The Syntax of Agreement and Concord 2008

This page intentionally left blank
T H E S Y N TA X O F AG R E E M E N T A N D C O N C O R D
In this series
75 l i l i a n e h a e g e m a n: The syntax of negation
76 p a u l g o r r e l: Syntax and parsing
77 g u g l i e l m o c i n q u e: Italian syntax and universal grammar
78 h e n r y s m i t h: Restrictiveness in case theory
79 d . r o b e r t l a d d: Intonational morphology
80 a n d r e a m o r o: The raising of predicates: predicative noun phrases and the theory
of clause structure
81 r o g e r l a s s: Historical linguistics and language change
82 j o h n m . a n d e r s o n: A notional theory of syntactic categories
83 b e r n d h e i n e: Possession: cognitive sources, forces and grammaticalization
84 n o m t e r t e s c h i k - s h i r: The dynamics of focus structure
85 j o h n c o l e m a n: Phonological representations: their names, forms and powers
86 c h r i s t i n a y. b e t h i n: Slavic prosody: language change and phonological theory
87 b a r b a r a d a n c y g i e r: Conditionals and prediction
88 c l a i r e l e f e b v r e: Creole genesis and the acquisition of grammar: the case of
Haitian creole
89 h e i n z g i e g e r i c h: Lexical strata in English
90 k e r e n r i c e: Morpheme order and semantic scope
91 a p r i l m c m a h o n: Lexical phonology and the history of English
92 m at t h e w y. c h e n: Tone Sandhi: patterns across Chinese dialects
93 g r e g o r y t. s t u m p: Inflectional morphology: a theory of paradigm structure
94 j o a n b y b e e: Phonology and language use
95 l a u r i e b a u e r: Morphological productivity
96 t h o m a s e r n s t: The syntax of adjuncts
97 e l i z a b e t h c l o s s t r a u g o t t and r i c h a r d b. d a s h e r: Regularity in
semantic change
98 m aya h i c k m a n n: Children’s discourse: Person, space and time across languages
99 d i a n e b l a k e m o r e: Relevance and linguistic meaning: the semantics and
pragmatics of discourse markers
100 i a n r o b e r t s and a n n a r o u s s o u: Syntactic change: a minimalist approach to
grammaticalization
101 d o n k a m i n k o va: Alliteration and sound change in early English
102 m a r k c . b a k e r: Lexical categories: verbs, nouns and adjectives
103 c a r l o ta s . s m i t h: Modes of discourse: the local structure of texts
104 r o c h e l l e l i e b e r: Morphology and lexical semantics
105 h o l g e r d i e s s e l: The acquisition of complex sentences
106 s h a r o n i n k e l a s and c h e r y l z o l l: Reduplication: doubling in morphology
107 s u s a n e d wa r d s: Fluent aphasia
108 b a r b a r a d a n c y g i e r and e v e s w e e t s e r: Mental spaces in grammar:
conditional constructions
109 h e w b a e r m a n, d u n s ta n b r ow n and g r e v i l l e g . c o r b e t t: The
syntax-morphology interface: a study of syncretism
110 m a r c u s t o m a l i n: Linguistics and the formal sciences: the origins of generative
grammar
111 s a u m u e l d . e p s t e i n and t. d a n i e l s e e ly: Derivations in minimalism
112 pa u l d e l a c y: Markedness: reduction and preservation in phonology
113 y e h u d a n. f a l k: Subjects and their properties
114 p. h . m at t h e w s: Syntactic relations: a critical survey
115 m a r k c . b a k e r: The syntax of agreement and concord
Earlier issues not listed are also available

CAMBRIDGE STUDIES IN LINGUISTICS
General editors: p. a u s t i n, j . b r e s n a n, b. c o m r i e ,
s . c r a i n, w. d r e s s l e r , c . j . e w e n, r . l a s s ,
d . l i g h t f o o t, k . r i c e , i . r o b e r t s , s . r o m a i n e ,
n. v. s m i t h
The Syntax of Agreement and Concord

T H E S Y N TA X O F
AG R E E M E N T A N D
CONCORD
M A R K C . BA K E R
Rutgers University
CAMBRIDGE UNIVERSITY PRESS
Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore, São Paulo
Cambridge University Press

The Edinburgh Building, Cambridge CB2 8RU, UK
Published in the United States of America by Cambridge University Press, New York
www.cambridge.org
Information on this title: www.cambridge.org/9780521855471
© Mark Baker 2008
This publication is in copyright. Subject to statutory exception and to the provision of

relevant collective licensing agreements, no reproduction of any part may take place
without the written permission of Cambridge University Press.
First published in print format 2008
ISBN-13 978-0-511-38884-2 eBook (NetLibrary)
ISBN-13 978-0-521-85547-1 hardback
ISBN-13 978-0-521-67156-9 paperback
Cambridge University Press has no responsibility for the persistence or accuracy of urls
for external or third-party internet websites referred to in this publication, and does not
guarantee that any content on such websites is, or will remain, accurate or appropriate.
This book is dedicated to the three wonderful children – Catherine,
Nicholas, and Julia – that God has given me to enliven my journey
through this life. I only ask that they not fight about who gets to be
the noun, who the verb, and who the adjective.
Contents
Acknowledgments page xi
List of abbreviations and conventions xiv
1 Introduction: category distinctions as a window on the

theory of agreement 1
1.1 A generalization to be explained 1
1.2 The incompleteness of previous discussions 3
1.3 What a better theory could look like 6
1.4 What is in this book 6
1.5 What is not in this book 7
2 Basic agreement and category distinctions 12

2.1 The generality of the categorical asymmetries in agreement 13
2.2 The category-theoretic infrastructure 27
2.3 The agreement-theoretic contribution 40
2.4 Explaining the basic categorical asymmetries in agreement 48
2.5 Issues arising 56
2.6 Conclusion 64
3 The unity of verbal and adjectival agreement 65

3.1 Downward agreement on adjectives 67
3.2 Upward agreement on verbs 74
3.3 Verbs that cannot agree in person, like adjectives 85
3.4 Conclusion 107
4 Explaining the restriction on person agreement 111

4.1 Person agreement and other categories 112
4.2 Operator-variable agreement and Agree 121
4.3 A locality condition on first and second person variables 124
4.4 On the strictness of locality conditions involving heads 138
4.5 Deriving the SCOPA 142
4.6 Conclusion 148
ix
x Contents
5 Parameters of agreement 153

5.1 Introduction: parameters and other kinds of variation 153
5.2 Agreement on tense 157
5.3 Agreement on FA and the formulation of the parameters 171
5.4 Agreement on complementizers 178
5.5 Agreement on determiners 184
5.6 Agreement on adpositions 191
5.7 Agreement on v 196
5.8 Agreement on the linker head 206
5.9 Agreement in auxiliary constructions 207
5.10 A third value for the Direction of Agreement Parameter 214
5.11 Many little parameters or two big parameters? 219
5.12 General conclusion 244
Appendix: Table of languages and their

agreement properties 246
References 254
Index 264
Acknowledgments
As I get older, I seem to be losing my sense of history. As in many families,

I have fewer pictures documenting every developmental stage of my youngest
child than I do of my oldest child. In the same way, I feel like I have less to say
by way of preface and acknowledgments with each book that I write. But there
are still plenty of people to thank, so here goes.
I do not think that I would have started another book so soon after finishing
Lexical Categories (Baker 2003a) if Andrew Winnard of Cambridge Univer-
sity Press had not approached me about the possibility of publishing some
manuscripts he had seen on my website. Talking to him made me realize that
there was in fact a common theme underlying many of those studies of par-
ticular languages, which I had not fully realized, and that common theme had
to do with agreement. I further realized that my theory of lexical categories
raised some huge unanswered questions about agreement – questions that had
not been answered by other people’s theories either. So I decided to delve into
this topic with some gusto. As I sought to write a capstone essay that would
draw together my little discoveries about agreement in particular languages
(Mapudungun, Kinande, Lokaa, Icelandic), the outlines of a broader theory
began to emerge, and the capstone essay took over the book as a whole. I fear
that some of the original papers are still unpublished (or published in less-visible
venues), but I hope that the final book is more consistent and unified, and paints
a bigger picture than it otherwise would have. It was also personally rewarding
for me to dive back fully into pure linguistic research after some time spent as
department chair and being involved in some cognitive science projects. I thank
Andrew and Cambridge University Press for the excuse to do this.
Of course, one needs not only a project worth working on, but also some
time to work on it. For meeting that need, I thank Rutgers University for pro-
viding a sabbatical leave and a competitive leave fellowship, and the American
Philosophical Society for awarding me a sabbatical leave fellowship (funded
by the Mellon Foundation) for 2005–2006. I would not have had the freedom
xi
xii Acknowledgments
to do the survey of 108 languages reported in chapter 5 if I had not received

this special support.
One also needs some facts to work with, and it is nice when some of them are
new ones that people have not had access to before. In this most recent period of
my career, I have been privileged to work with native-speaker linguists on their
fascinating languages in ways that have been inspiring and helpful to me. This
group includes Elisa Loncon (Mapudungun), Alexander Iwara (Lokaa), Willie
Udo Willie (Ibibio), and – of special significance to this particular study –
Philip Mutaka (Kinande). Kinande gets the most press in what follows, but
meditating on the challenges and wonders of all of these languages has provided
the impetus to do this work. I also wish to thank many linguists who have
generously answered email questions arriving from me “out of the blue” about
the languages that they know when I felt some small but crucial piece of a
puzzle was missing. These are acknowledged individually at the relevant points
in the text, but here I single out Halldór Sigurð sson for service beyond the call
of duty, since Icelandic turned out to be particularly important at several points.
And one needs some inspiring colleagues, who can stimulate you, challenge
you, and help you put your ideas to the test. I have been lucky enough to
present this material in a number of colloquiums and more extended forums,
including graduate seminars at Rutgers University in spring 2005 and fall 2006,
an LSA summer institute class at MIT in 2005, extended colloquia at UCLA and
Georgetown University in 2006, and week-long classes at the University of the
Basque Lands and the LOT summer school at the University of Amsterdam in
2006. I thank all of these audiences for their encouragement and input, including
José Camacho, Liliana Sanchez, Roger Schwarzchild, David Pesetsky, Esther
Torrego, Seth Cable, Hilda Koopman, Anoop Mahajan, Philippe Schlenker,
Carson Schütze, Raffaella Zanuttini, Bob Franks, Michael Diercks, Itziar Laka,
Javier Ormazabal, Myriam Uribe-Extebarria, Hans Broekenhuis, Jenneke van
der Wal, and others that I forgot – or whose names I never even learned. I thank
Jessica Rett and Cedric Boeckx for sending me written comments on parts of
the manuscript. I thank my students who have also been interested in issues of
case and agreement, including Vita Markman, Natalia Kariaeva, Jessica Rett,
and Carlos Fasola. Three people stand out for special thanks in this category.
First, I thank Chris Collins for early collaborative work on what we originally
called “the Bantu Parameter.” This provided the seeds for what became chapter
5, once I was finally able to investigate how that parameter might apply to other
languages. Second, I thank Ken Safir for his enormous influence on chapter 4,
helping me to get up to speed (or at least closer to speed) on the issues of binding,
person, and pronoun interpretation that are crucial there, and showing me how to
Acknowledgments xiii
say what I wanted to say in a much cleaner and more straightforward way. Third,
I thank Carlos Fasola, who in the guise of being my research assistant helped me
to discover the properties of many of the 108 languages discussed in chapter 5,
and helped to nurture in us both a common pleasure in grammar-reading.
One might not literally need a loving and supportive family in order to write
a book like this, but I certainly would not want to do it any other way. Many
thanks to my wife Linda and my three children for much help, support, prayers,
and companionship along the way.
Finally, I am convinced that I needed the grace of the Lord Jesus Christ, the
love of God, and the fellowship of the Holy Spirit to do something like this. I
thank my God for providing all of the people and resources listed above, and
everything else besides.
Abbreviations and conventions
In this book, I cite examples from a large number of languages, some of which
I do not know well. This presents certain challenges for effective glossing. The
safer course would be to simply follow the glossing practice of the source that
the example is taken from. The problem with this is that it multiplies greatly the
number of abbreviations used, and can obscure comparison by giving similar
morphemes very different glosses. This problem could be addressed by trying
to impose a uniform system of glossing on all of the languages considered. But
that creates other problems: in particular, languages might have morphemes
that are similar in their usage but not identical in all respects, and I might not
know enough to do it accurately. I have tried to strike a middle path between
these two courses, making the glosses more uniform when I thought I could do
it with reasonable accuracy and when the morphemes are relatively important
to my topic – in particular, when they are agreement morphemes. I am not fully
satisfied with the results, and experts on the relevant languages may be even
less so. But that is what I did.
Agreement morphemes (particularly those on verbs) are glossed by a com-
plex symbol that begins with a number indicating the person of the agreed-with
phrase (1, 2, or 3), then has a lower-case letter indicating the number of the
agreed-with phrase (s, singular; d, dual; p, plural), and then a capital letter indi-
cating the grammatical function of the agreed-with phrase (S, subject; O, object;
P, possessor; A, absolutive; D, dative/goal; E, ergative). Thus, 1pS means first
person plural subject agreement, 3sO means third singular object agreement,
and so on. Sometimes one member of this triple is missing when the correspond-
ing category is not marked – for example, when the agreement indicates person
but not number, or vice versa. When two agreement factors are expressed with
a single portmanteau morpheme, their features are separated with a slash.
The reader should also note that 1, 2, and 3 have two meanings in agreement
morphemes: they can mean first, second, or third person (all languages), or
they can mean a third person noun phrase in class 1 (human singular), class 2
(human plural), or class 3 (singular) in a Bantu language or Lokaa. Thus 1sS
xiv
Abbreviations and conventions xv
always means first person singular subject, but 1S in the gloss of a Niger Congo
language means subject agreement with a noun of class 1 (third person human
singular). (1S in the gloss of a non-Bantu language could mean first person
subject agreement, with number unspecified.) I hope this will not be unduly
confusing.
Other abbreviations used in the glosses of linguistic examples are as follows.
Readers should consult the original sources for more on what these categories
amount to in particular languages.
1–19 noun class (Bantu)

a,b,c gender/number categories (Southern Tiwa)
abs absolutive case
acc accusative case
act actor
adj adjectival
adv adverbial
aff affirmative
agr agreement
an animate
aor aorist tense
appl applicative morpheme
art article
asp aspect
assoc associative marker
aug augmented
aux auxiliary
ben benefactive applicative marker
caus causative
cl noun class (Bantu), classifier (Tariana)
comp complementizer
compl completive
cond conditional
conj conjunction
cont continuous aspect
contr contrastive
cs construct state (Berber)
dat dative case
decl declarative
dem demonstrative
xvi Abbreviations and conventions
desid desiderative
det determiner
dir direct
disj disjunctive prefix
dr directional
dur durative
dyn dynamic
erg ergative case
eu euphonic
ext extended aspect (Bantu)
f feminine gender
fam familiar
foc focus
fut future
fv final vowel (Bantu, indicative mood marker?)
gen genitive case
ger gerund
hab habitual aspect
i irrational
imp imperative
impf imperfective
inan inanimate
ind indicative
indf indefinite
inf infinitive
instr instrumental
int intentional
intrans intransitive
inv inverse
irr irrealis
lk linker
loc locative
log logophoric
m masculine gender
n neuter gender
neg negative
ni noun incorporation
nom nominative case
noml nominalizer
np nonpast tense
Abbreviations and conventions xvii
npst nonpast tense

nsf noun suffix (determiner?)
obj objective case marker
obl oblique
opt optative
pass passive
past past tense
perf perfective aspect
pl plural number
poss possessor
pp past perfective
pred predicative head
pres present tense
prev preverb
prog progressive aspect
ptpl participle
q question particle
r rational
rcp recent past
refl reflexive
rel relative
rem remote past
rep reported
rpst recent past
sbjn subjunctive
sg singular number
ss same subject
stat stative aspect
sub subordinate
subj subject
t Tense, unspecified tense marker
ta Transitive animate (Algonquian)
th thematic
thsy hearsay particle
top topic
tr transitive
unposs unpossessed
val validator
vbzr verbalizer
veg vegetable gender (Mayali)
xviii Abbreviations and conventions
vis visible
x special gender class in Burushaski
The following are abbreviations of linguistic terms: names of principles, gram-

matical categories, language families, and construction types.
a, ap adjective, adjective phrase

c, cp complementizer, complementizer phrase
d, dp determiner, determiner phrase
deg degree head
ecm Exceptional Case Marking
ecp Empty Category Principle
epp “Extended Projection Principle” feature (triggers the
movement of a phrase to the category that bears it)
fp functional phrase
ie Indo-European languages
n, np noun, noun phrase
nc Niger-Congo languages
p, pp adposition (preposition or postposition), adpositional
phrase
pf Phonological Form
plc Person Licensing Condition
plc(h) Person Licensing Condition applied to Heads
scopa Structural Condition on Person Agreement
spec, xp Specifier of XP
t, tp tense head, tense phrase
v, vp light verb (abstract verbal element, assigner of external
argument), light verb phrase
v, vp verb, verb phrase
vso, sov, etc. Verb-subject-object word order; subject-object-verb order,
etc.
wals World Atlas of Language Structures
Finally, the following are some conventions used in presenting examples:
(x) The example has the same grammatical status with or

without X included.
(*x) The example is good without X, but bad when it is included.
*(x) The example is bad unless X is included.
In some cases, an agreement morpheme and the NP that it agrees with are both
underlined.
1 Introduction: category
distinctions as a window on the
theory of agreement
1.1 A generalization to be explained

Even though agreement phenomena are some of the most familiar and well-
studied aspects of grammar, there are certain basic questions that have rarely
been asked, let alone answered. One such question concerns the fact that, in
many languages, the three major lexical categories – noun, verb, and adjective –
behave quite differently with respect to agreement. Words of all three categories
can bear similar, even cognate, inflectional affixes. But verbs are consistently the
most prolific agreers, often agreeing with their subjects in person, number, and
gender features, and sometimes agreeing with their objects in these features as
well. Adjectives clearly participate in agreement, but they do so more modestly:
they rarely or never agree with more than a single nominal, and they can agree
in number and gender but they typically do not show first or second person
agreement forms. In light of these differences, traditional grammar often says
that adjectives participate in concord, a distinct phenomenon from the agree-
ment that involves verbs. Nouns are more conservative still. Although they are
often inflected, sometimes with the same affixes that adjectives take, nouns do
not need to agree with another NP in their environment the way that an adjec-
tive in a similar structural configuration does. My thesis in this book is that the
true theory of agreement should be able to explain these robust cross-categorial
differences, and that much can be learned about agreement by seeking a theory
that meets this condition.
To illustrate the fundamental contrasts in agreement behavior that I have in
mind, consider (1) from Swahili.
(1) a. Ni-li-kuwa ni-ki-som-a. (Ashton 1949)
1sS-past-be 1sS-cont-read-fv
‘I was reading.’
b. Ni-Ø m-refu.
1sS-be cl1-tall
‘I am tall.’
1
2 Introduction
c. Ni-li-po-kuwa ki-jana . . . sasa ni-li-po m-tu m-zima, . . .

1sS-past-when-be cl7-child now 1sS-be-when cl1-man cl1-whole
‘When I was a child . . . Now that I am a man . . .’
The copular verbs in all three examples manifest agreement with an understood
subject that is first person, singular, and animate; let us put that aside.1 The
interesting differences are found on the postcopular lexical category. The main
verb in (1a) also shows agreement that is first person, singular, and animate; its
prefix is thus identical to the one on the copula. (2) shows a transitive verb in
Swahili, in which the verb agrees with its understood object in person, number,
and gender as well as with its subject.
(2) Juma a-li-ni-ambia kwamba . . . (Vitale 1981:62)
Juma 1S-past-1sO-tell that
‘Juma told me that . . .’
In contrast to the main verb in (1a), the predicate adjective in (1b) cannot
bear the first person singular prefix ni. Rather it must bear the class 1 prefix
m. This morpheme expresses singular number and human/animate gender, in
partial agreement with the subject, but it does not express first person. The
same prefix is thus used on the adjective when the subject is third person, as
in (3a). (3b) shows an adjective with a different prefix, manifesting a different
combination of number and gender in agreement with its subject. This confirms
there is partial agreement on adjectives in Swahili, but not complete agreement
with first and second person subjects.
(3) a. Hamisi yu-Ø m-refu (Ashton 1949)
Hamisi cl1-be cl1-tall
‘Hamisi is tall.’
b. Mi-zigo hii mi-zito.
cl4-loads these cl4-heavy
‘These loads are heavy.’
Finally, the first predicate nominal ki-jana ‘child’ in (1c) is different from
both the verb and the adjective in that it does not even agree with the subject
in gender. It bears the class 7 agreement prefix ki, reflecting the diminutive
meaning of the predicate noun, not the class 1 agreement prefix m that the
1 In contemporary Swahili, the stem of the verbal copula is normally omitted in the simple present,
when it bears no morphology other than the subject agreement marker. (The stem li does show up
in some archaic expressions, such as proverbs, suggesting that this is an innovation.) Swahili also
has an uninflected copular particle ni. It is not clear whether the ni in (1b) is an agreement marker
or this copular particle, but this is orthogonal to the point I want to make using this example.
The incompleteness of previous discussions 3
first person singular subject triggers on the predicate adjective in (1b) (which
also appears on the second predicate nominal m-tu m-zima ‘full-grown man’ in
(1c)). A more striking example of a predicate nominal that does not agree with
its subject is shown in (4), where the predicate noun ‘clouds’ fails to agree with
the subject ‘sign of rain’ in gender or in number.
(4) Dalili y-a mvua ni ma-wingu. (Ashton 1949)

cl9.sign cl9-assoc cl9.rain pred cl8-clouds
‘Clouds are a sign of rain.’
So verbs in Swahili show full agreement, adjectives show partial agreement, and
nouns are inflected but do not agree. This is also true for many other, unrelated
languages, as we shall see in chapter 2.
1.2 The incompleteness of previous discussions

Although this pattern of facts is reasonably well known, at least for Indo-
European languages and Semitic languages, linguists have rarely attempted to
explain it. Consider, for example, Stassen’s (1997) extensive typological study.
Stassen clearly recognizes the empirical connection between being a verb and
bearing person agreement, stating it as follows (p. 38):
(5) The Agreement Universal

If a language has person agreement in intransitive main clauses, this
agreement will at least be used in sentences with event predicates [i.e. with
prototypical verbs – MCB].
In connection with this universal, he writes:
As far as I am aware, no known principle of linguistic theory prevents us from

imagining a possible natural language with some form of person agreement
where this agreement can be used only with predicate adjectives or only with
predicate nouns, but not with predicate verbs. However, in my sample [of 410
languages – MCB] I have not found a single instance of a language in which
such a situation can be observed. In this book, no attempt at an explanation
for the Agreement Universal will be made.
In contrast, this book will make an attempt at explaining the Agreement Uni-
versal and related facts, using the tools of formal generative linguistics.2
2 Stassen sees this as only a one-way implication, from verbiness to person agreement, not a
bidirectional implication as I argue. For a discussion of languages in which adjectives and/or
nouns bear person agreement as well as verbs, see section 2.5.1 below.
4 Introduction
Within the Chomskian tradition, the (often tacit) state of the art has been
simply to stipulate which feature slots are present but unvalued on a particular
lexical item, thereby specifying explicitly its agreement potential (Chomsky
2000, Chomsky 2001). Fleshing out this background assumption, the Swahili
pattern in (1) could be described by saying that verbs in Swahili have one or
two empty slots for person, number, and gender features, depending on their
transitivity. Values for these slots are then filled in by a process of agreement. In
contrast, adjectives in Swahili would have only a single set of feature slots, and
they happen to lack a slot for a person feature. Nouns in Swahili would have no
empty feature slots: either they have no slots at all, or (better) they have slots
for gender and number, but those features are already valued and hence are not
open for agreement. On this baseline view, the Swahili words used in (1) could
have representations like the following:
(6) soma refu jana

‘read’ ‘tall’ ‘child’
verb adjective noun
Person: Number: (Person: 3rd)
Number: Gender: (Number: sg)
Gender: (Gender: 7)
(Person: )
(Number: )
(Gender: )
A variant of this view would put the feature slots, valued or unvalued, not on
the lexical category itself, but rather on some functional category associated
with it. For example, the subject agreement features might be attributed not
to the verb itself, but to the Tense/Infl node that selects VP. Similarly, object
features might be attributed to a v projection distinct from verb, the valued
number feature might be on a Number head rather than on the noun itself, and
so on.
Although it may be descriptively accurate, this baseline generative view is
not very satisfying theoretically. If the lexical categories varied randomly in
their behavior with respect to agreement within and across languages, then
this sort of theory is the best one could hope for. But they do not. The pattern
shown in (1) and described in (6) is in fact the norm for languages that use
the relevant features as part of their grammatical system. Given this, some
deeper theoretical relationship should be found between saying what lexical
category a word belongs to and saying what its behavior with respect to agree-
ment is.
The incompleteness of previous discussions 5
It might seem tempting to take the agreement potential of a word as being

basic, and define its lexical category in terms of that. For instance, one could
define an adjective in Swahili as a word that can agree in number and gen-
der, but not person, whereas verbs are defined as words that can show person
agreement and nouns as words that do not undergo agreement at all. (Ouhalla
(2005b) makes a proposal along these lines, defining verbs as categories that
have person features and nouns as categories that have a class (gender) fea-
ture.) To the extent that there are other morphological and syntactic differences
between nouns, verbs, and adjectives in Swahili, one might attempt to derive
them from this inflectional difference. For example, one might derive the fact
that adjectives can function as attributive modifiers but nouns generally can-
not from these definitions plus the stipulation that attributive modifiers must
agree with the noun that they modify. This approach has some intuitive appeal,
because the inflectional differences among the lexical categories are often very
salient in richly inflected languages. Indeed, awareness of the noun/verb dis-
tinction in Western grammar comes historically from the observation of Ancient
Greek and Latin grammarians that some words inflect for number, gender, and
case whereas others inflect for tense and subject agreement (Robins 1989).
This is often how the noun/verb/adjective distinctions are presented in modern
descriptive grammars as well.
If, however, one’s linguistic theory aspires to give a general account of the
human language capacity, and if it seeks to have explanatory depth, it becomes
clear that this temptation should be resisted. One reason is that some languages
clearly distinguish nouns, verbs, and adjectives even though they have little or
no agreement. Modern English is very nearly a language of this type: agreement
has been lost entirely on adjectives, and only survives on third singular verbs
in the present tense. Japanese and Edo are even better examples (see Baker
2003a for discussion, e.g. pp. 240–5 on Japanese). A definition of the lexical
categories that is rooted in their differing agreement potentials seems artificial
and unenlightening for these languages. A second reason is that, even in lan-
guages that have a substantial amount of agreement, it is difficult to see how
some of the purely syntactic differences between nouns, verbs, and adjectives
can plausibly be derived from the inflectional differences. For example, only
adjective phrases can be the complements of dedicated degree heads like so,
as, too, and how in English (Chris is too hungry versus *Chris too likes wine,
*Chris is too (a) fool). It is hard to see any plausible direct connection between
a fact like this and the agreement asymmetry in (1), such that the former could
be explained in terms of the latter.
6 Introduction
1.3 What a better theory could look like

A more promising approach is to work the other way around. Suppose that we
had a syntactically oriented theory of the lexical categories that was designed
to explain why nouns, verbs, and adjectives occur in systematically different
structures. That theory could be combined with a structural theory of agreement
that did not explicitly mention the lexical categories at all. If everything worked
properly, the agreement differences among the lexical categories might follow
from the independently established syntactic differences together with general
principles of agreement that are blind to category labels. Such a theory could
explain why the noun/verb/adjective distinction does not depend on a language
having agreement, but shows up in systematic ways in languages that do have
agreement. The goal of this book is to develop this sort of approach, and to
learn what can be learned about the general theory of agreement by doing so.
In fact, the needed subtheories already exist, for the most part. In Baker
2003a, I developed a theory of the noun/verb/adjective distinctions designed
to account for their syntactic differences across languages. That work did not
consider agreement at all, however. Meanwhile, Chomsky (2000, 2001) and
other minimalist theorists have developed a sophisticated syntactic theory of
agreement, looking primarily at the agreement properties of verbs. This book
considers how these theories can be combined – with a few adjustments – to
achieve a unified theory of agreement that applies to all three lexical categories
and explains their differences in agreement behavior in terms of their more
basic syntactic differences.
1.4 What is in this book

I pursue this goal as follows. Chapter 2 is concerned with documenting more
fully the agreement asymmetries sketched in (1), and showing how the core
cases of agreement on simple predicates and modifiers can be explained by the
conjunction of my theory of lexical categories and the Chomskian theory of
Agree. Two key additions to the theory of agreement are found to be necessary.
The first is the idea that a head can search upward through the syntactic structure
for something to agree with as well as downward. The second is the idea that
agreement in first and second person features is subject to a more stringent
locality condition than other sorts of agreement are (the Structural Condition
on Person Agreement, or SCOPA).
Chapter 3 extends the account to other, less canonical constructions, including
agreement on unaccusative and raising adjectives, agreement with wh-phrases,
What is not in this book 7
agreement in oblique subject constructions, agreement in double object con-

structions, and long distance agreement. In addition to proving that the ideas
introduced in chapter 2 are general and robust, this chapter proves that adjec-
tives and verbs should fall under the same theory of agreement, rather than
having distinct theories of agreement (for verbs) and concord (for adjectives).
For example, I show that agreement on raising adjectives is sensitive to all
the same factors as agreement on raising verbs, and that verbs lose the ability
to agree in first and second person when they happen to appear in syntactic
configurations that are similar to those that adjectives always appear in.
Chapter 4 turns to a closer examination of the SCOPA, considering in a
deeper way why agreement in first and second person should be subject to a
tighter locality condition than other sorts of agreement. I claim that this is the
projection onto agreement of a general condition that applies also to first and
second person pronouns as opposed to third person pronouns: first and second
person pronouns must be bound by the closest relevant operator, whereas third
person pronouns need not be. Thus, the SCOPA is not an ad hoc stipulation,
but rather something rooted in the fundamentals of what it is for any linguistic
expression to be first or second person.
Chapter 5 then turns from the universal aspects of the theory of agreement to
those aspects that vary parametrically. I claim that there are two such aspects.
First, in some languages agreement depends on there being a unidirectional c-
command relationship, whereas in other languages it does not. Second, in some
languages agreement is dependent on the case features of the two expressions
being co-valued whereas in other languages it is not. These parameters are
motivated by a close comparison of agreement in Niger-Congo languages with
agreement in Indo-European languages. Their validity is then tested against a
sample of 108 languages taken from around the world. I conclude that agreement
is a domain of grammar in which there are both substantive universal constraints
and a degree of highly patterned and systematic crosslinguistic variation.
1.5 What is not in this book

Almost as important as saying what this book is about is being explicit at
the beginning about what this book is not about. This book is almost entirely
about the syntax of agreement – not about the morphology of agreement or
about the semantics of agreement. This is not to say that agreement is a purely
syntactic phenomenon; it certainly is not. Nor is it always easy to tease apart the
syntactic aspects of agreement from its morphological or semantic aspects. But
I believe that there is a syntactic aspect to this topic, which can and should be
8 Introduction
distinguished from the purely morphological and the primarily semantic. Here
is a brief sketch of why.
Consider first the distinction between morphology and syntax. It is a fact
about Spanish and other Western European languages that verbal agreement
shows number and person but not gender. There is, then, no difference in the
verbal agreements in (7a) and (7b), although there is in the adjectival agreements
in (7c) and (7d).
(7) a. Nosotros com-emos las manzanas.

we.m.pl eat-1pS the apples
b. Nosotras com-emos las manzanas.
we.f.pl eat-1pS the apples
c. Nosotros estamos list-o-s.
we.m.pl are.1pS ready-m-pl
d. Nosotras estamos list-a-s.
we.f.pl are.1pS ready-f-pl
If one looked only at a single language or language family, one might consider
this fact about verb agreement to be on a par with the fact that adjectives inflect
for number and gender but not person. But crosslinguistic comparison reveals
a clear difference. The inability of verbs to manifest agreement in gender is a
property of certain IE languages, but it is clearly not universal. Many languages
do have verbs that agree with their subjects in gender as well as in number and
person; examples include Swahili, Arabic, Yimas, and Mohawk. In contrast,
the failure of adjectives to agree in person is much more general, arguably
universal. Furthermore, as we shall see, the ability of verbs to agree in person
is contingent on the kind of syntactic configuration that holds between the
verb and the agreed-with argument (see chapter 3). In contrast, whether gender
agreement is present on verbs does not (as far as I know) interact with syntactic
configurations in any interesting way. Taken together, these facts show that
whether or not person agreement is possible is often a syntactic fact, whereas
the presence versus absence of gender agreement on verbs is not syntactic in the
same way. I therefore take the latter fact to be purely morphological in nature:
the gender features are present on the verb in Spanish, but the morphemes that
happen to spell out a [1st person plural feminine] feature bundle are no different
from the morphemes that spell out a [1st person plural masculine] feature bundle
in this language.
This is not meant to deny that the absence of gender in verb agreement in
some languages is an interesting and grammatically significant feature of the
language. For example, there could be a general rule of impoverishment of

the sort studied by Bonet (1991) and other Distributed Morphologists, which
systematically removes the gender feature from the feature bundles created by
agreement prior to the insertion of actual morphemes. Such rules of impoverish-
ment may obey laws of their own and have a major impact on what agreement
looks like in particular languages. But whatever principles govern these phe-
nomena, they are very likely to be different from those that underlie the contrasts
in (1).3 More generally, I assume that the apparatus of Distributed Morphology
is present to interpret in nontrivial ways the feature bundles that are placed
on syntactic heads by the principles of agreement that I discuss, but I do not
consider that aspect of agreement here in any depth.
Bobaljik (to appear) has constructed an argument that all of agreement takes
place after the syntax, in the morphological part of the PF component (see also
Marantz 1991, Halle and Marantz 1993). Whereas I follow a fairly standard
Minimalist division of labor between morphology and syntax, Bobaljik claims
that even the matter of which NP a given word agrees with is determined in the
postsyntactic morphological component. I do not follow Bobaljik in this for a
variety of reasons: (a) I am not convinced that agreement relationships never
feed other syntactic and semantic processes, (b) I think they are sensitive to
more details of syntactic structure than other PF processes are, and (c) I am too
conservative.4 But the difference between the two conceptions is not as large or
important as it might seem. Bobaljik acknowledges that agreement is sensitive
to syntactic relations, including a notion of which of two NPs is the higher one
(c-command) and some notion of local domains (clausal constituents, phrases).
He simply claims that this information is still available at PF, inherited from the
syntax. As a result, the difference in framework does not matter much in practice:
people who are attracted to Bobaljik’s Distributed Morphology architecture
can interpret my title The Syntax of Agreement and Concord as meaning “those
3 There is probably nothing special about gender such that it alone can undergo impoverishment
within the verbal inflectional system. There are a number of New World languages (e.g., Diegeño,
Zoque, Hixkaryana) in which verbs inflect for person but not for number, suggesting that the
number feature is deleted from the PF representation in those languages. Tsez is a language in
which verbs agree with subjects or objects in number and gender but not in person (Polinsky
and Potsdam 2001); a possible analysis could be that the person feature is deleted by a rule of
impoverishment in this language. I assume then that any feature can in principle be involved in
postsyntactic adjustments of this kind.
4 It is also not clear to me how to generalize Bobaljik’s approach from the one type of agreement
he considers in detail (single agreement on the finite verb), to agreement on the full range of
functional heads I am concerned with here. It would undoubtedly be interesting to try to do this
and then compare the results with my theory, but I leave at least the first step of this process to
others.
10 Introduction
aspects of the syntactic representation that agreement and concord make use of,”
rather than “what happens in the syntactic component with regard to agreement
and concord.”
Consider now the distinction between syntax and semantics. There are
instances of “agreement” that are rather clearly semantic in nature. The most
famous is the possibility of collective nouns to trigger plural agreement on verbs
(but not demonstratives) in many British varieties of English. An example is:
(8) a. (*) This band are brilliant live. (Wechsler and Zlatić 2003:76–9)
b. This band is brilliant live.
c. *These band are brilliant live.
This sort of agreement is semantic in that it depends on semantic properties

of the subject – in particular, the ability of a singular noun to denote a group
consisting of several members. British speakers are also reported to detect a
semantic distinction between using the plural form are and the singular form is
in a sentence like (8): is goes with a collective reading of the predicate and are
with a distributive reading. Somehow, then, a semantically licensed form can
replace a morphosyntactically licensed form under certain conditions in British
English. Nothing in this book will tell you why.5
More directly relevant to the range of data I have been focusing on, there is
a kind of semantic agreement that can be seen in (9c–d).
(9) a. Those women are a committee.
b. The committee on diversity is three women and two men.
c. Those women are theoretical linguists.
d. *Those women are a theoretical linguist.
(9a–b) are like the Swahili examples (1c) and (4), which I have been using to
show that there is no agreement between a predicate nominal and its subject; one
is singular and the other plural. Nevertheless, a kind of agreement is enforced in
(9c–d), such that (9d) is unacceptable. How do we resolve the paradox? What
seems to be going on here is that nouns like committee and woman individuate
things in very different ways. What counts as a single committee also counts as
several distinct women. In contrast, nouns like woman and linguist individuate
things in a similar way. If all the women are linguists and all the linguists are
women, then there could never be a different number of linguists and women.
5 The issue of semantic agreement can arise for gender features as well as for number. Thus in
Swahili (but not in Chichewa) nouns that denote humans but belong to some gender class other
than class 1 nevertheless trigger class 1 agreement on verbs and other heads (Ashton 1949).
This case also has the earmarks of semantic agreement. (I thank an audience at Georgetown
University, especially Michael Diercks, for discussion.)
Given this, I take the appearance of agreement in (9c–d) to be an epiphenomenon

of there being two NPs that pick out the same set of people independently, but
which happen to individuate the members of that set in the same way. The
two noun phrases thus both turn out to be plural, each by its own independent
calculation. In contrast, (9a–b) show that when the two NPs individuate the
set in question differently, they can have different number values. I take the
less-common sentences in (4) and (9a–b) to be telling the truth that there is no
grammatical agreement between predicate nouns and their subjects the way that
there is between predicate adjectives and their subjects. (In contrast to (9a–b),
one never finds a predicate adjective that does not agree with the subject in
languages like Spanish, where adjectives are inflected.) I concede that (9c–d)
shows a kind of agreement, but claim that it is semantic agreement rather than
syntactic agreement, and the notions needed to explicate it are different.
It is certainly an incompleteness and perhaps a deficiency of this work that
it does not provide a theory of the morphological and semantic aspects of
agreement, complete with a battery of diagnostic tests for discerning which
is which, and an overarching theory of how all three types of agreement are
integrated together. But if this work attempted those achievements, it would
have another serious deficiency: it would be way over its page limit. There is
plenty to say for one book about the narrower topic of those aspects of agreement
that are arguably syntactic, as will soon be evident.
2 Basic agreement and category
distinctions
In chapter 1, I presented the Swahili data in (1) to illustrate that verbs can
show agreement in person, number, and gender, adjectives show agreement
(concord) in number and gender but not person, and nouns do not undergo
syntactic agreement at all.
(1) a. Ni-li-kuwa ni-ki-som-a. (Ashton 1949)

1sS-past-be 1sS-cont-read-fv
‘I was reading.’
b. Ni-Ø m-refu.
1sS-be cl1-tall
‘I am tall.’
c. Ni-li-po-kuwa ki-jana . . .
1sS-past-when-be cl7-child
‘When I was a child . . .’
I also claimed that these differences in the agreement properties of the lexical
categories are found in a wide variety of languages and deserve a general
explanation in terms of more basic principles of syntactic theory. And I held
out the hope that the effort to do so would help to refine and elaborate the theory
of agreement, which has been a topic of great interest and attention in recent
Chomskian work, especially since Chomsky 2000 posited Agree as one of the
most basic syntactic processes. It is the burden of this chapter to develop and
support these claims.
I begin by showing as briefly as possible that the pattern in (1) is also found
in other agreement-rich languages from around the world, providing empirical
justification for the goal of deriving it from general principles (section 2.1). I
then introduce and briefly motivate the necessary parts of my (2003a) theory of
lexical categories, culminating in a precise understanding of the different syn-
tactic structures that the lexical categories appear in (section 2.2). The next step
is to present the Minimalist theory of agreement, together with a generaliza-
tion of that theory that is needed to apply it to structures containing adjectives
12
The generality of the categorical asymmetries in agreement 13
(section 2.3). The crucial adjustment is that agreement-bearing heads must be

able to search upward through the syntactic structure to find something to agree
with as well as downward. I then apply the theory of agreement to the differing
syntactic structures that contain instances of lexical categories, to show how the
agreement asymmetries are accounted for (section 2.4). For the noun–adjective
contrast, nothing new is required, but the adjective–verb contrast requires a
novel principle that expresses what is special about first and second person
agreement as opposed to agreement in other features. Finally, I discuss some
apparent counterexamples to the claim that nouns and adjectives cannot show
agreement in first or second person, and some cases where adjectives agree
in some syntactic contexts but not others (section 2.5). Throughout this chap-
ter, I concentrate on the most familiar syntactic categories (nouns, verbs, and
adjectives) used in the most canonical ways (as predicates and as modifiers)
and present the simplest versions of the principles needed to account for them.
Subsequent chapters extend this core account empirically to other syntactic
structures and other parts of speech, as well as deepening the principles theo-
retically and showing how they vary parametrically. For now, though, I focus
on telling the basic story for the basic facts as directly as possible.
2.1 The generality of the categorical asymmetries in agreement

Suppose that agreement in Swahili were the way I have described it, but other
languages had very different patterns of agreement. Suppose, for example, that
it were just as common to find languages in which predicate adjectives agree in
person with their subjects, predicate nouns must agree syntactically in number
and gender, and verbs do not agree at all. Still other languages might have
other imaginable categorical asymmetries in agreement, or they might treat all
categories in the same way. If that were the case, then the right theory would
be to stipulate on a language-by-language basis which words or word classes
can agree with which features. In the limit, it could be that there is no general
theory of agreement, and this is a topic for the lexicographer of a particular
language rather than for the comparative syntactician.
But that is not what a look at other languages reveals. On the contrary, the
Swahili pattern proves to be robust and widespread, found in agreement-rich
languages from different language families and different parts of the world.
To show this, I put alongside the Swahili facts facts drawn from Mayali
(Australian), Syrian Arabic (Afro-Asiatic), Tariana (Arawakan, from South
America), and Spanish (Indo-European), showing that they are all essentially
the same at this level of description. Furthermore, I believe the patterns to be
14 Basic agreement and category distinctions
presented are consistent across the language families that these languages rep-
resent to the extent that the related languages show agreement at all. Since some
of these language families are quite large, the pattern in question is indeed a
widespread one.1
Note that I am not (yet) arguing that the pattern is universal – only that it
is much more common than one would expect to arise by mere chance, and
hence worthy of being explained. Once we have gained some sophistication in
treating these matters theoretically, I return to the question of whether these
categorical asymmetries in agreement are truly universal (section 2.5).
How to present this data is something of a practical problem. Given that
I need to cover quite a bit of ground in order to make my point, it would
become tedious to most readers for me to present a sketch of each language,
describing the essentials of its gender and number system, the morphological
structure of its words, and other basic information needed to fully understand
the examples presented. Furthermore, even if I did this, it would not eliminate
readers’ need to either trust my presentation or check the facts for themselves.
Therefore, I opt for a minimal presentation, noting that published and gen-
erally available grammars exist for all of these languages. It thus should be
perfectly feasible for most skeptical or curious readers to check whether I got it
right.
2.1.1 Adjectives versus nouns: predicative structures

The first asymmetry to consider is the fact that predicate adjectives must agree
with their subjects in number and gender, but predicate nouns need not. Predicate
nouns often seem to agree with their subjects in number and gender, but that is a
semantic effect that arises when their number or gender features are semantically
interpretable and both nouns happen to carve up the world in the same way,
as discussed in section 1.5. When two nouns have different inherent genders
or individuate entities differently, then mismatches in gender and number are
tolerated in predicate nominal constructions. This is shown in the English and
Swahili examples in (2), repeated from chapter 1.
(2) a. Dalili y-a mvua ni ma-wingu.
cl9(sg).sign cl9-assoc cl9.rain pred cl6(pl)-clouds
b. Those women are a committee
1 The generalizations also hold for Yimas, a New Guinean language with a three-way number
distinction and more than ten genders; see Foley 1991 for data and description. I omit citing
examples for reasons of space.
Such mismatches of number and gender are not tolerated in Swahili when the
predicate is an adjective, as shown in (3) (also from chapter 1).
(3) Mi-zigo hii mi-zito (*ki-zito).

cl4(pl)-loads these cl4(pl)-heavy (*cl7(sg)-heavy)
‘These loads are heavy.’
Example (4) illustrates the same contrast in Spanish, a language that distin-
guishes two grammatical genders and has a two-way number system.
(4) a. Est-as mujeres son gord-as. (*gord-o)

These-f.pl women(f.pl) are fat-f.pl (*m.sg)
(Predicate A, agreement)
b. Est-as mujeres son un grup-o.
These-f.pl women(f.pl) are a m.sg group-m.sg
(Predicate N, no agreement)
Examples (5) and (6) illustrate the contrast in Syrian Arabic, which also has
two grammatical genders and distinguishes singular from plural (Cowell 1964:
420–1).
Predicate adjectives:
(5) a. L-walad žūc ān (*žc ān-ı̄n). (number agreement)
the-child.m.sg hungry.m.sg hungry-pl
‘The child is hungry.’
b. L-bə nt žūc ān-e (*žūc ān). (gender agreement)
the-girl.f.sg hungry-f.sg hungry(m.sg)
‘The girl is hungry.’
Predicate nouns:
(6) a. Mac būd-ak ə l-masāri. (no number agreement)
idol.m.sg -2sP the-money.pl
‘Your idol is money.’
b. Hayy modēl ə ždı̄d. (no gender agreement)
this.f.sg model.m.sg new.m.sg
‘This one (referring to a car) is a new model.’
The Australian language Mayali has four genders (masculine, feminine,

vegetable, and neuter), although little or no number is represented on adjec-
tives and nouns. (7) shows agreement on predicate adjectives for gender;
(8) shows an absence of agreement on predicate nouns for number or
gender.
(7) a. Na-meke bininj na-kimuk. (Evans 2003:557)

m-that man.m m-big
‘That man is big.’
b. Ngal-eke daluk ngal-kimuk.
f-that woman.f f-big
‘That woman is big.’
(8) a. Ngaleh ngal-yuhyungki ngurrurdu ngaleng bininj, Bulanjdjan.

f.dem f-ancestor emu.f she human.m subsection
‘The female ancestor emu, she was a human of Bulanjdjan subsection.’
b. Ngad kun-kanj bedberre bininj. (Evans 2003:555–6)
we.pl n-meat.sg 3p.obl Aborigine
‘We are meat for Aborigines.’
The South American language Tariana has an unusual gender system.

Aikhenvald (2003) presents it as an open class system, with predicate adjectives
taking any of more than one hundred distinct “classifier” suffixes in order to
show agreement with the subject noun phrase along some semantic dimension.
(9a) versus (9b) shows agreement with the subject in number in Tariana; (9a)
versus (9c) shows how agreement with a female-human-denoting subject is
different from agreement with a plate-denoting subject.
(9) a. Wyume-ma-se matSa-ma-pidana. (Aikhenvald 2003:499)

last-cl:f-contr good-cl:f-rem. p.rep
‘The last one (a female) was beautiful.’
b. Keru-ma-pe na-pidana. (Aikhenvald 2003:252)
angry-cl:f-pl 3pS.become-rem. p.rep
‘They (women) became angry.’
c. Nuha karapi hala-karapi-niki. (Aikhenvald 2003:119)
I plate open-cl:plate-compl
‘My plate is completely full of holes.’
In contrast, (10) shows that predicate nouns in Tariana do not need to have a
classifier suffix that agrees with their subject in gender or number. In (10a),
the predicate nominal ‘rat’ does not bear the feminine and plural markers ma
and pe in agreement with its subject, whereas the adjectives ‘good’ and ‘angry’
in (9a) and (9b) do. In (10b), the predicate nominal does not bear the animate
plural classifier peni that an adjective like ‘good’ would in a similar context
(see (19c)).
(10) a. Pedalie-ma-pe inari na-matSi-ka

old-cl:f-pl mucura.rat 3pS-transform-th
na-pidana. (Aikhenvald 2003:252)
3pS.go-rem. r.rep
‘The old women transformed into mucura rats.’
b. AtSa ihya alia-ka-naka (Aikhenvald 2003:491)
men 2.pl exist-decl-pres.vis
‘You are real men.’
Notice that there are significant differences in the systems of grammatical

features that are used in this range of languages, to say nothing of the obvious
differences in how those ϕ-features are spelled out as affixes. Some languages
distinguish three values of a number feature (Yimas; see note 1), some two
values (Swahili), and some do not spell out a number feature at all (Mayali).
Some languages have two distinct genders (Arabic), some three (Greek), some
four (Mayali), some more than five (Swahili), some more than ten (Yimas),
and some more than a hundred (Tariana). These important differences should
be attributed to a rich and interesting theory of which features are present in
a language, and how those features are spelled out in the morphology. (On
the latter, see Halle and Marantz 1993 and subsequent work in Distributed
Morphology; for steps toward the former, see Harley and Ritter 2002.) But at the
same time, there is a consistent difference between nouns and adjectives when it
comes to agreement that crosscuts these differences in feature systems and how
they are manifested in morphemes. Inasmuch as the lexical category distinctions
are primarily syntactic in nature (see Baker 2003a:ch. 5, for discussion), this
consistent difference should probably be explained in terms of the syntax of
agreement.
2.1.2 Adjectives versus nouns: modifying structures

The same difference between nouns and adjectives can be seen in a different
syntactic environment, when the noun or adjective is used as an attributive
modifier, adjoined directly to a noun phrase. In this environment too, adjectives
show gender and number agreement with the head noun, whereas nominal
modifiers do not. (11) and (12) show this in Spanish; (13) and (14) show it in
Swahili; (15) and (16) are from Arabic.2
2 Attributive adjectives are also said to agree with the noun that they modify in definiteness in
Arabic. Predicate adjectives, however, do not agree with their subjects in definiteness. This
suggests to me that it is wrong to fully subsume the feature ± definite into the other ϕ–features
in this language. A plausible alternative is to say that the modificational structure is really one of
(11) a. el hombre gord-o (Spanish, agreeing adjectives)

the.m.sg man.m.sg fat-m.sg
‘the fat man’
b. las mujer-es gord-as
the.f.pl women-f.pl fat-f.pl
‘the fat women’
(12) a. la mujer-vampiro (Spanish, nonagreeing nouns)3
the.f.sg woman.f.sg-vampire.m.sg
‘the vampire woman’
b. el hombre-araña
the.m.sg man.m.sg-spider.f.sg
‘the spider-man’
(13) a. m-tu m-zuri (Swahili, agreeing adjectives)
cl1-man cl1-good (Ashton 1949:11)
‘a fine man’
b. wa-toto wa-zuri
cl2-child cl2-good
‘beautiful children’
c. mi-ti mi-zuri
cl4-tree cl4-good
‘fine trees’
(14) a. m-tu Ø-tajiri (Swahili, nonagreeing nouns)
cl1-man cl9-wealth/merchant (Ashton 1949:47)
‘a rich man, merchant’
b. Ø-bata ma-ji
cl9-fowl cl6-water
‘water fowl’
(15) a. ?ə bno ž-žūc ān (Arabic, agreeing adjectives)
son.his det-hungry.m.sg (Cowell 1964:500–1)
‘His hungry son’
b. wə rt-e zgı̄r-e
inheritance-f.sg small-f.sg
‘a small inheritance’
c. l-c arab ə s-sūriyy-ı̄n
det-arabs.pl det-syrian-pl
‘the Syrian Arabs’
apposition between two DPs: [DP D NP] [DP D A ØN ]. The second DP in apposition to the first
naturally has the same definiteness value.
3 I thank José Camacho (personal communication) for calling these examples to my attention.
(16) s-sakak-ı̄n ə l-fə dd-a (Arabic, nonagreeing noun modifier)

det-knife-pl det-silver-f (Cowell 1964:507)
‘the silver knives’
(17) shows attributive adjectives agreeing with the head noun in Mayali; (18)
shows comparable noun-noun structures in which there is no agreement in
gender.
(17) a. na-rangem na-mak ‘good boy’ (Evans 2003:182)

m-boy m-good
b. man-me man-mak ‘good food’
veg-food veg-good
c. kun-wardde kun-mak ‘good rock’
n-rock n-good
(18) a. an-bornde gun-yarl (Evans 2003:247–8)
veg-banyan n-string
‘banyan (bark) string’
b. gunj an-djomborl
kangaroo.m veg-pad
‘pad made by kangaroos’
c. gun-mogen ngan-gorle
n-bundle veg-spears
‘bundle of bamboo spears’
Only in Tariana there is a complication. Tariana’s attributive adjectives do

agree with the noun they modify through the medium of a “classifier” suffix,
as shown in (19).
(19) a. heku-na hanu-na (Aikhenvald 2003:73)

wood-cl:vertical big-cl:vertical
‘a big tree’
b. kule-kha matSa-kha (Aikhenvald 2003:88)
fishing.tool-cl:curved good-cl:curved
‘a good fishing line’
c. tSãri matS-i:te; ãtSa matSa-peni (Aikhenvald 2003:174)
man good-cl:animate men good-cl:animate.pl
‘a good man’ ‘good men’
The surprise is that when a noun modifies another noun in Tariana, it too seems
to have a classifier suffix:
(20) a. heku-na [tsuli i-tape]-na (Aikhenvald 2003:85)

tree-cl:vertical diarrhea indf-medicine-cl:vertical
‘a tree, (which is) a medicine for diarrhea’
b. heku-na [pana-phe matSa-phe]-na
tree-cl:vertical leaf-cl:leaf.like good-cl:leaf.like-cl:vertical
‘a tree which has beautiful leaves; a good-leafed tree’
This seems like a counterexample to my otherwise robust generalization. But

(20b) illustrates two important features of this structure: first, the agreeing
classifier is peripheral to the NP, not necessarily attached to the noun itself, and
second, the agreeing classifier does not replace the normal classifier associated
with the noun but rather is added to it. There is no such stacking of classifiers
on the agreeing adjectives seen in (19). I interpret this as being consistent with
my generalization, as follows. The fact that we see pana-phe and not pana-na
(leaf-CL:vertical) in (20b) shows that nouns do not themselves agree with other
nouns in Tariana. The second na in (20a–b) is a realization of agreement with the
head noun on a higher functional head that appears in this sort of modificational
structure. The agreement properties of this sort of functional head are interesting
in themselves, but are not directly relevant to the generalization now under
consideration.
What is this mysterious additional functional category? I claim that it is an
instance of the head that linguists like den Dikken and Singhapreecha (2004)
have studied under the name “linker.” In many languages, a noun (phrase)
usually cannot modify another noun (phrase) directly; some additional head
is required to make the structure work. This is true in at least Spanish and
Swahili among the languages under consideration. Some N-N structures like
(12) and (14) are possible, but this is not the most common or productive pattern
(they may be listed compounds, in fact). Rather, the particle de (Spanish) or a
(Swahili) must usually come between the head noun and the nominal modifier:
(21) a. una pulsera *(de) oro

a.f.sg bracelet.f.sg of gold.m.sg
‘a bracelet of gold, a gold bracelet’
b. ki-ti *(ch-a) m-ti
cl7-chair cl7-assoc cl3-wood
‘a chair of wood, a wood chair’
In Baker (2003a) I discussed the badness of these examples without the extra
head. I attributed this to the Noun Licensing Condition (a generalization of part
of the θ -Criterion, see the discussion of (51) for a brief review); in intuitive
terms, the modifying noun cannot receive a θ -role when it is adjoined directly
to another noun. Thus a preposition or similar particle is needed to θ -mark

the nominal modifier. From this perspective, it is not at all surprising that an
additional functional particle would be necessary in Tariana structures like
(20); if anything, the surprise is that no such particle is present in the Arabic
and Mayali examples cited. (I have no information about the productivity of
these constructions in those languages; I cherish the hope that, like Spanish and
Swahili, they are only possible in a restricted set of cases.)4
Consider now the agreement properties of the examples in (21). The modify-
ing noun clearly does not agree with the modified noun: we do not see *ki-ti ch-a
ki-ti for ‘chair of wood’ for example. But in Swahili the linking particle does
agree in gender and number with the head noun – a fact to be explained when I
generalize the theory of agreement to a wider range of categories (see section
4.1.4). I claim that the Tariana structure in (20b) is identical to (21b) except
that Tariana is a head-final language, so the linking particle is postpositional to
the NP (not prepositional), and the linking particle has no phonological content
beyond that of the agreement.
Overall, then, the generalization that adjectives undergo agreement and com-
parable nouns do not holds for attributive structures as well as for predicative
structures. This is significant because it shows that the theoretical explanation
for the difference should not be tied too closely to the syntax of predication.
The correct analysis must be general enough to apply to attributive modification
structures as well. (There are some languages in which adjectives seem to agree
in predicative environments but not in attributive ones; I return to this in section
2.5.2.)
2.1.3 Adjectives versus verbs

Next, I document the generality of the agreement asymmetry between verbs
and adjectives – the fact that verbs can show agreement in first and second
person as well as in number and gender but adjectives cannot.5 Many linguists
already realize that this is true for many languages, and it is reflected to some
degree in typological generalizations such as Stassen’s (1997) Agreement Uni-
versal, quoted in section 1.3. Here I illustrate the breadth of this observation
4 Another possibility is that linking particles are required for nominal modification in these lan-
guages too, but the particles happen to be phonologically null. It is not immediately obvious to
me how to tell if that is true or not, but I leave the matter aside; it is an issue more for my 2003a
book than for the current project.
5 It is also true, a fortiori, that nouns do not agree in first or second person any more than they
agree in number and gender. I do not illustrate this separately, but see examples (1c), (8b), and
(10b) above. See section 2.5.1 for discussion of apparent counterexamples to this generalization.
using the same range of languages as I used in the previous two subsections.
In the interests of brevity, I only present verbs showing distinctive first and
second person agreement with their subjects, but the reader should bear in
mind that several of these languages also have object agreement, and this also
has distinctive first and second person forms. (There is object agreement in
at least Swahili, Mayali, and Yimas – and maybe also in Spanish and Arabic
depending on whether object “clitics” in these languages are really instances of
agreement (see Shlonsky 1997 for Arabic and Ormazabal and Romero 2006 for
Spanish).)
Examples (1a–b) already showed that this is true for Swahili. (22) shows a
similar contrast with a first person plural subject; notice that the adjective agrees
with the subject in number and gender (animacy), but not in person.
(22) a. Tu-li-(vi)-pot-ez-a vi-tabu vy-ote. (Vitale 1981:17)

1pS-past-8O-lose-caus-fv cl8-book cl8-all
‘We lost all of the books.’
b. Tu-Ø wa-refu.
1pS-be cl2-tall
‘We are tall.’
This difference is also familiar from Indo-European languages. For example,

verbs inflect for first and second person in Spanish, but adjectives do not:
(23) a. (Nosotras) com-emos las manzanas.

we.f.pl eat-1pS the apples
b. (Nosotras) somos gord-as. (*gord-amos)
we.f.pl are.1pS fat-f.pl fat-1p
In Syrian Arabic, verbs are like adjectives in agreeing with their subjects in
gender and number:
(24) a. Wə sl-et bə rnētt-i? (Cowell 1964:420)

arrived-3sS.f hat.f.sg-my
‘Has my hat arrived?’
b. Wə sl-u l-ə ktāb-ēn taba-c i?
arrived-3pS det-book-pl two-my
‘Have my two books arrived?’
But verbs also have endings that agree with subjects in first or second person:
(25) smə c -t, smə c -ti, smə c -tu, smə c -na (Cowell 1964:193)
hear-1sS hear-2sS.f hear-2pS hear-1pS
‘I heard’ ‘you(F) heard’ ‘you(PL) heard’ ‘we heard’
Adjectives cannot take these endings, nor is there any equivalent in the adjectival
paradigm. For example, the adjectival (participial) form ‘dined’ in (26) takes
the suffix ı̄n, expressing the plurality of its first person plural subject, but there
is no expression of the first person feature. This ı̄n suffix is the same as would
be used with a third person plural subject, and a form like *mətc ašš-na with a
distinctive first person affix is apparently impossible.
(26) Katter xērak, nə hna mə tc ašš-yı̄n ya bēk. (Cowell 1964:266)
thank you we dined.ptpl-pl sir
‘Thank you, but we have already dined, sir.’
A similar difference between verbs and adjectives exists also in Tariana.

Recall that adjectives in Tariana bear classifier suffixes that agree with nouns
in various semantic features, as well as agreeing in number. In general, verbs
do not take these classifier affixes; rather, most verbs take prefixes that agree
with the subject argument.6 Included in this prefixal agreement paradigm are
special forms for first and second person subjects:
(27) a. Nuha siruri nu-walita nu-a. (Aikhenvald 2003:236–7)
I cumatá.leaves 1sS-offer 1sS-go
‘I am making an offering of cumatá leaves.’
b. Wa-na i-walita na:-pidana.
1.pl-obj 2pS-offer 3pS.say-rem.p.rep
‘You make an offering to us, they say.’
(28) shows that no comparable first or second person morphology appears on

predicate adjectives, even when their subjects bear these features.
(28) a. Ha-ehkwapi-nuku ma:tS-ite hiku nhua. (Aikhenvald 2003:495)
dem:an-world-top bad-cl:an appear I
‘I look bad in this world.’
b. Ma:tSi-pu-naka pi-rena phia. (Aikhenvald 2003:249)
bad-aug-pres.vis 2sS-feel you
‘You are feeling bad (suffering).’
In (28a), the predicate adjective partially agrees with the subject ‘I’ in that it
bears the animate singular classifier ite. But it clearly does not bear the first
person prefix nu seen in (27a), or any equivalent. Aikhenvald (2003:76) states
explicitly that “Underived adjectives never take cross-referencing prefixes.”
6 Some stative nonagentive verbs in Tariana do not bear agreement at all, neither verb-like nor
adjective-like. I analyze this interesting phenomenon in section 3.2.2.2. Classifier suffixes do
show up on verbs in relative clauses and cleft constructions. I take this to be an instance of
adjective-like agreement on a complementizer element; compare section 4.1.4.
Attributive adjectives are also incapable of agreeing in first or second per-

son. Not all languages allow attributive adjectives to modify first and second
person pronouns, but some do, including Tariana and Kinande (Bantu, related
to Swahili). When such a structure occurs, the adjective again agrees with the
pronoun in number and animacy, but not in person:
(29) a. [Pi-na inasu-ite-nuku] nuhua-wya-ne ma:-kasu nu-ñha.

2s-obj lazy-cl:an-top.obj 1s-ext-foc.subj neg-let 1sS-eat
‘I am the only one who will not let [you the lazy one] eat.’
(Tariana, Aikhenvald 2003:188–9)
b. Itwe ba-kuhi mo-tu-a-gend-ire. (Kinande)
we cl2-short aff-1pS-t-go-ext
‘We short ones went.’ (*tu-kuhi 1sS-short)
This sort of fact is not so frequently reported in grammars, but I assume that it
is general.
This time Mayali is the language that presents an instructive complication.
Verbs clearly agree with subjects (and objects) in person and number, as shown
in (30).
(30) Ngayi nga-wurlebme, la ngudda wanjh ngune-bo-rro!

I 1sS-swim.np conj you then 2dS-water-strike.imp
‘I’m going to swim, and you two strike the water.’ (Evans 2003:692)
The complication is that adjectives – particularly adjectives that are given tem-
porary, stage level readings – can take the verbal prefixes (which do not show
gender features) instead of the adjectival prefixes (which do). (31a) and (31b)
are thus both possible, with a difference in meaning.
(31) a. Alekke daluk ngal-warre (Ø-dowi-men).

f.dem woman f[Adj]-bad 3sS-die-imp
‘That woman is bad/ugly (and must die).’
b. Alekke daluk ga-warre.
f.dem woman 3sS[Verb]-bad
‘That woman’s no good (right now, she is upset, sick, etc.).’
In similar semantic situations, adjectival roots can even bear first and second
person agreement prefixes:
(32) a. Ngayih nga-mak. (Evans 2003:354)

I 1sS-good
‘I am healthy.’
b. Yi-keb-mak.
2sS-face-good
‘You are good looking.’
Taken by themselves, the examples in (32) look like counterexamples to the

generalization that adjectives cannot be inflected for person. But the alternation
between (31a) and (31b) points to a different interpretation. Mayali allows
adjectival roots to be transformed into verbs by a process of zero-derivation,
the verbal version having a more temporary, stage-level meaning (see also Heath
1984 on related Nunggubuyu). Another sign that a process of zero-derivation
exists in Mayali is the fact that “adjectives” can host noun incorporation (see
(32b)), which is otherwise a property of verbs not adjectives in Mayali and
other languages (see (42) and Baker 2003a:70–2). Furthermore, “adjectives” –
and even nouns referring to life stages, but not others – can bear tense/aspect
suffixes that are normally attached to verbs:
(33) Ngaye nga-wurdurd-ni galuk ngaban-na-ng. (Evans 2003:357)

I 1sS-child-past bye.and.bye 1sS/3pO-see-pp
‘When I was a child, I would see them.’
The agreement in (32) thus falls into place. It is not at all surprising that “adjec-
tives” can show first and second person agreement when they are really verbs
created from adjectival roots by zero-derivation. The data in (32) are thus no
more problematic for my generalization about agreement than a sentence like
The sky usually clears at night is for the generalization that adjectives like clear
in English are not inflected for tense or agreement.
This point is important because it helps us to interpret the typological record.
Typologists with a functionalist orientation, such as Stassen (1997), have argued
for implicational universals concerning agreement, stateable as follows (see also
Croft 1991:82):
(34) a. If class-membership predicates agree in person, then so do property

predicates.
b. If property predicates agree in person, then so do event predicates.
If one equates “class-membership predicates” with nouns, “property predicates”

with adjectives, and “event predicates” with verbs, this research purports to find
the following range of languages and no others:
(35) a. languages without agreement (in person) on any predicates

b. languages with person agreement on verbs but not adjectives or nouns
c. languages with person agreement on verbs and adjectives but not nouns
d. languages with person agreement on all three lexical categories
The languages described in (35b) are very common, and are the type that I have
focused on in this section. Languages like (35a) are also common – East Asian
languages and some West African languages are obvious cases in point – but
they are of only negative interest for a theory of agreement. Languages that
superficially fit the description in (35d) are not so common, but they do exist:
Turkish, Lango, Nahuatl, Abaza, Guarani, and Jakaltek are some examples;
I return to these in section 2.5.1 below. The crucial question now is whether
typological studies like Croft’s and Stassen’s give us any reason to think that
languages of the type described in (35c) exist.
I claim that these studies do not in fact support the existence of such lan-
guages. Whereas it is fairly safe to equate “class-membership predicates” with
the syntactic category of noun and “event predicates” with the syntactic category
of verb, equating “property predicates” with the syntactic category of adjective
is notoriously unreliable. Many languages have property-denoting words that
are syntactically verbs or nouns rather than adjectives (Dixon 1982). Further-
more, it is common for verbs to be derived from adjectival roots, derivations
that may or may not be morphologically marked (see Baker 2003a:159–69 for
examples and discussion). Mayali is an excellent example. Given this, imagine
a language that does not have adjectives as a distinct class (at the relevant level),
and that has person agreement on verbs but not on nouns (as usual). Stassen
and other functionalist typologists would interpret this as a language in which
the antecedent clause of (34a) is false, but the antecedent clause of (34b) is
true. Hence it supports their implicational universals. Such a language does
not fit the description in (35c), however. It would not be a language in which
adjectives agree with their subjects in person; rather it would be a language
that has verbs where other languages have adjectives. One can in principle tell
the difference between the two categories whenever there is independent evi-
dence as to whether a property-denoting predicate is an adjective or a verb –
for example, when there is evidence from the distribution of copulas, or from
the attachment of tense morphology, or from the form of negation. (These are
Stassen’s (1997) other tests for a “verbal encoding strategy.”) Suppose that we
remove from the sample all those cases where “property denoting” words are
verbs as opposed to adjectives by criteria that do not involve agreement. Is it
still true of this residue that it contains languages in which “property denoting
words” (adjectives) agree in person but nouns do not? As far as I can tell, there
is no evidence that it is.7 So the statement that adjectives never show person
7 Stassen (1997) gives three distinct criteria for identifying a “predication strategy” as being
nonverbal: a predicate is nonverbal, (i) if it does not have person agreement in a language
The category-theoretic infrastructure 27
agreement could very well be universal (with a qualification to be given in

section 2.5.1).
2.2 The category-theoretic infrastructure

Section 2.1 documented that the differences in how the major lexical categories
behave with respect to agreement are too consistent across languages for us to
be content with stipulating what feature slots are present on particular categories
on a language-by-language basis. Rather, the pattern should be derived from a
general theory of the lexical categories, in interaction with a general theory of
agreement. In this section, I present the first of these two subtheories, giving
a precise characterization of the syntactic structures over which the observed
agreement relationships hold. In fact, I believe that my (2003a) category theory
serves the need well, so I review the basics of that theory and some of its
motivation here. The only change that I propose is the addition of a layer of
functional structure above the lexical heads, where agreement morphemes can
be housed.
2.2.1 Essentials of category theory

It has been realized since the Middle Ages that the lexical categories have differ-
ent syntactic distributions. For example, in English and many other languages,
only verbs can be used alone as the main predicate of a matrix clause ((36)),
only nouns can be used as arguments in subject and object positions ((37)),
and only adjectives can be used as bare adnominal modifiers ((38)):
(36) a. Chris sings.

b. *Chris talls.
c. *Chris mans.
(37) a. Chris respects men.
b. *Chris respects tall.
c. *Chris respects sing.
where event predicates do have person agreement, (ii) if it needs a copular/auxiliary particle in
a language where event predicates do not need separate copular/auxiliary particles, and (iii) if
it is negated using a different “negation strategy” from that used with event predicates. Stassen
does not mention cases in which these criteria give conflicting results. In particular, he does
not mention any cases in which a language has predicates that show person agreement but are
nonverbal by the auxiliary criterion or the negation criterion. I interpret this as evidence in favor
of the claim that true adjectives never show person agreement. See also Croft 1991:131, where
the author observes that it is languages in which “adjectives” bear the tense-aspect morphology
of verbs that allow adjectives to agree like verbs, for example in person.
(38) a. The tall person arrived.

b. *The man person arrived.
c. *The sing person arrived.
Until recently, most generative work on the lexical categories has concen-
trated on capturing their similarities, through X-bar theory and the positing of
parallel systems of functional categories. In Baker 2003a, however, I sought
to explain these and other, more subtle morphosyntactic differences in a uni-
fied way. My theory is based on the definitions in (39), together with the basic
principles in (40).
(39) a. Verbs are lexical categories that license a specifier.

b. Nouns are lexical categories that have a referential index.
c. Adjectives are lexical categories that have neither a specifier nor a
referential index.
(40) a. All the θ -roles of a head must be coindexed with a maximal projection
immediately dominated by a projection of that head.
b. The Noun Licensing Condition (NLC):
A referential index must be coindexed with a dependent element that it
c-commands (a θ -role, a bound pronoun, or a trace of movement).
c. The Reference-Predication Constraint:
No syntactic node can have both a referential index and a specifier (cf.
the logical incompatibility of predication and reference, Geach 1962).
Rather than merely saying that verbs are +V – a meaningless feature that no
principle of grammar refers to – I say that verbs are the only lexical category that
has a specifier as well as a complement. This is seen most clearly in paradigms
like (41), from the Nigerian language Edo. Even in a small clause environment,
where Tense is not a factor, verbs can take subjects directly, but adjectives and
nouns can only be predicated of a subject if they are the complements of a
copular particle:
(41) a. Úyı̀ yá [èmátòn pèrhé].

Uyi made metal be.flatV
‘Uyi made the metal be flat.’
b. Úyı̀ yá [èmátòn *(yé) pèrhè].
Uyi made metal pred flatA
‘Uyi made the metal flat.’
c. Òzó yá [Úyı̀ *(rè) òkhaèmwèn].
Ozo made Uyi pred chief
‘Ozo made Uyi a chief.’
This basic difference between verbs and other categories is harder to see
in Indo-European languages and others in which the copular particle Pred is
phonologically null. The three lexical categories look structurally parallel in
(42), for example.
(42) a. Chris made [the metal shine].

b. Chris made [the metal smooth].
c. Chris made [them leaders].
Nevertheless, one can use indirect means to tell that there is an extra structural
projection in sentences like (42b) and (42c) that is not needed in (42a). For
example, a noun that expresses the sole argument of an intransitive verb can
incorporate into that verb in polysynthetic languages like Wichita and Mohawk
((43a)), but the sole argument of a predicate adjective or a predicate noun cannot
incorporate into the adjective or noun ((43b–c)).
(43) a. Hánnhirh ta:c-ehe:k-ʔ irhawi. (Wichita: NI into V)

ground.loc t/agr-cloth-be.lying (Rood 1976:5)
‘The cloth is lying on the ground.’
b. *né:rhir?as-tac ti-ʔ i. (OK: tac ti-ʔ i né:rhirʔ a)
buffalo-fat t/agr-be fat t/agr-be buffalo
‘The buffalo is fat.’ (Wichita: no NI into A; Rood 1976:13)
(Also OK: tac ti-ré:rhir?as-?i ‘The buffalo is fat.’)
c. *Ka-nerohkw-a-nuhs-a’ (OK: Ka-nuhs-a’ ne o-nerohkwa-kvha.)
Ns-box-Ø-house-nsf Ns-house-nsf ne Ns-box-former
‘That box is a house.’ (Mohawk: no NI into N)
(e.g., a child’s playhouse, or a homeless person’s shelter)
This contrast follows from the assumption that only the subject of the verb
is generated inside the maximal projection of the verb, together with the fact
that head movement can never lower a word to attach it to a word that did not
c-command it in the first place.
The same difference between verbs and the other lexical categories can often
be detected using unaccusativity diagnostics – phenomena in which the sole
argument of an unaccusative (nonagentive) verb behaves like a direct object.
A well-known case is (44a) from Italian, in which the sole argument of a verb
like ‘sink’ (but not one of an agentive verb like ‘telephone’) can be expressed
as a partitive clitic ne attached to the finite verb. In the Government-Binding
era, this was taken as evidence that the theme argument of ‘sink’ was generated
inside the VP, where its trace is properly governed by the verb (Burzio 1986).
(44) a. Ne sono affondate due. V: OK

of.them are sunk two (Burzio 1986, Cinque 1990)
‘Two of them sank.’
b. *Ne sono buoni pochi (dei suoi articoli). A: bad
of.them are good few (of his articles) (Cinque 1990:7)
‘Few of them are good.’
c. ?*Ne sono professori molti. N: bad
of.them are professors many (Baker 2003a)
‘Many of them are professors.’
The point of interest here is that the sole argument of thematically similar
adjectives and nouns cannot be expressed as a ne clitic attached to the finite
verb, as shown in (44b–c).8 I interpret this as showing that the subject of the
nonverbal predication is not inside the AP or NP where it would be lexically
governed and thus licensed by the adjective or noun. Rather, it is in the specifier
of a nonlexical projection, the PredP, and moving from there incurs a violation
(see Baker 2003a:62–9 for a full discussion, couched in more current theoretical
terms).
There is also indirect evidence of the structural distinction in the simple
paradigm in (36) from English. The agreement-bearing Tense node can merge
with the verb under adjacency, but it cannot merge with the predicate noun or
adjective. I take the inability of the predicate noun or adjective to inflect for
tense in English to be due to the presence of a null Pred head: this null head
blocks Tense from merging with the lexical head at PF.
If analyses like these are correct, then there is a systematic difference between
the phrase structures that verbs appear in and the phrase structures that nouns
and adjectives appear in. This structural difference can be used to explain the
special agreement properties of verbs, given that agreement is defined over
syntactic structure.
The defining property of nouns that distinguishes them from verbs and adjec-
tives is not in the kinds of phrase structures that can be built from them, but
rather in the fact that noun projections are associated with a referential index.
This can be seen most clearly in paradigms like (45), originally discussed by
Kayne (1984).
(45) a. Italy{i} ’s invasion of Albania (grieved the expatriate community).
b. The Italian invasion of Albania . . .
c. Italy{i} ’s destruction of itself{i} . . . .
8 Note that there are a few adjectives in Italian that do permit ne-cliticization; see section 3.1.1 for
discussion.
d. *The Italian destruction of itself{i} . . . .

(compare The Italian self-destruction . . .)
Inside a derived nominal, the name of a country and an adjective derived from it
can be nearly synonymous, either one serving to express the agent of the event
named by the derived nominal, as shown in (45a) and (45b) (see also Grimshaw
1990, Giorgi and Longobardi 1991, and many others). But when the object of
the derived nominal is a reflexive anaphor, a difference emerges: the genitive
noun phrase is fine ((45c)), but the nationality adjective is strongly degraded
((45d)). My interpretation of this fact is that the noun Italy has a referential
index, and hence can be the local syntactic binder that the anaphor requires.
The adjective Italian, in contrast, has no referential index, and hence cannot
be the local binder that itself requires. (45d) is thus ruled out as a violation of
Chomsky’s Binding Principle A.
This kind of account extends to the more elementary fact that nouns and their
projections can serve as the subjects and direct objects of verbs and other θ -
marking heads, given Williams’s (1989) view that θ -roles are a kind of anaphor.
Under this assumption, (46b) is ruled out for the same reason as (45d): the agent
θ -role of the verb destroy is an anaphor that has no syntactic binder.
(46) a. Italy{i} destroyed<Ag{i} , Th{k} > Albania{k}

b. *Italian destroyed<Ag{i} , Th{k} > Albania{k}
In the same way, (37b) and (37c) are ruled out because (unlike the noun men)
the adjective tall and the verb sing do not bear a referential index that can bind
the anaphoric internal θ -role of the verb respect.
The primary significance of nouns being the only lexical categories that have
referential indices for the theory of agreement comes from (47), a principle that
is implicit in Chomskian work and explicit in HPSG work, including Pollard
and Sag (1994) and Wechsler and Zlatić (2003:11).
(47) XP can have intrinsic ϕ-features (pre-specified values for person,

number, and gender) only if XP has a referential index.
One may speculate as to why (47) should be true. I suspect that the ulti-
mate answer has to do with an even more fundamental property of nouns than
their bearing a referential index. In asking why only common nouns can be
the restrictors of quantifiers (No dog barked vs. *No big barked and *No run
barked), Geach (1962) and Gupta (1980) proposed that only nouns have criteria
of identity. By this, they meant that only nouns have the kind of lexical seman-
tics that supports judgments of whether two objects are the same according to
some standard. The examples in (48a) are thus meaningful, but the examples
in (48b–c) are not.
(48) a. i. That is the same man as you saw yesterday.
ii. That is the same water as was in the cup this morning.
iii. The Chinese want to have the same liberty as the Americans have.
b. i. #That is the same long as this.
ii. #She is the same intelligent as he is.
c. i. #I saw Julia the same sing as Mary did.
ii. #I watched Nicholas the same perform a stunt as Kate performed.
This lexical semantic property underlies the fact that only nouns have referen-
tial indices, because referential indices are used as a grammatical expression
of (approximately) presupposed coreference, and coreference is ultimately a
judgment of sameness: X and Y are coreferential if the entity that X refers
to is judged to be the same entity that Y refers to according to a suitable
standard. Common noun meanings give the standard that makes this judg-
ment possible, so the linguistic expressions involved in the anaphoric rela-
tionship must be nouns (or pronouns – functional categories that also bear
indices).
It is reasonable to think that the same criterion of identity that underwrites
the referential index also underwrites the presence of ϕ-features. This is clear-
est for number features like singular, dual, and plural, because they concern
enumeration, just like the quantifiers that Geach and Gupta were interested
in. In an observation that goes back to Frege, Geach and Gupta observe that
counting – deciding whether there is one of a thing, or two of it, or more –
presupposes being able to individuate those things. This in turn requires being
able to decide which constitute the same thing. Since only common noun
meanings permit this, only nouns can be intrinsically singular or dual or
plural.
This reasoning also applies to person features such as first person and second
person. These features relate to the referential properties of a linguistic expres-
sion, so it makes sense that they too are dependent on a criterion of identity.
For example, all of the first person pronouns in a sentence like (49) include the
same designated person (the speaker) in their reference.
(49) I{i} told my{i} sister that our{i+k} mother would mail me{i} the document.
Therefore it makes sense that these person features are associated with the
same class of linguistic entities as referential indices are. (I return in detail to
the special qualities of first and second person elements in section 4.3.)
The connection between ϕ-features and the criterion of identity/referential

index that defines the category of noun is least obvious for gender features. In
part, this is because these do not have any stable semantic interpretation the way
that number and person features do (see Harley and Ritter 2002). Nevertheless,
it seems plausible to say that gender features have at least the function of placing
the referent of the expression within the natural folk taxonomy that is imposed
by the language. The gender 10 marking on the demonstrative pronoun izi in
(50) from Chichewa (Bantu), for example, has a semantic interpretation to the
extent that it constrains the demonstrative to refer to something that could be
described by a class 10 noun in the language – to bees or zebras or elephants,
say, but not to people, or lions, or knives.
(50) Awa zi-na-wá-lúm-a izi. (Mchombo 2004:52)

cl2.these 10S-past-2O-bite-fv cl10.these
‘These (e.g. bees) bit these (e.g. hunters).’
It follows that two pronouns of the same gender can potentially refer to the same
entity, whereas two pronouns of different genders generally cannot. Gender
can thus act as a kind of surrogate criterion of identity, inherited from the set
of criteria of identity that are associated with common nouns that have that
gender. From this perspective, it seems natural that the linguistic entities that
have criteria of identity are the same as those that have grammatical gender.
At least, that is what I assume, pending further inquiry into the syntax and
semantics of gender across languages.
The third and final lexical category is the adjective.9 Adjectives already differ
from verbs in not having a specifier, and from nouns in not having a referential
index. I claim that that is all one needs to say about adjectives. The specific
morphosyntactic properties of adjectives follow simply from this definition
in interaction with general principles such as the ones in (40). Consider, for
example, the fact that only adjectives can be used as bare attributive modifiers
in a noun phrase, as shown again in (51).
(51) a. The tall person arrived.

b. *The man person arrived.
c. *The sing person arrived.
9 Unlike Jackendoff (1977), I consider P to be a functional category, for reasons discussed in

Baker 2003a:303–11. Note that the Reference-Predication Constraint in (40c) insures that, from
a theoretical point of view, there is no fourth lexical category that has the positive qualities of
both a noun and a verb.
(51b) is ruled out by the Noun Licensing Condition in (40b) (essentially a

generalization of one half of Chomsky’s (1981) θ -Criterion): there is no θ -role
or other dependent element that man can be coindexed with in this structure. In
essence, man as an attributive modifier fails to receive a θ -role. (51c) is ruled
out by (40a), the other half of the traditional θ -Criterion: the verb sing has a
θ -role to assign to a subject, but there is no nominal phrase within its maximal
projection for it to assign that θ -role to. The adjective tall, in contrast, has no
θ -role to assign and no index that needs to be licensed. Therefore, the structure
in (51a) violates neither (40a) nor (40b), and is grammatical. This, then, is
the kind of explanation I give for why adjectives appear in a different range of
syntactic positions than other categories. Similar accounts can be given for why
only adjectives are used as resultative secondary predicates (e.g., I pounded the
metal flat/*shine/*a sword) and as the complements of dedicated degree heads
(e.g., Chris is too smart/*hunger/*a genius).
2.2.2 Functional heads as the loci of agreement

There is one additional assumption about structure, not made in Baker 2003a,
that is useful in explaining why the different lexical categories have different
behaviors with respect to agreement. Let us assume that it is not technically
the lexical category that agrees with something in its environment, but rather a
functional category that immediately dominates the lexical category, as stated
in (52).
(52) Any lexical category can be immediately dominated by the projection of a

functional head that matches it in gross categorical features. Functional
heads, unlike lexical heads, can manifest agreement.
For verbs, this is a familiar assumption. There are strong and well-known rea-
sons to say that “verbal agreement” is not an inherent property of V nodes per
se, but rather of Tense/Infl, a functional category that selects the verb phrase
and often fuses with the verb to form a single phonological word. For example,
there is no agreement on verbs in English when the Tense is nonfinite rather
than finite ((53a–b)). Furthermore, when Tense is kept separate from the main
verb by an intervening negation, agreement shows up on the Tense position,
not on the verb itself ((53c)).
(53) a. Chris likes swordfish.

b. For Chris to like(*s) swordfish (would be unfortunate).
c. Chris does not like(*s) swordfish
This is often generalized to object agreement as well: object agreement is on

the functional head v (or Aspect, or AgrO), which selects VP and fuses with
the verb. (52) is the assertion that this holds true for nouns and adjectives as
well as for verbs.
Empirical support for (52) comes from incorporation constructions, in which
one lexical category undergoes head movement to adjoin to another lexical
category (Baker 1988). Li (1990) shows that it is not possible for a lexical head
to move through a functional head position on its way to another lexical head
position. Standard effects of Li’s Generalization are the fact that only verbs that
are not marked for tense incorporate into higher verbs in causative constructions
and other complex verbal predications. Now, if the agreement that is often
realized on a verb is really a property of Tense or some other functional head
that dominates VP, Li’s Generalization predicts that verbs incorporated into
other verbs should be stripped of their agreement affixes as well as of their
tense/aspect markers. This is true; the minimal pair in (54) from Southern Tiwa
provides an illustration.
(54) a. Ow-t’am-ban hliawra-de u-napir-hi-’i. (Frantz 1993)

2sS/aO/cO-help-past lady-sg 3sS/cO-sew-fut-sub
‘You helped the lady sew.’
b. Ow-napir-t’am-ban hliawra-de.
2sS/aO/cO-sew-help-past lady-sg
‘You helped the lady sew.’
The unincorporated verb ‘sew’ in (54a) bears its own agreement prefix, as well
as tense and subordination suffixes. These tense and subordination suffixes are
predictably absent when ‘sew’ is incorporated into the higher verb ‘help’ in
(54b), in accordance with Li’s Generalization. The agreement prefix u seen in
(54a) also disappears in (54b), suggesting that it too is attributable to a functional
category that dominates VP. As a result, the complex verb is ow-napir-t’am-ban,
not something like *ow-[u-napir-hi-(‘i)]-t’am-ban.
Agreement affixes are also stripped of adjectives in the Hebrew adjective
incorporation construction in (55b), discussed by Borer (1991).
(55) a. Sney ha-sir-im šxor-im.

two the-pot-pl black-pl
‘The two pots are black.’

b. Sney sir-im hi-šxir-u [AP t haki še epšar].
two pot-pl V-black-m.pl most possible
‘Two pots blackened as much as possible.’
The predicate adjective in (55a) bears a masculine plural suffix im in agree-

ment with the subject ‘two pots’, but this affix does not appear in (55b) when
the adjective is incorporated into the inchoative verb (although the derived
verb as a whole does agree with its subject in the usual way for Hebrew). This
also follows from Li’s Generalization together with (52), the assumption that
agreement is not a property of the adjective proper, but rather of a functional
head that dominates AP. This functional head must be omitted when A-to-V
incorporation takes place, or the head-movement would be blocked. When the
functional head is omitted, so too is the number–gender agreement with the
subject that is otherwise characteristic of adjectives in Hebrew.10
Although nouns do not agree the way that verbs and adjectives do, they often
do have inflectional affixes that display the number and/or gender of the noun
itself. The incorporation test shows that this inflection too should be attributed to
a functional head distinct from the noun proper. For example, (56) from Mayali
shows that the prefix gun, which expresses the neuter gender of the direct object
ganj ‘meat’ in (56a), is necessarily absent when the noun incorporates into the
verb as in (56b).
(56) a. Aban-yawoih-warrgah-marne-ginje-ng gun-ganj. (Evans 2003:1)

1sS/3pO-again-wrong-ben-cook-pp n-meat
‘I cooked the wrong meat for them again.’
b. Aban-yawoih-warrgah-marne-(*gun)-ganj-ginje-ng.
1sS/3pO-again-wrong-ben-(n)-meat-cook-pp
‘I cooked the wrong meat for them again.’
This range of evidence thus supports the view that any lexical category can be
dominated by a functional category. Features of person, number, and gender
are spelled out on the functional category, not on the lexical category proper. In
informal statements throughout this book, I continue to use phrases like “lexical
category X agrees with Y,” but these expressions are always to be understood
as shorthand for the more accurate “the functional head that immediately dom-
inates X and fuses with it to form a single word at PF agrees with Y.”
Putting these ideas together, I posit the partial structures in (58) for basic
verbal, adjectival, and nominal predications like those in (57).
10 It is true quite generally that inchoative and causative verbs derived from adjectival roots do
not contain adjectival agreement morphology inside them; the verbal morphology replaces the
adjectival inflection, rather than being built on top of it. What is not so clear is which of these
inchoative and causative formations are derived by adjective incorporation in the syntax, and
which (if any) are formed by some kind of lexical derivation.
(57) a. Ni-li-anguka. (Swahili)

1sS-past-fall
‘I fell.’
b. Ni-Ø m-refu.
1sS-be cl1-tall
‘I am tall.’
c. Ni-li-po-kuwa ki-jana . . .
1sS-past-when-be cl7-child
‘When I was a child . . .
(58) a. FVP b. PredP
FV VP NP Pred´
NP V´ I Pred FAP
I V (PP) Ø FA AP
fall A
tall
c. PredP
NP Pred´
I Pred FNP
Ø FN NP{i}
N{i}
child
Not shown in (58b–c) is the projection of the copular/auxiliary verb and any
verbal functional heads that dominate that. Any such structure is higher than
PredP and largely irrelevant to the agreement dynamics currently under study.
The verbal clause in (58a) may well have additional higher functional heads as
well, of course.
Notice that there are differences in these structures that the differences in
agreement can plausibly be pinned to. First, since nouns are the only category
that has intrinsic ϕ-features, the complement of FN in (58c) has ϕ-features of
its own, whereas the complement of FA in (58b) does not. This could affect the
ability of the functional head to agree with the subject in Spec, PredP (known to
be possible in (58b) but not in (58c)). Second, since verbs are the only category
that take subjects (specifiers) directly, FV c-commands the subject of predication
in (58a) whereas FA in (58b) does not. The next step, then, will be to articulate
a general theory of agreement that is indeed sensitive to structural distinctions
such as these.
If FV in (58a) is another name for Tense/Infl, as I assumed above, what
are FA and FN ? An initially plausible answer might be that FA is Degree and
FN is Determiner, these being the most familiar functional categories that are
associated with AP and NP, respectively. But facts from Spanish suggest that
this is wrong. Spanish has independent words that are of category Degree and
Determiner, both of which can be seen in (59).
(59) L-as chic-as estab-an tan enferm-as que no pod-ı́an habl-ar.

the-f.pl girl-f.pl were-3pS so sick-f.pl that not could-3pS talk-inf
‘The girls were so sick that they couldn’t talk.’
These phonologically independent and semantically meaningful functional

heads are distinct from the inflectional affixes as that express gender and number
on the noun and the adjective. (There is an additional manifestation of feminine
plural as on the determiner; I return to this in section 4.1.2.) This suggests that
FN and FA are distinct from Determiner and Degree, and lower than them in
the functional structure. FN might be Number, in the sense of Ritter (1991) and
others. FA does not correspond to any other, independently known functional
category that I am aware of. I continue to refer to these heads simply as FN and
FA , leaving their exact nature open.11
The statement in (52) does not necessarily require that a functional head be
generated above each lexical head; it merely admits of the possibility. I thus
leave open the possibility that whether such heads are generated or not may vary
from language to language, and even from lexical item to lexical item within
a single language. For example, Swahili has both agreeing and nonagreeing
adjectives. The latter are said to come from Arabic, but Ashton (1949) reports
no syntactic difference between them. (60) compares the nonagreeing adjec-
tive safi ‘clean’ with the agreeing adjective zuri in predicative and attributive
environments.
11 One might think that FN and FA are the same thing as the n and a heads that take bare roots as
complements in Marantz 2000 and related work in Distributed Morphology. That is a conceivable
identification, but unlike Marantz I assume that the complements of these heads are fully and
intrinsically specified for syntactic category (see Baker 2003:265–75). I also permit there to
be instances of AP and NP that are not dominated by FN P and FA P, as discussed in the next
paragraph.
(60) a. Sahani (zi-0) safi. compare: Ch-akula hiki si ki-zuri.

plates cl10-be clean cl7-food this not cl7-nice
‘The plates are clean.’ ‘This food is not good.’
b. ch-umba safi compare: ch-akula ki-zuri
cl7-room clean cl7-food cl7-nice
‘a clean room’ ‘good food’
Similarly in Mayali, there are nouns that bear an overt gender prefix, and there
are other nouns from each gender class that do not. (61) displays examples of
both kinds from Evans 2003:182. (Agreeing adjectives are included in these
examples to make visible the gender that each noun has for purposes of agree-
ment.)
(61) a. na-rangem na-mak vs. bininj na-mak
m-boy m-good man m-good
b. ngal-kohbanj ngal-mak vs. daluk ngal-mak
f-old.woman f-good woman f-good
One way to model this type of variation is to assume that each lexical category
may or may not be immediately dominated by a functional category which
is a potential target for agreement. Whether a given functional head appears
or not can be regulated by the grammar of a language as a whole – plausibly
there are none of these functional projections in an agreementless language like
Chinese – or by individual lexical items within the language in question. When-
ever the functional head is absent, there is no agreement either, by (52).
This cannot be the whole story about why we observe variation in the presence
of agreement, however. It is also necessary to stipulate that an existing functional
head may or may not be designated as a probe (agreement-bearing head) that
seeks for features in its environment to agree with. To see this, consider the fact
that finite verbs agree with their subjects in Spanish but not in Japanese. One
cannot simply say that there is no Tense node above the verb phrase in Japanese,
because one can observe the presence of tense in Japanese both semantically
and morphologically. Rather, we must say that the Tense head is present but not
active for agreement (not a probe) in Japanese, whereas it is active in Spanish.
Similarly, both English and Spanish have phonologically overt articles, but
these are probes for agreement in Spanish but not in English.
An important moral of (60) and (61) is that it is not possible to predict in a
principled way whether or not there will be agreement in a given configuration
in a given language. Rather, the emphasis for syntactic theory must be on what
a given head agrees with and in what features it can agree, given that the head
is present and can agree at all.
2.3 The agreement-theoretic contribution

Now that we have a handle on the different structural configurations that the
lexical categories appear in, the next task is to introduce a general theory of
agreement that combines with these structural differences to give the systematic
differences observed in section 2.1. As my starting point, I adopt the theory of
Agree developed by Chomsky (2000, 2001), who had facts about agreement
in expletive and quirky subject constructions in English and other Western
European languages primarily in mind. The essential components of this theory
are summarized in (62), together with convenient names that I can use to refer
to each subcondition. Also included is a page number from Minimalist inquiries
(MI, Chomsky 2000) where the condition is discussed.
(62) A functional head F agrees with XP, XP a maximal projection, only if:
a. F c-commands XP (the c-command condition, MI:122).12
b. There is no YP such that F c-commands YP, YP c-commands XP, and YP
has ϕ-features (the intervention condition, MI:122).
c. F and XP are contained in all the same phases (e.g., full CPs) (the phase
condition, MI:108).13
d. XP is made active for agreement by having an unchecked case feature
(the activity condition, MI:123).
I first review briefly the original motivation for each clause in this condition.
I then propose a modification that allows it to be applied to constructions that
contain adjectives.
2.3.1 Characteristic properties of agreement

The requirement that the agreeing head be in a c-command relationship with the
agreed-with nominal in (62a) is largely taken for granted, c-command being a
requirement for many syntactic relationships, including binding and movement.
Perhaps the most striking evidence in its favor is the contrast in Hindi shown
in (63), as analyzed by Bhatt (2005:775) (see also Boeckx 2004:26).
12 An expression X c-commands another expression Y if and only if X does not dominate Y and
every phrase that dominates X dominates Y (Chomsky 1986:8; see also Reinhart 1983 and much
subsequent work). I further follow Chomksy 1986:7 in assuming that X dominates Y only if
every segment of X dominates Y. In a structure like [X Y [X . . .]] where Y is adjoined to X,
there are two segments of X, only one of which dominates Y, so the category X as a whole does
not dominate Y in the relevant sense.
13 Bobaljik and Wurmbrand (2005) argue that even smaller verbal projections than CPs are opaque
to agreement when they are the complements of lexical verbs. Their argument is based on the
claim that there are two kinds of restructuring, one that allows agreement at a distance and one
that does not. I do not go into the matter of agreement on restructuring predicates in this work.
The agreement-theoretic contribution 41
(63) a. Vivek-ne kitaab parh-nii chaah-ii.

Vivek-erg book.f.sg read-inf. f.sg want-perf. f.sg
‘Vivek wanted to read the book.’
b. Mona kuttõ-ko dekh-naa/*nii chaah-tii thii.
Mona dog.m.pl-acc see-inf/inf. f.sg want-hab.f.sg be-past. f.sg
‘Mona wanted to see the dogs.’
Bhatt shows that when the subjects of certain verbs that take infinitival comple-
ments are marked by a postposition (the ergative ne) and hence ineligible for
agreement, the matrix verb can agree downward with the object of the lower
verb, as in (63a). But the situation is not symmetrical: (63b) shows that when
the argument of the lower verb is marked by a postposition (dative/accusative
ko), it is nevertheless impossible for the lower verb to agree with the subject
of the higher verb, even though the higher verb does. This looks like evidence
that the agreeing head must c-command the agreed-with phrase at some point
in the derivation. Nevertheless, it is this uncontroversial and little-examined
assumption about agreement that I end up modifying most dramatically.
Example (62b) expresses the notion that, although agreement can be a fairly
long distance relationship, it is blocked if there is an intermediate phrase
between the agreeing head and the agreed-with phrase that bears the same
kinds of features as are involved in the agreement relationship. Perhaps the most
striking example of this condition at work in constraining agreement is the fol-
lowing, from Icelandic (Schütze 1997:108–9, Boeckx 2000, Bobaljik to appear).
(64) a. Jóni virð -ast/?*virð -ist vera taldir lı́ka hestarnir.

John.dat seem-3pS/seem-3sS to.be believed.pl to.like horses.nom.pl
‘John seems to be believed to like horses.’
b. Mér ?*virð -ast/virð -ist Jóni vera taldir lı́ka
Me.dat seem-3pS/seem-3sS John.dat to.be believed.pl to.like
hestarnir.
horses.nom.pl
‘John seems to me to be believed to like horses.’
(64a) is an example of the matrix verb ‘seem’, when it has a dative subject,
agreeing long distance with the plural nominative object ‘horses’ across three
other verbal heads (‘to be’, ‘believed’, and ‘to like’). (64b) is identical, except
that an extra dative argument has been added to the structure, the optional
experiencer of ‘seem’. This experiencer functions as the subject of ‘seem’,
causing the dative NP ‘John’ (thematically the subject of ‘like’) not to raise as
high, so that it remains inside the c-command domain of ‘seem’. In this structure,
it is impossible for ‘seem’ to agree with the nominative object ‘horses’, the
agreement apparently blocked by the presence of another NP between them.

Such facts motivate the intervention condition in (62b).14
The intervention condition has also been applied to double object construc-
tions by authors like Boeckx and Niinuma (2004) to explain why in many cases
the verb can show object agreement with the NP bearing the goal/benefactive
θ -role, but not with the NP bearing the theme θ -role. (65) is a typical example
of this widespread pattern, taken from Chichewa (Mchombo 2004:80–3).
(65) a. A-lenje a-ku-phı́k-ı́l-á anyanı́ zı́-túmbûwa.

2-hunters 2S-pres-cook-appl-fv 2-baboons 8-pancakes
‘The hunters are cooking the baboons some pancakes.’
b. A-lenje a-ku-wá-phı́k-ı́l-á zı́-túmbûwa (anyáni).
2-hunters 2S-pres-2O-cook-appl-fv 8-pancakes 2-baboons
‘The hunters are cooking them (the baboons) some pancakes.’
c. *A-lenje a-ku-zı́-phı́k-ı́l-á anyáni (zı́-túmbûwa).
2-hunters 2S-pres-8O-cook-appl-fv 2-baboons 8-pancakes
‘The hunters are cooking them (some pancakes) for the baboons.’
The asymmetry in (65) follows from (62b) given the assumptions that both
objects of the verb are inside the c-command domain of the agreement-
bearing functional head (here v), and that the goal argument asymmetrically
c-commands the theme argument. The second assumption is consistent with
Barss-Lasnik (1986) c-command tests applied to the base structure in (65a),
which generally show that the benefactive asymmetrically c-commands the
theme (see also Marantz 1993). Thus, the structure of a sentence like (65) is
14 There has been some controversy in the recent literature about the so-called “defective inter-
vention” effect in (64b), for both empirical and conceptual reasons. First, it is not always true
that a dative NP prevents Tense from agreeing with a lower nominative in Icelandic; there
is no intervention effect when Tense and both NPs are all in the same clause, for example
(Bobaljik, to appear). Second, dative arguments seem never to prevent Tense from agreeing
with a lower nominative in related languages, such as Dutch (Hans Broekhuis, personal com-
munication). Therefore, it is not clear that this intervention effect is general enough to motivate
a fundamental principle. Finally, from a conceptual viewpoint, it is not clear why a phrase
that cannot itself be agreed with should count as an intervener that can block agreement with
another NP.
For purposes of this work, I need not worry much about these puzzles, because none of the
cases that are crucial to my theory involve defective intervention. Rather, they all involve simple
intervention, in which the phrase X that prevents a head F from agreeing with a more remote
phrase Y is itself agreed with by F (see, for example, (73) and (76b) below). In this respect,
they are like (65), which is a crosslinguistically robust effect. Nor is this sort of intervention
effect conceptually problematic. So I take (62b) – with perhaps refinements about exactly what
features YP needs to have to count as an intervener – to be a true principle of agreement even
if one of the examples that initially motivated it should turn out to be dubious.
approximately (66) (not shown is the verb ‘cook’ moving through the applica-
tive head to land in v).15
(66) [ v(=FV) [ApplP baboons Appl [VP cook pancakes ]]]
Agree OK Agree blocked
The intervention condition plays an important role in what follows, explaining

why the FN associated with a predicate noun cannot agree with the subject,
whereas the FA associated with a predicate adjective can.
I discuss conditions (62c) and (62d) together, because they have some similar
effects and there may be some redundancy between them. They account for data
like the following, from expletive constructions in English:
(67) a. There seem [to be three unicorns in the garden].
b. It seems/*there seem [that three unicorns are in the garden].
c. It seems/*there seem to two maidens [that three unicorns are in the
garden].
(67a) shows that agreement on the Tense node associated with seem can see
into the nonfinite, raising-type clause in order to agree with the plural NP three
unicorns contained in that clause. In contrast, (67b) shows that this Tense cannot
see into the corresponding finite clause, and (67c) shows that it cannot see into
the PP in order to agree with the plural NP two maidens. The contrast between
(67a) and (67b) follows from (62c), a special case of Chomsky’s (2000, 2001)
Phase Impenetrability Condition, which holds that full CPs (and also some vPs)
are “chunks” for interpretation. Once formed, these tensed CPs are shipped off to
the interpretive components immediately, with the result that only a constituent
at the edge of one of these phases remains available for further computation (in
this case, for agreement). Nonfinite clauses of the raising type are not phases,
however. The phase condition could also account for (67c), if one held that
PPs also count as phases. Alternatively, (67c) could be ruled out by the activity
condition in (62d): once the case feature of two maidens is checked by the
preposition to, it becomes inactive with respect to agreement, and the Tense
associated with seem cannot agree with it. The activity condition could also
rule out (67b): the finite Tense of the embedded clause checks the nominative
case of three unicorns, making it ineligible to enter into agreement with the
matrix Tense. Given that both (67b) and (67c) can be ruled out by both (62c)
and (62d) under plausible assumptions, I am not always sure which is crucially
violated in a particular example. This need not concern us greatly, however,
15 See section 3.3.3 for more on agreement in double object constructions.

because for now I am primarily interested in the contrasts in (67) because

they illustrate a characteristic pattern of agreement. In practice, where either
condition will do I appeal primarily to the phase condition, because it is easier
to state and may be more fundamental.16
Chomsky assumes one further condition on the Agree relation that I suppress
here, which is that the agreeing head (the probe) and the agreed-with NP must
match in features. The idea behind this is that the probe has certain predefined
feature slots that need to receive a value from some other phrase in the structure
that has specified values for those same features. It makes sense that (for exam-
ple) a verb that needs a value for its person feature is not going to receive it by
agreeing with a PP or a CP, since those categories do not bear person features.
But one goal of this work is to eliminate, as much as possible, the idea that probe
heads can be arbitrarily specified as having certain feature slots as opposed to
others. Stipulating what unvalued features a given head has on a case-by-case
basis does not capture the systematic differences in how verbs, adjectives, and
nouns behave with respect to agreement. Instead, I am foregrounding the idea
that all Fs are potential agreers and they agree with whatever features they can
find in their environment according to structural principles. If agreeing func-
tional heads are not prespecified as agreeing in particular features, it follows
that there cannot be any matching condition of the kind Chomsky envisions.
This condition should become superfluous – at least if agreement in ϕ-features
is the only relevant kind of agreement in natural languages.
2.3.2 Generalizing agreement theory to adjectives

All of the paradigm-forming examples of agreement in the previous subsec-
tion were instances of agreement on verbs. That is characteristic of the recent
Chomskian literature on these matters, which has focused on this most prolific
of agreeing categories. Consider then how this existing theory of agreement can
be adapted to structures with other lexical categories.
We run into a problem immediately when we consider the fact that predicate
adjectives agree with their subjects in many languages. If my (2003a) approach
to lexical categories is at all correct, then the partial structure of an adjectival
predication is something like (68) (=(58b)).
16 In chapter 5, I show that the activity condition in (62d) is parameterized in a significant way,
holding in some languages but not in others. Languages in which (62d) does not apply allow
double agreement in structures like (67b) (see section 5.9), but they do not allow agreement
into a PP as in (67c) (section 5.6). This suggests to me that both tensed CPs and PPs are phases,
and that the phase condition is absolute and invariable across languages.
(68) PredP
NP Pred´
girls Pred FAP

[F,pl]
Ø FA AP
[F,pl]
A
tall
In some languages, this PredP structure can appear by itself as a matrix clause
(Edo, Arabic, Hebrew); in others it must be selected by a nonthematic auxiliary
verb (e.g., The girls are tall) or a small-clause-selecting verb (I consider the
girls tall, Good nutrition made the girls tall). The crucial thing to notice is that
the subject ‘girls’ is never in the c-command domain of the agreeing head FA .
This subject could move higher still (in a raising construction, like The girls
seem tall), but it cannot move lower, by the ban on downward movement. Thus,
either the structural analysis of adjectival predication needs to be changed, or the
conditions on agreement need to be changed. I propose to change the conditions
on agreement.17 In particular, I suggest that the c-command condition in (62a)
should be revised as follows:
(69) F agrees with XP, XP a maximal projection, only if:
a. F c-commands XP or XP c-commands F (the c-command condition).
(69) permits agreement in (68), since the subject of the predication c-commands
the functional head FA associated with the adjective.18
17 The more traditional view, which preserves the idea that heads always look downward for NPs
to agree with, is to say that the subject of predication is generated inside the maximal projection
of the adjective, as in (i) (see Chomsky 1993:8):
(i) [TP The girlsi are [VP <are> [ FA [AP ti tall ]]]]
In this structure, FA can probe downward, agreeing with the trace/copy of the subject inside AP.
The cost of this view is that it has no explanation for the fact that verbs behave like unaccusative
predicates whereas thematically similar adjectives do not: see the discussion of (43) and (44)
above, and Baker 2003a:ch. 2. More generally, the Baker 2003a system for explaining the
differences among the lexical categories would unravel.
Note that agreement with the subject could appear on Pred in (68) (see section 2.5.1). But that
would not be the right analysis for the kind of agreement that appears on both predicate adjectives
and attributive adjectives, since there is no Pred in structures of attributive modification.
18 A more radical and elegant possibility, pointed out to me by Itziar Laka (personal commu-
nication), would be to do away with the c-command condition on agreement altogether. It is
not clear that any examples are uniquely ruled out by this condition, and Laka observes that
Does Bhatt’s evidence from long distance agreement in Hindi, repeated here
as (70a–b), prove that (69) is a nonstarter? The answer is no, on two counts.
First, Bhatt himself points out that infinitival morphology in Hindi never initiates
an agreement relationship; it only shows agreement if it appears on the path
between the matrix verb and the NP that the matrix verb ultimately agrees with.
As a result, there is no agreement between the infinitival verb and its object in
(70c), even though agreement here would be downward.
(70) a. Vivek-ne kitaab parh-nii chaah-ii. (=(63a))
Vivek-erg book.f.sg read-inf. f.sg want-perf. f.sg
‘Vivek wanted to read the book.’
b. Mona kuttõ-ko dekh-naa/*nii chaah-tii thii.
Mona dog.m.pl-acc see-inf/inf. f.sg want-hab. f.sg be-past. f.sg
‘Mona wanted to see the dogs.’
c. [Imlii khaa-naa] achchh-aa hai. (Bhat 2005:771)
tamarind.f.sg eat-inf. m.sg good-m.sg be.pres.3sS
‘To eat tamarind is good.’
The infinitive in (70b) also does not intervene structurally between the matrix
verb and the NP that it agrees with (the matrix subject), so one would not expect
it to agree any more than the infinitive in (70c) does. Furthermore, embedded
verbs clearly do agree with matrix subjects in restructuring constructions in
other languages, such as the Serbian example given in (71) (Sandra Stjepanovic,
personal communication).
(71) Mi ga [VP zel-imo [VP da posjet-imo]].
we him want-1pS that visit-pres.1pS
‘We want to visit him.’
Of course there could be differences in the structure of complements of want-

type verbs in Hindi versus Serbian that one could relate the difference in
agreement to. But at least (71) shows that there is no obvious up-versus-down
asymmetry in the agreement properties of want-type constructions crosslin-
guistically, of the kind that would support a down-only theory of agreement
prior to in-depth analysis of a wide range of relevant cases.19
sentences such as A group of children walk to school together every day, although considered
substandard, are often produced in many languages, even though children does not c-command
walk. But despite the thinness of the evidence that it is required, I continue to use the c-command
condition, at least for expository purposes.
19 Some minimalists might declare the possibility of upward agreement out of bounds on theo-
retical grounds. Chomsky (2000) considers the c-command condition a “perfection” of the
language faculty because it restricts the space in which a head can find something to agree with,
Having generalized the c-command condition on agreement as in (69), it

behooves me to make corresponding changes to the other conditions on agree-
ment in (62), so that they properly regulate upward-probing agreement as well
as downward-probing agreement. The intervention condition can be generalized
as follows:
(72) F agrees with XP only if:

b. There is no YP such that YP comes between XP and F and YP
has ϕ-features (the intervention condition).
The notion “comes between” can be defined in terms of c-command: A comes

between B and C if and only if either (i) B c-commands A and A c-commands
C, or (ii) C c-commands A and A c-commands B.20 An application of this
condition to upward agreement could be Italian examples like those in (73)
(from Cinque 1990:28), which have a structure roughly like [NP1 verb [NP2
Pred [FA [Adjective]]]].
(73) a. Ne ha resi infelici (*infelice) i

of.it has.3sS rendered unhappy.m.pl (*unhappy.m.sg) the
sostenitori.
supporters.m.pl
‘He has rendered its supporters unhappy.’
b. Ne ritenevo opportuna la riapertura.
of.it consider.1sS appropriate.f.sg the opening.f.sg
‘I consider its opening appropriate.’
reducing computational load. Also, some forms of cyclicity require that the uninterpretable
features of a head be dealt with immediately upon introducing that head into the structure by
Merge (e.g., Chomsky 2000:132). This would imply that agreement must happen before higher
phrases have been included in the structure. I assume that the phase condition is adequate to
narrow the probe’s search space and avoid computational explosion, and that cyclicity requires
only that all features must be satisfied by the time that the phase is complete, not necessarily
immediately upon introducing a feature-bearing category.
20 An alternative formulation could be stated in terms of the m-command relation of Chomsky
1986: X m-commands Y if and only if X does not contain Y and the smallest maximal projection
that contains X also contains Y. Then A comes between F and NP if and only if either (i) F m-
commands A and A asymmetrically m-commands NP, or (ii) NP asymmetrically m-commands
A and A m-commands F. This alternative definition in terms of m-command implements a form
of equidistance, in which the specifier and complement of a single projection count as equally
close to a potentially agreeing head (because they m-command each other). In most cases the
difference does not matter, but this alternative may have some advantages when it comes to
agreement on Pred (section 2.5.1) and agreement on possessive determiners (section 4.1.2).
In such examples, the adjective clearly agrees with NP2 (the subject of the small
clause), which is the closest c-commanding noun phrase, and not with NP1 (the
subject of the matrix clause), which is farther away.21
Finally, the phase condition works as stated, given the understanding that
agreement is blocked when the functional head is in a phase that does not
contain the NP it might agree with, as well as when the NP is in a phase that
does not contain the functional head (see (67b)). The phase condition thus also
rules out upward agreement in configurations like the following:
(74) *Three women said [that there are likely [that it will rain]].
Putting the pieces together, the revised syntactic condition on agreement is
given in (75).
b. There is no YP such that YP comes between XP and F and YP has
ϕ-features (the intervention condition).
c. F and XP are contained in all the same phases (the phase condition).
(the activity condition).
I adopt this as my first significant modification to the theory of agreement.
2.4 Explaining the basic categorical asymmetries in agreement

Now that we have a theory that can account for the agreement between a pred-
icate adjective and its subject, we are ready to explain the crosscategorical
differences in agreement: the fact that predicate adjectives differ from verbs in
not showing first or second person agreement, and the fact that nouns differ
from adjectives in not agreeing at all. I begin with the second of these contrasts,
since it follows from the principles already given.
2.4.1 Explaining the noun–adjective contrast

In fact, the agreement differences between nouns and adjectives follow imme-
diately from the intervention condition in (75b), given the structures in (76)
(=(58)).
21 In similar sentences in Icelandic, an adjective can never agree with an NP that is more remote
than the subject of its predication, even if the subject of its predication has quirky case and
hence does not trigger agreement on the adjective itself. Thus, in the Icelandic version of ‘The
women(NOM) consider Mary(DAT) to be cold,’ ‘cold’ must be default masculine singular, not
feminine plural in agreement with ‘the women’ (Sigurð sson, personal communication). This is
an even closer analog to the “defective intervention” example in (64).
Explaining the basic categorical asymmetries in agreement 49
(76) a. PredP b. PredP
NP Pred´ NP Pred´
they Pred FAP they Pred FNP

[F,pl] [F,pl]
Ø FA AP Ø FN NP[M,sg]
[F,pl] *[F,pl]
A N
tall group[M,sg]
The one significant difference between these structures is that the complement
of the potentially agreeing F head, namely ‘group’, has a referential index and
therefore has ϕ-features of its own in (76b). Since this phrase c-commands
FN and is c-commanded by the subject of predication ‘they’, it counts as an
intervener, blocking an agreement relationship between FN and ‘they’. In con-
trast, the complement of FA in (76a) – the AP headed by ‘tall’ – does not have
ϕ-features. It thus does not count as an intervener, and it does not block agree-
ment between ‘they’ and FA . The asymmetry between A and N with respect to
agreement thus reduces to the independently motivated intervention condition
on agreement in (75b).
Of course, nothing prevents FN in (76b) from agreeing with its NP comple-
ment, thereby manifesting the number and gender features that are intrinsically
associated with that NP. This I claim is the source of the inflectional affixes
that appear on nouns in many languages. These are often cognate with those
that appear on adjectives, but are not used in the same way for agreement. In
intuitive terms, predicate nouns do not agree with their subjects because they
have to agree with themselves. There can be agreement in a predicate nominal
construction, in fact, but it is internal to what is usually considered to be the
noun itself.
Now imagine what would happen if a second FN were generated immedi-
ately above the first FN in (76b), giving a structure like [‘they’ Pred [FN [FN
[NP ‘group’]]]]. Could this higher FN agree with the subject? If so, one might
expect there to be languages like Swahili in which predicate nominals bear
two inflectional prefixes, an inner one that shows the inherent gender and num-
ber of the predicate nominal, and an outer one that agrees with the subject of
predication. But we did not observe that in any of the languages surveyed in
section 2.1.1. The structure itself should not be out of the question; after all,
there can be two agreement-bearing functional heads above VP, one the locus
of subject agreement and one the locus of object agreement (see (82b) below).
If it is not possible to have two such heads above an NP, that is something
we want to explain, not stipulate. But recall from (52) that the functional head
FN is, by definition, a functional head that shares the same essential categor-
ical properties as its NP complement – an idea borrowed from Grimshaw’s
(1991) notion of an extended projection. Now within my approach to lexical
categories, the essential categorical property of an NP is having a referential
index, and with that referential index come intrinsic ϕ-features. The lower FN P
in the structure under consideration must thus be nominal in the same sense of
having a referential index and ϕ-features; presumably it inherits the index and
ϕ-features of its NP complement. Consider then the effect of this on the higher
FN . Just like the lower FN , its complement is the closest phrase to it that has
ϕ-features; therefore its FN P complement prevents the higher FN from agreeing
with a more remote NP, such as the subject of predication. I conclude that even
iterating functional heads do not make it possible for a predicate nominal to
agree with its subject. Such iteration would only generate multiple redundant
agreements with the predicate nominal, and this is presumably ruled out for
economy reasons. There is an advance here over the baseline view, mentioned
in section 1.2, which says that the ϕ-feature slots of a noun phrase are already
valued and hence not available for agreement. This standard view does not
explain why a noun could not have a second set of ϕ-feature slots which are
unvalued and hence open for agreement. In contrast, the theory just sketched
does rule out the analog of this possibility without additional machinery.
Finally, recall that the agreement difference between nouns and adjectives is
found in attributive contexts as well as predicative ones. The explanation given
for (76) generalizes easily to this case. Following Baker 2003a, I assume that
attributive modifiers are simply adjoined to the phrase that they modify, giving
the structures in (77), enriched by the presence of an agreement-bearing F head
that dominates the attributive modifier.
(77) a. wə rt-e zgı̄r-e b. s-sakak-ı̄n ə l-fə dd-a (Arabic)

inheritance-f.sg small-f.sg the-knife-pl the-silver-f.sg
‘small inheritance’ ‘the silver knives’
NP NP
NP FAP NP FNP
inheritance FA AP knives FN NP[F,sg]

[F,sg] [F,sg] [pl] *[pl]
A N
small silver[F,sg]
Here again, the presence of ϕ-features on the complement of FN makes the

complement a closer goal for agreement than the NP that FN P is adjoined
to. Therefore, if an attributive N bears inflectional morphology at all, it man-
ifests its own features, not those of the noun that it modifies. In contrast,
there are no ϕ-features on the complement of FA , so there is no intervener
to prevent FA from agreeing with the NP that it adjoins to, which c-commands
it.22 The reasoning is the same as in (76), except that no Pred head is
present.
2.4.2 Explaining the adjective-verb contrast

Next let us turn to the difference between adjectives and verbs when it comes
to agreement in person. This requires a second addition to the basic theory of
agreement, one that draws a distinction among the various ϕ-features. The
empirical difference is that the functional heads associated with verbs can
show agreement in first and second person as well as in number and gen-
der, whereas FA is never first or second person. The most basic theoretical
difference between verbs and adjectives is that verbs license a specifier and
adjectives do not. This means that an FV typically c-commands the agreed-with
subject inside VP, whereas an FA does not c-command the subject, as shown
in (78).
(78) a. PredP b.
FVP (=TP)
NP Pred´ FV VP
[1,F,pl]
we Pred FAP NP V´
[1,Fpl]
Ø FA AP we V (PP)
[F,pl] [1,F,pl]
[*1] A fall
tall
22 Technicalities involving the c-command relationship in adjoined structures, introduced in

Chomsky 1986 and mentioned in note 12, are relevant here. In particular, the NP headed by
‘inheritance’ in (77a) does not dominate the FA P that is adjoined to it, since only one segment
of the NP contains FA P. Therefore, the NP does c-command FA , and so agreement between FA
and the NP it is adjoined to is consistent with the c-command condition. (There is a tension,
however, between this definition and my explanation of the lack of first and second person
agreement in attributive constructions given in chapter 4. In the end, the significance of the
segment-category distinction might be different for Agree than for binding; see section 4.5 for
discussion.)
Given this, one might reasonably entertain a stipulation that downward agree-
ment is required for agreement in first or second person, whereas upward
agreement is possible for other features. But this would not turn out to be
right, for reasons I discuss at length in the next chapter: there are structures in
which FA also agrees downward, but it still cannot bear first or second person
features.
Instead, I develop a version of the rather traditional idea that there is some-
thing special about the specifier–head relationship when it comes to agreement.
Inspired by Chomsky 1986, work around the early 1990s often assumed that
agreement took place only if there was a specifier–head relationship between
the agreed-with NP and the agreeing head (see Kayne 1989, Chomsky 1991,
Kinyalolo 1991, Koopman and Sportiche 1991; see also Koopman 2006 for a
recent defense of this idea). The pendulum has since swung the other way,
with the specifier–head relationship playing no role at all in agreement in
Chomsky 2000. I propose to take an intermediate position, claiming that a
particularly rich form of agreement becomes possible if and only if there is
a direct relationship of Merge between the controller of agreement and its
target:
(79) The Structural Condition on Person Agreement (SCOPA)

A functional category F can bear the features +1 or +2 if and only if a
projection of F merges with an NP that has that feature, and F is taken as the
label for the resulting phrase.
Notice that, in accordance with minimalist views about phrase structure, (79)
does not actually grant any special status to the specifier configuration: F can
agree with NP in +1 (first person) or +2 (second person) if NP is the speci-
fier of FP, but it can also agree in this richer way if NP is the complement of
F. In addition to being consistent with Chomsky’s (1995) Bare Phrase Struc-
ture, this formulation pays off when we consider agreement on adpositions and
determiners in section 4.1.
The SCOPA makes possible an account of the difference between adjectival
agreement and verbal agreement as follows. The verbal head T, like its VP
complement, is verbal in the sense of taking a specifier (another application
of (52)). Thus T generally has an “EPP” feature that causes one NP inside its
complement (typically the highest one, by the Minimal Link Condition) to move
to become its specifier, as shown in (80). Once that movement takes place, T
can agree with NP in all features, including +1 and +2:
(80) FVP (=TP)
NP FV´
we FV VP
[1Fpl][1Fpl]
NP V´
<we> V (PP)
fall
In contrast, FA cannot trigger movement of the subject of an adjectival predica-

tion to its specifier position. Probably it cannot have an EPP feature (by (52)),
but even if it did, this would constitute an instance of downward movement; the
NP would originate in a position that c-commands its destination, in violation
of the fundamental laws of movement:
(81) PredP
NP Pred´
we Pred FAP
[1Fpl]
Ø FA AP
[F,pl]
[*1] A
tall
FA can thus agree with the subject of predication, but only at a distance. Hence,
it can only agree in features other than +1 and +2, by the SCOPA. This then
gives us a way of thinking about the difference between adjectives and verbs
when it comes to agreement.
Transitive verbs in some languages can agree with an object as well as a
subject, and such agreement can be for first and second person features as well
as for number and gender. Let me now make explicit how this fits in. The
first step is to say that there can be two functional projections above VP, the
heads of which can potentially bear agreement. For concreteness, I follow much
current Chomskian literature in assuming that these heads are finite Tense (as
assumed above), and v – the light verb which θ -marks the subject and (in many
languages) licenses accusative case on the object. Tense is the usual locus of
subject agreement and v the usual locus of object agreement. The structure of a
transitive clause with double agreement on the verb, such as (82a) from Swahili,
is thus roughly as shown in (82b).
(82) a. Juma a-li-ni-ambia kwamba . . . (Vitale 1981:62)

Juma 1S-past-1sO-tell that
‘Juma told me that . . .’
b. TP
NP T´
Juma T vP
[3,sg, [3,sg,
an] an] NP v’
<Juma> NP v’
me v VP
[1,sg] [1,sg]
NP V’
<me> V CP
tell that . . .
In this implementation, v has the perhaps rather special property of being able
to license two specifiers: the thematic subject, which is base-generated there,
and an object which can arrive there by movement.23 Given that movement of
an object to Spec, vP is possible, a projection of v merges with this object,
so v can agree with the object in first or second person as well as in other
features, in keeping with the SCOPA. Note that the subject agreement in (82a)
23 Other assumptions are possible. For example, one could distinguish two separate heads, one
that creates the base position for the subject, and another immediately above or below it that
provides the landing site for the accusative object. I leave the details open, adopting (82) for
purposes of presentation, because of its relative simplicity and familiarity.
I also note that the arrangement in (82) is not necessarily universal. It could be that in other
languages the agreement-bearing heads are C, or Aspect, or Mood, instead of or in addition
to T and v. In general, it is not the label of the category that determines what sort of agree-
ment that category will manifest, but rather the geometry of the configuration that the category
appears in. See also section 3.2.2.1 for discussion of how T and v are involved in languages with
ergative agreement systems. (For evidence that T agrees with the subject and v agrees with the
object in the normal way in a language with an active agreement system – a language in which
some intransitive verbs bear “subject” agreement morphemes and others bear “object” agree-
ment morphemes – see Baker 1996:ch. 5 on Mohawk. Other languages with active agreement
(Choctaw, Guaranı́) seem similar to Mohawk in this respect, but I have not studied the matter
carefully.)
is left-adjacent to the tense morpheme li, whereas the object agreement is left-
adjacent to the transitive verb root ambia ‘tell’. This morphological structure
fits naturally with the syntactic structure in (82b), where the two agreements
pertain to two distinct functional heads.
The SCOPA raises many questions. For example, it is natural to ask what
happens to agreement on FV if something other than the agreed-with NP satisfies
the EPP feature of FV . The prediction is that in any such environment the verb
should become more adjective-like in its agreement properties, being able to
agree with an NP in gender and number but not in first or second person. Chapter
3 explores this prediction in detail, claiming that it is confirmed. However, the
SCOPA definitely does not look like a plausible candidate for a basic principle
of Universal Grammar. Rather, it is the kind of highly particular statement that
one wants to derive from more general principles. Chapter 4 explores how this
can be done in terms of the fundamental properties that distinguish first and
second person pronouns from other nominal categories.
Why does the SCOPA contain the additional qualification that F must be
taken as the label of the phrase that results from merging it with NP? This is
added to capture the observation made in section 2.1.3 that even attributive
adjectival modifiers do not show first or second person agreement when they
adjoin to a first or second person pronoun. One example of this is (83) from
Tariana, repeated from (29a) above.
(83) Pi-na inasu-ite-nuku nuhua-wya-ne ma:-kasu nu-ñha.

2s-obj lazy-cl:an-top.obj 1s-ext-foc.subj neg-let 1sS-eat
‘I am the only one (who) will not let you the lazy one eat.’
In examples like these, one cannot complain that the FA is too far from the
first or second person element, for the two do undergo Merge. The difference is
that in these attributive constructions the first or second person pronoun counts
as the head of the newly formed phrase, as shown in (84). This is unlike the
structure in (80), in which the agreement-bearing FV head provides the label
for the newly formed phrase.
(84) DP
DP FAP
you FA AP
[2,sg,an] [sg,an]
[*2] A
lazy
This minor-looking difference in structure apparently makes all the difference

when it comes to agreement, first and second person morphology being licensed
in (80) but not in (84). This shows that it matters which of the merged categories
projects in cases of first and second person agreement. It will be an important
virtue of the deeper theory that underlies the SCOPA that this odd qualification
makes more sense; see section 4.5 for discussion.
2.5 Issues arising

Before closing this chapter, I consider two kinds of languages that seem to
depart from the norm that I have described in this chapter when it comes to the
agreement properties of lexical heads. The first is a set of languages in which
all lexical heads – nouns and adjectives as well as verbs – seem to show person
agreement with the subject of the clause. The second kind is a set of languages
in which predicate adjectives agree with their subject in number and gender,
but attributive adjectives do not agree with the noun they modify. Neither turns
out to be problematic for my proposals once one looks into the details.
2.5.1 Person agreement on adjectives and nouns

My theory is designed to explain why verbs often agree in first and second
person features, whereas adjectives and nouns do not. As mentioned at the
end of section 2.1.3, the typological literature recognizes this as being a strong
tendency, but claims that it is not a universal feature of language. In particular, a
number of languages have been reported as allowing the same person agreement
affixes to attach to all three lexical categories. For example, in (85) from Turkish,
the same first person singular suffix attaches to the predicate nominal ‘teacher’,
the adjective ‘dirty’, and the verb ‘read’.
(85) a. Ben oku-r-um. (Kornfilt 1997:78–83)

I read-aor-1sS
‘I read.’
b. Ben temiz-im.
I dirty-1sS
‘I am dirty.’
c. Ben ögretmen-im.
I teacher-1sS
‘I am a teacher.’
Other language that fit this description include the Salish and Wakashan lan-
guages, Lango, Abaza, Nahautl, Guaranı́, and the Mayan languages.
Issues arising 57
The account of these languages that I offer is simply that the person agreement
in these languages is not on FA/N , the functional category that can be generated
immediately above all uses of A and N. Rather it is agreement on some more
verbal functional category higher in the structure of the clause, a functional
category that is generated above the subject and can trigger movement of the
subject to its specifier.
For Turkish, this interpretation is supported by the fact that when examples
like those in (85) appear in past tense rather than present tense, an overt past
tense morpheme di appears attached to the predicate. It is significant that the
person agreement marker appears outside of this tense marker, not inside of it,
attached directly to the nominal or adjectival root:
(86) a. Ben kitab-i oku-du-m. (Kornfilt 1997:78–83)

I book-acc read-past-1sS
‘I read.’
b. Ben temiz-di-m.
I dirty-past-1sS
‘I was dirty.’
c. Ben öğretmen-di-m.
I teacher-past-1sS
‘I was a teacher.’
This suggests that it is the Tense node that agrees with the subject in person in
Turkish, not the noun or adjective (i.e., not FN or FA ). This unproblematic fact
is merely obscured somewhat by the fact that Tense and the noun or adjective
sometimes form a single word on the surface in Turkish as a result of head
movement, PF merger, or cliticization. This makes it look like the noun or
adjective agrees with the subject – especially in the present tense, when the
tense marker is phonologically null.
That person agreement in Turkish is really a property of Tense, not FN or FA ,
is even clearer in the future tense. This tense does not merge with a nonverbal
root; rather an auxiliary root ol ‘be’ needs to be included to support it when
the predicate of the clause is a noun or adjective, as shown in (87). In this
case, when tense and the nonverbal predicate show up in different words, it is
perfectly clear that the person agreement belongs to the word that contains the
tense and not to the word that contains the nominal or adjectival root:
(87) a. gel-ecek-sin (Wetzer 1996)

come-fut-2sS
‘you will come’
b. Ben temiz ol-acağ-im. (Kornfilt 1997:78–83)

I dirty be-fut-1sS
‘I will be dirty.’
c. Ben öğretmen ol-acag-im.
I teacher be-fut-1sS
‘I will be a teacher.’
There is no logical necessity that this be so. It is easily imaginable that the
person agreement could have appeared on the noun or adjective instead of or in
addition to appearing on the auxiliary. But it does not, supporting my claim that
adjectives and nouns in themselves never show person agreement. The structure
for examples like (86b) and (87b) is shown in (88), where Tense affixes to the
adjacent adjective when it is present and auxiliary insertion takes place when
Tense is future.24
(88) TP
DP T´
(I) PredP T
[1,sg]
DP Pred´ Ø -PRES
dI-PAST
<I> FAP Pred EcEk-FUT
[1,sg]
AP (FA)
[sg, *1]
rich
Essentially the same analysis can be applied to Lango, Wakashan, Salish, Abaza,
Tzotzil, and similar languages. For example, Noonan (1992:144–6) shows that
person agreement prefixes attach directly to nouns, adjectives, and verbs in the
unmarked present tense in Lango, but crucially they do not attach to nouns
and adjectives in the past tense; then they appear only on a finite auxiliary.
These languages, then, do not count as real counterexamples to the claim that
adjectives and nouns themselves do not agree with their subjects in person but
only in number and gender.
One language that might call for a slightly different treatment is Classical
Nahuatl, as described by Launey (1981). First and second person agreement
prefixes can attach directly to nouns in this language, as shown in (89). (Launey
24 Indeed, there is evidence that an auxiliary root i is inserted in the present and past tenses as
well, prior to T cliticizing to the previous word. This form of the auxiliary, however, is elided
after consonant-final adjectives or nouns, for phonological reasons; see Kornfilt 1997:77–83
for discussion.
Issues arising 59
does not clearly distinguish adjectives from nouns in Nahuatl, so presumably

adjectival predicates like cualli ‘good’ behave similarly.)
(89) a. Ni-mexı̀ca-tl. (Launey 1981:26)
1sS-mexican-nsf
‘I am a Mexican.’
b. Am-mexı̀ca-’.
2pS-mexican-nsf.pl
‘You (pl.) are Mexicans.’
The presence of the determiner-like noun suffixes tl and ’ in these examples

shows that they are not verbs derived from nominal roots, for which agreement
in person would be expected; tl in particular shows up on most singular nouns
but not on verbs in Nahuatl. Nor is there any sign of tense marking in these
examples or others that involve predicate nominals. Tense suffixes never show
up on nouns in Nahuatl the way they do in Turkish and Abaza, nor do overt
auxiliaries appear in nonpresent tenses to bear the tense and agreement. This
seems like a clearer case of person agreement attaching directly to a noun, then.
Two interpretations of this data are possible within the theory I am developing,
and I do not have clear facts to tell which is true. One is that the agreement is
really housed syntactically on a null Tense node in Nahuatl, which then cliticizes
onto the adjacent noun word, just as in Turkish. On this view, the only difference
between Nahuatl and Turkish would be that Nahuatl does not have auxiliary
verbs that can be used to support other types of Tense. The second possibility
is that the examples in (89) are bare PredPs, not embedded under a TP at all,
as seems to be possible in Edo, Hebrew, and other languages. In that case, I
would claim that the first and second person prefixes are actually realizations of
agreement on the Pred head itself, but not on FN . Since Pred does merge directly
with the first or second person subject of the predication (see (88)), having Pred
bear person agreement features is a theoretical possibility, consistent with the
SCOPA.25 Which of these analyses is correct should be settled by doing a
detailed study of the temporal properties of examples like (89) to determine if
a Tense node is present semantically or not (see Benmamoun 2000 for such a
study in Arabic).
Even though we cannot inspect the location of agreement relative to Tense
directly in Nahuatl, there is a bit of confirming evidence that the person agree-
ment in question is on a higher, clause-like functional head and not on the
lower, purely nominal head that I call FN . This involves the location of person
25 For predicate nominals, taking this option might require adopting the alternative formulation
of intervention mentioned in note 20; on that definition, the ϕ-features on the FN P complement
of Pred do not prevent Pred from agreeing with its specifier by the intervention condition.
agreement with the subject on possessed nouns that are used predicatively. In
fact, the person agreement appears outside of the possessor agreement prefix
in Nahuatl, as shown in (90).
(90) a. Ti-no-cihuā-uh (Launey 1981:91)

2sS-1sP-woman-poss
‘You are my wife.’
b. An-to-pil-huān
2pS-1pP-child-poss
‘You (pl) are our children.’
Assuming that the morphological structure of the predicate typically reflects the
syntactic structure and derivation of the clause (the Mirror Principle of Baker
1985), this morpheme order indicates that the functional head that agrees with
the subject of predication is higher in the phrase structure than the functional
head that agrees with the possessor of the noun phrase. Now the functional head
that agrees with the possessor is presumably D (Determiner); see section 4.1.2
for examples, references, and discussion. FN is (by definition) the functional
head immediately above the NP projection. FN is thus lower in the phrasal
architecture than D – which is just what we need for Spanish noun phrases
like l-as chic-as (the-F.PL girls-F.PL) (see the discussion of (59)). (90) thus
suggests that FN does not agree in person with the subject even in Nahuatl. In
contrast, Tense and Pred are both above the DP projection in a predicational
sentence. Attributing the agreement in person to one of these heads, rather than
to FN , thus gives a better account of the morphological structure of predicate
nominals in Nahuatl, in addition to being consistent with the SCOPA.
I conclude that, although there are clear counterexamples to a superficial
statement such as “nouns and adjectives are never inflected for person,” there
seem to be no true counterexamples to my actual theoretical claim, which is
that the functional heads most intimately linked to adjectives and nouns cannot
be the loci for agreement in first or second person. Higher heads in a clause
containing a nominal or adjectival predicate clearly can, including Tense and
perhaps Pred. But that is not surprising, since those categories are intrinsically
verbal in the sense of licensing a specifier.
2.5.2 Agreement differences among adjectives

In presenting my theory above, I emphasized that predicate adjectives and
attributive adjectives typically have the same agreement behavior, even though
they are found in somewhat different syntactic environments. On the one hand,
both types of adjectives generally show number and gender agreement with
a nearby noun, whereas nouns in similar syntactic positions do not. On the
Issues arising 61
other hand, neither type of adjective can bear first or second person agreement,
whereas verbs can. I took it to be an important sign of the adequacy of my
theory that it was able to give a unified analysis of agreement on both predicate
adjectives and attributive adjectives. And indeed adjectives in both contexts do
show the same agreement in all of the test languages that I surveyed.
This is not always the case, however. There are also a reasonable number of
languages in which predicative adjectives agree with their subjects but attribu-
tive adjectives do not. Kannada (Dravidian) provides one example: (91) shows
that there is gender and number agreement on predicative adjectives; (92) shows
that the same adjective is invariant when it appears in attributive position (Srid-
har 1990:249–50).
(91) a. Avanu tuNTa(nu).

he naughty.m.sg
‘He is naughty.’
b. AvaLu tuNTa-Lu
she naughty-f.sg
‘She is naughty.’
c. Avaru tuNTa-ru
they naughty-pl
‘They are naughty.’
(92) a. a: tuNTa huDug-a (Sridhar 1990)
that naughty boy-m
‘that naughty boy’
b. a: tuNTa huDug-i
that naughty girl-f
‘that naughty girl
c. a: tuNTa huDug-a-ru
that naughty boy-m.pl
‘those naughty boys’
In fact, an analysis of this is readily available. I claim that adjectives in them-

selves never agree in Kannada; in more technical terms, FA is either absent or
not a probe. The facts in (92) follow immediately. What then is the source of
the agreement in (91)? I claim that these are not really predicate adjective con-
structions, but predicate nominal constructions in which the adjective modifies
a pronominal. Thus, a more literal gloss for (91c) would be ‘They are naughty
ones.’ The “agreement” in these examples is not a realization of FA , but rather
a realization of some part of a nominal structure that contains the adjective.
This analysis actually is fairly transparent, and similar to what Sridhar himself
says. He says that predicative adjectives in Kannada “behave syntactically like
nouns” and “appear in their nominal form.” Note that the “agreement” on
predicate adjectives is basically a pronominal form. The adjectives in (91) do
not end in a, i, and aru, the normal masculine, feminine, and plural endings in
Kannada (compare the nouns in (92)). Rather, they end in n(u), Lu, and aru –
the same endings that the demonstrative pronouns take in (91). So these are not
technically predicative adjectives, they are attributive adjectives that modify a
pronominal element. The adjectival part of the predicate does not agree, but the
pronoun part does agree with the subject, as pronouns normally do with their
antecedents. That sort of agreement is quite a different matter from the Agree
process under study here (see section 4.2 for a comparison).
To complete the analysis, we need to say why predicative adjectives must
be incorporated into a larger DP in Kannada but not in many other languages.
Put another way, we must say why this language allows predicate nominals
(which may contain an adjective) but not simple predicate adjectives. Baker
2003a:164–5, 210 provides an answer. There I showed that predicate nominals
and predicate adjectives require slightly different copulas: the copula associated
with a predicate nominal must θ -mark the predicate NP, whereas the copula
associated with a predicate adjective must not θ -mark its complement. Most
languages have both kinds of copula, but some do not: some have only the
θ -marking copula, and hence have only predicate nominal constructions (my
principal example was Vata). Kannada is another language of this less common
sort: it has only a copula that selects NP, not one that selects AP. That selectional
property is what prevents it from having a simple adjectival predicate, with no
agreement.
The Basque language illustrates a second reason why a language might have
agreement on predicative adjectives but not on attributive adjectives. (93) shows
in a single sentence that an attributive adjective cannot agree with a plural
noun phrase, whereas agreement in number is strongly preferred on a predicate
adjective (Maia Duguine, personal communication; cf. Saltarelli 1988:248).
(93) Katu lodi(*ak) hori-ek oso gaizto-ak dira.

cat fat-abs.pl those-abs.pl very bad-abs.pl 3pA.be
‘Those fat cats are very mischievous.’
The contrast in (94) suggests that the analysis proposed for Kannada is not
appropriate for Basque. (94a) shows that Basque is like English in that adjectives
cannot take PP complements when used in attributive position. However, (94b)
shows that a predicative adjective can have PP complement and still show
number agreement with the subject of the clause.
Issues arising 63
(94) a. liburu (*nire klase-rako) interesgarri hori-ek

book my class-for interesting those-abs.pl
‘those interesting (*to my class) books’
b. Liburu hori-ek nire klase-rako interesgarri-ak dira.
book those-abs.pl my class-for interesting-abs.pl 3pA.be
‘Those books are interesting to my class.’
If the plural suffix ak in (94b) were the realization of some nominal head in
a predicate nominal that contained ‘interesting’ as an attributive adjective, we
would expect the phrase ‘for my class’ to be ruled out, just as it is in (94a).
Basque thus seems to be a purer case of predicate adjectives agreeing and
attributive adjectives not.
The key to understanding these Basque facts is, I claim, observing that the
head nouns of the subjects in (93) and (94b) are themselves not marked for
number; there is no suffix like ak on these expressions either. This is a general
property of Basque: number (and case) morphology is spelled out exactly once
in the nominal, on the very last word of the phrase (the demonstrative in (93)
and (94b); see Saltarelli 1988:75–82 for other possibilities). This suggests that
the Basque noun does not in fact bear ϕ-features; rather intrinsic ϕ-features
(at least number) appear only on D, the highest head in the Basque nominal
structure. On this assumption, the structure for (93) would be (95).
(95) PredP
DP[pl] Pred´
NP D[pl] FAP Pred
NP[-] FAP those AP FA[pl] be
cat AP FA[-] bad

agree OK
fat * agree, no c-command
*agree, no ϕ -features
The attributive adjective could in principle agree with the NP that it adjoins
to, but agreement fails because that NP has no features to agree with. The
entire subject DP does have ϕ-features, but the FA associated with the attribu-
tive adjective cannot agree with this DP, because DP dominates FA and there-
fore does not c-command FA . (Recall also that the goal of an agreement rela-
tionship must be a maximal projection, so FA cannot simply agree with D
itself.) The predicate FA P, in contrast, is not contained in the plural DP. FA
is c-commanded by the plural DP, and can therefore agree with it in plural-
ity. For Basque, then, the agreement difference between predicate adjectives
and attributive adjectives follows from an independently observable property
of where ϕ-features are generated in the complex nominal. This is a para-
metric difference between Basque and (say) Spanish, where ϕ-features are
generated on the N itself and therefore can spread to other phrases inside the
DP by agreement with NP. A perusal of the 108 languages surveyed in chap-
ter 5 suggests that Basque is not alone in this respect; there are several other
languages in which ϕ-features are marked only once in a DP, on a peripheral
functional category rather than on the noun head. Possible languages of this
type include Khoekhoe (Africa), Alamblak (New Guinea), and various lan-
guages of Mesoamerica and South America (Otomi, Jacaltec, Sanuma, Yagua,
Canela-Krahô).26
2.6 Conclusion
The first part of this chapter documented evidence from a variety of languages
that attests to consistent differences in how the different lexical categories par-
ticipate in agreement: verbs can agree in person, number, and gender; adjectives
agree in number and gender only; nouns do not agree with anything outside of
themselves. The second part showed how these asymmetries can be explained
by combining my (2003a) theory of lexical categories with Chomsky’s (2000,
2001) theory of agreement. Once we say that functional heads can probe upward
as well as downward for something to agree with, the fact that adjectives agree
with their subjects but nouns do not follows from the intervention condition on
agreement. The fact that verbs agree in person but adjectives do not follows
from the fact that only verbal projections license specifiers, together with the
stipulation that a head can agree with an NP in first or second person features
only if it has merged directly with that NP (the SCOPA). Two novel assumptions
about agreement were thus needed to make the account work. It is the task of
the next chapter to provide independent support for those assumptions, proving
that the theory has a desirable degree of generality.
26 Dutch and German are said to have the opposite asymmetry, in which attributive adjectives
agree but predicative adjectives do not. Such cases are very rare; I do not know of any others.
Moreover, the “agreement” in question is an extremely impoverished kind: it does not vary
with the number and gender of the noun. (In Dutch, the “agreement” is an invariant schwa; in
German, it shows case but not ϕ-features.) This leads me to conjecture that these morphemes
are not really agreements, but rather “linker” morphemes of the kind that come between A and
N in Tagalog and other languages. I leave full consideration of this conjecture to linguists who
know more about Germanic languages than I do.
3 The unity of verbal and
adjectival agreement
In chapter 2, I developed the theory of agreement so as to explain a robust

crosslinguistic pattern: verbs agree with their subjects and objects in person,
number, and gender, adjectives agree in number and gender but not person, and
nouns do not agree at all. My goal was not to stipulate these differences directly.
Rather, I derived them from a general, structurally based theory of agreement
based on Chomsky 2000, 2001, together with an independently motivated theory
of the syntactic structures that nouns, verbs, and adjectives appear in. The same
theory of agreement thus applies to all the lexical categories, with the differences
falling out from differences in the configurations that the categories appear in.
The central principles of this approach are repeated in (1) and (2), with the
major innovations in italics.
b. There is no YP such that YP comes between XP and F and YP has
ϕ-features (the intervention condition).
c. F and XP are contained in all the same phases (the phase condition).
(the activity condition).
(2) The Structural Condition on Person Agreement (SCOPA)
A functional category F can bear the features +1 or +2 if and only if a
projection of F merges with a phrase that has that feature, and F is taken as
the label of the resulting phrase.
One can, however, still wonder whether having a unified theory of these
matters is truly an accomplishment, bringing us closer to the truth of the mat-
ter. There is a long tradition of treating agreement on verbs and agreement
on adjectives as two quite different phenomena. Indeed, the two are some-
times given different names: concord for the phenomenon of adjectives agree-
ing with the nouns they modify, as opposed to agreement proper for the relation
verbs have with their subjects and objects. For example, Chomsky (2001:34n.5)
65
66 The unity of verbal and adjectival agreement
writes “There is presumably a similar but distinct agreement relation, concord,

involving Merge alone.” And there are some good reasons for this traditional
distinction. First, the features manifested on the agreeing head are different in
the two cases: person features show up on verbs but not on adjectives. Sec-
ond, the typical syntactic configurations that trigger agreement are different in
the two cases: verbs search downward for something to agree with (Chomsky
2000, 2001), but adjectives search upward (see section 2.3.2). Third, adjectives
and verbs agreeing with the same NP occasionally show different values for
number and gender (see Wechsler and Zlatić 2003 on Croatian/Serbian). Given
these apparent differences, one can question whether it is really a good idea to
account for agreement and concord under the same formal theory. The fact that
it can be done does not necessarily imply that it is right to do so. It could be that
the erstwhile unification in (1) and (2) is a combination of formal tricks and
spurious generalizations, rather than the uncovering of a fundamental unity in
grammar.
This chapter argues in favor of the unified theory of agreement by looking at
some less-standard instances of agreement on adjectives and verbs. My goal is
to show that in some of these cases agreement on adjectives is more verb-like,
whereas in others agreement on verbs is more adjective-like. These examples
are intermediate between the most canonical instances of adjective agreement
and the most canonical instances of verb agreement, and they support the claim
that there is no intrinsic difference in how these categories agree. I consider
three kinds of noncanonical agreement configurations in particular. First, I look
at instances of adjectives searching downward for something to agree with,
rather than upward (section 3.1). This happens with a small class of ergative
and raising adjectives, which do not have a thematic subject. In these instances
it is clear that downward adjectival agreement is sensitive to the same factors as
downward verbal agreement is. Second, I look for instances of verbs searching
upward for something to agree with, in the manner of adjectives (section 3.2).
Such configurations are predictably rare, but I find two possible examples: verbs
that agree with a moved wh-phrase in Spec, CP, and agreement in ergative
languages. Third, I discuss instances of verbs agreeing with something that
is not in the specifier of their associated functional category at any level of
analysis (section 3.3). In these contexts, the verb is not able to agree with the
NP in person features; it shows agreement for number and gender only, in a
manner reminiscent of adjectives.
The last kind of example that could connect the two domains would be
special environments in which an adjective can show person agreement with a
Downward agreement on adjectives 67
nominal, the way verbs often do. This situation, however, can never arise (apart
from the kinds of examples discussed in section 2.5.1), for very simple reasons.
Despite this gap in the pattern, the big picture is clearly that agreement on
verbs and agreement on adjectives are not distinct phenomena. On the contrary,
there is ample evidence that the two should fall under the same general theory
of agreement. The fact that my unified theory generalizes from the simple
structures considered in chapter 2 to the less obvious structures surveyed in this
chapter provides strong support in its favor.
3.1 Downward agreement on adjectives

Perhaps my most startling theoretical innovation is the claim that a potentially
agreeing head can either probe downward or upward to find an NP with ϕ-
features to agree with ((1a)). The idea that agreement can be upward is necessary
to handle the agreement between a predicate adjective and its subject, since the
subject is generated in Spec, PredP and is never inside the c-command domain of
the adjective or its associated FA head. But (1a) does not specifically tie upward
probing to adjectival heads, nor does it limit downward probing to verbal heads.
Therefore, adjectival heads should also in principle be able to probe downward
for something to agree with. When they do, the agreement on adjectives should
be more obviously parallel to the agreement found on verbs. We will see that
this is true.
3.1.1 Adjectives with NP complements

One class of adjectives that is relevant to testing this theory is the so-called
ergative adjectives discovered by Cinque (1990). Cinque argues that Italian has
two distinct classes of adjectives, one that selects a subject argument and the
other that selects an object. This distinction is parallel to the much more familiar
distinction between unergative and unaccusative verbs, and it is revealed by
some of the same syntactic tests. For example, the sole argument of an unergative
(agentive) verb like telefonare ‘to telephone’ is like the subject of a transitive
verb in that it cannot be expressed as the clitic ne, which gets attracted to
the finite verb (compare (3c) with (3b)). In contrast, the sole argument of an
unaccusative (nonagentive) verb like arrivare ‘arrive’ is like the object of a
transitive verb in that it can undergo ne-cliticization (compare (3d) with (3a);
see Burzio 1986 and Belletti and Rizzi 1981 for extensive discussion).
(3) a. Ne ha affondato due –. (object of transitive verb)

of.them have sunk two (Cinque 1990:5)
‘He/she has sunk two of them.’
b. *Ne hanno avuto successo due –. (subject of transitive verb)
of.them have had success two (Cinque 1990:5)
‘Two of them had success.’
c. *Ne telefonano molti –. (sole argument of agentive verb)
of.them telephone.3pS many (Burzio 1986:20)
‘Many of them telephone.’
d. Ne arrivano molti –. (sole argument of nonagentive verb)
of.them arrive.3pS many (Burzio 1986:20)
‘Many of them arrive.’
In a similar way, the sole argument of adjectives like ‘good’ and ‘dangerous’
cannot undergo ne-cliticization, whereas the sole argument of adjectives like
‘well-known’ and ‘obscure’ can (Cinque 1990:7):
(4) a. *Ne sono buoni pochi (dei suoi articoli).

of.them are good.m.pl few of his articles.m.pl
‘Few of them (his articles) are good.’
b. *Ne sono pericolosi molti (di viaggi).
of.them are dangerous.m.pl many of journeys.m.pl
‘Many of them (journeys) are dangerous.’
(5) a. Ne sono note solo alcune (delle sue poesie).
of.them are well.known.f.pl only some of his poems.f.pl
‘Only some of them (his poems) are well known.’
b. Ne è oscuro più d’ uno (di motivo).
of.them is obscure.m.sg more than one of reason.m.sg
‘More than one of them (reasons) is obscure.’
Cinque (1990) also gives five other syntactic tests that reveal the same distinction
between ‘good’-class adjectives and ‘well-known’-class adjectives. For conve-
nience, I’ll refer to adjectives like ‘good’ as normal adjectives and adjectives
like ‘well-known’ as Cinque adjectives, after their discoverer.1
Updating Cinque’s structures in accordance with my (2003a) claim that adjec-
tives never have specifiers, the contrasting structures are as shown in (6).
1 Cinque himself calls adjectives like ‘well-known’ “ergative adjectives,” parallel to Burzio’s
(1986) “ergative verbs,” but I avoid Burzio’s terminology because it is too easily confused with
ergative case and agreement marking.
(6) a. [TP – be+TENSE [PredP few+ne PRED [FA [AP good ]]]]
b. [TP – be+TENSE [PredP – PRED [FA [AP well-known some+ne ]]]]
The clitic ne can move to adjoin to the tensed auxiliary in (6b) but not in (6a)
because only in (6b) is its trace properly governed by a lexical head (the adjective
‘well-known’; see Baker 2003a:63–9 for specific semi-Minimalist formulations
of the relevant principles).
The question of interest for us is what are the implications of this structural
difference for agreement on FA ? We can infer from Cinque’s failure to use
agreement as evidence for his distinction that normal adjectives and Cinque
adjectives do not differ in so salient a property. In fact, both classes of adjectives
agree in number and gender with their sole argument. (7) shows that there
is agreement with the sole argument of a Cinque adjective, even when ne-
cliticization proves that that argument has not moved higher, to Spec, TP (this
can also be observed in (5)).
(7) a. Ne sono ormai probabili le dimissioni.
of.them are already likely.m.pl the resignations.m.pl
‘Their resignations are already likely.’
b. Ne sono oscuri i motivi
of.them are obscure.m.pl the reasons.m.pl
‘Their reasons are obscure.’
The adjectival agreement in the normal adjective structure in (6a) is the kind that
motivated upward probing in chapter 2: FA does not c-command ‘few of them’
at any level of structure, but ‘few of them’ c-commands FA , and that is sufficient
for agreement to take place. Agreement in the Cinque adjective construction in
(6b) is an instance of the more familiar downward agreement: FA c-commands
NP but not vice versa in this structure. (It has been known since Burzio 1986 that
ne-cliticization is only possible with postverbal NPs in Italian; movement of
the argument to subject position renders the sentence ungrammatical. Thus, the
agreed-with argument in (6b) must remain in the c-command domain of FA at all
syntactic levels for ne-cliticization to be possible.) Moreover, the morphological
realization of this downward agreement is identical to that of upward adjectival
agreement. This shows that both upward and downward agreement are possible
for the FA associated with adjectives, and the two are equivalent from the
viewpoint of morphology.
3.1.2 Raising adjectives

Another relevant case to consider is raising adjectives such as likely. In English,
these can appear in expletive constructions in which their local subject has no
ϕ-features. These are precisely the kinds of constructions in which verbal predi-
cates like seem manifest longer distance, unambiguously downward agreement,
as shown in (8a).
(8) a. There seem [– to be three unicorns in the garden].
b. There are likely [– to be three unicorns in the garden].
Unfortunately, we cannot observe agreement on the adjective likely in (8b)

in English, for the obvious reason that adjectives never inflect for agreement
in English. (Either FA is not generated in English, or it is generated but is
always spelled out as Ø.) Fortunately, we can check similar structures in Ice-
landic, which is perhaps the only IE language that has both a raising adjec-
tive comparable to likely and robust inflection of adjectives.2 (9) is a relevant
example.
(9) fiað eru vı́st lı́klegir til að verð a valdir einhverjir komúnistar
there are thsy likely.m.pl to be elected some communists.m.pl
ı́ stjórnina.
to board.the
‘There are likely to be some communists elected to the board.’
Note that the adjective lı́klegir ‘likely’ does agree with the masculine plural
NP ‘some communists’ that is embedded inside its clausal complement. (I
thank Halldór Sigurð sson and Kjartan Ottosson for judgments and much valu-
able discussion relevant to this section and the next.) This then is a second
clear instance of FA probing downward for an NP to agree with, rather than
upward.
3.1.3 Comparing adjectival and verbal agreement

If downward adjectival agreement is the same phenomenon as verbal agreement,
we expect it to be subject to the same restrictions as verbal agreement. That
seems to be true. For example, a Cinque adjective in Italian can agree with its
bare NP complement ((7)), but it cannot agree with an NP embedded in a PP,
as shown in (10).
(10) a. Era oscura (*oscuri) a tutti la propria origine.
was obscure.f.sg (obscure.m.pl) to all.m.pl the their.own origin.f.sg.
‘Their own origin was obscure to everybody.’ (Cinque 1990:13)
2 Raising adjectives seem to be quite rare crosslinguistically, for unknown reasons. For example,
there are no such adjectives in the Romance languages or in Russian, even though those languages
have raising verbs comparable to seem.
b. Era chiaro (*chiare) alle donne che avrebbe

was.3sS clear.m.sg (*clear.f.pl) to.the women.f.pl that had
piovuto.
rained
‘It was clear to the women that it had rained.’
This shows that adjectival agreement is subject to the activity condition: there
is no agreement with an NP whose case has already been checked by P.3 (10)
is parallel to the absence of agreement between seem and the object of the
preposition in (11b).
(11) a. There arrive three new women each day.
b. *There seem to three new women that it will rain.
Similarly, the raising adjective ‘likely’ in Icelandic can agree with an NP

inside a nonfinite clause, as in (9), but it cannot agree with a similar NP inside
a finite clause:
(12) fiað er lı́klegt (*eru lı́klegar) að þrjár konur séu ı́ herberginu.
there is likely.m.sg (*are likely.f.pl) that three woman are in room.the
‘It is likely that three women are in the room.’
The contrast between (9) and (12) is parallel to the contrast between (13a),
where the raising verb seem agrees into a nonfinite clause, and (13b), where the
raising verb seem fails to agree into a finite clause.
(13) a. There seem to be three women in the garden.
b. *There seem that three women are in the garden.
This shows that downward adjectival agreement is subject to the phase condition
in (1c), since the tensed clause is a distinct phase from the matrix clause but
this type of infinitival clause is not.
Finally, we can consider the intervention condition in (1b). The canonical
evidence for this condition comes from examples like (14a) in Icelandic, where
agreement on the matrix raising verb ‘seem’ with the embedded nominative NP
‘horses’ is blocked by the presence of the dative NP Jóni that appears between
the two. (14b) is a similar example with a raising adjective; here too the presence
of the intervening dative NP (the subject of the embedded predicate) prevents
the adjective from agreeing with the nominative object of the lower predicate
(Sigurð sson, personal communication).4
3 Alternatively, (10) could illustrate the phase condition on adjectival agreement, if PP counts as
a phase, as suggested in section 2.3.1.
4 A complication is that Sigurð sson reports that (14b) “cannot even be constructed.” As far as any
theory that I know would have it, examples like this ought to be possible with default masculine
(14) a. ?*Mér virð ast Jóni vera taldir lı́ka hestarnir.

me.dat seemed.3pS John.dat to.be believed.pl like horses.nom.pl
‘I perceive John to be believed to like horses.’ (Schütze 1997:108)
b. fiað eru lı́klegar sumum mánnum til að leið ast
there are likely.f.pl some men.dat toward to find.boring
sumar konur.
some women.f.pl
‘Some men are likely to be bored with some women.’
I conclude that agreement on predicate adjectives can indeed probe downward,

the way that verbs do. When it does, it is transparently subject to exactly the
same factors that verbal agreement is subject to. This can even be seen internal
to each example I have given, because the tensed copular verb and the predicate
adjective always agree with the same noun phrase, or, if no NP is accessible,
they both show default agreement. I take this to be strong evidence in favor of
a unified account of agreement on adjectives and verbs.
If adjectives can take part in either upward or downward agreement, what
decides which way it will go in a given sentence? Are there ever cases of
optionality, in which the adjective can probe upward and agree with one NP or
probe downward and agree with another in free variation? Or does one kind of
agreement take priority over the other?
In fact, it is not clear that this possibility ever arises, because of case theory.
To investigate the question, we would need a configuration in which an adjective
has both a thematic subject and an NP complement. Near-candidates would be
adjectives like ‘certain,’ which take part in the alternation in (15), discussed by
Cinque (1990:2–3).
(15) a. Gianni è certo che verrò.
Gianni is certain that I.will.come
b. Gianni è certo di questo.
Gianni is certain of this
c. Che verrò è certo.
that I.will.come is certain
d. Questo è certo.
this is certain
When the thematic subject is not projected and the complement of the adjective
is nominal, as in (15d), the adjective agrees with its complement, as expected.
singular morphology on the copula and adjective. If they are not, some additional explanation is
required.
When the subject is projected, as in (15b), the adjective apparently agrees only
with its subject, never with its complement. But this does not prove that upward
agreement takes precedence over downward agreement when both are possible.
The reason is that the NP internal argument of the adjective in (15b) (unlike the
parallel clause in (15a)) actually appears inside a PP headed by de, and we know
that agreement never looks inside PPs. So agreement with the internal argument
of ‘certain’ is blocked in (15b) for the same reason as it is in (10a–b). The
question then is why de must appear in (15b) but not in (15a, c–d). The answer
(as anticipated by Cinque) presumably comes from case theory: adjectives do
not check/assign structural case. In (15d), the adjective’s internal argument can
have its case licensing by Tense, but in (15b) this case-licensing relationship is
established with the subject of predication. Therefore, the preposition must be
inserted to license (or spell out) case on the internal argument. This preposition
then prevents the adjective from agreeing with the internal argument. Therefore,
there is no simple adjectival construction in which both upward and downward
agreement are expected to be possible.
Finally, I need to make something explicit in my theory in order to ensure that
there is no person agreement even on Cinque adjectives in Italian examples like
‘We are well known’ or on raising adjectives in Icelandic examples like ‘You
are likely to win.’ In these examples, the surface subject is generated inside
the AP and moves to the matrix Spec, TP. The principles of movement would
allow the NP to move through Spec, FA P on its way to Spec, TP, resulting in
the following syntactic representations:
(16) a. [TP We[1,pl] be+T [PredP t[1,pl] Pred [ t[1,pl] FA [AP well-known t ]]]]
b. [TP You[2,sg] be+T [PredP t[2,sg] Pred [ t[2,sg] F A [AP likely [TP t to win]]]]
If these representations were possible, then FA could manifest first or second

person features in these configurations, in keeping with the SCOPA. Cinque-
type and raising-type adjectives could thus be exceptions to the general rule that
adjectives cannot agree in person. But Cinque adjectives and raising adjectives
are not exceptional in this respect; they support the same range of inflections
as other adjectives in the languages in question.
In fact, an answer to this problem was implicit in the convention stated in
(52) of chapter 2, repeated here as (17).
(17) Any lexical category can be immediately dominated by the projection of a

functional head that matches it in gross categorical features.
This is a version of the standard assumption that the functional heads that appear
with VPs are somehow verbal, those that appear with NPs are nominal, and those
that appear with APs are adjectival (compare Grimshaw’s (1991) notion of an
extended projection). Within my (2003a) theory of categories, however, being
adjectival as opposed to verbal has a precise syntactic meaning. Verbal phrases
have the defining property of having a specifier, whereas adjectival phrases
cannot have a specifier. (17) projects this property onto the functional heads
above VP or AP as well. Therefore, the FA that dominates AP in structures
like (16) is itself adjectival – meaning it must not have a specifier. Thus, there
cannot be a trace of the moved NP in Spec, FA P, and FA cannot have +1 or
+2 features (although it may agree at a distance with the moved NP in other
features). In more familiar terminology, FA never has an EPP feature, whereas
the functional heads associated with verbs do. With this clarification, we can
maintain the generalization that true adjectives never agree with an NP in first
and second person features, despite the fact that adjectives appear in a wider
range of syntactic structures than were considered in chapter 2.
3.2 Upward agreement on verbs

Next I consider whether it is possible for verbs, like adjectives, to agree upward
with something that they (and, more precisely, the functional heads associated
with them) do not c-command. Recall from section 2.4.2 that verbs can have
at least two distinct functional heads generated above them: Tense, the normal
locus of subject agreement, and v, the normal locus of object agreement. Either
of these functional heads could conceivably probe upward for something to
agree with, so I discuss them separately, beginning with Tense.
3.2.1 Upward agreement on tense

The prospects for finding agreement on Tense that is unambiguously upward
are very limited. Under standard assumptions, there are no θ -positions where
an NP can originate that are higher than Tense but still in the same clause. The
θ -positions of the clause are all internal to the vP complement of the Tense head.
It often happens that one of these NPs moves to Spec, TP, landing in a position
that c-commands Tense. Such NPs often agree with Tense, resulting in canonical
subject agreement. However, it is possible to say that the agreement relationship
Upward agreement on verbs 75
was established prior to the NP movement, when Tense c-commanded the NP,
as assumed in Chomsky 2000, 2001:5
(18) [TP The three womeni do+T [not [vP all ti V [VP like eggplant]]]]
AGREE
There could of course be NPs that are generated in a higher clause than the Tense
node, which would c-command Tense at all levels. We could thus contemplate
the possibility of upward agreement on the embedded verb in sentences like
those in (19).
(19) a. *Three women said that there seem that it will rain.
b. *I told three women that there seem that it will rain.
This sort of agreement is quite impossible. However, this does not prove that
verbal functional heads cannot search upward for something to agree with.
Agreement in examples like (19) is already ruled out by the phase condition in
(1c), since the CP complement of say or tell counts as a phase which contains
the Tense associated with seem but not the putative controller of the agreement.6
Virtually the only chance for T to manifest agreement that is unambiguously
upward, then, would be in structures that involve wh-movement. Wh-movement
might move (say) a plural NP from a lower phase (where T cannot agree with
it) to place it in Spec, CP. Once it reaches Spec, CP, the +wh NP would
asymmetrically c-command the agreeing T, and it would be contained in the
same phase as that T. Therefore, if upward verbal agreement is allowed by
Universal Grammar, it might show up in this configuration.
Kayne (2000:190) discusses examples like (20a) (originally from Kimball
and Aissen 1971), which are possible in certain dialects of English.
(20) a. the people who Clark think are in the garden

b. the person who Clark think*(s) is in the garden
In (20a), the verb think appears in its plural form, despite the fact that its subject
is third person singular. It apparently agrees instead with the plural relative
5 In chapter 5, I argue that Kinande has English-like subject–verb agreement, but never allows
agreeing heads to probe downward. If this analysis is accepted, then that language has many
instances of T (and v) probing unambiguously upward for something to agree with in a post-
movement structure (see section 5.2).
6 Occasionally TPs are not contained in CPs that constitute phases: this is true of raising and ECM
infinitives in Chomsky’s theory. But these TPs have Ts that do not manifest agreement, at least
in familiar languages like English. Hence, upward agreement with a phrase in the matrix clause
is not expected even here.
pronoun who in Spec, CP (which in turn agrees with the head of the relative
the people). This relative pronoun clearly c-commands the Tense associated
with the verb, and not vice versa. In the spirit of Kimball and Aissen’s original
analysis, I conjecture that, because a subject like Clark has only unmarked
ϕ-features (third person, singular), its features can be left unspecified in some
dialects. When this happens, T cannot agree with the subject. The wh-phrase
who is thus the phrase with explicit ϕ-features that is closest to T searching
upward, and it governs agreement on T.7
This account can explain why the opposite of (20), upward agreement with
a singular who over a plural subject, is not possible in these dialects (Kimball
and Aissen 1971; Kayne 2000:190):
(21) *the man who the girls thinks is in the garden
Here the NP the girls has a marked ϕ-feature, namely plural. This cannot be
left underspecified in the syntactic representation. Therefore the subject counts
as an intervener that is closer to the finite T than the singular who. Agreement
between Tense and who is thus blocked by the intervention condition ((1b) in
this example).
One might wonder if the agreement between T and who in (20a) could instead
be analyzed as T agreeing downward with who before who moves to the highest
Spec, CP. This alternative seems unlikely. Kimball and Aissen (1971:246) state
that “only verbs between the position of the relativized NP and the head of the
relative clause are subject to nonstandard agreement.” Agreement is thus not
found in examples like those in (22).
(22) a. The boy think*(s) the people are in the garden. (Kayne 2000:208)
b. Lucine know*(s) which people John think(s) are crazy.
(Linda Baker, personal communication; cf. Kimball and
Aissen 1971:246)
(22a) shows that the matrix T cannot agree with a plural NP in the argument
position that the wh-phrase in (20a) moves from. This is expected, since the
7 Hans Broekhuis (personal communication) points out that something needs to be said about why
this dialect does not allow T to agree with a plural direct object – the closest NP with ϕ-features
searching downward – when the subject is third person singular. This would wrongly allow
sentences like *Clark like the girls. I assume that this sort of agreement violates the activity
condition in (1d), the object already having its accusative case licensed by v.
This in turn raises the question of how the agreement with the wh-phrase in (20a) can be
squared with the activity condition. One possibility is that the uninterpretable features involved
in wh-movement render the wh-phrase eligible for agreement. For another conjecture, stated in
terms of the revised and parameterized version of the activity condition that I propose in chapter
5, see note 6 in chapter 5.
plural NP is in a phase (the embedded CP) that does not contain T. (22b)
shows that even when a plural NP moves to the specifier of the embedded CP,
agreement is not possible in the relevant dialect. Restating Kimball and Aissen’s
generalization in current terms, agreement takes place between T and the wh-
phrase only when the wh-phrase crosses over T to land in a higher position.8
This, then, is an instance of upward verbal agreement showing up in exactly the
situation where we predicted it might. See also Bruening 2001:sec. 4.3 on so-
called participle agreement in Passamaquoddy for another, more robust instance
of verbs agreeing in ϕ-features with an operator in Spec, CP. (It is not entirely
clear, however, whether the relevant agreement morphemes in Passamaquoddy
are realizations of agreement on the T node, or some other head, such as v
or C.)
3.2.2 Upward agreement on v

Consider next the possibility of upward agreement on v, the other agreement-
bearing functional head that can dominate VP. The prospects for this happening
are somewhat brighter, inasmuch as there is an NP that is regularly found above
v – the agent noun phrase, generated in Spec, vP in Chomsky 1995 and related
work. The question, then, is whether v ever agrees with the agentive subject,
or whether subject agreement is necessarily in Tense or some other higher
functional node.
3.2.2.1 Basque and other languages with ergative agreement

Bejar (2003) and Rezac (2003) have argued that agreement between v and
the subject happens in a class of languages with complex agreement systems,
including Basque, Georgian, and Nishnaabemwin. The Basque case is perhaps
the most transparent, so it is the one that I review here.
In Basque, there is a certain agreement slot, the prefix to the auxiliary root,
which normally registers agreement with the object of a transitive verb. Agree-
ment with the subject of the transitive verb is distinct from this, both in terms of
its position (it suffixes to the auxiliary) and in terms of phonological features,
as shown in (23).9
8 (20a) might not count as evidence for upward agreement if wh-movement has intermediate
landing sites between the lower Spec, CP and the higher Spec, CP, as Kayne (2000:208) assumes,
citing evidence from quantifier stranding. This theoretical device is not necessary to account for
agreement with a wh-phrase in Spec, CP in my theory, and I take this to be a good thing, given that
the evidence for other landing sites is weak and theory-internal at best. (I find Kayne’s example
quite bad, for example.)
9 For all but a small number of Basque verbs, tense and agreement appear on an auxiliary verb
rather than on the main verb. I assume that this is a phenomenon akin to do-support: perhaps most
(23) a. Andoni-k ni ikusi n-au. (Salaburu Etxeberria 1981:138)

Anthony-erg me.abs see 1sA-aux
‘Anthony sees me.’
b. Andonik zu ikusi z-aitu. (Salaburu Etxeberria 1981:138)
Anthony-erg you.abs see 2A-aux
‘Anthony sees you.’
c. Nik zu ikusi z-intu-da-n. (Saltarelli 1988:306)
I.erg you.abs see 2A-aux -1sE-past
‘I saw you.’
d. Ikusi n-indu-zu-n. (Saltarelli 1988:306)
see 1sA-aux-2E-past
‘You saw me.’
In these examples, prefixal n is object agreement with first person singular,

as seen in (23a, d), whereas suffixed da is subject agreement with first person
singular ((23c)). Similarly, the prefix z is object agreement with second person
singular polite ((23b–c)), and the suffix zu is subject agreement with second
person singular polite ((23d)). Rezac (2003) thus assumes that forms like n and z
are morphological realizations of agreement on v, and da and zu are realizations
of agreement on T, in a standard structure like (24) for (23c). (I suppress the
head-finality of Basque for expository convenience.)
(24) TP
NP T´
I T vP
[1,sg] [1,sg]
NP v’
t NP v’
you v VP
[2,sg] [2,sg]
NP V’
t V XP
see
verbs cannot raise up to v, and the auxiliary is inserted into v and raises to T, thereby spelling
out tense, transitivity, subject agreement, and object agreement, but not the lexical content of the
verb or some aspect distinctions. In any case, I abstract away from this. Note that the auxiliary
stem itself also shows complex allomorphy conditioned by various factors as well as bearing the
fairly regular and easily segmentable agreement affixes.
Based on this partial paradigm, one would expect to find forms like AUX-
da-n for ‘I verbed him’ and AUX-zu-n for ‘you verbed him’. But instead one
finds the forms in (25).
(25) a. Nik liburua erosi n-u-en. (Salaburu Etxeberria 1981:142)
I.erg book.abs read 1sA-aux-past
‘I read a book.’
b. Zuk liburua erosi z-en-u-en.
you.erg book.abs read 2sA.fam-aux-past
‘You read a book.’
In these examples, the thematic subject triggers the “object” agreement prefixes
n and z rather than the “subject” agreement suffixes da and zu. The agreement
affixes in (25a–b) are therefore identical to those in (23a–b) (although the sup-
pletive form of the auxiliary root is different; see note 9). This is known in
the Basque literature as ergative displacement. Building on Bejar’s work on
Georgian, Rezac analyzes this as follows. He assumes that third person argu-
ments are not marked for person features in Basque, but rather are unspecified
for this feature, much as I assumed for the subject Clark in (20). As a result,
when v probes downward for something to agree with in sentences like (25), it
finds no suitable goal. In this case, v probes upward instead, and agrees with
the agent NP in its specifier position.10 This agreement relationship makes the
NP inactive, presumably by valuing its case feature. As a result, T is not able
to agree with the subject, by the activity condition. The thematic subject thus
triggers “object” agreement but not subject agreement in these examples:
(26) TP
T vP
[---]
NP v’
I v VP
[1,sg][1,sg]
NP V’
AGREE
book V XP
[--]
read
10 Rezac (2003) himself proposes a Minimalist interpretation under which the agreement of a head
with its specifier does not really count as upward agreement in my sense (see also Bejar 2003).
He suggests that the copy of a head that is taken as the label of the projection in Chomsky’s Bare
Phrase Structure theory counts as a full-fledged instance of that head. Everything contained in
the maximal projection, including the specifier, then counts as being in the c-command domain
of that copy of the head. Taking his view would decrease the amount of evidence that counts
as unambiguous support for my theory of agreement. On my theory, however, no such radical
view about the nature of labels and their role in a syntactic derivation is required.
This then is a second apparent instance of upward agreement in the verbal

domain.11
Given the analysis in (26), we would also expect v to agree upward with
the thematic subject in intransitive clauses in Basque. These would be like the
transitive clauses just discussed inasmuch as there is no NP with ϕ-features
within the c-command domain of v that it can agree with. Therefore, v is forced
to search upward in these clauses too. This is correct: the sole argument of an
intransitive verb in Basque triggers “object” agreement forms such as prefixal
n and z, as shown in (27).
(27) a. Ibil-tzen n-intz-en. (Ortiz de Urbina 1989:9)
walk-hab 1sA-aux-past
‘I used to walk.’
b. Jaus-i-ko z-ara.
fall-fut 2sA.fam-aux.npst
‘You will fall.’
Since this type of agreement is triggered by intransitive subjects as well as by

transitive objects, it is usually – and more accurately – called absolutive agree-
ment rather than object agreement. But it can be nicely analyzed as agreement
on v, which probes downward when it can (in transitive sentences) and upward
when it cannot (in intransitive sentences). This is also one possible analysis
of ergative-absolutive agreement paradigms in other ergative languages, even
if they have no equivalent of ergative displacement. These languages could be
treated just like Basque, except that the object NP is never underspecified for ϕ-
features. As a result, v always probes downward to find the object in transitive
clauses and probes upward to find the subject in intransitive clauses, resulting
in an ergative-absolutive pattern of agreement.
There are other languages of the world that have ergative agreement patterns,
and there is no consensus about how they should be analyzed theoretically.
Nevertheless, it is likely that many of them also involve v agreeing upward with
the agent in Spec, vP. Rezac’s view that absolutive agreement is agreement
on the lower functional head (here v) is a descendant of one proposed by
Chomsky (1993:9) and argued for by Bobaljik (1993) for Inuit (see also Levin
and Massam 1984). In contrast, various authors have argued that absolutive
agreement is agreement on the higher functional head (here T) and ergative
11 Number agreement does not displace in Basque the way that person agreement does; see Rezac
2003 and references there for details. I abstract away from this here. Also ergative displacement
happens in some tenses (e.g., past) but not others (present), a fact that needs to be taken into
account in a fuller treatment.
agreement is agreement on the lower functional head (v), including Murasugi

(1992), Campana (1992), Bittner (1994), and O’Herin (2002) (see also Bittner
and Hale 1996). According to this second view, an ordinary transitive sentence
like ‘you see me’ would have a structure like the following:
(28) TP
NP T´
<me> T vP
[1,sg] [1,sg]
NP v’
you v VP
[2,sg] [2,sg]
NP V’
<me> V XP
[1,sg]
see
This structure presents at least one theoretical difficulty as compared to Rezac’s

(26): the Tense node either has to look past the agent NP in order to agree with
the object, or the object needs to move up past the agent NP to be agreed with
by T. The first option violates the intervention condition in (1b), whereas the
second might violate the Shortest Move condition, plus it is often not borne out
in the surface word order (for example, the word order in (29b) is Agent-Theme-
Verb, not Theme-Agent-Verb). But these theoretical problems might well be
solveable,12 and (28) has empirical advantages in some languages. For example,
consider the following range of data from Abaza, which O’Herin (2002) uses
to argue for a structure like (28):
12 It is interesting to note in this regard that ergative agreement patterns are not particularly common,
suggesting that there is something special and marked about them. For example, the World Atlas
of Language Structures (Haspelmath et al. 2005:map 100) identifies only 19 languages with
ergative agreement in its sample of 390 languages, as compared to 212 languages that have a
“normal” accusative pattern of agreement.
One marked possibility might be that the realization of ergative case on the agent NP makes
its ϕ-features invisible to T by something akin to P-insertion. Then the agent would not have
the features that would make it count as an intervener between T and the object. Alternatively,
it is possible that the object does move above the subject to Spec, TP, but the higher copy of the
chain deletes rather than the lower one, so one still observes Agent-Theme-Verb order in (29b).
(This is equivalent to O’Herin’s own view; see section 5.10.2.1 for evidence that something like
this happens in Berber.)
(29) a. A-phas d-ʕ ay-d. (p. 64)

def-woman 3sA.r-come-dyn
‘The woman came.’
b. Zaw ə a-phas də -y-ʕ a-y-d. (p. 67)
someone def-woman 3sA.r-3sE.m-hear-pres-dyn
‘Someone hears the woman.’
c. D-qac’a-b. (p. 76)
3sA.r-man-stat.pres
‘He is a man.’
d. s-ba-ra (p. 28)
1sE-see-noml
‘seeing me’
(29a) is a simple intransitive verbal clause and (29b) is a simple transitive

clause; they show that the transitive object triggers the same sort of agreement
on the verb as the intransitive subject does (here the third singular human prefix
d). The transitive subject is indicated by a different sort of agreement (for
example, third singular masculine y in (29b)). (29c) shows a predicate nominal
construction, which clearly has a T node (realized as the stative present suffix b)
but presumably not a v node. This form bears an absolutive agreement d but not
an ergative agreement y – evidence that absolutive agreement is agreement on
T. Conversely, (29d) is a nonfinite verbal construction formed from a transitive
verb; this example presumably contains a transitive v but no finite T. This
form can bear an ergative agreement but no absolutive agreement, suggesting
that ergative agreement is agreement on v (O’Herin 2002:76). Finally, O’Herin
points out that absolutive agreement is the outermost prefix on the verb in
examples like (29b), whereas ergative agreement appears closer to the verb
stem. Mirror Principle-style reasoning thus suggests that absolutive agreement
is on the higher functional head, the last one to unite with the verb – namely T.
These arguments also carry over quite well to Mayan languages, such as Tzotzil
(see Aissen 1987 for relevant data). So (28) is a serious analytic possibility.
It is not clear to me whether (28) is the correct analysis for languages with
ergative agreement, or whether (26) is, or whether Universal Grammar allows
two distinct kinds of ergative agreement. But for my current interests it is not
essential to have the last word on this topic, because both theories share the
assumption that v sometimes agrees upward rather than downward. For Rezac
the intransitive v agrees upward to give absolutive agreement, whereas for
O’Herin the transitive v agrees upward to give ergative agreement. Either way,
adjective-like upward agreement plays a role in the analysis of ergativity, so

upward agreement must be theoretically possible.13
3.2.2.2 Subject agreement in Tariana

Verbal agreement in Tariana provides an interesting variation on the theme of v
agreeing upward with the agentive subject. In transitive clauses, Tariana verbs
agree with their subjects but not with their objects, as shown by the examples
in (30) (all Tariana data come from Aikhenvald 2003).
(30) a. Na-na kuphe-nuku di-walita. (p. 236)
3.pl-obj fish-top.non.a/s 3sS-offer
‘He offered them fish.’
b. Nuha siɾ uɾ i nu-walita nu-a.
I cumatá.leaves 1sS-offer 1sS-go
‘I am making an offering of cumatá leaves.’
Many intransitive verbs show the same agreement with their sole argument:
(31) a. Nu-ɾ uku nu-a. (p. 239)
1sS-go.down 1sS-go
‘I am going downstream.’
b. Di-thuka-kha di-ɾ uku di-a diha-na-ne. (p. 235)
3sS-break-intrans 3sS-fall 3sS-go art-cl:vertical-foc.a/s
‘It (the penis of an evil spirit) was breaking and falling off.’
But there is a subclass of intransitive verbs that do not show any agreement with
their sole argument. These are stative/nonagentive verbs, presumably a subset
of the unaccusative verbs of Tariana:
(32) Lama-sina diha-dapana. (not: *di-lama-sina . . .) (p. 239)
burn-rep.nonvis art-cl:habitation
‘This house was burning.’
Why should the presence of subject agreement depend on the argument struc-
ture and lexical semantics of the verb in Tariana? This makes sense if the one
agreeing head in this language is v rather than T. Unlike all transitive verbs
13 There is one viable theory of ergative agreement systems that makes no use of upward agreement:
the view that both ergative and absolutive agreement are realizations of agreement on T (see
Bok-Bennama 1991 on Inuit and Bobaljik and Branigan 2006 on Chukchi). I argue that this is
true for Nez Perce (section 5.10.2.2), and tentatively accept it for Inuit and Chukchi as well.
This approach does not seem so plausible for Abaza or Tzotzil, however, given the existence of
syntactic environments like (29c–d) which have one sort of agreement but not the other.
and some intransitive verbs, stative intransitive verbs like ‘be burning’ plausi-
bly lack a v head. This would account for their inability to show agreement,
given the assumption that v is the only functional head with agreeing power in
Tariana.
Support for this idea comes from certain other details about agreement in
Tariana. First, agreement is not morphologically related to Tense or Mood
morphology in Tariana, as it is in many languages. Agreement is realized as
a prefix on the verb, whereas Tense is a suffix. Nor do variations in Tense
condition allomorphy of the agreement prefixes in this language.
Second, Tariana has nonfinite verb forms that are comparable to gerunds in
other languages. Gerundive verbs lack the ability to agree with their subjects in
languages where subject agreement is associated with the Tense node, includ-
ing English, Italian, Kinande, Lokaa, and Mapudungun, among others. But
Tariana is different in this respect: nonfinite gerundive verbs show exactly the
same subject agreement as other verbs in the language. Aikhenvald (2003:461)
thus states that “all nominalizations formed on prefixed verbs cross reference
the A/Sa [i.e. subject] constituent and can be derived from any verb.” Some
representative examples are:14
(33) a. [Hı̀ maɾ u na-ɾ apa-nipe] pi-na tuki nu-kalite-de.
dem:an dance.master 3pS-dance-noml 2.sg-obj a.little 1sS-tell-fut
‘I will tell you a little about the dance master’s dancing.’ (p. 462)
b. hinipuku w-ehpani-nipe; pe:the na-ni-nipe (p. 461)
garden 1pS-work-noml manioc.bread 3pS-do-noml
‘our working in gardens’ ‘their making manioc bread’
This fits very well with the idea that subject agreement is not a property of finite
Tense, which is missing in (33), but rather of v, which is present to assign the
agent θ -role to the subject in each example.
Third, passive verbs in Tariana systematically lack subject agreement. Rather,
the prefix slot where this agreement normally appears is filled with the invariant
prefix ka. Thus there is no 3sS prefix di on the passive verb in (34a), and no
first person nu in (34b).
(34) a. Ha-ne ka-ñha-kana-mhade di-a. (p. 259)
dem:inan rel-eat-pass-fut 3sS-aux
‘This one will be in the process of being eaten up by the jaguar.’
14 I do not know why ‘the dance master’ is translated as a singular noun but triggers plural agreement
on the verb here. Perhaps this is a plural of honorification.
Verbs that cannot agree in person, like adjectives 85
b. tʃ o! Nhua-sini ka-ñha-kana-kasu. (p. 259)

Oh I-too:act rel-eat-pass-int
‘I, too, am about to be eaten!’
Here the active v is missing and a finite Tense is present, and agreement is
also missing. Overall, subject agreement in Tariana is contingent not on the
verb being finite as opposed to nonfinite, but rather on the voice and lexical
semantics of the verb – exactly the distribution we expect if the sole agreement
factor in this language is attributed to v rather than to T. But even though this
agreement has the earmarks of being associated with the v node, it is nevertheless
agreement with the thematic subject. This then is another instance of v agreeing
upward with the agentive subject generated as its specifier.
Comparing Tariana with Basque gives us another chance to consider what
governs the choice of upward versus downward agreement when both are pos-
sible, a matter that I was unable to resolve in the domain of adjectives. There
seems to be a basic difference between the two languages on this point. On
Rezac’s analysis, the v in Basque prefers to agree downward with the thematic
object; it only agrees upward when there is no object that is marked for
ϕ-features. In contrast, the v in Tariana can only agree upward with the thematic
subject. This difference could be parametric in nature. In chapter 5, I argue
that Kinande and other Bantu languages have a more restrictive version of the
c-command condition in (1a) than the Indo-European languages do: Kinande
allows agreement on a head F only if the agreed-with NP asymmetrically
c-commands F. Given such a parameter, much of the difference between Tariana
and Basque can be captured by saying that Tariana is like Kinande in allowing
only upward agreement. (See section 5.6.2 for some confirming evidence for
this from the syntax of quantifiers in Tariana.) I conclude that v not only may
but actually must agree upward in some languages.
3.3 Verbs that cannot agree in person, like adjectives

So far I have shown that in certain configurations verbs can be like adjectives in
agreeing with a higher NP rather than a lower one. Now I turn to the other salient
difference between adjectival agreement and verbal agreement that could stand
in the way of a unified theory: the fact that verbs agree with their arguments
in person as well as number and gender, whereas adjectives never do. The
SCOPA stated in (2) says that this is because FA never merges directly with
an NP that has +1 or +2 features, whereas an FV can. For adjectives, the
consequence is a firm one: the complement of FA is AP (by definition) which
does not have ϕ-features, and inasmuch as FA is intrinsically adjectival it never

licenses a specifier (see (17)). Therefore, the functional head most immediately
associated with adjectives can never agree in person for fundamental reasons.15
But for verbs there should be chances of finding intermediate cases. There
might very well be instances in which a functional head dominating VP agrees
with something that it does not merge with, either because it exceptionally lacks
an EPP feature, or because its EPP feature is discharged by something that does
not have ϕ-features. In these situations, the expectation of my theory is that
the verbal head can still agree with a nearby NP in number and gender, but
should lose the ability to agree in person. Such instances of verbal agreement
would be adjective-like in this respect, and would support the notion that the
same category-neutral theory of agreement applies to both. In this section I
document a series of such cases.
3.3.1 Tense agreeing with Spec, CP

One case that should be relevant is the agreement between Tense and a wh-
phrase in the Spec, CP above it, found in the nonstandard dialect of English
discussed by Kimball and Aissen (1971) and Kayne (2000) (see section 3.2.1).
The crucial paradigm is in (35).
(35) a. ?the people who Clark think are in the garden

b. ?The people who Clark are hoping are in the garden
c. I who am (*is) in the garden
d. *I who Clark am hoping am in the garden
(35a) repeats a standard example of the effect, where the verb think shows
plural agreement with the wh-word who rather than singular agreement with its
subject Clark. (35b) is a similar example where the crucial verb is be, which
shows suppletive plural agreement with the moved wh-word. Two informants
for this dialect say that they do not find this sentence appreciably worse than
(35a). (35c) shows that a wh-phrase can count as first person singular when it is
anteceded by a first person singular pronoun in a nonrestrictive relative clause.
(35d) is the crucial test sentence. Here the verb be shows suppletive first person
singular agreement with the wh-phrase in Spec, CP, despite its grammatical
subject being third person singular. My informants agree that this is completely
impossible, clearly worse than the example with only number agreement in
15 But recall that a head higher than FA – T or Pred – in a sentence containing adjectival predi-
cation can agree in person with the subject, and in some languages these higher heads appear
morphologically joined with the adjective. See section 2.5.1 for discussion.
(35b). Apparently, then, T can agree in number with something in Spec, CP,
but it cannot agree in person with something in that position. This matches the
expectation of my theory: the wh-phrase never merges directly with a projection
of T, so the specific environment for agreement in first or second person does
not hold. T agreeing with Spec, CP is more adjective-like in this respect than
T agreeing with its own specifier. Similarly, when verbs agree with an operator
in Spec, CP in Passamaquoddy, they agree in the third person features singular
versus plural, animate versus inanimate, and proximate versus obviative – not
in first or second person (Bruening 2001:207).
3.3.2 Agreement in quirky subject constructions

Another relevant case to consider – and one that has been much discussed in the
recent literature – concerns quirky subject constructions in Icelandic. Icelandic
has many dyadic predicates in which the thematically higher argument (typically
the experiencer of a psych predicate) has a lexically specified case, often dative.
Since this argument is the highest, it is raised to Spec, TP to satisfy the EPP
feature of T. However, T is unable to agree with its quirky-case marked subject.
Under these conditions, T can (for many speakers, must) show a degree of
agreement with its nominative case object. (36a) is an example, where the plural
agreement on ‘was bored by’ is controlled by the plural nominative object, and
not by the dative subject (compare the singular agreement in the parallel (36b)).
(36) a. Henni leiddust þeir. (Taraldsen 1995:307)

her.dat was.bored.by.3pS they.nom
‘She was bored with them.’
b. Henni leið ist bókin sı́n. (Boeckx 2000:356)
her.dat was.bored.by.3sS book self’s
‘She finds her own book boring.’
This is another situation where T, a verbal functional head, agrees with some-
thing that is not its specifier, something that never merges with it. And in this
case too T can only agree with the nominative object in number, not in per-
son. Thus, examples like (37), which have first or second person nominative
pronouns, are ungrammatical.16
16 Bobaljik (to appear: note 27) notes that first and second person pronouns cannot be used as
objects of dative subject verbs even when the verb is used in a control infinitive. He therefore
suggests that the restriction in (37) has nothing to do with morphological agreement per se.
Instead, I tentatively follow Schütze (1997) and say that the same kind of syntactic agreement
takes place in infinitives as in tensed clauses in Icelandic, but the agreement is not spelled out
morphologically on the nonfinite verb.
(37) a. *Henni leiddumst við . (Taraldsen 1995:309)

her.dat was.bored.by.1pS we.nom
‘She was bored with us.’
b. *Ég veit að honum lı́kið þið . (Sigurdsson 2002:720)
I know that him.dat like.2pS you.nom.pl
‘I know that he likes you all.’
Again, this is just what my theory – which was designed primarily to capture
the fact that adjectives agree in number but not person – predicts should be the
case.
My account of these Icelandic facts is not radically different from theories that
have been proposed in the past. I submit, however, that it is a bit simpler and more
elegant. Many previous researchers, including Taraldsen (1995), Sigurð sson
(2000, 2002), Boeckx (2000), and Anagnostopoulou (2003) have developed
the intuition that the Icelandic verbs in (37) cannot undergo full agreement with
their objects because they have previously undergone a type of agreement with
the quirky case subject. One way to express this idea is that the dative subjects
are not completely devoid of ϕ-features; rather they bear a third person feature
and are unspecified for number only (Anagnostopoulou 2003:269–70). The
first time T searches, it finds the quirky case subject, establishes an Agree
relationship with it, and moves it to the Spec, TP position. Because the dative
subject is only partially specified for ϕ-features, however, only the person
feature of T is valued (as −1, −2); the number feature of T is left unspecified
by this first Agree relationship. Since it still has an unvalued feature, T can probe
again, this time locating the nominative object. Agreement with that object then
values the number feature of T, but cannot change the already valued person
feature. This explains why examples like (36) are grammatical, but examples
like (37) are not, according to this line of reasoning.
Although this line of research and mine account for the same facts, the
alternative theory rests squarely on one not very plausible assumption: that all
quirky-case-marked arguments are third person and unspecified for number. It
is far from obvious that this is true. If one considers dative-case arguments to
be simple NPs, then they seem able to have every possible value of person and
number. It is odd then to say that mér in (38) is third person and unmarked for
number when its interpretation is first person and singular.
(38) Mér lı́k-ar (*lı́k-a) bók-in. (Taraldsen 1995:310)

I.dat like-3sS like-1sS book-the.nom
‘I like the book.’
The alternative would be to analyze dative case expressions as PPs that con-
tain an NP with ordinary ϕ-features as their complement (see Markman 2005,
among others). On this second view, it makes sense that the PP as a whole is
unmarked for number, even though the NP it contains is singular or plural. But
it does not make sense that the PP is marked for third person. One would expect
such PPs to have no ϕ-features whatsoever. So there is no very plausible anal-
ysis of phrases in dative case that makes sense of the partial ϕ-feature specifi-
cation that the Boeckx–Sigurð sson–Anagnostopoulou analysis depends on.
In contrast, my account does not depend in any crucial way on the ϕ-features
of the dative subject; all that is required is that this have the minimal feature
needed to satisfy the EPP property of T in Icelandic. For me, person agree-
ment with the nominative object in (37) is not blocked because there is par-
tial agreement with the dative subject. Rather there is a direct configurational
explanation: the dative subject fills the specifier position but cannot be agreed
with (probably because of the activity condition; see chapter 5 for discussion).
Therefore, T agrees with a more remote NP. This precludes it from showing
person agreement, by the SCOPA. My theory is simpler in that it avoids making
an uncomfortable stipulation about the ϕ-features of dative expressions. It also
captures a similarity between these special verbal constructions in Icelandic and
the robust fact that adjectives do not show person agreement, which alternative
theories do not capture.17
These Icelandic examples vindicate the choice made in section 2.4.2 to say
that it is nonlocal agreement that cannot pick up +1 or +2 features. A plausible-
looking alternative was to stipulate that downward-probing agreement can
include agreement in first or second person, but upward-probing agreement
cannot.18 That would have been sufficient to explain why predicate adjectives
17 Sigurð sson (2000) mentions an empirical consideration that could weigh against an analysis like
mine. He shows that T can agree in person with a postverbal pronoun when the subject position
is occupied not by a dative subject, but by the expletive þ að in special copular constructions
like (i).
(i) það erum bara við .
It are.1pS only we.nom
‘It’s only us.’
Following the second suggestion in Sigurð sson’s note 25, I tentatively assume that þ að in (i) is
introduced higher than the T space, in a position like Spec, CP. This opens up the possibility that
the pronoun in (i) is really in Spec, TP, where it agrees in person with T, the T+verb combination
then moving to C. (This might not extend to Sigurð sson’s example (67), however, so I leave a
full analysis to future research.)
18 This alternative theory would also wrongly allow Cinque adjectives and raising adjectives to
agree with an NP in person.
do not agree in person, whereas verbs canonically do. The upward versus down-
ward proposal would also extend to the agreement with wh-phrases in Spec,
CP, shown in (35). But it does not extend to this Icelandic paradigm. Here
T is agreeing downward, with an object that is always within its c-command
domain, yet agreement is restricted to number. This shows that it is not whether
agreement is upward or downward that is crucial, but rather whether it is strictly
local, with F and its controller being merged together directly. Other examples
below confirm this.
The literature observes that this pattern of facts seems to be rather peculiar to
Icelandic. Other languages have similar-looking dative subject constructions,
but the verb shows full agreement with its nominative object. For instance, (39)
shows that a dative-subject verb like ‘like’ can show first or second person
agreement with its nominative object in Russian and Greek.
(39) a. Ja znaju, emu nravimsja my. (Russsian, Sigurð sson 2002:720)

I know him.dat like.1pS we.nom
‘I know that he likes us.’
b. Tu Petru tu aresis esi. (Greek, Anagnostopoulou 2003:91)
the Peter.gen cl.gen please.2sS you.nom
‘Peter likes you; you appeal to Peter.’
German and Kannada are also like Russian and Greek in this respect (on Kan-
nada, see Sridhar 1979, 1990). To the extent that my theory naturally explains
the Icelandic pattern, the existence of this full-agreement pattern is potentially
threatening to it.
Sigurð sson (2002), however, makes the crucial observation that dative subject
constructions in Icelandic differ from their Russian and German counterparts
in ways other than agreement. In particular, Icelandic quirky case subjects are
famous for being especially similar to canonical subjects in nominative case
when it comes to a range of syntactic properties, including word order, raising,
control, conjunction, and anaphora, as originally shown by Zaenen, Maling, and
Thráinsson (1985). In contrast, dative “subjects” are not nearly as subject-like
in German, but behave more like fronted topics. Even in Russian and Greek,
the oblique arguments have only a subset of the normal syntactic properties
of nominative subjects. For example, the dative experiencer can be elided in a
conjunction when it is parallel to a nominative subject in Icelandic, but not in
Greek or German.19
19 Oblique subjects do show some binding theoretic properties that are similar to those of ordinary
subjects in Greek and Russian. But see Franks 1995:254–5 for data showing that anteceding
(40) a. Hann segist vera saklaus en – hefur vı́st verið

he.nom says.refl to.be innocent but [dat] has apparently been
hjálpað ı́ prófinu.
helped on the.exam
‘He claims that he is innocent, but apparently has been helped
during the exam.’ (Icelandic, Zaenen et al. 1985:456–67)
b. *O Petros latrevi tin musiki ala dhen –
the Peter.nom adores.3sS the music.acc but not [gen]
aresun ta mathimatika.
please.3pS the mathematics.nom
‘Peter adores music but doesn’t like math.’ (Greek, Anagnostopoulou
2003:91)
This makes it very reasonable to suppose that the preverbal oblique expressions
in Greek, Russian, and German do not actually sit in Spec, TP. Rather, they could
sit in some higher position, such as Spec, TopicP, or they could be the specifier
of some decomposed functional category that has some of the properties of T
in English and Icelandic but not others. Now if subject properties show that
oblique NPs are not in the normal Spec, TP position in Greek, Russian, and
German, then it is perfectly possible that the nominative pronouns in examples
like (39) are in Spec, TP. If they are, then it is expected that T can bear +1
or +2 agreement. After this agreement is established, the V+T combination
could move on to a higher head position, such as C in verb-second clauses in
German, without destroying the agreement. One would certainly like to refine
this analysis, to learn more about exactly where fronted dative expressions
appear in different languages and why they appear there. But there is no strong
reason to fear that agreement in languages like Greek, German, and Russian will
be particularly problematic for my theory, even though they differ somewhat
from the Icelandic facts that support that theory so nicely.20
Another language that works like Icelandic – and so supports my theory – is
Gujarati, discussed by Bhatt (2005:801). (41a) shows that in perfective aspects
when the subject is marked ergative, the Gujarati verb can agree with the object,
anaphors and controlling participial clauses are not reliable diagnostics of being a subject in
Russian.
20 Similarly, the Indo-Aryan languages Maithili and Nepali allow full person agreement with
nominative “objects” in dative experiencer constructions, but Bickel and Yadava 2000 show
that the dative experiencers lack important subject properties in these languages (properties that
their Icelandic analogs have).
Taraldsen (1995) points out that transitive impersonal constructions in Italian are like quirky
subject constructions in Icelandic in not allowing the verb to agree with a first or second person
nominative object. Examples like this with featureless clitics or expletives in Spec, TP are
another possible source of evidence for the SCOPA.
even though it is marked accusative (unlike the better-known facts in Hindi).

On my theory, this suggests that the accusative particle ne is a pure case marker,
not an adposition which creates a barrier for agreement (and that agreement in
Gujarati is not dependent on case; see chapter 5). Crucially, though, this verb
is a participial form that agrees only in gender. Auxiliary verbs in Gujarati do
show person agreement as well as gender and number agreement; thus, ‘be’ has
a special second person form in agreement with the second person nominative
subject in (41b). The key example is (41c), in which there is an auxiliary along
with the perfective verb and the object is second person plural. Here the second
person form cho, seen in (41b), is impossible, and a default third person form
che is used instead.
(41) a. mEN tehmahri behEn-one bolawi. (Bhatt 2005:774)

I.erg your sisters.f-acc invited.f
‘I invited your sisters.’
b. tEhme āw-ya cho. (Bhatt 2005:801)
you.pl come-perf. m.pl be.pres.2pS
‘You have come.’
c. Mãi tam-ne mar-yā che. (Bhatt 2005:801)
I.erg you.pl-acc strike-perf. m.pl be.pres.3S
‘I have struck you.’
Gujarati is thus like Icelandic in allowing verbs to agree in gender but not in
person when the agreed-with NP is something other than the phrase in Spec,
TP.21
A non-Indo-European language in which this effect of the SCOPA can be
seen is the Muskogean language Chicasaw. Monroe and Gordon (1982) discuss
“dative subject” constructions like the following:
21 Bhatt took (41a) and (41c) from different sources, which apparently use different transliteration
systems for short and long vowels. This accounts for the different forms of the ergative first
person pronoun.
(41) leads Bhatt (2005:800) to posit what he calls the Person Generalization: “Dissociated
agreement does not involve Person,” where by “dissociated agreement” he means agreement
where the head does not license case on the agreed-with NP. Like my SCOPA, Bhatt’s Person
Generalization would cover the fact that adjectives do not agree in person, since adjectives are
not structural case licensers. The two generalizations are not identical, however. For example,
there is some reason to think that T in Icelandic licenses nominative case on the object in
examples like (36) (Taraldsen 1995, Schütze 1997), but it still cannot agree with the object in
person. Conversely, there are Cs in Kinande that agree with an operator in person even though
they do not assign case to those operators (see section 5.4.1).
(42) a. Hattak-at talowa’ [i-m]-alhkaniya-tok. (p. 90)

man-subj song 3-dat-forget-past
‘The man forgot the song.’
b. Talowa’ [a-m]-alhkaniya-tok. (p. 85)
song 1s-dat-forget-past
‘I forgot the song.’
c. Doris-at Claire-at [i-m]-ala-tok. (p. 99)
Doris-subj Claire-subj 3-dat-arrive-past
‘Doris gave birth to Claire.’
The overt subjects in (42a, c) do not bear overt dative case, but the “agree-
ment” prefix attached to the verb does. I tentatively assume that this prefix is
actually an incorporated oblique pronoun; when there is an overt “subject,” it
is really a topic licensed by binding this dative-case pronoun, which is the true
argument. Munro and Gordon give evidence from the switch reference system
of Chicasaw that the dative argument does count as the syntactic subject in
examples of this type (see also Davies 1986 on related Choctaw). Suppose then
that the dative pronoun is in Spec, TP before it cliticizes onto the verb. (42b)
shows that there is no additional agreement on T with this dative subject (the
verb is not a-m-alhkaniya-li-tok ‘1-DAT-forget-1sS-past’). The question, then,
is whether T agrees with the internal argument of the verb in these dative sub-
ject constructions in Chicasaw, the way it does in Icelandic. One cannot tell
simply by inspecting the morphology on the verb in (42), because third person
subject agreement is always Ø in Chicasaw. But evidence that it can agree is
the fact that the internal argument can bear nominative case in dative subject
constructions – and only in dative subject constructions – as shown in (42c). I
take this to be indirect evidence that this internal argument agrees with T, and
hence can get case from it. (This assumes that Chicasaw is a language in which
agreement is case-related in the sense of chapter 5. It also entails that there are
two sources for the SUBJ suffix in Chicasaw: it can be a topic marker, or it can
be the realization of case licensed by T; compare Japanese “nominative” ga.)
Imagine, then, how one would say ‘I forgot you’ in Chicasaw. The dative
subject would be in Spec, TP, and T would be left to search downward, inside VP
for something to agree with. But the only argument inside VP is second person,
and T cannot agree at a distance with such an argument, by the SCOPA. If these
assumptions are correct, one might expect ‘I forgot you’ to be ungrammatical in
Chicasaw, on a par with the badness of (37) in Icelandic. That is correct: Munro
and Gordon (1982) report that dative subject constructions are impossible if the
non-dative argument is first or second person. Speakers are thus driven to offer
complex paraphrases when asked how to say ‘I forgot you.’22
I conclude that there are languages both inside and outside the Indo-European
family in which the impossibility of nonlocal verbal agreement with first and
second person arguments can be observed in oblique subject constructions –
although further typological study of the matter would be most welcome.
3.3.3 Agreement in double object constructions

So far the focus has been on the agreement properties of T when something other
than what it agrees with occupies Spec, TP. Do similar considerations apply to
v, the other functional head with agreement-bearing potential associated with
verbs? Are there times when the Spec-head relationship between v and the
direct object is disrupted by some other syntactic element, with the result that
the verb can only show limited agreement with the direct object?
In fact, such cases exist and have been discussed in the recent literature, often
under the name of the Person Case Constraint (PCC), after Bonet 1991. The
effect shows up in double object and applicative constructions, in which there
is a benefactive/goal argument in the domain of v as well as the normal direct
object. In such configurations, v (in languages where v agrees at all) normally
shows full person and number agreement with the benefactive/goal argument.
This is expected, assuming that the benefactive/goal argument is structurally
the highest NP in the complement of vP (at the relevant level), hence the NP
that can most readily be attracted to Spec, vP (see section 2.3.1). When the goal
merges with the projection of v, agreement in first or second person as well as
number and gender is permitted by the SCOPA. The striking fact is that, in some
such languages, the verb can also show a degree of agreement with the second
internal argument, the theme/direct object. Crucially, however, this agreement
is for number and gender only, not for person.23
One language that fits this characterization is Classical Nahuatl (Launey
1981). (43) displays examples with a simple double object verb ‘give’. (43b)
22 The facts that Davies (1986) reports for closely related Choctaw are a little bit different, but
Choctaw also shows a difference between third person and non-third person arguments that can
plausibly be attributed to the SCOPA. In Choctaw, ‘I forgot you’ is sayable, but ‘you’ cannot
get nominative case and it triggers object agreement on the verb (Davies 1986:87, 92–93). This
shows that the second person argument cannot agree with T in the relevant construction, but
only with v.
23 For previous attempts to give a unified explanation for the agreement properties of dative sub-
ject constructions in Icelandic and double object constructions, see Boeckx 2000 and Anagnos-
topoulou 2003. For differences between their theories and mine, see discussion in the previous
section.
contains the third person object marker /k/ (orthographic c or qu) and the animate
plural object marker im. This form implies that both the goal and the theme
arguments are third person and that one or the other of them is plural. (43a)
contains the first person singular object marker nēch and the animate marker
plural im. This form can only be interpreted as having a first person singular
goal and an animate plural theme.
(43) a. Xi-nēch-im-maca huēhuèxōlô. (Launey 1981:174)

2sS.imp-1sO-pl-give turkeys
‘Give me some turkeys.’ (Not: ‘Give me to some turkeys.’)
b. Ni-qu-im-maca huēhuèxōlô in n-ocnı̄-uh. (Launey 1981:174)
1sS-3O-pl-give turkeys in 1sP-friend-poss
‘I gave some turkeys to my friend.’
(Also: Ni-qu-im-maca cē huèxōlotl in n-ocnı̄-huān. ‘I gave a turkey
to my friends’)
The constitutes a fairly standard PCC effect. The constraint has been formulated
as follows (Anagnostopoulou 2003:251, based on Bonet 1991:182; see also
Ormazabal and Romero 2007 for important empirical refinements).
(44) Person Case Constraint

Context: Ditransitives with phonologically weak direct and indirect objects.
Observation: If a direct and an indirect object co-occur, the direct object has
to be 3rd person (i.e., cannot be 1st or 2nd person).
This constraint as stated applies to weak pronouns, clitics, and null pronouns
related to agreement on the verb; I will be concerned only with agreement.
My theory of agreement contributes to an explanation of this pattern as
follows. In addition to assuming that the goal argument is the highest NP in the
complement of v (at the relevant level),24 let us also assume that v in Nahuatl
has exactly one EPP feature. This means that it can trigger the movement of one
and only one NP to remerge with it as a specifier. Since the goal NP is the higher
of the two internal arguments, it is the one that is attracted by v. The SCOPA
then implies that v can agree with the goal but not the theme in first or second
person. But the SCOPA does not constrain number and gender agreement; it
is permitted for v to agree at a distance with the theme NP that remains in its
complement for these features. The structure of (43a) then is (45).
24 I leave open the possibility that the goal starts lower in the VP and becomes the highest NP
as the result of some kind of movement process (“dative shift”), as in Larson 1988 and Baker
1997. Whether this prior movement occurs or not is orthogonal to the matter at hand.
(45) vP
NP vP
(you) NP v´
me v VP
[1,sg] [1,sg]
[an, pl] NP V´
<me> V NP
give turkeys
[an, pl]
I call this the two-and-a-half agreement phenomenon. This moniker expresses

the observation that Classical Nahuatl cannot show full agreement with three
distinct arguments (it cannot express things like ‘He gave you to me’), but it
can do a bit more than just agreeing with two arguments. One can also observe
a very similar two-and-a-half agreement effect in causative verbs derived from
transitive verb roots in Nahuatl (Launey 1981:191).
Another language with two-and-a-half agreement is Southern Tiwa, whose
complex agreement system is described in detail in Allen, Gardiner, Frantz, and
Perlmutter 1990. In this language, there are synthetic agreement forms for any
combination of subject and goal arguments. The relevant agreement forms also
vary with the gender and number of the theme, as shown in (46) (p. 335).
(46) a. Ka-‘u’u-wia-ban.
1sS/2sO/aO-baby-give-past
‘I gave you the baby.’
b. Kam-‘u’u-wia-ban.
1sS/2sO/bO-baby-give-past
‘I gave you the babies.’
c. Kow-keuap-wia-ban.
1sS/2sO/cO-shoe-give-past
‘I gave you (the) shoes.’
These particular examples all have a first person singular subject and a second
person singular goal, both expressed in the agreement prefix. In addition, (46b)
has a plural theme in contrast to the singular theme of (46a), and this also
affects the verbal prefix. Similarly, (46c) has a plural inanimate theme whereas
(46b) has a plural animate theme, and this difference too is registered in the verb
prefix. So there is a degree of agreement with the theme argument of ditransitive

verbs in Southern Tiwa. However, just as in Nahuatl, one can never have first
or second person agreement with the theme argument of a triadic verb: there
is no prefix form for ‘I gave you to him’ or ‘He gave me to you,’ for example.
The explanation is the same as for Nahuatl: the higher goal argument can move
to Spec, vP and trigger +1 or +2 agreement, but the lower theme argument
cannot. The v head can agree with the theme argument at a distance, but only
in number and gender – the same limited repertoire of features as adjectives
agree in.
Some Bantu languages may also provide instances of two-and-a-half agree-
ment, such as Shambala, whose object prefixes are described by Duranti (1979).
Duranti mentions that reflexive verb forms in Shambala cannot bear any (other)
object prefix (p. 36). I take this to be tentative evidence that the object prefixes in
this language are true instances of agreement on the active v head, not cliticized
pronouns derived by some kind of movement (see note 25, and also section
5.7.1 on object agreement in Bantu more generally). Nevertheless, the presence
of two internal arguments can be registered on the verb in some instances, as
in (47) (Duranti 1979:36).
(47) a. A-za-m-ni-et-e-a.
she-past-him-me-bring-appl-fv
‘She has brought him to me.’ Not: ‘She has brought me to him.’
b. A-i-wa-mw-et-e-e.
she-past-them-him-bring-appl-fv
‘She brought them to him.’ (marked number for theme)
b. Na-i-mw-itang-i-a.
I-it-him-call-appl-fv
‘I call it (the meeting) for him.’ (marked gender on theme)
Shambala thus permits agreement for more than just the subject and one internal
argument. But agreement with two internal arguments is not unrestricted. For
example, (47a) cannot be interpreted as having first person agreement with the
verb’s theme argument and third person agreement with the goal argument,
but only the other way around. The second imaginable interpretation is also
ruled out when the order of the object prefixes is reversed (*a-za-ni-mw-et-e-a,
intended to mean ‘she has brought me to him’ (Duranti 1979:36)). The only
way to render this meaning in Shambala is to project the goal argument as a PP,
the ϕ-features of which are inaccessible to v. Only then can v show first person
agreement with the theme argument in the presence of a goal:
(48) A-za-ni-et-a kwa yeye. (Duranti 1979:36)

she-past-me-bring-fv to him
‘She has brought me to him.’
This is a classic PCC effect, where the theme object must be third person when
the verb is ditransitive ((47)), but not when it is monotransitive ((48)). Put
another way, Shambala allows full agreement with the higher NP in the verb
phrase (the goal), and partial agreement – agreement in number and gender but
not in person – with the lower NP. Under this description, Shambala is very
much like Nahuatl and Southern Tiwa. Duranti (1979:36) also mentions that
first and second person object prefixes can never appear together on the same
verb in Shambala. This too is just what we would expect, given that v can only
show person agreement with the one argument that it attracts to its specifier
position.
Unlike restricted agreement in dative subject constructions, the PCC phe-
nomenon has been studied typologically and is known to be very general. Only
a handful of languages have come to light in which first or second person theme
arguments can be expressed on the verb in the presence of a goal NP (for exam-
ple, Haspelmath (2004) presents four possible counterexamples: Kabardian,
Lakhota, Haya, and Noon). Moreover, some of the known counterexamples
seem to involve clitics or incorporated elements rather than true agreement.25 I
thus make the bold claim, which I do not know to be false, that the PCC effect
25 One heuristic for telling whether a morpheme that expresses the object on a verb is a moved clitic
pronoun or a true agreement might be to see if the morpheme in question can appear on verbs
in the passive or in other nonactive voices (detransitivized reflexive, reciprocal, or anticausative
verbs). Typically it is only the active version of v – instances of v that assign an agent role to an
NP in their specifier – that is a legitimate bearer of object agreement (the agreement-oriented
version of Burzio’s Generalization). A verb in a nonactive voice is thus expected not to be able
to agree with a theme or goal object, even when the base verb is a ditransitive, so there is still one
internal argument that could in principle be agreed with. Indeed, in languages with fairly clear
instances of object agreement, such as Mohawk, Mayali, Mapudungun, and Nahuatl, there is no
object agreement with the nonsubject argument of a nonactive ditransitive verb (e.g., Nahuatl:
Ti-(*c)-mac-o in xōchi-tl (2sS-(*3sO)-give-PASS DET flower) ‘you were given the flowers’;
Launey 1981:175). In contrast, in languages that clearly have object clitics, such as the Romance
languages and Greek, an object clitic can perfectly well attach to the passive of a ditransitive
verb (e.g., Spanish Este libro le fue dado (a Juan) ‘This book him.DAT-was given to Juan’).
This heuristic can be applied to the minimal contrast between Shambala, which respects the
PCC, and Haya, which does not (Duranti 1979, Hyman and Duranti 1982). Duranti states that
Haya allows object prefixes to attach to passive and reflexive verb forms (Duranti 1979:43),
but Shambala does not (Duranti 1979:36). That suggests that object prefixes have the status of
clitics in Haya, but that of object agreement morphemes in Shambala. This encourages me in
my belief that the PCC is exceptionless for true agreement, though not for cliticization (perhaps
it holds in that domain only for the functionalist reasons that Haspelmath 2004 discusses).
is exceptionless in the area of agreement – although good tests to distinguish

instances of true agreement from instances of cliticization and weak pronouns
are clearly needed to substantiate this.
This analysis of two-and-a-half agreement involves an enrichment of the
theory. For examples like (43), (46), and (47), I must say that a single functional
head (v) can have multiple specifications for the same ϕ-feature. For example,
the v in the Nahuatl example diagrammed in (45) is both singular and plural.
(Perhaps it is also simultaneously first person and third person, if third person
is a real feature and not just a default value.) It contrasts with the form in (49),
which is doubly plural, expressing the plurality of both internal arguments
explicitly.
(49) An-tech-im-maca’. (Launey 1981:391)

2pS-1pO-pl-give.
‘You all give them to us.’
The active v in Nahuatl thus probes the phase twice in search for something to
agree with, finding the goal the first time and the theme NP the second time.
Not all languages allow v to probe twice in this way. For example, Mohawk
and Mayali are typologically similar to Nahuatl and Southern Tiwa (Baker
1996) and they have systems of gender and number marking. Yet in ditransitive
constructions they show full agreement with the agent and the goal, but no
additional gender/number agreement with the theme. Similarly, Chichewa is a
Bantu language that has the same sort of gender/number/noun-class system as
Shambala, but Chichewa only allows a single object marker that expresses the
goal, not a second one that expresses an additional theme argument. (50) gives
a minimal pair illustrating this difference.
(50) a. *Alenje a-ku-zı́-wá-phik-il-a (zı́túmbûwa anyâni).

2.hunters 2S-pres-8O-2O-cook-appl-fv 8.pancakes 2.baboons
‘The hunters are cooking them for them.’ (Chichewa, Mchombo 2004:82)
(also * . . .wa-zi . . .)
b. A-ya-i-dik-i-a. (Shambala, Duranti 1979:37)
she-them-it-cook-appl-fv
‘She cooks them for it.’
It is conceivable that these differences could be derived somehow from differ-

ences in the ϕ-feature systems of the languages in question or from general-
izations about how they are spelled out morphologically – but that does not
seem likely given (50).26 Probably there is a parametric difference at work that
determines whether v can search for something to agree with more than once
or not. This can be expressed as follows:
(51) A particular functional head F can search the phase for something to agree
with:
i. zero times, or
ii. one time, or
iii. two (or more?) times.
In essence, (51) says that a grammar can stipulate whether a given functional
head is a probe or not, although not (I assume) the particular feature structure
of the probe (contra Bejar 2003, Rezac 2003). Nahuatl, Southern Tiwa, and
Shambala are languages that adopt the (51iii) option; Mohawk, Mayali, and
Chichewa adopt the (51ii) option. The option in (51i) is for languages like
English and Chinese, which have transitive v heads, but where those heads do
not participate in agreement (see section 2.2.2). Note that even with this enrich-
ment my theory falls well short of stipulating for each grammatical category
which features it can and cannot agree with, the brute-force approach that is
unexplanatory and allows for more crosslinguistic variation than we actually
observe (see section 1.2). It is still predictable in this revised system which NPs
v will agree with and in which features. There is an arbitrary component to
the theory (does a language have two or two-and-a-half agreement?) but also
a principled component (all languages have two-and-a-half agreement rather
than three agreement).27
The general formulation of (51) suggests that it should apply to T too, allow-
ing T sometimes to agree twice as opposed to once or not at all. Suppose for
simplicity that this were to happen in a language or construction in which there
was no agreement-bearing v head. Then we would expect to see instances of
one-and-a-half agreement, where the finite verb agrees fully with the highest
argument and in number and gender only with a lower argument (assuming T
has only one EPP feature, so only the highest argument can remerge with the
26 Variation like that seen between Chichewa and Shambala in (50) is even found internal to the
Lakhota language (another language that Haspelmath (2004) says contains PCC violations).
Rood and Taylor (1996:sec. 9.3.8.3) state that “Some speakers inflect for both direct and bene-
factive objects, but others reject these forms as meaningless.”
27 Note that saying whether v is a single probe or a double probe is not the same as saying whether
v has one EPP feature or two. Nahuatl and Southern Tiwa are languages in which v is a double
probe but has only one EPP feature, resulting in two-and-a-half agreement. A language in which
v was a double probe and had two EPP features could be a language that showed full person
agreement with three arguments. I do not know that there are any such languages.
projection of T). In fact, this situation is attested in Southern Tiwa. Southern

Tiwa has a distinct one-and-a-half agreement paradigm that it uses both for the
passives of ‘give’-type verbs and for unaccusative motion verbs like ‘come’
when they take a goal argument. (52) shows that, in addition to full person
agreement with the goal subject, a passive verb also varies with the number and
gender of the theme argument.
(52) a. Hliawrade-ba in-khwian-wia-che-ban. (Allen et al. 1990:334–6)

lady-instr 1sS/A-dog-give-pass-past
‘I was given a dog by the lady.’
b. Im-khwian-wia-che-ban.
1sS/b-dog-give-pass-past
‘I was given (the) dogs.’
c. Iw-keuap-wia-che-ban.
1sS/c-shoe-give-pass-past
‘I was given (the) shoes.’
Exactly the same agreement pattern is found on an intransitive verb of the

unaccusative class when a goal argument is included. Thus, a sentence like
‘The dog came to me’ has the same prefix in as (52a), showing agreement with
the theme ‘dog’ in number and gender as well as with the goal ‘me’ in person,
number, and gender (Allen et al. 1990:356). These forms are, however, distinct
from the prefix found in a simple transitive like ‘I saw the dog’ (ti-mu-ban)
or the ‘The dog saw me’ (no prefix; passive required). Moreover, there is no
form for ‘You came to me’ or ‘I came to you.’ In these respects, the one-and-a-
half agreement seen on nonactive (v-less) clauses in Southern Tiwa is perfectly
parallel to the two-and-a-half agreement found in active clauses in the same
language.28
28 There is a theory of agreement in ergative languages that says that T agrees twice and v does
not agree at all (Bok-Bennema 1991, Bobaljik and Branigan 2006). In section 5.10.2.2, I adopt
such a theory for Nez Perce. Combining that approach with the reasoning outlined here, one
might expect that ergative languages of this type would also show a kind of one-and-a-half
agreement, agreeing with their objects in number and gender only. In fact, person agreement is
rather impoverished in Nez Perce in a way that might be attributable to the SCOPA: the language
shows overt agreement with third person arguments only, and does not encode a contrast between
first and second person agreement (see section 5.10.2.2 for examples and references).
In Inuit, on the other hand, there is full person agreement for both the subject and the object
of a transitive clause, so nothing akin to a PCC effect holds of T in that language. But Bittner
(1994) shows that agreed-with objects in Inuit must take wide scope with respect to sentential
operators (see also Bok-Bennema 1991). That could mean that T in Inuit has a second EPP
feature, so that objects as well as subjects are attracted to Spec, TP. In that case, full person
agreement with both arguments is expected (compare note 27).
In contrast, the passives of ditransitive verbs do not show agreement with a

plural theme object in Nahuatl. At least, Launey (1981) does not mention the
possibility, and there are no forms like (53) in his texts:29
(53) *T-im-mac-o huēhuèxōlô. (Launey 1981:175)

2sS-pl-give-pass turkeys
‘You were given some turkeys.’
(OK is Ti-mac-o in xōchitl ‘You were given the flowers’,
without plural im.)
(53) shows that im is not used to spell out an extra plural feature on T, nor is
there any other morpheme that does so. Apparently T in Nahuatl is specified
as probing once ((51ii)), whereas T in Southern Tiwa and v in Nahuatl are
specified for probing twice ((51iii)). How many times a given functional head
can agree thus varies independently both within and across languages. Of course
we already knew that this was true with respect to whether a functional head
agrees zero or one times: for example, v in Russian agrees zero times, in contrast
to T in Russian and v in Chichewa, which agree once each. So (51) correctly
captures the kind of head-by-head variation that one sees in this domain.30
Summing up this material, we see that, in a variety of languages, triadic verbs
allow a limited degree of object agreement with their third arguments. Specifi-
cally, such verbs can agree with a third person object, sometimes reflecting
differences in gender and number for that argument, but they cannot agree with
a first or second person object. My theory offers an explanation for this phe-
nomenon that relates it to the fact that adjectives can agree with an NP in number
and gender but not in first or second person. The common theme is that first and
second person agreement requires that the agreed-with NP merge directly with
the agreement-bearing functional head. In ditransitive constructions, this sort
of agreement is ruled out for the theme object because the specifier position of
the functional head is already occupied by the higher goal object; in adjectival
constructions it is ruled out by the inherent nature of adjectival projections,
which do not take specifiers by definition. The reason why the agreed-with
29 This was evaluated by typing Launey’s prose Nahuatl texts into a computer file, searching it
electronically for the relevant combinations of letters, and then doing a morphological analysis
of the hits to see if any of them had the morphological structure in question.
30 The ungrammaticality of (53) also confirms that im really is an object agreement morpheme
that spells out active v, not an agreement on some distinct third head like Appl (as in Adger and
Harbour’s 2005 analysis of Kiowa). If im were agreement on Appl, one would expect to find
it in passivized applicatives as well, contrary to fact. More generally, in Adger and Harbour’s
view there is no reason to expect that agreement on the Appl head can only be agreement for
number and gender but not for person.
argument does not merge with the functional category is different in the two
structures, but the consequences for agreement are the same.
3.3.4 Agreement into an embedded clause

So far the examples of nonlocal agreement with verbs have been examples in
which something else occupies the specifier of the functional category that the
agreed-with NP might otherwise have moved to. Thus, quirky dative subjects in
Icelandic and Gujarati block lower arguments from moving to Spec, TP; goals,
applied objects, and causees block lower arguments from moving to Spec, vP
in Nahuatl, Southern Tiwa, and Shambala. In all such cases, the verb can agree
with the lower argument in number and gender only, not in person.
Another relevant kind of case is the long distance agreement (LDA) that
verbs sometimes show with arguments of an embedded clause. A set of such
cases has come to light in the recent literature, in which the matrix verb agrees
with a constituent of the embedded clause which moves to the periphery of
the embedded clause (either overtly or covertly) but cannot move out of that
clause because of language-specific conditions on movement; see, for example,
Polinsky and Potsdam 2001 on LDA in Tsez. Since there is clear evidence that
the agreed-with NP is trapped in the embedded clause, we know that it does
not move to the specifier position of the relevant FV associated with the matrix
verb (although the clause that contains it might). The structures in question can
be schematized abstractly as in (54), where FV can stand for either T or v.
(54)
a. FV verb [CP C [TP . . . . NP . . . . ]] Agreement out by phase condition
b. FV verb [CP NPi C [TP . . . ti . . .]] LDA agreement configuration
c. *NPi FV verb [CP (ti ) C [TP . . . ti . . .]] * by conditions on movement
The SCOPA now predicts that the configuration in (54b) might allow agreement
in number and gender, but should not allow agreement in first or second person.
This prediction is consistent with the known facts. The well-known case of
Tsez is not directly relevant to this prediction, for the simple reason that verbs in
Tsez do not agree with an NP in person even in simple monoclausal structures.
As a result, no difference between ordinary verbal agreement and long-distance
verbal agreement would be observable in this language. Comparable data from
LDA in the Nigerian language Lokaa are relevant, however (taken from my own
research with Alex Iwara). Lokaa usually has subject-verb-object word order,
but in gerundive constructions the object fronts to a left-peripheral specifier
position (see Baker 2005 for detailed analysis). When this happens, the matrix
verb agrees in gender with the fronted object of the gerundive verb, not with
the gerundive phrase as a whole. (55a) shows an example where the gerundive
verb is intransitive, and the matrix verbs take the same noun class prefix as the
gerundive verb, in agreement with it. (55b) is a contrasting example in which
the gerundive verb is transitive. Its object ‘fish’ must front to come before the
gerund, and this time the matrix verbs take the same noun class prefix as this
object, not the noun class prefix of the nominalized verb:
(55) a. Ke-pa.a.la. ke-tum ke.-tawa.
ger/5-fly 5S-be.very 5S-be.difficult
‘Flying is very difficult.’
b. [E.-sau ke.-de.i] e-tum e.-tawa.
7-fish ger/5-buy 7S-be.very 7S-be.difficult
‘Buying fish is very difficult.’ (* . . .ke-tum ke-tawa)
Now Lokaa has distinct first and second person subject agreement prefixes (see
Iwara 1982 for a complete paradigm); (56) gives an example with the first
person singular form.
(56) Ami n-tum n-dam.
I 1sS-be.very 1sS-be.big
‘I am very big.’
The crucial question, then, is what happens when the object of a gerundive verb
is first person singular and the gerund as a whole is in subject position. The
distinctive prediction of my theory of agreement is that first person singular
agreement on the matrix verbs should be ruled out, because the first person NP
does not merge directly with T; only a larger phrase that contains it does. The
ungrammaticality of (57) shows that this prediction is correct.31
(57) *[Min ke-funna] n-tum n-tawa.
me 5-surprise 1sS-be.very 1sS-be.difficult
‘Surprising me is very difficult.’
Another language in which this effect can be seen clearly is Basque. Etxe-
pare (2006) documents the existence of LDA in number in certain substandard
varieties of Basque. In particular, he shows that certain verbs that select for non-
finite clauses with a controlled subject permit the matrix auxiliary verb to show
object agreement in number with the object of the lower verb. (58) reproduces
two examples of this:
31 The partial agreement structure in which the matrix verbs show singular animate agreement o
also seems to be out (see discussion below), as is agreement with the gerund as a whole. Alex
Iwara (personal communication) says that speakers would simply avoid using a gerund in this
case, using a subjunctive clause instead (‘It would be very difficult that one would surprise me.’).
(58) a. Nobela erromantiko-ak irakur-tze-a gusta-tzen za-(izk)-io.

novel romantic-abs.pl read-noml-abs.sg like-hab 3(p)A-aux-3sD
‘He/she likes to read romantic novels.’ (p. 1)
b. Liburu-ak eros-te-a erabaki d-(it)-u.
book-abs.pl buy-noml-abs.sg decide aux-3sE-3(p)A
‘He decided to buy books.’ (p. 33)
However, Etxepare also observes that this sort of LDA can only add extra
number features to the auxiliary in the matrix clause; it cannot add person
features. The examples in (59) contrast minimally with those in (58), but they
are ungrammatical.
(59) a. *Zu bertan ikus-te-a gusta-tzen zatzaizkio. (p. 39)

you.abs.sg there see-noml-abs.sg like-hab aux.2A.3sD
‘He/she likes to see you there.’
b. *Zu gonbida-tze-a baztertu zaituzte. (p. 39)
you-abs.sg invite-noml-abs.sg refused aux.3pE.2sA
‘He/she refused to invite you.’
Basque then is another language that confirms the prediction that true LDA can
only be agreement in number and gender.
Along with Tsez, the other well-known cases of LDA come from Algonquian
languages, including Bruening’s (2001) discussion of Passamaquoddy. At first
glance, this language seems problematic for my prediction, because the argu-
ment of the embedded clause appears to trigger any sort of agreement on the
matrix verb:
(60) a. N-kosiciy-a-k nuhuw-ok muwinuw-ok keq kis-temu-htit.

1-know.ta-dir-3.pl three-3.pl bear-3.pl what perf-eat-3.pl.conj
‘I know (of them) what the three bears ate.’ (p. 259) (number agreement)
b. Kosiciy-ul kis-ankuweht-uwon atomupil-ol Piyel naka Susehp
know.ta-1/2 perf-sell-2.conj car-inan.pl Piyel and Susehp
mil-osk-opon-il.
give-2.conj.inv-pret-part.pl (p. 266) (person agreement)
‘I know (of you) that you sold the cars Piyel and Sushep gave you.’
In (60a), the matrix verb ‘know’ shows plural object agreement with ‘three
bears,’ the topicalized subject of its clausal complement. That example is con-
sistent with my theory. (60b), however, looks problematic: here the matrix verb
‘know’ shows second person singular agreement with the subject of the embed-
ded clause, contrary to my prediction.
Bruening (2001) shows, however, that there is an important difference

between instances of LDA with third person, number, and animacy features
such as (60a) and putative instances that involve first and second person, such
as (60b). LDA with third person arguments is sensitive to islands. For example,
the matrix verb in (61) cannot agree with ‘Piyel and Sushep’ in the embedded
clause, because that NP is further embedded inside a relative clause.
(61) *N-kosiciy-a-k kis-ankuweht-uwon atomupil-ol Piyel naka Susehp

1-know.ta-dir-3.pl perf-sell-2.conj car-inan.pl Piyel and Susehp
mil-osk-opon-il.
give-2.conj.inv-pret-part.pl (p. 266)
‘I know (of them) that you sold the cars Piyel and Sushep gave you.’
This is more or less expected. Since ‘Piyel and Sushep’ is contained inside a
complex NP it cannot move to the edge of the clausal complement of ‘know’
via topicalization (even covertly). Therefore the v associated with the matrix
verb cannot access it, by the phase condition on agreement. The surprise is that
the matrix verb can show +2 agreement with an argument that is generated
inside a relative clause:
(62) Kosiciy-ul eli Susehp kis-ankuweht-aq atomupil mil-ot-opon.

know.ta-1/2 C Susehp perf-sell-3.conj car give-2.conj-pret
‘I know of you that Susehp sold the car you gave him.’ (p. 278)
Bruening concludes from this sort of data that “LDA” with first and second
person arguments is not the same phenomenon as LDA with third person argu-
ments. The third person cases are true instances of LDA, analyzed just like
Polinsky and Potsdam’s analysis of Tsez in (54), which I have been following.
But the first and second person cases are instances of proleptic objects, in which
a second internal argument is generated in the domain of the matrix verb ‘know’
and stands in a binding relationship (not a movement relationship) with some
NP in the embedded clause. This extra argument of the matrix verb is very much
like the phrase of you in the English gloss of (62). Since binding relationships
are not restricted by island conditions, the example in (62) is possible. Since
(61) is not possible, third person nominals cannot be base generated as objects
in the matrix clause, for some reason.
Bruening himself gives a rather complex explanation for this person asym-
metry in Passamaquoddy. His account hinges on the person-sensitive voice
system of Algonquian languages, in which first and second person objects trig-
ger inverse forms. Bruening analyzes this inverse system as a voice alternation
Conclusion 107
that is similar to the passive. Then if LDA with first and second person nominals
followed exactly the same derivation as third person nominals, it would require
the first or second person NP to move from the Topic position at the edge of
the lower clause to the Spec, TP position of the matrix clause, an instance of
improper movement (A-bar movement feeding A-movement). My theory offers
a simpler and more general account of the difference, not dependent on contro-
versial properties of the Algonquian voice-and-agreement system. My theory of
agreement states that an Agree relationship between T or v and something stuck
in the lower Spec, CP can only be agreement for number and gender (which
includes animacy and obviation in Algonquian), not for person. If one has full
agreement with a +1 or +2 expression, that NP must be a true argument of the
matrix clause, and hence able to move to Spec vP (or Spec TP).32 Therefore,
Passamoquoddy also supports my theory, initial appearances notwithstanding. I
predict that there cannot be full LDA with a first or second person argument, and
in fact putative cases of LDA with such arguments turn out to have a different
syntactic structure than those with third person arguments.
3.4 Conclusion
In the course of this chapter, we have found many reasons to believe that adjec-
tival agreement and verbal agreement are two instances of fundamentally the
32 Bruening does, however, have an explanation for a fact that does not follow from my account –
the fact that generating a third person argument as a proleptic object of the matrix verb seems
not to be possible in Passamaquoddy (so (61) is bad under any analysis). I have no formal
account for this; I can only appeal to a kind of economy principle, which says that the true LDA
derivation blocks the proleptic object derivation when both are consistent with the superficial
morphological facts.
It is interesting to compare Bruening’s analysis of Passamaquoddy with Branigan and
MacKenzie’s (2002) analysis of Innu-Aimun (IA). Branigan and MacKenzie’s description dif-
fers from Bruening’s on two points. First, they claim that constituents of the lower clause can
move into the matrix clause in IA, unlike in Passamaquoddy. Second, they do not present any
syntactic differences between LDA with third person NPs and LDA with first and second person
NPs. If they are correct and IA really differs from Passamaquoddy in these respects, the two
Algonquian languages constitute an interesting minimal pair that supports my theory. Given
that IA allows NPs to move into the matrix clause, movement of a +1 or +2 NP into Spec,
vP is possible, not blocked by language-specific constraints. When movement to this position
takes place, object agreement in +1 or +2 with an NP that originated in the embedded clause is
expected. (See also Bobaljik and Wurmbrand 2005:845–50, who show that (i) Itelmen matrix
verbs can sometimes agree in person with an argument of an embedded infinitive, but (ii) this
only happens when there is scopal evidence that the agreed-with argument has moved into the
matrix clause.)
same phenomenon. Both adjectives and verbs can agree with a c-commanding
NP, both can agree with a c-commanded NP, and both are restricted to gender-
number agreement when forced to agree at a distance. The only major difference
is that verbal heads can also agree with NPs in first and second person when they
merge directly with the NP, whereas this possibility is not available to adjectival
heads. But the fact that this is not possible follows from the general theory of
categories: adjectival projections by definition do not permit specifiers whereas
verbal projections do.33
The comparison between verbs and adjectives also gives strong reason to
think that the restrictions on person agreement are syntactic rather than mor-
phological in nature. If it were only adjectives that could not agree in first or
second person, one might think that some person feature was missing in their
lexical entry or happened not to be spelled out by the morphological rules
of the language. But in this chapter we have seen that verbs also sometimes
fail to show person agreement, and for verbs this clearly cannot be attributed
to general properties of the lexical category or its morphological paradigm.
Rather, whether there is person agreement or not clearly depends on the
33 A very different sort of argument that adjectival agreement is distinct from verbal agreement
can be found in Wechsler and Zlatić 2003:ch. 3. They point out that there are a few nouns in
Serbian/Croatian that trigger feminine singular agreement on attributive adjectives and demon-
stratives but trigger neuter plural agreement on auxiliaries and participial verbs. One such noun
is deca ‘children’:
(i) Ta dobr-a deca su došl-a. (p. 5)
that.f.sg good-f.sg children aux.3pS come.past.ptpl-n.pl
‘Those good children came.’
Wechsler and Zlatić conclude, following Pollard and Sag (1994), that nouns have two separate
sets of agreement features: concord features, which are accessed by a relation of concord
agreement that involves adjectives, and index features, which are accessed by a distinct relation
of index agreement that involves verbs (and pronouns in discourse). If they are right about this,
then there are two distinct kinds of agreement and not only one.
However, I find two things striking about this case. The first is how very rare are nouns
that have putative mismatches between concord features and index features; Wechsler and
Zlatić (2003:50) list only four. The second is how similar all of these nouns look to another
class of collective nouns in Serbian/Croatian (nouns that end in the morpheme ad), which
Wechsler and Zlatić treat as feminine singular nouns that have collective meanings, and thus
can trigger semantic/pragmatic agreement on the predicate (and on subsequent pronouns), just
as nouns like committee, government and band do in some varieties of English (see section 1.5).
The only difference reported is that the collective noun unuč-ad ‘grandchildren’ can trigger
singular or plural agreement on the finite auxiliary, whereas deca ‘children’ can apparently only
trigger plural agreement – a difference that Wechsler and Zlatić (2003:76) acknowledge to be
mysterious even in their account. I therefore conclude that nouns like deca in (i) also illustrate
a type of semantic/pragmatic agreement, and do not motivate a distinction between two kinds
of grammatical agreement, one that involves verbs and another that involves adjectives.
Conclusion 109
syntactic structure of the clause – whether there is a dative subject or not,

whether it is a monotransitive clause or a ditransitive clause, whether the agreed-
with phrase is trapped in a lower clause or not. It is a strength of my theory
that the syntactic characterization of when person agreement fails on verbs also
explains the general lack of person agreement on adjectives and nouns.
Before closing this chapter, I must mention an important residual issue that
this material raises, first articulated to me by David Pesetsky (personal com-
munication). The question is what happens to a structure when agreement in
person with a +1 or +2 nominal is ruled out by the SCOPA?
Descriptively speaking, there are two possible outcomes. One outcome is
that the grammar might make do with partial agreement: the functional head
in question agrees with the first or second person nominal as best it can and
leaves it at that. This is always the result when the agreeing head is adjectival,
as far as I know. Thus, Swahili, Hindi, Spanish, Arabic, Mayali, and Tariana all
allow sentences like ‘I am tall’ and ‘You are good’, where the adjective agrees
with the subject in gender and number but not person. Partial agreement also
seems to be good enough for agreement with wh-expressions in Spec, CP in
nonstandard English. Full person agreement with I is ruled out in this dialect,
as discussed in section 3.3.1, but number agreement with we seems better, on a
par with third person plural agreement, as shown in (63).
(63) a. ?the people who Clark are hoping will come

b. I, who Clark is/*am hoping will come
c. ?we who Clark are hoping will come
The second possible outcome when full agreement with a first or second
person argument is impossible is that the syntax crashes, the structure in ques-
tion being rendered ineffable. That seems to be the case for the double object
constructions discussed in section 3.3.3. Thus, structures like ‘She sent me
to him’ in Nahuatl, Southern Tiwa, and Shambala do not give rise to agree-
ment in number and gender/animacy with the first person theme; rather they
are unsayable (Mohawk), forcing some other structure to be used (e.g., one
with the goal expressed as an oblique PP, as in (48) from Shambala). Long dis-
tance agreement with a first person object in a gerund construction also leads
to ineffability in Lokaa (note 31). Finally, dative subject constructions with
a first or second person object are reported to be ineffable in Chicasaw and
Icelandic.
An interesting theoretical question then arises as to why partial agreement
with a first or second person is tolerated in some of these constructions and not
others. Unfortunately, I have no definitive answer to offer at this point. I simply
note that the difference does not seem to correlate neatly with the verb–adjective
distinction. While awaiting further insight into the matter, I assume that partial
agreement is generally allowed as far as the syntax of agreement is concerned,
and that the structures that crash do so because of other factors.34
34 The licensing needs of the null pronominal pro might be a relevant additional factor. It could
be significant that the types of agreement that tolerate partial agreement (adjectives and verbs
agreeing with a wh-phrase in Spec, CP) also do not license pro-drop. In contrast, the types of
agreement that do not tolerate partial agreement typically do license pro-drop. This is true for
agreement with subjects in Chicasaw and Lokaa, and for agreement with objects in Nahuatl,
Southern Tiwa, Shambala, Basque, and the Algonquian languages. So the following general-
ization might be valid:
(i) F can show agreement with a +1 or +2 NP in number and gender but not +1 or +2
if and only if F would not license pro-drop of NP.
If this generalization holds up, it suggests that full agreement is required in some configurations
not for the sake of agreement per se, but rather for the sake of the agreed-with NP, so that its
features will be recoverable when pro-drop occurs. Of course, working this out more fully would
require a better understanding of how agreement licenses pro-drop, and what the relationship is
between an overt pronoun and its pro-dropped cousin. But the direction seems promising.
4 Explaining the restriction on
person agreement
Most of the ingredients of the theory of agreement in the previous two chap-
ters are either fairly standard or generalizations of notions that are fairly stan-
dard. The notable exception is the Structural Condition on Person Agreement
(SCOPA), repeated in (1).
(1) F can agree with XP in +1 or +2 only if a projection of F merges with a +1

or +2 element and F projects.
This highly specific condition is not part of standard theory. Section 3.3 gave
a range of evidence showing that this condition correctly characterizes when
verbs can and cannot agree in person with their arguments, in addition to ruling
out first and second person agreement on adjectives quite generally. But as it
stands, (1) is merely a descriptive generalization. The purpose of this chapter
is to explore the SCOPA more thoroughly.
To accomplish this, I begin by investigating when person agreement is pos-
sible on categories other than nouns, verbs, and adjectives, confirming that the
SCOPA is descriptively correct (section 4.1). I then go on to derive the SCOPA
from more fundamental principles which go to the heart of what it is for any
linguistic expression to be first or second person. The first step in this derivation
is to distinguish the relationship of Agree that holds between a functional head
and a nearby NP from the sort of agreement that holds between operators and
the variables that they bind (section 4.2). I claim that first and second person
agreement belongs properly to operator-variable agreement and not to Agree
per se. It is not particularly surprising, then, that the two are subject to dif-
ferent formal conditions. The second step is to argue that all first and second
person elements – including pronouns – must be bound by the closest operator
of the relevant type (section 4.3). In this respect, first and second person ele-
ments differ from third person elements, which are not subject to any such local
binding condition. The final step is to show that locality conditions in general
apply to heads in a particularly strict fashion, such that any intervening head
blocks a local relationship, regardless of its features (section 4.4). Putting these
111
112 Explaining the restriction on person agreement
intermediate results together gives the overall result that any head that comes
between a first or second person head and the first or second person NP that it
is dependent on blocks the person-licensing relationship – a condition that is
equivalent to (1). The SCOPA is thus derived from a principle that regulates the
distribution of first and second person pronouns plus a general law concerning
how locality relationships apply to heads.
4.1 Person agreement and other categories

In previous chapters, I considered only the possibility of person agreement on
the lexical categories – or, more accurately, on the functional categories that
most immediately dominate those lexical categories. The primary result from
chapter 2 was that the functional categories directly associated with verbs,
Tense and v, can bear person agreement with subjects and objects, whereas the
functional categories associated with nouns and adjectives cannot bear person
agreement. Chapter 3 added the nuance that even Tense and v cannot agree
with an NP in person if their projection does not merge directly with that NP.
One thus finds only number and gender agreement with wh-phrases in Spec,
CP in an English dialect, with lower NPs in oblique subject constructions in
some languages, with second objects in double object constructions, and with
phrases trapped inside an embedded clause in LDA constructions. Overall,
person agreement is relatively fragile, and conditions need to be just right for
it to take place, as expressed in the SCOPA.
When we widen the picture to consider a fuller range of categories, we
find some other instances of person agreement. These arise in the range of
environments that the SCOPA predicts they should – with one very instructive
exception.
4.1.1 Adpositions
Consider first the category of adposition. The traditional generative view
since Jackendoff (1977) has been that this is a lexical category, but Baker
(2003a:appendix) argues that it is actually a functional category. As such, it is
a possible host for agreement. Moreover, adpositions normally select NP/DP
complements, which can bear ϕ-features, including the features +1 and +2.
Since adpositions merge directly with such complements, the SCOPA predicts
that adpositions should be like verbs and unlike the other lexical categories
in manifesting person agreement. Quite a few languages confirm this expec-
tation, including the examples in (2) from the Mayan language Jacaltec. (2a)
Person agreement and other categories 113
shows agreement with the third person object of the preposition; (2b) shows
contrasting first person agreement on a similar preposition.
(2) a. y-ul te’ n̈ah (Craig 1977:110)

3E-in the house
‘in the house’
b. X-Ø-colwa ix w-in̈. (Craig 1977:102)
asp-3A-help she 1sE-to
‘She gave me a hand.’
(3) shows similar examples from postpositions in Abaza (O’Herin 2002:54):
(3) a. awə y a-mʃj taxj

that 3s.i-after
‘after that’
b. (sara) s-pngə
I 1s-at
‘at my house, by me’
The SCOPA does, however, predict that Ps will not be able to agree in person
more remotely with (for example) the subject of its predication. This follows
as long as PPs do not have nominal specifiers, and only become predicates by
adjoining to a verbal projection (see Baker 2003a:appendix). Since the P does
not merge directly with the subject of the predication, P cannot agree in person
with that subject (see also Baker 1996:401–4). (4) shows that this is true for
Nahuatl, a language in which predicates of all other categories do seem to agree
in person with their subjects (see section 2.5.1).
(4) Petla-pan icpal-pan ni-ca’. (Launey 1981:384)

mat-on chair-on 1sS-be
‘I am on a mat, on a chair.’ (i.e. I govern.)
Not: *Ni-petla-pan, n-icpal-pan ni-ca’.
Adpositions are interesting because they provide our first examples of a cat-
egory agreeing in person with its complement rather than its specifier. Tense, v,
and FA all take nonnominal complements that do not have ϕ-features. There-
fore if they also merge with something that does have ϕ-features – as Tense
and v often do – that is a second merge, creating a specifier. As a result, person
agreement is in practice usually agreement with a specifier. In part because
of facts like these, generative theories of the early 1990s attributed a special
role to the specifier–head relationship in the theory of agreement, claiming that
full agreement occurred only in specifier–head relationships (see Koopman
2006 for a recent defense). But the existence of person agreement inside PPs
weighs against this, showing that one does find full person agreement between
heads and complements whenever conditions are right for it. This suggests that
Minimalist sensibilities are correct, in that there is no fundamental theoretical
difference between complements and specifiers (Chomsky 1995). It is only a
matter of what merges with what that is important for the theory of agreement,
as in other domains.1
4.1.2 Determiners
Next, let us consider determiners (Ds) and similar functional heads in the nom-
inal system. The expectations of my theory for this category are similar to those
just reviewed for the category P: since D is a functional category that merges
directly with a complement that has ϕ-features, it should be able to agree with
that complement. And of course articles, demonstratives, and quantifiers do
agree with their NP complements in many languages, including Spanish:
(5) a. el libro (the book.m.sg)
b. los libros (the books.m.pl)
c. la mesa (the table.f.sg)
d. las mesas (the tables.f.pl)
The definite article agrees with its complement only in number and gender,
whereas the SCOPA predicts that person agreement should be possible as well.
But this simply reflects the fact that common nouns in Spanish are always third
person, never first or second person. Therefore, the complement of this kind
of determiner never has a +1 or +2 feature for the determiner to agree with.
I assume that (5) is a case of total agreement, not partial agreement, but the
person feature is trivial, because articles and quantifiers do not usually take
pronouns as arguments (*the we, *every I, *some you).
Some Bantu languages have D-like heads that can bear person agreement as
well as number and gender agreement. An example is onke ‘all’ in Zulu:2
1 A proponent of Spec–head agreement can, of course, insist that the first person pronouns move
to Spec, PP or some other specifier in the extended projection of P in examples like (2b) and (3b).
I do not attempt to construct arguments that this is not the case. I simply point out that my theory
does not need this extra assumption, for which there is no obvious evidence in the languages
cited.
2 Wechsler and Zlatić (2003:15) mention the similar case of ote ‘all’ in Swahili, pointed out to them
by Manfred Krifka. They must treat these cases as exceptional: they are instances of concord
agreement (as opposed to index agreement) that nevertheless show person, a feature not normally
included in the concord feature bundle. Since my theory does not distinguish concord from true
agreement, these examples are not exceptional but expected.
On the fact that the agreed-with complement of onke ‘all’ comes before it in (6), see
section 5.5.1.
(6) Thina s-onke si-fik-ile. (Doke 1963:94)

we 1p-all 1pS-arrive-perf
‘We have all arrived.’
(compare: b-onke (cl2-all) ‘all of them’; izinkomo z-onke
(cattle.5.pl 5.pl-all) ‘all the cattle’)
This makes sense given that onke ‘all’ is different from most simple deter-
miners in that it selects DP complements, including pronouns, as well as NP
complements. Selectional properties like these need to be stated in the grammar
anyway; compare English all (the) men vs. every (*the) man. Once it is stipu-
lated that ‘all’ can select a first person pronoun then it follows immediately from
the SCOPA that ‘all’ can agree in person with that pronoun. Other determiner-
like elements that can inflect for person as well as gender and number in Zulu
are dwa ‘only, alone’ and bili ‘both’ (Doke 1963:94–6).
Another sort of determiner that clearly shows person agreement in many
languages is the possessive determiner, which merges with a specifier (the
semantic possessor) as well as a complement to form structures like [DP John
DPoss [NP book]] (Abney 1987). There are in general no selectional restrictions
on the features of the possessor; in particular, first and second person pronouns
can perfectly well function as possessors. Given this, the SCOPA permits D
to show person agreement with the possessor. And indeed possessive deter-
miners do agree with the possessor in many languages. (7) shows particularly
clear examples from Mapudungun, where the possessive determiner is real-
ized as a separate word from the noun (note that the possessor itself is often
pro-dropped).3
(7) a. mi füchá ruka (Smeets 1989:129–30)

2s.poss big house
‘your big house’
b. yiñ füchá ruka
1p.poss big house
‘our big house’
3 A question that arises here is why doesn’t the complement NP of the possessive determiner
block it from agreeing with its specifier by the intervention condition, given that the complement
also has ϕ-features. There are two possibilities. One is that the possessive determiner actually
agrees with its complement as well as with its specifier in (7), but this is not spelled out since
Mapudungun does not have gender distinctions and rarely spells out number agreement. The
second possibility is that the intervention condition needs to be defined in terms of m-command
rather than c-command, so that the specifier and the complement come out as equally close to
the head of the projection (see note 20 to chapter 2). I leave open which of these is correct.
c. (inché) ñi ruka

I 1s.poss house
‘my house’
In many languages, nouns themselves are said to agree with their possessors.
I am committed to the view that this commonplace description is not quite
accurate, and it is really always the possessive determiner that agrees with the
possessor. This can, however, be disguised on the surface if the noun incor-
porates into or merges with the determiner to form a single word at PF, as is
often the case. Nahuatl is an instructive case in point. Possessive agreement is
realized as a prefix attached to the noun root in this language:
(8) a. no-te-uh (compare: te-tl) (Launey 1981:90)
1sP-rock-poss rock-nsf
‘my rock(s)’ ‘a/the rock’ *esta forma recuerda mucho
a las lenguas campa!
b. am-e-uh (compare: e-tl)
2pP-bean-poss bean-nsf
‘your bean(s)’ ‘(the) bean(s)’
However, there is also a different suffix on the noun when it is possessed as

compared to when it is not possessed (uh/Ø vs. tl and its allomorphs). I take this
suffix to be the possessive determiner, with the noun root having incorporated
into it. As a result, the agreement with the possessor – which is clearly associated
with the D node, not the N node in languages like Mapudungun where the two
remain separate – comes to appear in the same word as the noun in Nahuatl,
where N and D are not separate. This is perfectly parallel to the familiar fact
that subject agreement is on Tense rather than the verb, but it often comes to
appear on the verb at PF because Tense and the verb combine either by head
movement or by PF merger. Many other languages, like Mohawk, do not show
any overt difference between possessor suffixes and absolute suffixes (see (9)),
but I take them to be syntactically like Nahuatl in these respects.
(9) a. ake-núhs-a’ (cf. ka-núhs-a’)
1sP-house-nsf Ns-house-nsf
‘my house’ ‘a/the house.’
b. sa-sahét-a’ (cf. o-sahét-a’)
2sP-bean-nsf Ns-bean-nsf
‘your beans’ ‘(the) beans’
4.1.3 Degree heads

So far, we have seen that adpositions and determiners can show agreement in first
and second person features, much as the verbal heads T and v can. In contrast,
the SCOPA makes the opposite prediction for degree heads, the functional
heads that appear with adjectives. Degree heads appear above AP (and FA P)
and below PredP to form predicative structures like [PredP NP Pred [DegP Deg
[(FA ) [AP]]]] (see Baker 2003a:214). Since the complement of a degree head is
an adjective phrase, and adjective phrases do not have ϕ-features, they will not
trigger +1 or +2 agreement on a Degree head. Moreover, inasmuch as Degree
is an adjectival functional head, it is like FA and A in not licensing a specifier.
Therefore, Degree heads cannot show +1 or +2 agreement with a specifier
either. Degree heads do not merge directly with any ϕ-feature bearing element;
therefore the SCOPA implies that they should never show first or second person
agreement.
I do not have much crosslinguistic evidence that bears on this prediction.
Most of the languages that I know do not have clear instances of the Degree
category, and in some of those that do, the degree heads do not happen to be
probes for agreement (e.g., Spanish muy ‘very’ and tan ‘so,’ which do not inflect
at all). But what little evidence I do have supports the prediction. In Russian, for
example, the degree head tak ‘so’ does manifest gender and number agreement
with the subject, as shown in (10a). This shows that it is a probe for agreement.
But tak does not have a distinct first person form, as shown in (10b) (Nerea
Madariaga, personal communication).
(10) a. Ivan tak-oj xoroš-ij; Nadja tak-aja xoroš-aja.

Ivan.m.sg so-m.sg good-m.sg Nadia so-f.sg good-f.sg
‘Ivan is so good; Nadia is so good.’
b. Ja tak-oj xoroš-ij.
I so-m.sg good-m.sg
‘I am so good (male speaker).’
This confirms that the degree head cannot agree in person at a distance, as
expected. The SCOPA thus correctly predicts that P and D can bear person
agreement morphology, but Degree cannot.
4.1.4 Complementizers: a favorable case

The last well-established type of functional category to discuss is complemen-
tizer and complementizer-like heads. The complement of C is TP, which does
not have ϕ-features. Thus, C is expected to show first or second person agree-
ment if and only if there is an operator in Spec, CP that has these features. In all
other cases, the C should either agree in number and gender only, or not agree
at all. Complementizers in English, the Romance languages, and most others
take the no-agreement option, which is trivially consistent with my theory.
Some other languages do have inflected complementizer-like heads, however.

One example is the particle that introduces a relative clause in the Niger-Congo
language Lokaa. This agrees in number and gender with the NP that the relative
clause modifies:
(11) a. Yı̀-sòwá. f-á. úbı̀ ó.ó.-dè.?i (my field notes)

12-pot 12-rel Ubi 1S-buy
‘the pot that Ubi bought’
b. kà-kó.ò k-á. é-fém é-jı́ı́
14-pig 14-rel 7-crocodile 7S-eat
‘the pig that the crocodile ate’
c. lı̀-kó.ò j-á. m-fém n-jı́ı́
11-pig 11-rel 8-crocodile 8S-eat
‘the pigs that the crocodiles ate’
Relative clauses can also be adjoined to pronouns in Lokaa. When this happens,
the complementizer agrees with the pronoun in number and gender, but not in
person. Thus, the first person pronoun in (12c) triggers the same class 1 (animate,
singular) agreement on the relative complementizer as the third person pronoun
in (12a). (12b) contains a first person plural pronoun; this triggers plural marking
on the complementizer (class 2), but not person marking.
(12) a. Si: eá. w-á. o-bi liman o:-yeni. (Iwara 1982)

despise him 1-rel 1S-not money 1S.neg-have
‘Despise him who has no money.’
b. Si: moon b-á. yo-bi liman yo:-yeni.
despise us 2-rel 1pS-not money 1pS.neg-have
‘Despise us who have no money.’
c. ami w-á. m-biila
I 1-rel 1sS-be.dark
‘I who am dark skinned’
These facts follow from the SCOPA as long as we can convince ourselves
that there is no relative operator that bears first person features in Spec, RelP
in (12b) or (12c). If there were, we would expect first person agreement on C
with that operator to be possible. In fact, there probably is a relative operator in
these constructions, for standard reasons having to do with island phenomena
and other diagnostics of A-bar movement relationships. Moreover, the relative
operator probably does count as being first person: notice that it triggers first
person subject agreement on the verbs inside the relative clause in (12b) and
(12c). However, the relative operator is probably not in Spec, RelP. The category
I have marked as REL in (11) and (12) is really an associative head, also found
in noun-noun modification constructions in which there is no operator, such as
(13).
(13) kà-kóò k-á ó.:-bòl; lı̀-kó.ò j-á ó.:-bòl (fieldnotes)

14-pig 14-rel 1-chief 11-pigs 11-rel 1-chief
‘the chief’s pig’ ‘the chief’s pigs’
(13) suggests that the relative particle does not have a +wh-feature that triggers
operator movement. Assuming that the very same particle is in the relative
clauses in (11) and (12), then there is presumably no operator movement to
Spec, RelP there either. The operator movement must be triggered by a distinct
head, which we may call C, giving a structure like (14) for the example in (12c).
(14) DP
DPi RelP
I Rel CP
[1,sg,cl1] [sg,cl1]
[*1] Opi C’
[1,sg,cl1]
C TP
Ø ti T’
[1,sg,cl1]
T VP
[1,sg,cl1]
be.dark
Given (14), the SCOPA correctly describes the fact that the relative particle can
agree with the head of the relative clause (or the operator below it) in number
and gender but not in person.
Overall, then, we see that there is much evidence for the SCOPA as a descrip-
tive generalization that characterizes when agreement in person features is
possible. The categories T, v, P, and D can show person agreement with
their complements or specifiers; the categories FA , Degree, and certain
complementizer-like particles cannot show person agreement because they do
not have specifiers and their complements do not have ϕ-features.
4.1.5 Complementizers: problematic cases

Before I go on to derive the SCOPA from more general considerations, I must
acknowledge that there are other configurations in which complementizers show
a surprising kind of agreement, even in first and second person features. I know
of at least two. One well-known case is the complementizer agreement found
with subjects in some varieties of Continental West Germanic. (15) shows an
unproblematic case where the complementizer agrees with the lower subject in
number (from Carstens 2003:393).
(15) Kvinden dan die boeken te diere zyn. (West Flemish)

I.find that.pl the books too expensive are
‘I find that those books are too expensive.’
But the complementizer can also agree with first or second person subjects
in some of these varieties, even though the subjects do not raise to Spec, CP
(Carstens 2003:393):
(16) Kpeinzen dank (ik) morgen goa-n. (West Flemish)

I.think that.1s I tomorrow go-1sS
‘I think that I’ll go tomorrow.’
This agreement is unexpected given the SCOPA.

An even more striking problem for the SCOPA is posed by another kind of
complementizer agreement in the Niger-Congo languages Lokaa and Kinande.
In these languages, the complement of a verb of speech is headed by a comple-
mentizer that seems to agree with the matrix subject. The complementizer can
even agree with the matrix subject in person, as shown by (18b) from Kinande.
(17) a. Ubı̀ ó.-jà.yı̀ ò-bı́ nè. yá-yàà máà. (Lokaa)

Ubi 1S-say 1-that fut 2S-come here
‘Ubi said that they will come here.’
b. Yá-jà.y yá-bı́/*ó.-bı́ Ubı̀ nè. ó.-yàà máà.
2S-say 2-that/*1-that Ubi fut 1S-come here
‘They said that Ubi will come here.’
(18) a. Mo-ba-nyi-bw-ire ba-ti Kambale mo-a-gul-ire eritunda. (Kinande)
aff-2S-1sO-tell-ext 2-that Kambale aff-1S-buy-ext fruit.5
‘They told me that Kambale bought fruit.’
b. Mo-n-a-layir-ire Kambale in-di a-gul-e amatunda.
aff-1sS-t-convince-ext Kambale.1 1s-that 1S-buy-sbjn fruits.6
‘I convinced Kambale that he should buy fruits.’
So there are attestations of apparently nonlocal agreement in person on com-

plementizers, both with the complementizer probing downward (West Flemish)
and with the complementizer probing upward (Kinande).
These examples cease to be problematic if there is a first person operator
in Spec, CP in (16) and (18b). In section 4.3, I present evidence that there are
such operators, not only in West Flemish and Kinande, but more generally. In
Operator-variable agreement and Agree 121
fact, these operators turn out to be essential to defining what it means for any
linguistic element – pronoun or agreeing head – to be first or second person,
providing the key to a deeper understanding of the SCOPA. Let us then turn to
the project of understanding the fundamental nature of first and second person
elements, returning to these cases of unexpected person agreement on C in the
course of the discussion.
4.2 Operator-variable agreement and Agree

The first step is to acknowledge that there is at least one other type of agreement
at work in natural language, apart from the relationship between functional
heads and nearby NPs that I have been studying throughout this book. This
second sort of agreement is the not-so-often-discussed agreement between an
operator and the variable that it binds.4 Operator–variable agreement is illus-
trated in (19).
(19) a. Every boyk hopes that hek (*shek , *theyk , *Ik ) will pass the test.
b. Every girlk hopes that someone will hire herk (*himk , *themk , *mek ).
c. Only the Yankeesk think that theyk (*hek , *wek ) will win the championship.
When a pronoun is interpreted as being bound by a masculine singular quantifier

like every boy, the pronoun must be realized as a masculine singular form he,
him, or his, as in (19a). In contrast, when a pronoun is bound by a feminine
singular quantifier like every girl, it must be a feminine singular form (her in
(19b)), and when it is bound by a plural quantifier like only the Yankees, it must
be third person plural (they in (19c)). This then is also a kind of agreement in a
broad sense.
Operator–variable agreement is clearly a somewhat different phenomenon
from the agreement that applies to functional heads, however, because it does
not obey the same formal restrictions. For example, we know that a functional
head can agree with an NP only if the two are contained in the same tensed
CP (the phase condition). That constraint does not apply to operator–variable
agreement: in all of the examples in (19) the bound pronoun is properly con-
tained in the finite embedded CP, a phase that does not contain the quantified
4 I thank Ken Safir (personal communication) for calling my attention to the phenomenon of
operator–variable agreement, and for suggesting that I couch the ideas developed in this chapter
in those terms. He is not to be blamed, however, for how I have acted on his suggestion.
The agreement between an operator and a bound pronoun may need to be distinguished from
the agreement that holds between any pronoun and its antecedent in the discourse. The latter is
probably an instance of semantic/pragmatic agreement of the type put aside in this book (see
section 1.5).
antecedent. Nevertheless, agreement between the pronoun and its binder is not
only possible, but required.
Second, we know that in English and many other languages a functional head
can agree with an NP only if a case–valuation relationship holds between them
(the activity condition; see section 5.3 for a new formulation). But no such
constraint restricts operator–variable agreement. For example, the variable in
(19b) is in accusative case and the quantified expression that binds it is in
nominative case. Nevertheless, ϕ-feature agreement takes place. This kind of
agreement, then, is not dependent on case-valuation the way that other forms
of agreement are.
Third, a functional head can agree with a given NP Y only if there is no
other NP Z that is closer to the functional head than Y is (the intervention
condition). (20) shows that this condition also does not apply to operator-
variable agreement.
(20) Every girlk told every boy about herk troubles with herk parents.
The two pronouns her in (20) agree with the subject every girl in the feature fem-
inine despite the fact that there is a closer quantified NP every boy in the object
position that has a different value for the same feature (masculine). Agreement
is clearly with the NP that binds the pronoun, regardless of its syntactic position,
for this type of agreement. That is quite different from agreement on functional
heads, where relative syntactic position is paramount.
Even the c-command condition does not apply to operator–variable agree-
ment, since there are examples like (21) in which the quantified NP does not
c-command the pronoun that is referentially dependent on it (see Safir 2004:34).
(21) Someone in every cityk loves itsk (*theirk , *hisk ) weather.
In short, none of the syntactic conditions that are characteristic of the Agree
relation between a functional head and an agreed-with NP apply to the agree-
ment relation between a variable and the operator that binds it. The two must
be fundamentally different relationships, although both result in the features
that are inherent on one expression also being realized on another expression.
I state the second kind of agreement as in (22).
(22) If variable X is referentially dependent on operator Y (directly or indirectly),
then X has the same ϕ-features as Y.
The qualification “directly or indirectly” in (22) is motivated by examples like

(23). Here the first pronoun in the embedded clause is directly dependent on the
matrix subject and agrees with it in ϕ-features (singular, masculine) in accor-
dance with (22). The second pronoun his also agrees in ϕ-features. It is probably
Operator-variable agreement and Agree 123
not directly dependent on the matrix subject, but rather takes the embedded sub-
ject he as its immediate antecedent. But it is dependent on something that is
itself dependent on the quantified matrix subject. This indirect dependency also
involves agreement in ϕ-features.
(23) Every boyk says that hek finished hisk (*herk , *theirk , *ourk ) homework.
More precisely, the term “referentially dependent” in (22) can be defined

inductively: X is referentially dependent on Y if (i) X is immediately ref-
erentially dependent on Y, or (ii) there is some Z such that X is immedi-
ately referentially dependent on Z and Z is referentially dependent on Y.
(The equivalence of direct and indirect dependencies plays a role in what
follows.)
Exactly which ϕ-features participate in operator–variable agreement? The
examples that I have used so far illustrate that number features (it vs. them)
and gender features (he vs. she) are involved. But first and second person
features also participate in this kind of agreement, as shown by the examples
in (24), a type of example that has been discussed in unpublished work by
Irene Heim (see also Kratzer 1998, von Stechow 2003, Schlenker 2005, and
others).
(24) a. Only Ik finished myk (*hisk , *herk , *ourk ) homework.
(For x = I, x finished x’s homework;
For all x, x = I, not: x finished x’s homework.)
b. All of youk are hoping that youk (*Ik , *shek , *theyk ) will win the single’s
title.
(For all x, x one of you, x hopes x will win the single’s title.)
In these examples, the pronoun can be interpreted as a variable bound by the

highest subject if and only if it agrees with that subject in person as well as
in number. So person features are full participants in the system of operator–
variable agreement.
Given this, let us adopt the conjecture in (25).
(25) Agree never puts the features +1 or +2 on a head; +1 and +2 features are
always the result of operator–variable agreement.
In other words, I claim that first and second person features are only involved
in the operator–variable agreement system. The sort of agreement that applies
to ordinary functional heads like T and v and FA and P is always “partial”
agreement, initially copying number and gender values only. How then can
there ever be first or second person features on T or v? The answer, I claim, is
that Agree creates a context where operator–variable agreement can apply, as
expressed in (26).
(26) If F agrees with XP, then F counts as a variable that referentially depends on
XP.
Therefore, in addition to copying number and gender features from XP to

F, Agree makes F dependent on XP. Many linguists have had the intuition
that there is something pronominal about agreement morphemes. For example,
there is no doubt that agreement often makes use of the same stock of features
as independent pronouns do, and the two are often historically related (see,
for example, Givó.n 1976, Bresnan and Mchombo 1987, and references cited
there). (26) is a way of capturing formally an aspect of this intuition. Then if an
agreeing functional head F is dependent on the XP it agrees with, and if that XP
is a pronoun that is dependent on some higher operator Z, then F is (indirectly)
dependent on Z and must agree with it by (22). If Z is first or second person,
then F will also be first or second person. That is how an agreeing head can
come to have a first or second person feature, not directly as a result of Agree,
but indirectly as a result of (22) and (26).
What is gained by this shift in perspective? Only this: now we have the
beginnings of an idea as to why agreement in first and second person might
pattern differently from agreement in other features, since it is really part of
another kind of agreement. Furthermore, attributing first and second person
agreement to operator–variable agreement gives us an idea of what we need to
look at next – the special operators that bind first and second person pronouns.
4.3 A locality condition on first and second person variables

A deep answer to the question of why first and second person features are
subject to a stricter condition on agreement than other features would be one
that related this fact to other, more basic truths about first and second person
elements. Let us consider then the behavior of first and second person pronouns,
taking them to be more basic bearers of the features +1 and +2 than agreeing
functional heads.
4.3.1 The Person Licensing Condition

What is the most basic truth about first and second person pronouns? At first
glance, it is the idea that (putting aside direct quotation) first person pronouns
always include the person who spoke (or wrote) the sentence in their reference,
and second person pronouns always include the person to whom the sentence
was addressed. This received wisdom has been influentially articulated in the
work of the philosopher David Kaplan (1989). But the recent literature has
A locality condition on first and second person variables 125
brought to light two kinds of counterexamples to this simple characterization,

which suggest that there is a more specific syntactic licensing condition at work.
First, Kayne’s (2000:154) example in (27) shows a situation in which a first
person pronoun cannot refer to the speaker; this is ruled out when the subject
of the sentence is a third person noun phrase that itself refers to the speaker.
The pronoun me cannot have any interpretation in this example, and is simply
ungrammatical.
(27) The man who is talking to you wants you to give him/*me some money.
Conversely, Schlenker’s (2003) example in (28) from Amharic shows a situation

in which a first person pronoun (the pro-dropped subject of the embedded
clause) can be interpreted as someone other than the speaker – namely the
subject of the matrix clause.
(28) ˇon ˇə gna nə -ññ y l-all.

John hero be.pf-1sO 3m.say-aux.3m
‘John says that he is a hero.’ (lit. ‘John says that I be a hero.’)
(Note that Schlenker reviews evidence that the embedded clause in (28) is not a
direct quotation.) However, a first person pronoun in Amharic can only refer to
someone other than the speaker in a very specific situation, when it is embedded
in the complement of the verb ‘to say.’ Thus, there is need for a nontrivial
grammatical condition that regulates the occurrence and interpretation of first
and second person pronouns.
Building on important insights of Schlenker (2003, 2005), Sigurð sson (2005),
and Anand and Nevins (2004), I propose that there are special empty categories
introduced at the CP level which designate the speaker and the addressee of the
sentence, as stated in (29).5
(29) a. All matrix clauses and certain embedded clauses have two special null
arguments generated within the CP projection, one designated S (for
speaker) and the other A (for addressee).
5 The specific syntax-oriented implementation that I adopt is most like “Theory II” of Schlenker
2005, which Schlenker attributes particularly to his dissertation, unpublished work by Irene Heim,
and von Stechow 2003. Another precedent for this idea is the logophoric operators associated
with Spec, CP, which lie at the heart of syntactic explanations of logophoricity in West African
languages in Koopman and Sportiche 1989, Adesola 2004, and related work. See Safir 2005 for
a dissenting view, on which first and second person pronouns are not bound by logophoric-like
operators. For clarification of the relationship between first and second person licensing and the
licensing of logophoric pronouns, see section 4.3.3.
b. In the absence of an overriding control relationship, S designates the

person who produced the CP and A designates the person who the CP
was addressed to.
The idea is that certain key aspects of the “point of view” from which the clause
is interpreted are expressed by syntactically represented elements.6 All uses of a
first person pronoun must then be interpreted by being bound by the S operator,
and all uses of a second person pronoun must be interpreted by being bound by
the A operator, in accordance with (30).
(30) The Person Licensing Condition (PLC)
a. A DP/NP is first person only if it is locally bound by the closest
c-commanding S or by another element that is first person.
b. A DP/NP is second person only if it is locally bound by the closest
c-commanding A or by another element that is itself second person.
c. Otherwise, a DP/NP is third person.
One minor pay-off of the PLC is that it explains the fact that ordinary non-
pronominal NPs are never first or second person, even when they refer to the
speaker or hearer. This can be seen in (31), where ordinary DP subjects cannot
trigger distinctive first or second person agreement on the verb be.
(31) a. The man who is talking to you is/*am hoping to get some money.
b. Sorry honey, but Daddy is/*am too tired to play with you tonight.
c. Your honor was/*were misinformed by the counsel for the defense.
It is a general fact that pronouns can be bound by operators in, for example,
resumptive pronoun constructions, but lexical NPs generally cannot be (unless
they contain a pronominal element; see Aoun and Choueiri 2000 for data and
discussion):
(32) a. ?Johnk , whok we all wonder whether hek will actually show up, . . .
b. *Johnk , whok we all wonder whether the boyk will actually show up, . . .
Now if lexical NPs cannot be bound by operators ((32b)) and expressions can
only be first or second person if they are bound by a special operator ((30a–b)),
then it follows that lexical NPs cannot be first or second person. They always
fall under the elsewhere case in (30c). This explains the facts in (31).7
6 There may be other such elements too, related to the location where the CP was spoken and the
time at which it was spoken, relevant to the interpretation of deictic expressions like now and
here. I leave this matter open. Unlike Schlenker (2003), I do not package all of these parameters
together as a single semantic operator, because there are contexts in which first person shifts but
second person does not in Slave (see the discussion of (44)).
7 Unlike English and French, Spanish allows sentences like Las mujeres trabajamos mucho (the
women work.1pS much) ‘We women work a lot.’ I assume that the possibility of such a sentence
The PLC might also be used to explain the fact that a first person pronoun
cannot refer to the speaker in examples like Kayne’s (27), repeated as (33).
(33) The man who is talking to you wants you to give him/*me some money.
As a matrix clause, the sentence as a whole has an S operator that refers to

the speaker. What is special about this example is that it also has a matrix
subject that is a definite description that refers to the speaker without being
dependent on S. Suppose that we represent this fact by giving both S and the
matrix subject the same index i. Now suppose that we want an interpretation
in which a pronoun in the indirect object position of the embedded clause also
refers to the same individual i. If it is assigned this index, then it is bound by S,
but it is not locally bound by S; the matrix subject is a closer NP that bears the
same index. Therefore, the pronoun in question cannot be first person according
to (30a); rather, it must be third person by default:8
(34) [CP Si [TP [NPi The man who is talking to you] wants you to give himi /*mei
money]]
Of course if the matrix subject refers not to the speaker, but rather to some
other guy who happens to be talking to the addressee at the same time, then
using me in the embedded clause to refer to the speaker is grammatical and
using him would be at best very strange. In that case, the matrix subject has
a different index, say k, and S is the closest binder of the pronoun. Then first
person features are licensed on the pronoun by (30a). Moreover, if the indexing
remains as in (34) but the matrix subject is replaced by I, then too a first person
pronoun is licensed in the lower clause:
(35) [CP Si [TP [NPi I] want you to give mei /*himi some money]]
is related to the fact that Spanish is a pro-drop language whereas English and French are not.
Thus, one need not say that the verb ‘work’ is agreeing with a first person feature borne by the DP
‘the women’ in Spanish. Rather, we can say that the verb ‘work’ agrees with a null first person
pronoun ‘we’, and ‘the women’ is a kind of topic or adjunct loosely related to the sentence by
some kind of aboutness relation.
8 Raffaella Zanuttini (personal communication) points out to me that imperatives are different in
this respect: (i) is fine, with a proper name that refers to the addressee apparently binding a second
person pronoun.
(i) Nicholas Baker wash your (*his) hands!
A possible analysis of this is that imperatives, as a consequence of their special semantics and
pragmatics, license an A operator that is lower in the clause, in the TP space rather than at the
CP level. If that is correct, then the subject in (i) does not intervene between the A operator and
the genitive pronoun, and the pronoun can have second person features.
(35) is like (34) in that the embedded pronoun cannot be anchored directly to
the S at the head of the matrix clause, there being a closer antecedent. But unlike
in (34), the closer antecedent in (35) is a first person pronoun which is itself
locally bound to the S in the proper way. This is enough to license a first person
feature on the embedded pronoun. The intuition is that first and second person
pronouns need to be bound by the designated operator, either directly, or else
indirectly through a sequence of local binding relationships to a pronoun that is
so bound. On this analysis, (34) illustrates a kind of strong crossover effect that
arises with first person pronouns. The ungrammaticality of (34) is thus akin to
the ungrammaticality of (36), with the indexing indicated.
(36) *[CP Whoi does [TP [NPi he] want you to give ti some money]]
(compare: [CP Whoi C [TP ti wants you to give [NPi him] some money]])
The trace in (36) must be locally bound by the operator in Spec, CP, and this is
violated if the matrix subject is also interpreted as having this index. This, then,
is a significant similarity between first and second person pronouns and other,
less-controversial operator-variable structures, supporting the treatment of first
and second person pronouns given in (29) and (30). This analysis of (33) raises
some serious theoretical issues concerning indices and their interpretations,
which are largely irrelevant to problems of agreement, so I do not push it
strongly here. But I find the facts suggestive.9
9 In particular, one has to say that the DP in matrix subject position is not referentially dependent
on the S operator (since it is not first person; see (31a)), but it is nevertheless coindexed with it
in whatever sense is needed to create the strong crossover effect in (34). This is a very delicate
distinction – perhaps ultimately not coherent.
My diagnosis of (34) as a type of strong crossover predicts that the phenomenon should be
sensitive to c-command: using a third person DP to refer to the speaker should only prevent a
first person pronoun in the same sentence from referring to the speaker when the third person
DP c-commands the pronoun, therefore counting as its local binder. I believe that this prediction
is correct, although natural-sounding sentences are somewhat difficult to construct, and not all
speakers agree. I consider (iia), in which Daddy does not c-command I, to be more or less
possible, whereas (iib), where Daddy c-commands I is as bad as (34).
(ii) a. (Sk ) Because Daddyk forgot something at the office, hek /I?k have to go back there.
b. (Sk ) Daddyk has to go back to the office because hek /I*k forgot something there.
Similarly, I find the first person pronoun to be marginally possible in (iiia) but completely out
in (iiib).
(iii) a. (Sk ) This old picture of Daddyk shows himk /me?k with long hair.
b. (Sk ) Daddyk finally showed hisk /my*k boss the new contract today.
4.3.2 Locality and shifted interpretations of first and

second person pronouns
Consider next how (29) and (30) account for the fact that a first person pronoun
can refer to someone other than the speaker of the sentence as a whole in
Schlenker’s Amharic example shown in (37) (repeated from (28)).

John hero be.pf-1sS 3m.say.aux.3m
‘John says that he is a hero.’ (lit. ‘John says that I be a hero’)
The shifted interpretation of the first person pronoun in the embedded clause in
(37), in which it refers to the subject of the matrix clause, can be accounted for
by saying that the embedded CP in (37) has an S operator of its own. Moreover,
that S operator is controlled by the subject of the matrix clause, a possibility
alluded to in (29b). (This is a syntactic version of Schlenker’s own view; see
also Koopman and Sportiche 1989 and Adesola 2004 for control of logophoric
operators in Abe and Yoruba.) The representation of (37) is thus (38).
(38) [S1k Johni [S2i Ii hero be+T] say+T]
Here the pronoun in the embedded clause is locally bound by the closest S
operator, namely S2. Hence it qualifies as being first person, by (30a). But S2
does not designate the same individual as S1, namely the speaker. Rather, it
designates John, by virtue of the control relationship. In this way, shifted and
unshifted instances of first and second person pronouns can be accounted for
within the same formal system for licensing first and second person features.10
The full potential of this treatment of first and second person features can
be seen in Slave, as described by Rice (1989) (see Anand and Nevins 2004 for
a similar analysis). (39) gives examples that show that in the complement of
certain verbs, first person pronouns shift to refer to the matrix subject rather
than to the speaker. The matrix subject itself can be any person – first, second
((39c)) or third ((39a–b)).
(39) a. Tony lue gho shéohtı́ enı́dhe . (Rice 1989:1273)

Tony fish 1sS.opt.eat 3sS.wants
‘Tony wants to eat fish.’
(lit. ‘Tony wants I eat fish,’ 1s=matrix subject, not speaker)
10 (37) can also have an unshifted reading, in which it means ‘John said that I am a hero’; see (48)
below for discussion.
b. Judóné ri nurse Teddy gho beghárayuhdá sudeli?

when q nurse Teddy about 1sS.opt.see.3sO 3sS.want.1sO
‘When does the nurse want to see me again about Teddy?’
(lit. ‘When does the nurse want I see him about Teddy?’
1s=matrix subject, not speaker; 3s=speaker) (Rice 1989:1274)
c. Se-ts’e náodı́ yenenewe . (Rice 1989:1282)
1s-to 3sS.opt.help 2sS.want
‘Do you want him/me to help you?’
(lit. ‘Do you want he help me?’ 1s=matrix subject;
3s= speaker or other)
Rice shows that these complements are not direct quotations, which are syn-
tactically independent of the matrix clause. For example, (39b) shows that it
is possible for wh-phrases to move out of the embedded clause in Slave even
when first person pronouns have the shifted reading. (In contrast, direct quota-
tions are islands for extraction in English: compare What did Mary say that she
will buy? with *What did Mary say “I will buy”?) Just as in Amharic, these
embedded clauses have an S operator that is controlled by the matrix subject.
The representation for (39c) would be as in (40) (presented with English word
order for convenience).
(40) [Si , Ak [youk want [Sk [hei,n help mek,*i ]]]]
Rice (1989) shows that whether a verb induces this shift in the interpretation
of first person pronouns is a matter of semi-idiosyncratic lexical selection.
Alongside verbs like those meaning ‘want’ in (39) is another class of verbs
that do not induce a shift in the interpretation of pronouns inside their clausal
complements. (41) is an example; here the first person subject of the embedded
clause refers to the speaker of the sentence, not the matrix subject.
(41) John ʔ erákeʔ ée wihsı́ gú kodı́hsho . (Rice 1989:1273)
John parka 1sS.made comp 3sS.know.area
‘John knows that I made a parka.’ (1s=speaker)
[Si , Ak [Johnn know [(Si ) [Ii make parka]]]]
Slave verbs that do not induce pronoun shift have English glosses such as
‘know,’ ‘hear,’ ‘teach,’ ‘see,’ ‘think, worry about,’ ‘say,’ ‘remember,’ and ‘find
out’ (Rice 1989:1274–5). Verbs that do induce pronoun shift have English
glosses such as ‘say,’ ‘tell,’ ‘ask,’ ‘want,’ and ‘think’ (Rice 1989:1276). Note
that near-synonyms can appear on different lists. This supports a syntax-oriented
approach to the phenomenon like the one I have given. In these terms, the
difference reduces to whether the S operator in the embedded clause is controlled
by the matrix subject or not. The CP complement of ‘know’ in (41) either has
no S operator at all, or it has one that is indexed to the speaker, just as the
S in a matrix clause invariably is. As a result, any first person pronoun in this
complement is locally bound by an S that designates the speaker. This difference
between the two kinds of verbs in Slave is not significantly different from the
familiar fact that some English verbs take finite complements and others take
nonfinite control complements – a difference that also does not seem to be
entirely predictable from the meanings of the verbs involved:11
(42) a. I insist that I teach syntax next year.

*?I insist to teach syntax next year.
b. I demand that I teach syntax next year.
I demand to teach syntax next year.
c. *I want that I teach syntax next year.
I want to teach syntax next year.
Rice (1989) further shows that some verbs – notably the verb meaning
‘tell’ – induce a shift in the interpretation of second person pronouns as well
as first person pronouns. A first person pronoun in a clause embedded under
this verb refers to the matrix subject rather than the speaker, and a second per-
son pronoun in the embedded clause refers to the matrix object rather than the
hearer:
(43) John ʔ aranı́la séhdi. (Rice 1989:1277)

John 2sS.go.home 3sS.tell.1sO
‘John told me to go home.’
(lit. ‘John told me you go home’ 2s=matrix object)
In contrast, verbs like intransitive ‘say’ induce a shift of first person pronouns
in their complement clause, but not a shift of second person pronouns. Thus,
the first person object in (44) refers to the matrix subject (not the speaker) but
the second person subject refers to the addressee of the whole sentence, not to
some covert argument of the matrix verb.
(44) Simon náseneineht’u hadi. (Rice 1989:1279)

Simon 2sS.hit.1sO 3sS.say
‘Simon said that you hit him.’
(lit. ‘Simon said you hit me’ 1s=matrix subject, 2s=addressee)
11 One regularity that Rice mentions is that the complements of verbs that do not shift pronouns
can have overt complementizers; gú in (41) is a case in point. In contrast, the complements of
verbs that do shift pronouns never have an overt complementizer. This relationship between the
presence of S and A operators and the type of C that is selected fits well with the idea that these
operators occur at the CP level of the clause.
Note that the pronouns in the embedded clause in (44) are not interpreted
as they would be in normal indirect discourse, nor as they would be in nor-
mal direct quotation; rather they are a mixture of the two. This confirms
that we need a richer typology of embedded clauses than this simple binary
distinction.
I account for this difference between ‘tell’ and ‘say’ by saying that the com-
plement of ‘tell’ has an A operator that is controlled by the matrix object, as
shown in (45a). In contrast, the complement of ‘say’ has an A operator that
is not controlled, there being no suitable controller in the matrix clause. This
A operator designates the addressee, as usual whenever there is no control
((29b)). The representations in (45) show that the observed facts follow from
these assumptions about structure plus the PLC in (30).
(45) a. [Si , Ak [Johnn tell mei [Sn , Ai [youi,*k /*Ii go home.]]]]

b. [Si , Ak [Simonn say [Sn (Ak ) [youk hit men/*i /him*n ]]]]
These examples make several significant points. First, they show that the system
generalizes properly from first person to second person pronouns, as we would
hope. Second, they show that there are similar but distinct operators for second
person and first person pronouns, since second person and first person do not
shift in exactly the same environments.12 Third, they give further support for
a syntactic approach modeled on control, in that object control exists in both
domains, and similar sorts of verbs trigger object control of both kinds.
Now we come to the aspect of the Person Licensing Condition that is the
most crucial for the project of deriving the SCOPA. Notice that the second
person pronoun in (43)/(45a) cannot refer to the addressee of the sentence as a
whole, but can only refer to the object of the matrix clause. This shows that it
is not enough for a second person pronoun to be bound by any old A operator;
rather it must be bound by the closest A operator – the one in the embedded CP,
not the one in the matrix CP. Similarly, the first person pronoun in (44)/(45b)
cannot refer to the speaker of the sentence as a whole, but only to the matrix
subject (Simon). This shows that the first person pronoun can only be bound
by the closest S operator. Rice’s extensive Slave data is entirely consistent on
this point. Indeed, the same thing could be seen back in example (39c)/(40).
There too the first person pronoun in the embedded clause can only be bound
by the closest S; therefore it ends up designating the addressee rather than the
speaker, given that the embedded S is controlled by the matrix subject. Note
12 This is crucially different from Schlenker’s (2003) implementation, in which all the index
parameters are packaged together in a single operator.
also that the third person subject ‘he’ of the embedded clause in (40) can refer
to the speaker: even though this pronoun is assigned the index i associated with
the matrix S, it is not locally bound by the closest S, so first person features are
not licensed on it. It is therefore third person, by default, in accordance with
the PLC.
Rice (1989:1289) also briefly discusses the interpretation of pronouns in
sentences with two levels of clausal embedding, such as (46). This provides
further evidence for the minimality clause in the PLC:
(46) John Susan tle go lı́ ʔ aohde enı́we ʔ adi.

John Susan Norman Wells 1sS.opt.go 3sS.want 3sS.say
‘John said that Susan wants (Susan/*John/*me) to go to Norman Wells.’
The notable feature of this example is that the first person pronoun in the
most deeply embedded clause must refer to the subject of the immediately
higher clause, not to the matrix subject, and not to the speaker. This is addi-
tional evidence that a pronoun is only first person if it is bound by the closest
c-commanding S operator. The representation of (46) in (47) should make this
clear.
(47) [Si Johnk said [Sk Susann wants [Sn In,*k,*i go.to Norman Wells]]]
At first glance, the other languages discussed in the literature on the shifting
of first and second person pronouns seem to be different from Slave in this
respect. The first person pronoun in Schlenker’s Amharic example, repeated
again in (48), can refer to either the matrix subject or the speaker.

John hero be.pf-1so 3m.say.aux.3m
‘John says that he is a hero’ or ‘John says that I am a hero.’
The same ambiguity is found in Anand and Nevins’s (2004) examples from
Zazaki, including (49).
(49) Hεseni va kε εz dεwletia.

Hesen.obl said that I rich.be
‘Hesen said that he (Hesen) is rich’ or ‘Hesen said that I am a rich.’
But at least for Zazaki the problem is only apparent. The source of the ambiguity
in (49) is not the ability of a first person pronoun to be bound by a more remote S
operator; rather it comes from the fact that the Zazaki verb ‘say’ only optionally
selects for a controlled S operator in its CP complement. Thus, (49) can have
either of the two representations in (50).
(50) a. [Si Hesenk say [Sk that [Ik rich.be]]]

b. [Si Hesenk say [that [Ii rich.be]]]
In both representations, the first person pronoun is bound by the closest S, in

accordance with the PLC. Slave is different only in that its verb meaning ‘say’
selects only the equivalent of the CP found in (50a).
Crucial evidence that the same minimality condition holds in Zazaki as in
Slave comes from examples like (51), which Anand and Nevins discuss as
illustrating their “shift together” generalization.
(51) Hεsen va kε pyaay kε m -ra hes kene

Hesen says That people that me-obl like do
pyaay kε m -ra hes ne kene ame zuja.
people that me-obl like neg do came together
*‘Hesen said that the people who like him/me and the people
who don’t like him/me met.’
Like (49), this sentence is two-ways ambiguous: the first person pronouns in
the embedded clause can refer to the speaker or to the subject of the matrix
clause, Hesen. But Anand and Nevin point out that the sentence is not four-ways
ambiguous: it is impossible for one of the first person pronouns to designate the
matrix subject Hesen while the other designates the speaker. The minimality
condition in the PLC explains why this is so. Suppose one of the first person
pronouns is interpreted as referring to Hesen. This is only possible when the
verb ‘say’ exercises its option of selecting a CP with a controlled S operator, so
the representation is one like (50a), not one like (50b). Given this, consider the
options for the second first person pronoun. It must be bound by the controlled
embedded S, not by the matrix S, according to the minimality clause of the
PLC. Hence it too is interpreted as referring to Hesen. The representation is
given schematically in (52a). (The other, less-relevant representation is given
in (52b), for comparison.)
(52) a. [Si Hesenk say [Sk that [people who like mek,*i ] & [people who don’t like
mek,*i ] met]]
b. [Si Hesenk say [that [people who like mei,*k ] & [people who don’t like
mei,*k ] met]]
Therefore, (30) applies as written to Zazaki and as well to Slave. I assume that
Amharic is to be analyzed similarly to Zazaki, although the crucial evidence is
not available. Indeed, I assume that the PLC is universal.
4.3.3 Comparison with logophoric pronouns

To conclude this background investigation into first and second person pro-
nouns, it is instructive to compare what we have learned about them with the
phenomenon of logophoric pronouns found in West African languages. Most
people working on this topic have seen similarities between the two phenom-
ena, and for some pronoun shifting is analyzed as a type of logophoricity (e.g.,
Safir 2005). I agree that there are many important parallels – but there is also
one crucial difference that is highly relevant to my topic.
Logophoricity in African languages is the use of a special pronoun in an
embedded clause to refer to the subject of a main clause containing a verb
of speaking or cognition. I illustrate the phenomenon from Edo (data from
O. T. Stewart, personal communication); see also Koopman and Sportiche 1989
on Abe and Adesola 2005 on Yoruba. For example, the ordinary third person
pronoun ó. in (53a) may not be coreferential with the matrix subject Òzó., whereas
the distinctive logophoric pronoun ı́rè.n in (53b) must be.
(53) a. Òzó Miànmián wè.é. ó. kı̀é èkhù.

Ozo forgot that he opened door
‘Ozoi forgot that he*i,k (someone else) opened the door.’
b. Òzó. miànmián wè.é. ı́rè.n kı̀é èkhù.
Ozo forgot that he.log opened door
‘Ozoi forgot that hei,*k (Ozo) opened the door.’
The logophoric pronoun cannot, however, be used in argument position of an

ordinary matrix clause, nor is it licensed in infinitival embedded clauses:
(54) a. *Írè.n mié.n ènı́.

he.log see elephant
‘He saw an elephant.’
b. Òzó miànmián ya tı̀é èbé é.rè /*ı́rè.n
Ozo forget to read book his/his.log
‘Ozo forgot to read his book.’
Following Koopman and Sportiche (1989), this range of data can be captured
by saying that logophoric pronouns need to be locally bound by a designated
logophoric operator, which is found in the specifier of certain CPs. This is
stated in (55) (essentially the same as the Strong Pronouns Licensing Principle
of Adesola 2005:190).
(55) A logophoric pronoun must be locally bound by a logophoric operator, or by

another logophoric pronoun.
Verbs of speaking and cognition select for CPs that have such an operator, and
one of their arguments controls that operator. This gives a representation like
(56) for (53b).
(56) [Ozok forgot [LOGk that [hek opened the door]]]
In contrast, matrix clauses do not have a logophoric operator, nor do nonfinite

clauses like the one in (54b). Therefore the examples in (54) are ruled out by
the condition in (55).
The parallels between logophoric pronouns and shifted first person pronouns
in Slave are evident once one realizes that the logophoric pronoun in Edo is
used in the same contexts in which a first person pronoun in Slave has a shifted
interpretation. Both logophoric pronouns and first person pronouns are used
to designate the subject of the higher speech verb, but not the subject of other
sorts of verbs. Moreover, Adesola (2005) shows that logophoric pronouns are
subject to a kind of strong crossover effect, much like first person pronouns are
in my analysis of (34) in section 4.3.1. In Yoruba (but not in Edo) a normal
pronoun can be interpreted as referring to the matrix subject in an example
like (57).13
(57) Olúi ti kéde pé ói,k ń bò. ló.la. (Adesola 2005:184)
Olu asp announce that he prog come tomorrow
‘Olu announced that he (Olu or someone else) is coming tomorrow.’
But the existence of such a pronoun in the embedded clause can prevent a
logophoric pronoun from designating the matrix subject the way that it normally
would:
(58) Olúi so. [LOGi pé [ó*i,k rı́ bàbá òuni ]] (Adesola 2004:185)
Olu say that he see father him.log
‘Olu said that he (someone else) saw his (Olu’s) father.’
The deviance of (58) when ó is understood as bound by the matrix subject can be
attributed to (55): although the logophoric pronoun is bound by the logophoric
operator, it is not locally bound by that operator; the nonlogophoric pronoun is
a closer binder. The badness of (58) is thus comparable to the badness of [Si
Daddyi lost myi wallet] in English. If the closer binder is itself a logophoric
13 This difference between Yoruba and Edo can be modeled by saying that ‘say’-type verbs in Edo
require a logophoric operator in their complement, whereas ‘say’-type verbs in Yoruba allow
one but do not require one. Note that this is exactly parallel to the difference between Slave
and Zazaki when it comes to the shifting of first person pronouns: some Slave verbs require
a controlled S operator in their complement, whereas ‘say’ in Zazaki allows one but does not
require it.
pronoun, however, the sentence is fine with the relevant interpretation (Adesola
2005:185, 192).
(59) Olúi so. [LOGi pé [óuni rı́ bàbá òuni ]]

Olu say that he.log see father him.log
‘Olu said that he (Olu) saw his (Olu’s) father.’
The acceptability of examples like this are the reason why (55) states that a
logophoric pronoun can be locally bound by another logophoric pronoun as
well as by a logophoric operator. This is parallel to the fact that first person
pronouns can be locally bound by an S operator or by another first person
pronoun, as shown by the goodness of [Si Ii lost myi wallet]. In all these respects,
logophoric pronouns are like first and second person pronouns – similarities
that are captured by saying that both must be locally bound by a particular sort
of operator (the PLC and (55)).
There is one crucial difference between the PLC and (55), however: (55) does
not contain the same minimality condition that the PLC does. (55) does not stipu-
late that a logophoric pronoun must be bound by the closest logophoric operator.
The difference is empirically motivated. In a doubly embedded structure like
(60) from Edo, the logophoric pronoun in the lowest clause is ambiguous; it
can refer to either the subject of the highest clause or to the subject of the
intermediate clause.
(60) Òzó ròró wè.é. Úyı̀ tá wè.é. Adesuwa bàá ı́rè.n òhó! ghé.
Ozo thinks that Uyi say that Adesuwa accuse him.log of.lying
‘Ozo thinks that Uyi said that Adesuwa accused him (Ozo or Uyi) of lying.’
The same is true in other West African languages, including Yoruba, Abe, and
Ibibio. In contrast, we have seen that shifted first person pronouns in similar
contexts in Slave are unambiguous: a first person pronoun in the lowest clause
can only refer to the subject in the clause immediately above it, not to the subject
of the highest clause, as shown in (46), repeated here as (61).
(61) John Susan tle go lı́ ʔ aohde enı́we ʔ adi.

‘John said that Susan wants (Susan/*John) to go to Norman Wells.’
It follows that when there are two logophoric pronouns in the same clause, they
need not refer to the same person; (62) is an example in which they do not (see
also Koopman and Sportiche 1989:571–2).
(62) Òzó tá wè.é. Àdésúwà ròró wè.é. ı́rè.n hoè.mwé.n ı́rè.n.
Ozo say that Adesuwa think that he.log like her.log
‘Ozo said that Adesuwa thinks that he (Ozo) likes her (Adesuwa).’
This contrasts with the Zazaki example in (51), where two first person pronouns
in the same clause must refer to the same person. This contrast, assuming
that it proves systematic across languages, is problematic for Safir’s (2005)
view that “shifted” first person pronouns in languages like Amharic are simply
logophoric pronouns that happen to be homophonous with ordinary first person
pronouns.
I conclude that logophoric pronouns are very similar to shifted first and
second person pronouns, but they are not identical to them. All three must be
locally bound by a designated operator. The difference is that first and second
person pronouns must be bound by the closest operator of the relevant kind,
whereas third person pronouns (including logophoric ones) need not be. I do
not have any deep insight into why first and second person pronouns should
differ from third person pronouns in this way. But apparently they do – and it is
this difference that I use to derive the differences between first/second person
agreement and other kinds of agreement.
4.4 On the strictness of locality conditions involving heads

So far I have established that a pronoun is first or second person only if it
is locally bound by the closest particular element that is also first or second
person, as stated in the Person Licensing Condition (PLC) in (30). Moreover,
agreeing functional heads are like pronouns in that they count as bound variables
dependent on the agreed-with NP, according to (26). Therefore it is natural to
think that the PLC applies to these agreeing functional heads as well. Hopefully
this is the source of the difference in first and second person agreement as
opposed to agreement in other features, the descriptive generalization expressed
in the SCOPA.
But there is still something that needs to be added before this works. While
it may be true that agreement markers associated with first or second person
pronouns are subject to the local binding condition in (30), it seems like this
condition will always be satisfied in a rather trivial way. By hypothesis, a
functional head F has first or second person features only if it enters into an
Agree relationship with some pronoun in the same phase that is first or second
person. The agreed-with pronoun is itself only first or second person if it is
locally bound by an S or A operator, or by some other first or second person
On the strictness of locality conditions involving heads 139
pronoun, given the PLC. Then either the S or A operator that binds the pronoun
also locally binds the agreement-bearing head (if the head agreed downwards,
with a pronoun in its c-command domain), or else the agreed-with pronoun
itself locally binds the agreement-bearing head (if the head agreed upwards,
with a pronoun that c-commands it). Either way, any configuration that has a
first or second person pronoun to agree with will also be a configuration in
which the PLC is satisfied for the agreeing head. Therefore, the PLC as stated
puts no new restrictions on first and second person agreement that go beyond
those built into the Agree relationship itself. In particular, it does not require
that the first or second person pronoun merge directly with the agreeing head.
So it cannot be used to explain the SCOPA unless something is added.
Fortunately there is a new assumption that can be added to this picture. When
one considers other sorts of locality conditions, there is evidence that, even when
recognizably the same condition applies to both phrases and heads, the locality
holds in a stronger form in the case of heads. One case in point is the binding
condition that regulates anaphors. All anaphors must be bound by an antecedent
within a suitable local domain. For full NP anaphors, this local domain can be a
rather large one, such as the smallest indicative clause that contains the anaphor.
Thus, the anaphor in object position in (63a) from Icelandic need not be bound
by the closest subject, but can be bound by the subject of the matrix clause,
given that the embedded clause is subjunctive. This kind of latitude is never
granted to anaphors that are bare heads, such as the clitic anaphors that attach
to Tense in the Romance languages. These can only take as their antecedent the
most local NP imaginable: the specifier of the very T that they attach to. (63b)
thus differs from (63a) in this respect (Pica 1991).
(63) a. Jón sagð i þeim að Marı́a elski sig. (Icelandic, p. 119)

Jon told them that Maria love.sbjn self
‘Johnk told them that Mariai loves him/herselfi,k .’
b. Paul souhaite que Jean se photographie. (French, p. 130)
Paul wishes that Jean self photograph.sbjn
‘Paulk wishes that Jeani would photograph himselfi,*k .’
Both sig and se are anaphors, distinguished from comparable pronominals

by their need to have a binder within a local domain. But there is also a
difference: the relevant domain is much smaller for the anaphoric head than
for the anaphoric DP.
Another illustration of this observation comes from the theory of movement.
All sorts of movement are subject to locality conditions of one kind or another.
Moved wh-phrases can only move out of one clause at a time, and only if there
is an intermediate landing site at the edge of the clause; this results in the famil-
iar range of island phenomena (Chomsky 1973). Similarly NPs undergoing
A-movement for case-and-agreement reasons usually stay within the same
clause, and can only move into the next higher clause when very specific con-
ditions are met. Both of these sorts of movements are local to something like a
clause. Moved heads are also subject to a locality condition, but a much stricter
one. Staying within the same clause/phase is not good enough for a moving
head; rather it must land in the immediately superordinate phrase, whether it is
a phase or not (the Head Movement Constraint; see Travis 1984, Baker 1988).
This results in contrasts like the following:
(64) a. Such opportunities must be taken advantage of –. NP movement

b. Which car do you like the color of –? Wh-movement
c. *K-ather-a-nuhwe’-s [ne ohsohkwa –]. Noun incorporation
1sS-basket-Ø-like-hab ne color (Mohawk (Baker 1991:371))
‘I like the color of the basket.’
In (64a), a kind of pseudopassive, the NP apparently raises out of PP, NP, and
VP into the subject position. In (64b), the wh-phrase moves out of PP, NP, VP,
and TP into Spec, CP. But there is no noun incorporation structure like (64c) in
which a noun moves out of a larger NP to adjoin to V. The movement in (64c)
is, if anything, shorter than the movements in (64a) and (64b), but yet it is too
long. Similar contrasts can be found in complex verbal constructions, as shown
in (65) and (66).
(65) a. Le nuove case furono iniziate a costruire negli anni ’20.

The new houses were started to build in.the 20s
‘They started to build these houses in the 20s.’
(Italian, Wurmbrand 2003:20)
b. Which book did you start to read – ?
(66) a. ə tlə g-e lə waw-ə rkə -nen rə -maraw-at-ə k ekə k.
father-erg cannot-pres-3sS/3sO caus-scold-caus-inf son.abs
‘The father cannot scold the son.’ (Chukchi, Baker 1999:369)
b. Ga-nae l-in-au payitt-o-k.
past2-start-past2–3pS blackberry-eat-inf
‘They started to eat blackberries.’
Not: *Ga-payitt-nae l-in-au o-k ‘(They) blackberry-started to-eat’
In these restructuring cases, the object of the lower verb can move out of the
lower VP and out of the higher VP, into the matrix Spec, TP, as in (65a) (the
so-called “long passive”). Also in such cases, the +wh object of the lower verb
can easily move out of the lower VP and the higher VP, into the matrix Spec, CP,
On the strictness of locality conditions involving heads 141
as in (65b). But restructuring seems never to allow “long noun incorporation.”

Chukchi provides a good illustration: clitics/agreements that express the object
of the lower verb can appear attached to the higher verb, as in (66a), but an
incorporated noun root that expresses the object of the lower verb cannot ((66b)).
Again, the movement in (66b) is, if anything, shorter than the movements in
(65), and yet it is too long.
It is helpful to see how this difference between head movement and phrasal
movement was captured in Rizzi’s (1990) theory of Relativized Minimality.
Rizzi’s idea was that movement relationships were blocked if (among other
things) the movement skipped over a position of the same type as the landing
site of the movement. Crucial building blocks toward achieving this result are
quoted in (67) and (68) (Rizzi 1990:7).
(67) Relativized Minimality: X α-governs Y only if there is no Z such that
(i) Z is a typical potential α-governor for Y,
(ii) Z c-commands Y and does not c-command X.
(68) a. Z is a typical potential antecedent governor for Y, Y in an A chain =
Z is an A specifier c-commanding Y.
b. Z is a typical potential antecedent governor for Y, Y in an A chain =
Z is an A specifier c-commanding Y.
c. Z is a typical potential antecedent governor for Y, Y in an Xo chain =
Z is a head c-commanding Y.
Notice that the movement of phrases divides into two distinct kinds, A-
movement and A-bar movement. The difference between the two is defined not
only by the syntactic configuration (whether there is an intervening specifier) but
also by something about the intrinsic feature content of the position – whether it
is an A-bar specifier or an A-specifier, a specifier with quantificational/operator
features or one with case/agreement features. Thus NP raising can cross a floated
quantifier but not a nominative subject, whereas wh-movement can cross over
a nominative subject but not a floated quantifier. Subsequent work identified
still more types; for example, Müller and Sternefeld (1993) identify scrambling
and topicalization as additional types of movement, which do not interact with
each other, or with wh-movement and NP-movement.
In contrast, the movement of heads does not divide up into two or more distinct
subtypes. There is little or no evidence that some kinds of head movement can
skip one type of head position whereas other kinds can skip other types of
head positions. This implies that the intrinsic features of heads do not matter
when evaluating the locality of a head movement, the way that the features
of specifiers matter when evaluating the locality of a phrasal movement. Only
a specifier of the same type blocks movement to specifier position, whereas
heads of all types block movement to head positions. The upshot is that head
movement obeys a more restrictive form of locality than phrasal movement,
as observed in (64)–(66). The movement of phrases is local to (roughly) the
clause, but the movement of heads is bounded by a single phrase.
Suppose that we generalize over the case of anaphor binding and the case of
movement, to posit a general law, stated roughly as in (69).
(69) For a locality relationship involving a head, all intervening heads count as
interveners, regardless of their intrinsic features.
(69) could be the underlying reason why there are multiple clauses dealing with
phrasal movement in (68), but only one clause dealing with head movement.
4.5 Deriving the SCOPA

We are now ready to put the pieces together by applying the general law in (69)
to the PLC, the locality condition on the licensing of first and second person
features motivated by the data in section 4.3. The PLC is repeated here as (70),
with the crucial locality condition unpacked and italicized.
(70) Person Licensing Condition

a. A phrase X is first person only if it is locally bound by an S or another
element that is first person, and there is no other S that c-commands X but
not its local binder.
b. A phrase X is second person only if it is locally bound by an A or another
element that is second person, and there is no other A that c-commands X
but not its local binder.
c. Otherwise, X is third person.
(70a–b) is similar to (68a–b) in that the intrinsic features of the operators matter
for whether they count as interveners as well as their syntactic positions. Thus,
the S operator blocks the more remote binding of first person pronouns but not
that of second person pronouns, whereas the A operator blocks the binding of
second person pronouns but not first person pronouns. This is seen most clearly
in an example like (44) from Slave, repeated here as (71).
(71) Simon náseneineht’u hadi.

Simon 2sS.hit.1sO 3sS.say
‘Simon said that you hit him.’ (1s=matrix subject, 2s=hearer)
[Si , Ak [Simonn say [Sn [youk hit men/*i ]]]]
Here the lower S operator prevents the pronoun ‘me’ from referring to the matrix
S (i.e., to the speaker), but it does not block the pronoun ‘you’ from referring
Deriving the SCOPA 143
to the matrix A (the addressee). So intrinsic features and relative positions both
matter in these relations involving pronouns. But according to (69), intrinsic
features should not matter for relations involving heads. This means that any
intervening head, regardless of its features, should block the licensing of first
and second person on a head, much as any head, regardless of its features,
blocks head movement. The version of (70) that applies to heads is thus (72),
with the italicized locality condition replaced by a non-relativized one.
(72) Person Licensing Condition (Head Version):

a. A head X is first person only if it is locally bound by an S or another
element that is first person, and there is no other head that c-commands X
b. A head X is second person only if it is locally bound by an A or another
element that is second person, and there is no other head that
c-commands X but not its local binder.
c. Otherwise, X is third person.
I now claim success, inasmuch as (72), the PLC(H), is equivalent to the SCOPA
in (1), repeated here as (73).
(73) F can agree with XP in +1, +2, only if a projection of F merges with a
+1, +2 element and F projects.
To see the equivalence of (72) and (73), consider first a typical configuration
that satisfies the SCOPA, such as a finite T agreeing in person with the nomi-
native subject in Spec, TP in Icelandic or some other language. The syntactic
configuration would be approximately (74).
(74) . . . [CP Si Ak that [TP Ii [1,sg] Tense[1,sg]i [vP <I> v . . .]]]
The SCOPA is satisfied because a projection of T merges with a copy of the

first person subject that T agrees with. But (72) is also satisfied. The pro-
noun ‘I’ is a first person element that locally binds the first person head Tense
as a result of Agree (see (26)). Moreover, there is no head Y, distinct from
Tense and I, that c-commands Tense but not I. (The first head that c-commands
Tense – other than perhaps I itself – is the complementizer C. But C also
c-commands the subject I, since I is contained in the TP complement of C.) The
same is true for a configuration in which P merges with (say) a first or second
person pronoun and agrees with it; the pronoun locally binds the agreeing P,
and there is no distinct head that c-commands P without also c-commanding
that pronoun. More generally, suppose that any agreeing head F merges with a
+1 or +2 NP. Then the +1 or +2 NP locally binds the agreeing F. Moreover,
there could never be a head G that c-commands F but not the +1 or +2 NP. If
G c-commands F, then G merges with something that contains FP, the maximal
projection of F. But F merged with the pronoun, by hypothesis, so the pronoun
is inside FP and hence also inside the complement of G. So G c-commands
the pronoun as well as the head, and does not count as an intervener. Thus,
configurations that satisfy the SCOPA also satisfy the head version of PLC in
(72).14
Next, consider configurations which do not satisfy the SCOPA because the
pronoun is generated higher than the agreement bearing functional head F.
The canonical case is predicate adjective agreement, where the first or second
person pronoun is generated in Spec, PredP, and FA is the complement of Pred,
as shown in (75).
(75) [CP Si Ak C [TP Wei [1,pl] be+T[1,pl]i [PredP <we> Pred [FAP FA [*1,pl]
[AP good]]]]]
The condition in (72) is also violated in (75). The closest binder of the first
person FA is the copy of ‘we’ in Spec, PredP. But Pred is a head distinct from
either FA and ‘we’ that c-commands FA and not ‘we’. Therefore, it prevents
‘we’ from licensing a first person feature on FA (even though it has no relevant
feature of the same type). The same result is true more generally. Suppose
that the first or second person pronoun does not merge with FP, but also is
not c-commanded by F. Then the pronoun must c-command F, or even basic
gender-number agreement would fail. As a maximal projection, the pronoun
must merge with some other head G, by hypothesis distinct from F, where
G projects. That the pronoun c-commands F implies that F is contained in
the smallest projection that contains the pronoun G*, a projection of G. More
particularly, if the pronoun is the specifier of G*, then F must be contained in
the complement of G.15 If F is in the complement of G, then G c-commands
F but not the pronoun. Therefore, G is an intervening head that prevents the
pronoun from licensing +1 or +2 features on the functional head. These cases
that violate the SCOPA thus violate (72) as well.
14 A possible complication could arise if G merges with F, but F projects. Then the higher projection
of F merges with the pronoun. Then G might c-command F, but not the pronoun, and count as
an intervener. The relationship of projection might need to be stated in a certain way to render
this unproblematic (or one could state intervention in terms of asymmetrical m-command, as in
note 20 to chapter 2).
15 As usual, a little more care is needed if the pronoun is adjoined to G*. Then F could be in a
specifier of G*, or a lower adjunct to G*, and G would m-command F but not c-command it.
One might also have to take into account the possibility of multiple specifier constructions; I
leave this open.
The other broad class of cases to consider is configurations that do not satisfy
the SCOPA in which the agreeing functional head c-commands the agreed-with
DP. Typical cases in point are the downward adjectival agreement found on
Cinque adjectives in Italian (section 3.1.1) and the downward verbal agreement
found in oblique subject constructions in Icelandic (section 3.3.2). The relevant
syntactic configurations are sketched in (76).
(76) a. [CP Si Ak C [TP <EX> be+T [<EX> Pred [FAP FA [*1,sg] [AP
well.known [I, 1sg]]]]]]
b. [CP Si Ak C [TP Mary-DAT Tense[*1,sg]i [vP <Mary> v [VP bored.with
[I,1sg]i ]]]]
These configurations also clearly violate (72). This time the agreed-with pro-
nouns do not bind the agreeing heads because they do not c-command them
(by hypothesis). Therefore, the pronouns cannot license first and second person
features on FA or Tense, even apart from the “closer head” clause of (72). This
will be true for any configuration that meets the description of this case: if F
c-commands the pronoun and F does not merge with the pronoun, then the
pronoun does not c-command F and hence cannot license +1 or +2 features on
F. Putting this together with the previous two cases, we find that all syntactic
configurations that satisfy the SCOPA satisfy (72) as well, and all syntactic
configurations that violate the SCOPA violate (72) as well. The two are thus
linguistically equivalent conditions. Inasmuch as (72) is the result of indepen-
dently motivated conditions ((30) and (69)), we have explained the SCOPA.
There is one other matter to consider with regard to structures like (76),
however: could first or second person agreement be licensed on FA or T by
virtue of FA or T being locally bound by the S or A operators that have scope
over the whole clause? In principle, this could be possible: we know that S
and A can license +1 and +2 on pronouns that they bind, and an important
assumption of this whole line of analysis is that the agreement on F is essentially
a bound variable, subject to versions of the same conditions. For the specific
configurations in (76), this is not an issue: S cannot license +1 features on FA or
T in (76) because the complementizer C is an intervening head that c-commands
FA and T but not the S and A operators. Therefore the super-strict style locality
that heads require does not hold.
It is, however, perfectly imaginable that the S and A operators could license
first or second person morphology on the complementizer itself, there being
no other head that intervenes between the operators and the complementizer.
This is a good result. Recall from section 4.1.5 that complementizers are the
one category on which we found some apparent exceptions to the SCOPA.
Exceptions came to light in at least two language families: the Continental

West Germanic languages, and the Niger-Congo languages. Consider first the
West Germanic languages. A typical case of complementizer agreement in these
languages is repeated in (77), from West Flemish.
(77) Kpeinzen dan-k (ik) morgen goa-n.
I-think that-1s I tomorrow go-1s
‘I think that I’ll go tomorrow.’
Here the complementizer appears to be agreeing downward with the first person
pronoun in Spec, TP, in violation of the SCOPA. But our investigation into the
details of first and second person pronouns has revealed that CPs of this sort
can have S and A operators generated within them. I claim that the agreement
on the complementizer in (77) is made possible by the S operator generated
inside CP, as shown in (78).
(78) . . . think [CP Si that[1sg]i [TP I[1sg]i [vP tomorrow <go>] go+T[1sg]i ]]
This is compatible with (72) (and the SCOPA): the pronoun in Spec, TP locally
licenses +1 features on T, and the S operator in CP locally licenses the +1
on C.
The Kinande example repeated in (79) is similar: there is an operator in the
projection of the embedded speech complementizer ati, and it locally triggers
first person agreement on that complementizer.
(79) Mo-n-a-layir-ire Kambale in-di a-gul-e amatunda.
aff-1sS-t-convince-ext Kambale.1 1sS-that 1S-buy-sbjn fruits.6
[Si I i [1sg] T convinced . . . [LOGi [1sg] that[1sg]i [TP he buy fruits]]]
At first, this looked like nonlocal upward agreement on C with the matrix sub-
ject – an NP that is not even in the same phase as the C. Now it can be analyzed
as local upward agreement with an operator in C, that operator itself being con-
trolled by the matrix subject.16 These examples thus cease to be problematic;
on the contrary, they constitute additional evidence in favor of the syntactic
reality of S and A operators in the CP. That S and A operators can license
agreement on C shows that they are indeed part of the syntactic representation.
The fact that they can license agreement only on C – not on T or FA in examples
16 For Kinande, it is better to say that the operator in Spec, CP is a logophoric operator controlled
by the matrix subject than to say it is an S or A operator. The reason is because first and second
person pronouns in the embedded CP do not shift their interpretation in Kinande the way they
do in Slave. It is not the exact nature of the operator that is crucial here, but rather its location
and the ϕ-features that it bears.
like (76) – both confirms precisely where these operators are in the syntactic
representation and bears further witness to the extra-tight locality condition on
first and second person agreement. (Note that the agreement on C in Kinande
agrees with the matrix subject, whereas in West Flemish it agrees with the
embedded subject. This difference relates to a very general parametric differ-
ence that distinguishes agreement in Niger-Congo languages from agreement
in Indo-European languages – the topic of the next chapter.)
Finally, a syntactic configuration of special interest is one in which the pro-
jection of an agreeing functional head is adjoined to a first or second person
pronoun. We saw in chapter 2 that in such a configuration the functional head
can agree with the pronoun it is adjoined to in number and gender, but not in
person. The canonical case is an adjective adjoined to a pronoun in Tariana or
Zulu, as in (80).
(80) a. Thin’ aba-khulu si-ya-khuluma. (*si-khulu) (Doke 1955:51)

we cl1.pl-big 1pS-disj-be.speaking 1p-big
‘We big ones are speaking.’
b. [DP [DP we[1pl]k ] [FAP FA [pl,*1]k [AP big]]]
Another case is the projection of the associative head/relative marker in Niger-

Congo languages (see section 4.1.4). Here person agreement fails, even though
the target and the goal are as close to each other as one could ask for; they
merge with each other, and no other head or phrase intervenes. For the SCOPA,
this configuration motivated the extra condition that when the agreeing head
F and the agreed-with pronoun merge, it must be F that labels the resulting
projection for person agreement to take place. Does the shift from the SCOPA
to the PLC(H) in (72) shed any additional light on this matter?
I claim that the answer is yes. When framed in terms of (72), it is clear
that the real condition on person agreement is not just a locality condition:
the agreed-with phrase must also bind the agreeing head. Here binding is to
be understood in the way it is understood in binding theory and the theory of
operator–variable relationships, as involving c-command plus coindexing of the
sort that represents a referential dependency. This is pressed upon us by the view
that (72) is not an independent principle of agreement; rather it is a version of
a general operator-binding condition that applies to pronouns/variables more
generally. Now, does the pronoun in (80) bind the agreeing head FA in the
required sense? The answer is arguably no: it does not c-command FA , rather
only a segment of it does. The maximal projection of the pronoun is the ultimate
bearer of the pronoun’s index, and this category contains FA P, and hence does
not c-command it. The distinction between c-command and containment is
known to be a crucial one in binding theory: a phrase cannot be the antecedent

of an anaphor or pronoun that is contained within it (the i-within-i condition of
Chomsky 1981). Thus, (81a) is acceptable, but (81b) is bad.
(81) a. [Johnk ’s picture of himselfk ] hangs in the front hall.
b. *[NPk M. C. Escher’s picture of itselfk ] hangs in the front hall.
Since the NP in brackets in (81b) contains the anaphor, it cannot be the

antecedent of that anaphor; it does not bind the anaphor. Taking this back to (80),
it is fair to say that the first person plural pronoun does not bind the agreeing FA
here. Therefore, it cannot license first person features on FA via (72), a binding
theoretic principle. That aspect of agreement that is pure Agree can succeed
in (80), but the aspect of agreement that is ultimately part of operator-variable
binding cannot – which is the part of agreement that licenses first and second
person features.17 I claim, then, that no special stipulation (only a clarifica-
tion) is needed to account for structures like (80) within the new theory of the
SCOPA. This is an encouraging sign that the quest to derive the SCOPA from
more fundamental principles was worth the effort and has been successful.
4.6 Conclusion
In the previous section, I showed that the configurations that satisfy the SCOPA
also satisfy the Person Licensing Condition for Heads in (72), and the config-
urations that violate the SCOPA also violate this PLC(H). Thus, the two are
17 Note that the c-command condition on Agree has to be understood a little differently: the first
person pronoun does count as c-commanding FA for purposes of pure Agree. This is a delicate
distinction, but not too unnatural, I think. It is equivalent to saying that all segments of a category
(including the lower one) have ϕ-features needed for agreement, but only the highest one has
the index relevant to binding. Technically, we can say that segments participate in c-command,
but X binds Y only if X c-commands Y, X is coindexed with Y, and X is the largest phrase
that bears the relevant index. (And one can hope that these choices will make sense in an even
deeper, more minimalist understanding of these matters.)
Notice that an anaphor can seem to depend on the head of a noun phrase that it is contained in
in relative clauses, e.g., The man who is proud of himself will fall into error. In such examples,
the anaphor is actually directly dependent on a relative operator that c-commands it. In exactly
parallel circumstances, a verb or other head inside a relative clause can appear to agree in person
with the head of the relative clause, as in the Lokaa example repeated here:
(i) amon b-ȧ yo-bi:la
we 2-rel 1pS-be.dark
‘we who are darkskinned’
Here, too, agreement is really with the trace of the relative operator, which properly binds the
agreeing T. (This reasoning also implies that the relationship between the head of a relative
clause and the operator of the relative clause cannot be binding-theoretic in nature.)
Conclusion 149
linguistically (nearly) equivalent, and we can replace the SCOPA in the theory
of agreement with the PLC(H). But the PLC(H) is the coming together of two
independently motivated principles, namely the PLC in (30) – the general lin-
guistic rule of what it means for a grammatical element to be first or second
person – and (69), a general law about how locality conditions apply to heads.
In short, I have successfully derived the special behavior of agreement in first
and second person from the fundamental defining property of first and second
person features.
What this ultimately amounts to (beyond theoretical and rhetorical tricks) is
a claim to have uncovered a very abstract but significant parallel between the
special syntactic behavior of first and second person pronouns and the special
behavior of first and second person in agreement. When it comes to pronouns,
first and second person pronouns are marked forms. They obey a very specific
licensing condition, such that they must be locally bound by the closest operator
of a specific type. Third person pronouns are unmarked forms, which are not
subject to any such licensing condition; they simply appear whenever a first
or second person pronoun is not licensed. The locality condition on first and
second person pronouns is seen clearly in examples like (82) from Slave, where
the first person pronoun in the most deeply embedded clause can only take the
subject of the immediately higher clause as its antecedent.
(82) John Susan tle go lı́ ʔ aohde enı́we ʔ adi. (p. 1289) (=(46))
‘John said that Susan wants (Susan/*John) to go to Norman Wells.’
In contrast, third person pronouns – even logophoric ones that need to be bound
by an operator – do not need to be bound by the closest operator. Thus, exam-
ples like (83) from Edo are possible with the indicated interpretation (see also
Adesola 2004:207 for Yoruba).
(83) Òzó ròró wè.é. Úyı̀ tá wè.é. Adesuwa bàá ı́rè.n òhó! ghé.
Ozo thinks that Uyi say that Adesuwa accuse him.log of.lying
‘Ozo thinks that Uyi said that Adesuwa accused him (Ozo or Uyi) of lying.’
So there is no condition on third person pronouns that they must be bound by the
closest operator of a particular kind. When taken into the domain of heads that
have first, second, and third person features (agreeing heads), this means that
first and second person heads are subject to a strengthened form of the locality
condition, in which every head blocks feature licensing regardless of its own
features. However, third person heads are not subject to a strengthened form of
the locality condition, because there is no locality condition on the licensing of
third person features in the first place; they are simply the default values. Thus,
the agreement asymmetry in (84) is found to be of a piece with the binding
theoretic asymmetry in (82) versus (83).
(84) a. Hamisi yu-Ø m-refu. (Swahili)

Hamisi cl1-be cl1-tall
‘Hamisi is tall.’
b. *Ni-Ø ni-refu. (OK: ni-Ø m-refu)
1sS-be 1sS-tall
‘I am tall.’
To the extent that you believe that there is a real parallelism here – that it is
no coincidence that first and second person elements differ from third person
elements in both of these respects – I have succeeded in explaining the spe-
cial properties of first and second person agreement in terms of more general
facts.
Appendix: a note on anaphoric agreeement and the SCOPA

Some languages seem to have a special form of object agreement that is triggered
by anaphoric pronouns as opposed to ordinary pronouns. The Swahili example
in (85) is a case in point (see Woolford 1999 for discussion).
(85) Ahmed a-na-ji-penda (mwenyewe). (Vitale 1981:137)

Ahmed 1S-pres-refl.O-love himself.
‘Ahmed loves himself.’
A possible approach to this phenomenon is to say that the +anaphoric feature

of the (overt or covert) object in (85) can count as a ϕ-feature, and v agrees
with the object in that feature.
If this is the correct analysis of at least some instances of anaphoric agreement,
we can then ask what sort of ϕ-feature +anaphoric is. Is it akin to the features
+1 and +2, and thus subject to the SCOPA, or is it akin to number and gender
features and not subject to the SCOPA?
The existing empirical evidence suggests that agreement in +anaphoric is
restricted by the SCOPA. For example, the Bantu languages have special reflex-
ive forms of object agreement, but they do not have special reflexive forms of
adjectival agreement. This is shown in (86b) from Chichewa.
(86) a. Ndi-na-i-khal-its-a pro[CL4] y-a-i-kali.

1sS-past-4O-become-caus-fv cl4-assoc-cl4-fierce.
‘I made them (e.g. lions) fierce.’
Appendix: a note on anaphoric agreeement and the SCOPA 151
b. Ndi-na-dzi-khal-its-a pro[+ana] w-a-m-kali.

1sS-past-refl-become-caus-fv cl1-assoc-cl1-fierce
(*dz-a-dzi-kali)
refl-assoc-refl-fierce.
‘I made myself fierce.’
(86a) shows an example of a nonanaphoric null pronoun that functions as both

the object of the causativized verb ‘become’ and the subject of the predicate
adjective ‘fierce’. (Probably ‘become’ selects a small clause complement that
has ‘fierce’ as its predicate, and the subject of the small clause is exceptionally
case-marked by the causative morpheme. The details of the structure are not
crucial, however.) (86b) displays a minimally different example in which the
null object is anaphoric, understood as being coreferential with the subject of
the sentence. This null anaphor triggers special anaphoric agreement on the
verb, but not on the predicate adjective. Rather, the agreement on the adjective
can only show the number and gender of the null anaphor (here singular and
human). This looks like another instance of partial agreement: just as verbs can
agree with NPs in the feature +1 but adjectives cannot, so verbs can agree with
NPs in the feature +anaphoric but adjectives cannot. This parallelism suggests
that the SCOPA can be generalized to read as follows:
(87) F can agree with XP in +1, +2, or +anaphoric only if a projection of F

merges with a +1, +2, or +anaphoric element and F projects.
A wider range of data also supports the generalization in (87). For example,
it is known in the literature on the Person Case Constraint that just as the first
object of a double object construction can trigger person agreement on the verb
and the second object cannot, so the first object of a double object construction
can trigger a reflexive form but the second one cannot (e.g., Bonet 1991:192,
Anagnostopoulou 2003:254). Thus, Southern Tiwa has inflected verbs that mean
‘I gave myself a shirt’, but it does not have inflected forms that can mean ‘I
gave myself to John’ (Allen et al. 1990). Similarly, Bruening (2001) shows
that true long distance agreement is not possible with anaphoric arguments
in Passamoquoddy, any more than it is possible with first and second person
arguments. Thus, agreement in +anaphoric is absent in the same range of
syntactic structures that agreement in +1 and +2 is absent in. In contrast,
possessive determiners can show special +anaphoric forms in languages like
Greenlandic (Bittner 1994) and adpositions can show special +anaphoric forms
in languages like Slave (Rice 1989). These are functional categories that can
manifest first and second person agreement as well (see sections 4.1.1 and 4.1.2).
So the evidence that +anaphoric agreement has the same limited distribution
as first and second person agreement seems quite good.
Given this, it is natural to ask whether the SCOPA as it applies to the
+anaphoric feature can be derived by techniques similar to the ones I used to
derive the SCOPA for the features +1 and +2. The answer is “I hope so.” The
extension is far from trivial, because the licensing conditions in (30) clearly do
not extend to anaphoric pronouns. In other words, there is no condition exactly
like (88) in the grammar:
(88) A phrase X is +anaphoric only if it is locally bound by a designated
operator or by another element that is itself +anaphoric, and there is no
other operator of that sort that c-commands X but not its local binder.
In general, anaphors need to be bound by a nearby NP in argument position,

not by an operator, and the antecedent NP itself does not need to be anaphoric
in any sense. The good news is that the general condition on anaphors does not
have to be exactly (88) in order to have the desired effect. Any condition of the
form in (89a) should turn into the SCOPA-equivalent statement in (89b) when
the law of strict locality for heads ((69)) applies to it.18
(89) a. An element X is +anaphoric only if it is locally bound by SOMETHING
or by an another element that is itself +anaphoric, and there is no other
THING OF THAT SORT that c-commands X but not its local binder.
b. A head X is +anaphoric only if it is locally bound by another element
that is itself +anaphoric, and there is no other head that c-commands X
I am optimistic that a suitable condition like (89a) can be found to achieve this
result, since anaphors are by definition elements that need nearby antecedents.
But completing the account is complex and leads into the deep waters of prin-
ciple A of the binding theory – in part because natural languages have different
kinds of anaphors (e.g. the SE, SELF, and SE-SELF anaphors of Reinhart and
Reuland 1991), each of which is subject to a somewhat different locality condi-
tion. Taking this on properly would take us far astray, into topics that I am not
well qualified to discuss. Therefore, I leave the task of explaining this aspect of
the SCOPA to future research.
18 Another challenge for deriving (89b) would be to make sure that the “something” mentioned
in (89a) is never close enough to the functional head to trigger +anaphoric agreement on the
functional head directly, without there being a +anaphoric NP nearby for the head to agree with.
5 Parameters of agreement
5.1 Introduction: parameters and other kinds of variation

In the preceding chapters of this book, I concentrated on the putatively uni-
versal conditions on agreement – those conditions that shape agreement and
concord in all natural human languages. But agreement is a relatively superficial
phenomenon, where one expects to see variation as well as uniformity.
It is perfectly clear that there is significant variation at some levels. The most
obvious type is the variation between languages with a great deal of agreement,
where agreement is central to the expression of grammatical relationships, and
languages where agreement is either absent or peripheral to the expression of
grammatical relationships. The first class of languages are the head-marking
languages of Nichols 1986, including the polysynthetic languages of Baker
1996; the second class of languages includes the isolating languages and pure
dependent-marking languages. I have modeled this dimension of variation in
part by saying that languages can vary as to whether or not functional heads
like FA or FN are generated above APs and NPs, and in part by saying that
functional heads may or may not be specified as being probes, which look for
features to agree with.1 For example, Ps, Ts, and vs are functional heads with
semantic content that agree with nearby DPs/NPs in some languages but not
others. A language has lots of agreement if it includes suitable functional heads
in its syntactic structures and those functional heads are designated as being
probes. In contrast, a language has little or no agreement if either functional
heads are not projected, or if the functional heads that are present do not count
as probes. We might then expect to find a continuum in how much agreement
behavior languages manifest, and we more or less do. Of the 108 languages
surveyed below, 26 have no agreement, 9 have agreement on one functional
head (T or in one case D), 30 have agreement on two heads (T and v or D), 26
1 Functional heads can also be specified as probing more than once, as discussed in section 3.3.3
in the context of double object constructions. The possibility of T agreeing more than once plays
an important role in what follows, especially in the analysis of Nez Perce in section 5.11.2.2.
153
have agreement on three heads (T, v, and one other), 12 have agreement on four
functional heads, and 4 have agreement on all five functional heads considered
(T, v, D, P, and C). This looks like a normal distribution, centered on languages
in which about 50% of the functional heads are agreers, plus an unexpectedly
high number of languages that have no agreement at all.2
There is also a great deal of variation when it comes to how agreement is real-
ized morphologically in particular languages. Languages obviously vary as to
which phonological features are used to represent agreement in a particular cat-
egory. For example, first person plural subject agreement on verbs is realized as
mos in Spanish, yakwa in Mohawk, tu in Kinande, iñ in Mapudungun, and so on.
Other superficial aspects of morphological variation include whether agreement
is realized as a prefix or a suffix, details (or at least idiosyncracies) as to where
in the morphological word it appears, which forms have zero exponence, and
instances of syncretism, which may be patterned or idiosyncratic. I assume that
these matters are handled by a postsyntactic realizational morphology, like that
developed by the Distributed Morphologists (Halle and Marantz 1993). I thus
continue to assume that the syntax decides which head agrees with which NP in
which features, whereas the postsyntactic morphology decides how the features
that a head acquires by agreement are spelled in morphemes.3 There may be
cases in which it is somewhat unclear which component is responsible for
handling a particular aspect of agreement. For example, one might debate
whether a head fails to agree with a particular noun phrase, or whether it agrees
with that noun phrase but the agreement is spelled out morphologically as zero
(see Baker 2006 for one case study). One might also debate whether there is a
syntactic basis for certain patterns of syncretism (for example, if a voice alter-
nation is at work in a language with animacy hierarchy effects), or whether
it is purely a matter of morphology (see, for example, Baker 2003b on the
animacy hierarchy and agreement in Mapudungun). Relevant cases have been
discussed in the literature, and I do not intend to add to the discussion here. I
simply assume that there is both a syntactic and a morphological component
2 The relatively large number of languages with no agreement at all suggests that there is also a
grammatical parameter that specifies whether or not Agree is operative in the syntax of a given
language. I do not consider this parameter in any detail here, however.
Although the amount of agreement present in a given language may be randomly distributed,
which heads bear the agreement clearly is not. As is well known, T is by far the most common
agreeing head, v and D are intermediate, and P and especially C are uncommon agreement
bearers. I have no explanation as to why this is so.
3 See section 1.5 for a brief comparison of this view with that of Marantz (1991) and Bobaljik (to
appear), who hold that agreement is entirely a PF phenomenon.
Introduction 155
to agreement phenomena, that the distinction between the two is clear enough
in many cases, and that boundary issues will tend to be resolved as we learn
more about both the syntax of agreement and the structure of the morphological
component. In this work, I continue to pursue this overall vision by focusing
on the syntactic aspect of agreement.
The question at hand, then, is whether there are any significant parameters
within the syntax of Agree, apart from issues of whether a particular head agrees
at all and of how any agreement it has is realized morphologically. Something
deserves to be called a grammatical parameter if it concerns what syntactic
configurations undergo agreement, and if it is a relatively general feature of the
language, not one that is tied to a particular head or construction. I argue that
there are at least two such parameters, stated in (1) and (2).
(1) The Direction of Agreement Parameter (preliminary)

F agrees with DP/NP only if DP/NP asymmetrically c-commands F.
(2) The Case-Dependency of Agreement Parameter
F agrees with DP/NP only if F values the case feature of DP/NP or
vice versa.
These parameters are language-particular narrowings of the universal condi-

tions on agreement. (1) is a more specific version of the c-command condition,
which states that F may probe upward or downward for an NP to agree with.
Some languages restrict this further, saying that agreement can only be upward.
See section 5.10 for a refinement of parameter (1); there I show that there is
also at least one language (Burushaski) that requires agreement to be downward,
not upward. (2) can be thought of as a version of Chomsky’s activity condition,
which says that an NP is visible to agreement only if it has an unvalued case
feature. Both conditions are general parameters in that I take them to hold of
all agreement relationships, not just those that involve a particular functional
head. (1) is also syntactic inasmuch as it depends on the syntactic relation-
ship of c-command, defined over phrase structures. (2) is syntactic inasmuch as
the assignment of case is a syntactic matter, although this can be debated (see
Bobaljik, to appear).
It is very controversial whether grammatical parameters of this sort exist
at all. Much current research in the field has been guided by the idea that
all parameters are fundamentally lexical in nature. More specifically, many
researchers hold that all syntactic variation is attributable to variation in the
features of individual lexical items, the syntactic principles themselves being
invariant, as stated in (3).
(3) The Borer-Chomsky Conjecture

All parameters of variation are attributable to differences in the features of
particular items (e.g., the functional heads) in the lexicon.
I refer to this view as the Borer-Chomsky Conjecture for the following reasons.
It was first suggested, so far as I know, by Borer (1984), shortly after the
introduction of the original GB notion of parameter (Chomsky 1981), when
cracks in the classical Pro-drop Parameter were already beginning to show (see
also Fukui and Speas 1986, Webelhuth 1992). Chomsky (1995) then adopted it,
incorporating it into his Minimalist Program, and many learned of it from there.
Richard Kayne has also done much to promote it, clarify what it amounts to,
and find interesting examples that illustrate it (see, for example, Kayne 2005).
I refer to (3) as a conjecture because it clearly has that honorable status: it
was posited in advance of the wealth of research in comparative syntax done
over the last 10–20 years, and has guided how much of that work has been
done. It is an intriguing generalization over a few suggestive examples that
were available in the 1980s and early 1990s, which makes sense and has some
attractive conceptual properties. Part of my interest in defending (1) and (2)
is that they raise the very real possibility that the Borer-Chomsky Conjecture
might turn out to be false – that there is variation in the principles of syntax as
well as in the lexicon.
My discussion proceeds as follows. First I consider the major functional
categories one at a time, to see how the parameters in (1) and (2) apply to each.
For each category, I begin by contrasting agreement in two large and relatively
well-studied language families, Bantu (Niger Congo, NC) and Indo-European
(IE). I show that many Bantu languages systematically obey (1) and not (2),
whereas many Indo-European languages obey (2) and not (1).4 By first looking
at examples from these two families in some depth, it is possible to see the
full range of consequences that the choice of (1) versus (2) can have. Then for
4 Neither the Bantu family nor the Indo-European family is completely homogeneous with respect
to these parameters, however. While most NC languages that I have checked obey parameter
(1), Makhuwa on the eastern periphery of the Bantu area does not; see section 5.11.3.1 for
discussion. Similarly, while most IE languages obey parameter (2), Nepali and Maithili on the
eastern periphery of the IE area seem not to, based on the data discussed by Bickel and Yadava
(2000). Subject agreement on T in Nepali can agree with an ergative case subject as well as
with a nominative subject (p. 348). “Object” (nonsubject) agreement in Maithili can agree with a
direct object, an experiencer, an indirect object, or even an NP in ablative case (p. 349). Finally,
Maithili has raising constructions in which the raised nominal triggers agreement in both the
embedded clause and in the matrix clause (p. 362). Nepali and Maithili are different in all these
ways from Hindi, not to mention the European IE languages. In the text I often say for simplicity
that NC languages work one way and IE languages work another way, and that is largely true,
but is a bit of an idealization.
Agreement on tense 157
each functional category, I discuss how the same parameter values show up in
other relevant languages from the core 100-language sample of the World Atlas
of Language Structures (WALS) (Haspelmath et al. 2005). This accomplishes
several goals. First, it shows that the issues raised are not parochial to the NC and
IE languages. Second, it gives a sense of how the same agreement parameters can
be realized in otherwise typologically distinct languages, which have different
basic word orders and case-marking systems. Third, the additional languages
may show combinations of parameter settings that are not attested in the NC
languages and the IE languages – languages in which both parameters are set as
yes, or both are set as no. Once my tour of the functional categories is complete,
I come to the crucial question for the theory of parameters: the question of
whether languages have consistent settings for the parameters in (1) and (2),
or whether different functional categories can have different parameter settings
within the same language. To evaluate this, I present statistical information
from a study of the 100 languages of the WALS core language sample, showing
that languages with consistent parameter settings are much more common than
languages with inconsistent parameter settings. From this I conclude that the
parameters in (1) and (2) are classical GB-style grammatical parameters, which
regulate languages as wholes, and should not be reduced to stipulations about
the feature content of individual lexical items the way that the Borer-Chomsky
Conjecture would have it.
5.2 Agreement on tense

5.2.1 Tense agreement in NC versus IE languages
It is attractive to compare NC languages and IE languages for several reasons.
Both are large language families, so one can be sure that the properties in ques-
tion are robust if they survive or are replicated in many daughter languages
of the original stock. Both are relatively well studied (although the genera-
tive literature on IE languages is huge compared to that on Bantu languages).
Finally, the languages are not wildly different from each other when it comes to
syntactic properties other than agreement. The Bantu languages have the same
general head-initial phrase structure as IE languages like English, Scandina-
vian, Celtic, and the Romance languages, they have roughly the same range of
syntactic categories, and they have many of the same core syntactic processes
(passive, wh-movement, control, raising). These baseline similarities make the
differences that exist with respect to agreement stand out all the more clearly.
Coming then to agreement, another similarity between the Bantu languages
and the IE languages is that in both the finite verb agrees with the preverbal
thematic subject in person, number, and (in Bantu) gender. Simple illustrative
examples from Kinande are:5
(4) a. Omukali mo-a-seny-ire olukwi (lw’-omo-mbasa).
woman.1 aff-1S/t-chop-ext wood.11 lk11-loc.18-axe.9
‘The woman chopped wood (with an axe).’
b. Abakali ba-[a]-gul-a amatunda.
woman.2 2S-t-buy-fv fruit.6
‘The women bought fruits.’
The differences in agreement appear when something other than the thematic
subject moves to Spec, TP, as is allowed to varying degrees in both Bantu
languages and IE languages. For example, some Bantu languages allow locative
inversion, where a locative PP rather than the theme NP moves to the subject
position of a passive or unaccusative verb (Bresnan and Kanerva 1989). When
this happens, T agrees with the fronted locative:
(5) a. Ku-mu-dzi ku-li chi-tsı̂me. (Chichewa, Bresnan and Kanerva 1989:2)
17-3-village 17-be 7-well
‘In the village is a well.’
(compare: Chi-tsı̂me chi-le ku-mu-dzi ‘The well is in the village.’)
b. Pa-m-sikǎ-pa pá-bádw-a
16-3-market-16.this 16/fut-be.born-ind
nkhonya. (Bresnan and Kanerva 1989:9)
10.fist
‘At this market will break out a fight.’
(6) a. Omo-mulongo mw-a-hik-a mukali. (Kinande, Baker 2003c)
loc.18-village.3 18S-t-arrive-fv woman.1
‘At the village arrived a woman.’
b. Oko-mesa kw-a-hir-aw-a ehilanga.
loc.17-table 17S-t-put-pass-fv peanuts.19
‘On the table were put peanuts.’
5 My strategy for picking examples from Bantu languages is the following. Wherever possible, I
illustrate each property of agreement from two languages. One of these is a language on which
some important article has been published on the phenomenon. The other is always Kinande, a
language that I worked on myself and on which I have data for the whole range of agreement
configurations within a single language. I have not studied the NC family in detail to determine
what are the exact boundaries of each phenomenon within that family. The claim is only that
these phenomena are found in multiple Bantu languages and seem to be characteristic of how
many Bantu languages work as opposed to how IE languages normally work.
Similar remarks apply to how I have chosen examples from IE languages: I illustrate with
English examples where possible, and otherwise from the published literature on some other
language. I am not as concerned about showing that constructions in IE languages are widespread,
because this is for the most part well known and uncontroversial.
Also like Chichewa and Kinande in this respect are Swahili and Kilega
(Kinyalolo 1991:17–30).6
Locative inversion also exists in English, and has syntactic properties that
are similar to those of the Bantu construction in many respects (Bresnan 1994).
But there is an obvious difference in agreement: in locative inversion structures
in English, the finite verb agrees with the postverbal nominal, not with the
preverbal locative expression:
(7) a. On the table were/*was (put) some peanuts.

b. On the table was/*were (put) a peanut.
The oblique subject constructions found in many IE languages are similar, with
the dative or ergative subject acting like a locative PP in agreement-relevant
respects. The oblique subject does not trigger agreement on T, whereas the
theme argument bearing nominative case does, as shown in (8) from Icelandic.
(8) a. Henni leiddust þeir. (Taraldsen 1995:307)

her.dat was.bored.by.3pS they.nom
‘She was bored with them.’
b. Henni leið ist bókin sı́n. (Boeckx 2000:356)
her.dat was.bored.by.3sS book self’s
‘She finds her own book boring.’
Oblique subject constructions with these agreement properties also exist in Ital-
ian, Russian (Franks 1995, Sigurdsson 2002), and Hindi (Bhatt 2005), among
others.
There are various ways in which one might state this difference theoretically,
but the parameters in (1) and (2) are one straightforward way. Agreement in
Bantu languages is sensitive to c-command but not case. The moved PP in Spec,
TP c-commands T, but the theme argument that remains in the verb phrase does
not. Therefore, T in Bantu can agree with the PP, but not with the theme NP. IE
languages are not subject to (1), so T can probe downward and agree with the
6 Southern Bantu languages like Sesotho and Zulu are consistent with this, but do not show it as
clearly. In these languages, all locative expressions trigger a single kind of agreement (ho) and
this is also the expletive agreement found in subjectless constructions. As a result, in an example
like (i) it is not clear if ho is T agreeing with the preposed locative, or if it is dummy/default
agreement.
(i) Setófó-ng hó-phéh-el-o-a nama.
7.stove-loc 17S-cook-appl-pass-fv 9.meat
‘On the stove is cooked meat.’ (Sesotho, Machobane 1993:15)
It is clear, however, that the verb does not agree with the postverbal NP nama, as it does in IE
languages. This confirms that the Southern Bantu languages also obey (1) and not (2).
theme NP, even when it remains in the verb phrase. However, T in IE languages
can only agree with a phrase if it values the case feature of that phrase as
nominative. It does this for the theme NP, but not for the fronted PP or quirky-
case-marked subject. Therefore in IE languages agreement can only be with the
theme, not with the PP or oblique NP.7
Another way to think about the contrast between (5)–(6) and (7)–(8) is to say
that locative expressions bear ϕ-features in (some) Bantu languages but not in
IE languages; this is the view of Bresnan and Kanerva 1989 and related work.
But this alternative approach does not generalize to a second kind of inversion
construction, found to varying degrees in some central Bantu languages. These
languages allow the object to move to Spec, TP over the subject under certain
discourse conditions, often to express contrastive focus on the subject. When
this happens, T agrees with the fronted object, not with the thematic subject.
Well-studied examples of this subject-object reversal construction exist in Kin-
yarwanda (Kimenyi 1980) and Kirundi (Ndayiragije 1999); it is also found in
Swahili and to a limited extent in Kilega (Kinyalolo 1991) and Kinande (Baker
2003c).
(9) a. Abâna ba-á-ra-nyôye amatá (Kirundi, Ndayiragije 1999)

children 2S-past-foc-drink.perf milk
‘Children drank milk?’
b. Amatá y-á-nyôye abâna.
milk 3S-past-drink.perf children
‘Children (not parents) drank milk.’
(10) Olukwi si-lu-li-seny-a bakali (omo-mbasa). (Kinande)
wood.11 neg-11S-pres-chop-fv women.2 loc.18-axe.9
‘WOMEN do not chop wood (with an axe).’ (compare (4a))
It is quite foreign to IE grammatical intuitions that there could be “subject”

agreement with the direct object. Direct objects usually cannot move to Spec,
TP in IE languages, but there are some plausible cases. For example, Icelandic
and Yiddish are Germanic languages that show verb second phenomena even in
embedded clauses with an overt complementizer. Diesing (1990) (for Yiddish)
and Rögnvaldsson and Thráinsson (1990) (for Icelandic) argue that the object
7 Note that I do not assume that Agree is a prerequisite for movement, as proposed in Chomsky
2000 and related work. In Bantu examples like (4)–(6), movement to Spec, TP is a prerequisite
for agreement, rather than the other way around in my implementation (see Carstens 2005 for a
different view). In English examples like (7), movement is clearly independent of morphological
agreement. I assume that movement is triggered by a distinct EPP feature of heads like T, and has
nothing directly to do with agreement (on this point, see also Bobaljik and Wurmbrand 2005).
has moved to Spec, TP in sentences that have C–Object–Verb[finite]–Subject

order, such as (11).
(11) . . . az vayn ken men makhn fun troybn oykh. (Yiddish, Diesing 1990:44)
that wine can one make from grapes also
‘(You should know) . . . that one can make wine from grapes also.’
In this respect, then, (11) is syntactically comparable to (9) and (10) in Bantu
languages.8 But the agreement properties are different: in (11), the finite verb
must agree with the postverbal subject, whereas in (9) and (10) the finite verb
must agree with the preverbal thematic object.
This difference between Bantu and IE is parallel to the difference seen in
locative inversion sentences. In both domains, the Bantu languages agree with
the fronted, preverbal phrase whereas the IE languages agree with the postverbal
thematic subject that bears nominative case. But it is not plausible to say that
direct objects in Bantu languages have ϕ-features that can be agreed with,
whereas direct objects in IE languages do not. The NPs that can be in object
position in IE languages clearly do have ϕ-features: these features are active
in NP-internal concord, and they can trigger agreement whenever they appear
in a nominative subject position. In contrast, my parametric proposal based
on (1) and (2) correctly captures the patterns. Whatever lands in the Spec, TP
position in Bantu governs the agreement on T, be it the thematic subject ((4)),
the thematic object ((9) and (10)), or a locative expression ((5) and (6)), because
that phrase alone c-commands T. In contrast, T in IE agrees with the NP which
has nominative case, regardless of whether that NP is above T or below it.
These examples also show the importance of the case condition in (2) in
IE languages. Why does the IE finite verb agree with the subject and not the
object in examples like (11)? The standard answer within Minimalist syntax
involves locality: the thematic subject is closer to T than the object is before
anything moves to Spec, TP. But the Bantu languages show that agreement can
be sensitive to postmovement configurations: movement to Spec, TP clearly
feeds agreement in the Bantu languages, because no phrase meets the condition
of c-commanding T prior to movement. After the direct object moves to Spec,
TP in sentences like (11) in IE (triggered perhaps by topic-focus features), the
subject is no closer to T than the object is. The subject is the closest NP to T
8 There may be a difference in the type of movement undergone: A-bar movement in the Germanic
languages, as opposed to A-movement in the Bantu languages. But it is not entirely clear what
this difference amounts to apart from the differences in case and agreement under discussion.
In any case, the A/A-bar distinction has no direct relevance to agreement in the theory I am
developing.
searching downward, but the object is the closest NP to T searching upward.

So some condition other than simple minimality is needed to determine that T
agrees with the subject and not the direct object in this structure. One could
stipulate that T needs to agree downward in IE languages, but that turns out to be
inadequately general; upward agreement is perfectly fine in predicate adjective
constructions, for example. The case-dependence condition in (2) does the trick,
and proves useful in other domains as well.9
These object-preposing structures also confirm that the case condition in (2)
does not apply in the Bantu languages. How case is assigned in the Bantu inver-
sion constructions is a recalcitrant problem for P&P/Minimalist style theories,
especially in the subject-object reversal construction. Suppose that one carries
over from IE languages the assumption that the NP that agrees with T is the one
that has nominative case, as Ndayiragije (1999:422–4) does. Then the thematic
object gets nominative case in (9) and (10). But what case does the subject get?
There is no comfortable answer. On the one hand, there is no reason to say that
there is a second nominative case. On the other hand, it also seems wrong to
say that it has accusative case. This would lead to a theoretically undesirable
under-restriction of syntactic possibilities. Nor is there any empirical motivation
9 This does not answer the question of why it is the thematic subject that gets nominative case
from T. That could be a straightforward minimality effect: the thematic subject is the closest NP
to T prior to movement. Alternatively, it could be the result of a case-marking algorithm like the
one sketched in Marantz 1991. I do not undertake a full investigation of how case is assigned
here.
In chapter 3, I discussed examples like (i), in which the T node in some nonstandard dialects
of English agrees with the wh-phrase in Spec, CP examples that Kayne (2000) makes much of.
(i) ?the people who Clark think – are in the garden
Now we have to ask in the light of parameter (2) how the matrix T realized on think can agree
with the plural who in Spec, CP, even though that T does not assign nominative case to this
wh-phrase (but rather to the subject Clark). Note, however, that the wh-phrase who in (i) does
bear nominative case, assigned by the finite T in the embedded clause. This seems to be important
to the relative acceptability of (i), since even speakers that accept (i) find (ii) worse (see Kayne
2000:210n.4).
(ii) *the people who(m) Clark think he saw – in the garden
In (ii) the wh-phrase bears accusative case, a different case from the one that the matrix T would
assign, and agreement with this T is bad. (ii) is thus correctly ruled out by the parameter in (2).
So case is relevant even to these marked examples in English. I tentatively assume that, because
the wh-phrase in (i) is nominative, the same case that the matrix T assigns, this T can mistake the
wh-phrase for being its own case assignee, allowing agreement. The fact that the actual assigner
of nominative case is different from the reckoned assigner of nominative case might help account
for why (i) is marginal and limited to particular dialects. (In the standard dialect, (i) is ruled out,
its ungrammaticality perhaps attributable to the parameter in (2), interpreted strictly.)
for saying that the postverbal subject has accusative case (it cannot be realized
as an “accusative” object clitic attached to the verb, for example).
These problems are not so severe if we decouple case from agreement in the
Bantu languages (following in part Ura (1996)) by saying that these languages
are not subject to (2). Two possibilities can now be explored. The first is that the
postverbal subject gets nominative case and the fronted object gets accusative
case in Bantu, just as in the Yiddish example in (11). Case assignment would then
be the same in both language families, although its relationship to agreement
varies. The second possibility is that there is simply no requirement that NPs
get case in Bantu languages. Either path avoids the artificial difficulties created
by assuming that case and agreement are intimately related in all languages
simply because they are related in some.
A third difference between agreement on T in NC versus IE languages is
seen in sentences in which no argument of the verb moves to Spec, TP, that
position either being left empty or occupied by a (possibly null) expletive. In IE
languages, the finite verb in such constructions still agrees with the nominative
postverbal subject:
(12) a. There is/*are a peanut on the table.

b. There are/*is some peanuts on the table.
In contrast, many Bantu languages show only a pleonastic locative agreement

in similar constructions. Zulu seems to be particularly rich in these expletive
subject constructions. In general, when the thematic subject is indefinite it can
remain in Spec, vP, while the verb moves past it to T. When this happens, the
verb does not agree with its subject, but rather displays pleonastic ho agreement
(Doke 1955, 1963). In contrast, definite NPs do move to Spec, TP, and therefore
come before the finite verb. These preverbal NPs do trigger agreement on the
verb:
(13) a. Ku-khuluma aba-khulu. (Doke 1955:13–14)

10S-speak 1.pl-elders
‘There are speaking elders.’
b. Aba-khulu ba-ya-khuluma.
1.pl-elders 1.pl.S-disj-speak
‘(The) elders are speaking.’
Kinande has similar verb-subject sentences, although they are relatively

restricted, being found only in negative or emphatic affirmative sentences. As
in Zulu, the verb agrees with the preverbal subject but not the postverbal one:
(14) a. Mo-ha-teta-sat-a mukali (omo-soko).

aff-there-neg/past-dance-fv woman.1 loc.18-market
‘No woman danced in the market.’
b. Omukali mo-a-sat-ire (omo-soko).
woman aff-1S/t-dance-ext loc.18-market
‘The woman danced in the market.’
See also Collins 2004 on existential constructions in Swahili, Horton 1949 on

Luvale, and Ndayiragije 1999 on transitive expletive constructions in Kirundi,
which have VOS order and no agreement on V with either S or O. Overall, the
frequency and pragmatics of (expletive)-verb-subject constructions seems to
vary across the Bantu languages, but the agreement pattern stays quite stable,
consistently different from the familiar IE pattern. Examples like (13) and (14)
show clearly that the position of the subject relative to T is crucial to whether
T agrees with the subject in Bantu languages in a way that it is not in IE
languages.
There is of course no necessity that a language have any or all of these
inversion constructions. The agreement parameters do not determine whether
these constructions exist, but only what their agreement properties are when
they do exist. The possibility of moving an object to Spec, TP, for example,
seems to be quite restricted in both IE and NC languages: one cannot move the
object to Spec, TP in the presence of a syntactic subject in English, nor as far as
I know in Zulu, Luvale, or Chichewa. Similarly, expletive subject constructions
are allowed with unergative verbs in some IE languages but not others, and the
same holds true for NC languages. A limiting case is Lokaa, a non-Bantu NC
language of Nigeria. Lokaa is typologically similar to Bantu languages in many
ways, including the overall nature of its agreement system (Iwara 1982). But
Lokaa happens not to allow locative inversion, object preposing, or expletive
subject constructions, regardless of what agreement is used. Thus, there are
close equivalents in Lokaa for ‘A fish is in the water’ and ‘Women buy cups,’
but not for ‘In the water is a fish’ or ‘There is a fish in the water’ or ‘Cups buy
WOMEN’ (meaning ‘only women buy cups’) (Ijaja Eno and Alexander Iwara,
personal communication). It is thus impossible to find evidence about the setting
of the agreement parameters from this range of constructions in Lokaa. This
point is of practical importance as we turn to broader typological investigation
of agreement on T. On the one hand, inversion constructions may not be so
common (or so commonly described), limiting the diagnostic utility of this test.
On the other hand, when inversion constructions do exist, they provide some of
the clearest evidence about how the agreement parameters are set.
5.2.2 Agreement on tense in other languages

Now that we have solid motivation for the parameters in (1) and (2), let us
consider how these parameters can be seen in a wider range of languages.
Many languages are like the IE languages in that there is no clear relationship
between word order and subject agreement. Alternative word orders might be
possible, but they do not correlate with changes in what the finite verb agrees
with. In head-initial languages, clauses in which the thematic subject has not
moved to Spec, TP can often be recognized by the presence of verb-subject
order, with or without something else appearing before the finite verb. Such
orders are often the result of either the verb itself moving to T (see, for example,
McCloskey 1996 on VSO order in Irish) or a projection of the verb phrase
moving to Spec, TP (see Massam 2001 on verb-initial order in Niuean), while
the subject stays in a position lower than Spec, TP. When the verb agrees with
the subject in such configurations, it is an indication that downward agreement
is allowed, the Direction of Agreement Parameter in (1) being set to “no.”
Languages of this sort include Hebrew, Arabic, Finnish, Makah, Halkomelem,
some Mesoamerican languages (Nahuatl, Otomi, Jakaltek), and some South
American languages (Yagua, Guaranı́, Wari, Wichı́).
In head-final languages, the order of subject and finite verb will not be so
instructive. One can, however, compare agreement in Subject-XP-Verb clauses,
where the subject comes before everything but scene-setting adverbs, with XP-
Subject-Verb clauses, where the subject is closer to the verb and something
else has scrambled before it. At least some clauses of the latter sort could be
instances of moving XP to Spec, TP, while the subject stays inside vP. If so, and
if the finite verb still agrees with the thematic subject in the same way that it
does when the order is Subject-XP-Verb, then downward agreement is possible
in this language. A likely example of this sort is Amele (New Guinean):
(15) a. Uqa jo ceh-ade-i-a. (Roberts 1987:162)

he house build-3pO-3sS-rpst
‘He built houses.’
b. Jo mel age doda cehe-(Ø)-gi-na. (Roberts 1987:142)
house boy 3p self build-(3sO)-3pS-pres
‘The boys built the house themselves.’
c. Ija wen Ø-te-i-a (Roberts 1987:146)
I hunger give-1sO-3sS-rpst
‘I am hungry.’
(15a) illustrates normal SOV order, with the verb showing subject agreement
with the agentive subject (as well as object agreement with the direct object).
(15b) shows an alternative OSV order, a possible analysis of which is that the
object moved to Spec, TP instead of the subject, as in Kinande and Yiddish.
Nevertheless, the thematic subject still triggers third person plural subject agree-
ment on the verb in this example. (15c) shows a psych predication, where the
reversal of the thematic subject (the state-denoting noun ‘hunger’) and the the-
matic object (the experiencer ‘I’) is normal; here too the thematic subject con-
trols subject agreement related to T rather than the thematic object, despite the
surface positions of these NPs. Other SOV languages in which scrambling does
not affect subject agreement include Khoekhoe, Basque, Abkhaz, Burushaski,
Kannada, Choctaw, Ika, Quechua, and many New Guinean languages. So in
many languages changes in word order do not induce changes in agreement,
suggesting that downward agreement is permitted.
There are other languages that do show a Bantu-like dependency of agreement
on word order, however. The tense markers in Canela-Krahô are a case in
point (Popjes and Popjes 1986). Canela-Krahô is largely head final, but T is
a head-initial particle that remains separate from the verb (compare Vata on
the analysis of Koopman (1984)). Some Ts apparently have an EPP feature that
triggers movement of the subject to Spec, TP, whereas others do not. As a result,
in some tenses one sees Subject-T-VP order (the simple past in (16)), and in
others one sees T-Subject-VP order (the remote past in (17)).
(16) a. Wa i-te po pupun. (p. 176)

I 1sS-past deer see
‘I saw a deer.’
b. Ca a-te ton. (p. 192)
you 2sS-past make
‘You made it.’
(17) a. Pê wa rop cakwı̃. (p. 180)
rem I dog beat
‘I beat the dog long ago.’
b. Pê ca to. (p. 192)
rem you make
‘You made it long ago.’
(16) and (17) also show that there is a difference in agreement that goes along
with the difference in word order. Some of the tenses that trigger NP movement
to Spec, TP also agree with the moved subject, including the past marker te
in (16). In contrast, the tenses that do not trigger NP movement to Spec, TP
never show agreement with the subject, including the remote past marker pê in
(17). This suggests that T can only probe upward for something to agree with
in Canela-Krahô, as in Bantu.10
Jarawara (Amazonian) is like Bantu in that T affixes to the verb, although it
is like Canela-Krahô in being head final. Dixon (2004) describes the difference
between actor constructions and object constructions in this language. In the
actor construction, the agentive subject is the topic of the sentence, it is usually
the first phrase in the clause (85% of the time), and the tense/mood marker
agrees with it in gender. In the object construction, the object is the topic of the
sentence, the object is usually the first phrase in the clause (73% of the time),
and the tense/mood marker agrees with the object in gender.11 (18) shows one
sentence of each type (Dixon 2004:418–19).
(18) a. Mioto Watati awa-ka.

Mioto.m Watati.f see-decl.m
‘Mioto saw Watati.’
b. Watati Mioto hi-wa hi-ke.
Watati.f Mioto.m 3/3.inv-see 3/3.inv-decl.f
‘Watati, Mioto saw.’
I thus claim that whether the object or the subject moves to the specifier of the
highest functional head (Tense or Mood) in Jarawara determines whether the
object or the subject triggers agreement on that head. Jarawara is minimally
different from Amele in this respect, the alternation between SOV and OSV
having no consequences for agreement in Amele (compare (15)).
Apurinã provides another glimpse of how essentially the same distinction
can show up in different ways depending on other syntactic properties of the
language. Abstracting away from right dislocation of the object, Apurinã clauses
alternate between the two very unusual word orders OSV and OVS, as shown
in (19) (Facundes 2000:549–50).
(19) a. Oposo kema pitxi unawa muna txa.

Then tapir penis they bring aux
‘Then they brought the tapir’s penis.’
10 Canela-Krahô also has Ts that trigger NP movement to Spec, TP but do not agree with the
moved NP. This simply shows again that some members of a category might show agreement
and others not in the same language. The point is that agreement is a live option if the NP
c-commands T, but not if the T c-commands the NP.
11 This statement is slightly oversimplified: the higher head Mood always agrees with the “topic”
in Jarawara, but the lower functional head Tense occasionally agrees with the thematic subject
even when the object is fronted. See Dixon 2004 for a full description.
b. Txipoko-ru o-txima-ãpo-ta-pe hãtako-ro.

Fruit-unposs 3S.f-eat.fruit-random-vbzr-perf youth-f
‘The girl would go eating fruit.’
Concomitant with this difference in word order is a difference in agreement: the

verb bears a prefix that agrees with the subject in person, number, and gender in
the OVS order in (19b), but not in the OSV order in (19a). I take this language to
be the mirror image of Bantu. Heads come after their complements in Apurinã,
and (unusually) specifiers are on the right edge of the phrase instead of on the
left edge. Like Bantu, the verb moves to Tense. If the verb moves but the subject
stays in Spec, vP, the word order is OSV as in (19a), just as moving V to T
and leaving the subject in vP derives VSO order in various IE languages. In
addition, there is the option of moving the subject from Spec, vP to Spec, TP,
thereby placing the subject after the finite verb. This derives the OVS order in
(19b), just as moving the subject to Spec, TP derives SVO order in Bantu and
many IE languages. Finally, the moved subject in (19b) triggers agreement on
T and the unmoved subject in (19a) does not; this shows that parameter (1) is
set as “yes” in Apurinã. The Apurinã contrast in (19) is exactly like the Zulu
contrast in (13), once the differences in word order are recognized.
So far I have only considered languages in which one of the agreement param-
eters is active and the other is not. But one would expect the two parameters
to be logically independent. For example, one can readily imagine a language
in which T cannot agree with NP unless NP c-commands T and T values the
case of NP. With this possibility in mind, consider (20) from Turkish (Kornfilt
1997:385).
(20) a. Haydut-lar köy-ü bas-miş-lar.
robber-pl village-acc raid-rep.past-3pS
‘They say that the robbers raided the village.’
b. Köy-ü haydut-lar bas-miş.
village-acc robber-pl raid-rep.past
‘They say that the robbers raided the village.’
In the normal SOV order shown in (20a), the definite subject triggers third person
plural agreement on the verb. But nonspecific indefinite subjects stay inside the
verb phrase in Turkish, showing up left-adjacent to the verb, as shown in (20b).
Along with this word order constraint is the fact that these “low” subjects do not
trigger agreement on the verb.12 Turkish is different in this respect from Amele,
where the low subject continues to trigger agreement on the verb (see (15)).
12 Sentences like (21b) are usually called “subject incorporation” in Turkish. But there is little
evidence that there is any real syntactic incorporation in these examples (although there is
This suggests that parameter (1) is set “yes” in Turkish. But Turkish is also
different from Jarawara and Kinande in that the verb does not agree with the
fronted object in the inverted OSV structures either. Moving the object to Spec,
TP bleeds T-agreement with the thematic subject, but does not feed agreement
with the thematic object. Why not? This can be attributed to parameter (2). The
fronted object in (20b) clearly still has accusative case. The case feature of this
NP is thus not valued by T. If the Case-Dependency of Agreement Parameter in
(2) is also set “yes” in Turkish (but not in Jarawara), then it follows that T cannot
agree with the object in this structure either. T thus has to be a default form
in these “subject incorporation” sentences in Turkish. This then is a plausible
instance of a language in which both agreement parameters are set positively.
The typology of T-agreement would be complete if we found languages in
which both agreement parameters were set “no”. The “no” value of parameter
(1) would imply that agreement does not change with word order. A “no” value
of parameter (2) would imply that agreement does not change with changes in
case marking either. A language that fits this description nicely is Burushaski,
an isolate spoken in the Himalayas (Lorimer 1935). Burushaski has an ergative
case system, in which the subject of a transitive verb has a different case marking
than the subject of an intransitive verb. However, subject agreement on T is not
sensitive to this variation in case; both the nominative subject in (21a) and the
ergative subject in (21b) trigger the same a form of agreement on the auxiliary
and main verb.
(21) a. Jε u:ņε xidmt ε č-a b-a. (p. 317)
I.nom your service do-1sS be-1sS
‘(For these many years) I have been at your service.’
b. Ja be.dpi.ε n ε t-a b-a. (p. 321)
I.erg discourtesy do-1sS be-1sS
‘I have committed a discourtesy.’
Burushaski is markedly different from Hindi in this respect, where verbs agree
with nominative subjects but not with ergative ones, as shown in (22) (Mohanan
1995:100–1).13
“semantic incorporation” in the sense of van Geenhoven 1998). Even if indefinite, vP-internal
subjects in Turkish are incorporated in some sense, that does not by itself explain why the
verb cannot agree with them; indefinite objects in Hindi undergo a similar kind of (pseudo)
incorporation, but the verb can nevertheless agree with them (Mohanan 1995).
13 Also like Burushaski in this respect is the IE language Nepali (Bickel and Yadava 2000:348;
Bobaljik to appear).
Bobaljik emphasizes the fact that T in Nepali can agree with an ergative case subject but not
a dative case subject. This is important motivation for his claim that agreement is sensitive to
(22) a. Niina baalak-ko ut.h aa-eg-ii.

Nina.f.nom boy.m-acc lift-fut-f.sg
‘Nina will lift up the boy.’
b. Niinaa-ne baalak-ko ut.h aa-y-aa. (*uthaa-y-ii)
Nina.f-erg boy.m-acc lift-perf-m.sg lift-perf-f.sg
‘Nina lifted up the boy.’
Hindi is a typical IE language in which agreement is dependent on case valuation

in the way described in (2). The fact that Burushaski behaves differently suggests
that its parameter (2) is set “no.” But agreement is also not dependent on word
order in Burushaski. (23) shows an example with OSV order, where the object
not the subject has moved to the left-periphery; nevertheless, the agreement
suffix still registers plural agreement with the thematic subject.
(23) Ja au.u Pisn-kuts-ε ε -sqnu-mn.
My father Pisan-people-erg 3sO-kill-3pS
‘The people of Pisan slew my father.’
Hence, agreement is not dependent on either structural position or case-sharing

in Burushaski, the fourth logical possibility.14 Other languages of this type
include Georgian and Warlpiri, both of which are languages with free word
order, in which ergative subjects and nominative/absolutive subjects trigger the
same agreement on the verb:
(24) a. Šina.ber-a jagl-s jval-s mi-s-c-em-s. (Georgian)
spinster-nom dog-dat bone-dat(acc) prev-3sO-give-t-3sS.impf
‘The spinster will give the dog a bone.’ (AgrS with nominative ubject)
b. Šina.ber-am jagl-s jval-i mi-s-c-a. (Hewitt 1995:549)
spinster-erg dog-dat bone-nom prev-3sO-give-3sS.perf
‘The spinster gave the dog a bone.’ (AgrS with ergative subject)
a three-way distinction among unmarked case, dependent case, and oblique case (cf. Marantz
1991). I interpret the facts differently. According to my typology, languages that agree with
ergative as well as nominative should be the same as those that agree with dative as well as
accusative, and there is some evidence that this is true (see sections 5.7.2 and 5.11.1). Rather
I accept Bickel and Yadava’s arguments that dative experiencers in Nepali are not subjects for
purposes of control and raising. T fails to agree with them not because they have the wrong kind
of case, but because they are in the wrong syntactic position (consistent with them being PPs,
and hence phase barriers to agreement).
14 This discussion leaves open what does determine which NP T agrees with in a “no-no” language
like Burushaski. Agreement in such languages is not random and unconstrained. The easiest
answer would be to say that T simply probes downward in the pre-movement structure, agreeing
with the first NP it finds – the thematic subject in Spec, vP – regardless of how it gets case or
whether it moves. I do not know any of these languages well enough to know if there are
problems with this simple hypothesis.
Agreement on FA and the formulation of the parameters 171
(25) a. Ngaju Ø-rna wangka-ja. (Warlpiri, Simpson 1991:150)

I.abs past-1sS speak-past
‘I spoke.’ (AgrS with nominative subject)
b. Ngajulu-rlu ka-rna nya-nyi kurdu. (Simpson 1991:100)
I-erg pres-1sS see-npst child
‘I see the child.’ (AgrS with ergative subject)
I conclude that the two parameters in (1) and (2) give us a four-way typology
of the agreement properties of Tense, and that this matches well with the kind
of variation that we observe in the behavior of “subject” agreement in the
languages of the world. The next task is to see whether the same parameters
apply to other functional heads.
5.3 Agreement on FA and the formulation of the parameters

Next let us consider FA , the functional head that realizes agreement on adjec-
tives. This is something of an interlude to the main line of discussion, because
in fact there is little or no observable difference in adjectival agreement between
the NC languages and the IE languages.15 In both families, predicate adjectives
agree with the subject of predication in number and gender (but not person)
and attributive adjectives agree with the noun that they modify in number and
gender (but not person), as seen by comparing (26) in Kinande with (27) in
Spanish. This has been discussed at length in the previous chapters.
(26) a. Aba-kali ni ba-kuhi.
2-women pred 2-short
‘The women are short.’
b. aba-kali ba-kuhi
2-women 2-short
‘the short women’
(27) a. L-as mujer-es son alt-a-s.

the-f.pl women.f-pl are.3pS tall-f-pl
‘The women are tall.’
b. l-as mujer-es alt-a-s
the-f.pl women.f-pl tall-f-pl
‘the tall women’
15 It is conceivable that more data might reveal some second-order differences in adjectival agree-
ment between NC languages and IE languages. For example, I would predict that a raising
adjective could not agree with an NP that stays within its CP complement in a Bantu language,
the way it can in Icelandic (section 3.1.2). But raising adjectives are rare crosslinguistically, so
there may well never be one in this family, making it impossible to check the prediction.
Even though FA does not reveal the existence of the parametric differences
in (1) and (2), these structures do have important implications for how the
parameters in (1) and (2) are stated. It is worth interrupting the presentation
of the parametric differences, then, to make sure that we have an accurate
formulation of these parameters.
First consider the Bantu languages. The facts surveyed in the previous sec-
tion have been known for some time, and others have discussed them. The most
common interpretation has been that all agreement in Bantu languages is agree-
ment between a specifer and the functional category that it is the specifier of.
The theoretical context and exact way of formulating this have varied over time.
For Kinyalolo (1991) and others in the early 1990s, agreement was always the
effect of a specifier–head relationship holding at some level. For him, there was
no parameter of agreement per se; Kilega simply showed more perspicuously
what is fundamentally true of all languages. IE languages were assimilated to
the Bantu case by positing a process of expletive replacement via subject raising
to Spec, TP at LF, so that an agreement configuration between the nominative
NP and T is held by LF in IE sentences like (12) (Chomsky 1986b). More
recent work cast in terms of Chomsky’s (2000, 2001) theory of Agree puts this
another way. Baker (2003c), Collins (2004), and Cartsens (2005) have all stated
parameters roughly like the following:
(28) X agrees (downward) with Y only if Y satisfies an EPP feature of X.
On this view, functional heads agree with NPs inside their c-command domain in
Bantu languages, just as in IE languages. However, whenever there is ϕ-feature
checking between a head and a nominal, there must also be EPP checking. This
means that the agreed-with NP raises to become a specifier of the agreeing head
subsequent to the agreement relationship being established. The end result is
much the same as the earlier theory: agreement holds only between a specifer
and its head in Bantu.
The adjectival constructions in (26) show that this cannot be quite right, at
least if Baker (2003a) and the current work are at all right about the structure
of adjectival phrases. The subject of predication in (26a) is not in a specifier–
head relationship with FA (or A) at any stage of the derivation; rather it is
base-generated higher, in Spec, PredP. In my view, this is crucial to explaining
the fact that most predicate adjectives do not behave like thematically similar
unaccusative verbs with respect to unaccusativity diagnostics. I have also used
it to explain the fact that adjectives cannot agree with their subjects in first
and second person, although thematically similar verbs can. These and other
syntactic differences between adjectives and verbs are unexplained if we say
that the subject of predicate adjectives can be the specifier of FA P/AP. So (28)
cannot be correct, within this framework of assumptions.
Instead, predicate adjective constructions show that functional heads can
probe upward for something to agree with as well as downward. The more
general way to state the parameter that distinguishes Bantu and IE languages,
then, is to say that functional heads always look upward for something to agree
with in Bantu, never downward. This covers the canonical examples in the
previous section, because the specifier of a phrase c-commands the head of that
phrase (and not vice versa). The formulation in (1) is also a little broader than
the one in (28), in that it allows agreement in adjectival constructions such as
(26). This is why I take (1) to be superior to alternatives that have been proposed
in the literature.
Consider next adjectival agreement in IE languages, and the proper formula-
tion of the Case-Dependency of Agreement Parameter in (2). To succeed in my
goal of accounting for concord within the same theory as subject–verb agree-
ment, the Activity Condition of Chomsky (2000, 2001) needs to be considered
more carefully. The core case of agreement for generative theories has always
been agreement between the subject and the finite verb. A robust fact about IE
languages from Icelandic to Hindi (but not Nepali) is that the finite verb only
agrees with a nominative case noun phrase. This generalization is captured by
Chomsky’s Activity Condition: X agrees with Y only if Y has an unchecked
case feature – which is then checked/valued (as nominative) by X (here Tense).
This insures that the finite verb only agrees with the subject, regardless of details
about where the subject appears. This can also be extended smoothly to certain
other situations: for example, the transitive verb (technically v) might only show
object agreement with the NP that it licenses accusative case on, the possessed
noun (technically D) might only show agreement with the NP that it licenses
genitive case on, and an adposition might only show agreement with the NP
that it licenses oblique case on.
But the situation is somewhat different when it comes to agreement on adjec-
tives. The predicate adjective in (27a) agrees with the subject, even though it
does not determine the case of the subject (rather the inflected copula does). In
another environment, the same A could perfectly well agree with an NP that
gets accusative case – for example in the exceptional case-marking structure
shown in (29b) from Icelandic (Thráinsson 1979:361).
(29) a. Marı́a er góð .

Maria.nom is good.f.sg.nom
‘Maria is good.’
b. Ég tel Marı́u vera góð a/ * góð .

I believe Maria.acc to.be good.f.sg.acc good.f.sg.nom
‘I believe Maria to be good.’
Why is case assignment required for agreement with T in IE languages,

but not for agreement with FA ? Previous work has tended to say that there
is something defective about the agreement on adjectives, picking up on the
fact that they do not show person agreement. For example, Chomsky (2000,
2001) claims that an NP must have an unvalued case feature to be visible
for adjectival agreement as well as for verbal agreement. The difference is that
because adjectival agreement is incomplete, lacking a person feature, it does not
value the case feature of the associated NP, whereas T is “ϕ-complete” and does
value the case feature of NP. This proposal is designed to relate two differences
between agreement on adjectives (and participles) and agreement on verbs: the
fact that adjectives do not determine the case of the NP they agree with, and the
fact that agreement between an adjective and an NP does not prevent another
head from agreeing with that same NP (see (27a), where there is agreement
also on the verbal copula). On this view, then, adjective agreement is seen as a
kind of degenerate case, where case-checking is attempted by Agree but does
not in fact take place. But this view about the distinction between adjectival and
verbal agreement unravels when one attempts to derive rather than stipulate the
fact that verbal heads agree in person but adjectival ones do not – a desirable
move for the reasons laid out in chapters 1 and 2.16
In contrast, I want to foreground an obvious fact about adjectival agreement
and case in IE languages that is at best peripheral to Chomsky’s account.17 There
is clearly an important case relationship between the NP and the adjective that
16 See also Carstens 2001 for criticism of Chomsky’s idea that adjectives and participles do not
assign case because they are defective agreers, based on data from Bantu languages similar
to that considered in this chapter. Carstens shows that participles in Bantu languages show
ϕ-complete agreement (see section 5.9.1); nevertheless, they do not assign case to the subject
NP or bleed further agreement between the NP and higher auxiliary verbs any more than
ϕ-incomplete participles in IE languages do. Conversely, she shows that finite Tense is not
really ϕ-complete in IE languages, because it does not show gender agreement. Nevertheless,
finite Tense does assign case and bleeds the establishment of further agreement relationships.
Carstens’s conclusion is that the case-assigning properties of functional heads are not predictable
from the type of agreement that those heads enter into. She thus urges a return to the older view
that some functional heads are simply stipulated as being case assigners (finite T, transitive v)
whereas others are not (FA , participial inflection). I follow Carstens in this respect.
17 In one of the few passages in which Chomsky talks explicitly about case concord (Chomsky
2001:17–18), it works as follows (for participle constructions in Icelandic). First the partici-
ple agrees with the NP in ϕ-features, but there is no case valuation either way. Then a higher
functional head (say T) probes its domain to find something with ϕ-features it can agree with.
The participle head has a partial set of ϕ-features (number and gender, but not person) so the T
agrees with it in (29). Although it is true that the adjective does not assign
case to the noun, the adjective does agree with the noun in case. Intuitively
speaking, the case relationship here has the features flow the other way. With
T–NP agreement, the inherent nominative feature of finite T becomes a feature
of the NP as well; we call this case assignment. With FA –NP agreement, the
nominative or accusative feature already associated with the NP (assigned by
T or v) becomes a feature of the FA P as well; we call this case concord. But the
two processes can both be considered instances of a more general process of
case co-valuation (or case feature unification), in which a category that did not
have a case value acquires one. This is why (2) was stated as it was, repeated
here as (30).
(30) F agrees with DP/NP only if F values the case feature of DP/NP or vice versa.
Adding “or vice versa” to the end of this condition generalizes it from ordinary
instances of subject–verb agreement to cover adjective–noun concord as well.
This innovation is parallel to the innovation I proposed in section 2.3.2 with
respect to the c-command condition on agreement. In standard treatments, F
must c-command NP to agree with it, but I argued that NP c-commanding F is
also sufficient. What is crucial is that the two stand in a c-command relationship,
not the direction of that relationship. Similarly, the standard view has been that
F must value the case feature of NP in order to agree with it. Now I am claiming
that NP valuing the case feature of F is also sufficient. Once again, the essential
requirement (for IE languages) is that there be a case-sharing relationship, not
which way the determination of case goes.
This idea extends easily to attributive adjectives. Why can an attributive
adjective agree with the NP that it is adjoined to in an IE language, even though
it does not determine the case of that NP? The answer is that this is possible
because the modified NP determines the case of the adjoined adjective. The
agrees with it, but it also keeps probing to find a more perfect match. Eventually it finds the NP,
and agrees with that as well. Since T agrees with both the participle and the NP, it assigns the
same case (nominative) to both. On this conception, there is no direct case concord between NP
and the adjective or participle; both get the same case by agreeing with the same verbal head.
A technical problem arises in extending Chomsky’s account to predicate adjective construc-
tions like (29), which also show case concord. This time when the T associated with the copular
verb probes downward, the first ϕ-feature-bearer it encounters is the NP subject of the predi-
cation. Since this has a complete set of ϕ-features, it is not clear why the T should continue to
probe downward to find the additional ϕ-features on FA . Thus, it is not clear why there should be
case concord in this construction. Nor do Chomsky’s remarks about case concord have anything
to offer in explaining the difference between primary and secondary predicates shown in (31).
idea, then, is that agreement in ϕ-features depends on sharing of case features

in a certain class of languages.
Support for this view comes from some details about agreement with quirky
case subjects in IE languages. Sigurð sson (2002) points out that some adjec-
tives can agree with such subjects, whereas others cannot. More specifically,
when the adjective is the main predicate of the sentence, it cannot agree with
a dative case subject. In contrast, adjectival secondary predicates can agree
with the dative case subject in case, number, and gender. Agreeing and non-
agreeing adjectives appear side by side in the same sentences in (31) (Sigurð sson
2002:709-10).
(31) a. Strákunum var mjög kalt, svona fáklæddum.
boys.the.dat. m.pl was.3sS very cold.n.sg so few.clad.dat.pl
‘The boys were very cold so scantily dressed.’
b. Henni verð ur kalt svona fáklæddri.
her.dat. f.sg becomes.3sS cold.n.sg so few.clad.dat. f.sg
‘She will be cold so scantily dressed.’
Why can ‘few-clad’ agree with the subject in (31), whereas ‘cold’ cannot? Part
of the answer I adopt from Sigurð sson’s own discussion. He suggests that a
predicate X cannot agree in case with Y if X determines quirky case on Y. He
writes (p. 710):
The reason for this difference has to do with case-assignment or case-matching.
A primary predicate with a non-nominative subject is itself an assigner or
matcher of the inherent case, whereas a secondary predicate of an argument
so case-marked is not its case-assigner. That is, a predicate cannot agree with
its own case assignee.
It is a lexical property of the adjective kalt in Icelandic that it selects a dative

case subject. Since kalt is responsible for the dative case on the subject, it cannot
turn around and agree in case with that subject. Case can flow either direction,
from the adjective to the NP or from the NP to the adjective, but it cannot flow
both directions at once, from the A to the NP and back to the same A. In contrast,
the secondary predicate ‘few-clad’ in (31) is not responsible for the dative case
on the subject; it can equally well be predicated of a nominative subject. Dative
case does not flow from the secondary predicate to the subject; therefore it can
flow from the subject to the secondary predicate.
Suppose that we take this generalization of Sigurð sson’s to be a theorem
of the true case theory. My primary interest here is to demonstrate the nature
of (2), which expresses the dependence of ϕ-feature agreement on case in IE
languages. The examples in (31) show this dependency rather clearly. When case
theory allows the adjective to agree with the dative subject in case, the adjective
also agrees with the subject in number and gender. In contrast, when case
theory forbids the adjective to agree with the dative subject in case, agreement
in number and gender also fails. Here we see the intimate relationship between
case and agreement in IE languages in a very different way from the familiar
relationship between subject agreement and nominative case. The parameter in
(2) captures the fact that these two types of case dependence occur in the same
class of languages, a pay-off of the quest to unify the theories of agreement
and concord. Vita Markman (personal communication) and Natalia Kariaeva
(personal communication) tell me that the same distinction between primary
predicates and secondary predicates is found in Russian and Ukrainian as well.
This suggests that (31) is not caused by some of the rather unusual properties
of quirky subjects in Icelandic, but may illustrate principles of agreement that
hold more generally of IE-type languages.18
With these clarifications and justifications of (1) and (2) in place, let us return
to the claim that they hold parametrically, not universally. If that is correct, then
we should not see the Icelandic contrast between primary predicates and sec-
ondary predicates in languages that are not subject to (2). The Bantu languages
are not directly relevant to this issue, because they do not have any overt case
marking. But Georgian does have morphologically marked case, and it is like
Bantu in not obeying (2). In particular, verbs agree with their subjects in Geor-
gian regardless of whether they are in nominative case or ergative case (see
(24)). So agreement does not depend on the sharing of a certain kind of case
in Georgian, as it does in IE languages.19 Now predicate adjectives agree with
their antecedents in case and optionally in number in Georgian, both when they
are primary predicates ((32a)) and when they are not ((32b–c)).
18 However, Natalia Kariaeva (personal communication) points out that quirky dative subjects are
less unambiguously subjects in Russian and Ukrainian than in Icelandic (see section 3.3.2 for
some discussion). Given this, it could be that primary predicates agree with a null expletive
subject, and the dative expression has simply been topicalized to clause initial position. If so,
the Slavic analog of (31) does not give crucial support for (2).
See Baker and Kariaeva 2006 for an analysis of instrumental case predicates in the East
Slavic languages within this theory. These instrumental predicates seem to show ϕ-feature
agreement with their subjects even though they do not agree with them in case. The analysis
(due to Kariaeva) is that the instrumental predicates really have a PRO subject, and they agree
with this subject in default case (spelled out as instrumental on As) and hence also in ϕ-features.
19 Other signs of Georgian’s “no” value for parameter (2) are its having full person agreement
on both the auxiliary and the main verb in some complex tenses (section 5.9.2) and its object
agreement not being sensitive to whether the object has dative or nominative case (section
5.7.2).
(32) a. Čven uk’ve did-(eb)-i vart. (Hewitt 1995:50)

we.nom already big-(pl)-nom be.1pS
‘We are already big.’
b. K’ar-s nu da-t’ov-eb gia-s. (Hewitt 1995:562)
door-dat neg prev-leave-fut open-dat
‘Don’t leave the door open.’
c. Me, še-šin-eb-ul-ma, mo-v-Ø-k’al-i. (Hewitt 1995:50)
I.erg alarmed-erg prev-1sS-3sO-kill-aor
‘It was I, alarmed, who killed it.’
This is no different from Icelandic. Georgian also has dative subject construc-
tions that are induced by lexical properties of the primary predicate. But unlike
IE languages, primary predicates can agree with dative subjects in Georgian,
as shown in (33).20
(33) a. Bavšv-eb-s mo-(Ø)-e-c’on-eb-a-t es sačukar-i.
child-pl-dat prev-3O-appl-like-fut-3sS-pO this present-nom
‘The children will like this present.’ (Hewitt 1995:556)
b. g-civ-a; g-cx-el-a (Hewitt 1995:382)
2sO-cold-3sS 2sO-hot-3sS
‘you are cold’ ‘you are hot’ (Literally: ‘it hots to you’)
This is what we expect given that (2) does not hold in Georgian: predicates can
agree with NPs in person and number, whether they agree with them in case
(as in (32)) or not (as in (33)).
5.4 Agreement on complementizers

Next let us survey the effect that the parameters in (1) and (2) have on other
types of agreement, beginning with the relatively rare case of agreement on C.
5.4.1 C-agreement in NC and IE languages

We already saw in sections 4.1.5 and 4.5 that there is person agreement on C
both in certain NC languages and in certain IE languages. But what is agreed
with is different in the two language families. In the NC languages, agreement
20 The details of agreement in Georgian – including theoretical questions about how T and v
contribute to that agreement – are notoriously complex. Note in this regard that the agreement
with dative subjects in (33) is formally “object” agreement, not “subject” agreement. This
interesting fact needs to be explained in a fuller treatment, but should be orthogonal to the point
at hand. Also the primary predicates in (33) are apparently verbs, not adjectives, but that too
should also be irrelevant; verbs that take quirky case subjects do not agree with those subjects
any more than adjectives do in (2)=“yes” languages like Icelandic.
Agreement on complementizers 179
seems to be with the matrix subject of a verb of speech (Kinande; see also
section 4.1.5 for Lokaa examples):21
(34) a. Mo-ba-nyi-bw-ire ba-ti Kambale mo-a-gul-ire eritunda.

aff-2S-1sO-tell-ext 2S-that Kambale aff-1S-buy-ext fruit
‘They told me that Kambale bought fruit.’
b. Mo-n-a-layir-ire Kambale in-di a-gul-e amatunda.
aff-1sS-t-convince-ext Kambale.1 1sS-that 1S-buy-sbjn fruits.6
In the West Germanic languages, agreement is with the subject of the embedded
clause – the same NP that the lower verb agrees with:
(35) a. Kvinden dan die boeken te diere zyn. (West Flemish)

I.find that.pl the books too expensive are
‘I find that those books are too expensive.’ (Carstens 2003:393)
b. datt-e wiej noar ’t park loop-t (Hellendoorn)
that-pl we to the park walk-1pS
‘that we are walking to the park’ (Carstens 2003:397)
It is appropriate, then, to ask if this difference in the agreement behavior of

complementizers is related to the known difference in agreement on T between
the two language families.
To answer this question, we need to add in more details about the structures.
Given the analysis in chapter 4, the complementizer in Kinande actually agrees
with a logophoric operator in Spec, CP. This logophoric operator is controlled
by the subject of the matrix verb, and therefore inherits the person, number,
21 As in many languages of Africa and elsewhere, the complementizer in (34) is cognate with
the verb meaning ‘say.’ One could thus question whether (34) really shows complementizer
agreement, or whether these are merely instances of subject agreement on the second verb of
some kind of complex verbal construction. The problem with this alternative is that it is very
unclear what the syntactic relationship between the second VP and the first could be. One can
extract wh-phrases out of the embedded clause in Kinande, just as one can from complement
clauses in English:
(i) Eki-hi ky-o Kambale a-ku-bwir-a a-ti n-a-gul-a?
7-what 7-foc Kambale 1S/t-2sO-tell-fv 1-that/say 1sS-t-buy-fv
‘What did Kambale tell you that I bought?’
This rules out the possibility that ti heads a VP that is conjoined with or adjoined to the VP
headed by ‘tell’; if so, (i) would violate the Coordinate Structure Constraint or the Adjunct
Island Condition. Conceivably (34) could be some kind of serial verb construction, but Kinande
does not otherwise have such constructions. I therefore assume that the most plausible analysis
is that ti heads a CP and its relationship to the verb ‘say’ is purely historical. (The same issues
arise with respect to the Arapesh example in (40) below.)
and gender features of that subject. The syntactic representation of (34b) is thus
approximately (36).
(36) [Ii convinced Kambalek [CP Logi C<agri > [TP hek T<agrk >-buy fruits]]]
In contrast, the complementizer in West Flemish agrees with the subject below
it. This by itself is enough for simple number–gender agreement. When first or
second person agreement is seen, I claimed in section 4.5 that the agreement
gets enriched with first and second person features by virtue of C’s proximity
to the S (speaker) or A (addressee) operator generated in CP. The C counts as
a variable indirectly dependent on this operator, given that Agree makes it a
variable dependent on the lower subject and that subject is first person in (35b),
hence bound by S. So schematic structures for (35) are as in (37).
(37) a. [Ii find [CP Si that<agrk > [TP the booksk too expensive are-T<agrk >]]]
b. [(Theyk think) [CP Si that<agri > [TP wei to park walk-T<agri >]]]
The first parametric question that these structures raise is why can’t C agree
with S or A in IE languages when the lower subject is something other than a first
or second person pronoun? A plausible answer is that this is a reflex of the case
dependence of agreement in IE languages. Many researchers have proposed that
finite that-type complementizers participate with finite Tense in the assignment
of nominative case (Stowell 1981, Watanabe 1996, Pesetsky and Torrego 2001).
Without settling on a particular implementation, suppose that some idea in this
family is correct. Then, given (2), C agrees with X only if C assigns (or, better,
contributes to the assignment of) nominative case to X. That is true for C and
the lower subject. But it is not true for C and the S, A, or Log operators in CP.
As null elements with no thematic role, these operators presumably have no
need for case. Therefore, complementizers in IE languages cannot agree with
these operators directly; they can only agree with them indirectly, if they bind
the NP that C is in a case relationship with. The impossibility of examples in
IE that are perfectly analogous to (34) is thus another instance of the dependence
of agreement on case valuation in IE languages.
Consider now Kinande. The complementizer in this language can agree with
an operator in CP without there being any case relationship, because (2) does
not hold in Bantu languages. Rather, the agreement parameter that is relevant to
Kinande is (1), which implies that the C must be asymmetrically c-commanded
by what it agrees with. This means that the Bantu C could only agree with an
operator in Spec, CP or some similar CP-peripheral position. It could not agree
with (say) a third person subject that is internal to the IP complement of C and
distinct from the matrix subject. Such a noun phrase would not c-command C,
and it would have no relationship to any operator immediately contained in CP.
So a structure like (36) is possible in Bantu, but one like (37a).
In summary, we see that IE complementizers can agree with nominative
subjects but not (directly) with operators in Spec, CP, and Bantu languages
can agree with operators in Spec, CP but not (directly) with subjects. On my
analysis, this follows from the same parameters that force T to agree with its
specifier in Bantu but with the nominative subject in IE languages. T and C
thus undergo the same parametric variation. This suggests that the agreement
parameters are not keyed just to one particular functional head, but influence
whole languages.
Kinande also has a second type of complementizer agreement. In focus/wh-
movement constructions there is a C-like particle that comes after the fronted
NP and before the subject of the clause and the finite verb. This particle agrees
in number and gender with the fronted NP:
(38) a. Eritunda ry-o n-a-h-a omukali.

fruit.5 5.foc 1sS-t-give-fv woman.1
‘It’s a fruit that I gave to a woman.’
b. Amatunda w’ omukali a-gul-a.
fruits.6 6.foc 1.woman 1S/t-buy-fv
‘It’s fruits that the woman bought.
c. Ebi-hi by-o Kambale a-gul-a?
8-what 8-foc Kambale 1S/t-buy-fv
‘What did Kambale buy?’
This is entirely consistent with what we know about agreement parameters

in the Bantu languages. Given that it is sitting in Spec, CP (or Spec, FocP),
the focused NP c-commands the C/Focus head, making agreement with that
head possible. In contrast, the Direction of Agreement Parameter predicts that
a complementizer in Kinande could not agree downward with (say) a wh-
phrase that remained in situ. (39) shows that this is correct (compare with the
synonymous (38c)).22
22 See Carstens 2005 for discussion of the same asymmetry in Kilega, where C agrees with fronted
wh-phrases but not with unfronted ones. The Kilega examples are less transparent, however,
because the finite verb also moves to C, with the result that agreement on C is hard to distinguish
from subject agreement on the finite verb (Kinyalolo 1991, Carstens 2005).
If anything, the question particle in (39) agrees with the A operator in CP, since uti comes
historically from u-ti (2sS-say) ‘you say.’ This could be just a frozen form, however, not a true
instance of agreement.
(39) Uti/*bi-ti Kambale a-gul-a ebi-hi?

q/8-q Kambale.1 1S/t-buy-fv 8-what
‘What did Kambale buy?’
There is presumably no case relationship between the Focus head and the
focused NP in its specifier, that NP getting case in the ordinary object posi-
tion prior to wh-movement (if case assignment happens in Bantu at all). But
case is not an issue, because we know that agreement is not contingent on case
in Bantu languages. In contrast, +wh or +Focus complementizers in IE lan-
guages do not agree in ϕ-features with the moved phrase. That follows from
(2), the stipulation that agreement is contingent on a case-valuation relationship
in the IE languages.23
Again it must be pointed out that languages can perfectly well be subject to
the parameters in (1) and (2) without this showing up on their complementizers.
In fact, agreeing complementizers seem to be fairly rare. In the whole IE family,
C-agreement is only known to occur in a few continental West Germanic lan-
guages. C-agreement also seems to be fairly rare in NC languages: Chichewa,
Zulu, Swahili, and Luvale do not have the C-agreement found in Kinande, for
example. Like Kinande, the complementizer in these languages is historically
related to the verb ‘to say,’ but in these languages it is the infinitival form of the
verb that is used, a form that does not vary with the ϕ-features of operators in
CP. This shows us once again that the grammatical parameters do not determine
whether a complementizer agrees or not, but rather what it agrees with if it does
agree.
5.4.2 C-agreement in other languages

A look at C-agreement outside of the IE and NC languages provides some rein-
forcement for these themes. Most languages simply do not have C-agreement,
and this gives no information about how the agreement parameters are set in that
language. I found agreement on C in at most 9 of the 108 languages I surveyed.
Among the few languages that have C-agreement, the one that most clearly
has Bantu-like upward agreement with an operator in CP is the New Guinean
language Arapesh (Fortune 1977:98):
23 In some approaches to complementizer-trace phenomena, it is crucial that IE Cs agree with their

specifiers so that they can license a trace in the subject position immediately below them (Rizzi
1990). Such an approach is not necessarily incompatible with my parameter. The instances of
C that need to have agreement in order to ameliorate complementizer-trace effects are always
in structures that have local subject extraction. The finite C participates in assigning nominative
case to the extracted subject, so the IE agreement condition in (2) is satisfied. Other wh-phrases
could not trigger agreement on C in an IE language, if I am right, but their traces are also not at
risk of creating a complementizer-trace violation.
(40) De ja-kalipw-e ja-ka akir urur re-igure.

now 1sS-tell-2sO 1sS-that(say) your vulva 3sS-cry.out
‘I told you that your vulva cried out.’
Agreement in Arapesh is like agreement in the Bantu languages in other respects

as well, as I show below.
Another possible instance is the so-called allocative agreement in Basque,
where matrix verbs can show a kind of second person agreement with the
addressee of the sentence, in addition to standard agreement with the subject,
object, and direct object (Ortiz de Urbina 1989:62n.6).
(41) Z-e-ki-na-t (compare: d-a-ki-t)
3A-past-know-2s.dat-1sE 3A-pres-know-1sE
‘He knows me, y’know.’ ‘He knows me.’
This could be analyzed as agreement between the matrix C and the A operator
at the head of the sentence, which is realized on the inflected verb as a result of
V-to-C movement.
In certain other languages, agreement on C seems to be downward, with the
subject of the TP complement of C. The New Guinean language Daga has no
C with indirect speech complements, but in relative clauses there is a second
suffix on the verb, in addition to T, that (like T) agrees with the subject of the
relative clause in person and number. Some examples are (Murane 1974):
(42) a. pa Dani tu-n-i
house Dani build-3sS.past-3s.that
‘the house that Dani built’
b. at gega gat yon-an-a
place you just stand-2sS.past-2s.that
‘the place where you stood’
Another language with downward agreement on C could be Jakaltek (Craig

1977), assuming that the very high “aspect” head that bears agreement with the
absolutive argument in this language is a type of C, a head higher than T.
Next let us consider what patterns of C-agreement we would expect to see
in languages that set both agreement parameters as “yes” or both as “no.” In
a “yes-yes” language, a C would agree with an NP only if that NP was in
Spec, CP and if C was in a case-valuation relationship with that NP. But Cs
don’t generally enter into case relationships of their own: they neither assign
case (apart from T) nor manifest it in relationships of case concord. To the
extent that this observation is universal, a “yes-yes” language could not have
agreement on Cs. That might be a mildly positive result in some cases. For
example, I claimed in section 5.2.2 that Turkish was a “yes-yes” language,
based on the behavior of T agreement. It is also true that Turkish does not have
agreement on C, as predicted. But this is hardly a stunning confirmation, given

that C agreement is rare in languages of all types.
What about a “no-no” language, in which agreement can be downward and
is not sensitive to case? Cs in that sort of language could agree with NPs inside
their CP complement, but they would not necessarily be restricted to agreeing
with the nominative subject, the way Cs in IE languages are. One language that
might take advantage of this freedom is Nez Perce. In addition to the agreement
for subject and object found on finite verbs (see section 5.11.2.2), this language
has a variety of initial particles that mark negation, questions, and relative
clauses (Aoki 1973:128-9). These C-like particles agree with something inside
the clause that they introduce. The agreed-with NP is often the subject, as in
(43a) and (43b), but question and relative particles can agree with the object
instead of the subject when the subject is third person and the object is first or
second person, as shown in (43c).
(43) a. Mı́ʔ s-e:x ʔ ı́:n manmaʔ ı́ Ø-ki-yúʔ .
not-1s I anyway 1sS-go-fut
‘I cannot go in any way.’
b. We:t-e:x ʔ e-né:s-tiwı́kce?
q-1s 1S/3O-pl-accompany
‘Shall I go with them?’
c. We:t-e:x hi-twı́:kc-ix?
q-1s 3S/1O-accompany-pS
‘Would they go with me?’
Thus, agreement on Cs in Nez Perce is not restricted by case or by direction.

Overall, then, we find the same four types of agreement with complementizers
that we do with Tenses, even though C-agreement is much rarer, with the caveat
that agreement on C cannot arise in a “yes-yes” language.
5.5 Agreement on determiners

Next I consider the possibility of agreement on determiners and similar func-
tional heads in the extended projection of NP, beginning as usual with NC and
IE languages.
5.5.1 D agreement in Bantu and IE languages

A widespread pattern in Bantu languages is for words inside the noun phrase to
follow the noun and agree with it in noun class (gender and number). This is what
happens with adjectives, for example, in Kinande, Swahili, Zulu, Chichewa,
Luvale, and others:
Agreement on determiners 185
(44) aba-kali ba-kuhi (Kinande)

2-woman 2-short
‘a/the short woman’
I have analyzed these adjectives as adjuncts to NP. A sign of their adjunct status
within my theory is that they cannot agree in person with a first or second person
pronoun, although they do agree with such pronouns in number and gender (see
sections 2.4.2 and 4.5):
(45) Thin’ aba-khulu si-ya-khuluma. (*si-khulu) (Zulu, Doke 1955:51)

we cl1.pl-big 1pS-disj-speak 1p-big
‘We big ones are speaking.’
Many other NP-internal elements in Bantu languages show the same basic word
order and agreement properties. This is consistent with the view that they may
simply be adjectives syntactically, even though they correspond semantically
to determiners or other functional heads in other languages. This class typically
includes numerals, demonstratives, and possessive pronouns. I tentatively put
all these categories aside as not being relevant to the question of agreement on
D-like heads.
Some Bantu languages also have a small number of elements that have a
syntax that is distinct from adjectives. One such element is ‘all’ in Swahili,
Kinande, and Zulu. Like adjectives, ‘all’ follows the noun and agrees with it:
(46) a. vi-tabu vy-ote; wa-tu w-ote (Swahili, Ashton 1949)

8-book 8-all 2-person 2-all
‘all the books’ ‘all the people’
b. aba-kali b-osi; ama-tunda w-osi (Kinande)
2-woman 2-all 6-fruit 6-all
‘all the women’ ‘all the fruits’
The difference is that ‘all’ can also agree with a pronoun in first or second
person features, at least in Swahili and Zulu.
(47) a. Thina s-onke si-fik-ile. (Zulu, Doke 1963:94)

we 1p-all 1pS-arrive-perf
‘We have all arrived.’ (compare: b-onke ‘all of them’)
b. sisi s-ote; ninyi ny-ote (Swahili, Ashton 1949)
we 1p-all you 2p-all
‘we all’ ‘all of you’
These elements were discussed in section 4.1.2. Within my theory, the difference
in agreement shows that ‘all’ is a true functional head, with the DP ‘we’ in its
specifier position. So the structure of (47a) is (48a), as compared to the structure

for adjectival modifications like (45) shown in (48b).
(48) a. [DP [ProP we] [D all X]]

b. [ProP [ProP we] [AP big]]
(48a) is also consistent with what we now know about agreement parameters in
the Bantu languages. The nominal in Spec, DP c-commands the D head ‘all’,
so agreement is allowed by (1). ‘All’ presumably does not assign case to the
NP in its Spec, but that is not required in a Bantu language. Other words like
‘all’ in Zulu include dwa ‘only, alone’ and bili ‘both’.
There is a third class of noun phrase elements that contrasts with both of the
above two. Examples of this third class are kila ‘every’ in Swahili and obuli
‘every’ in Kinande. Although similar in meaning, kila and obuli differ from ote
and osi ‘all’ in two ways, both illustrated in (49). First, kila and obuli come
before the principal noun of the nominal, not after it. Second, kila and obuli
stand out among the NP internal elements in Swahili and Kinande in that they
do not show agreement with the head noun:
(49) a. kila ki-tabu ; kila m-tu (Swahili, Collins 2004)

every 7-book every 1-person
‘every book’ ‘every person’
b. obuli mu-kali; obuli ri-tunda (Kinande)
every 1-woman every 5-fruit
‘every woman’ ‘every fruit’
The difference in word order (as well as their meanings) suggests that kila and
obuli are not adjectival modifiers; nor are they D-like heads with the NP in
their specifier. The simplest analysis given that these languages have specifier-
head-complement word order is that kila and obuli are functional heads and the
co-occurring NPs are their complements, as shown in (50).
(50) [DP every [NP woman]]
What does this have to do with the agreement difference? Why do NPs before
the D agree with D, but NPs after the D do not in these languages? There is a
clear parallel between this and the fact that subjects agree with the tensed verb
when they come before it in Bantu languages, but not when they come after it
(right dislocation aside). There is also a parallel between this and the fact that
complementizers in Kinande agree with moved NPs that appear to their left,
but not with in-situ wh-phrases that remain to their right. I claim that this is
a third reflection of the parameter in (1), which says that agreement requires
c-command in Bantu. When NP is in Spec, DP, it c-commands D and agreement

is possible. When NP is the complement of D, it does not asymmetrically c-
command D and agreement is not permitted in most Bantu languages.
To get this symmetry to follow from the same principles as the others that
we have considered, some fine print becomes relevant. The background to (1)
is that Universal Grammar allows a functional head to search for something to
agree with in two directions: upward through the structure for something that
c-commands it, or downward for something that it c-commands. The idea of (1)
is that some languages restrict agreement to the first case only, permitting only
an upward search. The question of detail that has not arisen yet is what happens
when the agreed-with nominal is the complement of the agreeing functional
head and hence both c-commands it and is c-commanded by it. Does this count
as an upward search, or a downward search, or can it count as either? We
now see that there is an empirical answer to this question. What we observe in
these Bantu languages is that agreement seems not to be possible in a relevant
configuration like (50). So we need a head agreeing with its complement to be
an instance of downward search, but not an instance of upward search. A simple
way to do this, already stated in (1) but not needed until now, is to require that
the agreed-with NP asymmetrically c-command the agreeing functional head
in this type of language.
Now compare Swahili and Kinande with the Indo-European languages. In
the IE languages, there is no apparent correlation between word order inside
DP and whether the D agrees with the noun or not. In particular, many IE
languages have D-NP order, in which the D agrees with NP – the forbidden
structure in Bantu. This is found even in English with demonstratives (this book
vs. these books), and more robustly with all sorts of determiners in the Romance
languages:
(51) a. el muchacho ‘the boy’ (M, sg) (Spanish)

b. la muchacha ‘the girl’ (F, sg)
c. los muchachos ‘the boys’ (M, pl)
d. las muchachas ‘the girls’ (F, pl)
This is what we expect given the agreement parameters in (1) and (2). There
is no requirement that the agreement trigger asymmetrically c-commands the
agreement bearer in IE, so agreement is not ruled out on these grounds. There is
a requirement that one of the members of the Agree relationship values the case
feature of the other. But this is satisfied in IE languages, as seen by the fact that
determiners agree in case with their NP complements in IE languages that have
both overt determiners and overt case marking, such as Greek:
(52) a. I Maria aghapuse ton

the.f.sg.nom Maria.f.sg.nom loved.3sS the.m.sg.acc
Petro . . .
Peter.m.sg.acc
‘Mary loved Peter . . .’ (Anagnostopoulou 2003:90)
b. O Giani-s estile tis
the.m.sg.nom John-m.sg.nom sent.3sS the.f.sg.gen
Maria-s to grama.
Maria-f.sg.gen the.acc letter.acc
‘John sent Mary the letter.’ (Anagnostopoulou 2003:9)
Case agreement thus makes ϕ-feature agreement possible on determiners in

IE, much as it does with adjectives.
The statement of the parameter in (1) creates one complication elsewhere in
my analysis of Bantu. The problem arises in the arena of predicate nominals. In
section 2.4.1, I assumed that the FN head associated with a predicate nominal
can agree with its NP complement. Given our current understanding of these
principles, this agreement should not happen in Bantu languages, although it
could in IE languages. But this sort of agreement does take place in Bantu: I
assume that it is the source of the noun class prefixes that manifest the gender and
number features inherent in the noun root. Fortunately, a mechanical solution
is available. We can say that the predicate nominal moves to adjoin to FN P in
Bantu, much as the pronoun might move to Spec, DP in (48a). The structure of
a sentence like (53a) would thus have to be (53b) in the Bantu languages.
(53) a. Dalili y-a mvua ni ma-wingu. (Ashton 1949)

cl9.sign cl9-assoc cl9.rain pred cl8-clouds
b. PredP
NP Pred´
sign of rain Pred FNP

[9,sg]
ni NP FNP
cloud FN NP
[8,pl] [8,pl]
[*9,sg] cloud
AGREE
MOV’T
Since FN is realized as an affix on NP, not as an independent word,24 there is no

direct evidence for or against this proposal, so I adopt it to save the phenomenon.
If, however, languages turn up which are subject to (1) and have FN realized as a
separate particle, this derivation will not necessarily work. It will be important
to check whether in such a language word order and agreement covary for FN
the way they do for determiners in Kinande and Swahili.
5.5.2 D agreement in other languages

What do we find when we look at D heads in languages outside the IE and NC
families? It is relatively common for a language to have agreement on at least
some Ds; this is found in about 40 of my 108 languages. The most common type
is the unremarkable agreement of D with its complement NP in number and
perhaps gender and case (e.g., Yagua, Jakaltek, Nahuatl, Ojibwa, Halkomelom,
Maung, Amele, Georgian, Khoekhoe, Finnish, Arabic). Languages with this
sort of agreement need to be analyzed as allowing downward agreement, as in
Indo-European.
A few languages, however, seem to show an interaction between word order
and agreement similar to what we observed in Bantu. One possible case is
quantificational heads in Tariana. (54a) shows that the quantifier ‘all’ comes
before the NP that it quantifies over and does not agree with it, while (54b)
shows that the quantifier ‘many’ comes after the relevant NP and can bear
number and classifier agreement with it.
(54) a. thuya nawiki (Aikhenvald 2003:219)

all people
‘all (the) people’
b. nu-ita-kanape hanupe-(ma-pe) (Aikhenvald 2003:181)
1sP-daughter-pl many-f-pl
‘my many daughters’
This contrast looks very much like the one in (46) and (49) and should be subject
to the same analysis.
Somewhat similar is Slave. (55a) shows a quantifier that can come before
the associated NP and is invariant; (55b–c) show quantifiers that follow the
associated NP and agree with it. The quantifier in (55c) can even agree in
24 What must be assumed about the morphological union between FN and the noun in Bantu?
One possibility is that FN is a clitic that cliticizes to the left edge of its specifier at PF. Another
is that the head of NP incorporates into FN , and then the NP as a whole undergoes remnant
movement to Spec, FN P. Either could be true, as far as I know, although both raise interesting
further questions.
person, showing that it must be analyzed as a functional head, not as an adjectival

modifier.
(55) a. hoyí t’áhsi ; hoyí i dene (Rice 1989:263)

any thing any person
‘anything’ ‘anybody’
b. dene naáne; tue ndahe (Rice 1989:262)
people some fish some
‘some people’ ‘some fish’
c. dene ʔó hla; ʔó l a-dı́t’e (Rice 1989:264)
people both both-1pS.number
‘both people’ ‘both of us’
Overall, most D-like words in Slave do not show agreement, regardless of

the word order, but the few that do always come after the agreed-with NP.
This suggests that agreement requires upward search, as in Bantu. Greenlandic
Eskimo provides a third example: most nominal-internal functional heads come
after the associated NP and agree with it in number and case, but there are a
few “emphatic words” that can come before the associated NP, and these do not
agree with the NP (Fortescue 1984:110). (See also section 5.11.2.1 on Berber.)
Less striking but potentially relevant are languages in which only one order of
D and NP is attested, but there is agreement and the observed order is consistent
with NP being in Spec, DP. These languages can also be analyzed as obeying
parameter (1), even though language-internal minimal pairs like (54) and (55)
cannot be produced, because of lexical factors. Some of the Bantu languages
fit this description, lacking a word comparable to ‘every’ in Swahili (Zulu,
Luvale); so too do Jarawara, Apurinã, and Arapesh.25
Finally, there are languages which allow both D-NP order and NP-D order
in apparently free variation, the D agreeing with the NP in both orders (e.g.,
Warao, Zoque, Yimas). I tentatively assume that these putative Ds are really
adjectival modifiers, with the syntactic status of adjuncts rather than that of
heads that select NP complements, and thus fall outside the purview of the
headedness parameter. They have a structure like (48b), and are not directly
relevant to the topic of this section.
25 A further complication is that in languages that are generally head final and that have Ds that
follow NP it is not clear from the word order whether the NP is the complement of D or its
specifier. Agreement could thus be analyzed as being either upward or downward in languages
like Wari, Lakhota, Gooniyandi, and Alamblak. I tentatively assume that agreement is downward
unless there is evidence to the contrary, figuring that the NP would only be in Spec if a marked
feature forced it to move there.
Agreement on adpositions 191
Throughout this section, the focus has been on the implications of parameter
(1) for agreement on Ds. In principle, one should consider parameter (2) as
well. In point of fact, however, I do not know of languages that have overt case
marking on both nouns and determiners in which the two ever fail to agree in
case. I thus tentatively assume that D and the NP it selects always share a single
case feature, quite apart from the agreement parameters (perhaps because they
are part of the same extended projection). If this holds universally, then the “only
if” clause in (2) is always satisfied, and the Case-Dependency of Agreement
Parameter has no detectable consequences in this domain.
5.6 Agreement on adpositions

Next let us consider the possibility of agreement on adpositions. My prediction
is that agreement on adpositions should be very similar to agreement on deter-
miners. Like determiners, adpositions normally take NP/DP complements, so
they will generally agree with that NP only in languages where the Direction of
Agreement Parameter is set to “no.” Moreover, since adpositions always case-
mark their NP complements, the Case-Dependency of Agreement Parameter is
always satisfied, so (2) will not give rise to observable crosslinguistic variation.
I show that these expectations are borne out.
5.6.1 P agreement in Bantu and IE

Beginning with Bantu languages, the expectation is that they should not have Ps
that agree with their NP/DP complements in person, number, or gender, because
the complement does not asymmetrically c-command the head. In fact, they
typically do not. Bantu languages have relatively few prepositions, but most
have an instrumental/comitative preposition, such as na in Kinande, and its
cognates in other languages. This P does not agree with its NP complement in
noun class (see also Kinyalolo 1991:111 for Kilega):
(56) Kambale a-ka-kanay-a na-(*bo) aba-syakulu.

Kambale 1S-pres-speak-fv with-2 2-old.people
‘Kambale is speaking with the old people.’
Some, like Chichewa, also have a dative preposition kwa. This too fails to agree
with its complement:
(57) Mbidzi zi-na-perek-a msampha kwa nkhandwe. (Baker 1988:229)

zebras 10S-past-hand-fv trap to fox
‘The zebras handed the trap to the fox.’
The absence of agreement on Ps is rather conspicuous, inasmuch as agreement

is otherwise so pervasive in Bantu languages; almost every other argument
relationship is either obligatorily or optionally marked by agreement. In the
absence of a theory, it might seem odd that Bantu languages do not generalize
agreement to the relationship between a P and its argument. But within a true
understanding of the syntax of agreement in Bantu, based on (1) and (2), this
gap makes perfect sense.
Kinyalolo (1991:106–30) in his discussion of agreement on Ps in Kilega
makes an important observation that increases the interest of this case consid-
erably. He points out that Ps in Kilega do agree with their complements if and
only if the complement has been moved out of PP, either by focus movement
or by NP movement in a passive construction. (58) replicates his observation
in Kinande.
(58) a. Aba-syakulu b-o Kambale a-ka-kanay-a na-bo.

2-old.people 2-foc Kambale 1S-pres-speak-fv with-2
‘It’s old people that Kambale is speaking with.’
b. Abasyakulu si-ba-li-kan-ibaw-a na-bo.
2.old.people neg-2S-pres-speak-pass-fv with-2
‘Old people are not spoken with.’
This shows that it is not enough to rule out agreement in (56) simply by saying
that P is not a head that initiates agreement, or that there is no realization for
third person animate plural agreement on P in Kinande and Kilega. In fact,
there are agreeing forms of the P. The issue is simply that P cannot agree with
NP when the NP is its complement, but only when it has moved to a higher
position in the clause. The Direction of Agreement Parameter in (1) explains
why: agreement requires asymmetric c-command in these languages. Note also
that the contrast between (56) and (58) is parallel to what we have seen with
other agreeing heads in the Bantu languages: like T, C, and D, a P in Kinande
can agree with a phrase that precedes it, but not with a phrase that follows it in
the surface word order of the sentence.
In (58b), the P apparently agrees with the surface subject in Spec, TP. This
raises the question of why the P in an active sentence like (56) cannot show class
1, human singular agreement with its subject in Kinande, resulting in the PP
*na-go abasyakulu. The subject certainly c-commands the P, so the parameter
in (1) is satisfied. It is true that P does not assign case to the subject, but we know
that agreement is not dependent on there being a case-valuation relationship in
Bantu. The most promising explanation for this is probably to say that PPs, like
CPs and some vPs, count as phases in Chomsky’s (2000, 2001) terms. Chomsky
mentions this in passing as a possible way of explaining the fact that movement
out of PPs is banned in many languages. I have also entertained this possibility
as the most straightforward way of explaining the robust fact that T and v never
agree with the NP complement of a P (section 2.3.1). (56) then illustrates the
converse case: just as no outside functional head can agree with an NP inside PP,
so P cannot agree with an NP that is outside PP. Both sorts of agreement across
a PP boundary are universally ruled out by the phase condition on agreement,
a corollary of the Phase Impenetrability Condition.
Why then is agreement allowed in (58)? Inspired by van Riemsdijk’s classic
(1978) study of extraction from PPs in Dutch, the obvious answer is that the
complement of P adjoins to PP on its way to Spec, TP or Spec, FocusP.26
This intermediate stop is needed anyway to satisfy the Phase Impenetrability
Condition, given that PP is a phase. The intermediate position then is one in
which the agreed-with NP asymmetrically c-commands P and is still in the
same phase as P. It is only NPs that occupy this position at some point in the
derivation that can trigger agreement on P in a Bantu language. The crucial
structures are compared in (59).
(59) a. [Kambale T [VP speak [PP with [NP old people]]]]

agreement out by parameter (1)
b. [Kambale T [VP speak [PP with [NP old people]]]]

agreement out by phase condition
c. [old.peoplei T [VP are-spoken [PP t i [PP with t i ]]]]

agreement allowed
In contrast with what we have seen in Bantu languages, we expect that IE

languages could allow a P to agree with its unmoved NP complement. Asym-
metrical c-command is not required in these languages, since (1) is set to “no.”
Moreover, Ps in IE languages assign case to their NP/DP complements, so
(2) should be satisfied. In fact, not many IE languages take advantage of this
26 The reader might wonder why I say that the intermediate position that NPs move through in
pseudopassives and focus constructions is adjoined to PP rather than Spec, PP. Theoretical
motivation for this is my claim that prepositions are functional heads that have the same major
category properties as adjectives, and adjectives by definition do not have specifier positions
(Baker 2003a). Empirical motivation for this is Kinyalolo’s (1991) observation that Ps in Kilega
agree with moved NPs in number and gender but not in person. This suggests that the NP moves
through an adjoined position rather than a Spec position, since agreement in person is possible
with specifiers.
opportunity, but a few do. Welsh in particular has many Ps that agree with
unmoved pronominal complements:27
(60) a. Soniais i amdan-o ef. (Harlow 1981:220)

talked I about-3s.m him
‘I talked about him.’
b. amdanaf i; amdanat (ti); amdani (hi); amdanynt (hwy)
about-1s me about-2s you about-3s.f her about-3p them
(Harlow 1981:249)
This is yet another manifestation of the systematic difference between Bantu

and IE when it comes to agreement.
5.6.2 P agreement in other languages

Turning to other languages, we find that a significant minority of the 108 lan-
guages have some form of agreement on Ps – between 20 and 30, depending on
what one makes of languages in which Ps agree with pronouns only (see note
27). Agreement on P is thus significantly more common than agreement on C,
but not as common as agreement on D. Languages in which P agrees with its
NP argument and there is nothing unusual about the word order that suggests
that the NP has moved must generally be analyzed as permitting downward
agreement. (61) and (62) give some examples.
(61) a. y-ul te’ n̈ah (Jakaltek, Craig 1977:110)

3-in the house
‘in the house’
b. j-iban̈, w-et
1p-on 1s-to
‘on top of us’ ‘to me’
(62) a. awə y a-mʃj taxj (Abaza, O’Herin 2002:54)
that 3s.i-after
‘after that’
b. (sara) s-pnə
I 1s-at
‘at my house, by me’
27 A complication is that Ps do not agree with nonpronominal objects in Welsh: there is no third
person masculine agreement on the P in soniais i am y dyn ‘talked I about the man,’ for
example. Languages that allow agreement with pronouns but not full NPs are not uncommon
(see WALS:map 48); other languages like this in my sample include Persian, Canela-Krahô,
Hixkaryana, Guarani, Choctaw, and Lakhota. In some cases this can be explained by reanalyzing
the putative agreement morpheme on P as a pronoun that cliticizes to the P, but that does not
seem right for all the relevant languages, including Welsh. I leave open the explanation for this
phenomenon.
Other languages that have agreement on P with in-situ objects include Nahuatl,
Arabic, Finnish, Georgian, Burushaski, Tauya, Dani, Lavukaleve, Daga, Yagua,
Pirahã, Wari, and perhaps Greenlandic, Quechua, Wichı́, and Yimas.28
The most compelling evidence for a “yes” setting to the Direction of Agree-
ment Parameter comes from languages that permit an NP to appear outside of
PP, the P agreeing with NP only when this happens. This is not common in my
sample, but is attested. One such language is Slave (Rice 1990:1197):
(63) e ] gondeh.
a. Yahti [ts’éli a ts’´
priest girl to 3.talks
‘The priest talks to the girl.’
e ] gondeh.
b. Ts’éli ai yahti [ei ye-ts’´
girl priest 3-to 3.talks
‘The priest talks to the girl.’
Another is Fijian (Dixon 1988:42):

(64) a. ‘Eimami saa qaaqaa a ‘ai-Boumaa [i-na drano].
we asp victorious art native-place about-art lake
‘We, the natives of Boumaa, were victorious concerning the lake.’
b. A drano ‘eimami saa qaaqaa [‘i-na --] a ‘ai-Boumaa.
art lake we asp victorious about+3.sg art native-place
‘The lake, we the Boumaa people were victorious concerning it.’
These are the languages that are most clearly like the Bantu languages when it
comes to agreement on P.
But it is not too surprising that we do not observe this pattern very often,
because relatively few languages allow adposition-stranding in the first place.
Within the IE languages, for example, P-stranding is found mainly in some
languages of the Germanic branch. A language that does not allow P-stranding
but has the Direction of Agreement Parameter set “yes” will simply be a lan-
guage in which Ps do not agree with their NP complements. Such languages are
common. It is difficult in practice to distinguish languages in which Ps fail to
agree for systematic grammatical reasons (parameter (1) is set as “yes”) from
languages in which Ps fail to agree for accidental lexical reasons (the Ps of that
language happen not to be agreement bearers). Nevertheless, we can conclude
28 Another issue (motivating the “perhaps” in this statement) is the challenge of distinguishing true
adpositions from locative nouns. In quite a few languages, words which are used functionally as
locative adpositions are at least cognate with and perhaps reducible to location-denoting nouns
that show possessive agreement with their NP argument. I have not worried about this distinction
too much, because even though examples of this type might technically be agreement on N rather
than on P, they probably still count as downward agreement of a head with its complement, and
hence provide evidence of a “no” setting for parameter (1).
that the range of variation in how Ps agree in languages of the world is in

keeping with what one would expect given the parameters in (1) and (2).
5.7 Agreement on v
I have saved almost until last the investigation of “object” agreement – the
agreement that one often finds on v. This is probably the second most com-
mon kind of agreement after T agreement, being found in roughly 50% of the
languages in my sample. It is thus much more prominent and grammatically
significant than agreement on C or P. My parameters do indeed make clear
predictions about how object agreement should be different in NC languages
as compared to IE languages. The reason that I have put off discussion until
now is an empirical one: object agreement happens to be rare or nonexistent
in the IE languages, and there are some complications concerning it in the NC
languages. Nevertheless, I can work through the logic of the parameters as they
apply in this domain, and point to some possible instances in the NC and IE
languages. Then when we turn to other language families we will find clearer
examples that illustrate the expected range of possibilities.
5.7.1 v agreement in NC and IE languages

In Bantu languages, the implications of (1) for object agreement are exactly par-
allel to its implications for agreement on Ps: there should be no object agreement
on v unless the agreed-with NP moves to Spec, vP or higher. The object could
in principle stay in Spec, vP, or it could move on to a peripheral position, as in a
dislocation structure. For some Bantu languages, such as Chichewa, this looks
promising. There is indeed no object prefix if the direct object is internal to the
verb phrase, but there is an object prefix if the object is dislocated to a right or
left-peripheral position, as discussed in detail by Bresnan and Mchombo (1987)
(data from Mchombo 2004:80–1).29
(65) a. Alenje [a-ku-phı́k-ı́l-á anyanı́ zı́-túmbûwa].
2.hunters 2S-pres-cook-appl-fv 2.baboons 8-pancakes
‘The hunters are cooking the baboons pancakes.’
29 Note that “object markers” in Chichewa and Kinande cannot appear on verbs in passive or
reciprocal voice (Mchombo 2004:91,107), which I tentatively take to be evidence that they are
manifestations of agreement on active v, not cliticized pronouns comparable to those in Romance
languages and Greek (see section 3.3.3, especially note 24). Another relevant difference is that
in complex tenses object markers always stay on the main verb in the Bantu languages, whereas
in IE languages object clitics climb onto the finite auxiliary, as shown by the contrast between
Kinande and French in (i). Thus, object markers stay close to the v projection in Bantu, as
expected if they are agreement on v, in contrast to object clitics in IE languages, which are
attracted to the T node.
Agreement on v 197
b. Alenje [a-ku-wá-phı́k-ı́l-á - - zı́-túmbûwa] anyanı́.

2.hunters 2S-pres-2O-cook-appl-fv 8-pancakes 2.baboons
‘The hunters are cooking the baboons pancakes.’
Essentially the same is true in Kinande, except that left dislocation of the agreed-
with object is fairly rare and marked, and right dislocation is forbidden. One
usually finds object agreement only with a null object in Kinande, but (66b) is
possible.
(66) a. N-a-(*ri)-gul-a eritunda. (Baker 2003c)

1sS-t-5O-buy-fv fruit.5
‘I bought a fruit.’
b. Eritunda, n-a-*(ri)-gul-a.
fruit.5 1sS-t-5O-buy-fv
‘The fruit, I bought it.’
In contrast, an IE language could in principle have object agreement with an

object that remains inside VP, as long as v values the case of the agreed-
with NP, presumably as accusative. This prediction is largely moot, since most
IE languages do not have agreement on v at all. It is possible, however, that
what were historically clitics have been reanalyzed as true object agreement in
some varieties. Ormazabal and Romero (2006a) claim that this is true of leı́sta
dialects of Spanish, in which the formerly dative clitics le and les are used (also)
with animate/human objects in accusative case. Ormazabal and Romero argue
that le and les (in contrast to lo/los/la/las) have the status of object agreement
markers. Part of their evidence is that these morphemes can appear with even
an indefinite or quantified NP in the normal direct object position, without it
undergoing movement or dislocation. (67) is an example, where both the word
order and the indefinite interpretation of a unos jóvenes ‘some young people’
suggest that it is still in VP.
(67) (Yo) les-llevé a unos jóvenes al pueblo. (p. 18)

I 3pO-carry acc a.pl youngsters to.the town
‘I gave some young people a ride to the town.’
(i) a. Tu-a-by-a tu-ka-ri-ly-a. (Not: *tu-a-ri-by-a tu-ka-ly-a.)

1pS-t-be-fv 1pS-ptpl-5O-eat-fv 1pS-t-5O-be-fv 1pS-ptpl-eat-fv
‘We were eating it.’
b. Je l’ai mangé. (not: *J’ai le mangé)
I it-have eaten
‘I ate it.’
If Ormazabal and Romero’s interpretation of examples like these is correct,

then it confirms that v can in principle look downward into VP for something
to agree with in IE languages, as expected.30
Zulu reveals another important factor in evaluating predictions in this domain.
Like Chichewa, Zulu has optional object agreement on the verb:
(68) a. Ngi-bona umu-ntu. (Doke 1963:299)

1sS-see cl1.sg-person
‘I see a person.’
b. Ngi-ya-m-bona umu-ntu.
1sS-disj-1.sg.O-see cl1.sg-person
‘I see the person.’
But unlike Chichewa, the agreed-with object is not dislocated to the periphery
of the clause. Thus in Zulu it is possible to have V-NP1-NP2 order where the
verb agrees with NP1 but not NP2, as in (69b) (contrast this with (65b) from
Chichewa).
(69) a. Ngi-leth-el-a umfundisi incwadi. (Doke 1963:299)

1sS-bring-appl-fv teacher.1.sg book.5.sg
‘I am bringing the teacher a book.’
b. Ngi-ya-m-leth-el-a umfundisi incwadi.
1sS-disj-1.sg.O-bring-appl-fv teacher.1.sg book.5.sg
‘I am bringing the teacher – the one who told me to do so – a book.’
At first, this seems problematic. It is, however, perfectly possible to maintain

that the (first) object in (68b) and (69b) has moved to Spec, vP. This is defensible
as long as we believe that the verb in Zulu has moved past v into the T node (or
perhaps to some aspect or mood head above v but below T, as in Julien 2002
and Carstens 2005:227). Moving the object to Spec, vP would put the object
before the verb, but subsequent movement of the verb to T (or Aspect/Mood)
restores the V-NP order:
(70) [TP pro T+v+bring [vP <pro> teacher <v+bring> [VP <teacher> <bring> book]]]
30 Kayne (1989) discusses a rather different kind of object agreement in Romance languages –
agreement on past participles – which seems to be possible only if the object moves to a position
higher than the participle. I have no explanation for why this kind of agreement apparently needs
to be upward, despite the “no” setting for parameter (1) in IE languages. I only suggest that this
phenomenon needs to be studied in the light of the broader properties of participle agreement,
including those mentioned in section 5.9.
Agreement on v 199
Indeed there is good independent reason to think that verb-raising happens

in Zulu and other Bantu languages. First, the verb is morphologically united
with tense-aspect-mood morphology into a single morphological word; head
movement is one of the primary mechanisms by which such morphologically
complex words can be formed (Baker 2002). Second, the finite verb comes
before indefinite subjects that stay in vP in Zulu and other Bantu languages,
resulting in V-S-X order (see (13a) above). It is expected that the same verb-
raising that takes the verb to the left of unraised subjects in Spec, vP will also
take the verb to the left of shifted objects, which are also in a specifier of vP. This
is at least consistent with the idea that Bantu languages have V-to-T movement.
There is some semantic evidence that agreed-with objects have undergone
movement in Zulu, even though that movement is concealed by other, indepen-
dently motivated movements. Doke observes that whether the object is agreed
with or not corresponds to something like definiteness: agreed-with objects are
interpreted as definite, whereas unagreed-with objects are interpreted as indef-
inite, as indicated in the glosses of (68)–(69). We know that in languages like
Dutch and German leftward object shift produces strong, definite readings of
the shifted object (from Broekhuis 1992:81):
(71) a. dat Jan waarschijnlijk morgen iemand zal bezoeken

that Jan probably tomorrow someone will visit
‘that Jan probably will visit someone (or other) tomorrow’
b. dat Jan iemand waarschijnlijk morgen zal bezoeken
that Jan someone probably tomorrow will visit
‘that Jan probably will visit someone (a particular person) tomorrow’
A standard account of this sort of phenomenon since Diesing 1992 is to say

that the verb phrase is the domain of existential closure. NPs that remain
in the verb phrase are within the domain of existential closure, and get
weak/indefinite/nonspecific readings, whereas NPs that escape the verb phrase
get strong generic, specific, or (in articleless languages) definite readings. If
something like this picture is right, then it is attractive to analyze the difference
between (68a) and (68b) in Zulu in the same way. The NP stays in VP and gets a
weak indefinite reading in (68a), but leaves the VP and gets a strong or definite
reading in (68b). The different structures do not result in different surface word
orders, because the verb moves to T (and there are no adverbs that left-adjoin
to VP in most Bantu languages). But the difference is revealed indirectly by
agreement on v, given the parameter in (1). When the NP moves to Spec, vP
(or higher) it can agree with v, whereas when it stays inside VP (the domain of
existential closure) it does not c-command v and cannot agree with it:
(72)
a. b.
TP TP
DPi T´ DPi T´
pro T vP pro T vP
ti v´ ti v´
v ∃ VP NP v´
see V NP person v ∃ VP
[*CL1,sg] [CL1,sg]
<see> person see V NP
[CL1,sg]
*Agree by (1) Agree is OK <see> <person>
Thus, there are at least two ways to observe whether or not an NP has moved
higher than v: one is word order, and the other is whether it must have
a strong/generic/definite reading. Given these background assumptions, the
impossibility of a verb showing object agreement with a weak/indefinite NP
in Zulu is a result of the restriction on agreement in Bantu languages stated in
(1). The minimal contrast between the Zulu example in (68a) and the Spanish
example in (67) thus testifies again to the parametric difference in the syntax
of agreement between the two families. Diercks (2006) shows that this analysis
of Zulu can also be applied to the very similar “optional” object markers found
in Swahili.31
5.7.2 v agreement in other languages

Let us then briefly survey the behavior of object agreement on transitive verbs
in other languages. There are two things to look for. The crucial question with
31 There is one situation in which the Bantu v might agree downward with something in its
c-command domain. In section 3.3.3, I suggested that the Bantu language Shambala has two-
and-a-half agreement, in which v agrees with both the goal object in its Spec and the theme
object still in VP. Assuming for the sake of argument that Shambala has the same setting for
parameter (1) as most other Bantu languages, this stands out as an anomaly. What is special
about this case is that a single functional head (v) agrees twice, once in the proper way (probing
upward) and once not. This suggests that it is a “minimal compliance” effect in the sense of
Richards 1998: if the parametric condition in (1) has been satisfied once for a given head, that
head becomes free of the condition for further applications of Agree. (An alternative would be
to say that the object markers in Shambala are clitics, not true instances of agreement. In that
case Shambala is not relevant to the issues discussed in section 3.3.3.)
Agreement on v 201
respect to parameter (1) is whether object agreement requires the agreed-with

NP to move out of VP or not, as evidenced either by word order or by the
strong/definite reading of the object NP. The crucial question with respect to
parameter (2) is whether object agreement requires that the agreed-with NP
bear a particular case, such as accusative, or if agreement with oblique-case
objects is also tolerated. In fact, the full space of possibilities is realized for v
agreement, just as it was for T agreement and C agreement.
Unfortunately, even languages that clearly have object agreement do not
always provide clear evidence as to whether object agreement is contingent on
movement. The reason is because object agreement with third person objects
is often phonologically null, whereas agreed-with first and second person pro-
nouns are usually pro-dropped except in special discourse conditions (topic
and focus) which can themselves affect word order. When this is true, one
cannot expect to observe that object agreement depends on word order as in
(65) from Chichewa or on definiteness as in (68) from Zulu, because the form
with agreement and the form without agreement would be homophonous. Lan-
guages of this indeterminate kind include quite a few languages of the Amer-
icas (Mapudungun, Maricopa, Choctaw, Acoma, Halkomelem, Wichı́, Pirahã,
Quechua, Guaranı́, Ika) and some languages of Australia (e.g., Gooniyandi).
But agreement with third person objects is not always phonologically null,
and when it is not we find both languages in which object agreement is con-
tingent on word order and/or definiteness and languages in which it is not. A
good example of the former is Nahuatl. (73) compares an example in which the
direct object is indefinite and adjacent to the verb with an example in which
the direct object is definite and not adjacent to the verb. Both word order and
semantic interpretation suggest that the direct object is inside VP in (73a) but
outside it in (73b); nevertheless, the verb shows explicit object agreement with
its direct object in both examples (Launey 1981:38).
(73) a. Ø-quim-itta cōcōhua in pilli.

3sS-3pO-see snakes det child
‘The child saw (some) snakes.’
b. Ø -qu-itta in cihuātl in calli.
3sS-3sO-see det woman det house
‘The woman saw the house.’
(73a) implies that downward agreement is permitted in Nahuatl. Other lan-

guages in which object agreement is possible with indefinite objects that stay
close to the verb include Basque, Chukchi, Abkhaz, Georgian, Burushaski,
Tauya, Amele, Alamblak, Yimas, Mangarayi, Wari, Ojibwa, Maung, Tiwi, Dani,
Warlpiri, Lavukaleve, Mohawk, Wichita, Hixkaryana, Lakota, Kiowa, Makah,

and Pirahã. This seems to be the larger class.
There are also a reasonable number of languages in which agreement does
vary with the word order and/or interpretation of the object, however. One is
Canela-Krahô. (74a) shows that when the object is adjacent to the verb – its
normal position – the verb does not agree with it. The verb does, however, agree
with the object when it is displaced to the left, either in a focus-like construction
in which the object lands above the subject and tense marker ((74b)), or in a
scrambling-like construction in which the object lands at the left edge of the
verb phrase, below the subject and the tense maker ((74c)).
(74) a. Hũmre te po curan. (Popjes and Popjes 1986:139)

man past deer kill
‘The man killed a deer.’
b. Po wa i-te ih-curan. (p. 136)
deer I 1sS-past 3O-kill
‘It was a deer that I killed.’
c. Wa ka capi pı̃ na cu-mẽ. (p. 140)
I fut Capri wood about 3O-order
‘I ordered Capri about the wood (to cut it).’
Another language that shows a Bantu-like relationship between word order and
object agreement is the Athapaskan language Slave (Rice 1989:1197):
(75) a. Li [ʔ ehkee ka-yi hshu]

dog boy adv-3S.bit
‘The dog bit the boy.’
b. ʔ ehkeei li [ei ka-ye-yi hshu]
dog boy adv-3O-3S.bit
‘The boy, a dog bit (him).’
Yet another is Fijian (see Massam 2001 for a detailed analysis of the difference
between these sorts of sentences in the related language Niuean):
(76) a. [E’au.i vola mai] a cauravou. (Dixon 1988:49)

deliver letter to.here the youth
‘The youth is delivering letters.’
b. [E’au-ta - - mai] a-i-vola yai a cauravou.
deliver-tr.3O to.here the-letter this the youth
‘The youth is delivering the letter.’
As a final example, I cite Apurinã. Recall from section 5.2.2 that this language
has a very unusual word order, alternating between OSV and OVS. I analyzed
Agreement on v 203
this as a head-final language in which specifiers appear on the right – the

mirror image of English and the Bantu languages. So movement upward in the
structure is movement to the right in Apurinã. We find the same thing with
object movement: low objects that appear before the verb are not agreed with,
whereas high objects that appear after the verb are agreed with, as shown in
(77) (Facundes 2000:547).
(77) a. Uwa ata maporoka.

3.m.sg 1.pl root.up
‘We pulled it up.’
b. Owa maporoka-ru uwa.
3.f.sg root.up-3O.m 3.m.sg
‘She pulled it up.’
So the kind of relationship between word order and object agreement that
is found in the Bantu languages is also attested in other parts of the world.
Moreover – anticipating section 5.11 below – it is very suggestive that the
languages mentioned here are also languages that were seen to be like Bantu
languages rather than like IE languages in other domains: Canela-Krahô and
Apurinã are also like Bantu in showing a dependency between word order
and subject agreement (section 5.2.2); Slave and Fijian are also like Bantu in
showing agreement on Ps only when the object of P is outside the PP (section
5.6.2). This encourages the view that (1) is a single unified parameter, governing
all the functional categories of the language.32
The crucial issue for the Case-Dependency of Agreement Parameter is
whether or not v can agree with objects that bear oblique case as well as with
objects that bear the normal structural case. This too will not be observable
in all languages, but only in those that overtly mark a distinction between
(say) accusative case and dative or instrumental case. Quite a few agreement-
rich languages do not have this sort of morphological case system, includ-
ing most languages of Africa and the Americas. But among the subset that
32 A different sort of language in which v can only agree upward is Tariana, as analyzed in section
3.2.2.2. In Tariana, v agrees with the thematic subject, rather than with a moved object. As far as
I know, this is a rare pattern. Perhaps it only arises in languages in which v has no EPP feature
(so agreeing upward with a shifted object is not an option) and T is not a probe for agreement
(so it does not compete with v for the privilege of agreeing with the thematic subject). Some
ergative languages may also have vs that only agree upward with the thematic subject, such as
Abaza (section 3.2.2.1). The easiest way to explain this would be to say that agreement in these
languages is case-dependent: v agrees with NP only if v values NP’s case as ergative. If there are
any languages that have this sort of ergative agreement but show other evidence of agreement
not being case-dependent in the sense of (2), they might be problematic for my proposals.
do show case distinctions, we find both kinds of languages. For example,

Georgian is a language in which object agreement is not dependent on case.
Some verbs in Georgian take nominative objects in the aorist tense, whereas
others take objects in the dative case. Some verbs, such as ‘hit,’ even alternate
between the two case frames, depending on the dialect and register (Hewitt
1995:550):
(78) a. Mc’vrtnel-ma lom-s s-cem-a (formal/conservative)

trainer-erg lion-dat 3sO-hit-3sS
‘The trainer hit the lion.’
b. Mc’vrtnel-ma lom-i s-cem-a (colloquial/innovative)
trainer-erg lion-nom 3sO-hit-3sS
‘The trainer hit the lion.’
This difference in the case-marking of the object has no effect on verbal

agreement: the same third person singular agreement with the internal argu-
ment ‘lion’ is present on the verb in both (78a) and (78b). Similarly, Georgian
verbs show the same kind of agreement with dative objects of ditransitive verbs
as they do with nominative or dative/accusative objects of monotransitive verbs.
Burushaski ((79)) and Warlpiri ((80)) also fit this profile:33
(79) a. (U:n ) gu-yε ts-m. (Agr with absolutive object)

you-abs 2sO-see-1sS (Lorimer 1935)
‘I saw you.’
b. U:n ə r hik trn gu-čič-m. (Agr with dative object)
you-dat one half 2sS-give-1sS (Lorimer 1935)
‘I shall give a half to you.’
(80) a. Ngajulu-rlu ka-rna-ngku nyuntu nya-nyi. (Simpson 1991:158)
I-erg pres-1sS-2sO you see-npst
‘I see you.’
b. Yurrkunyu-rlu-lpa-ngku yu-ngkarla mangarri manu
police-erg-impf-2sO give-irr food and
nalija rdaku-ngka-ku-ju. (Simpson 1991:206)
tea jail-loc-dat-eu
‘The police must give you food and tea in jail.’
33 Interestingly, dative objects can sometimes be replaced with allative-marked NPs in Warlpiri;
these apparently never trigger agreement on the Aux node. I assume that allative case markers
are really members of category P; as such the phrase they head is a phase and hence impenetrable
to agreement from the outside.
There is a difference between agreement with an absolutive object and agreement with a
dative object only in the third person singular in Warlpiri. I put this aside as a secondary effect,
perhaps a matter of how agreement is realized at PF.
Agreement on v 205
Note that these are the very same languages in which subject agreement is not
sensitive to whether the subject has ergative case or nominative case (section
5.2.2). Thus, the languages that do not obey (2) for agreement on T also do not
obey it for agreement on v. Other languages of this type are Basque, Chukchi,
Imbabura Quechua, and perhaps Ika.34
In contrast to these languages, there are also languages in which the verb
agrees only with an object that has a particular case. Greenlandic Eskimo is
a clear example, where only absolutive case objects trigger agreement on the
verb, never internal arguments that bear dative or instrumental case. This is seen
in the minimal pair in (81): in (81a) the theme argument is in absolutive case
and triggers plural agreement on the verb, whereas in (81b) the goal argument
is in absolutive case and triggers singular agreement on a closely related verb
(Fortescue 1984:88–9).
(81) a. Aningaasa-t Niisi-mut tunniup-pai

money-abs.pl Nisi-dat give-3sS/3pO
‘He gave money to Nisi.’
b. Niisi aningaasa-nik tuni-vaa
Nisi.abs.sg money-instr.pl give-3sS/3sO
‘He gave Nisi money.’
Apurinã provides another example. In this language, most ditransitive verbs

take an oblique goal argument, and the verb cannot agree with that argument
((82a)). The only ditransitive verb that can take two non-oblique arguments is
‘give’ ((82b)); this is also the only verb that can show object agreement with
the notional goal ((82c)) (Facundes 2000:287–90).
(82) a. Mipake owara-pira-ta-ru/*i u-sãkire pita-monhi.

Mipake teach-language-3O.m/*2O 3.m-language you-goal
‘Mipake taught his language to you.’
b. Nhi-yowata-ne pita nota suka.
1sP-knife-poss you I give
‘I gave you my knife.’
34 My typology is more restrictive on this point than Bobaljik’s (to appear). Bobaljik distinguishes
between unmarked case (nominative, absolutive), dependent case (accusative, ergative), and
oblique case (all others). In his theory, a language can be specified as allowing agreement
with dependent cases but not oblique cases, so he expects to find languages in which ergative
subjects trigger agreement but dative NPs do not. I distinguish only between languages in which
agreement on a particular head requires the trigger to have a particular case and languages that
do not. I agree that there are a few languages that seem to match Bobaljik’s prediction, but I
would analyze them as languages that have goal PPs, not dative case NPs (see, for example,
note 13 on Nepali) – a distinction that the theory needs anyway (see note 31 on dative versus
allative in Warlpiri).
c. Pu-suka-no nota.
2sS-give-1sO me
‘Give (it) away to me.’
The possibility of agreement with the goal in (88c) but not (88a) testifies to the
case-dependence of agreement in Apurinã. (Recall that agreement is sensitive
to word order in Apurinã because of parameter (1); that is why there is no
agreement with either object in (82b).)35
Overall, then, we find the range of object agreement systems that we expect
to find, given the existence of the two independent parameters for agreement in
(1) and (2).
5.8 Agreement on the linker head

The last functional head whose agreement properties I consider is a rather
unusual one that is unique to Kinande among the languages in my 108-language
sample. Baker and Collins (2006) describe a particle that appears between the
two complements of a triadic verb in Kinande; some examples are given in (83).
(83) a. Mo-n-a-hir-ire okugulu k’ omo-kihuna.

aff-1sS-t-put-ext leg.15 lk.15 loc.18-hole.7
‘I put the leg in the hole.’
b. Kambale a-seng-er-a omwami y’ ehilanga.
Kambale 1S/t-pack-appl-fv chief.1 lk.1 peanuts.19
‘Kambale packed peanuts for the chief.’
Collins and I analyze this as a head that we call linker (Lk), which selects a
VP complement and is itself the head of the complement of v. One NP from
within VP then moves to Spec, LkP, where it gets accusative case from v. If
necessary, the other complement gets case from Lk itself. Interestingly, either
NP can move to Spec, LkP. Thus, the sentences in (84) are also possible, in
more or less free variation with the corresponding sentences in (83).
(84) a. Mo-n-a-hir-ire omo-kihuna m’ okugulu.

aff-1sS-t-put-ext loc.18-hole.7 lk.18 leg.15
‘I put the leg in the hole.’
35 There is a serious practical problem in distinguishing NPs marked with dative (or oblique) case
from PPs. I suggested that agreement with the goal argument in (82a) is bad because Arupinã
has parameter (2) set as “yes.” An alternative interpretation is that monhi is a P and hence blocks
agreement with the goal by the phase condition. This indeterminacy of analysis is a general
problem, and clearer and more reliable ways to distinguish case and adpositions are needed.
(In Greenlandic it is relatively clear that dative is a case marking in the sense familiar from IE
languages, because modifiers agree with the head in case, including dative.)
Agreement in auxiliary constructions 207
b. Kambale a-seng-er-a ehilanga hy’ omwami.

Kambale 1S/t-pack-appl-fv peanuts.19 lk.19 chief.1
‘Kambale packed peanuts for the chief.’
Consider now the agreement properties of this particle. It always agrees with
the NP to its left, and never with the NP to its right. In structural terms, it agrees
with its specifier, which asymmetrically c-commands it, never with something
inside its VP complement. This is in accordance with (1). Furthermore, case
assignment does not play a crucial role in this agreement: if Baker and Collins
are correct, the first NP in (83) and (84) gets its case from v, not from Lk;
nevertheless, Lk agrees with it. This shows once again that (2) does not apply
in Kinande.
Since IE languages have the opposite values for the agreement parameters in
(1) and (2) from Kinande, I expect that IE languages could not have verb-phrase-
internal particles with the same agreement properties as Kinande’s linker. That
is consistent with the facts: to the extent that IE languages have VP-internal
particles at all (e.g., English I gave Chris back the book) they do not bear
agreement. An IE particle could conceivably agree with an NP that it assigned
case to, given its “yes” value for (2). But if particles never assign case in
IE languages – and never undergo case concord – then they can never have
agreement.
The value of this functional head for broad typological investigation is essen-
tially nil. The only languages that are known to have overt linkers outside the
NC languages are certain Khoisan languages studied by Collins and his stu-
dents. But the linkers in those languages do not bear agreement. The value of
mentioning this unusual element in Kinande comes from its contribution to
my argument that the agreement parameters apply to languages as wholes, not
just to particular heads in those languages. The linker particle in Kinande has
recognizably the same agreement behavior as many other, more familiar heads
in Kinande, including Tense, Complementizer, Determiner, Preposition, and v.
The inventory of functional heads might vary to some extent from one language
to another, but the parameters in (1) and (2) apply automatically to whatever
agreement-bearing functional heads a language happens to have, I claim.
5.9 Agreement in auxiliary constructions

5.9.1 Auxiliary constructions in Bantu vs. IE languages
There is one more rather prominent difference in agreement between Bantu
languages and IE languages that can be studied in terms of my parameters. This
difference is seen in complex tenses that are made up of an auxiliary plus a main
verb. IE languages consistently have full verbal agreement on the auxiliary verb
but not on the main verb:36
(85) a. I am crying
b. You are crying.
c. He is crying(*s)
(86) a. Las muchachas están leyendo el libro. (*leyendas)
the girls.f.pl are reading(m.sg) the book reading.f.pl
b. Las muchachas han leı́do el libro. (*leı́das)
the girls.f.pl have read(m.sg) the book read.f.pl
In contrast, Bantu languages have full person-number-gender agreement on

both verbs in a complex tense. This construction has been studied with par-
ticular care from a generative perspective by Kinyalolo (1991) for Kilega and
Swahili; it also exists in Kinande and Zulu. (See also Carstens 2001 for important
implications of this construction for aspects of Chomsky’s Minimalist theory of
agreement.)
(87) (Mimi) ni-li-kuwa ni-ngali ni-ki-fanya kazi. (Swahili)
(1.sg) 1sS-past-be 1sS-still 1sS-perf-do 9.work
‘I was still working.’ (Carstens 2001:150)
(88) a. Abakali ba-lwe ba-ka-ly-a amatunda. (Kinande)
women.2 2S-leave 2S-ptpl-eat-fv fruits.6
‘The women were eating fruits.’
b. Tú-lwé tú-ká-ly-a.
1pS-leave 1pS-ptpl-eat-fv
‘We were eating.’
This sort of double agreement is also found in Lokaa and Ibibio, agreement-rich
non-Bantu languages of the Niger-Congo family. The question then is why are
two full agreements with the same NP argument tolerated in NC but not in IE
languages.
My answer centers on the fact that IE languages are subject to (2), whereas
Bantu languages are not. Suppose that the T-like participial head associated
with the lower verb in a complex verbal construction agrees with a particular
36 Past participles in IE languages sometimes agree with moved direct objects in number and
gender (although not in person) under complex conditions that vary from language to language
(see Kayne 1989 and many others). I have nothing to say about this complex topic here. The
question at hand is whether a language allows two verbs in the clause to agree with the same
noun phrase, and the possibility of object agreement on participles is not directly relevant to
this.
NP X. In an IE language, this will only be possible if the participial head values

the case feature of X as (say) nominative. X then raises into the domain of the
higher T associated with the auxiliary verb. This T cannot value the case feature
of X, because X’s case was already valued by the lower T. Therefore, the upper
T cannot agree with X. In short, the sort of double agreement with the same NP
that is found in NC languages is ruled out in most IE languages because both
agreeing Ts must value the case feature of NP and case valuation can only be
done once for each NP.37
The NC languages, in contrast, are not subject to the Case-Dependency of
Agreement Parameter in (2). Therefore, there is nothing to rule out two distinct
T nodes from agreeing with the same NP in these languages. All that would
be required would be for the NP to asymmetrically c-command both Ts. This
is accomplished by NP movement. That explains why examples like (87) and
(88) are possible in languages of this family.
This reasoning extends to raising-to-subject constructions involving verbs
like seem. In IE languages, these have agreement only on the matrix verb; the
lower verb must be an uniflected infinitive:38
(89) a. The children seem – to be tired.

b. *The children seem – are tired.
In Kinande, it is possible to have full agreement on both the matrix verb and
the embedded verb, as shown in (90b).
(90) a. A-li-nga aba-kali mo-ba-gend-ire.

1S-be-if 2-woman aff-2S-go-ext
‘It seems that the women have left.’
b. Aba-kali ba-li-nga mo-ba-gend-ire.
2-women 2S-be-if aff-2S-go-ext
‘The women seem to have left.’
37 Double agreement in these auxiliary-plus-verb constructions can be contrasted with having

agreement on both C and T, which is possible in some West Germanic languages (see (35)). The
difference, in my view, is that C and T work together to assign a single instance of nominative
case (Watanabe 1996), whereas the two T-like nodes in a complex tense structure are competing
to assign nominative case twice.
38 European Portuguese is unusual among the IE languages in that it has infinitives that can show
agreement. Interestingly, the complement of a raising verb in Portuguese can be an agreeing
infinitive if and only if the subject does not raise into the matrix clause, where it agrees with
the matrix T (Schütze 1997:127). This proves that it is the agreement/no-agreement distinction
that is crucial in IE languages, not the finite/nonfinite distinction per se, just as my theory would
expect.
In (90b), the lower verb can agree with the subject ‘woman’ without necessarily
valuing its case as nominative. So can the higher verb, and double agreement
is therefore allowed in raising structures in Kinande.
While the parameter in (2) is the primary factor at work in this domain, one
can observe some influence of the Direction of Agreement Parameter as well.
The matrix verb in Kinande can agree with the thematic subject of the lower
clause only if that subject appears in the matrix clause. When the thematic
subject stays in the lower clause, the ‘seem’-class predicate cannot agree with
it, and must be a default third person singular form. Thus, (91) is ungrammatical,
in contrast to (90a).
(91) *Ba-li-nga aba-kali mo-ba-gend-ire.

2S-be-if 2-woman aff-2S-go-ext
‘It seems that the women have left.’
This follows from the fact that T cannot search downward to find an agreement
trigger in Bantu, the way it does in IE structures like (92).
(92) There seem to be some children in the basement.
Returning to complex tense constructions, it must be acknowledged that these

sometimes do show a limited form of double agreement even in IE languages.
But this never happens when the lower verb is a fully verbal, finite form. It
only happens when the lower verb is an adjective-like participle, which agrees
with the subject in number and gender but not in person. (93) gives examples
from Serbian, where the phenomenon is quite general. (Many western European
languages have this sort of double agreement only in passive or unaccusative
clauses, for reasons that I do not investigate.)
(93) a. Já sam čı̀ta-o/ čı̀ta-la. (Magner 1991:268–9)

I am read-m.sg read-f.sg
‘I was reading; I read.’
b. On/ona je čı̀ta-o/ čı̀ta-la.
he/she was read-m.sg read-f.sg
‘He/she was reading; he/she read.’
c. Mi smo čı̀ta-li/ čı̀ta-le
we are.1pS read-m.pl read-f.pl
‘We were reading; we read.’
Why is this possible? The answer is that participial heads do not value the
case of the NP that they agree with. Thus they do not compete with the finite
T associated with the auxiliary verb in this respect, so nothing prevents that
T from both assigning case to and agreeing with the NP. Why then can the
participle agree with NP without valuing NP’s case, given (2)? Because, like
adjectives and determiners, participles have an unvalued case feature of their
own, which the NP can value. In other words, they can agree with the NP in
case. In many situations the case on the participle will be unmarked nominative,
but it can sometimes bear other cases; for example, accusative participles show
up in ECM constructions in Icelandic (Thráinsson 1979:362–4):
(94) a. Marı́a var tekin af lögreglunni.

Maria was taken.f.sg.nom by police.the
‘Maria was apprehended by the police.’
b. Ég tel Marı́u hafa verið tekna (*tekin) af lögreglunni.
I believe Mary.acc to.have been taken.f.sg.acc by police
‘I believe Maria to have been apprehended by the police.’
There are thus at least two types of T when it comes to case properties: finite
Ts that assign (nominative) case to NPs, and participial Ts that can receive case
from NPs. Bantu languages can have complex tenses with multiple Ts of the
first kind; IE languages can only have complex tenses in which there is one T
of the first kind and all the other Ts are of the second kind.
Of course, lexical factors also play into whether the potential for multiple
agreement in complex tenses is realized or not. The “participial” T on the main
verb in a Bantu language could, for example, happen not to be a functional
head that undergoes agreement. This happens sometimes in Kinande: some
auxiliaries select an infinitival form of the main verb, with prefix eri, rather than
the participial form shown in (88), with prefix ka. Verbs with eri do not agree,
so there is only IE-like single agreement on the auxiliary in these particular
constructions (e.g., tu-lwé tu-ka-lya ‘we-are we-eating’ vs. tu-lwe b’eri-rya
‘we-are to-eat’, meaning ‘we ought to eat’). The Bantu language Luvale seems
to have only complex tenses based on the infinitival form of the verb. As a result,
it never has double agreement:
(95) Tu-li na-(*tu)-ku-va-ngula. (Horton 1949)

1pS-be with-(1pS)-inf-2O-teach
‘We are teaching them.’
The implications we derive in this domain thus go only one way. It follows from
a positive setting of parameter (2) that a language will not have full double
agreement in auxiliary-plus-main verb constructions, but it does not follow
from a negative setting of parameter (2) that a language must have double
agreement in auxiliary-plus-main verb constructions. That depends also on the

lexical properties of the morphemes involved in the construction.
5.9.2 Auxiliary constructions in other languages

Checking whether other languages have multiple agreement in auxiliary con-
structions is relatively straightforward. Some languages do not appear to have
any relevant auxiliary constructions; either these languages simply lack non-
thematic verbs that select VP complements, or the main verb incorporates into
the auxiliary verb to form a single complex head, which cannot bear more than
one set of agreements. Such languages do not have evidence of this sort for how
parameter (2) is set. Examples include Mohawk, Abaza, Mapudungun, Mayali,
Gooniyandi, Yimas, Nez Perce, Mixtec, Karok, Wichita, Pirahã, and Wari.
Of the languages that do have some sort of auxiliary constructions, a healthy
minority (about 23) are like Bantu languages in having full person-number-
gender agreement on the main verb as well as the auxiliary in at least some
cases. Examples of this type include Berber, Arabic, Georgian, Burushaski,
Arapesh, Fijian, Maricopa, Tukang Besi, Halkomelem, Ojibwa, Slave, Otomi,
Imbabura Quechua, Guaranı́, Canela-Krahô, Jarawara, Jakaltek, and Zoque. For
example, Cherchi (1999:25–6) lists some contexts in which an auxiliary verb is
used in Georgian, and his examples show double agreement with a first person
subject:
(96) a. v-my.er-i-v-ar.
1sS-sing-ptpl-1sS-be
‘I am singing.’
b. c.a-v-sul-v-ar.
prev-1sS-go-1sS-be
‘I have gone.’
Thus, agreement with the first person argument appearing on the main verb does
not disqualify the auxiliary verb from agreeing in person with the very same
argument, as expected if Georgian is not subject to (2). Similarly, (97) shows
that both the auxiliary and the main verb bear full agreement in Burushaski
(Lorimer 1935).
(97) a. Jε u:n ε xidmt ε č-a ba. (p. 317)

I.nom your service do-1sS be.1sS
‘(for these many years) I have been at your service.’
b. ε č-u bo; ε č-i bi; ε č-a ba:n (p. 274)
do-FsS be.FsS do-xS be.xS do-1S be.1pS
‘she is doing it; X is doing it; we are doing it’
Burushaski and Georgian are also languages in which T can agree with ergative
subjects as well as nominative ones, and v can agree with dative objects as well
as with nominative ones. So there is converging evidence that these languages
are not subject to the parameter in (2).39
There are also many languages (about 44) in which the auxiliary verb bears
agreement but the main verb does not; these include Bagmiri, Lango, Hausa,
Hebrew, Finnish, Basque, Turkish, Chukchi, Kannada, Asmat, Tauya, Amele,
Alamblak, Kewa, Mangarrayi, Maung, Tiwi, Dani, Lavukaleve, Chamorro,
Greenlandic, Acoma, Yagua, Warao, Barasano, Choctaw, Hixkaryana, and
Wichı́. Crucially we expect those languages in which agreement is tied to case
to be in this group, and that is true. For example, Greenlandic verbs do not agree
with oblique-case arguments, but only with arguments in structural case (abso-
lutive and ergative), as shown in (81). This goes along with the fact that only
the auxiliary verb bears agreement in complex tenses in Greenlandic (Fortescue
1984:48):
(98) Niqi niri-niq ajur-pa-a-t.

meat eat-noml neg.hab-ind-3sO-3pS
‘They don’t eat meat.’
Apurinã was another language in which agreement seems to be case dependent

in other respects (see (82)), and it too has agreement only on the auxiliary when
one is present:
(99) Pitxeka i-txa komeru. (Facundes 2000:295)

grow 3S.m-aux manioc.m.sg
‘The manioc grew.’
Some readers may have noticed that nothing in my account requires that the
sole agreement in a language subject to (2) be on the auxiliary rather than on the
main verb. My line of reasoning implies that there can only be one agreement
per argument in such a language, but it says nothing about where that agreement
must be. In fact, having agreement on the main verb but not the auxiliary does
not seem to be common, but is attested in, for example, the Mayan language
Tzotzil:
(100) a. Ba j-ta-tikotik j7ilol. (Aissen 1987:17)

went 1E-find-1.pl.ex shaman
‘We went to find the shaman.’
39 Warlpiri was another language of this type, but it does not have multiple agreement in auxiliary
constructions. I assume that this is a result of lexical factors, not the parameter settings, as in
Luvale.
b. Laj j-maj-ot.
end 1E-strike-2sA
‘I’m done hitting you.’
The impossibility of agreement appearing on the higher verb as well is consistent

with other evidence that Tzotzil is a case-dependent language; its verbs never
agree with oblique complements marked with ta, for example:
(101) 7i-tal ta Saklum. (Aissen 1987:11)
compl-come to Saklum
‘They came to Saklum.’
(Not: *7i-s-tal ta Saklum ‘compl-3E-come-3A to Saklum’)
I thus leave open the possibility that agreement can be on the main verb rather
than on the auxiliary verb – although one would like to know why this is not very
common (see section 5.11.3.2 for another example of this kind from Huallaga
Quechua).
Finally, I point out that there could be constructions in which the same argu-
ment triggers agreement on more than one head other than the auxiliary con-
structions focused on here. The existence of any such construction will tend to
show that (2) is not obeyed in the language, given that an NP can only be case-
valued once. Another situation where this arises is the long distance agreement
constructions in Passamaquoddy and other Algonquian languages, discussed
briefly in section 3.3.4. In these constructions the same argument triggers sub-
ject or object agreement on the lower verb and object agreement on the higher
verb. Neither the lower T nor the matrix v undergoes case concord, and both of
them agree, so these Algonquian languages must not be subject to (2). Similar
multiple agreement in LDA constructions is found in Arabic, Canela-Krahô,
and Quechua. However, this type of agreement is too rare to be especially useful
for typological investigation; I therefore omit a full discussion here.
5.10 A third value for the Direction of Agreement Parameter

There is one other case of multiple agreement with a single argument that is
worthy of special discussion, however, because it points to a third possible
setting for the Direction of Agreement Parameter.40 This is the interesting fact
40 The analysis of Burushaski presented in this section, and its implications for the statement of
the Direction of Agreement Parameter, was a late-breaking discovery made possible by my
becoming aware of sources on this language other than Lorimer 1935. It was added to the book
while it was already in production, so its implications could not be fully integrated into the text
as a whole. I thank Beatriz Fernández for calling the additional sources to my attention, and for
relevant discussion.
A third value for the Direction of Agreement Parameter 215
that unaccusative verbs show both subject and object agreement with their sole
argument in Burushaski, as seen in (102).
(102) Dası́n há-e le mó-yan-umo. (Willson 1996:3, 19, 21)
girl(abs) house-obl in 3sO.f-sleep-3sS.f/past
‘The girl slept in the house.’ (similarly, ‘die,’ ‘wake up,’ ‘rot,’ ‘be lost’ . . .)
At a minimum, examples like (102) show us that the non-theta-marking v

present in unaccusative structures can be a probe for agreement in Burushaski,
just as the theta-marking v found in transitive constructions is. Burushaski is
different in this respect from many languages with object agreement, in which
only the active transitive v is an agreement-bearer. But given this, the possibility
of (102) is not so surprising, since we already know that the Case Dependency
of Agreement Parameter is set “no” in Burushaski (see (21), (79), and (97)).
Thus v need not assign objective case to the theme argument in order to agree
with it in Burushaski, and v agreeing with this NP does not prevent T from also
agreeing with the same NP, the way it would in an Indo-European language in
which parameter (2) is set “yes.”
It is notable, however, that unergative verbs in Burushaski do not show object
agreement as well as subject agreement with their sole argument, as seen in
(103).
(103) Dası́n há-e le huruT-umo. (Willson 1996:3, 19, 21)
girl(abs) house-obl in sit-3sS.f/past
‘The girl sat in the house.’ (similarly, ‘dance,’ ‘walk,’ ‘cry,’ ‘come out’ . . .)
The question is why. The v used in unergative constructions is no different from

the v used in transitive constructions in most conceptions, and we know that the
latter does bear agreement – so why can’t the former? An interesting possibility
is that v in unergative constructions cannot agree with the subject because the
syntactic configuration is wrong for agreement to take place in this language.
That would be the case if, in fact, functional heads in Burushaski need to probe
downward to find a goal to agree with, rather than upward. Suppose then that
we revise the Direction of Agreement Parameter stated in (1) so that it includes
a third possible setting, as expressed in (104).
(104) The Direction of Agreement Parameter
(i) F agrees with DP/NP only if DP/NP asymmetrically c-commands F, or
(ii) F agrees with DP/NP only if F c-commands DP/NP, or
(iii) F agrees with DP/NP only if F c-commands DP/NP or vice versa.
(104i) is the now-familiar Bantu setting of this parameter and (104iii) is the IE
setting; (104ii) is the third logical possibility, which I propose for Burushaski.
This version of the parameter is more symmetrical than the original one: it says
that a language can limit itself to downward agreement as well as to upward
agreement. It is, therefore, to be preferred on conceptual grounds.
The parameter setting in (104ii) accounts for the difference between (102) and
(103) as follows. The sole NP argument of an unaccusative verb originates inside
the VP complement of v; hence v c-commands NP (prior to any movement)
and can agree with it in accordance with (104ii). In contrast, the sole overt
argument of an unergative verb originates in Spec, vP, a position that v does
not c-command. Therefore, v cannot agree in this structure, given (104ii). The
two structures are compared in (105).
(105) a. [TP -- [vP [VP girl sleep ] v+AGR ] Past+AGR ]
b. [TP -- [vP girl [VP sit ] v+*AGR ] T+AGR ]
Object agreement in Burushaski has a second somewhat unusual property,

which can also be attributed to this parameter setting. Willson (1996) shows
that verbs like ‘tell’ and ‘give’ that select arguments in dative case show object
agreement with their dative argument ((106a)). In contrast, verbs that appear
with a freely added benefactive expression in dative case do not show object
agreement with the benefactee ((106b)).
(106) a. Hilés-e dası́n-mo-r toofá mu-ú-imi. (Willson 1996:3)

boy-erg girl-obl-dat gift.x(abs) 3sO.f-give.x-3sS.m/past
‘The boy gave the girl a gift.’
b. Jé-e gó-e-r cái tayáar é-t-am (Willson 1996:36)
I-erg you-obl-dat tea.y(abs) prepare 3sO.y-do-1sS/past
‘I prepared tea for you.’
Most other languages treat examples like (106) the same: either they show object
agreement with neither oblique phrase (if agreement is case-dependent in the
language, or if the oblique phrase counts as a PP), or they show agreement
with both dative expressions (e.g., Basque). Although the Case Dependency
of Agreement Parameter cannot draw this distinction, the parameter setting
in (104ii) arguably can. Verbs like ‘give’ select goal expressions as a lexical
property; therefore, the goal phrase is generated inside the VP complement
of v along with the theme argument. In contrast, verbs like ‘prepare’ do not
inherently select a benefactive expression; this phrase is arguably theta-marked
instead by an applicative head (Marantz 1993), which is phonologically null in
A third value for the Direction of Agreement Parameter 217
Burushaski but overt in many other languages. Suppose then that we add the
assumption that the applicative head takes vP as its complement rather than the
other way around in Burushaski. Then the structures of (106a,b) are roughly
(107a,b).
(107) a. [TP -- [ vP boy [VP girl-DAT gift give ] v+AGR ] Past+AGR ]
b. [TP -- [ApplP you-DAT [ vP I [ VP tea prepare ] do+AGR ] Appl ] Past+AGR ]
The v c-commands the dative phrase in (107a) but not in (107b); therefore it can
agree with the dative phrase in (107a) but not in (107b), given (104ii). The same
parameter that explains why there is object agreement with “low” subjects but
not higher ones ((102) vs. (103)) can also explain why there is object agreement
with dative phrases selected by V but not with unselected dative phrases.
There is some reason to say that the parameter setting in (104ii) applies to
other functional heads in Burushaski as well. Some postpositions and quantifiers
(‘both,’ ‘all’) agree with their NP complements in Burushaski; this confirms that
downward agreement is possible in this language, although not that it is required.
A hint that T also must agree downward comes from transitive sentences with
a “scrambled” OSV order, such as (108); I tentatively analyze these as being
the result of moving the object to Spec, TP rather than the thematic subject (see
section 9.5.2 above).
(108) a. Ne hir cél-e Ø-cú-m-i (Grune 1998:16)

the man(abs) water-erg 3sO.M-carry.away-past.II-3sS.n
‘The man was carried away by the water.’
b. [TP man [vP water-ERG [VP <man> carry ] v+AGR ] Past+AGR]
Notice that the agreement suffix associated with finite T still agrees with the
thematic subject in Spec, vP, not with the moved object in Spec, TP (see also
(23)). Burushaski is like the IE languages rather than like Kinande in this
respect. The question is why. For an IE language, agreement between T and the
fronted object is blocked by the Case Dependency of Agreement Parameter:
the fronted object has accusative case, not nominative case assigned by T; there-
fore T cannot agree with it. But this reasoning does not extend to Burushaski,
because the Case Dependency of Agreement Parameter is set “no”; T clearly can
agree with NPs that bear absolutive case as well as those that bear ergative case
(see (102) and (103)). Fortunately the parameter setting in (104ii) can explain
this pattern. T cannot agree with the thematic object in the post-movement struc-
ture shown in (108b), because T does not c-command the object in Spec, TP. T
does c-command the thematic object in its base position inside VP prior to move-
ment. However, T also c-commands the thematic subject in Spec, vP, and the
subject therefore blocks T from agreeing with the object in its pre-movement
position, by the intervention condition on agreement. The “downward only”
setting of the Direction of Agreement Parameter thus explains why T always
agrees with the subject in Burushaski.
If this analysis is correct, it should also have implications for agreement on
predicate adjectives in Burushaski, since the subject of predication is never in
the c-command domain of FA , the functional head associated with adjectives
(see chapter 2 and section 5.3 above). And there is something interesting to
explain in this area. Whereas Lorimer (1935:111) says only that adjectives
“sometimes appear in plural forms when the noun to which they refer is in the
plural,” Tiffou and Pesot (1989:53–4) observe a distinction between attributive
adjectives and predicate adjectives. They imply that agreement in number is
required when the adjective is an attributive modifier, but say that agreement is
optional when the adjective is predicative:
(109) a. iskó éser-um-iŋ (*éser-um) γatéha-ŋ

three sharp-adj-pl sharp-adj sword-pl.y
‘three sharp swords’
b. We guš-iŋ a kančéni/ kančéni-tiŋ ban.
these women-f.pl flighty flighty-f.pl be.f.pl
‘These women are flighty.’
I interpret this as follows. When an AP is an attributive modifier of NP, the

projection of FA merges directly with NP and hence FA can agree with NP in
accordance with (104ii).41 When AP is the complement of Pred, FA does not
merge with NP, and the subject NP is never in its c-command domain; hence
FA cannot agree with NP in these circumstances, in accordance with (104ii).
(104ii) thus induces a difference in agreement between attributive structures
and predicative structures in Burushaski that is not seen in Bantu languages or
IE languages.
41 The technical definition of c-command would need to be adjusted somewhat so that FA counts
as c-commanding the NP (or as being c-commanded by it) in the relevant adjunction structure.
While I cannot enter into the technicalities here, the crucial structural difference between the
two structures involving APs should be clear.
Many little parameters or two big parameters? 219
To complete this account, I need to say why agreement is possible as an option

in (109b), whereas agreement is either required or forbidden in most other
contexts. I claim that the reason is because (109b) is structurally ambiguous: it
can be parsed as a simple predicate adjective structure, or it can be parsed as
a predicate nominal structure in which the adjective is an attributive modifier
of a null noun. The latter structure is possible in Burushaski by all accounts;
Lorimer (1935:107) observes that when the subject is singular, the predicate
adjective is often followed by the indefinite determiner n, and says that such
sentences are properly translated as in (110).
(110) Huq mat-um-n bi (cf. mt-um huk)

dog black-adj-one be.3.sg.n black-adj dog
‘The dog is a black one.’
The version of (109b) with agreement is, I claim, analogous to (110): the null
N head is plural in agreement with its antecedent (not part of Agree: see section
4.2), and FA agrees with the N it modifies in number, just as it does in (109a). In
contrast, when the complement of the copula is a simple AP, agreement never
takes place. If this is right, it provides converging evidence that functional heads
only probe downward for something to agree with in Burushaski – in contrast
to Kinande, in which they only probe upward.
The properties of Burushaski thus lead us to a more symmetrical and attrac-
tive conception of the Direction of Agreement Parameter. We no longer need to
ask what is special about upward agreement, such that some languages would
restrict themselves to that kind of agreement. It turns out that there is nothing
particularly special about upward agreement; languages can also restrict them-
selves to downward agreement. Unfortunately, I do not have the kind of data
that would distinguish the (104ii) setting of the parameter from the (104iii)
setting for languages other than Burushaski and IE languages. Therefore, I
recombine these two logically distinct cases back into one for purposes of the
broad typological survey that the next section engages in.
5.11 Many little parameters or two big parameters?

I have presented a detailed comparison between agreement in NC languages
and agreement in IE languages, showing that there are consistent differences
that show up in a wide range of syntactic environments. This suggests that the
differences are created by two overarching parameters; any other view seems to
be missing a generalization. I have also shown that a similar range of phenomena
can be seen in various other languages, filling out the range of possibilities and
proving that the issues are not unique to these two families. I also noted in
passing that in several instances the parameter values in these other languages
seem to be consistent across various functional heads.
The skeptic could still say that it is a coincidence that all of the functional
categories in NC languages work in a consistent way, and that all of the func-
tional categories in IE languages work consistently in the opposite way. These
are possible languages, of course, but perhaps languages that have some func-
tional heads that behave one way with respect to agreement and other functional
heads that behave another way are equally possible. The same skeptic might
be suspicious that I have cherry-picked those languages that look best for my
view. And the skeptic would be right. To really nail down the question, we have
to find a way to look more broadly at what kind of languages are possible and
common versus what kinds (if any) are impossible or surprisingly rare. In other
words, we need to do some statistics.
5.11.1 A 108-language survey of agreement behavior

One way to pose the question is to ask how many kinds of languages there are
with respect to the syntax of agreement. Are there roughly the four kinds of
languages predicted by the two logically independent, binary-valued parame-
ters in (1) and (2)? Or are there the hundreds of different kinds that could in
principle result from applying (1) and (2) to each functional head in a language
individually?
I investigated this by considering the 100 languages in the World Altas of
Language Structures core language sample, with a few substitutions, plus a
few additional languages that I had good information on and found interesting.
More specifically, I replaced Ambrat with Tauya, Rama with Ika, and Kutenai
with Nez Perce, because I could not find sources on those languages. Also in
a handful of cases I switched a language in the WALS sample for another in
the same family that I know better: Mohawk for Oneida, Choctaw for Koasati,
Ojibwa for Cree, Tzotzil for Jakaltek. Finally, the eight languages that I added
were: Halkomelem, Mayali, Warlpiri, Yimas, Nahuatl, Kinande, Tariana, and
Jarawara. The result is a relatively large sample, genetically and areally diverse,
and not personally constructed by me to prove a point. For each language, I
looked for evidence relevant to the following eight specific issues:
(111) a. Is subject agreement dependent on the position of the subject?

b. Is subject agreement sensitive to the case of the subject?
c. Is object agreement dependent on the position or definiteness of the
object?
Table 1
Agree must be up
((1) = Yes) Agree can be up or down ((1) = No)
Agreement Turkish, Lango, IE languages (7), Hausa, Finnish,

dependent Greenlandic, Apurinã, Abkhaz-Abaza, Kannada, Asmat, Amele,
on case Chamorro, Mapudungun Alamblak, Maung, Mangarrayi, Tiwi,
((2) = Yes) (n = 6) Lavukaleve, Daga, Yimas, Lakhota,
Tzotzil, Warao, Barasano, Yagua, Wichı́,
Choctaw, Hixkaryana, Hebrew, Wari’,
Chukchi, Makah
(n = 32)
Agreement Zulu, Swahili, Kinande, Georgian, Arabic, Persian, Warlpiri, Dani,
independent Berber*, Arapesh, Kewa, Burushaski, Mayali, Halkomelem,
of case Tariana, Fijian, Tukang Tauya, Ojibwa, Nez Perce, Karok, Otomi,
((2) = No) Besi, Slave, Zoque, Ika, Basque, I. Quechua, Guaranı́
Canela-Krahô, Jarawara (n = 19)
(n = 11)
r Downward agreement possible, case dependence indeterminate: Khoekhoe,
Kiowa, Nahuatl, Pirahã, Gooniyandi
r Upward agreement required, case dependence indeterminate: Luvale, Mohawk?
r Not case dependent, direction indeterminate: Maricopa
r Case dependent, direction indeterminate: Bagirmi, Acoma
r Both parameters indeterminate: Wichita
r No Agreement: Sango, Yoruba, Supyire, Grebo, Krongo, Koyra Chiini, Harar
Oromo, Khalkha, Japanese, Korean, Lezgian, Mandarin, Burmese, Meithei, Hmong,
Thai, Vietnamese, Paiwan, Indonesian, Malagasy, Rapanui, Tagalog, Imonda,
Kayardild, Martuthunira, Ngiyambaa, Yaqui, Mixtec, Sanuma (n = 29)
d. Is object agreement sensitive to the case of the object?

e. Is there agreement on complementizers, and if so what kind?
f. Is there agreement on adpositions with an unmoved complement?
g. Is there agreement between a determiner/quantifier and the associated
NP, and if so, does it depend on word order?
h. Is there double agreement with the subject in auxiliary-verb
constructions?
In general, the answers to (111a,c,e–g) are relevant to the setting of parameter

(1), and the answers to (111b,d,h) are relevant to the setting of parameter (2).
The details of what I found are summarized in the table in the appendix.
It turns out that most of the languages can be categorized into one of five
types, as shown in table 1.
Not all parameter settings turn out to be equally common in this survey. Lan-
guages in which agreement is dependent on case are about 33% more common
than languages in which agreement is not dependent on case (40 to 31). More
strikingly, languages that allow downward agreement between a probe and its
goal outnumber languages that require upward agreement by an almost three-to-
one ratio (53 to 19). Languages in which agreement is subject to both restrictions
seem to be particularly rare, with only a handful of plausible examples attested.
It would be nice to understand these asymmetries in the distribution of language
types. But for now I take them to be relatively insignificant to the point at hand.
There is no strong expectation that both values of a binary parameter are equally
distributed in the languages that are currently spoken in the world. The world
is too small a place, and the way that languages diffuse through it and interact
with one another is too nonrandom to expect complete statistical independence.
What I take to be significant about table 1 is the fact that each of the four agree-
ment types is found in a variety of language families and in different areas of
the world. Even the rarest kind, the “Yes-Yes” languages, are attested on four
continents and in six distinct language families. Thus, we are not dealing with
a unique areal phenomenon, like the use of clicks in the languages of Southern
Africa, or the existence of object-initial word order in Amazonian languages.
Rather, the different agreement types are robustly attested in languages of the
world in a way that is comparable to the famous Greenbergian word order types
of SOV, SVO, and VSO (Greenberg 1963, Dryer 1992).
Even more important for the theory of agreement – and the theory of param-
eters – is the question of how consistent are the parameter settings across func-
tional categories. Suppose there was a language that only had agreement on
T. One could use the properties of that agreement to classify the language in
terms of table 1, following the model of section 5.2. But such a language would
not provide evidence that all functional categories are governed by the same
parameter settings. One might also wonder about the (fairly few) languages that
I found to be indeterminate with respect to one parameter or the other. Were
the results indeterminate because I found no convincing evidence for how the
parameter in question is set, or because I found conflicting evidence regarding
how the parameter is set? If it was the latter, then a “microparametric” theory
of agreement that respected the Borer-Chomsky Conjecture might turn out to
be better than the “macroparametric” version in terms of grammatical param-
eters that I have presented. What we need to know in this regard is how many
languages have consistent behavior across more than one functional category,
versus how many languages have inconsistent behavior.
Consider first the Case Dependency of Agreement Parameter. In practice,
there are three principal opportunities to observe the value of this parameter:
whether T agrees with oblique subjects or not, whether v agrees with oblique
objects or not, and whether the language allows more than one head to agree
with the same argument. What then is the degree of convergence between these
three data points?
For T agreeing with an oblique subject and v agreeing with an oblique object,
there is not a lot of relevant data. To evaluate this, we need languages that have
overt subject agreement, overt object agreement, ergative or dative subjects,
and a dative/accusative distinction for objects. Not that many languages have
all four properties. I found approximately sixteen, which pattern as follows:
(112) a. Consistently case-sensitive: 7 (Greenlandic?, Choctaw, Tzotzil?,

Yimas?, Daga, Wari’, Mangarrayi, also Peruvian Quechua – see section 5.11.3.2)
b. Consistently insensitive to case: 7 (Georgian, Burushaski, Warlpiri,
Ika, Basque, Canela-Krahô, Imbabura Quechua – see section 5.11.3.2)
c. T case-sensitive, v not: 1? (Nez Perce – see section 5.11.2.2)
d. v case-sensitive, T not: 1 (Gooniyandi)
There are more consistent languages than inconsistent ones, but there are a
couple of candidates for the latter, and the numbers are too small to draw any
firm conclusions.
Better results come from collapsing the two possible instances of agreement
with oblique NPs into a single data point and correlating it with instances of
multiple agreement with the same NP. The implicational universal predicted
by my parameters is that languages which rule out agreement with oblique
arguments should also not permit multiple agreement with the same argument
(see section 5.9). This is nicely confirmed by the data:
(113) a. No agreement with obliques; no multiple agreement: 20 languages

(Hindi, Apurinã, Daga, Choctaw, Hixkaryana . . .)
b. Agreement with obliques; multiple agreement OK: 11 languages
(Georgian, Burushaski, Ika, I. Quechua, Maricopa, Jarawara . . .)
c. Agreement with obliques; no multiple agreement: 6 languages
(Warlpiri, Basque, Chukchi, Tauya, Dani, Kewa)
d. No agreement with obliques; multiple agreement OK: 0 languages
I take this to be a very positive result for the parametric theory. Note also that
the “mixed” type in (113c) is somewhat less common than either of the consis-
tent types in (113a) and (113b). Such examples can arise when the parameter
values allow multiple agreement, but a suitable situation for multiple agreement
to appear never arises because of lexical factors. Apparently such lexical con-
spiracies to hide the parameter value are not impossible, but are not especially
common.
Next consider the Direction of Agreement Parameter. Here the results are
even more impressive. There are five opportunities to observe this: agreement
on T, on v, on D, on C, and on P. Most languages do not have all five kinds
of agreement, but many have at least two, and quite a few have three or more.
Call a language consistent up if at least two functional heads only agree with
an NP that asymmetrically c-commands them, and all other functional heads
either confirm this value, or are indeterminate, or are not applicable because
they do not bear agreement. Conversely, call a language consistent down if at
least two functional heads can agree with an NP that they c-command, and
all other functional heads either confirm this value, or are indeterminate, or
are not applicable because they do not bear agreement. Call all other com-
binations inconsistent. Obviously there is a large variety of imaginable lan-
guages that would count as inconsistent by this criterion. But there are not
many actual languages that are inconsistent. On the contrary, there are almost
none:42
(114) a. Consistent up languages: 16
b. Consistent down languages: 43
c. Inconsistent languages: maybe 2
(Berber, Nez Perce)
I take this to be strong support for the idea that the parametric values for the
agreement properties of functional heads are not independent of each other. On
the contrary, languages tend to be remarkably uniform in these respects. This
supports the classical GB notion of a parameter as variation in the formulation
of a principle of grammar, rather than the Borer-Chomsky Conjecture that
all parameters reduce to variation in the feature-specifications of individual
functional items.
5.11.2 Two apparent exceptions

Having already cherry-picked some of the more interesting positive examples
for my grammatical parameters in previous sections, let me now cherry-pick
two of the more interesting problematic examples for a bit more discussion.
42 There are two languages, Abkhaz-Abaza and Tzotzil, that have ergative agreement systems
which can be analyzed as v agreeing upward with the agent in Spec, vP while T agrees downward
with the object (see section 3.2.2.1). I assume that v is forced to agree upward in these languages
not by parameter (1) but by parameter (2): v agrees only with an NP that it gives ergative case.
Ergative case is not marked overtly on NPs in these languages, but following O’Herin (2002) I
assume it is present in the syntactic representation and guides agreement. If this analysis proves
untenable, these two languages might be switched from the consistent-down category to the
inconsistent category.
Showing why they seem problematic hopefully illustrates what my thesis is,
for better or for worse. These languages are, in my current understanding, as
bad as it gets for my hypothesis, and I believe that they are not so bad. There
are other things going on in the languages which, while far from being fully
understood, might very well affect how the parameters apply. I am not so bold
as to say that these languages support my parameters. I am open to the idea
that they are neutral or even genuine counterexamples to it – either because
lexical parameters happen to conceal the grammatical parameter settings, or
for some other reason. But the macroparametric view has at least the virtue of
pressing one to dig deeper into the syntax of these languages, and when one does
one discovers some interesting things. I consider the two possibly inconsistent
languages mentioned in (114), Berber and Nez Perce.
5.11.2.1 Berber
Berber is an Afroasiatic language, related to Hebrew, Arabic, and Hausa. It
is like the Niger-Congo languages and Arabic (but unlike Hebrew or Hausa)
in allowing multiple agreement in complex tense constructions, as shown in
(115).
(115) La-n la teddu-n . . . (Ouali and Pires to appear)

past-be.perf-3pS pres leave-impf-3pS
‘They were leaving/had been leaving . . .’
This suggests that agreement in Berber (like Arabic) is not case-dependent in

the sense of parameter (2). However, Berber does not have an overt dative case
or other oblique cases, so we cannot confirm this parameter setting by checking
whether agreement is possible with oblique noun phrases or not. Berber is
thus consistent with respect to this parameter, although not in a particularly
interesting way.
The difficult issue concerns the Direction of Agreement Parameter in (1).
Berber is a VSO language, in which the finite verb routinely agrees with the
postverbal subject:
(116) a. I-zra urgaz lfilm. (Sadiqi 1997)

3sS.m-see man.cs film
‘The man saw a film.’
b. N-zra-t nuKni.
1pS-see-him we
‘We saw him.’
This strongly suggests that Berber allows downward agreement from T to the
subject in Spec, vP (or some other specifier position lower than Spec, TP).
In contrast, D-like heads in Berber only agree with NPs to their left, not with
NPs to their right. For example, maTa ‘which’ precedes the associated NP and
is invariant, whereas the ordinal numeral ‘second’ follows the NP and agrees
with it:
(117) a. maTa urgaz (Sadiqi 1997: 128)

which man.m.sg.cs
‘which man?’
b. argaz wiSin; tamtut tisnat
man.m.sg second.m.sg woman.f.sg second.f.sg
‘the second man; the second woman’
Even more striking, the word kuLu ‘all’ can come either before the associated
nominal or after it. When it comes before the nominal, it is invariant, but when
it follows the nominal it agrees with that nominal:
(118) a. kuLu irgzn (Sadiqi 1997:128–9)

all men.m.pl
‘all (the) men’
b. irgzn kuLu-tn; tajtšin kuLu-tnt
men.m.pl all-m.pl women.f.pl all-f.pl
‘all the men; all the women’
This is strongly reminiscent of the behavior of D-like heads in Kinande and

Swahili, and suggests that D-like heads in Berber can only agree with some-
thing that c-commands them. We have, then, a possible instance of inconsistent
parameter settings in Berber: parameter (1) is set “no” for finite T and “yes”
for D. (Other functional heads in Berber give no additional data on this issue,
since they are not agreement bearers.)
Can anything be made of this relatively rare anomaly, beyond identifying it
is as an apparent counterexample? One dull but reasonable possibility is that
Berber does in general allow downward agreement, as IE languages do. That
covers (116) in a straightforward way. One can then say that it is simply a
coincidence (synchronically speaking) that the Ds that precede the associated
NPs happen not to be agreement-bearers and (some of) the Ds that follow NP
are. The word kuLu ‘all’ could have two distinct lexical entries, one that agrees
and one that does not. Such an analysis would be descriptively adequate. It
may fail to capture a generalization about the relationship between word order
and agreement within DP, but since D is a closed class with a relatively small
number of members the loss would be a fairly small one.43 (Note that there
are also Ds that follow NP and do not manifest agreement with it, such as the
demonstratives ad ‘this’ and ann ‘that’ (afrux-ad ‘this boy’ vs. ifruxn-ad ‘these
boys’ (Ouhalla 2005a: 628)). Thus lexical properties that determine whether a
D agrees or not must come into the story on all accounts.)
A riskier but potentially more rewarding possibility is that the Ds show that
the true setting of the Direction of Agreement Parameter in Berber is “yes,”
and something special is going on in VSO examples like (116). What could
it be? Conceivably the subject NP moves from Spec, vP to Spec, TP, thereby
triggering upward agreement on T, just as in a Bantu language. The difference
could be that (exceptionally) the lower copy of the subject NP is spelled out
in Berber and the higher copy is deleted, whereas it is the higher copy that is
spelled out in Kinande and most other languages. This derivation is sketched in
(119), where strikethrough indicates the copy of a moved item that is deleted
at PF.
(119) [TP mani T{AGRi }+seek [vP mani v [VP seek film]]]
Bobaljik (2002) defends the idea that sometimes it is the lower copy of a
movement chain that is spelled out at PF rather than the higher one, and explores
some of the ramifications of this idea (see his article for other references). This
analysis would be a kind of revival in current terms of an earlier proposal
concerning VSO in Berber by H.-S. Choe (1986), who claimed that the subject
moves downward from the normal subject position so as to appear after the
finite verb in Berber.
One should probably not consider adopting this second, riskier analysis unless
one has at least a glimmer of an answer to why the subject chain should be
spelled out differently in Berber than in most other languages. Fortunately,
Choe had an idea about this that still seems relevant. She claimed that principles
of case assignment determine where the subject NP appears in Berber. More
specifically, she claimed that case is only assigned by the agreement-bearing
verb to its right in Berber. And indeed there are some features of subject case-
marking in Berber that do seem typologically unusual and could plausibly
43 Probably I would take this route for related Hebrew. For Hebrew, the evidence seems quite
strong that T can agree downward with the subject, even if something else has moved to Spec,
TP, given detailed studies like Borer 1995 and Shlonsky 1997. There are a couple of individual
lexical items in Hebrew that seem to agree only with a DP on their left (’eyn ‘not’ and kol ‘all’;
Shlonsky 1997) but there are no large classes of items that systematically behave this way. I thus
assume that this agreement behavior reflects the lexical properties of those particular items, not
the general parameter settings of Hebrew.
be related to the language’s anomalous agreement and word order properties.

Subject case in Berber is definitely not the familiar nominative case known
from Indo-European languages. Rather, the postverbal subject appears in the
so-called construct state (Guerssel 1983, Sadiqi 1997). This form is also found
on the objects of prepositions and on NPs within NP, so it corresponds not
only to the IE nominative but also to the IE dative and genitive. In each of
these domains, the NP in construct state immediately follows its case-assigner
(V, P, or N). The phonological realization of construct state is also somewhat
unusual: whereas nominative case is usually unmarked, the construct state is a
marked form, realized as a phonological change at the beginning of the NP that
involves deleting the first vowel of the word along with sometimes inserting
a glide (see Guerssel 1983 for details). Perhaps then construct case is really a
form of phonological weakening that results from the case-marked NP forming
a prosodic unit with the word that immediately precedes it. If so, it is not
implausible that the PF rules that spell out case in this unusual way might take
precedence over the normal rules of chain realization, causing the lower copy
to be spelled out rather than the higher one, consistent with Bobaljik’s (2002)
vision.
Another hint that this might be so involves the placement of overt subjects
in complex tenses. This seems not to be entirely fixed: the subject can appear
immediately to the right of the auxiliary verb or immediately to the right of the
main verb, as shown in (120).
(120) a. Ila Ali da thiddu. (Ouali and Pires to appear (14))

past.be.3sS.m Ali fut go.aor.3sS.m
‘Ali was going to leave.’
b. Da illi-n la teddu-n lwashun.
fut be.aor-3pS pres go-impf.3pS children
‘The children will be leaving.’
(Ouali and Pires to appear (19))
Still, the freedom of placement of the subject is not unrestricted; it seems never
to appear between the tense markers da and la and the main verb, for example.
This limited freedom is captured by the descriptive statement in (121).
(121) The case-marked subject must be spelled out right-adjacent to an

inflected verb (with which it forms a phonological unit?).
But the statement in (121) also explains why the copy of the subject in Spec,
TP cannot be spelled out in (119), whereas the copy in Spec, vP can be. To the
extent then that the PF spell-out rule in (121) is already motivated by (120), we
have independent reason to think that PF spell-out also governs the appearance
of VSO rather than SVO word order in Berber.
Finally, this theory makes the prediction that objects and PPs should not be
able to move to Spec, TP to derive X-V-S orders in which the verb agrees with
the subject in Berber. Such sentences are sometimes possible in languages that
really do allow downward agreement, including Icelandic, Yiddish, Finnish,
and Hebrew. The reason why this should not be possible in Berber is simply
because the Spec, TP position is occupied in the syntax by the highest copy of
the agreed-with subject; otherwise the subject would not be agreed with, given
a “yes” setting of parameter (1). If the subject is in Spec, TP in Berber, then it
follows that that position is not an open landing site for movement of another
phrase. This prediction seems correct: phrases other than the subject can appear
on the left edge of the clause, but only as a result of clefting or left dislocation,
not as a result of simple movement to Spec, TP. Thus, the fronted object in (122)
requires there to be an object clitic on the verb, whereas in (123) the fronted
object lands in the specifier of a C/Focus head rather than in Spec, TP.
(122) Tabrat i-ara-*(T) hmed. (Sadiqi 1997:148)

letter.f.sg 3sS.m-wrote-it.f Ahmed.
‘The letter, Ahmed wrote it.’
(123) Argaz a t-zra. (Sadiqi 1997: 217)
man c 3sS.f-see
‘It’s a man that she saw.’
There are of course many other reasons why a particular language might forbid
the movement of a direct object to Spec, TP. But there are languages that do
allow such movement, and it is consistent with the proposed analysis that Berber
is not one of them.
The important point here is probably not that I can think of ways to save
my theory; that might always be the case. The most important point about
problematic-looking languages like Berber is how rare they are, as documented
in section 5.11.1. But one does also learn more when one looks into the details
about why a given language might seem anomalous. In this particular instance,
one learns something about the special way that “case” is realized in Berber,
and how that might interact with issues of agreement and word order.
5.11.2.2 Nez Perce

Nez Perce, a Sahaptian language spoken in Idaho, has an interesting and unusual
case and agreement system discussed originally by Noel Rude (1982, 1986a,
1986b, 1988) and in generative terms based on his work by Ellen Woolford
(1997) (see also Carnie and Cash Cash 2003). The puzzle for my parameters
stands out clearly in Woolford’s discussion, so I begin by following her analysis.
(Throughout the discussion, I concentrate on the person-marking affixes of Nez
Perce. The language also has affixes that mark plural number for subjects and
definite objects, as seen for example in (125b), but I leave these aside because
they appear to raise no new issues.)
Consider first subject agreement, which is presumably on T (note that it is
present only in finite clauses, not in nonfinite ones (Rude 1988)). In Nez Perce,
this is primarily realized by the alternation between Ø for first or second person
versus hi for third person.
(124) a. Ø-páay-na ‘ı́in. (Rude 1986a:126)

1S-arrive-asp I
‘I arrived.’
b. Hi-páay-na háama. (Rude 1986a:126)
3S-arrive-asp man
‘The man arrived.’
Downward agreement on T is presumably allowed because T can agree with

a postverbal subject, as it does in Spanish but not Zulu. More generally, word
order is extremely free in Nez Perce, and has no effect on agreement (Rude
1986a:125). Thus, there is no evidence that the position of an NP is crucial
to whether it is agreed with or not, making Nez Perce quite different from
Kinande, Apurinã, and other languages in which (1) is set as “yes.” The Nez
Perce verb also shows hi agreement with its third person subject regardless of
whether that subject is in nominative case or ergative case, as shown in (125)
(Rude 1988:Ex. (31)) (Woolford 1997:Ex. (36)) (as long as it is not blocked by
the more specific morpheme pee; see (127b)).
(125) a. Háama hi-‘wı́-ye wewúkiye. (Rude 1988:552)

man.nom 3S-shoot-asp elk
‘The man shot an elk.’
b. Háama-nm hi-néec-‘wi-ye wewúkiye-ne. (Rude 1986a:127)
man-erg 3S-pl.O-shoot-asp elk-obj
‘The man shot the elk (pl).’
Therefore, subject agreement is independent of case marking in Nez Perce,

as argued by Woolford (1997:213). In my terms, the Case Dependency of
Agreement Parameter in Nez Perce is also set as “no.” So far then, Nez
Perce seems to be a No-No language, in the same category as Georgian and
Burushaski. Moreover, these are the same parameter values that we observed
for agreement on C-like heads in Nez Perce; see the discussion of (43) in section
5.4.2. (P and D do not manifest any agreement in ϕ-features with their associated
NPs, so they give no information about the setting of these parameters.)
The problems arise when one considers agreement with objects. Object agree-
ment is shown most clearly by the alternation between Ø (for 1/2 objects) and
‘e (for third person objects) (Rude 1986a):
(126) a. ‘ı́in Ø-‘ewı́-ye ‘ime-né. (Rude 1986a:126)
I (1S/2O)-shoot-asp you-obj
‘I shot you.’
b. ‘ı́in ‘e-‘wı́-ye wewúkiye-né. (Rude 1986a:126)
I (1S)/3O-shoot-asp elk-obj
‘I shot the elk.’
When the subject is third person and the object is first or second person, the
prefix is hi as expected ((127a)). A slight (but significant) wrinkle is that when
the subject and object are both third person, one does not get the sequence of
morphemes hi+‘e, but rather the portmanteau morpheme pée ((127b)).
(127) a. Háama-nm hi-‘wı́-ye ‘iin-e. (Rude 1986a:126)
man-erg 3S/(1O)-shoot-asp I-obj
‘The man shot me.’
b. Háama-nm pée-‘wi-ye wewúkiye-ne. (Rude 1986a:126)
man-erg 3S/3O-shoot-asp elk-obj
‘The man shot the elk.’
Now comes the complication: the morphemes ‘e and pée, which explicitly
show third person object agreement, are in a sense optional. They appear if
and only if the object is definite/topical in the discourse, and if it is marked by
the objective case marker ne as in (126b) and (127b). If the object is indefi-
nite/nontopical and not marked by ne, the agreement prefix is Ø or hi, the same
prefixes used on intransitive verbs as in (124):
(128) a. Kawá taxc qáamsit Ø-wiyáamk-o’ (Rude 1986a:23)
then soon qáamsit 1S-peel-asp
‘Then soon I will peel the qáamsit fruit . . .’
b. Háama hi-‘wı́-ye wewúkiye. (= (125a))
man.nom 3S-shoot-asp elk
‘The man shot an elk.’
So unlike the subject prefix hi, which is used regardless of whether the subject
bears ergative case or unmarked nominative, the object prefix ‘e is only used
when the object bears objective case, not when it is unmarked accusative (in
Woolford’s terminology). It thus seems that agreement on v is case-dependent
in a way that agreement on T is not. Furthermore, the fact that verbs agree with
the object only when the object is somehow definite/topical in Nez Perce recalls
the behavior of “optional” object agreement in Zulu, analyzed in section 5.7.1.
I analyzed the Zulu facts as indicating that v could only agree upward, with
an object that had moved to Spec, vP, out of the domain of existential closure.
By parity of reasoning, it would be reasonable to view object agreement in
Nez Perce in the same light, and indeed Carnie and Cash Cash (2003) propose
essentially that analysis. If so, then agreement on v must probe upward, whereas
agreement on T can probe downward. It seems, then, that v in Nez Perce has quite
different agreement properties from T and C in the same language, contrary to
what one would expect if the agreement parameters are grammatical in nature.
But once again, if one uses this cognitive dissonance as a spur to look more
deeply into the language, one finds other interesting properties that might be
relevant. Notice that I moved very quickly from “agreement with the object”
to “agreement on the v head” in the previous paragraph, without citing any
evidence for this equivalence. The v head is the usual locus for agreement with
an object, but that is not necessarily universal. In fact, there is some interesting
evidence that object agreement, like subject agreement, is a property of T in
Nez Perce. If so, the picture changes dramatically.
Some of the evidence that object agreement is on T rather than v is relatively
superficial. First, there are never two distinct person agreement morphemes
on the Nez Perce verb; rather there is one morpheme that expresses both (see
especially pée in (127b)). This systematic fusion could be a sign that there is
only one head (T) that agrees with both arguments. Second, I mentioned that
subject agreement does not appear on nonfinite verbs in Nez Perce, as one
would expect. But neither do object agreement morphemes like ‘e (Noel Rude,
personal communication). Object agreement is just as dependent on finiteness
as subject agreement is, suggesting that both are housed in T. In this respect, Nez
Perce is different from Bantu languages, Mapudungun, and Quechua, in which
object agreement is found in nonfinite clauses (evidence that object agreement
is agreement on v, not T in those languages). Third, reflexive morphology in
Nez Perce replaces both subject agreement and object agreement (Aoki 1973),
rather than bleeding ordinary object agreement and leaving subject agreement
unaffected, as in many other languages.
The most sophisticated evidence that object agreement is on T in Nez Perce
comes from clauses that have a possessed subject. A verb that has an intransitive
third person subject that is not possessed bears the prefix hi, as in (124b). But a
verb that has an intransitive subject that contains a third person possessor may
bear the ‘e prefix, and if the subject contains a first or second person possessor,
the verb may have the Ø prefix:
(129) a. (‘ı́inim) pike Ø-wı́ine. (Rude 1986b:124–5)

my.gen mother 3+1-weep.asp
‘My mother wept.’
b. Kaa cáalii-nm pike ‘e-wéeyikse. (Rude 1986b:113)
and Charlie-gen mother 3+3-cross.asp
‘And Charlie’s mother went across.’
Woolford (1997) puts this type of agreement aside; Rude (1986b) discusses it
but does not fully integrate it with other kinds of agreement in Nez Perce. This
is understandable: it is an unusual sort of agreement, and not easy to integrate
into the system. Nevertheless, the ‘e and Ø morphemes in (129) are presumably
the same morphemes that are used as object agreement morphemes in (126),
since they are both homophonous with them and sensitive to the same person
features (‘e signals that there is a third person NP distinct from the subject in
both structures; Ø signals the presence of a first or second person NP in both).
It is thus desirable to collapse the two into a unified analysis.
What then is the nature of the ‘e prefix in (129b)? If it were a realization of
agreement on v, then it could not manifest agreement with the possessor of the
thematic subject, because neither one would c-command the other at any point
in the derivation. The only way that the c-command condition on agreement is
satisfied is if ‘e is a spell-out of agreement on T, because T c-commands the
possessor when the possessed subject is still in Spec, vP. In other words, the
structure of (129b) must be something like (130).
(130) TP
T vP
Agree(1)
NP v´
Agree(2)
NP N v VP
Charlie mother V
(GEN)
go.across
The same is true for the null allomorph of agreement found with a nonthird
person possessor in (129a). Now assuming that the occurrences of ‘e and Ø as
object agreement in (126) are instances of the same morphemes, then these too
must be realizations of agreement on T. More specifically, I propose that finite

T probes its domain twice in Nez Perce. The first time, it finds the subject and
agrees with it; the second time it finds either the direct object or the possessor
of the subject. The result is an ordered pair of features on T. T is then spelled
out at PF according to a set of disjunctive realization rules something like the
following:44
(131) a. T <3, 3> → pee-
b. T <X, 3> → ‘e-
c. T <X, 1/2> → Ø-
d. T <3 . . .> → hi-
e. T <1/2 . . .> → Ø-
What are the implications of this analysis for the agreement parameters? One
clear implication is that there is no longer any question of agreement on v having
different syntactic behavior than agreement on T, simply because we now hold
that there is no agreement on v. The question now is whether the two agreement
relationships that T enters into are consistent with respect to the parameters in (1)
and (2). The answer seems to be yes. The object/possessor agreement realized
as ‘e or Ø may be downward agreement, as shown in (130). That is consistent
with the parameter setting that we observed for simple subject agreement on
T and for agreement on C. Also the object/possessor agreement realized as ‘e
or Ø can be agreement with an NP marked with the objective case ne, as in
(126), or it can be agreement with an NP marked with the genitive/ergative
case nm, as in (129). Thus, the second T agreement is not case-dependent,
consistent with the parameter setting that we saw for simple subject agreement
(hi) and for agreement on C. Once we realize that agreement with objects and
agreement with possessors is (at least partly) the same phenomenon, it becomes
clear that this sort of agreement can be downward and is not case-sensitive.
There is thus no inconsistency in the parameter settings in Nez Perce after
all.
One remaining question is why indefinite objects which remain in VP and
do not bear the objective case ne cannot trigger object agreement on the verb.
44 This is not a complete solution to the mysteries of Nez Perce agreement. Unlike (131b-c), the
rule in (131a) must be limited to transitive clauses: it spells out Ts where the object as well
as the subject is third person, but not Ts in which the possessor as well as the subject is third
person. Second, these rules do not account for the hi prefix found with a third person subject
and a first or second person object in (127a). Perhaps this needs to be a second hi morpheme,
ordered before (131c). The underlying question here is whether it is better to treat hi or Ø as the
default morpheme in this system, since neither is used in a natural class of environments.
I do not have a definitive answer to this, but a promising suggestion is that one
might make use of Chomsky’s (2000, 2001) idea that transitive vPs are phases.
As a phase, the vP should be opaque to agreement relationships. In the normal
case, when object agreement is agreement on v, the Agree relationship between
v and the object is internal to the vP phase, so there is no problem. But when
object agreement is on T, as in Nez Perce, it is in a different phase from the base
position of the object. The phase condition on agreement then implies that T
can agree with the object (downward) if and only if the object raises to the edge
of the vP phase, taking it outside the domain of existential closure, as shown in
(132).
(132) TP
T Agree(1) vP
NP vP
man NP v´
(ERG)
elk v ∃ VP
Agree(2) OK (OBJ)
V NP
*Agree(2) by phase condition
shoot <elk>
object shift
Therefore object agreement is possible only with strong/definite objects in Nez

Perce, not because agreement must be upward (as in Zulu), but rather because
the agreeing head is in a different phase from the agreed-with NP. This makes
the strong prediction that (barring parameterization in whether vP is a phase or
not) all languages in which T agrees with the object must be languages in which
only definite objects are agreed with, regardless of other parameter settings. I do
not know if this is true, but Inuit and Chukchi are two other languages that seem
to support this. They have also been analyzed as having Ts that agree with both
the subject and the object (Bok-Bennema 1991 for Inuit, Bobaljik and Branigan
2006 for Chukchi). Moreover the agreed-with object in absolutive case clearly
moves higher than propositional operators on the verb in Inuit (Bittner 1994)
and Bobajlik and Branigan observe that this seems to be true in Chukchi as well,
although the matter has not been studied as carefully. So the overall approach
to Nez Perce seems promising, and this language should not cause us to doubt
too much the existence of grammatical parameters for agreement.
5.11.3 Two minimal comparisons

One of the strongest tests that a macroparametric proposal can be put to is the
test of closely related languages. Suppose that, through a process of histori-
cal change, a language develops a setting for a grammatical parameter that is
different from that of the protolanguage, which is passed down to most of the
daughter languages. Such changes might not be very frequent, but they must be
possible given that even languages we think of as unrelated presumably came
from a single language at a more remote time (e.g., the languages of South
America). The strong prediction is that the effects of this parametric change
should tend to show up in different areas of the innovative language’s syntax. In
contrast, a purely microparametric analysis, in which grammatical parameters
are avoided in favor of lexical parameters, does not have this expectation. On
the microparametric view, the properties of one functional category could very
well change without having any effect on the behavior of another functional
category.
I have not looked systematically for evidence that bears on this question,
for both practical reasons and theoretical reasons. Practically speaking, the
choice to work with the 100 languages from the WALS core language sample
amounts to working with a set of languages that are more distantly related.
There are not many minimal pairs of closely related languages in that sample,
by virtue of its basic design. From a theoretical perspective, I would expect
lexical factors to be very prominent in cases where parametric change has
happened, perhaps hiding the grammatical parameter settings to an unusual
degree (otherwise parametric change might not be possible at all). Nevertheless,
I have stumbled across a few cases that are relevant to this issue, more or less
by accident. One is Makhuwa, a Bantu language that seems to differ from
its kin in allowing downward agreement, thereby having a different setting
of parameter (1). Another is Imbabura (Ecuadorian) Quechua as compared
to Huallaga (Peruvian) Quechua, the two languages apparently differing in the
setting of parameter (2).45 I bring this study to a close with a short presentation of
these two case studies, claiming that they illustrate in microcosm what the 108-
language survey illustrates on a larger scale – the fact that the two parameters
apply to a range of different functional heads.
45 Another case of parametric variation within a single family, not chosen for illustration here, is
Nepali and Maithili having a different setting for the Case-Dependency of Agreement parameter
from Hindi and other IE languages; see note 4 for an outline of the data relevant to this. There
seems to be a fair amount of variation in the Afroasiatic languages: Hausa and Hebrew differ
from Arabic in the setting of parameter (2), and Berber differs from Arabic in the setting of (1)
if the analysis proposed at the end of section 5.11.2.1 is correct.
5.11.3.1 Makhuwa versus Kinande/Chichewa

Makhuwa is spoken in Mozambique, on the eastern extreme of the Niger-Congo
area; I thank Jenneke van der Wal for calling it to my attention and sharing data
with me. Van der Wal (2006) observes that Makhuwa is different from many
other Bantu languages in that the verb does agree in noun class with the subject
in thetic clauses with verb-subject word order, as shown in (133).
(133) a. O-hoó-khwá mwanámwáne.

1S-perf.disj-die 1a.child
‘There has died a child.’
b. A-náá-ttónyá maátsi.
6S-pres.disj-drip 6.water
‘There is water leaking out.’
The finite verb also agrees with the postverbal subject in clauses that have a
PP-V-S order, in contrast to Kinande and Chichewa where agreement is with
the fronted PP (van der Wal, personal communication):
(134) Watarátú a-rı́ a-kwaatú.

16.roof 2S-be 2-cats
‘On the roof are cats.’
Van der Wal (2006) treats the difference between Makhuwa and other Bantu
languages like Sesotho in more or less microparametric terms. She claims
that other Bantu languages are like French in having an expletive (il) that has
ϕ-features, which the verb must agree with. In contrast, Makhuwa is like Span-
ish and Italian in not having an expletive with ϕ-features, so the verb is left free
to agree with the postverbal subject in sentences like (133). This may indeed be
a factor in the overall analysis. But her analysis also assumes that finite T can
agree with a subject that is still inside VP in (133) and (134), a kind of down-
ward agreement that we have seen to be banned in other Bantu languages. And
if there is a macroparametric difference between Makhuwa and other Bantu
languages in this respect, then we might expect to see the effects of this on
other aspects of agreement in the language.
In this light, it is striking that object agreement in Makhuwa also behaves very
differently from object agreement in languages like Kinande and Chichewa (van
der Wal, personal communication). Kinande and Chichewa have a complete set
of object markers, and these appear if and only if there is no overt object or
the object is dislocated to the edge of the clause. In contrast, Makhuwa has
lost its object markers for all but class 1 and class 2 objects. Moreover, the
object markers for these two classes act as obligatory object agreement: they
are required whenever the object of the verb is class 1 or 2, even when that
object occurs in immediate postverbal position and has an indefinite reading:
(135) a. Ki-n-ı́n-thúmá namarokoló.

1sS-pres-1O-buy 1a.rabbit
‘I am buying a rabbit.’
b. O-hi-n-thel-é nthı́yáná oowoótha
2sS-neg-1O-marry-opt 1.woman 1.lying
‘Don’t marry a lying woman.’
There is no reason from word order or definiteness to think that the object
has moved out of the verb phrase in (135b); nevertheless it triggers class 1
agreement on the transitive verb (technically, on v). This shows that v also agrees
downward in Makhuwa. Thus, both subject agreement and object agreement
work in a systematically different way in Makhuwa than in many of its Bantu
relatives. I take this as support for the macroparametric view that parameter (1)
governs the complete set of functional heads in the language.
The other functional heads in Makhuwa are less striking, but are consistent
with (1) being set as “no.” There is no agreement on C or P, hence no evidence
from those domains that agreeing heads must probe upward. The other relevant
head is D. Recall that most D-like heads in Kinande and Swahili follow the
associated NP and agree with it, but the one or two that precede NP cannot
agree with it. Given that Makhuwa allows heads to agree downward, it might
allow D-NP order with agreement on the D, as IE languages do. In fact, judging
by Woodward 1926, most D-like heads follow the NP in Makhuwa, so are not
relevant to the prediction. Woodward does, however, give some examples of
demonstrative doubling, where one token of the demonstrative comes before
the NP and the other one comes after it, as shown in (136).
(136) a. yo-la m-tu’ o-la (Woodward 1926:285)

1-this 1-man 1-this
‘this very man’
b. I-la inupa i-la y-ulupale. (Woodward 1926:300)
3-this 3.house 3-this 3-large
‘This house is large.’
Observe that the preceding demonstrative agrees with the NP in number and
noun class as much as the following demonstrative does. This is possible
additional support for the claim that Makhuwa allows downward agreement –
although one would like to know more about the structure of nominals in the
language to be sure that we know how to interpret the data in (136).
Makhuwa does not seem to differ from its Bantu relatives with regard to its
setting for parameter (2). In particular, it is like the other Bantu languages in
permitting complex tenses in which the auxiliary verb and the main verb both
show full person-number-gender agreement with the same argument (Jenneke
van der Wal, personal communication):
(137) a. Ki-háana ki-khumáka

1sS-have 1sS-go.out.dur
‘I have to go out.’
The fact that a Bantu language can show a consistent change with respect to the
direction of agreement without agreement developing case sensitivity confirms
that (1) and (2) are two independent parameters. One cluster of properties
changes together when the parameter setting changes, but the other cluster of
properties does not.
5.11.3.2 Ecuadorian and Peruvian Quechua

My second case study involves a change in the value of the Case Dependency
of Agreement Parameter. Ecuadorian Quechua – specifically the Imbabura
Quechua described by Cole (1985) and others – seems to differ from other
Quechua varieties in having a negative value for this parameter. (I do not
know for sure which is the innovative value of the parameter, but tenta-
tively assume that it is the Imbabura value, since it seems to have a narrower
distribution.)
One sign of the parametric difference comes from auxiliary constructions.
While agreement only on the auxiliary is the norm in both types of Quechua,
Cole points out that some Imbabura Quechua dialects have agreement on both
the main verb and the auxiliary in past conditional tenses:
(138) a. Shamu-y-man ka-rka-ni. (Cole 1985:155–6)

come-1sS-cond be-past-1sS
‘I would have come.’
b. Shamu-ngui-man ka-rka-ngui.
come-2sS-cond be-past-2sS
‘You would have come.’
Cole himself points out that other Quechuan languages have only default third
person agreement on the auxiliary in examples comparable to (138). That is
true in Huallaga Quechua, for example, as shown by (139), which is otherwise
perfectly cognate to (138b).
(139) Sha-mu-nki-man ka-ra-n. (Weber 1989:104)

come-afar-2sS-cond be-past-3S
‘You should have come.’
(138) shows that parameter (2) is set “no” in Imbabura Quechua, whereas the
systematic absence of such double agreement suggests that it is set “yes” in
Peruvian Quechua.
Given this difference, the macroparametric view predicts that Imbabura
Quechua should be more amenable to having the verb agree with oblique
NPs than Peruvian Quechua. This is true. Imbabura has verbs that take objects
marked with dative case man. Moreover, the same first singular object agree-
ment appears on the verb regardless of whether the object bears accusative case
or dative case, as shown in (140).
(140) a. Maria-ca (ñuca-ta) maca-wa-rca-mi. (Jake 1985:30)

Maria-top I-acc hit-1sO-3.past-val
‘Maria hit me.’
b. Maria-ca (ñuca-man) cara-wa-rca-mi.
Maria-top I-dat serve-1sO-3.past-val
‘Maria served me.’
Indeed, Jake (1985:30) shows that some Imbabura verbs can even agree with
objects in genitive case (benefactees) and instrumental case (comitatives). Thus,
agreement on v is clearly not sensitive to the accusative/oblique distinction
in Imbabura Quechua, as expected if parameter (2) is set “no”. In contrast,
there does seem to be a close relationship between case and object agreement
in Huallaga Quechua: an argument triggers object agreement only if it bears
accusative case. It is true that the goal argument of a ditransitive verb like ‘give’
can trigger object agreement in Huallaga, as in Imbabura (see (141a)), but that
is only because goal arguments bear the same accusative case marking ta as
theme arguments in Huallaga (unlike Imbabura, see (141b)).
(141) a. Pay qam-ta qu-ma-ra-n (Weber 1989:98)

he you-obj give-1sO-past-3S
‘He gave you to me.’
b. Hwan-ta papa-ta qo-ykusha. (Weber 1989:23)
Juan-obj potato-obj give-past.3S
‘He gave the potato to Juan.’
Only NPs marked with ta trigger object agreement in Huallaga Quechua, a sign
that its value for parameter (2) is “yes.” The two Quechuas thus differ not only
in the pattern of agreement in auxiliary constructions, but also in the ability to

agree with oblique NPs, as expected.
Finally, we want to check if the same contrast between the varieties is found
with respect to oblique subjects. I argue that the answer is yes. On the basis
of its “no” setting for parameter (2), one might expect Imbabura Quechua to
allow agreement with oblique subjects, whereas Peruvian Quechua would not.
In fact, there is an important difference between the languages when it comes
to oblique subjects, discussed especially by Hermon (1985, 2001). Imbabura
Quechua has experiencer constructions in which the experiencer argument bears
overt accusative case and there is no nominative subject (Cole 1985:107–8):
(142) a. Juzi-ta puñu-naya-n.

José-acc sleep-desid-3S
‘José wants to sleep; José is sleepy.’
b. Juzi-ta rupa-n.
José-acc be.hot-3S
‘José is hot.’
These accusative case experiencers behave like subjects in certain respects, and
not like the direct objects that they superficially resemble. For example, the
experiencer of a desiderative verb like (142a) can be controlled when the verb
itself is an infinitive selected by a suitable main clause verb (Hermon 2001:161):
(143) Ñuka-ka [- - puñu-naya-y-ta] kati-ju-rka-ni.

I-top sleep-desid-inf-acc continue-prog-past-1sS
‘I continued to desire to sleep.’
Similarly, the experiencer argument of such a verb can be treated as the subject
of the clause for the purposes of switch reference marking (Hermon 2001:162):
(144) [-- mika-naya-shpa] aycha-ta randi-rka-ni.

eat-desid-ss meat-acc buy-past-1sS
‘When I desired to eat, I bought meat.’
The inflection shpa on the dependent verb is normally used only when the
subject of that verb is coreferential with the subject of the matrix verb. Its use
in (144) shows that the experiencer argument of a desiderative verb counts as a
subject for these purposes, despite being case-marked like an object. So there
are accusative-marked subjects in Imbabura Quechua, as well as nominative-
marked ones.46
46 Hermon (1985, 2001) shows that there are some differences between the accusative subjects
of desiderative verbs like (142a) and those of simple psych verbs like (142b). Descriptively
In contrast, there seem not to be accusatively marked subjects in Peruvian

Quechua. Constructions that are equivalent to those in (142) seem to be absent in
Huallaga Quechua, for example. There is no mention of such a use for accusative
case in Weber (1989), and the cognate of the desiderative morpheme in (142a)
is not used in the same way in this variety (Weber 1989:170). Hermon (1985,
2001) shows that examples that are superficially similar to (142) are found in
Huanca Quechua, a Central Peruvian variety:
(145) a. Yaqa-kta tushu-na-ykaa-ma-n. (Hermon 2001:171)
me-acc dance-desid-prog-1sO-3S
‘I desire to dance.’
b. Yaqa-kta alalaa-ma-n.
me-acc cold-1sO-3S
‘I am cold.’
But it is striking that accusative case experiencers in Huanca Quechua do

not behave syntactically like subjects the way their equivalents in Imbabura
Quechua do. (146a) shows that the experiencer of a desiderative verb cannot be
controlled in Huanca; (146b) shows that it cannot license the use of same-subject
marking (Hermon 2001:171).
(146) a. *Yaqa muna-a-chu [-- mika-na-y-ta].
I want-pres.1-neg eat-desid-inf-acc
‘I want to desire to eat.’
b. *[-- kasara-na-la-l] alli wamla-kta ashi-lla-a.
marry-desid-stat-ss good girl-acc look.for-affect-pres.1sS
‘When I desire to marry, I will look for a good girl.’
Indeed, Hermon says that accusative case experiencers have virtually no sub-
ject properties in Huanca Quechua, and behave almost identically to the normal
objects of transitive verbs.47 Thus, Imbabura Quechua has accusative case sub-
jects and Peruvian Quechua does not.
I analyze these differences as being rooted in the setting of parameter (2).
Since (2) is set “yes” in Peruvian Quechua, T cannot agree with the experiencer,
which is assigned an inherent accusative case. Therefore, the experiencer does
speaking, the experiencers of desiderative verbs are fully subject-like in their syntactic behavior,
whereas those of simple psych verbs show a more mixed behavior. I offer no analysis of this
interesting contrast.
47 The one difference between accusative experiencers and normal direct objects in Huanca
Quechua that Hermon (1985, 2001) discovered is that accusative experiencers cannot be
extracted from embedded clauses in a manner reminiscent of the ECP of Chomsky (1981).
I have nothing to say about this fact.
not have the “subject” properties that go along with being in an agreement
relationship with T. In contrast, (2) is set “no” in Imbabura Quechua, so T
can agree with the accusative-marked experiencer in this variety. Therefore,
the experiencer is similar to an ordinary nominative subject in those respects
that relate to it having an Agree relationship with T. For example, nonfinite T
in Quechua makes it possible for the argument it agrees with to be PRO; that
extends to experiencers of desiderative verbs in Imbabura but not in Huanca.
Similarly, the dependent T in an adverbial clause is spelled out as a same-subject
marker if the NP it agrees with is referentially dependent on the NP that the T
in the main clause agrees with. That includes the experiencer of a desiderative
verb in Imbabura but not in Peruvian varieties. This is not a full theory of control
or switch reference in Quechuan languages, of course, and I cannot provide one
here. But on the (reasonable, I think) assumption that the Agree relationship
plays a role in the theory of control and switch reference in these languages,
then the differences in the syntax of accusative experiencers are reflections of
the difference in how parameter (2) is set in two branches of this family.
There is a complication to this story: T does not in fact appear to agree with
the accusative-case experiencer in Imbabura Quechua any more than it does in
Huanca Quechua. When the experiencer is first person in a sentence like (147),
the finite verb does not agree with it in person, but rather bears the third person
suffix n.
(147) ñuka-ta puñu-naya-n. (*puñu-naya-ni). (Cole 1985:111)

I-acc sleep-desid-3S sleep-desid-1sS
‘I want to sleep; I am sleepy.’
Cole, Jake, and Hermon all very reasonably interpret this as default agreement,
the third person ending being filled in when there is no argument to agree
with. But in light of the overarching pattern of facts reviewed in this section,
I interpret it as partial agreement, comparable to the agreement that a predi-
cate adjective would have with its subject. The details could go something like
this. Since experiencer subjects have a different θ -role than agentive subjects,
they are generated in a different structural position – say Spec, ApplP rather
than Spec, vP.48 In that position, they receive inherent accusative case from the
applicative head. This inherent case freezes them in place, so that (unlike NPs
generated in Spec, vP) they are unable to move to Spec, TP. The principles of
agreement still allow T to agree with the experiencer in Imbabura, because the
48 This specific proposal is based on the similarities between experiencers and goals, not only
crosslinguistically, but also in Imbabura (see especially Jake 1985 for relevant data).
universal conditions on Agree are satisfied, agreement is not case-dependent

in Imbabura, and all Quechuan languages permit downward argument. (The
primary evidence for this is that all Quechuan languages have reasonably free
word order, and alternations between (say) SOV and OSV order have no affect
on subject or object agreement.) This agreement relation between T and the
experiencer contributes to an understanding of the control and switch reference
properties reviewed above. But the principles of agreement do not allow T to
agree with the experiencer in person in this configuration; rather, agreement
can only be in number and gender, by the SCOPA. Now it so happens that gen-
der is never marked in Quechua, and there is no distinction between singular
and plural in the third person (Cole 1985:143, Hermon 1985:22). Therefore the
partial agreement forced by the SCOPA happens to be indistinguishable on the
surface from default agreement, simply because there happens to be only one
nonfirst/nonsecond inflectional ending in the language. Nevertheless, the exis-
tence of a genuine agreement relationship can be detected indirectly by its role
in control and switch reference. Positing it also allows us to identify a unified
source for the various observable differences in the relationship between agree-
ment and case that distinguish Imbabura Quechua from its Peruvian cousins –
a testimony to the reality of (2) as a unified grammatical parameter.49
5.12 General conclusion

The first four chapters of this book developed a universal theory of agreement
that applies equally well to all syntactic categories – to adjectives, nouns, adpo-
sitions, determiners, and complementizers as well as to verbs. This quest bore
theoretical fruit in two specific ways: it led to the result that agreement can be
upward as well as downward, and to the result that person agreement needs a
more strictly local syntactic relationship than other kinds of agreement do. In
this chapter, we have seen how having a broader theory of agreement can bear
fruit along another dimension. Once we have all agreement-bearing categories
fully in view, two typological results can also become clear. On the one hand,
the agreement-bearing functional heads behave rather differently from one lan-
guage to another with respect to the precise conditions under which they agree.
On the other hand, all of the agreement-bearing heads in a given language tend
to work the same way with respect to these conditions. This balance between
49 My analysis crucially predicts that if a dialect of Imbabura Quechua comes to light that distin-
guishes third plural agreement from third singular agreement, one would find the plural form
in a sentence like ‘We are sleepy,’ not the default singular form. But no such dialect has been
documented to my knowledge.
General conclusion 245
a limited degree of inter-language diversity and high degree of intra-language

consistency is exactly what one expects given Chomsky’s classic Principles and
Parameters vision of the nature of the human language faculty. But the pattern
can only be seen clearly if one has in view both a broad range of natural lan-
guages and a theory that applies to a wide range of syntactic categories. This
work, then, has culminated in an experiment in combining a formal-generative
depth of analysis, concerned with capturing significant generalizations across
different grammatical constructions, with a functionalist-typological method-
ology of putting claims to the test against a broad sample of languages. I like
to think that the experiment has been a successful one, giving a big picture that
is more robust and convincing than could be constructed simply by pursuing
either style of inquiry by itself. But I admit that you are likely to be a better
judge of that than I am.
Appendix: Table of languages and
their agreement properties
The following table displays the properties of agreement that are most relevant for assess-
ing how the Direction of Agreement Parameter and the Case Dependence of Agreement
Parameter are set, for each of the 108 languages in my sample. The languages are listed
in the same order as they are in the description of the World Atlas of Language Struc-
tures core sample, starting in Africa and moving northward and eastward in roughly the
presumed ways that language-speakers spread across the globe by migration, ending in
South America. The first two columns identify the primary word order and case-marking
properties of full noun phrases in the language; in almost every instance, these values
are taken from WALS maps 81–3 and 98. This information is included as background
and plays no direct role in the analysis. The last eight columns record the answers that
I found to the following eight questions concerning the syntax of agreement:
a. Is subject agreement dependent on the subject being in a particular syntactic

position?
b. Is subject agreement sensitive to the case of the subject? For example, does
the verb agree only with nominative subjects, or can it show the same sort of
agreement with ergative or dative subjects as well?
c. Is object agreement dependent on the position or definiteness of the object
(two kinds of evidence that the object has moved to a higher position)?
d. Is object agreement sensitive to the case of the object? For example, does the
verb agree only with accusative/absolutive/unmarked objects, or can it show
the same sort of agreement with dative or oblique complements?
e. Is there agreement on complementizers, and if so what kind?
f. Is there agreement on adpositions with an unmoved complement?
g. Is there agreement between a determiner/quantifier and the associated NP,
and if so does it depend on the word order of the NP with respect to the
determiner?
h. Is there full person-number-gender agreement with the subject on both the
auxiliary verb and the main verb in complex tense constructions, or is there
person agreement on one verb only?
In general, the answers to (a, c, e–g) are relevant to the setting of Direction of Agreement
Parameter, and the answers to (b, d, h) are relevant to the setting of the Case Dependence
of Agreement Parameter. Note that if WALS gives no indication that a language has
agreement with subjects, objects, or objects of adpositions, and does not indicate that
246
Table of languages and their agreement properties 247
it has grammatical gender (so adjectival agreement is unlikely), then I did not look up
the language, and scored all categories as “none.” Such languages are only included to
give an overall sense of how common each type of agreement is in the languages of the
world. See chapter 5 for extensive further discussion of these properties, how they relate
to the proposed parameters, and illustrative examples. (See also below for explanation
of some of the less obvious notations used in the table.)
Word Case- AgrS AgrS AgrO AgrO Agr on C Agr on P Agr on Det Agr in
Order marking needs depends only if depends with object Aux+Verb
Language pattern order? on case mov’t on case
Khoekhoe SOV accusative No (Yes) (clitic) N/A None None? Yes N/A
Sango SVO neutral None None None None None None None None
Luvale SVO neutral Yes N/A Yes N/A None Only if Yes+ N-D Single
moved
Swahili SVO neutral Yes N/A Yes N/A None (clitic) Only if N-D Double
Zulu SVO neutral Yes N/A Yes N/A None (clitic) Yes, N-D Double
Kinande SVO neutral Yes N/A Yes N/A Up Only if Only if N-D Double
moved
Yoruba SVO neutral None None None None None None None None
Supyire SOV neutral None None None None None None None None
Grebo SVO neutral None None None None None None None None
Krongo VSO neutral None None None None None None None None
Bagirmi SVO neutral Can’t tell N/A (clitic) N/A None (clitic) None Single
Lango SVO neutral Can’t tell N/A (clitic) N/A None (clitic) Yes if N-D Single
Koyra Chiini SVO neutral? None None None None None None None None
Berber VSO accusative No* N/A (clitic) N/A None? None Only if N-D Double
Hausa SVO neutral Can’t tell N/A None N/A None None Yes Single
Harar Oromo SOV accusative None None None None None None None None
Arabic VSO- accusative- No N/A No/ N/A Down Yes/ clitic Yes Double
SVO neutral clitic
Hebrew SVO accusative No Yes (clitic) N/A None (clitic) ? Single
Basque SOV ergative No No No No Up None Yes Single
English SVO neutral No N/A None None None None (Yes) Single
German Mixed accusative No Yes None None Down None Yes Single
Greek SVO accusative No Yes (clitic) None None None Yes Single
Hindi SOV split No Yes None None None None Yes Single
ergative
Persian SOV accusative No N/A? (clitic) None None (clitic) None Double
French SVO neutral No N/A? (clitic) None None None Yes Single
Spanish SVO accusative No Yes No/Clitic N/A None None Yes Single
Russian SVO accusative No Yes None None None None Yes Single
Finnish SVO accusative No Yes None None None Yes Yes Single
Khalkha SOV accusative None None None None None None None None
Turkish SOV accusative Yes Yes None None None None None Single
Chukchi Free ergative No(2) No(2) (none) (none) None None No D? Single
Japanese SOV accusative None None None None None None None None
Korean SOV accusative None None None None None None None None
Abkhaz/ SOV neutral No (Yes) No (Yes) None Yes None No aux
Abaza
Lezgian SOV ergative None None None None None None None None
Georgian SOV split No No No No None Some? Yes Double
ergative
Burushaski SOV ergative No No No No No C Some Yes Double
Kannada SOV accusative No Yes None None None None None Single
Mandarin SVO neutral None None None None None None None None
Burmese SOV accusative None None None None None None None None
Meithei SOV accusative None None None None None None None None
Hmong SVO neutral None None None None None None None None
(cont.)
Cont.
Thai SVO neutral None None None None None None None None
Vietnamese SVO neutral None None None None None None None None
Paiwan V-first neutral None None None None None None None None
Fijian V-first neutral Yes N/A Yes N/A None Only if None Double
moved
Rapanui VSO neutral None None None None None None None None
Malagasy VOS accusative None None None None None None None None
Chamorro VSO neutral Yes N/A None None (Up) None None Single
Tagalog V-first neutral None None None None None None None None
Tukang Besi VOS ergative Yes No Yes N/A None None None Double
Indonesian SVO neutral None None None None None None None None
Tauya SOV neutral No No No N/A None Yes None Single
Imonda V-last active None None None None None None None None
Arapesh SVO neutral Yes N/A Yes N/A Up None Yes if N-D? Double
Alamblak SOV neutral No (Yes) No N/A None None Yes Single
Yimas free neutral No (Yes) No Yes No C None Some No aux
Asmat SOV neutral No (Yes) No N/A None None None Single
Dani SOV ergative No No No N/A None Some None Single
Kewa SOV ergative No No None None None None None Single
Amele SOV neutral No (Yes) No N/A None None Yes Single
Daga SOV neutral No Yes ? Yes Down Some None Single
Lavukaleve SOV neutral Can’t tell N/A No Yes No C? Yes Yes Single
Gooniyandi V-last ergative No No No Yes No C None None No aux
Mayali Free (weak No No No N/A None None Yes No aux
ergative)
Maung SVO neutral No (Yes) No Yes None None Yes Single
Mangarrayi OV accusative No Yes No Yes None None Yes Single
Martuthu- SVO accusative None None None None None None None None
nira
Ngiyambaa V-last ergative (clitic) N/A (clitic) N/A No C No P No D None
Warlpiri Free ergative No No No No No C None No D Single
Kayardild Free accusative None None None None None None None None
Tiwi SVO neutral No Yes No N/A None None Yes Single
Greenlandic SOV ergative Yes(2) Yes (none) (none) No C None Yes if N-D Single
Slave SOV neutral Yes? N/A Yes N/A None Only if Yes if N-D Double
moved
Ojibwa SVO- neutral No N/A No No- None No P Yes Double
free LDA
Mohawk Free neutral Yes N/A Yes N/A None No P None No aux
Choctaw- SOV accusative No Yes Can’t Yes None Yes None Single
Chicasaw tell
Wichita Free neutral No N/A No N/A None No P None No aux
Lakhota SOV neutral No (Yes) No N/A None Yes Yes Single
Kiowa SOV neutral Can’t tell N/A No N/A None None Yes No aux
Acoma Free neutral Can’t tell N/A Can’t N/A None None None Single
tell
Yaqui SOV accusative None None None None None None None None
Nahuatl VO neutral No N/A No N/A None Yes Yes No aux
(cont.)
Cont.
Makah VSO neutral No N/A No Yes None No P None Single

Nez Perce Free tripartite No No Yes* Yes* Down None None No aux/
double
Halkomelem VSO neutral No N/A Can’t No- None None Yes Double
tell LDA
Maricopa SOV accusative Can’t tell No Can’t N/A None None None (yes) Double
tell
Karok Free neutral No No No N/A None None None No aux
Mixtec VSO neutral (clitic) None None None None (clitic) None No aux
Otomi VOS neutral No N/A No No? None None Yes Double
Zoque VOS ergative No No No N/A None None None Double
Tzotzil VOS neutral No (Yes) No Yes? None Yes Yes Single
Ika SOV ergative No No Can’t No No C None None Double
tell
Barasano OV accusative No Yes None None None None Yes? Single
Warao OSV neutral No Yes None None None Yes or Yes Single
clitic
Sanuma SOV ergative None None None None None None None None
Yagua VSO neutral No N/A (clitic) N/A None Yes Yes Single
I. Quechua SOV accusative No No Can’t No No C None None Some
tell double
Apurinã varies neutral Yes N/A Yes Yes None None Yes in D-N Single
Hixkaryana OVS neutral No? N/A No Yes No C Yes or None Single
clitic
Guaranı́ SVO accusative No N/A Can’t N/A None (clitic) Yes Double
tell
Canela- SOV neutral Yes No Yes No- None (clitic) None Double
Krahô LDA
Wari’ VOS neutral No Yes No Yes Up? Yes Yes No aux
Pirahã SOV neutral Can’t tell N/A No N/A None Yes None No aux
Wichı́ SVO neutral No (Yes) No N/A None Some None? Single
Mapudungun VO- neutral Yes Yes Yes? Yes No C None None No aux
free
Tariana SOV accusative None* None Yes* N/A None None Only if N-D Double
Jarawara SOV neutral Yes No None None None None Some if Double
N-D
Clarificatory notes:
r “None” means the relevant kind of agreement is not present in the language.
r N/A means that the agreement is present, but cannot be evaluated for the parameter for some other reason (e.g. no oblique case marking).
r “Can’t tell” means that the agreement is present, but I could not find evidence as to whether it is sensitive to the position of the agreed-with
NP (e.g. subject always in Spec, TP; subject or object agreement is null).
r (Yes) means there is no overt case, but case sensitivity could insure that T agrees only with thematic subject despite the subject not being
in a fixed position.
r N/Y(erg) means that there is agreement with a weak/limited/optional ergative as well as with an unmarked NP.
r (clitic) means that a bound pronominal form can attach to the relevant head. This was counted as equivalent to “none,” but conceivably
could be reanalyzed as being agreement given more data.
r (2) indicates a kind of ergative language analyzed in the literature as having two instances of agreement on T and none on v. Alternative
analyses of this are conceivable.
References
Abney, Steven. 1987. The English noun phrase in its sentential aspect. MIT, Cambridge,
Mass.: Doctoral dissertation.
Adesola, Oluseye. 2005. Pronouns and null operators: A-bar dependencies and relations
in Yoruba. Rutgers University: Ph.D. dissertation.
Adger, David, and Daniel Harbour. 2005. Syntax and syncretisms of the Person Case
Constraint. In Minimalist Approaches to Clause Structure (MITWPL #50), ed. Ken
Hiraiwa and Joey Sabbagh, 1–36. Cambridge, Mass.: MIT Linguistics Department.
Aikhenvald, Alexandra. 2003. A grammar of Tariana. Cambridge: Cambridge University
Press.
Aissen, Judith. 1987. Tzotzil clause structure. Dordrecht: Reidel.
Allen, Barbara, Donald Frantz, Donna Gardiner, and David Perlmutter. 1990. Verb
agreement, possessor ascension, and multistratal representation in Southern Tiwa.
In Studies in Relational Grammar 3, ed. Paul Postal and Brian Joseph, 321–84.
Chicago: University of Chicago Press.
Anagnostopoulou, Elena. 2003. The syntax of ditranstives: evidence from clitics. Berlin:
Mouton de Gruyter.
Anand, Pranav, and Andrew Nevins. 2004. Shifty operators in changing contexts. Paper
presented at SALT 14, Northwestern University.
Aoki, Haruo. 1973. Nez Perce grammar. Berkeley: University of California Press.
Aoun, Joseph, and Lina Choueiri. 2000. Epithets. Natural Language and Linguistic
Theory 18:1–39.
Ashton, E. O. 1949. Swahili grammar. New York: Longmans.
Baker, Mark. 1985. The Mirror Principle and morphosyntactic explanation. Linguistic
Inquiry 16:373–415.
Baker, Mark. 1988. Incorporation: a theory of grammatical function changing. Chicago:
University of Chicago Press.
Baker, Mark. 1991. Clitic climbing and the boundedness of head movement. In Clitics
in the languages of Europe, ed. Hank van Riemsdijk, 369–73. Berlin: Mouton de
Gruyter.
Baker, Mark. 1996. The polysynthesis parameter. New York: Oxford University Press.
Baker, Mark. 1997. Thematic roles and syntactic structure. In Elements of grammar, ed.
Liliane Haegeman, 73–137. Dordrecht: Kluwer.
Baker, Mark. 2002. Building and merging, not checking: the nonexistence of (Aux) SVO
languages. Linguistic Inquiry 33:321–8.
254
References 255
Baker, Mark. 2003a. Lexical categories: verbs, nouns, and adjectives. Cambridge:
Cambridge University Press.
Baker, Mark. 2003b. On the loci of agreement: inversion constructions in Mapudungun.
In Proceedings of NELS 33, ed. Makoto Kadowaki and Snigeto Kawahare, GLSA,
University of Massachusetts-Amherst.
Baker, Mark. 2003c. Agreement, dislocation, and partial configurationality. In Formal
approaches to function in grammar, ed. Andrew Carnie, Heidi Harley, and Mary
Ann Willie, 107–34. Amsterdam: John Benjamins.
Baker, Mark. 2005. On verb-initial and verb-final orders in Lokaa. Journal of African
Languages and Linguistics 26:125–64.
Baker, Mark, and Chris Collins. 2006. Linkers and the internal structure of vP. Natural
Language and Linguistic Theory 24:307–54.
Baker, Mark, and Natalia Kariaeva. 2006. On the relationship of case valuation and
phi-feature agreement. Ms., Rutgers University, New Brunswick, N.J.
Barss, Andrew, and Howard Lasnik. 1986. A note on anaphora and double objects.
Linguistic Inquiry 17:347–54.
Bejar, Susana. 2003. Phi-syntax: a theory of agreement. University of Toronto: Ph.D.
dissertation.
Belletti, Adriana, and Luigi Rizzi. 1981. The syntax of ne: some theoretical implications.
The Linguistic Review 1:117–54.
Benmamoun, Elabbas. 2000. The feature structure of functional categories: a compar-
ative study of Arabic dialects. New York: Oxford University Press.
Bhatt, Rajesh. 2005. Long distance agreement in Hindi-Urdu. Natural Language and
Linguistic Theory 23:757–807.
Bickel, Balthasar, and Yogendra Yadava. 2000. A fresh look at grammatical relations in
Indo-Aryan. Lingua 110:343–73.
Bittner, Maria. 1994. Case, scope, and binding. Dordrecht: Kluwer.
Bittner, Maria, and Kenneth Hale. 1996. Ergativity: toward a theory of a heterogeneous
class. Linguistic Inquiry 27:531–604.
Bobaljik, Jonathan. 1993. Nominally absolutive is not absolutely nominative. In Pro-
ceedings of the Eleventh West Coast Conference on Formal Linguistics, ed. Jonathan
Mead, 44–60. Stanford: CSLI.
Bobaljik, Jonathan. 2002. A-Chains at the PF-interface: copies and “covert” movement.
Natural Language and Linguistic Theory 20:197–267.
Bobaljik, Jonathan. to appear. Where’s Phi? Agreement as a post-syntactic operation. In
Phi theory: Phi features across interfaces and modules, ed. David Adger, Daniel
Harbour, and Susanna Béjar. Oxford: Oxford University Press.
Bobaljik, Jonathan, and Phil Branigan. 2006. Eccentric agreement and multiple case
checking. In Ergativity: Emerging Issues, eds. Alana Johns, Diane Massam and
Juvenal Ndayiragije, 47–77. Dordrecht: Springer.
Bobaljik, Jonathan, and Susi Wurmbrand. 2005. The domain of agreement. Natural
Boeckx, Cedric. 2000. Quirky agreement. Studia Linguistica 54:354–80.
Boeckx, Cedric. 2004. Long-distance agreement in Hindi: some theoretical implications.
Studia Linguistica 58:23–36.
256 References
Boeckx, Cedric, and Fumikazu Niinuma. 2004. Conditions on agreement in Japanese.

Bok-Bennema, Reineke. 1991. Case and agreement in Inuit. Berlin: Foris.
Bonet, Eulàlia. 1991. Morphology after syntax: pronominal clitics in Romance lan-
guages. MIT: Ph.D. dissertation.
Borer, Hagit. 1984. Parametric syntax: case studies in Semitic and Romance languages.
Dordrecht: Foris.
Borer, Hagit. 1991. The causative-inchoative alternation: a case study in parallel mor-
phology. The Linguistic Review 8:119–58.
Borer, Hagit. 1995. The ups and downs of Hebrew verb movement. Natural Language
and Linguistic Theory 13:527–606.
Branigan, Phil, and Marguerite MacKenzie. 2002. Altruism, A-bar movement, and object
agreement in Innu-aimun. Linguistic Inquiry 33:385–408.
Bresnan, Joan. 1994. Locative inversion and the architecture of Universal Grammar.
Language 70:72–131.
Bresnan, Joan, and Joni Kanerva. 1989. Locative inversion in Chichewa: a case study
of factorization in grammar. Linguistic Inquiry 20:1–50.
Bresnan, Joan, and Sam Mchombo. 1987. Topic, pronoun, and agreement in Chichewa.
Broekhuis, Hans. 1992. Chain-government: issues in Dutch syntax. University of
Amsterdam: Ph.D. dissertation.
Bruening, Benjamin. 2001. Syntax at the edge: cross-clausal phenomena and the syntax
of Passamaquoddy. MIT: Ph.D. dissertation.
Burzio, Luigi. 1986. Italian syntax: a government-binding approach. Dordrecht:
Reidel.
Campana, Mark. 1992. A movement theory of ergativity. McGill University: Ph.D.
dissertation.
Carnie, Andrew, and Phil Cash Cash. 2003. Nez Perce case: a note on Woolford 1997.
Ms., University of Arizona.
Carstens, Vicki. 2001. Multiple agreement and case deletion: against phi-(in)complete-
ness. Syntax 4:147–63.
Carstens, Vicki. 2003. Rethinking complementizer agreement: agree with a case-checked
goal. Linguistic Inquiry 34:393–412.
Carstens, Vicki. 2005. Agree and EPP in Bantu. Natural Language and Linguistic Theory
23:219–79.
Cherchi, Marcello. 1999. Georgian. Munich: LINCOM Europa.
Choe, Hyon-Sook. 1986. An SVO analysis of VSO languages and parameterization: a
study of Berber. Ms., MIT.
Chomsky, Noam. 1973. Conditions on transformations: a festschrift for Morris Halle.
New York: Holt, Rinehart and Winston.
Chomsky, Noam. 1981. Lectures on government and binding. Dordrecht: Foris.
Chomsky, Noam. 1986. Barriers. Cambridge, Mass.: MIT Press.
Chomsky, Noam. 1991. Some notes on economy of derivation and representation:
principles and parameters in comparative grammar. Cambridge, Mass.: MIT
Press.
References 257
Chomsky, Noam. 1993. A minimalist program for linguistic theory. In The view from
Building 20, ed. Kenneth Hale and Samuel Jay Keyser, 1–52. Cambridge, Mass.:
MIT Press.
Chomsky, Noam. 1995. The minimalist program. Cambridge, Mass.: MIT Press.
Chomsky, Noam. 2000. Minimalist inquiries: the framework. In Step by Step, ed. Roger
Martin, David Michaels, and Juan Uriagereka, 89–155. Cambridge, Mass.: MIT
Press.
Chomsky, Noam. 2001. Derivation by phase. In Ken Hale: a life in language, ed. Michael
Kenstowicz, 1–52. Cambridge, Mass.: MIT Press.
Cinque, Guglielmo. 1990. Types of A-bar dependencies. Cambridge, Mass.: MIT Press.
Cole, Peter. 1985. Imbabura Quechua. London: Croom Helm.
Collins, Chris. 2004. The agreement parameter. In Triggers, ed. Anne Breitbarth and
Hank van Riemsdijk, 115–36. Berlin: Mouton de Gruyter.
Cowell, Mark. 1964. A reference grammar of Syrian Arabic. Washington, D.C.: George-
town University Press.
Craig, Colette Grinevald. 1977. The structure of Jacaltec. Austin: University of Texas
Press.
Croft, William. 1991. Syntactic categories and grammatical relations. Chicago: Uni-
versity of Chicago Press.
Davies, William. 1986. Choctaw verb agreement and universal grammar. Dordrecht:
Reidel.
den Dikken, Marcel, and Pornsiri Singhapreecha. 2004. Complex noun phrases and
linkers. Syntax 7:1–54.
Diercks, Michael. 2006. Object agreement in Swahili and the interpretation of objects.
Ms. Georgetown University, Washington, D.C.
Diesing, Molly. 1990. Verb movement and the subject position in Yiddish. Natural
Diesing, Molly. 1992. Indefinites. Cambridge, Mass.: MIT Press.
Dixon, R. M. W. 1982. Where have all the adjectives gone? Berlin: de Gruyter.
Dixon, R. M. W. 1988. A grammar of Boumaa Fijian. Chicago: University of Chicago
Press.
Dixon, R. M. W. 2004. The Jarawara language of Southern Amazonia. Oxford: Oxford
University Press.
Doke, Clement. 1955. Zulu syntax and idiom. Cape Town: Longmans Southern Africa.
Doke, Clement. 1963. Textbook of Zulu grammar. London: Longmans.
Dryer, Matthew. 1992. The Greenbergian word order correlations. Language 68:81–
138.
Duranti, Alessandro. 1979. Object clitic pronouns in Bantu and the topicality hierarchy.
Studies in African Linguistics 10:31–45.
Etxepare, Ricardo. 2006. Number long distance agreement in (substandard) Basque.
Ms., CNRS-IKER.
Evans, Nicholas. 2003. Bininj Gun-wok: a pan-dialectal grammar of Mayali, Kunwinjku
and Kune. Canberra: Pacific Linguistics.
Facundes, Sidney da Silva. 2000. The language of the Apurinã people of Brazil
(Maipure/Arawak). State University of New York: Ph.D. dissertation.
258 References
Foley, William. 1991. The Yimas language of New Guinea. Stanford, CA: Stanford
University Press.
Fortescue, Michael. 1984. West Greenlandic. London: Croom Helm.
Fortune, Reo. 1977. Arapesh. New York: AMS Press.
Franks, Steven. 1995. Parameters of Slavic morphosyntax. New York: Oxford University
Press.
Frantz, Donald. 1993. Empty arguments in Southern Tiwa. Ms., University of
Lethbridge.
Fukui, Naoki, and Margaret Speas. 1986. Specifiers and projections. MIT Working Papers
in Linguistics 8:128–72.
Geach, Peter. 1962. Reference and generality. Ithaca, N. Y.: Cornell University Press.
Giorgi, Alessandra, and Giuseppe Longobardi. 1991. The syntax of noun phrases: con-
figuration, parameters and empty categories. Cambridge: Cambridge University
Press.
Givón, Talmy. 1976. Topic, pronoun, and grammatical agreement. In Subject and topic,
ed. Charles Li, 149–89. New York: Academic Press.
Greenberg, Joseph. 1963. Universals of language. Cambridge, Mass.: MIT Press.
Grimshaw, Jane. 1990. Argument structure. Cambridge, Mass.: MIT Press.
Grimshaw, Jane. 1991. Extended projection. Ms., Brandeis University.
Grune, Dick. 1998. Burushaski: an extraordinary language in the Karakoram mountains.
Ms., Vrije Universiteit.
Guerssel, Mohamed. 1983. Some notes on the structure of Berber. Ms., MIT.
Gupta, Anil. 1980. The logic of common nouns. New Haven, Conn.: Yale University
Press.
Halle, Morris, and Alec Marantz. 1993. Distributed morphology and the pieces of inflec-
tion. In The view from building 20, ed. Kenneth Hale and S. J. Keyser, 111–76.
Cambridge, Mass.: MIT Press.
Harley, Heidi, and Elizabeth Ritter. 2002. A feature-geometric analysis of person and
number. Language 78:482–526.
Harlow, Stephen. 1981. Government and relativization in Celtic. In Binding and filtering,
ed. Frank Heny, 213–54. Cambridge, Mass.: MIT Press.
Haspelmath, Martin. 2004. Explaining the ditransitive person-role constraint: a usage-
based approach. Ms., Max-Planck-Institut für evolutionäre Anthropologie.
Haspelmath, Martin, Matthew Dryer, David Gil, and Bernard Comrie, eds. 2005. The
world atlas of language structures. Oxford: Oxford University Press.
Heath, Jeffrey. 1984. Functional grammar of Nunggubuyu. Canberra: Australian Insti-
tute of Aboriginal Studies.
Hermon, Gabriella. 1985. Syntactic modularity. Dordrecht: Foris.
Hermon, Gabriella. 2001. Non-canonically marked A/S in Imbabura Quechua. In Non-
canonical marking of subjects and objects, ed. Alexandra Aikhenvald, R. M. W.
Dixon, and Masayuki Onishi, 149–76. Amsterdam: John Benjamins.
Hewitt, B. G. 1995. Georgian: a structural reference grammar. Amsterdam: John
Benjamins.
Horton, A. E. 1949. A grammar of Luvale. Johannesburg: Witwatersrand University
Press.
References 259
Hyman, Larry, and Alessandro Duranti. 1982. On the object relation in Bantu. In Syntax
and Semantics 15, 217–39. New York: Academic Press.
Iwara, Alexander. 1982. Phonology and Grammar of Lokaa, University of London:
Masters thesis.
Jackendoff, Ray. 1977. X-bar syntax. Cambridge, Mass.: MIT Press.
Jake, Janice. 1985. Grammatical relations in Imbabura Quechua. New York: Garland.
Julien, Marit. 2002. Syntactic heads and word formation. New York: Oxford University
Press.
Kaplan, David. 1989. Demonstratives. In Themes from Kaplan, ed. J. Almog, H.
Wettstein, and J. Perry, 563–81. New York: Oxford University Press.
Kayne, Richard. 1984. Connectedness and binary branching. Dordrecht: Foris.
Kayne, Richard. 1989. Facets of Romance past participle agreement. In Dialect variation
and the theory of grammar, ed. Paula Benincà, 85–105. Dordrecht: Foris.
Kayne, Richard. 2000. Parameters and universals. New York: Oxford University Press.
Kayne, Richard. 2005. Some notes on comparative syntax, with special reference to
English and French. In The Oxford handbook of comparative syntax, ed. Guglielmo
Cinque and Richard Kayne, 3–69. New York: Oxford University Press.
Kimball, J., and Judith Aissen. 1971. I think, you think, he think. Linguistic Inquiry
2:242–6.
Kimenyi, Alexandre. 1980. A relational grammar of Kinyarwanda. Berkeley, Calif.:
University of California Press.
Kinyalolo, Kasangati. 1991. Syntactic dependencies and the SPEC-head agreement
hypothesis in KiLega. UCLA: Ph.D. dissertation.
Koopman, Hilda. 1984. The syntax of verbs. Dordrecht: Foris.
Koopman, Hilda. 2006. Agreement configurations: in defense of “Spec head”. In Agree-
ment systems, ed. Cedric Boeckx, 159–200. Amsterdam: John Benjamins.
Koopman, Hilda, and Dominique Sportiche. 1989. Pronouns, logical variables, and
logophoricity in Abe. Linguistic Inquiry 20:555–89.
Koopman, Hilda, and Dominique Sportiche. 1991. The position of subjects. Lingua
85:211–58.
Kornfilt, Jaklin. 1997. Turkish. New York: Routledge.
Kratzer, Angelika. 1998. More structural analogies between pronouns and tenses. In
Proceedings of SALT VIII, ed. Devon Strolovitch and Aaron Lawson, 92–110.
Ithaca, N.Y.: Cornell Linguistics Club.
Larson, Richard. 1988. On the double object construction. Linguistic Inquiry 19:335–92.
Launey, Michel. 1981. Introduction à la langue et à la littérature aztèques. vol. 1. Paris:
L’Harmattan.
Levin, Juliette, and Diane Massam. 1984. Surface ergativity: case/theta relations reexam-
ined. In Proceedings of the Fifteenth Annual Meeting of the North-East Linguistics
Society, 286–301. Amherst, Mass.: GLSA.
Li, Yafei. 1990. Xy-binding and verb incorporation. Linguistic Inquiry 21:399–426.
Lorimer, D. L. R. 1935. The Burushaski language 1: introduction and grammar. vol. 1.
Cambridge, Mass.: Harvard University Press.
Machobane, Malillo. 1993. The Sesotho locative constructions. Ms., National University
of Lesotho.
260 References
Magner, Thomas. 1991. Introduction to the Croatian and Serbian language. University
Park, Penn.: Pennsylvania State University Press.
Marantz, Alec. 1991. Case and licensing. Paper presented at The 8th Eastern States
Conference on Linguistics, University of Maryland, Baltimore.
Marantz, Alec. 1993. Implications of asymmetries in double object constructions. In
Aspects of Bantu Grammar 1, ed. Sam Mchombo, 113–50. Stanford, Calif.: CSLI.
Marantz, Alec. 2000. Reconstructing the lexical domain with a single generative engine.
Ms., MIT.
Markman, Vita. 2005. The syntax of case and agreement: its relationship to morphology
and argument structure. Rutgers University: Ph.D. dissertation.
Massam, Diane. 2001. Pseudo noun incorporation in Niuean. Natural Language and
Linguistic Theory 19:153–97.
McCloskey, James. 1996. On the scope of verb movement in Irish. Natural Language
and Linguistic Theory 14:47–104.
Mchombo, Sam. 2004. The syntax of Chichewa. Cambridge: Cambridge University
Press.
Mohanan, Tara. 1995. Woodhood and lexicality: noun incorporation in Hindi. Natural
Müller, Gereon, and Wolfgang Sternefeld. 1993. Improper movement and unambiguous
binding. Linguistic Inquiry 24:461–507.
Munro, Pamela, and Lynn Gordon. 1982. Syntactic relations in Western Muskogean.
Murane, Elizabeth. 1974. Daga grammar: from morpheme to discourse. Norman, Okla.:
Summer Institute of Linguistics.
Murasugi, K. 1992. NP-movement and the ergative parameter. MIT: Ph.D. dissertation.
Ndayiragije, Juvénal. 1999. Checking economy. Linguistic Inquiry 30:399–444.
Nichols, Johanna. 1986. Head-marking and dependent-marking grammar. Language
62:56–119.
Noonan, Michael. 1992. A grammar of Lango. Berlin: Mouton de Gruyter.
O’Herin, Brian. 2002. Case and agreement in Abaza. Dallas, Tex.: Summer Institute of
Linguistics.
Ormazabal, Javier, and Juan Romero. 2006. Object clitics and agreement. Ms., University
of the Basque Country and University of Alcalá.
Ormazabal, Javier, and Juan Romero. 2007. The object agreement constraint. Natural
Ortiz de Urbina, Jon. 1989. Parameters in the grammar of Basque: a GB approach to
Basque syntax. Dordrecht: Foris.
Ouali, Hamid, and Acrisio Pires. to appear. Complex tenses, agreement, and wh-
extraction. In Proceedings of the Berkeley Linguistics Society 31. Berkeley, Calif.:
Berkeley Linguistics Society.
Ouhalla, Jamal. 2005a. Clitic placement, grammaticalization, and reanalysis in Berber.
In The Oxford handbook of comparative syntax, ed. Guglielmo Cinque and Richard
Kayne, 607–38. New York: Oxford University Press.
Ouhalla, Jamal. 2005b. Agreement features, agreement and antiagreement. Natural Lan-
guage and Linguistic Theory 23:655–86.
References 261
Pesetsky, David, and Esther Torrego. 2001. T-to-C movement: causes and consequences.
In Ken Hale: a life in language, ed. Michael Kenstowicz, 355–426. Cambridge,
Mass.: MIT Press.
Pica, Pierre. 1991. On the interaction between antecedent-government and binding: the
case of long-distance reflexivization. In Long-distance anaphora, ed. Jan Koster
and Eric Reuland, 119–35. Cambridge: Cambridge University Press.
Polinsky, Maria, and Eric Potsdam. 2001. Long distance agreement and topic in Tsez.
Pollard, Carl, and Ivan Sag. 1994. Head-driven phrase structure grammar. Chicago:
University of Chicago Press.
Popjes, Jack, and Jo Popjes. 1986. Canela-Kraho. In Handbook of Amazonian languages,
ed. Desmond Derbyshire and Geoffery Pullum, 128–99. Berlin: Mouton de Gruyter.
Reinhart, Tanya. 1983. Anaphora and semantic interpretation. Chicago: University of
Chicago Press.
Reinhart, Tanya, and Eric Reuland. 1991. Anaphors and logophors: an argument struc-
ture perspective. In Long-distance anaphora, ed. Jan Koster and Eric Reuland.
Cambridge: Cambridge University Press.
Rezac, Milan. 2003. The fine structure of cyclic Agree. Syntax 6:156–82.
Rice, Keren. 1989. A grammar of Slave. Berlin: Mouton de Gruyter.
Richards, Norvin. 1998. The principle of minimal compliance. Linguistic Inquiry
29:599–629.
Riemsdijk, Hank van. 1978. A case study in syntactic markedness. Dordrecht: Foris.
Ritter, Elizabeth. 1991. Two functional categories in noun phrases: evidence from Mod-
ern Hebrew. In Syntax and Semantics 26, ed. Susan Rothstein, 37–62. San Diego,
Calif.: Academic Press.
Rizzi, Luigi. 1990. Relativized minimality. Cambridge, Mass.: MIT Press.
Roberts, John. 1987. Amele. London: Croom Helm.
Robins, R. H. 1989. A short history of linguistics. London: Longman.
Rögnvaldsson, E., and H. Thráinsson. 1990. On Icelandic word order once more. In
Modern Icelandic syntax, ed. Joan Maling and Annie Zaenen, 3–40. San Diego:
Academic Press.
Rood, David. 1976. Wichita grammar. New York: Garland.
Rood, David, and Allan R. Taylor. 1996. Sketch of Lakhota, a Siouan language. Hand-
book of North American languages 17 (Languages), 440–82.
Rude, Noel. 1982. Promotion and topicality of Nez Perce objects. In Proceedings of the
Eighth Annual Meeting of the Berkeley Linguistics Society, ed. Monica Macaulay,
463–83. Berkeley: University of California.
Rude, Noel. 1986a. Topicality, transitivity, and the direct object in Nez Perce. Interna-
tional Journal of American Linguistics 52:124–53.
Rude, Noel. 1986b. Discourse-pragmatic context for genitive promotion in Nez Perce.
Studies in Language 10:109–36.
Rude, Noel. 1988. Ergative, passive, and antipassive in Nez Perce. In Passive and Voice,
ed. M. Shibatani, 547–60. Amsterdam: John Benjamins.
Sadiqi, Fatima. 1997. Grammaire du berbère. Paris: Editions L’Harmattan.
Safir, Ken. 2004. The syntax of anaphora. New York: Oxford University Press.
262 References
Safir, Ken. 2005. Person, context, and perspective. Rivista di Linguistica 16:107–53.
Salaburu Etxeberria, Pello. 1981. Ikaslearen Esku-gramatika (Basque grammar).
Bilbao: Mensajero.
Saltarelli, Mario. 1988. Basque. London: Croom Helm.
Schlenker, Philippe. 2003. A plea for monsters. Linguistics and Philosophy 26:29–
120.
Schlenker, Philippe. 2005. Person and binding. Rivista di Linguistica 16:155–218.
Schütze, Carson. 1997. Infl in child and adult language: agreement, case and licensing.
MIT: Ph.D. dissertation.
Shlonsky, Ur. 1997. Clause structure and word order in Hebrew and Arabic. New York:
Oxford University Press.
Sigurð sson, Halldór. 2000. The locus of case and agreement. Working Papers in Scan-
dinavian Syntax 67:103–51.
Sigurð sson, Halldór. 2002. To be an oblique subject: Russian vs. Icelandic. Natural
Sigurð sson, Halldór. 2005. The syntax of person, tense, and speech features. Rivista di
Linguistica 16:219–51.
Simpson, Jane. 1991. Warlpiri morpho-syntax: a lexicalist approach. Dordrecht: Kluwer.
Smeets, Ineke. 1989. A Mapuche grammar. University of Leiden: Ph.D. dissertation.
Sridhar, S. N. 1979. Dative subjects and the notion of subject. Lingua 49:99–125.
Sridhar, S. N. 1990. Kannada. London: Routledge.
Stassen, Leon. 1997. Intransitive predication. Oxford: Oxford University Press.
Stechow, Armin von. 2003. Feature deletion under binding. In Proceedings of the North
East Linguistic Society 33, ed. Makoto Kadowaki and Shigeto Kawahara, 379–403.
Amherst, Mass.: GSLA.
Stowell, Timothy. 1981. Origins of phrase structure. MIT: Ph.D. dissertation.
Taraldsen, Knut. 1995. On agreement and nominative objects in Icelandic. In Studies
in comparative Germanic syntax, ed. Hubert Haider, S. Olsen, and Sten Vikner,
307–27. Dordrecht: Kluwer.
Thráinsson, Höskuldur. 1979. On complementation in Icelandic. New York: Garland.
Travis, Lisa. 1984. Parameters and effects of word order variation. MIT: Ph.D.
dissertation.
Ura, Hiroyuki. 1996. Multiple feature-checking: a theory of grammatical function split-
ting. MIT: Ph.D. dissertation.
Van Geenhoven, Veerle. 1998. Semantic incorporation and indefinite descriptions:
semantic and syntactic aspects of noun incorporation in West Greenlandic. Stan-
ford, Calif.: CSLI Publications.
Vitale, Anthony. 1981. Swahili syntax. Dordrecht: Foris.
Wal, Jenneke van der. 2006. How object-like is the postverbal subject? Paper presented
at The Bantu-Romance Connection, Leeds.
Watanabe, Akira. 1996. Case absorption and WH-agreement. Dordrecht: Kluwer.
Webelhuth, Gert. 1992. Principles and parameters of syntactic saturation: Oxford stud-
ies in comparative syntax. New York: Oxford University Press.
Weber, David. 1989. A grammar of Huallaga (Huanuco) Quechua. Berkeley: University
of California Press.
References 263
Wechsler, Stephen, and Larisa Zlatić. 2003. The many faces of agreement. Stanford,
Calif.: CLSI.
Williams, Edwin. 1989. The anaphoric nature of θ -roles. Linguistic Inquiry 20:425–56.
Willson, Stephen. 1996. Verb agreement and case marking in Burushaski. Work Papers
of the Summer Institute of Linguistics North Dakota 40:1–71.
Woodward, Herbert. 1926. An outline of Makua grammar. Bantu Studies 2:269–325.
Woolford, Ellen. 1997. Four-way case systems: ergative, nominative, objective, and
accusative. Natural Language and Linguistic Theory 15:181–227.
Woolford, Ellen. 1999. More on the anaphor agreement effect. Linguistic Inquiry
30:257–88.
Wurmbrand, Susanne. 2003. Infinitives: restructuring and clause structure. Berlin:
Mouton de Gruyter.
Zaenen, Annie, Joan Maling, and Höskuldur Thráinsson. 1985. Case and grammatical
functions: the Icelandic passive. Natural Language and Linguistic Theory 3:441–
83.
Index
Abaza-Abkhaz agreement
agreement on P in 113, 194 active patterns in 54
agreement parameters in 166, 201 with anaphors 150–1, 152
ergative agreement in 81–2, 83, 203, 224 in case 175–8
active agreement systems 54 crosslinguistic variation in 7, 153–5
activity default 72, 159
blocking agreement with oblique subjects in definiteness 17
89 on different lexical categories 1–3, 13–27,
as condition on agreement 40, 43–4, 71, 76, 48–56
79, 80–2 ergative-absolutive patterns in 79, 80–1, 82,
subsumed by Case Dependency of 83, 101
Agreement Parameter 155, 173–4 independent of movement 160
Adesola, Oluseye 125, 129, 135–7 languages without 39, 153–4
Adger, David 102 long distance 103–7, 151, 214
adjectives morphological aspects of 7–9, 154–5
agreeing but not in person 2, 14–26, 27, multiple, with single NP 223
38–9, 85–6, 109, 144, 147–8 not used to define lexical categories 5
attributive versus predicative 17, 60–4, partial 104, 109–10, 243–4
218–19 as property of functional heads 34–9
compared to determiners 184–6 between operators and variables 121–4
compared to verbs 6, 65–7, 107–8 semantic aspects of 10–11, 108
definition of 28, 33–4 theoretical principles of 6, 40, 45, 46–7, 48,
ergative and raising type 67–74, 89, 145, 65–6, 149–50
171 Aikhenvald, Alexandra 16, 23, 83, 84
functional heads associated with 35–6, 38 Aissen, Judith 75–7, 82, 86
not case assigners 174 Allen, Barbara 96
partial agreement with anaphors 150–1 Amele 165–6, 213
person agreement on 56–8, 86 Amharic 125, 129, 134, 138
and principles of agreement 44–5, 48, 56, Anagnostopoulou, Elena 88–9, 94, 95, 151
171–5, 177 Anand, Pranav 125, 129, 133–4
structure of, as predicates 36–8 anaphora
transformed into verb 25, 26, 27–34, 36 locality conditions on 139, 148
adjunction only nouns involved in 31, 32
and agreement 147–8, 185–6, 193 special agreement with anaphors 150–2
and c-command 51, 144, 218 as test for subjecthood 91
as intermediate landing site 193 Aoun, Joseph 126
adpositions, see prepositions, prepositional applicatives 43, 94, 102, 216–17, 243
phrases Apurinã 167–8, 190, 202–3, 205–6, 213
264
Index 265
Arabic Borer, Hagit 35, 156, 227

agreement in definiteness in 17 Borer-Chomsky Conjecture 156, 219–24
gender agreement in 8 Branigan, Phil 83, 101, 107, 235
noun versus adjective agreement in 15, Bresnan, Joan
18–19, 50 on locative inversion 158, 159, 160
parameter settings in 165, 195, 212, 214, on object markers 124, 196
236 Broekhuis, Hans 42, 76
verb versus adjective agreement in 22–3 Bruening, Benjamin 77, 87, 105–6, 107, 151
Arapesh 179, 182, 190, 212 Burushaski
Ashton, E.O. 38 agreement only downward in 215–19
attributive modifiers, see modifiers agreement on postpositions in 195
auxiliary verbs object agreement in 201, 204
and multiple subject agreement 174, parameter settings in 166, 169–70, 212–13
208–14, 225, 239–40 Burzio, Luigi 67, 69
and person agreement 57–8
and position of the subject 228 Campana, Mark 81
role in structure 45, 77–8 Canela-Krahô 166–7, 194, 202, 203, 212, 214
Carnie, Andrew 232
Baker, Mark Carstens, Vicki 160, 172, 174, 181, 208
on copula types 62 case
on head movement 140, 199 agreement in, see concord, in case
Mirror principle 60 assignment of 162, 174, 180
on Mohawk 54 compared to adpositions 204, 206
particular analyses of agreement 154, 172, determining position of moved NP 227–8
206–7 ergative, effect on agreement 81, 169–70
on prepositions 112, 113 marking definite objects 231–2
principles of category theory 20, 26, 27–34, and operator–variable agreement 122
38, 45, 172–3 quirky, on subjects 87–93
on trace licensing 69 relationship to agreement 7, 40, 92, 155,
Basque 173–8
auxiliaries in 77–8 shared between noun and determiner 191
ergative agreement in 77–80, 85 types of 205
long distance agreement in 104–5 uniqueness of assignment 162–3, 209
parameter settings in 166, 183, 201, 205, see also Case Dependency of Agreement
213 Parameter, Case theory
Bejar, Susana 77, 79, 100 Case Dependency of Agreement Parameter
Belletti, Adriana 67 applied to auxiliary constructions 208–10,
Benmamoun, Elabbas 59 211, 212–14
Berber 212, 225, 236 applied to complementizer 180, 182, 183–4
Bhatt, Rajesh 40–1, 46, 91, 92 applied to determiner 187–8, 191
Bickel, Balthasar 91, 156, 170 applied to linker 207
Bittner, Maria 81, 101, 235 applied to preposition 191, 193
Bobaljik, Jonathan applied to tense 159–60, 161–2, 163,
on agreement at PF 9, 41, 87, 155, 205 169–71, 230
on agreement domains 40, 42, 107, 160 applied to v 201, 203–6, 215, 232
on copy deletion 227–8 consistency of 156, 222–3, 224, 225, 234,
on ergativity 80, 83, 101, 169, 235 239–44
Boeckx, Cedric 40, 41, 42, 88–9, 94 distribution of settings of 221–2, 236
Bok-Bennama, Reineke 83, 101, 235 evidence relevant to 246–7
Bonet, Eulalia 94, 151 statement of 155, 173–8
266 Index
Case theory 72–3, 79, 176 Cole, Peter 239

Cash Cash, Phil 232 Collins, Christopher 164, 172, 206–7
causatives, agreement in 96 complementizer phrases (CPs)
c-command agreement into 43, 44, 71, 77, 103–7
and agreement in person 51–2 hosting person operators 125–6, 129–31,
as condition on agreement 7, 40–1, 45–6, 138, 146
63–4, 148, 233 complementizers
as condition on movement 53 agreement on 23, 54, 117–20, 145–7, 181,
definition of 40, 51, 79, 115, 144, 147, 218 184, 209
in definition of intervention 47, 128, 144 distinguished from verbs 179
downward, for agreement 67–70, 89–90, licensing subject traces 182
200, 215, 224 complements
not required for operator–variable agreement with 52, 113–14, 187–8, 191–2
agreement 122 not allowed with modifier 62
symmetrical 187 relationship to specifier 47
upward, for agreement 42, 45–7, 74–85, concord
155, 224 in case 174–5, 211
see also Direction of Agreement Parameter reduced to agreement 7, 65–6, 108, 114,
Chicasaw 92–3, 109 173, 177
Chichewa control
agreement with anaphors in 150–1 of logophoric operators 146
gender in 33 of person operators 126, 129–32, 133–4
no agreement on Ps in 191 copula
object agreement in 42, 99, 196–7 agreement on 72, 89
subject agreement in 158–9 needed with nonverbal predicates 27, 28–9,
Chinese 39, 100 37
Choctaw 94 types of 62
Choe, Hyon-Sook 227 criterion of identity 31–3
Chomsky, Noam Croft, William 25, 26, 27
bare phrase structure 52, 79, 114 cyclicity 47
on c-command 40, 46–7, 51
on ergativity 80 Daga 183
on parameters 156 Davies, William 93, 94
on phases 43, 192, 235 definiteness
theory of Agree 6, 12, 40, 43, 44, 45, 52, and agreement 169
65–6, 75, 160, 172, 174–5 agreement in 17
theory of unvalued features 4 of object, as sign of movement 199–200,
Choueiri, Lina 126 201, 232, 234–5
Chukchi 83, 140–1, 201, 205, 213, 235 degree, agreement on 17, 117
Cinque, Guglielmo 67–9, 72–3 deletion of higher copies 81, 227–9
classifiers, as gender agreement 16, 19–20, den Dikken, Marcel 20
23–4 derivation, morphological 25, 26, 27–34,
clauses 36
as domain for phrasal movement 140–1 determiners
structure of 54 agreeing with complement 38, 114–16,
see also complementizer phrases 184–91, 217, 226–7, 238
clitics agreeing with possessor 48, 60, 115–16,
anaphoric 139 151, 173
distinguished from agreement 98, 194, as locus of phi-features 63–4
196–7, 200 Diercks, Michael 200
functional heads as 189 Diesing, Molly 160, 199
Index 267
Direction of Agreement Parameter expletive constructions

applied to complementizer 180–1, agreement in 43, 69, 89, 91, 159, 163–4,
183–4 172, 177
applied to determiner 186–9, 190 phi-features of expletive 237
applied to linker 207 extended projection 34, 50, 52, 53, 74, 86
applied to preposition 191, 192, 195
applied to tense 159–60, 161, 165–9, 210, features
230 different systems of 17, 108
applied to v 196, 198–9, 201–3, 232 expressed multiple times 99–100
consistency of 198, 200, 224, 225–7, 234, morphological effects on realization of 8–9
238 role in defining local domains 42, 141–3
distribution of settings of 221–2, 236 stipulated on individual lexical items 4, 13,
evidence relevant to 246–7 44, 50, 174
statement of 155, 173, 214–19 see also gender, number, person,
Distributed Morphology 9, 17, 38, 154 phi-features
ditransitive constructions, see double object Fijian 195, 202, 203, 212
construction Foley, William 14
Dixon, R. M. W. 26, 167 Franks, Steven 90
Doke, Clement 199 Frantz, Donald 96
domination 40, 51 French 139, 196
double object constructions Fukui, Naoki 156
agreement in 42–3, 94–8, 100, 151, 200, functional categories
204–6, 240 associated with nouns and adjectives 38,
contrasted with applicatives 216–17 74
Duranti, Alessandro 97, 98 as bearers of agreement 4, 20–1, 34–9,
Dutch 42, 64, 199 53–4, 55
bearing person agreement 57–60, 112–21,
Edo 135–6, 137–8 143–5
English and case assignment 174
agreement in expletive constructions in 43, having consistent properties in a language
69–70 220, 222–4, 236
agreement on demonstratives in 187 iteration of 49–50, 53–5, 91
agreement on predicate nominals in 10–11, made into variables by agreement 123–4
14 Person Licensing Condition applied to
agreement on verbs in 75–7, 86–7, 109, 138–9, 143–5
159, 162, 163, 208–14 probehood of 100, 102, 153–4
properties of lexical categories in 27–8
semantic agreement in 10 Gardiner, Donna 96
EPP features Geach, Peter 31, 32
not on adjectival heads 53, 74 gender
related to agreement, in Bantu 172 and agreement on verbs 8–9
on Tense, causing movement 52, 55, 100, different systems of 17
160, 166 expressed on some nouns only 39
two on same head 100, 101 inherent only on nouns 33
on v 95, 203 long distance agreement in 103–4
ergativity mismatches of 66
in agreement patterns 79, 80–1, 82, 83, 101, and semantic agreement 10
203, 224 Georgian 170, 177–8, 195, 201, 204, 212
effects of ergative case marking 169–70, German 64, 90–1, 199
203 Germanic languages, complementizer
Etxepare, Ricardo 104–5 agreement in 120, 146, 179–80, 209
268 Index
Givón, Talmy 124 ineffability 109

goals infinitives, agreement on 46, 209, 211, 232
experiencer subjects as 243 Innu-Aimun 107
as NP or PP 205 intervention
object agreement with 94–9, 205–6, apparent violations of 81, 115
216–17, 240 definition of 47, 144
position of 95 and operator–variable agreement 122
subject agreement with 101 restricting agreement 40, 41–2, 43, 47–8,
see also double object constructions 59, 71–2, 76, 161, 218
Gordon, Lynn 92–3 source of adjective versus noun difference
Greek 90–1, 98, 187 48–51
Greenlandic, see Inuit Inuit
Grimshaw, Jane 50, 74 agreement on determiners in 151, 190
Gujarati 91–2 agreement with objects in 235
Gupta 31, 32 case in 206
ergativity in 83, 101
Hale, Kenneth 81 parameter settings in 195, 205, 213
Halle, Morris 9, 17 Italian
Harbour, Daniel 102 agreement on adjectives in 47–8, 67–9,
Harley, Heidi 17, 33 70–1, 72–3, 145
Haspelmath, Martin 98 impersonal sentences in 91
Hausa 236 oblique subjects in 159
Haya 98 unaccusativity in 29–30
Hebrew 35–6, 213, 227, 236 Itelmen 107
Heim, Irene 123, 125 Iwara, Alex 103, 104, 164
Hermon, Gabriella 241–2
Hindi Jacaltec 112–13, 165, 183, 194, 212
incorporated objects in 169 Jackendoff, Ray 33, 112
long distance agreement in 40–1, Jake, Janice 240, 243
46 Japanese 39
oblique subjects in 159, 169–70 Jarawara 167, 190, 212
Icelandic Kanerva, Joni 158, 160

agreement on adjectives in 173–4, Kannada 61–2, 90, 166, 213
176–7 Kaplan, David 124
agreement on participles in 174, 211 Kariaeva, Natalia 177
anaphora in 139 Kayne, Richard 30, 52, 125
copular clauses in 89 on agreement 75–6, 77, 86, 162, 198, 208
defective intervention in 41–2, 48 on crosslinguistic variation 156
object fronting in 160 Kilega
quirky case subjects in 87–90, 109, 145, agreement on complementizers in 181
159, 176–7 agreement on prepositions in 193
raising adjectives in 70, 71–2 subject agreement in 159, 160, 191, 208
imperatives 127 Kimball, J. 75–7, 86
impoverishment, rules of 9 Kinande
incorporation adjectival agreement in 24, 171
and agreement 168 agreement on C in 120, 146, 147, 179,
locality of 140–1 180–2
as test for functional categories 35–6 linker head in 206–7
into verbs only 29 object agreement in 196, 197
Index 269
subject agreement in 158–9, 160, 163–4, Mayali

208, 209–10 adjective versus verb agreement in 24–5
upward agreement in 75, 187, 191–3 gender in 36, 39, 99
Kinyalolo, Kasangati 52, 172, 191, 192, 193, noun versus adjective agreement in 15–16,
208 19
Kinyarwanda 160 Mchombo, Sam 124, 196
Kirundi 160, 164 Merge 47, 52, 112–14, 143
Koopman, Hilda 52, 113, 125, 129, 135, Minimal Compliance 200
166 Minimalism 46, 114, 156
Kornfilt, Jaqueline 58 see also Chomsky, Noam
Kratzer, Angelika 123 Mirror Principle 60, 82
modifiers
labels, of categories 52, 55–6, 79 adjectives as 33–4
Laka, Itziar 45 agreement properties of 17–21, 24, 50–1,
Lakhota 100 60–4, 147–8, 218
Lango 58 as appositional structures 17
Launey, Michel 58, 102 nouns as 21
Levin, Juliette 80 Mohawk
lexical categories active agreement in 54
definitions of 5, 26, 27–34 agreement on nouns in 116
differing agreement properties of 1–3, 4, agreement parameters in 99, 201
12, 13–27 incorporation in 29, 140
not true bearers of agreement 34–6 morphology, influence on agreement 7–9, 17,
Li, Yafei 35 87, 154–5, 204, 234
linkers movement
in attributive constructions 64 conditions on 53, 139–42
as locus of agreement in NPs 20–1, creating agreement configurations 161,
119 163, 188, 192, 196, 198–9, 227
in verb phrases 206–7 of goal arguments 95
Li’s Generalization 35–6 interactions with agreement 74–5, 81,
locality conditions 139–43 165–6, 198, 206
see also intervention, Person Licensing not dependent on Agree 160, 167
Condition out of PP 193
locative inversion 158–9 as precondition for agreement 77, 81, 103,
Lokaa 106–7, 235
agreement on complementizer in 118–19, types of 161
120, 148 Müller, Gereon 141
agreement on tense in 164, 208 Munroe, Pamela 92–3
long distance agreement in 103–4, 109 Murasugi, Kumiko 81
Luvale 164, 190, 211
Nahuatl
MacKenzie, Marguerite 107 agreement in ditransitive constructions in
Madariaga, Nerea 117 94–6, 98, 99, 102, 109
Maithili 91, 156 agreement on nouns in 58–60, 116
Makhuwa 237–9 agreement on prepositions in 113, 195
Mapudungun 115–16, 154 parameter settings in 165, 201
Marantz, Alec 9, 17, 38, 42, 162 Ndayiragije, Juvénal 162, 164
Markman, Vita 89, 177 negation 27
Massam, Diane 80, 202 Nepali 91, 156, 169–70
matching, as condition on agreement 44 Nevins, Andrew 125, 129, 133–4
270 Index
Nez Perce 101, 184, 229–34, 235 Ottosson, Kjartan 70

Niinuma, Fumikazu 42 Ouhalla, Jamal 5
Noonan, Michael 58
Noun Licensing Condition 20, 28, 34 parameters
nouns concerning agreement 7, 44
absence of agreement on 2–3, 17–21, 48–51 independence of 168–9, 239
agreeing with possessor 116 location of phi-features as 64
definition of 28, 30–2 nature of 155–7, 219–24, 236, 245
functional heads associated with 36, 38, 39, in probing up or down 85
188–9 in whether agreement happens 100, 154
as having intrinsic phi-features 31–3, 126, see also Case Dependence of Agreement
127 Parameter, Direction of Agreement
locative, compared to adpositions 195 Parameter
as modifiers 21 participles, agreement on 174–5, 198, 208,
person agreement on 56–60 209, 210–11
see also predicate nominals particles 207
number Passamaquoddy 77, 87, 105–7, 151, 214
agreement in 86–7, 105, 109 passive voice
different systems of 17 and agreement 84–5, 101–2, 107
impoverishment of 9 distinguishing clitics and agreement 98, 196
location of, in NP 38, 62–3 pseudopassives 192
mismatches of 66 Perlmutter, David 96
not affected by ergative displacement 80 person
only nouns specified for 32 distribution of agreement in 2, 3, 21, 85–91,
and semantic agreement 10–11 103–7, 108–9
in ergative agreement systems 101
objects on functional heads 112–21, 185
agreement with 3, 22, 35, 53–5, 101, 156, impoverishment of 9
173, 196–204, 206, 223, 237–8, 240 inherent only on nouns 32, 126–7
agreement with, versus clitics 98, 196–7 marked on verbs but not adjectives 21–7,
compared to oblique subjects 242 51–6, 73–4, 89, 174
complementizer agreeing with 184 more restricted than number and gender 7,
definiteness of, affecting agreement 149–50
199–200, 231–2, 234–5 on nouns and adjectives 56–60, 86
moved to Spec, TP 160–3, 164, 229 participles agreeing in 208
participles agreeing with 198, 208 as result of operator–variable agreement
position of 62 123–4, 148
tense agreeing with 76, 87, 91, 224, 232–5 special locality conditions on 92, 124–38
see also double object constructions underspecification of 77
O’Herin, Brian 81–2, 224 see also Person Licensing Condition,
operators Structural Condition on Person
agreeing with variables 121–4 Agreement
agreement with 118–19, 180, 181, 182–3 Person Case Constraint (PCC) 94–8, 151
first and second person 120–1, 125–8, 131, Person Licensing Condition (PLC)
136, 138, 142–3, 180 motivation for 126–8, 129, 132–4, 137–8
licensing person agreement on heads SCOPA derived from 138–9, 142–5, 147–9
145–7 phases
logophoric 125, 135–7, 146, 179 clauses as 75
see also wh-phrases not restricting operator–variable agreement
Ormazabal, Javier 95, 197–8 121
Index 271
prepositional phrases as 44, 71, 192–3, 204, prepositions

206 agreement on 112–13, 114, 143, 151,
as restricting agreement 40, 43–4, 47, 48, 173, 191–4, 196, 217
71, 75, 77 compared to case markers 204, 205,
vP as 235 206
phi-features compared to locative nouns 195
+anaphor as 150 as functional categories 33
completeness of, and case assignment 174 probes
definiteness not included among 17 functional heads stipulated as being 39,
in different kinds of agreement 123 100, 102, 153, 234
of expletive subjects 237 not defined by unvalued features 44
intrinsic only on nouns 31–3 searching upward or downward 47
left unspecified 76, 79, 80 pro-drop 110, 127
of locative phrases 160–1 pronouns
of oblique subjects 88–9 agreeing with their antecedent 121–3
position of in structure 37, 63–4, 148 agreement with 24, 89, 114–15, 118,
on predicate nominals 49–50 138–9, 194
related to EPP, in Bantu 172 compared to agreement 7, 149–50
spelled out on functional heads 36 licensing person features of 124–7, 128,
see also gender, number, person 131, 138, 142–3
Pica, Pierre 139 logophoric 125, 135–8
Polinsky, Maria 103 modification of 147–8
Pollard, Carl 31, 108 as source of agreement in predication
Portuguese 209 structures 61–2
possessors properties 25–6
adjectival 185
agreement on determiners with 60, 115–16 quantifiers, see determiners
agreement on tense with 232–4 Quechua
postpositions, see prepositions differences among dialects of 239–44
Potsdam, Eric 103 parameter settings in 166, 195, 201, 205,
Pred 212, 214
blocking person agreement on adjectives
144, 218 raising 156, 209–10
in nonverbal predication 29–30, 37, 45 Reference-Predication Constraint 28, 33
possibility of agreement on 45, 47, 59–60 referential indices 30–1, 127–8, 148
predicate, nonverbal Reimsdijk, Henk van 193
agreement on 45, 60–4, 188–9, 219 Reinhart, Tanya 40, 152
diagnostics for 26 relative clauses 118–19, 147, 148, 183
in instrumental case 177 restructuring 40, 46, 140–1
predicate nominals Reuland, Eric 152
absence of agreement on 2–3, 14–17, 49–50 Rezac, Milan 77–9, 80, 81, 100
person agreement on 56, 59 Rice, Keren 129–33
semantic agreement on 10–11, 61–2 Richards, Norvin 200
structure of 28–9, 36–8 Ritter, Elizabeth 17, 33, 38
tense marked on 30 Rizzi, Luigi 67, 141, 182
prepositional phrases Rögnvaldsson, E. 160
as islands for agreement 43, 44, 70–1, 73 Romero, Juan 95, 197–8
movement out of 193 Rood, David 100
oblique subjects as 89, 90–1 Rude, Noel 229, 230
as phases 71 Russian 90–1, 117, 159, 177
272 Index
Safir, Kenneth 121, 125, 135, 138 Stjepanovic, Sandra 46

Sag, Ivan 31, 108 Stowell, Timothy 180
Saltarelli, Mario 63 Structural Condition on Person Agreement
Schlenker, Philippe 123, 125, 126, 129, 132 and agreement on adjectives 73, 85
Schütze, Carson 41, 87, 209 and agreement on nouns 59–60
scrambling 165, 202, 217 and agreement on verbs 89, 93, 94–5, 103,
semantics, influence on agreement of 10–11, 244
108, 121 and anaphoric features 150–2
Serbian 46, 66, 108, 210 applied to functional heads 112–21
Sesotho 159 compared to other conditions 92
Shambala 97, 98, 99, 109, 200 definition of 52–6, 65
Shlonsky, Ur 227 derivation of 111, 138–9, 143–6, 147–9
Sigurð sson, Halldór 48, 70, 71 subjects
on person operators 125 agreement with 158–71, 173, 230, 237
on quirky subjects 88–9, 90–1, 176 case marking of 162
Singhapreecha, Pornsiri 20 complementizer agreeing with 180–1, 183,
Slave 184
parameter settings in 151, 189–90, 195, with oblique case 87–91, 94, 145, 159, 160,
202, 203, 212 169–70, 176–7, 223, 241–4
shifted pronouns in 126, 129–33, 136, 137, postverbal 225–6, 227–9
142 v agreeing with 77–85, 215–16
small clauses 48 see also ergativity
Spanish Swahili
agreement on functional heads in 38, 114, agreement on determiners in 114, 187
117, 187 agreement on lexical categories in 1–3, 12,
clitics/object agreement in 98, 197–8 14–15, 18, 22, 36, 38–9
linking particle in 20 copulas in 2
location of phi-features in 64 gender agreement in 8, 10
noun versus adjective agreement in 15, linking particle in 20–1
17–18 object agreement in 54–5, 150, 200
verbal agreement in 8, 39, 126–7, 208 subject agreement in 159, 160, 164, 208
verb versus adjective agreement in 22,
171 Taraldsen, Knut 87–93
Speas, Margaret 156 Tariana
specifiers nonagreeing verbs in 23
adjectives not having 68, 74 noun versus adjective agreement in 15,
of CP, agreement with 75–7 16–17, 19–21
as defining property of verbs 28–30, 37, 51, parameter settings in 189
52 v agreeing with subject in 83–5, 203
as distinguished position for agreement 52, verb versus adjective agreement in 23–4,
79, 113–14, 172–3 55, 147
multiple 54, 144 Taylor, Allan 100
properties of 47, 190 tense
on right edge of phrase 168, 203 agreeing downward 217–18, 227
Sportiche, Dominique 52, 125, 129, 135 agreeing upward 74–7, 86–7, 162
Sridhar, S. N. 61 agreeing with object 76, 81–2, 87–93, 94,
Stassen, Leon 3, 25–7 224
Stechow, Armin von 123, 125 agreeing with possessor 232–4
Sternefeld, Wolfgang 141 agreeing with postverbal subjects 81, 89,
Stewart, O. T. 135–7 159, 226, 237
Index 273
agreement parameters applied to 156, as locus of object agreement 35, 42–3,

158–70, 171, 208–11, 223, 229, 230, 53–4, 196–7, 223
242–4 Van der Wal, Jenneke 237–9
and assignment of case 162, 174 variables
differences in, affecting agreement 80 agreeing with their operators 121–4
double agreement on 83, 100–1, 102, 232–5 created by agreement 124
and locality of agreement 42, 81–2 verbs
as locus of subject agreement 34–5, 52, 54, agreeing upward 74–85
57–8, 59 without agreement 23
nonfinite 75, 87 and agreement in person 2, 85–107, 109
not agreeing with subject 84 agreement on, compared to adjectives 21–6,
only verbs inflect for 27, 30 51–5, 65–7, 107–8, 173
person agreement on 143 definition of 28–30
variable probehood of 39, 100–2, 167 derived from adjectives 36
verb moving to 198–9 gender agreement on 8–9
thematic (theta) roles 28, 31 impoverishment of features on 8–9
Theta Criterion 34 moving to Tense 198–9
Thráinsson, Höskuldur 160 structure of clause headed by 34–5, 36–8
Tiwa, Southern subject agreement on 157–71
agreement in ditransitive constructions in
96–7, 101–2, 109, 151 Warlpiri 170–1, 201, 204, 213
verb incorporation in 35 Watanabe, Akira 180
topics 90–1, 93 Webelhuth, Gert 156
Travis, Lisa 140 Wechsler, Stephen 31, 66, 108, 114
Tsez 103 Welsh 194
Turkish wh-phrases
agreement on all categories in 56–8 agreement on complementizer with 181–2
parameter settings in 168–9, 183, 213 agreement on verb with 75–7, 86–7, 109,
Tzotzil 162
ergative agreement in 82, 83 Wichita 29
parameter settings in 213–14, 224 Williams, Edwin 31
Woolford, Ellen 150, 229–30
unaccusativity word order, influence on agreement 165–9,
of adjectives 67–8 186, 189–90, 192, 202–3
special agreement on unaccusative verbs Wurmbrand, Susanne 40, 107, 160
83–4, 101, 215–16
of verbs versus adjectives and nouns 29–30, Yadava, Yogendra 91, 156, 170
45 Yiddish 160
unergative verbs 215–16 Yimas 14, 195, 201
universals of agreement 3, 26, 27–34 Yoruba 136–7
Ura, Hiroyuki 163
Zanuttini, Rafaella 127
v Zazaki 133–4, 136
agreeing upward with subject 77–85, 203, Zlatić, Larisa 31, 66, 108, 114
224 Zulu
agreement parameters applied to 196–206, agreement on modifiers in 147, 190
215–17, 232, 238, 240 agreement on quantifiers in 114–15, 185–6
in double object constructions 94–100, 102, object agreement in 198–200
200 subject agreement in 159, 163, 208

The Syntax of Agreement and Concord 2008

Enviado por

Dados do documento

Direitos autorais

Formatos disponíveis

Compartilhar este documento

Compartilhar ou incorporar documento

Opções de compartilhamento

Você considera este documento útil?

Este conteúdo é inapropriado?

Direitos autorais:

Formatos disponíveis

The Syntax of Agreement and Concord 2008

Enviado por

Direitos autorais:

Formatos disponíveis

This page intentionally left blank

Earlier issues not listed are also available

The Syntax of Agreement and Concord

Cambridge University Press

© Mark Baker 2008

This publication is in copyright. Subject to statutory exception and to the provision of

ISBN-13 978-0-511-38884-2 eBook (NetLibrary)

ISBN-13 978-0-521-85547-1 hardback

ISBN-13 978-0-521-67156-9 paperback

1 Introduction: category distinctions as a window on the

2 Basic agreement and category distinctions 12

3 The unity of verbal and adjectival agreement 65

4 Explaining the restriction on person agreement 111

5 Parameters of agreement 153

Appendix: Table of languages and their

As I get older, I seem to be losing my sense of history. As in many families,

to do the survey of 108 languages reported in chapter 5 if I had not received

1–19 noun class (Bantu)

npst nonpast tense

The following are abbreviations of linguistic terms: names of principles, gram-

a, ap adjective, adjective phrase

Finally, the following are some conventions used in presenting examples:

(x) The example has the same grammatical status with or

1.1 A generalization to be explained

c. Ni-li-po-kuwa ki-jana . . . sasa ni-li-po m-tu m-zima, . . .

(4) Dalili y-a mvua ni ma-wingu. (Ashton 1949)

1.2 The incompleteness of previous discussions

(5) The Agreement Universal

In connection with this universal, he writes:

As far as I am aware, no known principle of linguistic theory prevents us from

(6) soma refu jana

It might seem tempting to take the agreement potential of a word as being

1.3 What a better theory could look like

1.4 What is in this book

agreement in oblique subject constructions, agreement in double object con-

1.5 What is not in this book

(7) a. Nosotros com-emos las manzanas.

language. For example, there could be a general rule of impoverishment of

This sort of agreement is semantic in that it depends on semantic properties

Given this, I take the appearance of agreement in (9c–d) to be an epiphenomenon

(1) a. Ni-li-kuwa ni-ki-som-a. (Ashton 1949)

(section 2.3). The crucial adjustment is that agreement-bearing heads must be

2.1 The generality of the categorical asymmetries in agreement

2.1.1 Adjectives versus nouns: predicative structures

(3) Mi-zigo hii mi-zito (*ki-zito).

(4) a. Est-as mujeres son gord-as. (*gord-o)

The Australian language Mayali has four genders (masculine, feminine,

(7) a. Na-meke bininj na-kimuk. (Evans 2003:557)

(8) a. Ngaleh ngal-yuhyungki ngurrurdu ngaleng bininj, Bulanjdjan.

The South American language Tariana has an unusual gender system.

(9) a. Wyume-ma-se matSa-ma-pidana. (Aikhenvald 2003:499)

(10) a. Pedalie-ma-pe inari na-matSi-ka

Notice that there are significant differences in the systems of grammatical

2.1.2 Adjectives versus nouns: modifying structures

(11) a. el hombre gord-o (Spanish, agreeing adjectives)

(16) s-sakak-ı̄n ə l-fə dd-a (Arabic, nonagreeing noun modifier)

(17) a. na-rangem na-mak ‘good boy’ (Evans 2003:182)

Only in Tariana there is a complication. Tariana’s attributive adjectives do

(19) a. heku-na hanu-na (Aikhenvald 2003:73)

(20) a. heku-na [tsuli i-tape]-na (Aikhenvald 2003:85)

This seems like a counterexample to my otherwise robust generalization. But

(21) a. una pulsera *(de) oro

to another noun. Thus a preposition or similar particle is needed to θ -mark

2.1.3 Adjectives versus verbs