Escolar Documentos
Profissional Documentos
Cultura Documentos
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
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 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
ix
x Contents
xi
xii Acknowledgments
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.
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
vis visible
x special gender class in Burushaski
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
2 Introduction
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.
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.
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:
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
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).
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
What is not in this book 9
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.
(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.)
What is not in this book 11
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.
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
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.
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.
The generality of the categorical asymmetries in agreement 15
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).
Example (4) illustrates the same contrast in Spanish, a language that distin-
guishes two grammatical genders and has a two-way number system.
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.’
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)).
The generality of the categorical asymmetries in agreement 17
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
18 Basic agreement and category distinctions
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.
The generality of the categorical asymmetries in agreement 19
(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.
The surprise is that when a noun modifies another noun in Tariana, it too seems
to have a classifier suffix:
20 Basic agreement and category distinctions
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
The generality of the categorical asymmetries in agreement 21
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.
22 Basic agreement and category distinctions
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.
In Syrian Arabic, verbs are like adjectives in agreeing with their subjects in
gender and number:
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’
The generality of the categorical asymmetries in agreement 23
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.’
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.
24 Basic agreement and category distinctions
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).
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.
In similar semantic situations, adjectival roots can even bear first and second
person agreement prefixes:
b. Yi-keb-mak.
2sS-face-good
‘You are good looking.’
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):
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
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.
28 Basic agreement and category distinctions
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).
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:
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.
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)).
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).
30 Basic agreement and category distinctions
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.
The category-theoretic infrastructure 31
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.
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).
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
32 Basic agreement and category distinctions
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 category-theoretic infrastructure 33
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).
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)).
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).
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.
The category-theoretic infrastructure 37
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
38 Basic agreement and category distinctions
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).
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.
The category-theoretic infrastructure 39
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.
40 Basic agreement and category distinctions
(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.
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
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).
(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
42 Basic agreement and category distinctions
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.
The agreement-theoretic contribution 43
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 ]]]
(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,
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.
The agreement-theoretic contribution 45
(68) PredP
NP Pred´
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
46 Basic agreement and category distinctions
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.’
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,
The agreement-theoretic contribution 47
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).
48 Basic agreement and category distinctions
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).
(75) F agrees with XP, XP a maximal projection, only if:
a. F c-commands XP or XP c-commands F (the c-command condition).
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).
d. XP is made active for agreement by having an unchecked case feature
(the activity condition).
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
NP Pred´ NP Pred´
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
50 Basic agreement and category distinctions
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.
NP NP
NP FAP NP FNP
small silver[F,sg]
Explaining the basic categorical asymmetries in agreement 51
(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
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:
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:
Explaining the basic categorical asymmetries in agreement 53
NP FV´
we FV VP
[1Fpl][1Fpl]
NP V´
<we> V (PP)
fall
(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
54 Basic agreement and category distinctions
transitive clause with double agreement on the verb, such as (82a) from Swahili,
is thus roughly as shown in (82b).
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.)
Explaining the basic categorical asymmetries in agreement 55
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.
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
56 Basic agreement and category distinctions
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:
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:
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
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.
60 Basic agreement and category distinctions
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).
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.
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).
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).
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
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´
*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
64 Basic agreement and category distinctions
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
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
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.
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):
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.
Downward agreement on adjectives 69
(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.
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].
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.
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.
Downward agreement on adjectives 71
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.
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
72 The unity of verbal and adjectival agreement
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.
Downward agreement on adjectives 73
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]]]]
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.
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.
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.
76 The unity of verbal and adjectival agreement
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.
Upward agreement on verbs 77
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.)
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
78 The unity of verbal and adjectival agreement
(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.
Upward agreement on verbs 79
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.
80 The unity of verbal and adjectival agreement
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.
Upward agreement on verbs 81
(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
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.)
82 The unity of verbal and adjectival agreement
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.
84 The unity of verbal and adjectival agreement
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
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.
(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.
Verbs that cannot agree in person, like adjectives 87
(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).
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.
88 The unity of verbal and adjectival agreement
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.
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.
90 The unity of verbal and adjectival agreement
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.
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
Verbs that cannot agree in person, like adjectives 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.
92 The unity of verbal and adjectival agreement
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).
Verbs that cannot agree in person, like adjectives 93
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
94 The unity of verbal and adjectival agreement
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.
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.
Verbs that cannot agree in person, like adjectives 95
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.
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).
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.
96 The unity of verbal and adjectival agreement
(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]
(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
Verbs that cannot agree in person, like adjectives 97
(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:
98 The unity of verbal and adjectival agreement
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).
Verbs that cannot agree in person, like adjectives 99
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.
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.
Verbs that cannot agree in person, like adjectives 101
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).
102 The unity of verbal and adjectival agreement
(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.
Verbs that cannot agree in person, like adjectives 103
argument does not merge with the functional category is different in the two
structures, but the consequences for agreement are the same.
(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
104 The unity of verbal and adjectival agreement
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.’).
Verbs that cannot agree in person, like adjectives 105
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.
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:
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.
106 The unity of verbal and adjectival agreement
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:
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.)
108 The unity of verbal and adjectival agreement
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
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
110 The unity of verbal and adjectival agreement
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).
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.
4.1.1 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.
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).
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
114 Explaining the restriction on person agreement
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.
Person agreement and other categories 115
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
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.
116 Explaining the restriction on person agreement
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)’
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).
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.
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.
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
Person agreement and other categories 119
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) 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.
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).
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):
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.
(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.
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).
122 Explaining the restriction on person agreement
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.
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.
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).
124 Explaining the restriction on person agreement
(26) If F agrees with XP, then F counts as a variable that referentially depends on
XP.
(27) The man who is talking to you wants you to give him/*me some money.
(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.
126 Explaining the restriction on person agreement
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
A locality condition on first and second person variables 127
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.
(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.
128 Explaining the restriction on person agreement
(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.
A locality condition on first and second person variables 129
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).
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)).
10 (37) can also have an unshifted reading, in which it means ‘John said that I am a hero’; see (48)
below for discussion.
130 Explaining the restriction on person agreement
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
A locality condition on first and second person variables 131
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
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:
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.
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.
132 Explaining the restriction on person agreement
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).
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.
A locality condition on first and second person variables 133
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:
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.
The same ambiguity is found in Anand and Nevins’s (2004) examples from
Zazaki, including (49).
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).
134 Explaining the restriction on person agreement
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.
A locality condition on first and second person variables 135
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).
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]]]
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.
A locality condition on first and second person variables 137
pronoun, however, the sentence is fine with the relevant interpretation (Adesola
2005:185, 192).
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).
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).
138 Explaining the restriction on person agreement
(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.
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).
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:
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).
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
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
142 Explaining the restriction on person agreement
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.
(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).
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.
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).
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.
Deriving the SCOPA 145
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.
146 Explaining the restriction on person agreement
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
‘I convinced Kambale that he should buy fruits.’
[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.
Deriving the SCOPA 147
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).
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 Susan Norman Wells 1sS.opt.go 3sS.want 3sS.say
‘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
150 Explaining the restriction on person agreement
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).
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.
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).
152 Explaining the restriction on person agreement
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.
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
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
154 Parameters of agreement
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).
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.
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.
Agreement on tense 159
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:
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.
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).
160 Parameters of agreement
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).
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).
Agreement on tense 161
(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.
162 Parameters of agreement
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.)
Agreement on tense 163
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:
(15a) illustrates normal SOV order, with the verb showing subject agreement
with the agentive subject (as well as object agreement with the direct object).
166 Parameters of agreement
(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) 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
Agreement on tense 167
(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).
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).
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.
168 Parameters of agreement
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
Agreement on tense 169
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
170 Parameters of agreement
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
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.
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.
172 Parameters of agreement
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
Agreement on FA and the formulation of the parameters 173
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).
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
Agreement on FA and the formulation of the parameters 175
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).
176 Parameters of agreement
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.
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).
178 Parameters of agreement
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)).
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
In the West Germanic languages, agreement is with the subject of the embedded
clause – the same NP that the lower verb agrees with:
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.)
180 Parameters of agreement
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
Agreement on complementizers 181
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:
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.
182 Parameters of agreement
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.
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’
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):
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:
The difference is that ‘all’ can also agree with a pronoun in first or second
person features, at least in Swahili and Zulu.
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
186 Parameters of agreement
(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:
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).
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
Agreement on determiners 187
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:
188 Parameters of agreement
NP Pred´
cloud FN NP
[8,pl] [8,pl]
[*9,sg] cloud
AGREE
MOV’T
Agreement on determiners 189
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.
190 Parameters of agreement
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.
Some, like Chichewa, also have a dative preposition kwa. This too fails to agree
with its complement:
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
Agreement on adpositions 193
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).
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.
194 Parameters of agreement
opportunity, but a few do. Welsh in particular has many Ps that agree with
unmoved pronominal complements:27
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.
Agreement on adpositions 195
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.’
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).
196 Parameters of agreement
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.
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
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.
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).
(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
(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
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
Another language that shows a Bantu-like relationship between word order and
object agreement is the Athapaskan language Slave (Rice 1989:1197):
Yet another is Fijian (see Massam 2001 for a detailed analysis of the difference
between these sorts of sentences in the related language Niuean):
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
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.
204 Parameters of agreement
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).
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).
206 Parameters of agreement
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).
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).
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
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.
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
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.
Agreement in auxiliary constructions 209
In Kinande, it is possible to have full agreement on both the matrix verb and
the embedded verb, as shown in (90b).
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).
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).
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
Agreement in auxiliary constructions 211
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):
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:
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
212 Parameters of agreement
(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).
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):
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:
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.
214 Parameters of agreement
b. Laj j-maj-ot.
end 1E-strike-2sA
‘I’m done hitting you.’
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.
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’ . . .)
(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.
216 Parameters of agreement
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).
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 ]
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).
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
218 Parameters of agreement
(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:
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
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.
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.
Table 1
Agree must be up
((1) = Yes) Agree can be up or down ((1) = No)
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:
Many little parameters or two big parameters? 223
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:
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:
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.
224 Parameters of agreement
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.
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.
Many little parameters or two big parameters? 225
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).
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:
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:
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.
228 Parameters of agreement
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).
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
Many little parameters or two big parameters? 229
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.
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.
(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.
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
232 Parameters of agreement
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
Many little parameters or two big parameters? 233
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:
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
234 Parameters of agreement
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.
Many little parameters or two big parameters? 235
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
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.
Many little parameters or two big parameters? 237
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):
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
238 Parameters of agreement
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:
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).
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).
Many little parameters or two big parameters? 239
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):
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.
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).
240 Parameters of agreement
(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).
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)).
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
Many little parameters or two big parameters? 241
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):
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):
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
242 Parameters of agreement
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.
Many little parameters or two big parameters? 243
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.
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).
244 Parameters of agreement
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
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:
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.
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
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.
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
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
Language and Linguistic Theory 23:809–65.
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
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
Language and Linguistic Theory 8:41–80.
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
Language and Linguistic Theory 13:75–134.
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.
Language 58:81–115.
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
Language and Linguistic Theory 25:315–47.
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.
Natural Language and Linguistic Theory 19:583–646.
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
Language and Linguistic Theory 20:691–724.
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