Thus,
the theory is syntactic at its core, even if PF operations are required for the
analysis of certain phenomena.
Beyond having a syntax that derives complex structures, it is also neces-
sary for grammars to list certain types of objects (or information about ob-
jects) that play a role in the architecture in (1). For example, the morphemes
themselves, which are the basic building blocks of the theory, are not de-
rived by any operation—they exist in memory, on a list that is accessed in the
course of a syntactic derivation.
Questions about how lists and the property of listedness fit into the theory
of grammar are extremely important: first, because the theory must specify
precisely how listed information interacts with different aspects of the archi-
tecture in (1); and second, because of commonly held understandings (and
misunderstandings) about the nature of lists and what is often associated with
the notion of lexical information in the grammar. In the remainder of this sec-
Lists and Decomposition 13
tion, I briefly review some key assumptions about the lists that play a role in
Distributed Morphology, beginning with some background discussion con-
cerning some of the senses of the term lexical that are employed grammatical
theory.
Because it builds all complex objects in the syntax, the theory of morphology
that is presented in this book is sometimes referred to as a non-lexicalist the-
ory. What this means is something that must be approached in a few steps,
since there are many different theories and ideas to which the term lexicalist
has been applied. I will focus on two key senses of the term: one that concerns
architectural assumptions about the derivation of complex forms, and another
that is relevant for listed information and how (and where) it is represented.
In the way that I use the term, a non-lexicalist theory stands in contrast
to theories that posit a non-syntactic generative system that is responsible for
the derivation of words. Many approaches that derive words in a component
of the grammar that is not the syntax originated in the 1970s, and this idea
is still an active part of many frameworks. In theories that attribute a special
architectural status to words, the term lexicon is used for the non-syntactic
generative component of the grammar; accordingly, having a lexicon in this
specific technical sense results in what is called a lexicalist view of grammar
(see e.g. Carstairs-McCarthy 1992 for an overview and background discus-
sion).
As mentioned above, the term lexical is used in many ways in addition
to the technical sense “having a generative Lexicon” introduced immediately
above (Aronoff 1994 examines this point in its historical context). Of these
further uses of the term, perhaps the most important one uses lexical to de-
scribe information that must be listed. As it turns out, this list-related sense
of what it means to be lexical actually subsumes two distinct notions. The
first is that the lexicon is a list of the basic building blocks of the gram-
mar: that is, the primitives that are employed in the construction of complex
forms. Primitive building blocks have to be listed in some way, since they
are, by definition, not derived. The second type of listing associated with the
term lexical is related to certain kinds of unpredictability that are associated
with prima facie non-primitives. In this second sense, the lexicon is seen as
a repository for information about complex forms whose behavior is unpre-
14 Morphemes in the Grammar
dictable in some way, and which therefore must be recorded. For instance,
the fact that a certain noun (e.g. ox) takes an irregular plural form (ox-en),
or the fact that a certain phrase (e.g. buy the farm) has a “special” meaning
(something like ‘die’), is not predictable from anything else in the system;
therefore such special properties must be listed.
The two types of information just described—(i) the representation of the
primitive building blocks, and (ii) the irregularities of larger objects—are log-
ically distinct from one another. This means that, in principle, these two types
of information could be encoded in distinct lists; this is the approach that is
advocated in the present work. However, a frequently encountered proposal,
one which plays a role in many theories of grammar, is that there is a sin-
gle list that performs both of the functions identified in the last paragraph.
This view stems from the influential discussion of Bloomfield (1933), whose
definition of the lexicon is taken at face value in much later work. A central
passage expressing his view is the following:
It is clear from the second half of this quote that Bloomfield’s lexicon per-
forms the two distinct functions that are discussed earlier: it is both a list of
morphemes, and, in addition, a list of (the unpredictable properties of) irreg-
ular complex forms.
Putting things together, there are now three different ways in which the
lexicon and lexical information have been implicated in the discussion to this
point. These correspond to the different notions or senses of lexical that are
outlined in (8); and it bears repeating that (8a-c) are logically distinct from
each other.
(8) Notions of Lexical
a. The idea that the lexicon is a generative system in which words (as
opposed to syntactic objects) are derived.
b. The idea that basic elements (morphemes) must be listed, because
they are underived.
c. The idea that the unpredictable behavior of complex objects must
be listed.
Lists and Decomposition 15
With (8) at hand, we are now in a position to understand one of the impor-
tant architectural claims of many lexicalist theories: the idea that the grammar
contains a lexicon that—in addition to combining (8b,c), as Bloomfield’s lex-
icon does—also derives words (8a). In terms of the overall organization of the
grammar, a further assumption that characterizes theories of this type is that
the lexicon is “prior” to the syntax, in the sense that it provides the syntax
with its terminal elements (which, in theories of this type, are words).5 With
some non-trivial differences, theories that incorporate a single lexicon cover-
ing all of (8) are explored in Lieber (1980), Selkirk (1982), Kiparsky (1982),
and much related work.
This is not the place to review the details of how the functions in (8) are
accounted for in lexicalist theories, or the distinct ways in which the lexicon
has been defined, or the manner in which different lexicalist theories have
defined crucial notions like listedness, wordhood, etc.6 The point that is cru-
cial for present purposes is this: the non-lexicalist theory developed in this
book does not have a lexicon of the type identified in the last paragraph, i.e.,
a single component of the grammar that performs all of the functions in (8).
Nor does it have a lexicon that performs the two functions associated with
Bloomfield’s lexicon, i.e., it does not have a single list covering (8b) and
(8c). But—and this point is crucial—this does not mean that the approach
dispenses with the need to list certain types of information. A non-lexicalist
theory must still make use of lists, and must in some form account for all of
the list-related phenomena discussed in this section (that is, (8b) and (8c)).
As will now be shown, Distributed Morphology employs three distinct lists,
and situates access to these lists in different places in the architecture in (1).
Two of the three lists have already been introduced above: first, the list of
Syntactic Terminals, which consists of the functional morphemes and Roots;
and second, the Vocabulary, the list of Vocabulary Items that insert phonolog-
ical content into functional morphemes. These lists are accessed at different
points in the model in (1). The list of Syntactic Terminals is accessed by the
syntax, when complex objects are built. The Vocabulary is consulted after
syntactic structures are sent to the PF component of the grammar. The list of
morphemes (= Syntactic Terminals) and the Vocabulary Items are both cen-
tral to morphological concerns. Using separate lists for them in the way just
described and putting these lists in different parts of the grammar are thus
ways in which morphology is distributed in the present approach.
Moving beyond the Syntactic Terminals and the Vocabulary, something
remains to be said about (8c): unpredictable information associated with com-
16 Morphemes in the Grammar
of language are distinct, and require distinct representations (see e.g. Marantz
2010). Numerous questions of this type could be examined in a more sus-
tained study of this part of the theory of meaning; but since my concerns here
are more morphosyntactic and morphophonological, I will put these to the
side.
In summary, the theory developed here does not have a Lexicon in the way
that many lexicalist theories of grammar do. Rather, it employs three lists: the
list of Syntactic Terminals, the Vocabulary, and the Encyclopedia.
1.4.2 Decomposition
The previous subsection looks at the kinds of information that must be listed
in the grammar, and outlines an approach that employs three distinct lists.
Specifying what must be listed, though, is only part of the picture. Distributed
Morphology also takes a stance on what kinds of objects cannot be listed. In
particular, the theory holds that all complex objects—i.e., all objects that
consist of more than one morpheme—are derived syntactically, every time
they are employed. In other words, every word and every phrase is built every
time it is used. This position manifests a strongly decompositionalist view of
language, in which there is no possibility of storing non-primitives. It is stated
as Full Decomposition in (9).
part—that is, when they can be decomposed into a shared element. In the case
of plays and played, for example, pthe relatedness of these forms is found in the
fact that both contain the Root P LAY. Full Decomposition holds that there
is no possibility in the theory of storing plays and played as separate objects
in memory, and stating relationships of “relatedness” between them.pRather,
they are related because they are built from a shared piece, the Root P LAY;
that is, relatedness is defined decompositionally, in terms of morphemes.
While Full Decomposition might appear obvious (at least in some quar-
ters) when applied to “regular” forms like plays and played, its strength is
much more apparent in the case of “irregular” past tense forms like sang. In
particular, Full Decomposition p rules out an analysis in which sang is stored
as a combination of the Root S ING plus the past tense morpheme T[+past].
By way of contrast, this kind of “whole word storage” is a property of many
other theories of morphology, where it is assumed that any form that is irreg-
ular must be “stored”, not “derived by rule”.7
With respect to sing/sang, implementing an analysis that complies p with
Full Decomposition begins with the observation that both the Root S ING
and past tense morpheme pT[+past] exist as independent morphemes. This is
clear from the fact that S ING appears in a number of other contexts (e.g.
sing-Ø, sing-s, sing-ing, etc.), while T[+past] appears with a number of other
verbs (play-ed,ppass-ed, plaster-ed, etc.). Thus, by Full Decomposition, the
only way that S ING and T[+past] can be combined is in a syntactic deriva-
tion: one that produces the complex head shown in (10). (Also included here
is a categorizing v head that creates verbs (see chapter 2); the T[+past] mor-
pheme is shown adjoined to v, for reasons having to do with how English
tense-affixation works):
p
(10) Past Tense of S ING
v
�H
� HH
�
v T[+past]
p ��HH
S ING v
they can be decomposed into a sharedppart: each of these forms is the result
of a derivation that contains the Root S ING.8
The predictions that derive from Full Decomposition are the topic of on-
going research both in the theoretical domain, and in psycho- and neuro-
linguistic studies (for the latter see e.g. Embick and Marantz 2005 and Stock-
all and Marantz 2006 for perspectives that connect with the present work).
The importance of Full Decomposition is seen in many of the case studies
that are examined later in this book: particularly in the analysis of contex-
tual allomorphy (where irregular forms are at issue) and of blocking effects
(where lexical relatedness is crucial); see chapter 7.
The framework of Distributed Morphology uses three lists: the list of Syntac-
tic Terminals; the Vocabulary; and the Encyclopedia. It is worth mentioning
that with respect to “conceptual” questions about the nature of the grammar,
these lists might not all be on equal footing. Two of them are more or less im-
peccable as far as considerations of parsimony (“Minimalist” or otherwise)
are concerned: the Syntactic Terminals and the Encyclopedia. This is because
every theory needs to have a list of building blocks and to list certain types of
unpredictable interpretations.
The motivation for the Vocabulary, on the other hand, is not as self-evident
as the motivation for the other two lists is. Even though it deals with in-
formation that has to be listed—since the phonological forms that realize
morphemes are not predictable—there are ways of encoding this information
that do not require a third list; by representing phonological information with
morphemes in the list of Syntactic Terminals, for example. That is, admitting
the Vocabulary into the ontology of the theory is a consequence of denying
that functional heads are “traditional” morphemes (= morphemes that pos-
sess both sound and meaning from the beginning) in favor of a realizational
approach to sound/meaning connections.
It is not clear whether or not there is any sort of conceptual motivation
for the Vocabulary. However, the motivation for Vocabulary Insertion is not
meant to be conceptual; it is empirical, and is based on the need to state
certain cases of morphological identity—instances of the phenomenon of
syncretism—systematically. Section 1.6.2 of this chapter provides an intro-
duction to syncretism, which is then examined in detail in chapters 4-6.
20 Morphemes in the Grammar
The theory developed in this book is centered on the idea that morphemes
are the primitive elements of derivations in the grammar in (1). The theory
employs the three lists in (11):
(11) Lists
a. The Syntactic Terminals: The list containing the Roots and the
Functional Morphemes.
b. The Vocabulary: The list of Vocabulary Items, objects that provide
phonological content to functional morphemes.
c. The Encyclopedia: The list of special semantic information.
These lists are accessed at distinct stages, as shown in the diagram in (12):
(Spell Out)
Access to
The Vocabulary
PF! LF!
Access to
The Encyclopedia (Interpretation)
Summary and Core Properties 21
Morphemes are drawn from the list of Syntactic Terminals, and serve as the
terminal elements of syntactic derivations. The Vocabulary is consulted at PF,
where phonological exponents are added to functional morphemes. Finally,
the Encyclopedia is accessed in the interface with interpretation. For conve-
nience, I have shown this after the LF interface (which I take to encode the
more “structural” aspects of compositional interpretation). There are some
oversimplifications involved in this treatment of Encyclopedic knowledge,
but these do not impact the central claims of this work.
The rest of this book articulates the central claims of Distributed Morphol-
ogy, with respect to the morpheme and the Vocabulary Insertion operation in
particular. Although Distributed Morphology is a framework in which dif-
ferent theories can be explored, there are some core positions that define its
boundaries. As I see it, these are as follows:
Note: This section outlines a number of themes that recur throughout the
book, in a way that might require a more advanced background than the pre-
ceding sections. Much of this background is presented in later chapters of
the book. Thus, readers who are directly interested in the development of the
nuts-and-bolts of the theory might wish to continue on to chapter 2, and re-
turn to this section at a later point (e.g. when specific themes are referred to
later, or in conjunction with the concluding remarks in chapter 8).
a lexicon of the type defined in 1.4.1), or deny that the form of words is based
on a decomposition into morphemes in the first place (e.g., Anderson (1992)
who develops an affixless or amorphous theory, for inflectional morphology
anyway). Theories of the syntax/morphology interface thus range from the
more opaque (where the structure of words and the structure of clauses in
which those words appear have nothing to do with one another) to the more
transparent (where morphological structures are syntactic structures). The lat-
ter type of view, which I refer to as interface transparency, constitutes one
of the main tenets of Distributed Morphology. In essence, it amounts to the
claim that mismatches between syntax and morphology are minimal. More
precisely, the idea is that transparent syntax/morphology connections are the
norm, and that deviations from the norm are the result of operations that apply
in the PF component of the grammar.
Within transparent theories of the syntax/morphology interface, a standard
view is that complex heads are the most important syntactic objects of mor-
phological interest. A complex head consists of two or more morphemes that
have been combined by an affixation operation. For example, a Root that ap-
pears in a syntactic structure with heads X and Y (13) can be combined into a
complex head (14) with these heads. The affixation shown in (14) could, p for
example, be produced
p by the process of head movement, taking the ROOT
to X, and then [ ROOT X] to Y :
YP Y
�
�H
� H
H ��HH
Y XP X Y
�H p ��HH
�� HH p ROOT X
X P
p ��HH
ROOT ...
While Latin shows distinct agreement morphemes for the six combinations
of person and number features shown in (15), Latin American Spanish shows
only five distinct phonological exponents, with an -n appearing in both the
second person plural and third person plural contexts. The fact that -n appears
in these two morphosyntactically distinct environments is not an accident;
that is to say, it is not a case of homophony of the type found with bank
‘side of river’ and bank ‘financial institution’, or with plural /-z/ versus third
person singular present agreement /-z/ in English. Rather, in the analysis of
Latin American Spanish there needs to be a way of saying that the same
-n appears with two distinct sets of synsem features; this effect is what is
referred to as a (systematic) syncretism in this work.
Syncretisms are pervasive in natural language, and the theory of the mor-
pheme must be able to analyze them as non-accidents. The Vocabulary In-
sertion operation, introduced in section 1.3 of this chapter, is designed to
provide a systematic (that is to say, non-accidental) analysis of syncretism. It
does this by providing a mechanism by which a single phonological exponent
can occur in multiple synsem contexts; i.e., as the phonological realization of
distinct morphemes.
The key idea is that Vocabulary Items may be underspecified with respect
to the functional morphemes that they apply to. In order to see this point in
a specific example, consider the functional morphemes involved in Spanish
agreement (AGR) in (16); the synsem features are [±1] and [±2] for person,
and [±pl] for number:
[+1,-2,+pl] $ -mos
[+1,-2,-pl] $ -o
[-1,+2,-pl] $ -s
[-1,-2,-pl] $ -Ø
[-1,+pl] $ -n
A final thread that extends through the following pages is extremely general;
in some sense, it is one of the main arguments of Distributed Morphology.
This is the idea that the morpheme is essential for explaining a number of
phenomena in the grammar. This may sound unsurprising, especially given
the assumptions that are articulated above; at a minimum, the morpheme must
have an important status in a theory like the one outlined here, because it is a
syntactic theory, and morphemes have the privileged status of being the ter-
minal elements in syntactic derivations. But if the central claim of this book
is correct, then the morpheme is special in ways that go beyond this. In par-
ticular, it appears that the morpheme as defined in this work is the object that
is referred to by a number of formal operations and representations which,
taken together, cover a large part of the theory of morphology. To highlight
this facet of the theory, I refer to the convergence of distinct subtheories on
morphemes (and not objects of other sizes) as the Morpheme-Locus Effect.
(P)review: Main Themes 29
very general architectural points. For instance, much of what is called Inter-
face Transparency above is centered on the morpheme, and would be impos-
sible without it.
In sum, the Morpheme-Locus Effect is a shorthand way of referring to
the central claim of this book: the theory of morphology requires morphemes,
because it is only in this way that a number of grammatical phenomena can
be explained in a principled way.
Chapter 2
Morphemes and Features
2.1 Introduction
The first chapter of this book outlines the essential properties of a piece-based
view of morphology in which morphemes are the terminal nodes in syntactic
structures. A central idea in the theory is that morphemes are of two distinct
types: the functional heads and the Roots. The present chapter looks in greater
detail at each of these types of morpheme, with particular emphasis on how
they relate to different types of features.
As stressed in chapter 1, morphemes are the primitive elements of syntac-
tic derivation; that is to say, they are atomic, as far as the syntax is concerned.
This means that the morpheme is the smallest object manipulated by syntac-
tic operations. But the idea that morphemes are atomic does not mean that
they have no subparts. Indeed, according to the view that I will adopt here,
some morphemes (functional morphemes in particular) are bundles of fea-
tures. In the words of Chomsky (2007:6), such morphemes are “a structured
array of properties (features) to which Merge and other operations apply to
form expressions”.
The precise sense in which morphemes are structured objects is of some
interest. As an initial hypothesis, a morpheme consisting of different features
Fi can be represented as a set like (1):
(1) {F1 , ... , Fn }
Beyond the minimum of (1), how much internal structure morphemes pos-
sess is an empirical question. The simplest position is (presumably) that mor-
phemes are sets of features and nothing more, so that positing any additional
structure (e.g. “geometries”, or other hierarchical representations of features)
requires significant empirical motivation.1
While (1) is the basic way in which morphemes are represented, some
notational conventions are often employed in representing morphemes, par-
ticularly with respect to how category is treated. Following standard practice,
morphemes in this book are sometimes represented with a category label sep-
arate from further specification of feature content. So, for instance, when this
convention is employed, a past tense morpheme is represented as in (2):
32 Morphemes and Features
(2) T[+past]
The idea that morphemes are sets of features is most clearly relevant to func-
tional morphemes, where the approach developed here extends a view that
is standard in syntactic theory: functional morphemes are the familiar func-
tional categories, by definition, bundles of grammatical features. As will be
seen below, however, Distributed Morphology expands the use of functional
categories, by making use of functional heads to provide the “lexical cate-
gories” like noun, verb, etc. to Roots; see section 2.3.4 below.
This section looks at the non-phonological aspect of functional morphemes:
their synsem features. Questions about the phonological representation of
functional morphemes are put off until chapter 4, where the Vocabulary In-
sertion process is introduced.
In this morpheme, the two features together produce the relevant interpreta-
tion: the [+1] feature is responsible for the first person part, and [+pl] provides
the plurality.
Many languages possess the functional morpheme in (3). But it is by no
means necessary for the two features in (3) to be bundled together into one
morpheme. As far as the claims of this chapter go, it could very well be the
case that some languages have [±1] and [±pl] in separate morphemes. This
latter scenario is perhaps what is found in languages like Mandarin Chinese,
where personal pronouns show separate morphemes for person and number:
p/n form
1s wŏ
2s nı̆
3s tā
1p wŏ-men
2p nı̆-men
3p tā-men
So, whereas English has a single morpheme (3) that receives the pronunci-
ation we, Mandarin Chinese has two separate morphemes, [+1] and [+pl],
which are used to derive the meaning of first person plural. The general idea
that languages differ in terms of how they bundle features—and that these
differences have interesting empirical consequences—has been explored in a
number of domains.3
While it will be assumed throughout this work that languages may differ
with respect to how they bundle features along the lines just illustrated, some
Functional Morphemes 35
this kind of coherence is defined, and what motivation(s) there might be for
it, are important questions that are at present open for further investigation.
singular n´ı̄kē
dual n´ı̄kā
plural n´ı̄kai
The same number distinctions are found in verbal agreement as well. The
present tense verb forms in (6) show different endings that reflect the three-
way division of number:
Functional Morphemes 37
p/n form
1s lū´ō
2s lū´eis
3s lū´ei
1d lū´omen
2d lū´eton
3d lū´eton
1p lū´omen
2p lū´ete
3p lū´ousi
The number system of Classical Greek can be compared with that found in a
language like English, where there are only two numbers: singular and plural.
In a language of the latter type, a single binary feature suffices to make the
needed synsem (and hence morphological) distinctions; for convenience, we
can take this feature to be [±pl]. Then, [-pl] and [+pl] encode “singular” and
“plural” respectively (the terms in quotes can be understood as a shorthand
way of referring to the more basic feature distinction).
Clearly, a single binary feature [±pl] is insufficient to make the semanti-
cally relevant distinctions in a language that has a dual in addition to singular
and plural. On the assumption that features are maximally binary, another
feature is needed. Here, it will be assumed that dual number arises in systems
with two binary features, [±sg] and [±pl]; the dual is found when the value
for each such feature is negative, as shown in (7):4
+sg -sg
+pl –– “plural”
-pl “singular” “dual”
Classical Greek, on the other hand, employs functional morphemes with the
number features shown in (9):
In this analysis, English and Greek differ in terms of features selected from
the UFI: English uses [±pl], but makes no use of [±sg]. In terms that are
employed at the beginning of this section, this means that a feature that is ac-
tive in one language (Greek) is not active in another (English). (Secondarily,
Greek also differs from English on this analysis in having [±sg] and [±pl]
features bundled in the same node).
It should be evident from even this introductory discussion that the pre-
cise nature of the difference between these languages depends on assump-
tions about the representation of features. Suppose, for example, that (per-
haps because of some other assumptions about uniformity of features) En-
glish were treated with [±sg] and [±pl] features. Then, English singular and
plural would be defined as in Classical Greek: that is, they would be [+sg,-pl]
and [-sg,+pl] respectively, while [-sg,-pl] would not exist as a morpheme in
the language. Under this second scenario, English and Classical Greek would
both have the same features from the UFI, in the sense that both would em-
ploy [±sg] and [±pl]. The difference between them would then be that Classi-
cal Greek would have functional morphemes with [-sg,-pl], whereas English
would not. Thus, the languages would differ in how features are bundled into
morphemes, but not in terms of which features are active in the languages in
the first place.
On the basis of just the information considered above, it is difficult to
distinguish between the first (inventory-based) and second (bundling-based)
treatments. There might be additional empirical and theoretical considera-
tions favoring one or the other when systems of number are analyzed in detail.
However, because the comparison of Greek and English suffices to illustrate
Functional Morphemes 39
the basics of how individual languages make use of the UFI, I will not exam-
ine this point further here.
Speaking informally, it is often clear when one language makes a distinc-
tion that is not found in another, as in the case of the dual above. Sometimes,
though, many difficult questions arise when even simple-looking differences
are examined in detail. Consider, for example, the domain of “tense”, broadly
construed for present purposes so as to refer to tense and aspect together.
Verbs in even relatively similar languages often differ in terms of the tense
distinctions they express. For example, English has present and past tense
forms of the verb, whereas Spanish verbs manifest a further distinction be-
tween imperfective and preterite (= perfective) past tenses:
Clearly, English and Spanish differ in terms of the number of distinct finite
verb forms they possess. What this means at the level of synsem features ac-
tive in the language is, however, less obvious. For instance, one of the mean-
ings of the Spanish past imperfective form is past progressive. In English,
past progressive meaning is expressed with what is called the progressive
form, which consists of a form of be along with an -ing form of the main
verb:5
Determining how the Tense (and Aspect) morphemes of English and Spanish
might differ thus ultimately depends on an analysis of how the meaning of the
English (past) progressive relates to the meaning of the Spanish imperfective.
It is possible that the languages use very similar features that are “packaged”
differently: that is, so that the same features are realized as a single “verb”
(= complex head) in Spanish, but as an auxiliary plus a participle in English.
40 Morphemes and Features
There are, of course, alternatives to this analysis, according to which, for ex-
ample, English and Spanish employ different (though semantically similar)
features and morphemes in this part of the tense system. Answering ques-
tions of this type requires an integrated theory of semantics and syntax along
with morphology; it is a good example of why many topics of morphological
interest cannot be investigated in isolation.
The question of how morphemes are packaged into complex heads is
an important topic and will be taken up again in chapter 3. For immedi-
ate purposes, the comparison based on (10) shows that superficial differ-
ences in morphosyntax are not necessarily always reducible to differences
in synsem feature inventories or morphemes in a simple way. Rather, such
differences might involve either differences in features and their bundling
into morphemes, or the arrangement (through affixation operations) of mor-
phemes into complex heads, or both.
2.3 Roots
A hypothesis that has been adopted in much work is that Roots are repre-
sented with phonological features underlyingly. In terms of the standard no-
tation that is used for this type of morpheme, the Root underlying the noun
cat is shown in (12):
p
(12) C AT
p
On the further assumption that the Root C AT has a phonological matrix
underlyingly, (12) can be thought of a shorthand for a more explicit repre-
sentation that is minimally the one in (13) (using the set representation of
morphemes employed earlier in this chapter for functional morphemes):
(13) {/kæt/}
The indices are chosen arbitrarily; all that is needed is a minimal way of dis-
tinguishing these two Roots from each other, since the phonology is not able
to perform this function. Various other means might be appealed to in or-
der to distinguish homophonous Roots from each other. For example, there
are languages in which homophonous Roots behave differently in terms of
conjugation, declension, gender, or other such “morphological” features (see
below); when this occurs the morphological features can be used to identify
each of the homophonous Roots uniquely, such that arbitrary numerical in-
dices are in principle unnecessary. For example, different Roots with the same
underlying phonology are found in Latin aud-ı̄-re ‘to hear’, and aud-ē-re, ‘to
dare’. However, these two Roots belong to different conjugation classes, as
is reflected in the different theme vowels that appear with them (-ı̄ and -ē
respectively). In a theory that encodes conjugation classes with diacritic fea-
tures (see section 2.3.5), the diacritics could be used to uniquely identify the
two Roots.
If (at least some) Roots are subject to Late Insertion, numerical indices are
essential for all Roots, since only these distinguish one Root from another.
More specifically, Late Insertion of Roots can be formalized as the insertion
of a phonological form that makes reference to a numerical index, so that a
Vocabulary Item realizing the Root in cat would be represented as follows:
p
(15) 766 $ /kæt/
p
In an approach that employs Vocabulary Items like (15), the object 766
functions as a unique identifier of the relevant Root, which then receives its
phonological form via Vocabulary Insertion. Using Vocabulary Insertion for
Roots might be useful for the analysis of (stem) suppletion, the phenomenon
Roots 43
seen in light verbs (= bundles of grammatical features; see section 2.3.6) like
go/went, if this truly exists for Roots. However, the primary motivation for
Vocabulary Insertion is syncretism; and, because syncretism is most evident
in the realization of functional morphemes, I will put the question of Vocabu-
lary Insertion for Roots to the side in the pages to come, with the understand-
ing that Vocabulary Insertion can be extended to them if necessary.
n
p ��HH
C AT n
x
�HH
�� H p
x P
⇣P
⇣⇣ PP
⇣p P
... ROOT...
x
p ��HH
ROOT x
Another possibility is that Roots are merged directly with functional heads,
so that something like (17b) is created in the absence of a phrasal projection
of a Root (the two possibilities are not mutually exclusive). The exact range
of structures like (17a) in which complex heads like (17b) are formed is a
topic of ongoing research. For the most part, the analyses presented in this
book will concentrate on complex head structures like those in (17b), without
dwelling on details of the pre-affixation structure (see chapter 3 for further
comments).
The assumption that Roots must always be categorized by a functional
head is called the Categorization Assumption:
v n a
p ��HH p ��HH p ��HH
P LAY v C AT n R ED a
According to this theory, then, Roots are not words, where word is under-
stood in its informal sense to refer to cat etc. Rather, (some) words contain
Roots, but Roots by themselves can never surface on their own.
While this claim might seem prima facie counterintuitive for languages
like English, where, for various reasons, the underlying form of Roots is of-
ten
p quite close to what we think of as a word in certain cases (e.g., the Root
C AT underlies the noun cat), in other languages it is more immediately
evident that Roots do not surface by themselves. For example, in languages p
with theme vowels or “word markers”, such as Spanish, a Root like A LT
‘high’ cannot appear as *alt on the surface. Rather, this Root always appears
with further morphemes, associated noun class (which is related to gender):
alt-o ‘high-MASC’, alt-a ‘high-FEM’, and so on. The overt -a and -o suf-
fixes express features of noun class, an “arbitrary” morphological property of
nouns
p in many languages (cp. section 2.3.5 below). The fact that Roots like
A LT cannot appear without them illustrates the point that what we identify
as the Root must appear with other morphemes in forms that actually surface.
When these additional morphemes have overt phonology, the point that Roots
always surface in combination with other morphemes is easier to see than it
is in English.
In other languages, the differences between Roots and words are even
more striking. In Semitic languages like Arabic, for example, p Roots are rep-
resented phonologically as sequences of consonants like KTB for ‘write’.
These consonant sequences are syllabified and put into a prosodic shape (de-
scriptively, a template) of consonants and vowels, in a way that is determined
by morphemes inpthe local context of the Root. For example, kataba ‘to write’
is derived from KTBp , as is kitaab ‘book’ (see section 2.3.3 for more exam-
ples). A Root of like KTB cannot be pronounced on its own p (unlike what
we might think about the phonological representation of C AT); rather, it
is capable of being pronounced only when it has combined with other mor-
phemes.
As stated in (18), it is not just any functional morpheme that can categorize
a Root. There is a special set of morphemes in the grammar that perform this
46 Morphemes and Features
v n a
�H
� H ��H HH ��HHH
p � H p � p �
DARK [v,-en] M ARRY [n,-age] G LOBE [a,-al]
In words like vaporization, there are two overt categorizing heads, as shown
in (25):
(25) vaporization
n
��HH
�� HH
v [n,-ation]
�H
p �� HH
VAPOR [v,-ize]
Category-defining heads are subject to Vocabulary Insertion, just like other
functional morphemes. This is how the -en, -age, and -al exponents come
to be inserted into v, n, and a in (22-24) respectively. With respect to forms
like play etc., in which there is no overt realization of a categorizer, some
different analyses are possible. For present purposes, it can be assumed for
these examples that the category-defining
p head has a null (-Ø) exponent; so,
for example, the verbal form of P LAY can be represented as play-Ø when
this level of detail is required.7
In addition to showing category-defining heads with overt exponents, the
structure in (25) for vapor-iz-ation illustrates an additional point: category-
defining heads can appear outside of other category-defining heads, to pro-
duce words with multiple morphemes of this type (compare break-abil-ity,
fool-ish-ness, etc.). When multiple category-defining heads of different types
appear in a single word, true “category-changing” occurs, because an object
from one category (i.e., the verb vapor-ize in (25)) is converted into a mem-
ber of another category (in the case of (25), the noun vapor-iz-ation). The
Roots 47
idea that multiple category defining heads can appear together in a derivation
is particularly important when phase-cyclic derivation is implemented; see
chapter 7 for discussion.
A final point concerning Roots and their categorizers is that the structures
containing these objects are built syntactically; they are not stored. Because
it treats the derivation of all complex forms syntactically, the theory has no
modular split between derivational morphology and inflectional morphology.
Rather, derivational morphemes like n, v, etc. are syntactic objects, just like
morphemes like Determiner (D), Tense (T), and so on.
In addition to being grammatical objects that are categorized in the way de-
scribed in the preceding section, Roots are related to both lexical semantic
and conceptual representations (note that I am making no claims about where
the lines might be drawn between these two types of information). Because
questions about Root meaning intersect both with general questions about the
nature of concepts, and with more specific questions about how the grammar
treats lexical semantic information in the first place—both extremely com-
plex questions—meaning will be treated in only a cursory fashion here.
Part of the theory of Root meanings is directed at the Root itself, and part
of it is directed at immediate grammatical contexts in which a Root appears.
By the former, I mean that Roots possess an inherent meaning (at least in
the typical case); by the latter, I mean that categorizers (and perhaps other
morphemes local to a Root) play a role in determining which meaning(s) of
a Root are active in a given grammatical context.
With these two components of Root meaning in mind, a working hypothe-
sis about Root meaning that will be adopted here is that Roots possess lexical
semantic representations that are legible to a semantic system that interprets
syntactic objects. (Recall from chapter 1 that the Encyclopedia is sometimes
referred to as encoding meanings of this type; other assumptions are possible
as well.) The system in question determines how to interpret a Root when
it appears in different grammatical contexts: e.g., when it appears as a noun
(with n), or as a verb (with v), or as a member of some other category.
In these different grammatical environments there is a shared component
of interpretation that is centered on the Root’s “basic” meaning, but per-
haps some context-specific differences as well. For example, when the Root
48 Morphemes and Features
p
F EATHER in English is employed as the noun feather, it refers to an object
that has a number of properties. Part of what English speakers know about
this Root, though, is that in another context, as the adjective feather-y,
p it can
also have a more restricted meaning, where the weight aspect of F EATHER
is prominent: (feather-y in this sense means “light or airy”, with no actual ob-
jects (feathers) involved (example from Taft (2004)). Collectively, the differ-
ent facets of lexical semantic meaning that are found with Roots in different
contexts constitute the phenomenon of Root polysemy, a phenomenon that is
widespread in human language.8
The way in which Root meanings are manipulated in different grammati-
cal contexts is particularly striking and easy to illustrate in Semitic languages
like Arabic, where, famously, a single Root can typically appear in a num-
ber of different “templates” (sequences of consonants and vowels), with in-
terpretations related to the Root’s core meaning. These templates are a mor-
phophonological manifestation of the different grammatical contexts (defined
by different functional morphemes) pin which the Root is found.
Some illustrations for the Root KTB, which is roughly ‘write’, are shown
in (26) (see Wehr (1976), Kaye (2007)); note that some of these examples
involve overt affixes in addition to changes in templatic pattern, but for the
sake of simplicity the examples are given without segmentation:9
p
(26) Some Forms of the Root KTB
The second position (27b) is closely related to the first. It says that what-
ever the semantic or conceptual content of Roots might be (in terms of R-
meanings), Root meanings are not reducible to the synsem feature inventory
of human language (i.e., to features in the UFI). While it is possible that R-
meanings can be defined in terms of a feature system that classifies Roots
along different semantic dimensions (e.g. certain Roots are semantically sta-
tive, others relate to entities, and so on), this system of Root classification
is not the synsem feature system. Another way of putting this is that the R-
meaning system and the synsem meaning systems are disjoint, even though
these systems might have to interact in a way that accounts for the fact that
Roots have different distributions (i.e., appear in different syntactic contexts;
see below).
The claim in (27b) prohibits a particular kind of lexical decomposition in
which Roots are collections of grammatical features. Some care must be taken
terminologically here, because the term “lexical decomposition” is (unfortu-
nately) used in distinct and incompatible ways in the literature (this confusion
of uses relates to the many senses of lexical that are mentioned in chapter 1).
The sense of “lexical decomposition” that is prohibited by (27b) amounts to
the claim that Roots (Root meanings) cannot be broken down (i.e., constitu-
tively decomposed) into synsem features. It is worth emphasizing the termi-
nological point because “lexical decomposition” is also used in another way
to refer to the idea that “words” as typically understood can be decomposed
into a number of morphemes that are discrete syntactic objects. This latter
type of lexical decomposition is, of course, not only admitted by the theory;
it is at the heart of it.
Both of the statements in (27) are negative. This does not mean that Roots
have no properties whatsoever (although the idea that Roots are effectively
without syntactically visible properties or absent from syntactic derivations
has been explored in this framework; see Marantz (1995), Embick (1997,
1998, 2000) for some relevant views). At a minimum, there must be a theory
of Root meanings, i.e., of R representations, to account for the simple fact
that Roots possess inherent meanings.
A further question is how R-meanings relate to syntactic distributions.
Empirically, it is clear that
p some Roots are deviant in certain syntactic envi-
ronments. For example, RUG is impeccable as a noun, i.e. The rug is red,
but deviant as an intransitive verb in the author’s English, i.e. #John rugs. If
Roots are not specified for synsem features, then it cannot be the case that
this distributional fact is encoded by specifying the Root as [−v] (or [+n],
Roots 51
or whatever would achieve the desired result). Rather, the deviance must be
attributed to how the R-meaning of the Root connects with the syntactic con-
text in which it appears.
In the discussion of rug above, the symbol # is used for ‘deviance’ because
it is an open question whether this sentence is ungrammatical, or grammatical
but unacceptable (presumably due to being semantically deviant, i.e. infelici-
tous). Different positions on this point appear in the literature. Borer (2003),
for example, allows all Root/context combinations to be grammatical in the
technical sense. Another possibility is that restrictions on Root distributions
arise from the fact that they have inherent semantic properties (in the form
of R-meanings) that determine their distribution (see the comments in Em-
bick and Marantz 2008). According to the latter view, deviance arises when
the R-meaning of a Root cannot be reconciled with the meaning contribu-
tion of the syntactic environment in which the Root appears, in a way that
is part of the grammar, and not some other (e.g. conceptual or pragmatic)
system. There are many ways in which the differences between these views
could be explored, and the distinctions between competing approaches are
often quite subtle. On a unifying note, it is important to observe that on ei-
ther one of these views, some aspect of Root meaning (i.e., R meanings)
must interact with the grammatical meanings derived from morphemes in the
context of the Root; this reiterates the point (introduced in 2.3.1 above) that
R-meanings and synsem feature meanings must be visible to each other in
some way, even if they are distinct.
There are two components to the analysis of Roots that has been outlined to
this point. The first is that Roots are specified with phonological representa-
tions, and perhaps “numerical” indices. The second is that while Roots have
(R-) representations of lexical semantic meaning, Roots are not represented
with synsem features, nor are they decomposable into them.
These components of the theory revolve around the two types of features
introduced earlier in this work, phonological features and synsem features. It
seems, however, that Roots are sometimes related to (and perhaps specified
for) another type of feature that has not been introduced yet: a type that is
primarily morphological in nature. The need for such features is seen when
the Roots (and some functional morphemes) in a language belong to specific
52 Morphemes and Features
I II III III(i) IV
‘praise’ ‘warn’ ‘lead’ ‘seize’ ‘hear’
1s laud-ō mon-e-ō dūc-ō cap-i-ō aud-i-ō
2s laud-ā-s mon-ē-s dūc-i-s cap-i-s aud-ı̄-s
3s laud-a-t mon-e-t dūc-i-t cap-i-t aud-i-t
1p laud-ā-mus mon-ē-mus dūc-i-mus cap-i-mus aud-ı̄-mus
2p laud-ā-tis mon-ē-tis dūc-i-tis cap-i-tis aud-ı̄-tis
3p laud-a-nt mon-e-nt dūc-unt cap-i-unt aud-i-unt
many other languages) raises a number of important questions that have been
discussed in the literature, because gender features interact with semantically
interpretable features that encode the sex of a referent. In the case of Spanish,
gender also interacts with the realization of noun class features that deter-
mine the realization of the final segment of most nouns (typically -o or -a).
See Harris (1991) for extensive discussion of these points.
Beyond the specific examples just considered, a general question in the
theory of features concerns the limits of inherent specification. The idea that
Roots may not be specified for synsem features, which is advanced in section
2.3.4, is a substantive proposal in this vein. For general discussion along these
lines that takes into account further feature types, see Embick (1997, 2000).
Another very general question that can be asked is why languages employ
diacritic features in the first place, given that such features are (by definition)
semantically irrelevant. This question connects with issues in diachrony and
acquisition, and may very well implicate the properties of cognitive systems
(e.g., memory) that are external to the grammar.
Concerning details of implementation, it is also important to ask whether
the phenomena associated with inherent specification all require analyses in
which diacritic and other (e.g. gender) features are specified on Roots. As far
as this goes, it is probably safer to say that some sort of memorized informa-
tion about class is associated with Roots, but need not be represented with
the Root per se. For example, Kramer (2009) argues on the basis of data from
Amharic that gender features are not exclusively specified on Roots, but can
also be found on the n heads that categorize Roots.
Along these lines mentioned in the last paragraph, it should be noted that
there are some other mechanisms that could bepused to account for some ar-
bitrary class phenomena. Representations like AUD[IV] make arbitrary class
membership a featural property of the Root. An alternative to encoding arbi-
trary class information in this way would be to employ contextual allomor-
phy (see chapters 4 and 7). My view of this matter is that it is difficult to
completely eliminate diacritics by appealing to this or related alternatives.
However, I will not dwell on this matter further here, since distinguishing
alternative proposals in this domain is a subtle matter, and implicates many
questions that go beyond the scope of this work. In the pages to come, dia-
critic specification of Roots along the lines outlined here will be employed
whenever morphological features are required.
Roots 55
(30) vaporization
n
��HH
�� HH
v [n,-ation]
�H
p �� HH
VAPOR [v,-ize]
p
By virtue of its structural position in (30), the Root VAPOR happens to
correspond to the informal notion of “host for further affixation” mentioned
above.
In some other words, however, there is no Root in the technical sense,
even though such words could conceivably have an element that is the “host”
of affixes in the informal way of speaking. For example, the theory that is
adopted in this book holds that light verbs are simply bundles of synsem
features: i.e., varieties of v. Light verbs appear in structures that are Rootless
in the way that is mentioned above. For illustration, we may employ the verb
go, which is the realization of a functional head that can be referred to as vgo .
56 Morphemes and Features
In the word goes, which consists of vgo , T(ense), and the third person singular
agreement marker -s, there is no Root in the sense defined in this book:
(31) goes
T
�HH
�� H
[ vgo ,go] [T/Agr,/-z/]
p
There is no Root G O. Thus, the word in (31) (and others like it) consists
only of functional morphemes; even if go appears to host the affix -s, this
word is Rootless.
The idea that Roots are morphemes of a particular type—and that there
may be verbs and other words that have no Root in them—connects directly
with the theory of categorization outlined in section 2.3.2. Roots are category-
neutral, such that the grammatical categories “noun”, “verb” etc. are deriva-
tive of the functional vocabulary (types of n, types of v). By the Categoriza-
tion Assumption, it is not possible to have a Root without a category-defining
head. But there is no symmetrical requirement holding that categorizers must
always occur with Roots.
2.4 Conclusions
3.1 Introduction
When structures derived by the syntax are spelled out, they are subjected to
a number of computations in each of the interface components. Collectively,
the computations and representations that are found on the sound (and sign)
interface are referred to as the PF component of the grammar.
As stressed in chapter 1, some PF computations are quite general, whereas
other operations appear to be more specifically “morphological” in nature. In
terms of this distinction, the material covered in this chapter is mostly of
the more general variety. Specifically, I will look at some of the complex
structures that are assembled out of morphemes, and how linear relations
derived from these structures are represented. The guiding idea behind this
overview is that it is in terms of the structural relations between morphemes
and the linear relations derived from these that PF computations are defined.
A convenient starting point for this chapter is the architectural assump-
tion that all complex objects are derived syntactically (recall chapter 1). A
consequence of this view is that there is no architectural or modular division
between “word formation” and “phrase formation”. Rather, a single genera-
tive system is responsible for the construction of both types of objects.
One set of questions to be addressed concerns the nature of the complex
objects that are derived from morphemes. Here, the focus is on the structures
that are realized as what are ordinarily thought of as words, and the question
of whether the word is a privileged object in the theory. As noted in chapter
1, the term word is often employed in an informal (i.e., non-technical) sense
to refer to objects whose properties are normally taken to be of “morpholog-
ical” interest. Somewhat more precisely, this use clusters around objects that
show properties associated with phonological wordhood. Since the theoret-
ical framework developed here is based on morphemes, and not words, one
of the goals of this chapter is to outline the essential properties of the struc-
tures and linear relations in which morphemes appear, because it is in terms
of these that the formal properties of complex objects must be stated.
With respect to the narrow syntactic part of the equation, it is assumed here
that the syntax generates objects that are exclusively hierarchical in nature;
60 Structures and Linear Order
i.e., that syntactic structures do not encode linear order. The representation
of linear order among syntactic objects is introduced in the PF component of
the grammar.1 The organization of the initial parts of this chapter reflects this
view: after a preliminary look at some relevant structural notions in sections
3.2 and 3.3, I provide an overview of the main questions in the study of mor-
pheme order in section 3.4. Section 3.5 then outlines some basic points about
phonological wordhood, and how this notion relates to the theoretical objects
identified in prior sections.
3.2 Structures
Y
�H
� H
X Y
The object shown in (1) is a hierarchical structure, one that also shows linear
order for graphical convenience. In principle, the structure in (1) could be
linearized in one of two ways, to produce either X-Y or Y-X; see section 3.4
below.2
Head movement (sometimes referred to as X0 -movement) has been exam-
ined in an extensive syntactic literature from the 1980s onwards; see, for ex-
ample, Koopman (1984), Travis (1984), Chomsky (1986), and Baker (1988),
as well as the notion of “Merger” from Marantz (1984). For the purposes of
this work, it will be assumed that head movement is a syntactic operation.
Not too much should be read into this assumption; it is possible to formulate
the main results of this book in a framework that treats head movement as a
PF operation (as suggested by Chomsky 2001) and not in the narrow syntax.
Adopting a standard view, in head movement the head X of projection
XP moves upwards and adjoins to the head Y that takes XP as its sister.
Structures 61
YP
�H
�� HH
�� HH
RP Y
⇣⇣PP ��HHH
� � H
...R...
Y XP
�HH
�� H
WP X
⇣⇣PP �H
� H
...W... X RP
YP
�H
��� H HH
�� HH
RP Y
⇣⇣PP ��HH
� HH
...R... �� H
Y XP
�H
� H �HH
X Y �� H
WP X
⇣⇣PP �H
� H
...W... X RP
Y
�H
� H
X Y
These basic assumptions about head affixation suffice for many of the phe-
nomena to be analyzed below; further assumptions about the clausal struc-
tures in which complex heads are created will be introduced only as neces-
sary.
3.2.1 An Illustration
To this point it has been shown in relatively abstract terms how complex heads
are derived in phrasal structures. Here, I provide a more concrete example of
the role played by complex head structures in morphological analysis.
The examples in this section are drawn from the conjugation of verbs in
Spanish; I adapt and develop aspects of the (much more detailed) treatment
found in Arregi (1999) and Oltra-Massuet and Arregi (2005). As a first illus-
tration, consider the present and imperfect (a kind of past tense) indicative
forms of the verb hablar ‘speak’ in (5); these forms are from Latin American
Spanish, where the second and third plural are identical in form:
The forms in (5) are not segmented. A first step in the analysis of these verbs
is to develop a set of working hypotheses about how the forms are broken
Structures 63
down into morphemes, one that takes into account both morphological (i.e.
distributional) and syntacticosemantic (e.g., past tense forms have a T[+past]
morpheme) considerations.
It is clear that a number of relatively simple observations can be made
about the decomposition of the forms in (5). For instance, the imperfect forms
contain a morpheme -ba that is not seen in the present tense forms; this is
the marker of past (imperfective) tense in these verbs. It can be also seen that
almost all of the verb forms in (5) contain a theme vowel -a immediately after
the Root (recall the discussion of themes in section 2.3.5 in chapter 2), as well
as an agreement morpheme after the tense morpheme. The tense morpheme
is not realized overtly in the present tense, so with them agreement appears
after the theme vowel.
Assuming that T(ense) can be either T[-past] (present) or T[+past], and
assuming the person and number features [±1], [±2], and [±pl], the obser-
vations of the last paragraph lead to the hypothesis that Spanish contains the
morphemes in (6), which are shown with their phonological exponents:
a. Theme Vowel: -a
b. Tense Morphemes:
T[-past], -Ø
T[+past], -ba
c. Agreement Morphemes:
[+1,-2,-pl], -o; or -Ø in the imperfect
[-1,+2,-pl], -s
[-1,-2,-pl], -Ø
[+1,-2,+pl] -mos
[-1,+2,+pl], -n
[-1,-2,+pl], -n
Note that certain morphemes have no overt realization. This is the case with
T[-past], the first person Agr morpheme in the imperfect, and the third per-
son singular Agr morpheme in both tenses. These are treated as having -Ø
exponents.
64 Structures and Linear Order
Some further aspects of (6) will have to remain relatively vague at the
moment. For example, the first person singular morpheme shows allomor-
phy, with the form -o in the present, but -Ø in the imperfect. An analysis
of this effect can be given only after the Vocabulary Insertion operation has
been introduced (see chapter 4). In addition, (6c) shows two distinct -n mor-
phemes for second and third person plurals. As noted in chapter 1, though,
this identity in form appears to be systematic (a syncretism), an important
observation that needs to be accounted for in a comprehensive analysis. This
kind of identity in form is analyzed in detail in chapter 5.
Beyond the complications with allomorphy and syncretism, there is at
least one more minor deviation from the expected pattern seen in the forms
above. In particular, there is no theme vowel in the first person singular
present hablo. This effect can be treated as a phonological deletion (for clar-
ity, the deleted vowel is shown in parentheses in the segmentation given be-
low).
A segmentation of the forms in (5) that reflects the working analysis in (6)
is shown in (8); for reference, the linear order of morphemes in these forms
is given in (7):
p
(7) Linear Order: ROOT -TH-T-Agr
(8) Spanish Verbs, segmented
With respect to the primary aims of this chapter, the next question to be ad-
dressed is how the pieces identified in (8) (and, more generally, the linear
sequence shown in (7)) relate to a complex head structure of the type intro-
duced in the previous subsection. It can be assumed that the clause structure
in which the forms in (8) are derived is that in (9), which contains a Root, a
verbalizing morpheme v, and a Tense head; for purposes of illustration (and
to emphasize the fact that (9) shows a clausal structure) I have put XP and
WP phrases in the TP and vP specifier positions:
Structures 65
TP
�H
�� HH
XP T
�H
�� HH
T vP
�H
�� HH
WP v
�HH
�� H p
v P
⇣P
⇣⇣ PP
⇣p P
... ROOT...
In this structure,
pthe Root undergoes head movement and adjoins to v; sub-
sequently, the [ ROOT v] complex itself is moved and adjoined to T. These
movements result in the complex head in (10):
T
��HH
v T
p ��HH
ROOT v
On the assumption that the T head is specified as [±past], (10) produces ei-
ther present tense verbs ([-past]) or imperfective past tense verbs ([+past]).
One further addition to (10) is required beyond this: an Agr (agreement) mor-
pheme. I will assume that this type of morpheme is added to structures like
(10) at PF, prior to Vocabulary Insertion, in order to meet language-specific
well-formedness requirements.3 Morphemes added at PF are called disso-
ciated morphemes (or sometimes ornamental morphemes; see e.g. Embick
(1997) and Embick and Noyer (2007)).
With the addition of Agr, which is adjoined to T, the complex head is as
follows:
66 Structures and Linear Order
T
H
�� H
T Agr
��HH
v T
p ��HH
ROOT v
T
�H
� �� HH
H
�� HH
T Agr[+1,-2,+pl, -mos]
��HHH
�� H
v T[+past,-ba]
�H
� HH
p �
H ABL [v,-a]
With different feature values for Tense and for the Agr node, the structure in
(12) underlies all of the forms shown above.
In the case studies that are presented in this book it will be assumed that
complex head structures are created by an affixation operation that creates
adjunction structures of the type employed above. In the typical case, such as
in the analysis of the Spanish verb forms above, it will be assumed that the op-
eration that derives the complex heads is head movement. While head move-
ment is often implicated in the formation of complex heads, other operations
Some Definitions: M-Words and Subwords 67
In the theory of affixation, there are privileged structural objects that are im-
portant for the central concerns of this book. These objects can be defined and
illustrated with reference to the structure in (13), where a Root is adjoined to a
head X, followed by subsequent movement to Y . This is the kind of structure
that is typical of head movement, as discussed in the preceding section:
(13) Structure
Y (P)
�H
� �� HH H
� H
Y XP
��HH �H
�� HH
X Y �� HH
p ��HH WP X
ROOT X �H
�� HH
X ZP
p ��HH ⇣P
ROOT X ...
Embick and Noyer (2001) and Embick (2007b) argue that two objects in (13)
68 Structures and Linear Order
are special, in the sense that they are specifically referred to by PF rules.
These are defined in (14):
(14) Definitions
a. M-Word: (Potentially complex) head not dominated by a further
head-projection.
b. Subword: A terminal node; thus, a morpheme in the sense of chap-
ter 2 (either a functional morpheme, or a Root).
Illustrating
p with reference to (13), the complex head (15) that contains the
ROOT, X, and Y is an M-Word:
Y
��HH
X Y
p ��HH
ROOT X
Within this object, the terminal nodes—the Root, X, and Y —are Subwords.
According to (14b), a Subword is a terminal. Subwords and syntactic ter-
minals (i.e., morphemes) are equivalent in the normal case. The restriction
to “normal” here reflects the fact that if some morphemes are added at PF
(dissociated morphemes; recall the discussion of Agr(eement) in the Spanish
example above), then there are some morphemes at PF that are Subwords, but
not on the list of Syntactic Terminals.
Moving past the concerns of the syntax proper, Subwords are central to
the theory of the morpheme for two reasons: first, because Subwords and
morphemes are equivalent in most cases; and, second, because Subwords are
the target of the Vocabulary Insertion operation. This computation is defined
to provide phonological content to terminal nodes, and terminal nodes only:
that is, Subwords, in the sense of (14b). This is a crucial claim of the theory
presented in this book: it allows no Vocabulary Insertion at “intermediate”
nodes, at the M-Word level, or at any of the other phrasal nodes that are repre-
sented in syntactic structures. The idea that Vocabulary Insertion is restricted
to apply to terminals in this way is developed in chapter 4.
The M-Word corresponds directly to the notion of “complex head” em-
ployed above in section 3.2; in syntactic terms, it is equivalent to the H0max
Some Definitions: M-Words and Subwords 69
XP
��HH
X YP
��HH
Y ZP
�H
� H
Z WP
A first application of head movement in this structure adjoins Z to Y:
XP
�HH
�� H
X YP
�H
�� HH
Y ZP
�H
� H �H
� H
Z Y Z WP
Subsequently, the (complex) Y may adjoin to X to yield a complex head
containing X, Y, and Z:
XP
�H
�� HH
� H
X YP
��HH �H
� H
� H
Y X Y ZP
�H
� H �H
� H �H
� H
Z Y Z Y Z WP
In standard versions of head movement, only the complex head [Y Z Y] may
adjoin to X as shown in (18); it is not possible to move e.g. Z alone out of
70 Structures and Linear Order
the complex head [Y Z Y] and head-adjoin it to X.5 Thus, the objects that are
moved by head movement are M-Words in the sense defined in (14).
In addition to playing an important role in the composition of complex
objects, the M-Word is also directly implicated in the linear relations that
are established at PF (section 3.4), as well as in the mapping to domains of
phonological interaction (section 3.5).
It is often the case that the definitions of M-Word and Subword do not
overlap. For example, in the complex head used above to illustrate these no-
tions (the one based on (13)), there is no Subword that is also an M-Word:
(19) M-Word in (13)
Y
��HH
X Y
p ��HH
ROOT X
For much of the discussion of this book, where emphasis is placed on the
analysis of complex heads, I will be looking at objects in which the M-Words
and Subwords are distinct, as they are in (19). It is worth noting, though, that
there are many configurations that contain objects that are simultaneously M-
Words and Subwords according to (14). For example, in the DP the cat in
(20), where a determiner head D[+def] takes a noun phrase complement, the
D[+def] head is simultaneously an M-Word and a Subword:
(20) the cat
DP
��HH
� H
D[+def] nP
⇣⇣PP
cat
D[+def] is an M-Word because it is a head that is not dominated by further
head material; and it is a Subword because it is a terminal element, i.e., a
morpheme. By virtue of meeting both of these criteria, the D[+def] head is
linearized as an M-Word (see the following section); it is also a Subword,
and is therefore a target of Vocabulary Insertion. Further illustrations of how
the M-Word/Subword distinction plays a role in PF processes can be found
in Embick (2007b).
Linear Order 71
Y
��HH
X Y
p ��HH
ROOT X
Speaking figuratively, these are impossible because they require the “cross-
ing” (tangling) of branches of the syntactic tree; Y cannot be closer to the
Root than X without crossing lines in this way.6
For the purposes of this book, it will be assumed that only No Tangling
restricts the linearization of morphemes. Thus, linearization procedures are in
principle capable of realizing a syntactic object [X Y ] as either X-Y or Y -X,
depending on the language (and in some cases, the particular morphemes) in
question.7
We are now in a position to look closely at exactly how two morphemes
X and Y are connected linearly. It is quite likely that there are a number of
Linear Order 73
different linear relations that are relevant for PF concerns in the broad sense,
depending on what particular aspect of a representation is being analyzed (see
Embick 2010b for some specific proposals concerning morphophonology).
My interest in this chapter is on one specific type of linear relation: one that
relates M-Words to M-Words, and Subwords to Subwords. The operator that
does this encodes concatenation: a binary relation of immediate precedence.
Notationally, the operator _ is used for concatenation, so that X _Y is to be
understood as X immediately precedes Y.
With reference to the structure in (21), if both X and Y are suffixes, then
the linearization procedure generates the statements in (23), in which the Root
is concatenated with X, and X with Y :
A topic that has generated a great deal of discussion in the literature on the
syntax/morphology interface is how the order of morphemes inside words
sometimes appears to “reflect” or “mirror” the order of syntactic projections
in a clause. The latter terminology is employed by Baker (1985, 1988), where
the mirroring effect is attributed to the so-called Mirror Principle.
74 Structures and Linear Order
The observation that morpheme order and syntactic structure relate to each
other in this and other systematic ways is very important, particularly for the
question of how morphemes are assembled into complex forms. As will be
shown below, effects of this type do not call for an independent principle
of the grammar per se. Rather, mirror effects—and, more generally, system-
atic connections between morphology and syntax—are an architectural con-
sequence of a theory that constructs all complex objects syntactically.
An illustration of the mirroring effect is seen in (24), which employs
Quechua examples drawn from Muysken (1981):
(24) a. Maqa-naku-ya-chi-n
beat-RECIP-DUR-CAUS-3s
‘He is causing them to beat each other.’
b. Maqa-chi-naku-rka-n
beat-CAUS-RECIP-PL-3s
‘They let someone beat each other.’
CauseP
�HH
� �� H H
CAUS RecipP
�H
� H
�� HH
RECIP VP
⇣PP
⇣⇣ P
...VERB...
b. RECIP over CAUS
RecipP
H
�� H
�� HH
RECIP CauseP
�H
� H
�� HH
CAUSE VP
⇣PP
⇣⇣ P
...VERB...
The standard version of head movement applied to (25a,b) produces the com-
plex heads shown in (26a,b) respectively; these underlie the verbs in (24a,b):
(26) a. CAUS over RECIP
�H
� H
�� HH
�H CAUS
�� HH
RECIP
��HH
VERB ...
b. RECIP over CAUS
�H
�� HH
� H
�H
� H RECIP
� H
CAUS
��HH
VERB ...
The key to understanding these examples is the order of the CAUS and
RECIP morphemes. In Quechua, each of these affixes is linearized as a suf-
76 Structures and Linear Order
fix. Thus, linearizing the tree in (26a) yields the order VERB-RECIP-CAUS,
whereas for (26b) the order VERB-CAUS-RECIP is derived. The mirror ef-
fect is this: the phrasal syntax CAUS over RECIP over VERB (25a) is realized
“morphologically” as VERB-RECIP-CAUS (26a), while RECIP over CAUS
over VERB (25b) is realized “morphologically” as VERB-CAUSE-RECIP
(26b). In each case, the linear order of morphemes in the complex heads
in (26) is the “mirror image” of the order of the containing phrasal struc-
tures in (25): i.e., when RECIP dominates CAUS, the order of morphemes on
the verb is CAUS-RECIP; and when CAUS dominates RECIP, the order is
RECIP-CAUS.
Crucially, the mirror effect is derived as a consequence of constructing
affixed words syntactially, with head movement. As discussed above, head
movement moves a head to the immediately dominating head, and then moves
the resulting complex to the next immediately dominating head, and so on.
Starting with the different structures in (25), the locality conditions on head
movement produce the different complex heads in (26), from which the dif-
ferent morpheme orders follow.
Mirror orders are expected given how head movement works. But they are
only part of the picture. In a theory that linearizes complex heads in a way
that is contrained by No Tangling, transparent relations between syntax and
morphology may also be found in scenarios that do not involve strict mir-
roring. That is, whether or not morpheme order actually mirrors the order
of syntactic heads and projections depends on how the relevant morphemes
are linearized within the M-Word; and this is something that is (by hypoth-
esis) subject to cross-linguistic and perhaps (within a particular language)
morpheme-specific variation. Thus, mirror orders are found in one possible
way of linearizing a complex head, but they are not guaranteed to occur.
There are other ways of realizing complex heads that do not involve strict
mirror orders, but which are just as transparent as far as syntax/morphology
connections are concerned. p
Recall from section 3.4.1 that in a complex head like [[ ROOT X] Y ],
either X or Y could be linearizedp either as a prefix
p or a suffix. Thus, it is
possible to derive the orders X- ROOT-Y or Y - ROOT-X—neither of these
show a mirror effect, but both comply with No Tangling.
Linear Order 77
ZP
�H
�� HH
Z YP
�H
�� HH
Y XP
�H
�� HH p
X P
⇣P
p⇣ PP
⇣
ROOT
If the Root undergoes head movement to X, Y, and then Z, the resulting com-
plex head (i.e., the complex head that is found at the Z position in the tree)
has the structure in (28):
(28) Complex Head Formed in (27)
Z
��HH
Y Z
��HH
X Y
p ��HH
ROOT X
p
In terms of concatenation, a suffixal realization of (28) with ROOT -X-Y -Z
is produced with the following statements:
p
(29) ROOT_ X, X _Y , Y _ Z
The mirror effect is in evidence in (29), since the morpheme order reflects
the syntactic structure of the clause.
Thus far, the abstract example with X-Y -Z is, effectively, a review of what
is illustrated in the Quechua examples in the preceding subsection. The move
beyond strict “mirroring” is based on the observation that there are other pos-
sible ways of linearizing structures like (28). In particular, the linear orders
that comply with No Tangling are those in (30):
78 Structures and Linear Order
(31) a. Syntacticosemantic:
The mismatch is found in the fact that the morphemes in words like unhappier
appear to organized in different ways when different parts of the grammar are
considered. Sproat and Marantz argue that the mismatch comes about because
of the way in which hierarchical and linear relations differ. In particular, they
argue that linear relations allow for free rebracketing, a position that has been
explored in different directions in later work.9
Moving past bracketing, there are also apparent mismatches that impli-
cate morpheme order. These are directly relevant to the discussion of sec-
tions 3.4.2 and 3.4.3 above, because they have been argued to be exceptions
to (or to violate) the Mirror Principle. One argument to this effect is found
in Hyman (2003), and is based on data from the Bantu language Chichewa.
Chichewa has both causative (CAUS) and applicative (APPL) morphemes.
According to Hyman’s analysis, it is possible syntacticosemantically to form
both the applicative of a causative, and the causative of an applicative (re-
call the reasoning that was employed in the discussion of Quechua in section
3.4.3). However, these two different interpretations are both realized with the
same order of morphemes, CAUS-APPL its-il (Hyman 2003:248 and Hyman
and Mchombo 1992; in these examples PROG is for “progressive”, and FV
is for “final vowel”):
The example in (32b) is a prima facie mismatch. If the most obvious syn-
tacticosemantic analysis is correct (i.e. the analysis in which CAUS is above
APPL), the morpheme order in should be APPL-CAUS, but it is not.
With respect to this type of mismatch in general, there are two main
possibilities to be considered. One possibility is that the syntacticoseman-
tic analysis that predicts the non-occurring morpheme order is simply false,
and that the correct analysis does not produce a mismatch. For instance,
Pylkkänen (2002) suggests that the “fixed” ordering of CAUS and APPL
in Bantu has a syntacticosemantic basis. Because it maintains the simplest
interaction between syntax and morphology, this first option represents the
null hypothesis. As a general rule, it should thus be asked for any particular
putative mismatch whether the mismatch is only apparent, and derivative of
an incorrect synatico-semantic analysis. This might seem like it is placing a
rather stringent set of requirements on morphological theory broadly speak-
ing, since it requires close attention to syntactic and semantic detail; but this
is a strength of the approach, not a weakness.
The second option is that the syntacticosemantic analysis is indeed cor-
rect, so that some part of the morphophonological representation is in fact
distinct from the representation that is required for syntacticosemantic pur-
poses. With much prior work in this area, I assume that true mismatches like
this are the result of a small set of PF operations that apply to the output of
the syntactic derivation to alter it in limited ways.
One of the primary tasks of this part of the syntax/morphology interface
is to identify the set of PF operations that are responsible for these deviations
from the nondefault case (see Embick and Noyer 2007 for an overview). This
research program is a continuation of early work on the relationship between
syntax and phonology like Chomsky and Halle (1968), where questions about
mismatches are couched in terms of there being divergent notions of “surface
structure” (one syntactic, one phonological) that are related to each other by
rule. The nature of such rules is not at the center of Chomsky and Halle
(1968), though, and it is only later that a broad range of syntax/phonology
interactions are analyzed in their own right, in a large literature on prosodic
phonology. Closer to the focus of this chapter, the general theme of how to
constrain the divergences between the representation that is the output of
the syntax and the representation(s) relevant for PF computation is a pri-
mary component of Marantz’s (1984, 1988) work, which looks specifically
at rebracketing and affixation operations in this light. Within the framework
adopted in this book, it has been proposed that many mismatches are the re-
82 Structures and Linear Order
sult of PF movement operations; see Embick and Noyer (2001, 2007) and
Embick (2007b) for overviews.
The main theme that unifies the approaches to mismatches cited above is
the idea that syntactic structure and morphological structure are the same
in the default case, with the exceptions to this general pattern calling for
a restricted set of rules driven by language-specific PF requirements. This
type of theory allows for systematic investigation of syntax/PF mismatches,
while maintaining Interface Transparency in its strongest form possible. In-
deed, a main research intuition behind Distributed Morphology is that while
mismatches do exist, they do not compromise the general syntactic orienta-
tion of the theory, since the very fact that certain phenomena can be isolated
as mismatches in the first place testifies to the robust transparency of syn-
tax/morphology interactions more generally.
A final point for this chapter concerns the relationship between the mor-
phemes and syntactic structures that are at the center of the theory on the
one hand, and the intuitive notion of the word on the other.
In a theory that derives all complex objects syntactically, the term word is
often used informally to pick out a number of objects that have a shared set
of properties, properties that have some more or less direct relationship to the
notion phonological word. Technically speaking, whether or not the phono-
logical word is needed as part of the theory—i.e., whether it has ontologi-
cal status in the approach, and is not just a convenient descriptive term—is
an open question in phonological theory. However, for the purposes of the
present discussion, I will assume that there is a notion of “word-level phonol-
ogy” that is important for sound structure, because this will allow us to see
how phonological domains relate to structural notions, i.e., to objects like
Subwords and M-Words.10
The clearest connection to pursue along these lines identifies the M-Word
as an object of phonological interest. M-Words almost invariably show the
“close” types of phonological interaction that are associated with phonologi-
cal wordhood. As a working hypothesis, it appears that the M-Word is special
for phonological interaction, in the sense that it seems to correlate in the de-
fault case with the domain in which the word-level phonology takes place.
Part of this special status could be analyzed by taking the M-Word to be
A Note on Structure and Phonological Wordhood 83
where the non-cyclic phonology applies, for example (see Embick (2010b)
for some preliminary discussion).11
The M-Word/phonological word correspondence is a good starting point
for understanding how the informal notion of “word” can be understood struc-
turally. There are many additional questions that arise when both smaller and
larger phonological domains are considered.
In the case of smaller objects, asymmetries in the morphophonological be-
havior of certain (classes of) affixes, such as those associated with “Level 1”
and “Level 2” affixes in English could in principle be analyzed in a number
of different ways. For instance, this kind of difference might reflect impor-
tant structural differences in how Subwords attach to their hosts, or perhaps
whether certain morphemes are phase-defining or not; or it could be the case
that phonological behavior like this results from individual exponents being
specified as “cyclic” or “non-cyclic” in the phonological sense (cf. Halle and
Vergnaud (1987)).
In the case of objects that are larger than the M-Word, questions about
the trade-off between structures, morpheme-specific behavior, and locality
domains are salient in a large literature on the phonology of cliticization (see
e.g. Poser (1985), Booij and Rubach (1987), Hayes (1990), Halpern (1992),
Odden (1993), and Embick (1995) for some different perspectives on this
and some closely related matters). Beyond cliticization, the same sorts of
questions about how structure relates to domains of phonological interaction
arises for objects of larger sizes as well, in the area of research typically
referred to as prosodic phonology, or the syntax/phonology interface. See
Pak (2008) for an overview of the major research programs in this domain.
Programmatically, the idea at the center of this section is that domains of
phonological interaction (whether the phonological word, or some other do-
main) are not primitives of the theory; rather, they are to be understood in
terms of priviliged structural objects (as in the M-Word/phonological word
hypothesis), or in terms of other parts of the theory (phase theory, the prop-
erties of exponents, and so on). For a sample of work looking more directly
at some of the relevant phonological interactions in this framework, see e.g.
Marvin (2002), Oltra-Massuet and Arregi (2005), Newell (2008), and Embick
(2010b) for “word-internal” investigations, and for larger syntactic objects,
Wagner (2005) and Pak (2008).
84 Structures and Linear Order
3.6 Conclusions
The theory that is developed in this book is centered on two types of mor-
phemes: Roots and functional morphemes. Functional morphemes are bun-
dles of synsem features that do not possess phonological features underly-
ingly. Instead, they receive their phonological form in the PF component of
the grammar, via the Vocabulary Insertion operation. This chapter begins a
detailed examination of Vocabulary Insertion; this focus extends into chap-
ters 5 and 6, where some additional topics—including the phenomenon of
syncretism and its theoretical analysis—are examined.
As a starting point, it is convenient to begin with a review of some of the
central points about morphemes and their phonological forms that are intro-
duced in chapters 1 and 2. Recall that the grammar of every language contains
a list that is referred to as the Vocabulary of that language. The members of
this list are individual Vocabulary Items. According to the view that I will
develop here, Vocabulary Items are objects, objects that are accessed and ac-
tivated in the Vocabulary Insertion process.1
By definition, a Vocabulary Item is a pairing between a phonological expo-
nent and a set of synsem features that determine the privileges of occurrence
of that exponent. In essence, the synsem feature specification determines the
morphemes to which a Vocabulary Item could in principle apply. A Vocabu-
lary Item is schematized in (1), with its individual subparts labelled:
(1) Vocabulary Item
[ab g] $ /X/
|{z}
| {z }
synsem features phonological exponent
So, for example, the familiar past tense -ed of English is the exponent of the
Vocabulary Item in (2):
(2) T[+past] $ -ed
The effect of this Vocabulary Item is to give the phonological form -ed to the
node T[+past]. So, for example, if the T[+past] node is combined with the
86 Vocabulary Insertion: An Introduction
p
Root P LAY and the categorizing head v, the result is a “past tense verb”, as
shown prior to Vocabulary Insertion in (3):
�H
�� HH
T[+past]
p ��HH
P LAY v
The Vocabulary Insertion operation applies to the terminal nodes in the struc-
ture in (3), and inserts the phonological exponent -ed into the position of the
Tense node; the result is shown in (4), which also assumes that a -Ø exponent
is inserted into the v head:
��HH
�� HH
�H
� H T[+past,-ed]
p � H
P LAY [v,-Ø]
Ultimately, morphemes relate sound (or sign) and meaning. For functional
morphemes, sound and meaning come together via Vocabulary Insertion.
Thus, in a comprehensive theory of how phonological and synsem features
are related by this operation, two basic sets of questions must be addressed.
The first type of question concerns the “insertion” part of Vocabulary In-
sertion, and asks what it means for a phonological exponent to be added to a
terminal node. Specifically, there must be a theory of what happens when an
object from the Vocabulary (a Vocabulary Item, which pairs synsem features
with a phonological exponent) applies to a functional morpheme to yield a
node that has a phonological representation.
The second question concerns the synsem features that are referred to in
the insertion process, and what happens to such features when Vocabulary
Insertion applies to the morpheme that possesses them. The question here is
whether Vocabulary Insertion affects synsem features when phonological fea-
tures are added (for example, by deleting them), or whether synsem features
are unaffected by insertion, making them therefore potentially visible to sub-
sequent computations. These questions are addressed in turn in the following
subsections.
In the way that Vocabulary Insertion has been described to this point, the
Vocabulary Item (5) applies to the node T[+past] with regular verbs like play:
�H �H
�� HH �� HH
� H
T[+past] T[+past,-ed]
p ��HH p �
�H
� H H
P LAY v P LAY [v,-Ø]
The Vocabulary Insertion Process in Detail 89
While the process sketched in (5)-(7) is accurate in its gross contours, many
things remain to be said about the specifics of phonological instantiation.
To begin with, there are several different ways in which the procedure
for adding phonological exponents to nodes can be formalized. As shown
in (6) and (7), the process of Vocabulary Insertion is additive, in the sense
that it introduces a new element—a phonological matrix—to the morpheme
T[+past]; that is, it augments the T[+past] morpheme with material that it
does not possess underlyingly.
Another possibility for the analysis of phonological realization, which is
adopted here, is that Vocabulary Insertion is replacive. Halle (1990), for ex-
ample, develops an approach that treats phonological realization in this way.
In this theory, certain morphemes possess a place-holder Q as part of their
underlying representation, rather than an actual phonological exponent. Thus,
for example, the morpheme T[+past] is represented as T[+past,Q] prior to in-
sertion. In the version of the replacement view that I will propose here, the Q
element functions as a variable, such that the effect of Vocabulary Insertion
is to replace Q with a phonological exponent, which can be seen as the value
of the Q variable.
With this general picture in mind, the pre-insertion and post-insertion
stages of the past tense example are as follows:
T T
�H
� H ��HHH
�� HH
�� H
H T[+past,Q] T[+past,-ed]
p �� H p �
�H
� HH
P LAY [v,Q] P LAY [v,-Ø]
[a,Q]
phemes are often bundles of multiple features, so that, for example, [a,b ,Q]
is a possible functional morpheme (recall chapter 2). Crucially, though, each
functional morpheme contains only a single Q element. This latter assump-
tion has important consequences which will be seen as the discussion pro-
ceeds.
As mentioned earlier, the Q element can be treated as a variable. The
phonological realization part of Vocabulary Insertion can then be treated as
the substitution of a free variable, in which the phonological exponent of the
Vocabulary Item replaces Q. Using the notation [Q/X] for “phonological form
/X/ is substituted for Q” in the expression to its left, this part of the Vocabu-
lary Insertion process is shown in detail in (11) for the T[+past] example:
Morphemes like the one shown in (12) function as terminal nodes in syntactic
derivations, in the way described in the earlier chapters of this book. The
features a and b are, by hypothesis, features that are interpreted semantically.
Syntactic derivations containing morphemes like the one in (12) are spelled
out, such that structures containing (12) are sent to the PF and LF interface
components.
The synsem features of a morpheme are visible to computations at each
interface. On the meaning side, features like a and b play a role in the pro-
cedure that constructs semantic representations from syntactic objects. In the
PF component of the grammar, these features are visible to Vocabulary Inser-
tion, and they determine which Vocabulary Item applies to give a morpheme
its phonological form.
Having synsem features visible to the insertion process is a bare mini-
mum. Beyond this, there is a further question about what happens to synsem
features when Vocabulary Insertion occurs. More precisely, there are two op-
tions to consider here. The first is that synsem features are still present in a
morpheme, after Vocabulary Insertion has applied. The second is that (at least
some) synsem features are deleted when Vocabulary Insertion occurs.
In the representations that are employed in section 4.2.1 above, I adopted
the first of these two views: Vocabulary Insertion adds phonological features
to a morpheme, but it does not automatically delete that morpheme’s synsem
features when it does so. In the treatment of the past tense of played, for
example, insertion of the -ed exponent does not trigger the deletion of [+past]:
�H
�� HH
� H
�H
� H T[+past,-ed]
p � H
P LAY [v,-Ø]
92 Vocabulary Insertion: An Introduction
The non-deletion view will be adopted as the default case in the pages to
come. The qualification to the default case reflects the fact that across-the-
board non-deletion is probably an oversimplification; it is possible, for ex-
ample, that there are some complex phenomena for which Vocabulary-driven
deletion is required, as discussed in Noyer (1997) or Bobaljik (2000). How-
ever, for typical cases of Vocabulary Insertion non-deletion appears to be ex-
actly what is needed; some empirical illustrations of this point are advanced
in section 4.6.3 below.3
In the analysis of English past tense forms like played that is developed in
the preceding section, only one Vocabulary Item is considered: the one that
inserts -ed. However, it is often the case that there are multiple Vocabulary
Items that could in principle apply to a particular functional morpheme, such
that the winner is determined by a morpheme local to the morpheme under-
going insertion. This results in contextual allomorphy. For example, it can be
seen in (14) that the T[+past] morpheme in English has -t and -Ø phonologi-
cal forms in addition to the regular or default -ed:4
Based on the facts in (14), it is clear that the Vocabulary of English must
contain Vocabulary Items with -t and -Ø exponents in addition to the one
with the -ed exponent. Provisional versions of these Vocabulary Items are
shown in (15):
a. T[+past] $ -ed
b. T[+past] $ -t
c. T[+past] $ -Ø
Ordering of Vocabulary Items 93
The reason that the Vocabulary Items in (15) are provisional is that, as things
stand, they do not make correct predictions about the derivation of any partic-
ular past tense forms. Specifically, there is nothing in the Vocabulary Items in
(15) that ensures that the past tense form of leave is lef-t (and not *leav-ed),
while the past tense form of play is play-ed (and not *play-Ø), and so on.
Clearly, then, the Vocabulary Items in (15) need to be modified in a way that
produces the correct distribution of exponents.
The modifications that are required are of two types. The first type is rep-
resentational: the Vocabulary Items in (15) must be augmented so that the
non-defaults (those with the -t and -Ø exponents) apply only in the context of
some verbs, and not across the board. This additional information is shown in
the revised Vocabulary Items in (16a,b), where the notation / X means “in
the context of X”:
Consider, for concreteness, the derivation of left. The structure that is the
input to Vocabulary Insertion is shown in (17):
(17) left
�H
�� HH
T[+past]
p ��HH
L EAVE v
discussion, ordering must ensure that p the Vocabulary Item with -t wins out
over the one with -ed in the context of L EAVE.
A hypothesis that has been adopted in much work is that order among
Vocabulary Items is defined by the familiar principle that specificity deter-
mines order of application, such that the Vocabulary Item most specified for
insertion at a particular node applies in favor of less fully specified competi-
tors. One way of ordering Vocabulary Items along these lines is given in the
Subset Principle in (19), which also makes precise a version of the “compat-
ibility” conditions that are alluded to in the text under (17):
With reference to the derivation of left, it is clear that the Vocabulary Item
with the -t exponent is more specific than the Vocabulary Item with -ed. By
convention, more specified Vocabulary Items appear higher on lists than less
specified ones, as shown in (20):
The Vocabulary Item with -Ø is also more specific than the -ed one, and
is therefore ordered before it. Note, however, that the -t and -Ø Vocabulary
Items are equally specific, in the sense that each refers to T[+past] and has
a contextual condition. Because of this, one of them could not actually beat
the other. However, this is irrelevant, since their lists are disjoint, so for any
given verb, one will win. For notational convenience, in specificity “ties” of
this type, one of the two Vocabulary Items will be listed above the other by
fiat.
With the Vocabulary Items for T[+past] in English ordered as in (20), the
analysis of leave⇠lef-t/*leav-ed can be completed. When the T[+past] mor-
96 Vocabulary Insertion: An Introduction
pheme in (17) is worked on, the Vocabulary is searched, and the most spe-
cific item that can apply to this node is the one that inserts the exponent -t.
This derives lef-t. Thus, we can see form *leav-ed is not derived,pbecause the
Vocabulary Item with -t beats the one with -ed in the context of L EAVE.
The idea that Vocabulary Items are ordered and competing for application
is an essential one, and its consequences arise at numerous places in this book,
starting in section 4.4 immediately below.
With respect to the general theory of contextual allomorphy, a number of
important questions remain to be addressed. Of these, two in particular come
to the forefront. The first concerns what form a particular Vocabulary Item
could take; i.e., what kind of information could be included in the contextual
specifications that determine when a particular Vocabulary Item applies. The
second type of question concerns locality, and the conditions under which one
morpheme can see another for allomorphic purposes. These two questions are
examined in detail in chapter 7.
As a final point, it should be noted that while Ordering plays a partic-
ularly important role in the analysis of contextual allomorphy above, it is
important for other phenomena as well. As will be shown below in chapter 5,
the Vocabulary of a language will sometimes contain two Vocabulary Items
of the type shown in (21), where a and b are synsem features:
(21) a. [a,b ] $ -x
b. [a] $ -y
In an informal way of speaking, blocking effects are found when (i) one form
(or several) that might be expected to occur in a given context is not in fact
Vocabulary Insertion and Blocking: A First Look 97
found, and (ii) this failure can be related to the fact that some other form ap-
pears to take precedence over the non-existing one. The relevance of blocking
to the past tense examples can bepseen by observing that the ungrammaticality
of the regular past *leav-ed of L EAVE correlates with the grammaticality
of lef-t.
Part of the explanation for why lef-t is grammatical while *leav-ed is not
is implicated in the discussion of competition for insertion in the preceding
section. Consider the structure in (22), along with the Vocabulary Items in
(23):
(22) left
�HH
�� H
�H T[+past,Q]
p � H
L EAVE v
(23) Vocabulary Items for T[+past],
p p
a. T[+past] $ -t/{ B END, L EAVE,...}
p p
b. T[+past] $ -Ø/{ H IT, Q UIT,...}
c. T[+past] $ -ed
By the Subset Principle (19), the most specific Vocabulary Item that can apply
to T[+past] in (22) is the one with the -t exponent. This Vocabulary Item gives
phonological content to the T[+past] node, producing lef-t.
There are two aspects of this analysis to consider further with respect to
the notion of blocking. The first concerns the specifics of how the insertion
of -t prevents the insertion of -ed; or, more precisely, how the application
of the Vocabulary Item with the -t exponent prevents the application of the
Vocabulary Item with the -ed. The second concerns the general question of
what objects in the grammar Vocabulary Insertion applies to.
Beginning with the more specific question, it can be seen that Ordering
will result in the Vocabulary Item with -t beating the one with -ed, so that we
have lef-t instead of *leav-ed. But what insures that we do not see both of
the past tense exponents -t and -ed, as in *lef-t-ed? Or two -t exponents, as in
*lef-t-t, or three -ed exponents, as in *play-ed-ed-ed, and so on?
The basic answer to these questions is that these multiply-marked forms
are ungrammatical because they are not derived by the grammar. The assump-
98 Vocabulary Insertion: An Introduction
tion that is responsible for ruling the “multiply marked” forms mentioned
above is stated in (24) as Uniqueness:
In essence, Uniqueness is really just the assumption that each functional mor-
pheme possesses a single Q variable that is the target of Vocabulary Insertion.
Thus, when Vocabulary Insertion has applied once to a functional morpheme,
the process is done as far as that particular node is concerned, as there are no
further variables (i.e., Q positions) to be instantiated phonologically. Thus,
*lef-t-ed (or *play-ed-ed etc.) are not derived, because there are no mor-
phemes that the items in (23) could apply to after the T[+past] morpheme
has been realized once.
The general question posed above concerns which objects are targets for
Vocabulary Insertion. The answer is that it targets morphemes, as these are
the only objects with Q variables. The idea that the morpheme is special in
this way is stated here as Terminal Insertion:
T
�HH
�� H
v T[+past]
p ��HH
VAPOR v
On the assumption that the Root has a phonological form underlyingly, Vocabu-
lary Insertion applies first at the v node, inserting -ize:
(27) vapor-iz-ed
v
�H
�� HH
� H
v T[+past]
�H
� H
p � H
VAPOR [v,-ize]
After this, Vocabulary Insertion applies at the T[+past] node, and inserts -ed:
(28) vapor-iz-ed
v
�HH
� �� H
H
v T[+past,-ed]
�H
p �� HH
VAPOR [v,-ize]
100 Vocabulary Insertion: An Introduction
As a starting point, (29) shows the Latin verb laudāre ‘to praise’ in three
different tense forms, the present, imperfect, and perfect (all indicative):
Based on the synsem distinctions that are made in this system, there are a few
observations to be made concerning the morphemes (and features) that play a
role in the analysis of these verbs. The first is that the language distinguishes
present from past tense. On this basis we can posit a feature [±past], which is
negative in presents, and positive in imperfects. The status of [±past] in the
perfect is slightly more complicated. Based on the fact that the language also
Illustration: A Fragment from the Latin Conjugation 101
(31) a. non-perfects
H
�� H
Agr
��HH
T
p ��HH
ROOT v
102 Vocabulary Insertion: An Introduction
b. perfects
�HH
�� H
�H
� H Agr
�� HH
�H T
�� HH
Asp[perf]
p ��HH
ROOT v
The forms that are segmented in (30) and the structures in (31) can be related
to each other in a few steps. Beginning with v, Asp[perf], and T, a first set of
exponents is as follows:
a. v: -ā
b. Asp[perf]: -vi
c. Tense:
i. Present, Perfect: -Ø (although see below)
ii. Imperfect: -bā
For Agreement, the patterns are slightly more complicated. While some agree-
ment morphemes are realized only in one form, namely those in (33a), the
agreement morphemes in (33b) show contextual allomorphy:
a. No Allomorphy
i. 3s: -t
ii. 1p: -mus
b. Allomorphy:
i. 1s: -ō, -m, -ı̄
ii. 2s: -s, -stı̄
iii. 2p: -tis, -stis
iv. 3p: -nt, -ērunt
Illustration: A Fragment from the Latin Conjugation 103
pluperfect
1s laud-ā-ve-ra-m
2s laud-ā-ve-rā-s
3s laud-ā-ve-ra-t
1p laud-ā-ve-rā-mus
2p laud-ā-ve-rā-tis
3p laud-ā-ve-ra-nt
These forms introduce some additional facts that can be incorporated into the
analysis. Working from the inside out, it can be seen that the Asp[perf] head is
-vi, as it is in the perfects above in (30) (the vowel lowering that changes /vi/
to /ve/ can be assumed to be phonological; see e.g. Leumann et al. (1963)).
In the position of Tense, the exponent -rā appears. Finally, outside of Tense,
the Agreement morphemes are realized with the same exponents that appear
in the Imperfect, as can be seen by comparing (34) with the imperfects in
(30). Based on the last of these observations, it appears that insertion of the
-m allomorph of first person singular agreement in (33b) is conditioned by
T[+past].
Working outwards from the Root, for the theme vowel and Asp[perf], the sit-
uation is simple given what has been examined above, as there is one Vocabu-
lary Item for each of these:
104 Vocabulary Insertion: An Introduction
In point of fact, this part of the analysis is simple only because I have re-
stricted attention to one verb. When additional verbs are taken into account,
it can be seen that there are more theme vowels in the language. In addition,
there is contextual allomorphy for Asp[perf] with other verbs, a point which
is examined in section 4.6.3 below.
For the Tense head, it was observed above that the T[+past] forms, the Im-
perfect and the Pluperfect, show distinct allomorphs: -bā and -rā respectively.
An analysis of this effect is presented in (37), where the first Vocabulary Item
makes reference to the presence of Asp[perf]:
T[+past] $ -rā/Asp[perf]
T[+past] $ -bā
In the case of T[-past], which is found with Presents and Perfects, there is no
overt realization of tense. Rather than treating this as the insertion of a -Ø, I
will assume that this results from the deletion of T[-past], which is effected
by the rule (38):7
(38) T[-past] −! Ø
In part, this deletion rule is motivated by the fact that present tense T[-past]
does not show an overt morphological realization in Latin verbs. Further mo-
tivation for deleting T[-past] in this way can be seen in the contextual sensi-
tivities of Agr, which will become clear shortly when agreement allomorphy
in the perfect is analyzed.
Turning to agreement, a first instance of allomorphy is found in the first
person singular, where there are both -m and -ō realizations in the non-perfects.
The former appears in the [+past] tenses, suggesting the following Vocabu-
lary Items:8
All of the other cases of allomorphy of Agr seen in (33) share one property:
a special allomorph is inserted in Perfect forms. More precisely, the special
Illustration: A Fragment from the Latin Conjugation 105
forms appear in perfects, but not in pluperfects. This pattern can be seen
clearly when all of the forms under discussion are considered together as
in (40), where the agreement endings that are sensitive to the Asp[perf] are
shown in bold:
A straightforward way of accounting for the special forms seen in the perfect
is by making the boldfaced allomorphs refer contextually to Asp[perf]. Recall
that the structure of the perfect was assumed above to be the one shown in
(41), with a T[-past] node:
��HH
�� HH
� H Agr
�� HH
� H
��HH T[-past]
� H
Asp[perf]
p ��HH
ROOT v
In (38) it was proposed that T[-pres] is deleted. In the case of the perfect
forms, an effect of the deletion rule when it applies to (41) is that Agr is lin-
early adjacent to (i.e. concatenated with) Asp[perf], which makes the contex-
tual condition local in a way that will be discussed in chapter 7. The special
allomorphs in (40) can then be specified as shown in (42), which also includes
the first person singular -m and the rest of the agreement Vocabulary Items
required for Latin:
106 Vocabulary Insertion: An Introduction
The analysis in (42) illustrates the points about ordering of Vocabulary Items
and competition for insertion that have been emphasized throughout this chap-
ter. The more specific items—i.e., those that are contextually specified for
reference to Asp[perf]—take precedence over the less specified items in per-
fects. Thus, for example, the second person singular agreement morpheme
in the perfect (43) is realized as -stı̄, not as -s; (43) shows the structure with
the T[-past] node deleted, and with the exponents in positions that they are
realized in:
��HHH
� �� HH
� H
�H Agr[-1,+2,-pl,-stı̄]
� �� HH H
�H Asp[perf,-vi]
p � H
L AUD [v,-ā]
In this structure, the Vocabulary Item with -stı̄ for [-1,+2,-pl] beats the one
with -s because of the contextual specification that refers to Asp[perf]. In the
pluperfect, though, the Agr morpheme is next to the T[+past] head, and -s is
realized for second singular, not -stı̄, as shown in (44):
Illustration: A Fragment from the Latin Conjugation 107
�H
�� HH
�� HH
�� H H
�H Agr[-1,+2,-pl,-s]
� � HH
H
�
�� HH
� H
� HH T[+past,-rā]
�� H
�H Asp[perf,-vi]
p � H
L AUD [v,-ā]
That is, because Agr[-1,+2,-pl] does not “see” Asp[perf] (in a way to be
made precise in chapter 7), the Vocabulary Item that inserts -stı̄ cannot be
employed. Instead, the most specified item that can apply to this node is the
one that inserts -s. As can be seen in (42), this is the default Vocabulary Item
for second singular.
With the analysis of section 4.6.2 at hand, an additional point may be made
concerning the status of synsem features after Vocabulary Insertion. As a
starting point, recall that (as per section 4.5) Vocabulary Insertion applies
to the structures like (43) and (44) from the inside out; that is, first to the v
position, then to Aspect and Tense when they are present, and finally to Agr.
While this way of ordering Vocabulary Insertion is not crucial for the points
emphasized immediately above in 4.6.2, it relates directly to the assumption
made in 4.2.2 above, to the effect that synsem features are not automatically
deleted when Vocabulary Insertion takes place.
As a first example of this point, consider the realization of T[+past], which
is either -rā (when it occurs in the context of Asp[perf]) or -bā (in the Imper-
fect). The T[+past] head, which is realized in either of these two ways, has
in turn an effect on the realization of Agr: Agr[+1,-2,-pl] is realized as -m in
the context of T[+past], otherwise as -ō (see (39)). When these observations
are put together, the importance of non-deletion of features becomes clear.
On the assumption that Vocabulary Insertion proceeds from the inside out,
as specified in section 4.5, the Agr node undergoes insertion after insertion
takes place at the Tense node. If features are not deleted when Vocabulary In-
108 Vocabulary Insertion: An Introduction
a. amā-vi-ı̄ ‘love’
b. scrip-si-ı̄ ‘write’
c. vēn-i-ı̄ ‘come’
Of these forms, the -vi allomorph is the default; the other two have to be
contextually specified for lists of Roots that they are realized with (Embick
2000, 2010a).
Crucially, the same pattern of Agr allomorphy is found with all of these
different allomorphs of Asp[perf]. Again, this pattern can be accounted for
in a direct way if the relevant Vocabulary Items make reference to Asp[perf],
as is the case in (42) above. However, if synsem features like Asp[perf] were
deleted by insertion, the generalization about the distribution of the special
perfect allomorphs would have to be stated in another, less general way.
In summary, the two phenomena just considered provide important illus-
trations of what kinds of information are available when Vocabulary Inser-
tion occurs. In particular, it appears that a simper analysis of allomorphy pat-
Summary 109
terns is possible when it is assumed that synsem features are not obligatorily
deleted when Vocabulary Insertion occurs.
4.7 Summary
With this overview of Vocabulary Insertion now complete, the next chap-
ter turns to an examination of syncretism, the phenomenon that provides
the strongest motivation for treating sound/meaning connections in a real-
izational way.9
Chapter 5
Syncretism and (Under)specification
5.1 Introduction
[a,/X/]
112 Syncretism and (Under)specification
and third), and two numbers (singular and plural), there are not six distinct ex-
ponents found for the Agreement morpheme that is found on verbs. Instead,
the second and third person plurals are realized with the same exponent, -n:
p/n form
1s habl-o
2s habla-s
3s habla-Ø
1p habla-mos
2p habla-n
3p habla-n
Two main claims of the current approach (along with other realizational theo-
ries) are (i) that the identity in form seen in second and third person agreement
in (2) is systematic, not accidental; and (ii) that this systematicity must be en-
coded in the grammar, in a way that rules out the use of traditional morphemes
like (1). In terms of the maximization of generalizations (and minimization
of the Vocabulary), this means that the second and third person plural mor-
phemes in (2) have to be analyzed with a single Vocabulary Item, not with
two distinct Vocabulary Items.
The bulk of this chapter is devoted to an illustration of how Vocabulary
Insertion is used in the analysis of syncretism. Along the way, some key con-
cepts surrounding how morphemes are specified for features, and how, in
particular, Vocabulary Items may be underspecified, are introduced as well.
Taken together, this chapter develops a working analysis of syncretism that is
then applied to a number of examples in section 5.5.
Part of the motivation for defining syncretism in this way comes from the
need to distinguish systematic identities in form—which I refer to as (system-
atic) syncretisms—from cases of accidental homophony, where phonological
identity is accidental. For an example of the latter, the English (regular) plu-
ral morpheme has the phonology /z/ (dog-s), as does the third person singular
agreement morpheme (play-s). But this identity in form appears to be an ac-
cident, and analyzing the language with two different Vocabulary Items that
happen to have the phonological exponent /z/ does not miss significant gen-
eralizations (see section 5.4). On the other hand, when two feature bundles
that are similar in synsem feature content and in distribution show the same
exponent—as is the case with Spanish -n in section 5.1—generalizations
are missed when two distinct Vocabulary Items with identical exponents are
posited. A primary task of morphological theory is to give an account of the
syncretisms that is able to explain how it is that the same phonological ex-
ponent appears in more than one synsem environment in a way that is not
accidental as far as the grammar is concerned.
For present purposes it will be assumed that the lines between systematic
syncretism and accidental homophony are clear in most cases, so that empha-
sis can be placed the analysis of the former. Nevertheless, the dividing line
between syncretism and accidental homophony is an important and complex
issue. Some additional thoughts on this point are advanced in the concluding
section of chapter 6.
(3) Subjects
a. no:xoWtiW
no xw 1 W tiW
ADV 3s-OBJ 1s-SUBJ put
‘I put him down.’
Syncretism and Underspecification 115
b. s1nda
s1 n da
STAT 2s-SUBJ sit
‘You are sitting.’
c. k’y 1d1yaN
k’y 1 d1 yan
OBJ 1p-SUBJ eat
‘We are eating.’
d. Wohëc1s
W1 oh ë c1s
1s-OBJ 2p-SUBJ CAUS see
‘You-pl see me.’
(4) Objects
a. y1W1w1ëtehë
y1 W1 w1 ë teë
ADV 1s-OBJ ASP CAUS carry
‘It carries me along.’
b. n1w1ëtehë
n1 w1 ë teë
2s-OBJ ASP CAUS carry
‘It carries you-sg along.’
c. nohč’1ëca:n
noh č’1 ë can
1,2p-OBJ 3s-SUBJ CAUS see
‘He sees us/you-pl.’
116 Syncretism and (Under)specification
In the light of (4) it is important to recall that subject agreement shows distinct
morphological realizations for first person plural and second person plural
subjects: d1- versus oh-. From this contrast it can be seen that the language
does in fact distinguish first from second person in the plural. That is to say,
it is not the case that the language fails to make person distinctions in the
plural at the level of synsem features. Rather, the point is that in the object
agreement system, the distinction between first person and second person
plurals is not manifested on the surface. Instead, the single exponent noh-
appears whenever the object is either first plural or second plural.
Putting the facts in (3) and (4) together, the agreement prefixes in Hupa
can be arranged as follows:
(5) Subject and Object Exponents
subject object
1s W- W1-
2s n- n1-
1p d1- noh-
2p oh- noh-
Consider now the functional morphemes for Agreement, prior to insertion.
Hupa distinguishes first from second person, and singular from plural. The
analysis will therefore make use of the features [±1] and [±2] for person;
[±pl] for number; and, finally [+subj] and [+obj]. Regarding these features, a
few notes are in order.
First, technically speaking, only one binary feature is needed to distin-
guish first from second person. However, when third person arguments are
analyzed, it is useful to have both [±1] and [±2], so that third person can be
defined as [-1,-2]. For this reason (and for consistency with some of the other
analyses developed in this book), I will employ two features.
A second point is that the features [+subj] and [+obj] stand proxy for a
presumably more complicated feature system that is involved in representing
the relevant grammatical notions. Thus, not too much should be read into this
particular encoding.
As pointed out above, there is no incompatibility between the plural fea-
ture [+pl] and the person features that make up first and second person. Thus,
it will be assumed that plural object morphemes contain both person and
number features. The general principle that underlies this further assumption
is that functional morphemes are fully specified for synsem features; this idea
Syncretism and Underspecification 117
is made precise in section 5.3 below, where it is related to the general discus-
sion of features in chapter 2.
With reference to the table in (5), full specification of functional mor-
phemes means that verb forms in Hupa contain agreement morphemes with
the feature combinations shown in (6); (6) corresponds exactly to (5), but
shows only synsem features, not the exponents inserted by Vocabulary Inser-
tion:
subject object
1s [+1,-2,-pl,+subj] [+1,-2,-pl,+obj]
2s [-1,+2,-pl,+subj] [-1,+2,-pl,+obj]
1p [+1,-2,+pl,+subj] [+1,-2,+pl,+obj]
2p [-1,+2,+pl,+subj] [-1,+2,+pl,+obj]
The functional morphemes in (6) undergo Vocabulary Insertion in the indi-
vidual derivations in which they are present. Concentrating on the different
plural morphemes, the observation to be accounted for is as follows: while
there are distinct feature bundles at the synsem level for first and second per-
son plural objects, there is only a single phonological exponent associated
with these two different functional morphemes.
a. [+1,-2,+pl,+subj] (= d1-)
b. [-1,+2,+pl,+subj] (= oh-)
c. [+1,-2,+pl,+obj] (= noh-)
d. [-1,+2,+pl,+obj] (= noh-)
There are three different phonological forms in (7). With this in mind, con-
sider now the three Vocabulary Items in (8):
118 Syncretism and (Under)specification
a. [+1,-2,+pl,+subj] $ d1-
b. [-1,+2,+pl,+subj] $ oh-
c. [+pl,+obj] $ noh-
By (8), the first and second plural subject agreement morphemes (7a) and
(7b) are realized via distinct Vocabulary Items, (8a) and (8b) respectively.
However, realization of the [+obj] plurals is effected by a single Vocabulary
Item, (8c). The fact that this Vocabulary Item must apply with the [+obj]
morphemes follows from principles introduced in chapter 4. According to
the theory of Vocabulary Insertion developed there, neither (8a,b) can apply
to [+obj] morphemes, because they make reference to the feature [+subj].
Since neither of these Vocabulary Items in (8) can apply to (7c) or (7d), the
most specific Vocabulary Item that can apply to each of these morphemes is
(8c), which inserts noh-. Thus, the two distinct functional morphemes (7c,d)
are realized with the same Vocabulary Item (8c); this is what it means to have
a systematic syncretism in this theory.
An important component of the analysis based on (8) is that the Vocabu-
lary Item (8c) does not refer to the features [±1] or [±2]. Rather, it refers
only to the features [+pl] and [+obj]. Because it does not refer to person fea-
tures, it is able to apply to both [+1,-2] object plurals and to [-1,+2] object
plurals. In the type of situation just described, a Vocabulary Item is said to be
underspecified with respect to the feature bundles to which it applies.
The general idea behind this kind of underspecification is that Vocabulary
Items are able to apply to functional morphemes that contain a superset of the
features that are mentioned in the Vocabulary Item (see section 5.3 below). In
the particular case at hand, the effect of underspecifying the Vocabulary Item
with the noh- exponent is that it is able to apply to both (7c) and (7d).
Underspecifying Vocabulary Items allows for syncretisms to be analyzed
as systematic effects, and not as instances of accidental homophony. The
identity in form in the first and second plural object agreement morphemes is
systematic, since these two morphemes receive their phonological form from
the single Vocabulary Item (8c). Put slightly differently, there is a single mor-
phological object at the center of this analysis, one that allows the exponent
noh- to appear in more than one plural morpheme.
Syncretism and Underspecification 119
5.2.3 Comparison
[+1,-2,+pl,+obj] $ noh-
[-1,+2,+pl,+obj] $ noh-
The two Vocabulary Items in (9) are distinct objects, and the fact that they
have an identical phonological exponent is an accident. Using two Vocabu-
lary Items like the ones in (9) goes against the intuition that (all other things
being equal) generalizations are maximized when the number of items in the
Vocabulary is minimized. Recall moreover that in the case at hand, there is
sufficient motivation for treating this identity in form as systematic, since
the first person plural and second person plural object agreement morphemes
show a considerable overlap in their feature content. This overlap in features
speaks strongly against an analysis in terms of accidental homophony.4
We are now in a position to see why Distributed Morphology (and other
realizational theories) have dispensed with traditional morphemes. In a the-
ory that allows only traditional morphemes, with sound and meaning con-
nected from the beginning, the only choice in analyzing Hupa object agree-
ment would be to use the morphemes in (10):
These are distinct objects, just like the Vocabulary Items in (9). There is no
possibility of underspecifying sound with respect to meaning, since sound
and meaning are combined in a single object from the beginning. Traditional
morphemes are thus unable to account for syncretism in a systematic way.5
120 Syncretism and (Under)specification
With this analysis of the syncretism in the plural object morphemes at hand,
the analysis of Hupa Agreement can be completed with the addition of Vocabu-
lary Items for the subject morphemes. In the subject system, there is no syn-
cretism. Thus, the Vocabulary Items in (8) must be augmented with four ad-
ditional Vocabulary Items, as shown in (11):
[+1,-pl,+subj] $ W-
[+1,-pl,+obj] $ W1-
[+2,-pl,+subj] $ n-
[+2,-pl,+obj] $ n1-
[+1,+pl,+subj] $ d1-
[+2,+pl,+subj] $ oh-
[+pl,+obj] $ noh-
With respect to the details of (11), two points are in order.
First, an innovation in (11), which is retained and examined throughout
much of the subsequent discussion in this book, is that all of the Vocabulary
Items are underspecified. The first person singular subject Vocabulary Item,
for instance, does not make reference to the feature [-2]. The reason for this
is that the correct results can be derived without making reference to this
feature. The same is true of the other Vocabulary Items in (11). As a prac-
tical matter, it is often the case that Vocabulary Items are presented in this
way, with the minimum number of features required to ensure their correct
application. The empirical effects of this convention are not obvious in this
particular kind of example, although there are cases where minimizing the
specification of Vocabulary Items can have important consequences.6
Second, regarding the order of application, the order shown for the first six
Vocabulary Items in (11) is more or less arbitrary, as the correct results are
derived for any other order, as long as the Items are specified in this particular
way. The only crucial point of ordering for the present analysis is that the
Vocabulary Item inserting noh- must be underspecified in a way that allows it
to be beaten in the case of [+subj] plural morphemes by the Items that insert
d1- and oh-.
The details about specification of Vocabulary Items, and the interaction of
specification and ordering, are key themes that are revisited in the subsequent
Specification 121
parts of this book (see in particular the case studies in section 5.5 below, as
well as chapter 6).
With reference to the particular analysis in (11), many additional observa-
tions could be made along the general lines of inquiry outlined above. Many
of these implicate the idea that the Vocabulary should be minimized when
possible. To take a specific example, the first and second person singular
exponents in the subject and object categories are remarkably similar: they
differ only in the presence of a /1/ component in the object set (compare W-
/W1-, n-/n1-). If the presence of /1/ in the object set could be attributed to other
factors (in particular, to the phonology), then it would be possible to reduce
the number of Vocabulary Items in (11) further, with single Vocabulary Items
for first person singular and second person singular.7 In this revision, the
Vocabulary Items inserting W- and n- are underspecified for the subject/object
distinction, as shown in (12):
[+1,+pl,+subj] $ d1-
[+2,+pl,+subj] $ oh-
[+1,-pl] $ W-
[+2,-pl] $ n-
[+pl,+obj] $ noh-
The analysis in (12) maintains the intuition that the item inserting noh- is
underspecified, and further reduces the number of Vocabulary Items in a way
that improves on (11).
5.3 Specification
The main point of the last section is that a systematic analysis of syncretism
is produced by underspecifying Vocabulary Items. The specific case of syn-
cretism that is used to illustrate this point above, identity in Hupa first and sec-
ond person plural object agreement morphemes, is derived with a Vocabulary
Item that makes reference only to [+pl] and [+obj]. Because (i) this Vocabu-
lary Item does not mention person features, and (ii) because it is not beaten
by a more specific Vocabulary Item, it can apply to both [+1,+pl,+obj] and
[+2,+pl,+obj] nodes.
122 Syncretism and (Under)specification
One of the assumptions that makes this analysis possible concerns the rep-
resentation of functional morphemes prior to insertion. In particular, it is as-
sumed that while the first and second plural object agreement morphemes are
not distinct phonologically, these nodes are distinct in terms of their synsem
feature content. That is, a first person plural object morpheme is as in (13a),
and a second person plural object morpheme is as in (13b):
Another way of putting this is to say that the identity in form arises as a result
of the Vocabulary Insertion process, not because of something deeper.
The general principle at play here is a standard assumption of Distributed
Morphology: the idea that functional morphemes are always fully specified
for synsem features, whereas Vocabulary Items can potentially be underspec-
ified. Terminologically, it can be said that functional morphemes show Full
Specification, whereas Vocabulary Items may be subject to Underspecifica-
tion. Full Specification is defined in (14), which is presented with a definition
of Underspecification for purposes of comparison:
(14) Full Specification: Functional morphemes are fully specified for the
synsem features that are active in the language.
a. [+1,-2,+pl,+obj]
b. [-1,+2,+pl,+obj]
Full Specification is the reason why the functional morphemes must be spec-
ified in this way. Informally, the idea is that first and second plural object
morphemes are different objects as far as syntax (and semantics) are con-
cerned. In spite of their surface morphological identity (i.e., they are realized
by the same Vocabulary Item), they are different morphemes.
It is worth reflecting further on the motivation for positing the two distinct
morphemes in (16). Here, some other facts about the language help.
One such fact is noted in section 5.2 above. In the subject agreement sys-
tem, there are distinct exponents for all four combinations of person and num-
ber features; that is:
p/n exponent
1s W-
2s n-
1p d1-
2p oh-
From this observation, it is clear (i) that the language distinguishes two per-
sons (first and second) and two numbers (singular and plural); and (ii) that
these features are fully cross-classified in functional morphemes. That is, the
124 Syncretism and (Under)specification
pronoun gloss
We ‘I’
n1n ‘you’
nehe ‘we’
nohn1 ‘you-pl’
Taken together, the observations based on (17) and (18) indicate that Hupa
morphemes contain both person and number features. Since there is no reason
to believe that object morphemes should differ in this regard, Full Specifica-
tion requires an analysis of object agreement with the functional morphemes
in (16).
To this point, Full Specification has been viewed in a positive guise, i.e., as
a way of motivating particular representations of functional morphemes. This
perspective may be reversed, so that Full Specification is viewed negatively;
i.e., in such a way that it rules out certain types of morphemes: namely, those
that are simply “vague” with respect to a particular semantic dimension that
Syncretism versus Homophony 125
[+pl,+obj]
This hypothetical morpheme is vague at the synsem level, in the sense that
it does not contain person features. While it is a possible morpheme in some
other languages (e.g., in a language that does not have active person features,
or in one that does not bundle person and number features together), it is not
part of the grammar of Hupa, for the reasons that are adduced above.8
In summary, Full Specification places restrictions on the representation
of functional morphemes: it requires fully specified syntacticosemantic mor-
phemes, and rules out “vague” morphemes that fail to make distinctions that
are part of the grammar of a language. At the same time, Full Specification
does not say exactly what the required features are; this is something that can
only be determined in an analysis that takes into account syntax, semantics,
and morphology.
Along these lines, in many of the case studies that are examined in later
parts of this book it will become increasingly evident how the analysis of any
system of complex morphological interactions based on Vocabulary Insertion
depends crucially on assumptions about the nature of these different types of
features (recall also the discussion of the UFI in chapter 2).
As a final point, it is important to stress that Full Specification places re-
strictions on the representation of morphemes prior to any further operations
that might apply to them. As will be discussed below in chapter 6, it is pos-
sible to derive morphemes that look like (19) prior to Vocabulary Insertion
with Impoverishment rules, which delete features at PF.
The identities in form that motivate Late Insertion are systematic; and the
analytical tool of underspecification allows for syncretisms to be centered on
a single Vocabulary Item.
126 Syncretism and (Under)specification
According to this treatment, the two Vocabulary Items in (20) are distinct
entities; they have no more in common than any other pair of Vocabulary
Items selected at random from the Vocabulary of the language, except for the
fact that their exponents are identical.
There are, generally speaking, some salient differences between system-
atic syncretisms and accidental homophonies. To a first approximation, acci-
dental homophonies are found when there is no shared synsem feature con-
tent in functional morphemes that have the same phonological form. In syn-
cretism, on the other hand, there is typically shared feature content in the
Illustrations 127
syncretizing environments.9 For instance, in the case of Hupa noh-, the mor-
phemes realized as noh- are both syntacticosemantically plural (i.e., share
the synsem feature [+pl]). This may be contrasted with the English example
in (20), where there is no syntacticosemantic basis for thinking of the [+pl]
morpheme that attaches to nouns and the Agr[3sg] morpheme that attaches to
verbs as having similar feature content.
Even with underspecification at our disposal, it would be extremely diffi-
cult to treat the Agr[3sg] and [+pl] morphemes as subject to insertion by the
same Vocabulary Item. Thus, the best analysis is one in which there are two
distinct Vocabulary Items in memory, as in (20); while accidental homophony
should be minimized when possible (see chapter 6), there are many cases in
which it is unavoidable.
5.5 Illustrations
The functional morphemes that are active in Seychelles Creole are as follows.
In addition to [±1] and [±2] for person and the number feature [±pl], it can
be assumed for convenience that the “strong” and “possessive” columns are
defined by the features [+str] and [+poss] respectively. These features stand
proxy for finer-grained morphosyntactically defined features (related to case,
etc.) that do not play a crucial role in the main exposition. On the assumption
that these features are cross-classified to yield fully specified functional mor-
phemes, there are a number of observations to be made about the distribution
of forms in (21). Overall, (21) shows eight distinct exponents. Of these, four
appear in exactly one cell: mua, i, li, and sõ. There is no need to underspec-
ify the Vocabulary Items that insert these exponents, since their distribution is
such that each of them applies to only one functional morpheme. The remain-
ing four exponents, though, first singular mõ, second singular u, first plural
nu, and non-first person plural (or default plural) zot can be inserted into dif-
ferent morphemes to produce a variety of syncretisms. The Vocabulary Items
that insert these exponents need to be underspecified in order to account for
these distributions. Consider now (22):
a. [-1,-2,-pl,+poss] $ sõ
b. [-1,-2,-pl,+str] $ li
c. [-1,-2,-pl,+subj] $ i
d. [+1,-2,-pl,+str] $ mua
e. [+1,-pl] $ mõ
f. [+2,-pl] $ u
g. [+1,+pl] $ nu
h. [+pl] $ zot
Illustrations 129
There are a few crucial orderings in (22). For example, the Vocabulary Item
(22d) specified with [+1,-2,-pl,+str] to insert mua must beat the item (22e)
that inserts mõ. In addition, the first person plural nu is inserted by (22g),
which must beat (22h). Overall, the ordering of items in (22) reflects the gen-
eral principle introduced in chapter 4, according to which Vocabulary Items
specified for more features take precedence over less specified items.
The Mongolic language Oirat marks possessed nouns with suffixes that agree
with the person and number of the possessor. The possessive morphemes are
shown in (23), which is taken from Birtalan (2003:220):10
singular plural
1 -m -mdn
2 -cn -tn
3 -i -i
The first and second person possessors show distinct forms in the singular and
the plural; but third person possessors do not make this number distinction.
I will assume that in addition to person and number features, the posses-
sive functional morphemes have a feature [+poss]. Thus, for example, the
first singular possessor morpheme is [+1,-2,-pl,+poss], the second singular is
[-1,+2,-pl,+poss], and so on. Although [+poss] does little in the analysis of
(23), it assumes a more important role in a comparison to be made below.
The Vocabulary Items in (24) derive the distribution of exponents in (23):
a. [+1,+pl,+poss] $ -mdn
b. [+2,+pl, +poss] $ -tn
c. [+1,+poss] $ -m
d. [+2,+poss] $ -cn
e. [+poss] $ -i
The syncretism in third person forms arises because the Vocabulary Item
(24e) with the -i exponent is not specified with number features. This aspect
130 Syncretism and (Under)specification
singular plural
1 miny maany
2 ciny tany
3 ny ny
Just as in the case of Oirat, there is no distinction made between singular
and plural third person possessors. However, the exact manner in which un-
derspecification produces this result might be slightly different in the two
languages. It can be seen in the Khalkha forms in (25) that all of the pos-
sessive affixes end in -ny. A straightforward treatment of this pattern with a
-ny exponent is possible if it is assumed that the forms in (25) are composite,
with a person/number morpheme and a possessor morpheme:
�HH
�� H
[±1,±2,±pl] [+poss]
The -ny is then the exponent of a Vocabulary Item that refers to [+poss] alone:
And, completing the analysis, the Vocabulary Items for the person/number
component are as follows:
Illustrations 131
a. [+1,+pl] $ maa
b. [+2,+pl] $ ta
c. [+1] $ mi
d. [+2] $ ci
e. [ ] $Ø
In the analysis based on (27-28), the syncretism does not arise from under-
specifying a Vocabulary Item with [+poss]. Rather, the Vocabulary Item (27)
that realizes [+poss] applies to [+poss] in all of the forms in (25), on the
assumption that these consist of the two morphemes shown in (26). Then,
the syncretism arises in the person/number component of (26): in the case
of third person morphemes, the most specific item in (28) that can apply is
(28e). This Vocabulary Item inserts Ø for the person/number component both
in third person singulars and third person plurals, producing the syncretism in
the desired way: that is, Ø-ny for both third person singulars and third person
plurals.
In sum, the analyses of possessor syncretism in both Oirat and Khalkha
employs underspecified Vocabulary Items. However, the particular morphemes
that are targeted by underspecified Vocabulary Items differs in a way that de-
pends on further assumptions about how features are bundled into morphemes
in the two languages. (As it stands, the analysis treats Oirat and Khalka with
different structures; it would be possible to investigate the Khalka system fur-
ther with a structure that contains a distinct number morpheme; I leave this
and related moves to the reader.)
a. [-1,-2,+obj,-sg] $ -wun
b. [+obj,-sg] $ -iyuw
Illustrations 133
These items produce the correct results. However, the analysis does so in a
somewhat forced way. Rather than saying that -iyuw has a natural distribution,
covering first and second person non-singulars, it instead makes reference to
the third person [-1,-2] in the Vocabulary Item that inserts -wun. Given the
general idea that natural classes should be stated positively whenever possi-
ble, it is worth looking at other analyses.12
An alternative to (30) is to take -iyuw as expressing a natural class, some-
thing that might be preferred on general grounds. To do this in the case at
hand, an additional feature is required. As Noyer (1992) discusses, a natu-
ral class of “first or second person” cannot be stated directly using person
features like [±1] and [±2]. He suggests a feature [±part] for “participant”,
whose value is positive for speech act participants (i.e., first and second per-
son), and negative for non-participants (third person), and provides evidence
that [±part] is important for natural-class behavior in morphology. Other
ways of making the same distinction are conceivable (e.g., building on the
idea that the third person is a “non person” in some sense that makes [-1,-2]
a combination of features that cannot be referred to); but I will use [±part]
here, since my primary focus is on how alternative underspecifications of
Vocabulary Items can account for the distribution of forms in (29).13
The role of [±part] is easy to see in (29). The exponent -iyuw appears
in first and second person duals and plurals. Thus, the Vocabulary Item that
inserts it must be specified for [+part], and must be underspecified with re-
spect to the dual versus plural number distinction. A way of analyzing the
non-singular part of the system along these lines is shown in (31):
(31) Barbareño Chumash Vocabulary Items
a. [+part,+obj,-sg] $ -iyuw
b. [+obj,-sg] $ -wun
There are a number of different factors that have to be assessed in determining
the differences between the analyses in (30) and (31). For example, if gram-
mars are structured such that natural classes such as that defined by [+part]
are always to be preferred over negative specifications like those provided by
[-1,-2], then the analysis in (31) must be regarded as superior.14
Ultimately, the question of which of these analyses is to be preferred in-
teracts with larger assumptions about feature theory. The main point of this
illustration is to provide a further example of how underspecification pro-
duces syncretisms. Since both -iyuw and -wun are inserted into more than
134 Syncretism and (Under)specification
(32) Root-TH-POSS
Nouns are divided into four distinct classes, which differ in terms of the pos-
sessor allomorphs that they take. These are referred to as “class 1”, “class 2”,
etc. here; and diacritic features [I], [II], etc. are employed for these classes
in the analysis below. This treatment of the classes could be refined in vari-
ous ways in a more comprehensive study of Anêm nouns, but it suffices for
present purposes.
Singular possessed forms for nouns from the four different classes are
shown in (33):
Before looking at the third person possessors in greater detail, some points
are in order concerning the class features that determine the realization of the
distinct possessor allomorphs. The structure of possessed nouns is shown in
(34); for the purposes of this analysis, it will be assumed that the theme is a
realization of n:16
(34) Structure of Possessed Noun
�H
�� HH
[+poss]
p ��HH
ROOT n
I will assume that class features of the Root appear on the n morpheme, where
they condition the insertion of the different possessor allomorphs; it can be
assumed that the features are copied to n via concord.17
With respect to the latter, the [+poss] morphemes shown in (34) are spec-
ified for features of the possessor, which are derived from combinations of
[±1], [±2] for person, [±pl] for number, and [±fem] for gender. So, for ex-
ample, the first person singular possessed form of gi ‘child’ is as follows:
(35) Structure of 1s Possessed gi ‘child’
H
�� H
�� HH
�HH [+1,-2,-pl,+poss]
p �
G I[II] [n,II]
Although there is clearly much to be said about the relationships between
Roots, theme classes, and possessive allomorphy (not to mention further pos-
sible connections with gender and number), the role of underspecified Vocabu-
lary Items in the possessive morphology is relatively straightforward. As can
be seen in (33), classes 1 and 2 show distinct exponents for third person mas-
culine and third person feminine possessors, but in classes 3 and 4 there is a
single third person possessor exponent in each class. In terms of Full Speci-
fication, then, it is clear that distinctions for the gender of the possessor are
present on all [+poss] morphemes prior to Vocabulary Insertion. Thus, the
syncretism in classes 3 and 4 can and must be attributed to the underspecifi-
cation of Vocabulary Items.
Concentrating on the third person (i.e. [-1,-2]) part of the system, the fol-
lowing Vocabulary Items produce the correct results:
136 Syncretism and (Under)specification
Each Vocabulary Item in (36) bears a contextual specification for class. The
point to note is that class 1 and class 2 are associated with distinct Vocabulary
Items for [+fem] and [-fem], so that there are two distinct items for each of
these classes. On the other hand, classes three and four are referred to only
by one Vocabulary Item each in (36). As a result, there is only one item that
can apply to both [+fem] and [-fem] possessor morphemes in these classes;
and this produces the syncretism in (33).
In this case, unlike what we have seen in prior examples, the syncretism
involves Vocabulary Items that have a contextual specification for noun class
features. In this way, the syncretism is restricted to a particular subset of
nouns of the language. Underspecifying Vocabulary Items produces the cor-
rect results for these nouns. However, for other contextually determined syn-
cretisms, additional mechanisms are required; see chapter 6.
5.6 Conclusions
The primary goal of this chapter was to show how Vocabulary Insertion can
be used for the analysis of syncretisms. A key assumption is that Vocabulary
Items may be underspecified with respect to the functional morphemes that
they apply to. Because of this, it is sometimes the case that a single Vocabu-
lary Item can apply to multiple, distinct functional morphemes in a language.
When this happens, the resulting surface identities in form are treated as sys-
tematic effects in the grammar. They are systematic because they are derived
from the same Vocabulary Item. Without using Vocabulary Items (more gen-
erally, without a realizational approach to morphology), a straightforward
analysis of syncretisms is not possible.
Underspecified Vocabulary Items are the key to understanding how syn-
cretism is analyzed in Distributed Morphology. As mentioned at various points
Conclusions 137
above, though, there are some phenomena which appear to require addi-
tional mechanisms that work together with underspecification. These addi-
tional complications, and some of the mechanisms that have been proposed
to deal with them, are the topic of the next chapter.
Chapter 6
Further Topics in the Analysis of Syncretism
6.1 Introduction
6.2 Impoverishment
[a] −! Ø/ K
Item that makes reference to that feature cannot apply, with the result that (ii)
a less specified Vocabulary Item does apply.
The contexts in which an Impoverishment rule is specified to occur can
in principle be very specific, or very general. The illustrations in this section
concentrate on the first, specific application of Impoverishment rules. The
more general application is the topic of section 6.3.
The examples in (2) show the third person accusative and dative clitics in iso-
lation. The important observation at the center of Bonet’s analysis is that these
two clitics cannot co-occur; rather, in the equivalent of (2) with two clitic
pronouns—one for the theme argument and one for the recipient—neither
*le lo or *lo le surfaces; instead, the sequence se lo is found:
The second clitic, lo, is what is expected for the masculine accusative (recall
(2a)). The first clitic, se, is glossed simply as SE. This clitic has a complicated
distribution in Spanish (and, in fact, in other Romance languages as well). It
occurs in reflexive clauses, certain anticausatives (unaccusative intransitives),
142 Further Topics in the Analysis of Syncretism
and in a number of other contexts. This broad distribution (along with other
facts about Spanish discussed by Bonet and others) suggest that it is the de-
fault in the system of Spanish clitics. This aspect of SE plays an important
role in Bonet’s analysis, as will be seen immediately below.
Bonet’s observation is that while the behavior of the clitics in (3) is opaque,
since the expected outcome *le lo is not found, it is also not arbitrary: rather,
the expected and more specified dative clitic is realized instead with the de-
fault clitic se. The realization of an independently existing default in this con-
text suggests that clitic realization constitutes a closed system. It is exactly
with this kind of phenomenon that Impoverishment can work together with
underspecified Vocabulary Items to derive the correct results, because Impov-
erishment can be used to force the application of a less specified Vocabulary
Item within the clitic system.
A number of different analyses employing Impoverishment to derive the
spurious se effect appear in the literature; all share the same essential insight,
which is that contextual deletion of features of the dative clitic results in the
application of a highly underspecified (default) Vocabulary Item. To take one
particular analysis for purposes of illustration, in the treatment that is devel-
oped in Halle and Marantz (1994), Spanish clitics are analyzed as internally
complex, as shown in (4):
(4) Spanish Clitics (from Halle and Marantz 1994)
a. First Person: m-e (singular); n-o-s (plural)
[same for accusative, dative, and reflexive]
b. Second Person: t-e (singular); o-s (plural)
[same for accusative, dative, and reflexive]
c. Third Person
on the third person part of this system, Halle and Marantz suggest that the
l- component of the various third person clitics can only be inserted in the
context of a [+case] feature. The s- of the reflexive, on the other hand, is the
default. This is shown in the Vocabulary in (5), which looks only at third
person forms:
[cl +case] $ l
[cl ] $ s
With the Vocabulary Items in (5), the spurious se phenomenon is then ana-
lyzed with the contextual deletion of dative case features, here abbreviated
[+dat]:
This rule deletes the [+dat] feature in the context of [+acc]. With the Vocabu-
lary in (5), the result of the Impoverishment rule applying is that l cannot be
inserted, since its Vocabulary Item makes reference to case features; thus, the
default s is found.
Impoverishment accounts for the special behavior of particular clitic com-
binations in a way that directly accounts for the intuition that the action occurs
within a closed system, in which contextual deletion results in the application
of a less-specified Vocabulary Item.
-neut +neut
-pl -Ø -t
+pl -e -e
The singular shows sensitivity to [±neut], with -Ø for [-neut] and -t for
[+neut]. The plural is not sensitive to [±neut], and shows only -e.
As far as just these forms go, the Vocabulary Items in (8) can be employed
to derive the correct forms. The only point that is worthy of further attention
is that (8) treats -e as a default, not as the exponent of a Vocabulary Item
specified for [+pl]. The reasons for doing this will become clear below.
[adj +neut,-pl] $ -t
[adj -neut,-pl] $ -Ø
[adj ] $ -e
Moving past the strong adjectives, an important observation about the system
of adjective agreement as a whole is that the inflection of weak adjectives
involves only one exponent, -e:
-neut +neut
-pl -e -e
+pl -e -e
In light of (7) and (9), it is clear that the Vocabulary Item with -e is a typical
default: it applies in a range of environments that is not a natural class, cov-
ering both the weak/strong distinction, singulars and plurals, and neuters and
non-neuters. However, simply allowing the Vocabulary Items in (8) to apply
in the weak context does not generate the correct results, even if -e is under-
specified in the way that is shown in (8). The reason for this is that the -t and
-Ø Vocabulary Items should win out over -e in weak non-neuter and neuter
Impoverishment 145
singulars, just as they do with the strong adjectives. Thus, in the analysis de-
veloped to this point (i.e., (8)), it is predicted that strong and weak adjectives
should show exactly the same pattern of exponents, contrary to fact.
Looking at (7) and (9), it can be observed that the distinction between
[±neut] values plays a role in the realization of singulars in (7), but no role
in (9). Mechanically, what needs to happen for (9) to be derived is that the
Vocabulary Items with -t and -Ø exponents must not apply in the weak con-
text, so that the default Vocabulary Item, with -e, is used.
The Impoverishment rule (10) causes this to happen by deleting [±neut]
in the weak environment:
In the weak context, the feature [±neut] is deleted by (10), whatever its value.
As a result, neither of the first two Vocabulary Items in (8) can be employed
in that context, because each of these makes reference to one of the values
of [±neut]. Thus, the only Vocabulary Item from (8) that can apply in weak
contexts is the default, with the result that -e occurs across the board.
To this point, the reasoning follows Sauerland’s (1995) discussion in us-
ing Impoverishment to derive the distribution of -e in the weak adjectives.
The need to use Impoverishment in this example depends, however, on some
further assumptions on what features the Vocabulary Items are specified for.
Impoverishment deletes gender in the weak contexts, with the result that
neither of the first two Vocabulary Items in (8) can be used there. Consider
now what would happen if, rather than referring to [weak] in an Impoverish-
ment rule like (10), the analysis referred to [strong] in the Vocabulary Items
inserting -t and -Ø. The analysis would then be as in (11), which assumes that
[strong] is a feature on the node undergoing insertion:
[adj +neut,-pl,strong] $ -t
[adj -neut,-pl,strong] $ -Ø
[adj ] $ -e
In summary, both analyses developed above derive the correct results. For
comparing them, what is at issue is whether or not it is possible for Vocabu-
lary Items to refer directly to the feature [strong]. The distribution of weak
versus strong forms in Germanic is a complex issue, one that implicates the
interaction of syntactic structures with morphosyntactic features in ways that
warrant careful investigation. For immediate purposes, what is important is
how the two analyses discussed above illustrate how Impoverishment inter-
acts with other assumptions about the features that are referred to in Vocabu-
lary Items: for some phenomena, the need for Impoverishment is mitigated
when alternative specifications for Vocabulary Items are considered.
6.2.3.1 Ugaritic
p/n/g form
1s P-ktb
2s.masc. t-ktb
2s.fem. t-ktb-n
3s.masc. y-ktb
3s.fem. t-ktb
1d (n-ktb)
2d.masc. t-ktb-(n)
2d.fem. N/A
3d.masc. y/t-ktb-(n)
3d.fem. t-ktb-(n)
1p n-ktb
2p.masc. t-ktb-(n)
2p.fem. (t-ktb-n)
3p.masc. y/t-ktb
3p.fem. (t-ktb-n)
[Agr -1,-2,+masc,-pl] $ y-
[Agr +1,+pl] $ n-
[Agr +1] $ P-
[Agr ] $ t-
The intuition behind this analysis is as follows. First, all of the verbs in
the prefix conjugation are prefixed with an Agr(eement) morpheme, which
has features for person, number, and gender. Second, while there are a few
Vocabulary Items that realize specific feature combinations, as can be seen in
(13) for those that insert y-, n- and P-, the Vocabulary Item with t- realizes
the rest; that is, it applies in a non-natural set of environments, the defining
property of a default or elsewhere form.
148 Further Topics in the Analysis of Syncretism
6.2.3.2 Polish
Halle and Marantz (2008) analyzes the complex system of nominal case in-
flections in Polish (cf. Cameron-Faulker and Carstairs-McCarthy (2000)).
Their treatment is limited to the inflection of masculine singular nouns, which
can be divided into a number of distinct inflectional classes (1, 2, ... below)
depending on the set of case affixes that they appear with:
case/class 1 2 3 4 5 6 7
nom – – – – – – –
gen -a -a -a -a -a -u -u
dat -owi -owi -u -u -owi -owi -owi
inst -em -em -em -em -em -em -em
loc -e -u -e -u -u -e -u
voc -e -u -e -e -e -e -u
The aspect of (14) that is directly relevant to the discussion of defaults is ex-
hibited by the -u exponent, which can be seen in the genitive, dative, locative,
and vocative rows. In Halle and Marantz’s analysis, the Vocabulary Item with
-u is the default for realizing cases, as shown in (15).1
[nom] $ -Ø
[gen] $ -a
[dat] $ -owi
[inst] $ -em
[loc] $ -e
[voc] $ -e
[ ] $ -u
To account for these effects, Halle and Marantz propose a set of Impover-
ishment rules that are sensitive to the class that a noun belongs to. These rules,
which are shown in (16), have the effect of deleting particular case features
in the context of a noun from the listed class:2
a. [gen] −! Ø/{6,7}
b. [dat] −! Ø/{3,4}
c. [loc] −! Ø/{2,4,5,7}
d. [voc] −! Ø/{2,7}
These rules apply prior to Vocabulary Insertion, and delete case features in
the manner specified. As a result, the most specific Vocabulary Item that can
apply to the Impoverished case nodes is the default in the system, which
inserts -u. In this way, Impoverishment and underspecification work together
to derive the default distribution of the -u exponent.
Thus, the fact that [±b ] cannot be referred to in Vocabulary Insertion at [+a]
morphemes is accounted for directly. Syncretism for [+a,±b ] in multiple
environments (and with more than one surface realization) then follows from
the fact that [±b ] cannot be referred to by Vocabulary Items applying to [+a]
morphemes. In short, it is not simply an accident of the Vocabulary that the
two Vocabulary Items (17) and (18) fail to refer to values of [±b ]; rather,
the rule in (19) directly accounts for the generalization that the language sys-
tematically eliminates [±b ], in a way that has consequences for a number of
distinct Vocabulary Items.
In the following sections, some concrete examples of this use of Impover-
ishment are examined.
The inflection of Latin nouns provides an example of the deeper type of for-
mal identity introduced above. Nouns in the language are inflected for five
cases: nominative, genitive, dative, accusative, and ablative (an additional
vocative case is ignored in this example); they also show a distinction be-
tween singular and plural numbers. The nominals are organized into five de-
clension classes, given here as I-V; the declension to which a Root belongs is
an arbitrary property of that Root. The case and number forms of five nouns
from the different declensions is shown in (20) (the nouns are mēnsa ‘table’;
hortus ‘garden’; cōnsul ‘consul’; frūctus ‘fruit’; rēs ‘thing’):3
case I II III IV V
sg. nom mēns-a hort-us cōnsul frūct-us rē-s
gen mēns-ae hort-ı̄ cōnsul-is frūct-ūs re-ı̄
dat mēns-ae hort-ō cōnsul-ı̄ frūct-uı̄ re-ı̄
acc mēns-am hort-um cōnsul-em frūct-um re-m
abl mēns-ā hort-ō cōnsul-e frūct-ū rē
pl. nom mēns-ae hort-ı̄ cōnsul-ēs frūct-ūs rē-s
gen mēns-ārum hort-ōrum cōnsul-um frūct-uum rē-rum
dat mēns-ı̄s hort-ı̄s cōnsul-ibus frūct-ibus rē-bus
acc mēns-ās hort-ōs cōnsul-ēs frūct-ūs rē-s
abl mēns-ı̄s hort-ı̄s cōnsul-ibus frūct-ibus rē-bus
152 Further Topics in the Analysis of Syncretism
The features [±obl], [±str], and [±sup] are defined morphosyntactically (see
the works cited for details). In terms of (21), there is a single feature value that
distinguishes dative case morphemes from ablative case morphemes: dative
is [+obl,+str,+sup], while ablative is [+obl,-str,+sup].
To derive dative/ablative syncretism in the plural, the following Impover-
ishment rule can be used:
Impoverishment and Patterns of Syncretism 153
(22) Impoverishment:
For a second illustration, I return to the example from Latin American Span-
ish that is used to illustrate syncretism in chapter 1. Recall that there is a
154 Further Topics in the Analysis of Syncretism
general neutralization of the second and third persons in the plural, as shown
in (25a) for three tenses, the present, imperfect, and preterite (all indicative)
for the verb hablar ‘speak’; (25b) shows Peninsular Spanish for comparison:
(25) Three Tenses for Latin American and Peninsular Spanish
a. Latin American
�H
�� HH
H Agr
�� H
T
p ��HH
ROOT TH
156 Further Topics in the Analysis of Syncretism
Regarding aspect, I will assume that the imperfects are T[+past], while the
aorists are T[+past, perf], with [perf] a unary feature.4
The important pattern is found in the second and third person singulars in
the past tenses: they are identical, both in terms of how Agreement and Tense
are realized. Moreover, the identities in form go beyond the properties of par-
ticular Vocabulary Items, as they are found in both the past imperfective and
in the aorist. These facts make the system a prime candidate for an analysis
with Impoverishment.
Starting with agreement, the second singular Agr morphemes start as (29a),
whereas the third singular Agr is as in (29b):
(30) Impoverishment:
The result of this rule is that, prior to Vocabulary Insertion, the two mor-
phemes in (29) become identical: [-1,-pl].
With respect to agreement, no distinction between second and third per-
sons can be made in Vocabulary Items that apply to the Agr node itself. Thus,
the same Vocabulary Items will apply to second and third person singulars in
the past, with -še inserted in the imperfect, and -Ø in the aorist:
[-1,-pl] $ -Ø/T[+past,perf]
[-1,-pl] $ -še/T[+past]
An important part of this analysis is that deleting the [±2] feature does not
make second person morphemes behave like e.g. first person morphemes;
this is not possible, because (29a) does not possess the feature [+1]. Thus
Impoverishment and Patterns of Syncretism 157
The New Guinea language Amele, described and analyzed in Roberts (1987),
provides an example of a pattern of syncretism that appears to be quite broad
in the system. This language shows singular, dual, and plural numbers. The
main pattern of interest is that second and third persons are not distinct in non-
singulars. Thus, second person duals and third person duals are identical, as
are second person plurals and third person plurals. However, second and third
person duals are still distinct from second and third person plurals.
158 Further Topics in the Analysis of Syncretism
The same syncretisms are found in other parts of the language as well; for
example, in free pronouns and in pronominal (direct) object clitics:
plural, that occurs elsewhere in the language. The second and third persons
are identical in duals and plurals, exactly as in the verbal agreement above.
As in the examples studied earlier in this section, it is possible to account
for this general pattern of identity with Impoverishment. In the case at hand,
the Impoverishment rule must eliminate the distinction between second and
third person in the non-singular numbers. This can be done with the rule (35):
(35) Impoverishment:
The effects of this rule are shown in (36), which shows morphemes for the
different person/number combinations in (33) prior to and after Impoverish-
ment (– – is shown for singulars because the rule does not apply to them):
before after
1s [+1,-2,+sg,-pl] ––
1d [+1,-2,-sg,-pl] [+1,-sg,-pl]
1p [+1,-2,-sg,+pl] [+1,-sg,+pl]
2s [-1,+2,+sg,-pl] ––
2d [-1,+2,-sg,-pl] [-1,-sg,-pl]
2p [-1,+2,-sg,-pl] [-1,-sg,+pl]
3s [-1,-2,+sg,-pl] ––
3d [-1,-2,-sg,-pl] [-1,-sg,-pl]
3p [-1,-2,-sg,+pl] [-1,-sg,+pl]
As can be seen in the “after” column of (36), the first person non-singulars
are [+1], and distinct from second and third person non-singulars. Within the
second and third persons, there is no distinction for person. There is, however,
a distinction for number (dual versus plural), so that the difference between
duals and plurals can be referred to when Vocabulary Insertion applies.
Impoverishment clearly plays an important role in the analysis of this syn-
cretism. Relying on underspecification alone, it would be possible to derive
the correct forms, but the generalization that second and third persons are the
same in a number of places in the language would not be accounted for.
The pattern in Amele highlights a further question about the role of Im-
poverishment in the grammar. The question is based on the observation that
160 Further Topics in the Analysis of Syncretism
there does not seem to be any place in the grammar of Amele in which either
second and third person duals, or second and third person plurals actually
are distinguished from each other morphologically. Unlike e.g. Macedonian,
where second and third person singular forms are distinct elsewhere in the
verbal system, there appears to be no indication anywhere in the morphology
of Amele that the relevant feature combinations are realized distinctly.
These observations raise the question of whether another analysis should
be considered for “across the board” neutralizations of this type. The Impov-
erishment analysis holds that Amele person and number features are specified
as in the before column of (36), and Impoverished at PF to yield the after be-
fore Vocabulary Insertion occurs. A more radical alternative would be to say
that the language simply does not distinguish person in non-first person non-
singulars, at the level of its inventory of functional morphemes. According
to this view, the language does not possess functional morphemes with the
full cross-combination of person/number features in the first place; rather, its
morpheme inventory consists of the after column in (36) (more precisely: the
inventory would consist of the singulars from the before column and the non-
singulars from the after column). This can be called an inventory analysis.
Whether or not the inventory analysis is to be preferred over Impoverish-
ment depends on other assumptions about how languages bundle features into
morphemes in the first place. To start with, something would have to be said
about the semantic interpretation of morphemes that do not have [±2], and
which are therefore “vague”. Beyond this, if there were a general principle to
the effect that languages default to a full combination of features in their basic
inventory of morphemes, then this would be an argument in favor of main-
taining the first theory proposed above, with full feature combination plus
subsequent Impoverishment. On the other hand, if there is a general principle
to the effect that morphemes are only posited when their feature combination
is realized uniquely somewhere in the language, then this might provide mo-
tivation for the inventory analysis. As both of these principles seem to have
something in their favor, it is not clear whether there is an obvious preference,
at least at this level of abstraction. And, importantly, the two analyses under
consideration both derive the surface identity of forms, though in different
ways; this makes it difficult to see what types of empirical evidence could be
used to distinguish them from each other.
In summary, Amele shows a very systematic pattern of identity in form
that could be treated either with a completely general Impoverishment rule,
or with an inventory of morphemes that does not combine [±2] with non-
Underspecification and Impoverishment 161
subject object
1s [+1,-2,-pl,+subj] [+1,-2,-pl,+obj]
2s [-1,+2,-pl,+subj] [-1,+2,-pl,+obj]
1p [+1,-2,+pl,+subj] [+1,-2,+pl,+obj]
2p [-1,+2,+pl,+subj] [-1,+2,+pl,+obj]
[+1,+pl,+subj] $ d1-
[+2,+pl,+subj] $ oh-
[+1,-pl] $ W-
[+2,-pl] $ n-
[+pl,+obj] $ noh-
162 Further Topics in the Analysis of Syncretism
The question to consider in light of this chapter is why this system is not
treated with an Impoverishment rule that eliminates the distinction between
first and second person plural in the object system; that is:
(39) [±1/±2] −!Ø/[ ,+pl,+obj]
With underspecification alone, as in (38), the analysis says that person is not
referred to in the Vocabulary Item that applies to object plurals. In an analysis
with the Impoverishment rule (39), the further component is that it is part
of the grammar that Vocabulary Items could not refer to person features in
object plurals, because of deletion; this appears to be a stronger claim (see
Noyer (1998) for some pertinent discussion of the latter point).
The typical perspective, adopted here, is that Impoverishment is not needed
in examples like the Hupa one; rather, Impoverishment rules are posited only
when necessary. This necessity can, as we saw earlier in the chapter, take one
of two forms: it can be found either (i) when Impoverishment is needed to
make a less specified Vocabulary Item apply (section 6.2); or (ii) when Im-
poverishment is needed to account for a generalization that is deeper than the
Vocabulary (section 6.3). Since neither of these conditions is met with Hupa
object plurals, the analysis with underspecification alone is to be preferred.
There is in fact a way of testing the predictions of the “underspecification
only” versus Impoverishment analyses. I will first present the predictions ab-
stractly, and then turn to actual examples.
Schematically, and using the features [±a,±b ], and assuming in addition
that both [+a] morphemes are realized in the same way as -X, the effect to
concentrate on is as follows. When underspecification alone is used, the value
of [±b ] can be ignored for Vocabulary Insertion in the way shown in (40),
which realizes both [+a,+b ] and [+a,−b ] with the exponent -X:
(40) [+a] $ -X
Consider now what would happen in a language in which the morpheme with
[±a,±b ] is preceded or followed by a morpheme with the feature [+g]. Ac-
cording to the underspecification-only approach, the difference between [+b ]
and [−b ] is visible to the g-morpheme, even if [±b ] is not referred to in the
Vocabulary Item (40). Thus, it would be possible for [+g] to show contextual
allomorphy in a way that refers to [±b ], even though [±b ] is not referred to
in the Vocabulary Item (40) itself. The relevant Vocabulary Items for [+g] are
shown in (41) (I have put the both before and after [±b ], since in principle
the effect could happen in either position):
Underspecification and Impoverishment 163
Thus, this analysis predicts that [±b ] should have no effect on morphological
realization, whether in the morpheme in which [±b ] originates, or in nearby
morphemes.
With this schematization at hand, we may consider a few concrete exam-
ples that illustrate the empirical predictions of the different approaches.
One pertinent case study is provided by the Macedonian verbs considered
above in section 6.3.4. The relevant verb forms are repeated in (43), and the
structure that they are realized in is shown in (44):
�H
�� HH
H Agr
�� H
T
p ��HH
ROOT TH
164 Further Topics in the Analysis of Syncretism
Recall with respect to (43) that the imperfect and aorist are both past tenses,
with the former having T[+past], and the latter T[+past,perf].
The facts in (43) were discussed in section 6.3.4 with reference to two
effects: the realization of the Tense morpheme, which alternates between -v
and -Ø; and the syncretism of second and third person singulars. The main
point of section 6.3.4 was to show that a single Impoverishment rule that
deletes [±2] in the context of T[+past] produces both of the effects of interest.
It accounts for the second/third person singular syncretism, in a way that is
implemented with the Vocabulary Items in (45); and, by eliminating [±2],
there is also an effect on the Vocabulary Items that realize Tense shown in
(46), by causing -Ø to be inserted with second person singulars, and not the
expected -v:
[-1,-pl] $ -Ø/T[+past,perf]
[-1,-pl] $ -še/T[+past]
laufen kaufen
1s lauf-e kauf-e
2s läuf-st kauf-st
3s läuf-t kauf-t
1p lauf-en kauf-en
2p lauf-t kauf-t
3p lauf-en kauf-en
The second and third person singular forms are Umlauted. As I mentioned
above, Umlaut is a morphophonological rule: one that is triggered bypcertain
p and which applies only to certain hosts (in this example, L AUF
morphemes,
but not K AUF) (cf. Wiese (1996) and Embick and Halle (2005)). In verbs,
Umlaut is triggered by second and third singular morphemes ([-1,+2,-pl] and
[-1,-2,-pl]), but not by other Agr morphemes.
The realization of the suffixal Agr morpheme in these verbs shows a syn-
cretism that is important for the main argument of this section. In the third
person singular and the second person plural, Agr is realized as -t. If this
pattern is treated with underspecification alone, there is no issue with the
distribution of Umlaut in (47): it is possible to say that an underspecified
Vocabulary Item inserts -t into these two morphemes, but that the Umlaut
rule is triggered only by second and third person singular Agr, and not by
second plural. That is, second person plural is still different from third singu-
lar (and second singular) in a way that can be seen by the Umlaut rule, even
though the third singular and second plural morphemes are both realized with
-t. On the other hand, if Impoverishment were required for all syncretisms,
it would not be possible to account for the different behaviors of Vocabulary
Insertion and Umlaut in a straightforward way. The distinction between the
Agr morphemes realized as -t would have to be neutralized by Impoverish-
ment prior to Vocabulary Insertion. This would work as far as the insertion
of -t goes, but it would then not be possible to refer to the different feature
values for the triggering of Umlaut; 3s and 2p Agr should either both trigger
Umlaut or both not do so, contrary to fact.
In summary, examples like those considered just above argue that under-
specified Vocabulary Items alone are responsible for simple syncretisms, and
that Impoverishment is only used in a particular set of circumstances, as dis-
cussed earlier in the chapter.
166 Further Topics in the Analysis of Syncretism
Vocabulary Insertion, which is at the heart of the theory of the morpheme de-
veloped in this book, is motivated empirically by the existence of syncretism
in natural language. By providing a way of underspecifying phonology with
respect to syntax and semantics, the theory is able to account for syncretism
in a systematic (i.e. non-accidental) way. In concluding this look at Vocabu-
lary Insertion, there are two additional points to be addressed: one general,
and one specific.
On the more general level, the idea that syncretisms should be analyzed
with underspecified Vocabulary Items is, in effect, a manifestation of a gen-
eral economy principle along the lines of (48):
(48) M INIMIZE VOCABULARY: Analyze the system with the fewest Vocabu-
lary Items possible.
There are other ways of framing this principle that are more or less equiva-
lent for present purposes (e.g. in Embick 2003 it is called “Avoid Accidental
Homophony”); something along these lines is assumed in all work that tries
to account for syncretism systematically. In the theoretical framework that is
advanced in this book, the principle in (48) is not part of the grammar per se.
Rather, it is employed by the learner in the course of the acquisition process,
where it functions as the learner’s default hypothesis; one that can be overrid-
den in the event that no systematic analysis of an identity in form is possible
(in which case there is homophony).
On the more specific front, it is important that the identities in form that I
have called systematic syncretisms are treated in the theory as fundamentally
distinct from accidental homophony. However, having a way of representing
syncretism and homophony distinctly in the grammar does not mean that it is
always clear whether a particular identity in form is analyzed in the grammars
of speakers in one way versus the other. As emphasized at various points ear-
lier in the book, some of the boundary conditions for the theory of syncretism
come from the theory of synsem features, such that the analysis of whether a
particular pattern of formal identity is systematic or accidental must be linked
closely to syntax and semantics. As a result, the theory of syncretisms and the
theory of synsem feature-types rely crucially on one another, or, at the least,
are mutually informative. This is an essential property of the theory: it rein-
forces the point that the theory of morphology is responsible for connections
with several distinct subparts of the grammar, and, in fact, cannot be pursued
General Conclusions: Syncretism and Vocabulary Insertion 167
�H
�� HH
T[+past]
p ��HH
ROOT v
tions under which a morpheme may or may not see another morpheme in its
context comprise the theory of contextual allomorphy.
The second effect of interest in analyses like (1) is that the Vocabulary
Items with special exponents like -Ø and -t prevent the regular past tense
form from occurring, such that both “overregularized” *bend-ed and “dou-
bly marked” *ben-t-ed are ungrammatical. These effects are referred to as
blocking effects; and an important part of Distributed Morphology is the pro-
posal that Vocabulary Insertion as defined in this work produces the types of
blocking effects that are found in language.
Contextual allomorphy and blocking are central components of morpho-
logical theory. The rest of this chapter looks at each of these in turn, starting
with an overview of different types of allomorphic conditioning in section
7.2. Section 7.3 then outlines the principal aspects of a theory of contextual
allomorphy, with particular emphasis on the locality conditions that constrain
allomorphic interactions, and the predictions that derive from the assumption
that Vocabulary Insertion proceeds from the inside out. Section 7.4 examines
blocking effects in the grammar, and develops the idea that blocking is lim-
ited to interactions at the level of the morpheme; i.e., to interactions among
Vocabulary Items. Along with this discussion, I examine non-affixal mor-
phological alternations, like the one in sing/sang, from the perspective of a
morpheme-based theory of blocking. Section 7.5 concludes.
tive definition covers what are treated with two distinct mechanisms: one that
is phonological, and one that is morphological (see below). In this work, al-
lomorphy is used to refer to only the second of these two senses—that is, to
cases in which more than one Vocabulary Item realizes a given morpheme, as
in the case of the English past tense example considered above in (1). When it
is necessary to emphasize that a given alternation is morphological, the term
(suppletive) contextual allomorphy is employed.
As noted in the last paragraph, suppletive contextual allomorphy is only
one type of morphological alternation; there is also a phonological way in
which morphemes change their form. A simple example suffices to illustrate
what is at issue in the distinction between the morphological and phonolog-
ical alternations. Consider the fact that the English plural /-z/, which is the
default realization of the plural morpheme [+pl], does not always surface as
/z/; rather, it surfaces in the forms shown in (3):
As it turns out, the distribution of the different surface forms of the plural
morpheme in English is phonologically determined. If the underlying form
of the exponent of the default plural is assumed to be /z/, then the surface
forms /s/ and /@z/ can be derived via phonological rules. In this (standard)
analysis of this alternation in form, there is one object at the morphologi-
cal level of analysis (the Vocabulary Item with the exponent /z/), and three
different surface realizations at the phonological level of analysis:
One way of thinking about (4) is that it reduces the number of Vocabulary
Items stored in memory. That is, there is no need to posit Vocabulary Items
with the exponents /s/ and /@z/, because it is straightforward to derive these
surface forms from a single Vocabulary Item, in the way described above.
The situation with the alternations seen in (3) contrasts with (suppletive)
contextual allomorphy of the plural morpheme in e.g. ox, which has the plural
ox-en. This fact about [+pl] has to be memorized; it is not phonologically
172 Contextual Allomorphy, and Blocking
triggered, since e.g. box takes the default plural (box-es). Importantly, there
is no reason to think that -en should be derived from /-z/ in the phonology (or
vice versa). Rather, it appears that -en is part of a separate Vocabulary Item;
the same considerations apply to the -Ø realization of plural seen in moose
etc., so that we have the Vocabulary Items in (5):
(5) Vocabulary Items for English Plurals
p p
[+pl] $ -en/{p OX, pC HILD, ...}
[+pl] $ -Ø/{ F ISH, M OOSE, ...}
[+pl] $ /-z/
Thus, in this morphological type of alternation, there are multiple objects
at the morphological level of analysis—viz., the Vocabulary Items in (1) or
(5)—making this alternation an instance of allomorphy (i.e., suppletive con-
textual allomorphy), as the term is employed in this book.
To summarize, it will be said that the plural morpheme [+pl] has three
allomorphs, which are the exponents of the Vocabulary Items in (5). For the
differences in form that we saw with /-z/ when it attaches to different hosts,
as illustrated in (4), it may be said that there are different surface realizations
of the /-z/ exponent of plural.
��HH
Z
��HH
Y
p ��HH
ROOT X
Types of Allomorphic Conditioning 173
Thus far, we have seen examples of morphemes looking both inwards and
outwards for contextual allomorphy. The Vocabulary Insertion mechanism
provides a way of accounting for these effects, but more remains to be said
about potential asymmetries between looking inwards and looking outwards.
In particular, it will be shown in section 7.3 below that the assumption that
insertion proceeds from the inside out—which was introduced in chapter 4—
makes predictions about the types of information that can be referred to in
inwards sensitive versus outwards sensitive allomorphy.
Descriptively speaking, there are two different kinds of triggers for contextual
allomorphy: grammatical triggers and phonological triggers:
174 Contextual Allomorphy, and Blocking
In this kind of allomorphy, it is not necessary to list the Roots that take the
-i allomorph (or those that take the -ka allomorph) in the relevant Vocabulary
Items. Rather, the Vocabulary Items are specified so as to make reference to
the triggering phonological factor, as shown in (11):
[nom] $ -i /C
[nom] $ -ka /V
Types of Allomorphic Conditioning 175
The kind of distribution illustrated in (10) and analyzed with the Vocabu-
lary Items in (11) is not uncommon. For example, the language Seri, spoken
in Mexico, shows the same kind of C/V conditioning for its passive prefix
(Marlett and Stemberger (1983)):
Haitian Creole shows something that looks quite similar with its definite mor-
pheme (Klein (2003)):
This looks more or less like the cases of phonologically conditioned allomor-
phy seen in the different examples in section 7.2.2 above. At the same time,
since the two surface realizations of third person singular possessor are simi-
lar, in the sense that they overlap considerably in their phonological content,
it is conceivable that they could derive from a single exponent. Under the first
scenario, contextual allomorphy, the Vocabulary of Turkish would contain the
two different Vocabulary Items in (15):
Then, in the phonology, a rule that deletes the /s/ component of this mor-
pheme must apply when it is adjacent to a consonant (alternatively, a Vocabu-
lary Item with the exponent -I could be posited, with the /s/ inserted phono-
logically).
Each of these two analyses derives the correct results. In the absence of
other assumptions, it is not clear which is correct. For example, if minimizing
the size of the Vocabulary is paramount, then the second analysis would be
superior to the first. This (morpho)phonological analysis could receive some
further support from the fact that Turkish has a large number of morphemes
that show the deletion of an initial consonant when they are preceded by a
consonant (see e.g. Lewis (1967) and Kornfilt (1997)). However, the deletion
rule must be restricted so that it applies to certain morphemes, and not oth-
ers. While the third singular possessive alternates between -sI and -I, other
superficially similar morphemes do not; for instance, conditional -se always
surfaces in that form, and does not alternate with -e (Lewis (1967:130)). Thus,
if there were reasons for holding that morphologically conditioned phonolog-
ical rules are to be avoided whenever possible, then it would be preferable to
posit two Vocabulary Items along the lines of (15).
The general kind of question raised by the facts of Turkish examined im-
mediately above—allomorphy versus (morpho)phonology—arises in a num-
ber of languages and has been a disputed topic for many years (see e.g.
Kiparsky (1996) and Embick (2010b) for general perspectives). As can be
seen in even this preliminary discussion, a number of language-specific and
178 Contextual Allomorphy, and Blocking
(C1) Cyclic Locality: Two morphemes can see each other for allomorphic
purposes only if they are active in the same phase-cyclic domain.
(C2) Concatenation (linear adjacency): A morpheme X can see a mor-
pheme Y for allomorphy only when X is concatenated with Y : X _Y
or Y _ X.
Conditions on Contextual Allomorphy 179
In addition to (C1) and (C2), the idea that insertion in complex structures
occurs from the inside out, which was introduced in chapter 4, places further
constraints on contextual allomorphy. This is stated as (C3):
The “deepest” of the conditions (C1)-(C3) is (C1), which holds that mor-
phemes can interact at PF only if they are active in the same phase cycle of
spell out, assuming a syntactic theory of phases along the lines of Chomsky
(2000, 2001). If (C1) is correct, then at least this part of morphology is sen-
sitive to locality domains that are defined by syntactic computation. On the
other hand, (C2) is relatively superficial by comparison, since it invokes linear
order, which, by hypothesis, does not play a role in syntactic computation.
In the remainder of this section, the effects of (C1)-(C3) are outlined and
illustrated.
is simplified in the sense that it does not consider other heads that might occur
between v and n):
The intuition to be implemented is that patterns of this type are phase effects;
to a first approximation, if the head n is in the same cyclic domain with the
Root in derived nominals like (19), but in a distinct domain from the Root
when there is an intervening v as in (20), then important parts of the theory
of allomorphy can be made to follow from the more general theory of how
complex objects are built and spelled out.
The precise way in which phases operate is complex and requires a num-
ber of further assumptions about syntactic derivations. According to a hypoth-
esis that is developed in different forms in Marantz (2001, 2007), Embick and
Marantz (2008), and Embick (2010a), the category-defining heads n, v, a, etc.
play a special role in cyclic derivation by triggering spell out and by defining
the domains that are spelled out (these heads are introduced in section 2.3.4
in chapter 2). For this reason, these heads are sometimes referred to as cyclic
heads.
The pattern illustrated with English nouns in (18) suggests that a cyclic
head cannot see a Root across another cyclic head. However, other types of
morphemes that appear outside of a cyclic head—i.e., different non-cyclic
morphemes—do appear to be able to see the Root across a phase head. The
example of past tense allomorphy in English, which has been considered ex-
Conditions on Contextual Allomorphy 181
�H
�� HH
v T[+past]
p ��HH
ROOT v
Putting these points together, there appear to be two generalizations for the
cyclic part of the theory to account for:
The theory of phases that is developed in Embick (2010a) accounts for these
generalizations in a way that begins with two initial assumptions. The first
specifies under what conditions a cyclic domain is sent to the interfaces; the
second specifies what constitutes a cyclic domain that is spelled out:
Assumption 2: The domains that are spelled out are defined around cyclic
heads and their attendant material. A domain defined by cyclic x in-
cludes (i) x itself; and (ii) non-cyclic heads between x and the cyclic
trigger y; it does not include the cyclic y that triggers spell out of x.
182 Contextual Allomorphy, and Blocking
x Y
p ��HH ��HH
ROOT x x Y
p ��HH
ROOT x
On the other hand, in (26), the cycle in which the Root and x are active does
not contain y, so that when y is spelled out, the Root is inactive:
(26) Outer Cyclic y
y
��HH
x y
p ��HH
ROOT x
With the schemas (24)-(26) at hand, it can now be seen how the cyclic part
of the theory accounts for the important contrasts seen above: that between n
in derived nominals versus gerunds (which falls under Generalization 1), and
that between n in gerunds versus T[+past] in verbs (which falls under Gen-
eralization 2). Starting with the first comparison, n and the Root are clearly
Conditions on Contextual Allomorphy 183
7.3.2 Concatenation
It will be assumed here that the linearization mechanisms that apply to syn-
tactic structures at PF (cf. chapter 3) produce statements of concatenation
which order terminal nodes. For the purposes of looking at the effects of lin-
ear relations on contextual allomorphy, it suffices to begin with an example
like (27a), and the idea that if a structure like (27a) is linearized with X and
Y as suffixes, the concatenation statements in (27b) are derived:
(27) Structure and Linearization
a. Tree
��HH
Y
p ��HH
ROOT X
p
b. ROOT_ X, X _Y
In terms of statements like those in (27b), (C2) says that Vocabulary Items can
refer only to morphemes that are concatenated with the morpheme undergo-
184 Contextual Allomorphy, and Blocking
ing insertion. With this in mind, Vocabulary Items can be represented so that
they explicitly contain information about concatenation. For instance, in the
“outwards looking” allomorphy of the Hungarian plural morpheme consid-
ered in section 7.2, the Vocabulary Items are as follows, with concatenation
shown:
[+pl] $ -(V)k-
The intuition behind (C2) is that a PF-specific linear condition restricts pos-
sible interactions among morphemes. That is, the deep cyclic condition (C1)
defines a set of morphemes that could in principle interact at PF, and this in-
teracts with the PF-specific condition (C2), which places a further condition
on which morphemes can see each other within that cyclic domain.
The condition (C2) operates in ways that interact with further assumptions
about linear order, particularly as it concerns silent morphemes. As empha-
sized at various points in earlier chapters, it appears that many morphemes in
different languages have no overt phonological realization. In the most typi-
cal way of treating this effect, a Vocabulary Item is specified with a null (i.e.
-Ø) exponent. With respect to locality defined in terms of concatenation in
particular, it appears that many morphemes with no overt realization have a
special status. Consider, for example, the English past tense, which has been
examined at various places earlier in the book. The Vocabulary Items that
realize the past tense morpheme are shown in (29), modified now so that con-
catenation is represented:
�HH
�� H
�H T[+past]
p � H
L EAVE v
The issue here, which is discussed in Embick (2003, 2010a,b), is that the Root
and the T[+past] morpheme see each other for allomorphic purposes in spite
of the presence of the (by (31), linearly intervening) v morpheme.
Morphemes like the v in (31) seem to become transparent only when they
have no overt exponent. Thus, one way of analyzing what happens in (31) is
to say that (at least some) morphemes that have a -Ø exponent are deleted,
or Pruned. There are some different ways of thinking about how this Prun-
ing process might be formulated. For instance, if (certain) morphemes are
specified to be Pruned immediately after they are realized as -Ø, they will
be removed from concatenation
p statements. In the case of (31), Pruning v
in this way will make L EAVE and T[+past] immediately adjacent, which
is what is required for contextual allomorphy by (C2). However, rather than
deleting morphemes after they are realized as -Ø, it is also possible that Prun-
ing applies in much more general ways to particular types of morphemes
(something like this is implemented in section 4.6 in chapter 4). Which zero
morphemes are transparent in the way described above, and how the Pruning
of such morphemes should be formalized, are at present open questions.
With the basics of concatenation and its role in allomorphy at hand, I turn
in the next pages to two related dimensions of (C2). The first involves an
illustration of the linear locality effect enforced by this condition. The second
point concerns the idea that allomorphy is restricted linearly in such a way
that syntactic constituency is ignored; that is, concatenation (and therefore
contextual allomorphy) ignores syntactic brackets.
186 Contextual Allomorphy, and Blocking
By (C2), morphemes can interact for allomorphy only when they are con-
catenated. Thus, it should be possible to find cases in which (i) a morpheme
Y shows contextual allomorphy conditioned by a concatenated morpheme
X, but where (ii) Y does not alternate allomorphically when it is separated
from X by a linearly intervening morpheme Z. That is, contextual allomor-
phy should show linear intervention effects.
An example of this type is provided by the Latin perfect tense system;
the discussion here draws on Embick (2010a); see also Carstairs(-McCarthy)
(1987, 2001, 2003), Adger et al. (2003) and Lieber (1992). (Some initial
points about this part of the Latin verb system are analyzed in section 4.6
in chapter 4, but the discussion of this section is self-contained.)
Latin verbs that are marked for perfect aspect come in several types that
are differentiated by their Tense (present, past, future) and Mood (indicative,
subjunctive) specifications. For instance, in addition to the (simple) perfect
forms in (32), there are the pluperfect indicative, perfect subjunctive, plu-
perfect subjunctive, and future perfect forms in (33); to keep the exposition
simple, (32) and (33) do not segment the theme vowel:
(32) Perfect Indicative Forms
p/n form
1s laudā-v(i)-ı̄
2s laudā-v(i)-stı̄
3s laudā-vi-t
1p laudā-vi-mus
2p laudā-vi-stis
3p laudā-v(i)-ērunt
(33) Additional Perfect Forms of laudō ‘praise’
(34) Structure:
T
�HH
�� H
T Agr
�H
� H
�� HH
Asp T
�H
�� HH
v Asp[perf]
p ��HH
ROOT v
I will also assume that Tense comes in three varieties, which, in combination
with the Asp[perf] feature, yield (present) perfects, pluperfects, and future
perfects:
(35) Features of T:
For subjunctive forms, it will be assumed that Tense bears a feature [subj];
this part of the analysis could be modified in different ways, but they are
irrelevant to the main point of this section. This feature combines with those
in (35) to produce perfect subjunctives and pluperfect subjunctives in addition
to the three types of indicative perfect that are listed in (35).
The order of morphemes in Latin perfect forms is as shown in (36):
(36) Root-TH-Asp[perf]-Tense/Mood-Agr
(38) Generalizations
a. Asp[perf]_ Agr ! perfect-specific allomorphs of Agr possible
b. Tense_ Agr ! perfect-specific allomorphs of Agr not found
Taken together, the generalizations in (38) manifest (C2), with (38b) being the
linear intervention effect described abstractly at the beginning of this section.
The analysis of this part of Latin is based on Vocabulary Items that bear
contextual conditions stated in terms of concatenation. In particular, the spe-
cial, perfect-specific Vocabulary Items apply only when Agr is concatenated
with Asp[perf], as shown in (39):1
Conditions on Contextual Allomorphy 189
A second important aspect of (C2) concerns the relationship between the lin-
ear relations it refers to on the one hand, and structurally defined relations
of locality (e.g. sisterhood) on the other. In particular, by (C2), morphemes
can see each other for contextual allomorphy in a way that ignores syntactic
brackets.
Most, if not all, of the examples of contextual allomorphy that have been
analyzed earlier illustrate this point. To take a specific case that has figured in
many of the illustrations earlier, consider the fact that the English past tense
shows Root-specific allomorphy. This means that the T[+past] node must be
190 Contextual Allomorphy, and Blocking
able to see the Root (= be concatenated with the Root when Vocabulary In-
sertion occurs), when the underlying structure is that in (40):
�H
�� HH
T[+past]
p ��HH
ROOT v
In other words, T[+past] can see the Root even though the Root is part of
a constituent that excludes T[+past]; i.e., in spite of the fact that there is an
intervening syntactic bracket.
In a theory with (C2), the Root and T[+past] can see each other in the
linear representation derived from (40) because concatenation, as a linear no-
tion, is not sensitive to syntactic brackets (recall also that the v in this example
is Pruned).
The example (40) does not involve very many brackets between T[+past]
and the Root; especially when we take the Pruning of v is taken into account.
However, it is possible to make the point that contextual allomorphy ignores
syntactic brackets with other relatively straightforward examples. For exam-
ple, consider the verb understand, and its past tense understood; the structure
of the latter is shown (41):
(41) understood
�H
� HH
�� H
��HH T[+past]
� H
�H
� H v
� pH
under S TAND
p
Here, T[+past] must be able to see S TAND in order for the -Ø exponent
of T[+past] to be inserted (moreover T[+past] must be visible to the Root
in order for the Root to be changed morphophonologically from stand to
stood). These morphemes see each other because they are concatenated. The
fact that there is a further
p syntactic bracket in (41) relative to (40) (i.e., in the
constituent [under S TAND]) has no effect on allomorphy, as predicted in a
theory with (C2).
Conditions on Contextual Allomorphy 191
Z
��HH
Y Z
��HH
X Y
p ��HH
ROOT X
Putting to the side the question of Root insertion, assume that the morphemes
X, Y , and Z are suffixes, such that, prior p
to Vocabulary Insertion, the lineariza-
tion mechanisms yield the statements ROOT_ X, X _Y , and Y _ Z. When
Vocabulary Insertion begins, (C3) says that X receives its phonological ex-
ponent before Y . There are two consequences of ordering insertion in this
way. The first is that, if X is sensitive to Y , it can only be sensitive to Y ’s
synsem features, not to its phonological features; this is because Y receives
192 Contextual Allomorphy, and Blocking
its phonological form after Vocabulary Insertion applies to X.2 The second
consequence is for Y , which could in principle be sensitive to either X’s gram-
matical features, or to the features of the phonological exponent inserted at
X.
Taken together, the consequences of (C3) for allomorphy looking out-
wards and inwards are summarized in (43); since the attested cases in (43)
have all been illustrated above, either in earlier chapters or in section 7.2 of
this chapter, the predictions are not illustrated further here:
One point of difference among theories that assume (C3) concerns what hap-
pens to the features of a morpheme that has undergone Vocabulary Insertion.
Recall from the discussion of chapter 4 that Vocabulary Insertion can be for-
malized in various ways: with rewriting of features, substitution, etc. Among
these different options, a key question is whether the synsem features that are
referred to in a Vocabulary Item are deleted when the Vocabulary Item applies
to a morpheme.
In the theory developed in chapter 4, Vocabulary Insertion substitutes an
exponent for a Q variable, with the default assumption being that synsem
features are unaffected by this process. Thus, when a morpheme [a,Q] pre-
insertion is targeted by a Vocabulary Item [a] $ /-x/, the result is [a,-x]
post-insertion. However, as Bobaljik (2000) and others (e.g. Noyer 1997)
have discussed, deleting synsem features on insertion restricts the amount
of information that is in principle capable of being referred to when outer
morphemes undergo Vocabulary Insertion. With reference to (43c) in partic-
ular, the situation is as follows. If Vocabulary Insertion does not delete any
features, then (43c) is as it stands above. But if Vocabulary Insertion (or some
attendant operation) deletes either (i) all of the features on a morpheme, or (ii)
Conditions on Contextual Allomorphy 193
some of the features—e.g., those that are referred to by the Vocabulary Item
that applies—then (43c) does not hold. Rather, it is predicted that inwards-
looking allomorphy should see either (i) no synsem features (when all are
deleted), or (ii) only features that were not deleted during earlier Vocabulary
Insertion (if “partial” deletion is assumed).
For reasons discussed in chapter 4, I have adopted the “no deletion” as-
sumption in this work. But, as can be seen from even this cursory discussion,
a number of interesting directions in the theory of insertion can be explored in
terms of different assumptions about how insertion affects synsem features.
Many of these assumptions produce clear and interesting empirical predic-
tions, both for contextual allomorphy, along the lines outlined here, and also
for morphophonological processes, especially when phonological rules that
appear to refer to synsem features are taken into account. These topics are
active areas of current research.
The idea that Vocabulary Insertion proceeds from the inside out is inde-
pendent of the other conditions (C1) and (C2) that were introduced above. It
is also independent of other parts of the theory, in the sense that it is not clear
that it follows from other assumptions. However, early work in the theory
like Halle and Marantz (1993) connects with earlier models of morphology
and phonology in assuming inside-out application, and the assumption that
Vocabulary Insertion works this way has been examined fruitfully in much
subsequent work (see Bobaljik 2000 for discussion). Whether it can ulti-
mately be understood in terms of other parts of the grammar or not, it remains
an important theoretical proposal, one which is open to further investigation.
7.3.4 Summary
Chapter 4 introduces the idea that blocking exists in the grammar as a rela-
tion among Vocabulary Items. For purposes of review and illustration, recall
that the analysis of past tense verbs is based on the Vocabulary Items in (44)
applying to the T[+past] morpheme in (45):
�H
� H
� H
� H T[+past]
p � H
ROOT v
effects must be able to explain why lef-t and play-ed are grammatical, while
*leav-ed, *lef-t-ed, *play-ed-ed etc. are not.
As discussed in chapter 4, Vocabulary Insertion targets Q positions that are
possessed by morphemes and morphemes only (Terminal Insertion). More-
over, Vocabulary Insertion obeys Uniqueness, such that once the Q of a mor-
pheme has been provided with an exponent, there is no further Vocabulary
Insertion for that morpheme. Finally, Vocabulary Items are Ordered in a way
that determines the winner for application at each morpheme. Taken together,
these assumptions allow for the derivation of lef-t, and rule out the derivation
of the ungrammatical forms considered in the last paragraph. Ungrammati-
cal *leav-ed is not derived because Vocabulary Items compete for insertion,
and, due to Ordering, the Vocabulary
p Item that inserts -t wins this competition
when T[+past] is local to L EAVE. The “double application” seen in *lef-t-
ed is ruled out by Uniqueness: after the Vocabulary Item inserting -t applies
to T[+past] in (45), Vocabulary Insertion is done for that morpheme. Since
there is not a second T[+past] morpheme in the structure (45), there is no
morpheme that the Vocabulary Item with -ed could target a second time, so
that *lef-t-ed is not derived. The same reasoning extends, mutatis mutandis,
to rule out *play-ed-ed etc.
In a theory that implements blocking in terms of Vocabulary Insertion,
all of the action takes place in terms of morphemes. Thus, it is not the case
that the “word” left blocks the “word” *leaved. Rather, the Vocabulary Item
p inserts -t blocks the Vocabulary Item that inserts -ed in the context of
that
L EAVE, in the way described above. The (hypothetical) word leav-ed is
thus not itself blocked. Instead, it is not derived by the rules of the grammar
in the first place.
The reason to be precise about the locus of blocking effects is that block-
ing is often taken to be a relation that obtains between words (or perhaps
larger objects as well). This intuition, with respect to word/word blocking in
particular, stems from the influential work of Aronoff (1976), which looks at
complex patterns in derivational morphology. Moving beyond that particular
empirical domain, the theoretical issues that surround blocking are part of a
larger set of questions about the extent to which competition for grammati-
cality is part of the grammar of human language.3
In order to explicate further the essential properties of the morpheme-
based theory of blocking that is developed in this book, I will outline what is
at issue in putative cases word/word blocking in the rest of this section.
196 Contextual Allomorphy, and Blocking
Chapter 1 introduces the Morpheme Locus Effect: the idea that a number of
crucial grammatical generalizations converge on the morpheme as the key
object of analysis. One of the important manifestations of this effect is in
the idea that Vocabulary Insertion, which applies only to morphemes, derives
blocking. By contrast, many of the most influential discussions of blocking
in the field build on Aronoff (1976) in assuming that blocking relations ob-
tain between words, such that “otherwise expected” forms are ruled out via
word/word competition. In light of this, it is essential that a morpheme-based
theory of blocking be able to account for such generalizations in a way that re-
stricts competition to morphemes. The issues surrounding word/word block-
ing are complex, because they implicate many topics in the theory of deriva-
tional morphology. This section outlines some of the key intutions that have
been implemented in this domain, as a way of introducing readers to more
detailed discussions like Aronoff (1976) and Embick and Marantz (2008).
Aronoff’s (1976) theory of blocking is based on the idea that forms of
words fill cells, in such a way that when a “special” form occupies a cell, it
will prevent an otherwise expected word from being grammatical. The exam-
ple at the center of his argument involves the relationship between adjectives
with -ous and nouns with -ity, as shown in the following table:
an existing nominal, as it is with e.g. glory, then the -ity form based on that
Root is ungrammatical: hence, *gloriosity. It is important to note that on this
approach, *gloriosity is derived by the rules of the grammar, but rendered un-
grammatical by virtue of losing a (word/word) competition with glory. (On
the other hand, nouns in -ness are not listed in cells, because, on Aronoff’s
theory, they are fully regular; thus, they are always grammatical.)
The idea that competitions are waged between words like glory and *glorios-
ity has had a very strong influence on subsequent discussions of blocking. For
the purposes of this chapter, I will outline the analysis of glory/*gloriosity
from Embick and Marantz (2008), which employs blocking defined in terms
of Vocabulary Insertion. The essence of this analysis connects with the dis-
cussion of blocking earlier in this section: that is, in the same way that forms
like *leav-ed and *lef-t-ed are not derived (and thus do not need to be blocked),
the idea is that *gloriosity is not derived in the first place.
To start with, a simple noun (“nominal”) like glory is internally complex
in a Root-based theory; it consists of a Root and an n morpheme:
(47) glory-Ø
n
�H
p �� HH
G LORY [n,-Ø]
With this analysis of nouns in mind, there are two different structures to con-
sider for the (ungrammatical) abstract noun *gloriosity. The first structure,
(48), is a simple noun of the type that is considered in (47). The second struc-
ture is a noun that is derived from an adjective, and is shown in (49):
n n
p ��HH ��HH
ROOT n a n
p ��HH
ROOT a
198 Contextual Allomorphy, and Blocking
(50) glorious
a
�H
p �� HH
G LORY [a,-ous]
The salient question is what happens when this structure is nominalized with
n, along the lines of (49).
In order to answer this question, it is necessary to look closely at the dis-
tribution of the -ity exponent of n, since it forms part of *gloriosity. As shown
in the first Vocabulary Item in (51), -ity realizes n when n is local to a number
of different morphemes, both Roots and functional heads. For purposes of
comparison, -ness, which is the exponent of a Vocabulary Item that does not
have these contextual conditions, is shown as well:
p p
n $ -ity /{ S OLEMN, C URIOUS,...[a,-able]...}_
n $ -ness
The contextual condition for the -ity Vocabulary Item contains a number of
Roots. These are Roots that form an -ity nominal when n is Root attached;
that is:5
Blocking, Morphemes, and Vocabulary Insertion 199
a n
�H �H
p �� HH p �� HH
S OLEMN [a,-Ø] S OLEMN [n,-ity]
Thus far it has been established that the Vocabulary Item inserting -ity applies
to n only in the context of a highly restricted set of elements, as reflected in
(51). Returning to *gloriosity, it is important to observe that the list for the
insertion of -ity does not contain the a head with the exponent -ous. This point
is especially clear in forms like feroc-ious, feroc-ity, which show that -ity is
Root-attached when it forms abstract nouns:
a n
�H �H
p �� HH p �� HH
F EROC [a,-ous] F EROC [n,-ity]
(56) ferociousness
n
�H
�� HH
� H
a [n,-ness]
�H
� H
p � H
F EROC [a,-ous]
(57) gloriousness
n
�H
� �� HH H
a [n,-ness]
�H
p �� HH
G LORY [a,-ous]
Bringing the different components of this discussion together, then, the main
points are as follows. The task that was outlined at the beginning of this sec-
tion was to derive an analysis of the facts that motivated word/word blocking
in Aronoff (1976), but with blocking restricted to interactions among Vocabu-
lary Items. With this in mind, the analysis developed in this section
p concen-
trates on two possible structures forp*gloriosity, a basic noun [ G LORY n]
and a derived (deadjectival) noun [[ G LORY a] n].
In the first structure, the n head is realized as -Ø; assuming that there is
nop-osity exponent of n, *gloriosity is not derived. For the second structure,
[[ G LORY a] n], the focus is on the Vocabulary Item that inserts -ity. This
Vocabulary Item does not apply to n when it is outside of the a that is pro-
nounced -ous. Rather, the Vocabulary Item with -ness applies to n morphemes
that are attached to that a head, because the -ity-inserting Vocabulary Item ap-
plies only in the context of a specific list of morphemes.
p It therefore follows
that *gloriosity is not the output of the structure [[ G LORY a] n]. Instead,
gloriousness is derived.
Taken together, then, this analysis provides structures and Vocabulary Items
for the grammatical forms (glory, glorious, gloriousness, etc.). Moreover, it
does not derive the ungrammatical forms. And, crucially, it does this in terms
of the interaction of Vocabulary Items applying to morphemes, which is to
say that it does not involve words blocking other words.
There are many more things to be said about the full analysis of the facts
considered in Aronoff’s discussion, and about how morpheme-based block-
ing versus word-based blocking can be compared with each other. In addition,
there are a number of implications of the analysis outlined here that warrant
careful consideration.6 For detailed discussion of these matters, and for ar-
guments why morpheme-based blocking is superior to word-based blocking
when other phenomena are considered, see Embick and Marantz (2008) and
Embick (2010a).
Non-Affixal Morphology and Blocking 201
In concluding this chapter, I will introduce a topic that looks at a first glance
to provide a challenge for a morpheme-based morphology. The earlier parts
of this book are focused on morphemes and their phonological realization
via the Vocabulary Insertion operation. It is clear from even the most cur-
sory descriptions of what “morphology” is about, however, that there are
other phenomena at play in this part of language as well. In particular, most
languages show what is called (descriptively speaking) non-affixal or non-
concatenative morphology. Often, though not always, non-affixal morphol-
ogy takes the form of a change to a phonological representation, in a way that
is correlated with a particular morphological context. To take a specific ex-
ample, at many points above I have used the English past tense as a test case
for the illustration of the mechanics of Vocabulary Insertion. The last time
this system was looked at, in section 7.4 of this chapter, the working analysis
employed the Vocabulary Items that are shown in (58):
(58) Vocabulary Items
p p
a. T[+past] $ -t/{ B END, L EAVE,...}_
p p
b. T[+past] $ -Ø/{ H IT, Q UIT,...}_
c. T[+past] $ -ed
Although it correctly determines the different realizations of the T[+past]
morpheme in English, the analysis summarized in (58) does not take into
account forms like gave, sang, and so on, where, according to a standard type
of description, the verbs give and sing show stem allomorphy (or a stem al-
ternation) in the past tense. This is one of the phenomena that falls under the
heading of “non-affixal morphological change”.
The kind of stem allomorphy seen in sang etc. is only one type of non-
affixal morphology. Cross-linguistically, other phenomena, such as the tem-
platic effects seen in Semitic (recall the brief discussion of Arabic in chapter
2), would be part of a broader overview of this part of morphophonology.
However, the aims of this section are quite narrow: they are to show why it
is that theories like Distributed Morphology hold that many types of stem al-
lomorphy are not derived by Vocabulary Insertion, but are instead the result
of phonological operations. Importantly, this particular facet of the theory is
based on the behavior of non-affixal morphology with respect to blocking,
and thus follows directly on the discussion of the last section.
202 Contextual Allomorphy, and Blocking
In the next subsections, I will first outline an analysis of the sing/sang type
of alternation (section 7.5.1), and then look at why it is problematic to treat
non-affixal morphology of this type on a par with morpheme-based Vocabu-
lary Insertion (section 7.5.2). Section 7.5.3 concludes with some general re-
marks on the relationship between (morpho)phonological rules and Vocabu-
lary Insertion.
In extending the analysis in (58) to cover sang etc., there are two main facts
to bep accounted for. The first is that the default VI with -ed does not apply
with S ING; this is clear from the fact that *sing-ed is ungrammatical (as is
*sang-ed). The second effect to be accounted for is the stem form (where stem
is used descriptively for “phonologicalpform of a Root”): i.e., the analysis
must account for the fact that the Root S ING surfaces as sing in the present
tense, and sang in the past tense.
For reasons that are discussed in detail in 7.5.2 below, the standard anal-
ysis of these two types of effects in Distributed Morphology, which is de-
veloped in detail in Halle and Marantz p(1993), takes the blocking of -ed and
the change that produces sang from S ING to be the result of two distinct
mechanisms. The first one, blocking the Vocabulary p Item that inserts-ed, is
attributed to Vocabulary Insertion. Specifically, S ING is entered on the list
of the Vocabulary Items that take the -Ø exponent of past tense:
in sing, a Readjustment Rule (or a set of such rules) that is triggered by the
T[+past] morpheme changes the stem vowel to /æ/. In addition to being trig-
gered by T[+past], this rule must also be stated so that it applies only to cer-
tain Roots and not others. I am being neutral about the precise phonological
details involved in the sing/sang alternation because there are many ways in
which the alternation could be analyzed (see Halle and Mohanan (1985) for
one). Moreover, one need not assume that the underlying representation is the
one seen in sing. What is important is that the analysis separates Vocabulary
Insertion pat T[+past] from the change effected to the phonological form of
the Root S ING, as summarized in (60):7
��HH
�� HH
v T[+past,-Ø]
�HH
p �
S ING [v,-Ø]
b. Readjustment
p Rule(s): change vowel in phonological representation
of S ING with rules triggered by T[+past].
The analysis outlined in the preceding subsection treats the derivation of sang
with two distinct mechanisms: Vocabulary Insertion accounts for the absence
of the default -ed, and a Readjustment Rule accounts for the form of the stem.
The general claim behind this analysis is that morpheme-based Vocabulary
Insertion and (at least this type of) morphophonological change is executed
by different operations in the grammar. This section looks at the motivations
for making this distinction.
As a starting point, it is important to emphasize that the motivations are
empirical, and not conceptual. All other things being equal, it would be prima
204 Contextual Allomorphy, and Blocking
Taken together, these facts show that there is not always a stem change that
blocks an otherwise expected morpheme.
Although suggestive, the argument outlined in the last paragraph is still
relatively weak. It shows only that blocking of Vocabulary Items does not re-
quire a stem change—and, to the extent that we allow -Ø exponents, as we
have at several points in this book, this is relatively unsurprising. But there are
more interesting things to say about how morphemes and non-affixal changes
interact for blocking. This brings us to the more important and general ar-
gument in favor of distinguishing piece-based realization from non-affixal
changes. Rather than looking at whether non-affixal changes always block
expected Vocabulary Items, as we did immediately above, we can now look
at how non-affixal changes relate to overt realizations of morphemes. The
pertinent questions can be stated in two ways, since in principle non-affixal
changes could block affixation, or vice versa; that is:
(62) Questions:
a. Do non-affixal changes triggered by a feature X necessarily block
overt realization of X with an affix?
b. Does realization of morpheme X by an affix necessarily block non-
affixal changes triggered by X?
If the questions in (62) were answered positively, it would mean that mor-
phemes and non-affixal changes always block each other. This would be
strong evidence in favor of treating them with a single mechanism. For this
reason, the questions in (62) play a prominent role in Halle and Marantz’s
(1993) critique of Anderson’s (1992) single mechanism theory. In particular,
Halle and Marantz argue that there are strong reasons to think that non-affixal
changes and Vocabulary Insertion do not block each other. Continuing with
verbs (and participles) in English, this point can be seen in the fact that there
are many Roots which show both stem changes and overt affixation in their
past tense and participial forms:
For the reasons that were developed throughout this section and the last, these
generalizations can be accounted for in a principled way in a theory that treats
affixal and non-affixal morphology with distinct operations (Vocabulary In-
sertion and Readjustment Rules in Distributed Morphology), but not in a the-
ory that eschews this distinction.
The idea that Vocabulary Insertion and Readjustment Rules account for af-
fixal and non-affixal morphology respectively has been explored in a number
Non-Affixal Morphology and Blocking 207
The second type of analysis dispenses with the Readjustment Rule, and em-
ploys only Vocabulary Insertion. This analysis posits that the exponent of first
person singular Agr is an autosegment:
The floating [+nasal] feature, which is in the (prefixal) position of the Agr
morpheme, is then manipulated phonologically to produce the distributions
seen in (65). (This manipulation might involve a rule that is identical to the
Readjustment Rule required in the first solution).
Choosing between these analyses is not simple. The first type of analysis
employs tools that are clearly required elsewhere, since, as we saw above in
the discussion of English, both Readjustment Rules and Vocabulary Insertion
have independent motivation. Recall, in particular, the double marking seen
in e.g. tol-d. It might be possible to treat e.g. sang by inserting for T[+past]
some autosegment(s) to (i) derive sang, and (ii) block the insertion of -ed. But
it is not possible to derive the stem change in doubly marked forms like tol-d
in this way, since the stem change is effected in the presence of the default
-ed exponent of T[+past].
The second type of analysis, the one in (68), employs only Vocabulary
Insertion, and is thus, according to at least one metric, simpler than the first.
At the same time, though, allowing the exponents of Vocabulary Items to be
autosegments, rather than restricting exponents to larger objects, constitues a
modification to the theory that warrants careful investigation. Further study is
required to determine the predictions made by different theories of possible
exponents, and to determine which of these predictions are correct.
The general point that is at issue in this discussion, where to draw the line
between Vocabulary Insertion and (morpho)phonology, is an important topic
in current research (recall section 7.2.3 above). While it is not possible to
report on a consensus view at this point in time, the discussion above is, I
hope, sufficient to clarify what is at stake in this part of the theory.9
7.6 Conclusions
Contextual allomorphy and blocking are two of the most important topics
in morphological theory. They fit together, in the sense that one aspect of
contextual allomorphy is that a more specific Vocabulary Item must win out
over—i.e., “block”, in an intuitive sense—less specified competitors in par-
Conclusions 209
A common way of thinking about how grammars are organized holds that
sentences are built out of words, and that words are built out of morphemes.
The theory that is developed in this book takes another perspective: sentences
are built out of morphemes, and words (to the extent that they are needed at
all) are built out of syntactic structures; i.e., they reflect particular way(s) in
which morphemes are packaged.
In the particular morpheme-centered view that is developed in this book,
there are two types of morphemes in the grammar: the Roots and the func-
tional morphemes. The main emphasis of the book is on how sound and
meaning connections are established in terms of these morphemes, with a
particular emphasis on functional morphemes and Vocabulary Insertion.
The interplay of sound, meaning, and structure constitutes the fundamen-
tal issue in the theory of the morpheme. The particular interactions at the
center of this book are two phenomena showing that sound/meaning connec-
tions are not always one-to-one: allomorphy and syncretism. Allomorphy is
a departure from the one-to-one ideal because it shows that a single synsem
object can be realized in ways that are not phonologically related to one an-
other; i.e., with suppletive allomorphs. Syncretism is a departure from the
one-to-one ideal because it shows that a single phonological exponent can
realize distinct sets of synsem features.
Following the general intuition outlined in chapter 1, my goal has been
to present the basic properties of a theory in which sound and meaning are
connected transparently in the default case, but in which there is enough
flexibility to allow for limited departures from the one-to-one ideal, so that
constrained theories of allomorphy and syncretism can be developed. The
Vocabulary Insertion operation, which is at the center of chapter 4, and devel-
oped in subsequent chapters, serves this function by providing a mechanism
that adds phonological representations to morphemes. Because they can be
underspecified with respect to the morphemes that they apply to, Vocabulary
Items are able to account for syncretism; and because Vocabulary Insertion
targets morphemes at PF, after they have been assembled into complex ob-
jects, contextual allomorphy can be investigated in terms of different types of
locality conditions (phase-cyclic, linear) that are part of syntax and PF.
212 Concluding Remarks
Along the way, several additional elements of the theory are introduced
in addition to Vocabulary Insertion (e.g. Impoverishment rules, which are ex-
amined in chapter 6). Moreover, some questions about the scope of Vocabu-
lary Insertion are left unresolved; in particular, whether it applies to all mor-
phemes (including Roots); only to functional morphemes; or to some subset
of morphemes. With respect to these topics (and a number of related ones
that surface in the pages above), my goal has not been to try and provide
definitive arguments in favor of one view or another. Rather, I have tried to
provide enough theoretical and empirical context for understanding the im-
portant questions, so that interested theoreticians can appreciate what is at
issue, and direct their attention to the active primary literature in these areas.
One reason that I have not attempted to provide strong conclusions on these
topics is that the core topic of the book—Vocabulary Insertion, and its rela-
tion to syncretism and allomorphy—is compatible with a number of auxiliary
theories of related phenomena, and it is not my intention here to determine a
comprehensive theory that includes all of these.
My hope is that by outlining the core properties of a theory of the mor-
pheme, and that by taking an open approach towards a number of attendant
questions, the book will serve as a stimulus for further work in this part of the
theory of grammar. I also hope that in addition to helping readers understand
how to do things with the theory in one form, it will help them to undo it
(and redo it, and preferably outdo it) in constructive ways. Every aspect of
an active theory is subject to continuous question and re-examination; and to
the extent that presenting things as I have done here leads to innovations and
changes to the theory—even if this means fundamentally changing or reject-
ing parts that I consider now to be the most central and the most correct—the
book will have justified its existence.
Fusion 213
A.1 Introduction
Early work in Distributed Morphology introduced two types of rules that af-
fect the number of morphemes that are targeted by Vocabulary Insertion (see
in particular Noyer (1992) and Halle and Marantz (1993)). The rules are Fu-
sion, which creates one morpheme out of two; and Fission, which creates two
morphemes out of one. To a first approximation, these rules are motivated by
two different scenarios:
Some simple illustrations of Fusion and Fission rules are given in sections A.2
and A.3. These rules are controversial; there is some question as to whether
their effects can be reduced to independently motivated mechanisms (see e.g.
Trommer 1997). Since my intent here is primarily to illustrate what these
rules do, as an introduction to the primary literature, I will not dwell on pos-
sible reductions to other mechanisms here.
A.2 Fusion
Rules for Fusing morphemes prior to Vocabulary Insertion were initially dis-
cussed in Halle and Marantz (1993). As described in Case 1 above, Fusion is
motivated when what appear to be two distinct morphemes are compressed
214 Concluding Remarks
stance, the structure of the 1s Passive is shown in (4), which treats the theme
vowel as a realization of v, and omits the present tense morpheme (in this
particular form, laudor, the theme is deleted phonologically):
(4) Verb Structure
�HH
�� H
�� HH
�H [+pass,-r]
�� HH
� H
�H [+1,-2,-pl,-ō]
p � H
L AUD [v,-ā]
Other person/number combinations can be treated similarly, with one excep-
tion: in the 2pl, there is no relationship between passive -minı̄ and active -tis;
moreover, the passive -r that appears elsewhere in the system is absent here.
Thus, whereas each of the other Agr forms consists of the agreement found in
the active in conjunction with an -r component, it seems that the 2pl passive
is realized with only a single exponent, -minı̄.
The interaction of Agr and [+pass] in Latin falls under Case 1 of (1). The
two separate morphemes that are at issue in the majority of the system, Agr
and [+pass], are well motivated, and they are realized in most person/number
combinations as separate objects. However, in one specific feature combina-
tion, they are apparently realized together, such that a Vocabulary Item with
the exponent -minı̄ blocks both the -r-inserting passive Vocabulary Item and
the Vocabulary Item with -tis that realizes second person plural Agr.
Mechanically, Fusion creates one morpheme (with one Q position) from
two by combining the features of the two morphemes prior to the appli-
cation of Vocabulary Insertion. A rule for fusing second person plural Agr
with [+pass] in this way is shown in (5); in terms of locality, it can be assumed
that morphemes have to be concatenated in order to be fused:
(5) Fusion Rule for Latin
[+2,+pl,+pass] $ -minı̄
[+2,+pl] $ -tis
[+pass] $ -r
The most specific item that can apply to the fused node in (6) is the first one,
which inserts the -minı̄. This Vocabulary Item blocks both the one with -tis
and the one with -r, so that its special (“portmanteau”) exponent appears at
the expense of the two expected ones.
A.3 Fission
Noyer (1992) argues for morpheme “splitting” rules whose effects are in
some ways the inverse of Fusion. In Halle and Marantz (1993), such rules
are referred to as Fission rules.
Schematically, the effects of Fission can be illustrated as follows. Sup-
pose that we have a language that shows agreement for person, number, and
gender (±P, ±N, ±G) on verbs. Suppose moreover that there are reasons for
assuming that these features originate in a single morpheme (i.e. an agree-
ment morpheme), as shown in (7):
A Fission rule takes a single morpheme like this Agr and splits it into two
morphemes. For instance:
An important question concerns how Fission rules are ordered with respect
to Vocabulary Insertion. Early motivations for Fission (provided by Noyer
Fission 217
(1992) and Halle (1997), as well as Halle and Marantz (1993)) were closely
connected with the idea that morphemes first undergo Vocabulary Insertion,
with Fission then applying to the “left over” features; i.e., those that are not
affected by that first application of Vocabulary Insertion. (For this reason, Fis-
sion rules often work together with a theory of feature deletion or discharge
on insertion; recall the discussion of chapter 4). However, the phenomena mo-
tivating this type of “interleaved” Fission are too complex to illustrate here;
interested readers should consult the primary works.
For purposes of illustration, I will draw on an example in which Fission
is applied before Vocabulary Insertion, as this suffices to illustrate the basic
properties of this type of rule.
A simple case study is provided by the case inflection of the Australian
language Djabugay (see Hale (1976a,b) and Patz (1991); for the analysis pre-
sented here, see Embick (2010a)). Like many languages of Australia, this
language shows a great deal of phonologically conditioned allomorphy in the
case affixes that attach to nouns. Of particular interest in this connection are
the genitive and dative cases, which are shown in (9). The genitive alternates
between -:n (the colon because it lengthens a preceding vowel) and -Nun be-
fore V-final and C-final hosts, respectively. The dative case alternates in a
similar way; in fact, it appears to be simply the genitive exponent, followed
by an invariant -da:
(9) Djabugay Genitives and Datives
that labels for cases like genitive and dative are shorthand abbreviations for
combinations of features of the type shown in (10), which is taken from Halle
and Vaux (1998):
genitive dative
obl(ique) + +
str(uctural) + +
sup(erior) - +
free + +
It can be seen in (10) that genitive and dative case morphemes have a large
amount of shared feature content: they differ only in the value of [±sup].
To derive the facts in (9), there are two scenarios to consider. The first of
these involves what we can think of as “genitive syntax”, where the case mor-
pheme has the features shown in the Genitive column of (10). The second sce-
nario to consider is “dative syntax”: when this is found, the case morpheme
possesses the features shown in the dative column of (10).
The second scenario is the primary one of interest from the perspective
of Fission. As noted above with reference to (9), the fact to be accounted for
in datives is the presence of an (-:n or -Nun) morpheme—i.e., the exponent
realized in the genitive—followed by -da. With this in mind, consider the rule
of Dative Fission in (11), which applies to dative morphemes to create two
morphemes that are each subject to Vocabulary Insertion:
[+obl,+str] $ -:n /V
[+obl,+str] $ -Nun /C
[+sup,+free] $ -da
The most specific Vocabulary Items in (12) that can apply to the first mor-
pheme in (11) are the ones that refer to [+obl, +str]. In this first position,
then, the exponents -:n or -Nun are realized, depending on the phonological
Fission 219
context. For the second morpheme that is the output of Fission, with the fea-
tures [+sup, +free], the last Vocabulary Item inserts -da.
With genitive syntax, the case morpheme starts with the features [+obl,
+str, +sup, +free]. The Fission rule does not apply to this morpheme. The
most specific Vocabulary Items that can apply to it are also the first two in
(12); thus, genitive syntax yields a single morpheme with the exponent -:n or
-Nun, depending on the phonological context.
The analysis of the two cases considered above is summarized in (13-14):
The use of Fission in this analysis is made necessary by the assumption that
there is a single case morpheme on nouns prior to Vocabulary Insertion. Fis-
sion then provides a principled mechanical account of why it is that syntactic
datives are realized in a way that consists of two morphemes. Of course,
if there were substantial morphosyntactic evidence from other domains (or
other languages) for holding that “cases” of the relevant type are actually de-
rived from two morphemes in the normal case, then Fission would not be
required.
Notes
Notes to Chapter 1
4. This sense of spell out is distinct from the use of this term in syntactic
theory, where it refers to the process that sends syntactic representations to
the interfaces.
5. The claim that word formation precedes syntax produces some predictions
about the ordering of lexical and syntactic processes, of a type that are ex-
plored in works like Wasow (1977) and Bresnan (1982). These predictions
continue to be discussed in several domains; for example, in the analysis
of the differences between so-called adjectival and verbal passives, which
plays an important role in Wasow’s paper. See Embick (2004) for a non-
lexicalist point of view.
6. A useful overview of many of these themes can be found in Carstairs-
McCarthy (1992), chapter 2. Carstairs-McCarthy traces the development of
lexicalist theories from responses that arose to early work in morphology
by Halle (1973), which posits three distinct lists in the grammar. This move
is reacted to in various ways by lexicalist theories like Jackendoff (1975),
Aronoff (1976), Lieber (1980), and di Sciullo and Williams (1987).
7. See, for example, Pinker and Prince (1988), and many subsequent theories
in this vein, particularly those on the “words and rules” side (cf. Pinker and
Ullman (2002)) of the so-called past-tense debate (cf. Marslen-Wilson and
Tyler (1998)).
8. In a theory that allows Vocabulary Insertion for Roots, it might be possible
(depending on further assumptions about contextual allomorphy) p to hold
that sang is part of a Vocabulary Item that applies to a Root S ING in
the context of the past tense morpheme. On this approach, sing and sang
are treated as instances of suppletive allomorphy, along the lines of what
is seen in go/went. Although an analysis of this type might be necessary
for some alternations shown by Roots (i.e., if there is true Root supple-
tion), it should be approached carefully. Generalizing suppletion ignores
the shared phonological properties of sing and sang. More generally, sup-
pletion should probably be treated as a sort of “last resort” analytically; see
Embick and Halle (2005) and Embick (2010b) for some pertinent discus-
sion.
9. In fact, although underspecified Vocabulary Items like those in (17) are re-
quired for the analysis of Latin American Spanish, the connection between
second and third plural in Spanish is even deeper than the analysis in the
text indicates; see in particular section 6.3.3 in chapter 6.
10. Strictly speaking, theories that add phonological representations to mor-
phemes in this way this are only one type of realizational theory. There are
Notes 223
other realizational theories that have rules for realizing synsem features,
but that do not employ morphemes (e.g. Anderson 1992).
Notes to Chapter 2
the one-word Spanish type is called synthetic, while the two-word English
type is called analytic.
6. With respect to the UFI, it is safe to say that the features of category-
defining heads are understudied relative to the features of other domains
(e.g., tense, person/number, etc.) mentioned in section 2.2. Although most
theories seem to assume that a number of different features are active in
this domain, no comprehensive proposals have emerged in the literature.
7. For additional discussion of this matter see the notion of Pruning that is
discussed in chapter 7.
8. In addition to how polysemy arises from Root-categorizer interactions,
there is a further question of what to do when a Root is unacceptable in
a particular grammatical category; see section 2.3.4.
9. For discussion of Roots and templates see Arad (2005) subsequent work.
There is also an extensive experimental literature (psycho- and neurolin-
guistic) examining Root-based representation; see e.g. Prunet et al. (2000)
for representative discussion.
10. While the text and the examples employed above speak of diacritic features
as properties of Roots, such specification is not limited to Roots. For ex-
ample, the adjectival or participial head Asp[pres] in Latin that is realized
as -nt inflects as an adjective of the third declension; this information has
to be associated with this functional morpheme (or perhaps its exponent)
diacritically.
Notes to Chapter 3
1. For some discussion of this idea see Chomsky (1995:334 ff.), where it
is framed with reference to the idea that linear order appears to have no
effects in narrow syntax or LF.
2. As a notational convention, the label of non-terminal nodes in complex
heads will often be left out of tree structures in other parts of this book.
3. For the idea that Agreement morphemes are not syntactic objects, see Ia-
tridou (1990) and Chomsky (1995).
4. Regarding the option without head movement, Koopman (2005) and others
have argued that phrasal movement (augmented perhaps with a kind of
phonological “leaning” of morphemes) is responsible for the creation of
objects of morphological interest. Theories of this type need to be able to
give explicit accounts of the relationships between syntactic structures and
domains of phonological interaction (see section 3.5 below). But as long
Notes 225
several cyclic domains; see Marantz (2007) and Embick (2010a) for perti-
nent discussion. —)
11. Possible exceptions to the M-Word/phonological word correspondence are
not easy to find; Poser (1990) considers a potentially relevant example.
Notes to Chapter 4
Notes to Chapter 5
1. Of course, not all realizational theories share all of the assumptions that
are made in this work. In particular, many realizational approaches follow
Anderson (1992) in denying the existence of morphemes. See chapter 7 for
some discussion of this point.
2. The Spanish agreement example used for illustration in chapter 1 and sec-
tion 1 of this chapter is examined in further detail in chapter 6.
3. The glosses show underlying phonological representations, which are in
some cases altered in the surface forms by phonological rules. I focus in
this example only on the first and second person forms, as third person
morphemes have a slightly different status in the system (they involve de-
ictic and other semantic distinctions). Here and below, ADV = “adverb”;
228 Notes
14. Other facts from the language may provide some guidelines. For example,
Beeler (1976:255) notes that -wun also appears as a nominal plural with
certain elements (e.g. he-ẃun ‘these’, ho-ẃun ‘those’, čiči-ẃun ‘children’),
suggesting that it is a general purpose plural; this fits with (31), which treats
-wun as the default non-singular.
15. On the theme of natural classes and the [+part] feature, it should be noted
that both of the analyses in the text make use of the feature [-sg]. There
might be better ways of defining a natural class of “non-singulars”, but
these require a more detailed look at number systems. See Noyer (1992)
and Harbour (2008) for some relevant discussion.
16. While the class that the Root belongs to determines which set of possessive
endings appear, it does not determine how the theme morpheme surfaces.
Rather, the realization of the morpheme depends on the identity of the
Root; thus, the same theme exponent can be found in different classes:
So, for example, ei ‘eye’ and aN ‘firewood’ both have the -l themes, but the
former takes class 1 possessor inflection, and the latter class 2; kaua ‘dog’
and pom ‘chest’ both show -k themes, but the former is in class 2 while
the latter is in class 3. Thurston speculates (1982:37) that there are twenty
categories defined by theme realization combined with class behavior, but
makes no claims that his description is exhaustive (1982:87). Nouns also
have a grammatical gender, which does not relate in any predictable way
to theme class or possessive inflection class (1982:46). A further compli-
cation is that verbs show object agreement using the sets of possessive
suffixes illustrated in the text. Verbs, like nouns, show both unpredictable
theme elements and condition sets of agreement affixes in an unpredictable
way. As Thurston notes (1982:43) “[f]or the outsider, the morphology of
these suffixes gives a nightmarish quality equally to nouns and verbs in
Anêm.”
230 Notes
17. The relationship between Roots, themes, and possessor allomorphy is com-
plicated. In part this is motivated by the way in which the theme compo-
nent relates to the meaning of the Root. It appears to be the case that—at
least for some Roots—the theme functions as a sort of classifier, such that
the same Root may appear with different themes, each associated with a
different interpretation; e.g. aba-k-e is ‘my pig’, whereas aba-i-at is ‘my
pork’. Possessor class changes with the change in theme and interpretation
as well, in a way that could shed light on the nature of features like [I], [II],
etc.
Notes to Chapter 6
1. See Halle and Marantz for why two -e Vocabulary Items are posited for the
locative and the vocative.
2. Following Halle and Marantz, (16) simply enumerates the noun class fea-
tures contextually in these rules. Alternatives might involve decomposing
class features like 1, 2, etc. into primitive features, which are then referred
to in Impoverishment rules (Alexiadou and Müller (2008) treats inflec-
tional class features in this way).
3. The five nouns in (20) instantiate the five declensions, but represent a sub-
set of the full set of noun types that are found in the language. For the
purposes of this illustration I am putting to the side further complications
that are introduced when nominal gender and some other morphophono-
logical factors are taken into account.
4. The encoding of Tense and Aspect features could be changed in various
ways (e.g. by using binary [±perf]); the main points of this example are
not affected by these different options.
5. In (32), the context for the -v Vocabulary Item is given as either [+1] or
[+2]. With more developed assumptions about the representation of person
features (e.g., using a feature like [+participant] for [+1] and [+2] cate-
gories; recall section 5.5.3 in chapter 5), the disjunctive contextual condi-
tion on the -v allomorph could be removed and replaced by reference to a
single shared feature.
6. I have not included zero realizations. The past III is described as showing
a portmanteaux (“fused”) realization of tense and agreement. Even though
the overt forms are sometimes quite similar to overt agreement morphemes
elsewhere I have not attempted a more detailed segmentation.
Notes 231
Notes to Chapter 7
1. This analysis puts to the side some complications with the -ō and -m real-
izations of first person singular (in particular, why the latter appears with
both past tenses and subjunctives). See Footnote 7 of chapter 4.
2. Of course, later in the derivation, in the phonology, the phonological ex-
ponent of X could be affected phonologically (or morphophonologically)
by the phonological exponent of Y , when phonological rules apply. This
would be an alternation in the position of X induced by Y , but it would not
involve Vocabulary Insertion per se (i.e., it would not be suppletive con-
textual allomorphy). Some of the apparent cases of allomorphy looking
“outwards” to the phonology discussed in Carstairs (1987), for example,
appear to be caused by morphophonological rules operating in the manner
just described.
3. Beyond the issue of whether competition is found with word/word in-
teractions, another important line of discussion in the field asks whether
there are competitions for grammaticality among larger objects (e.g., words
versus phrases, phrases versus phrases, etc.) as well. See, for example,
Poser (1992), Andrews (1990), Hankamer and Mikkelsen (2005), Kiparsky
(2005), Embick (2007a, 2008), and, for a general discussion, Embick and
Marantz (2008). Another important question concerns how competition in
phonological theory relates to the themes that are examined in this section;
see the introductory discussion of Embick (2010a) for some comments.
4. Although there is an emphatic sense of -osity that appears to have entered
English—so that instances of gloriosity can be found on the web—this is
a marginal type of derivation.
5. The noun solemnness also exists in addition to solemnity. In p this word, the
n head realized as -ness is attached outside of an a head: [[ S OLEMN a]
n]. This outer n head cannot see the Root due to these morphemes being
in different phase cycles, as discussed earlier in this chapter. Thus, it is
realized as -ness, not -ity.
6. For example, it follows from the theory presented above p that the adjec-
tive curious is (on the simplest analysis) the realization of [ C URIOUS a],
where the a has the exponent -Ø, not -ous.
7. Although the structure in (60a) shows the v head (and a -Ø exponent for it),
recall
p from above that this head is Pruned, so that T[+past] and the Root
S ING are concatenated when Vocabulary Insertion applies to T[+past].
8. Unless, for example, blocking relations between morphological changes
232 Notes
Halle, M., and A. Marantz (1993) “Distributed Morphology and the Pieces
of Inflection,” in K. Hale and S. Keyser, eds., The View from Building
20: Essays in Linguistics in Honor of Sylvain Bromberger, MIT Press,
Cambridge, MA, 111–176.
Halle, M., and A. Marantz (1994) “Some Key Features of Distributed Mor-
phology,” in A. Carnie, H. Harley, and T. Bures, eds., Papers on Phonol-
ogy and Morphology, MITWPL 21, Cambridge, MA, 275–288.
Halle, M., and A. Marantz (2008) “Clarifying ‘‘Blur”: Paradigms, defaults,
and inflectional classes,” in A. Bachrach and A. Nevins, eds., Inflec-
tional Identity, Oxford University Press, Oxford and New York, 55–72.
Halle, M., and K. P. Mohanan (1985) “Segmental Phonology of Modern
English,” Linguistic Inquiry 16, 57–116.
Halle, M., and B. Vaux (1998) “Theoretical Aspects of Indo-European Nom-
inal Morphology: The Nominal Declensions of Latin and Armenian,” in
Mı́r Curad: Studies in Honor of Calvert Watkins, Innsbruck.
Halle, M., and J.-R. Vergnaud (1987) An Essay on Stress, MIT Press, Cam-
bridge, MA.
Halpern, A. (1992) Topics in the Placement and Morphology of Clitics, Doc-
toral dissertation, Stanford University.
Hankamer, J., and L. Mikkelsen (2005) “When Movement Must be Blocked:
A Response to Embick and Noyer,” Linguistic Inquiry 36:1, 85–125.
Harbour, D. (2008) Morphosemantic number: From Kiowa noun classes to
UG number features, Springer, Dordrecht.
Harbour, D. (2010) “Descriptive and explanatory markedness,” Morphology.
Harley, H., and E. Ritter (2002) “Person and number in pronouns: A feature-
geometric analysis,” Language 78:3, 482–526.
Harris, J. W. (1991) “The exponence of Gender in Spanish,” Linguistic In-
quiry 22:1, 27–62.
Harris, J. W. (1998) “Spanish imperatives: syntax meets morphology,” Jour-
nal of Linguistics 34, 27–52.
Hayes, B. (1990) “Precompiled Phrasal Phonology,” in S. Inkelas and D. Zec,
eds., The Syntax-Phonology Connection, University of Chicago Press,
Chicago, 85–108.
Hyman, L. (2003) “Suffix ordering in Bantu: A morphocentric approach,”
Yearbook of Morphology 2002 245–281.
Bibliography 239