0.1 Agreement and its Role
Rules of agreement, or concord, account for the special relationship between the verb and the `subject' that exists in most languages. In many traditional grammars, it is taken for granted that this agreement occurs and usually only the cases of `lack of concord' are discussed. Kellner (1905: 46) writes "[t]he first rule of every syntax, namely, that a finite verb agrees with its subject in number, is very often sinned against in all periods of English ... Of this concession made by grammar to psychology, there are instances from Old English down to our own day" (his italics). Others see agreement as less than essential. For instance, Jespersen (1922 : 335) remarks that verbal agreement is a superfluity and that languages would do well to get rid of it: "By getting rid of this [agreement] superfluity, Danish has got the start of the more archaic of its Aryan sister-tongues". Nominal agreement is "an heirloom from a primitive age" (p. 352).
In the present work, I will examine rules accounting for the agreement phenomena (mostly in a recent Chomskian framework) as well as account for the instances where agreement is `deficient' and show this is not due to psychological factors but to grammatical ones. Rather than regard the lack of agreement as `deficient' or as progressive, I will consider it as indicative of syntactic phenomena. For instance, when a verb follows the subject, the verb may display more agreement than when the verb precedes the subject. In languages with object agreement, when the object precedes the (participial) verb, the verb displays more agreement. This indicates that agreement between NP and V occurs when the NP precedes V. In cases where the verb precedes the subject or the object, there may be an expletive agreeing with the verb and, as a result, the agreement may be `reduced'. Words lose lexical content over time (as noticed, for instance, in Bopp 1816 and von der Gabelentz 1891). This gradual process is referred to as grammaticalization (cf. Heine 1984) and may also contribute to the `loss' of agreement since elements that grammaticalize typically lose person and number features. Structural configurations such as coordinate NPs are opaque or non-transparent and are, in many languages, the cause of a `breakdown' of agreement. In 0.3, I elaborate on these circumstances.
I have restricted myself to syntactic structures and do not deal with with collective and special NPs such as army, police, government, alms, bellows, or with agreement as in one kind of people are represented, nor why some languages consider some nouns plural whereas others consider them singular (see e.g. Quirk et al 1985: 757ff; Zandvoort 1945 : 305-313). It seems to me that these phenomena are semantic rather than syntactic.
In connection with agreement, the roles of subject and Case are important even though this will not be my main focus. The main focus will be the verb. There are languages where the verb agrees with the `subject' (e.g. English, Dutch, Arabic); those where it agrees (or agrees under certain circumstances) with the `object' (e.g. Basque, Inuktitut, Dyirbal, Urdu/Hindi, Georgian); and, those where verbal agreement indicates both subject and object (e.g. Diné).
In this introductory chapter, I will first (0.2) provide an overview of the different accounts of agreement in recent (transformational) syntax. Then (0.3), I show how the lack of agreement comes about in such a framework and why it is relevant and (0.4) how agreement features are represented. Finally (0.5), I provide an outline of the remainder of the book.
0.2 Transformational Accounts of Agreement
In this section, I indicate the changes in Phrase-Structure Rules and agreement that have occurred in recent years. I outline the system of the early 80s and subsequent changes up to Chomsky's 1995 The Minimalist Program. The latter work contains as chapters 2 and 3 a number of articles that circulated earlier, namely "Some Notes on Economy of Derivation and Representation" (1989) and "A Minimalist Program for Linguistic Theory" (1992). Since chapter 4, "Categories and Transformations", differs substantially from the other chapters, I refer to chapter 2 as Chomsky (1989), to chapter 3 as Chomsky (1992), and to chapter 4 as Chomsky (1995). There is also a Chomsky (1994), "Bare Phrase Structure", which is not included in The Minimalist Program.
In a `Government-Binding' framework, verbal agreement is linked with nominative Case. Chomsky (1981: 52) says: "Subjects are nominative when they agree with the verb" and Borer (1986: 378) argues that agreement with the verb is a manifestation of nominative Case. Except under special circumstances (predicative NPs, adjuncts), a non-agreeing NP cannot be nominative. It is unclear in these accounts whether nominative Case is a condition for verbal agreement or vice versa. The Phrase Structure Rules in Chomsky and Borer are as in (1) and all Case is assigned under a sister relationship, i.e. from the verb to the (object) NP and from AUX to the (subject) NP. Agreement is shared by the NPs with the V and AUX. A tree is provided in (2) below:
1. S ==> NP AUX VP
VP ==> V NP
NP AUX VP
In Chomsky (1986b), Functional Categories such as AUX are considered on a par with lexical categories and head their own projections. Thus, (2) is reformulated as (3), with AUX changed into I(nflection). In (3), I projects to a full maximal projection, namely IP, which also contains a specifier position. The subject occupies the specifier position and is no longer in a sister relationship with the AUX, now I(nflection). Therefore, a Specifier-Head (Spec-Head) relationship is introduced to account for nominative Case and verbal agreement between the NP in Specifier position and the verbal element in the Head I position:
Chomsky (1986b: 24) takes Spec-Head agreement to be "a form of `feature sharing' [...] in fact, sharing of the features person, number, gender, Case, etc.". So, in (3), the Specifier of IP and the head of IP share features of Case and agreement. If one takes government to be defined as m-command, government is a relation similar to Spec-Head agreement. An element a m-commands b if and only if every maximal projection dominating a also dominates b and vice versa (cf. Aoun & Sportiche 1983). Hence, a Head governs the Specifier position. In addition, however, a Head governs the complement and the definition of government is therefore broader than that of Spec-Head agreement. In the remainder of the book, I use c-command as a condition for government, but this is only crucial in 4.1. In the Chomsky (1986b) framework, as in (3), Case to the object inside the VP remains assigned under government by the sister verb. Hence, c-command would be required for this and m-command would be too broad.
Since Abney (1987), Functional Categories have become relevant to NPs. Most NPs are seen as dominated by D(eterminer)Phrases and as including NumberPhrases and other Functional Categories tied to features. I follow common practice (cf. Chomsky & Lasnik in Chomsky 1995: 59) in continuing to refer to them as NPs.
0.2.2 `Early' Minimalism
Chomsky (1992), based on Pollock (1989) and Chomsky (1989), i.e. `early' Minimalism, argues that all Case is checked (rather than assigned, see 0.4) in a Spec-Head relationship. For this purpose, several functional categories are introduced, such as AGRs and AGRo in (4). NPs move to the Specifier positions and verbs move to the Head positions. Nominative Case is checked against AGRs and objective is against AGRo:
Verbal agreement is checked in a Head-Head relationship between V and AGR after the verb incorporates into the AGR Head as in (5). The person and number features of the head are given `content' by the NP (just as the verb gives `content' to the Case: if in AGRs, nominative; if in AGRo accusative):
V AGR .......
The checking of Case and agreement occurs either overtly or covertly, depending on whether the features in the Functional Head are strong or weak. In Chomsky (1992), there are two types of features: N-features or V-features. The first type is responsible for triggering NP-movement and for checking Case; the second type for triggering V-movement and for checking agreement. Overt checking of the NP takes place in a Spec-Head relationship as in (6) before SPELL-OUT (or at s-structure in earlier frameworks); covert movement will mean that the element must wait till LF to check its features because this is `cheaper'. English is generally assumed to have weak V-features and (6) is listed as an illustration, rather than as a structure for English:
Zora AGR XP
V AGR .......
A number of principles regulate whether overt or covert movement occurs, namely `greed' and `procrastinate'. `Greed' says that elements only move because these elements need to check features; they do not move to `help out' another element. This is where a major problem occurs with strong features. Strong features are illicit at SPELL-OUT and need to be checked by an NP or V, but the latter elements only move out of self-interest. Wilder & _avar (1994) call this the problem of `early altruism'. In `later' Minimalism, i.e. Chomsky (1995), movement is reformulated as Attraction. Features and lexical items are attracted to a higher functional category and Greed and Procrastinate no longer seem to play the same role.
Other functional categories are introduced. Thus, T(ense)P accompanied by V- and N-features is also included in (7), which is a typical tree structure. Categories such as ASP(ect)P, VoiceP, Perf(ect)P, Num(ber)P, PersonP and others are also possible (cf. Rivero 1990; Marácz 1991). A debate occurs as to the universality of these (Iatridou 1991; van Gelderen 1993; 1996a):
I have argued elsewhere (van Gelderen 1993) that there is no evidence that English has an AGRsP as well as a T(ense)P and this argument still holds. Chomsky (1995: 349ff.) argues, in a similar vein, that there is no direct evidence for AGRs and AGRo in English. Agreement, tense and Case features would be checked through Head-Head and Spec-Head agreement in the IP in (3). Thus, the tense and agreement features are not necessarily connected to one particular functional head and that is the reason not all projections need be present. For the purposes of this book, I again assume that there is only one functional position and following Chomsky (1995), I call it IP. However, whether one or two functional categories between C and V actually occur in English is not relevant to this work. I do assume AGRo for reasons outlined below.
0.2.3 `Later' Minimalism
In Chomsky (1995), the following modifications relevant to agreement, features and functional categories occur: (a) Features are seen as intrinsic or optional; and as Interpretable or non-Interpretable, (b) The V- and N-features are reformulated as categorial features (only D-features are actually discussed), and Case and phi-features are added, (c) All movement at LF (i.e. covert), caused by weak features of a target, is replaced by feature movement, (d) As mentioned, neither AGRs nor AGRo occur, and (e) Move is replaced by Attract. I will elaborate on each of these changes.
There are many types of features: semantic (e.g. abstract object), phonological (e.g. the sounds), and formal (Chomsky 1995: 230ff; 236; 277ff). The formal ones are relevant to syntax and are divided into intrinsic or optional; and into Interpretable or non-Interpretable. The intrinsic ones are "listed explicitly in the lexical entry or strictly determined by properties so listed" (Chomsky 1995: 231) and include categorial features, the Case assigning features of the verb, and the person and gender features of the noun. Optional features are added arbitrarily and are predictable from UG Principles (e.g. nouns need Case). They include the tense and agreement features of verbs and the number and Case features of NPs.
The "much more important distinction" (p. 277) is that between interpretable and non-Interpretable. The Interpretable ones are relevant for interpretation at LF and include categorial and nominal phi-features. They are not deleted or erased after they are checked because they are relevant to the interpretative component. Non-Interpretable features are deleted and they involve the Case features of NPs and verbs and the phi-features of verbs. There are a number of reasons behind the distinction. Some features (e.g. phi-features of NPs) remain visible after checking and hence cannot be deleted. This is the reason an NP can move cyclically and provide the phi-features along the way (Chomsky 1995: 282). This is not true for the non-Interpretable Case. Once Case has been checked by an NP that same NP cannot move to check Case elsewhere.
The reason for abandoning V- and N-features in favor of categorial, Case and phi-features is that the Extended Projection Principle effects (i.e. that clauses have structural subjects) are accounted for by means of a strong D-feature in I (Chomsky 1995: 232). This is necessary since Minimalist trees do not automatically project a Specifier position and Spec IP must somehow be present. Hence, the assumption that a D-feature exists in I. Wh-movement is triggered by a D-feature in C rather than by a [wh]-feature. The argument that it is D-features that trigger movement comes from expletives. If the expletive there in (8) was present to check the Case features, the Case features of the postverbal five coyotes would not be attracted. As a result, the non-Interpretable Case features of the NP would remain unchecked and the sentence would not be well-formed (the term used for well-formedness is convergence):
8. There are five coyotes in our backyard.
If the expletive was present to check the phi-features, the Interpretable plural phi-features of the noun would not be attracted to I(nflection) and again, (8) would not converge. Since (8) is grammatical, there is only inserted to check the categorial features. This will be worked out in chapter 6. In chapters 7 and 8, I indicate that languages may vary as to what features they check.
NP-movement is deemed not to be economical since it is only the feature of the target that needs to be checked and NP-movement at LF is therefore replaced by feature movement (Chomsky 1995: 261ff). Feature movement will sometimes result in non-convergence at PF and then full NP-movement (pied piping) is required. "Just how broadly considerations of PF-convergence might extend is unclear, pending better understanding of morphology and the internal structure of phrases" (Idem: 264). Since what counts as PF-convergence is unclear, I mainly continue to use the (descriptive) phrases overt and covert movement, rather than pied piping and feature movement.
Chomsky notes a difference between functional categories such as D(eterminer), I(nflection), or T(ense) and C(omplementizer) on the one hand and AGR(eement with the)s(ubject) and AGR(eement with the)o(bject) on the other hand. AGR lacks phi- and Case-features since V and T provide the latter by moving to AGR. In addition, AGR never has lexical content, unlike D, I and C. The evidence that AGRs and T both exist comes from the analysis for an Icelandic construction. However, as I have argued in van Gelderen (1993), in most languages there is no evidence for AGRs. Overt object raising as in French (cf. chapter 2 below) and possibly in English (cf. chapter 7 below) provides structural evidence for AGRo. Chomsky argues that the Case to the object features can be present in the light verb v in (9). A similar construction is used in Larson (1988) to account for sentences with three arguments and is adopted by Chomsky for all VPs (cf. Chomsky 1995: 352). I will continue to assume AGRo since that is common in the literature but agree that there is a difference between D, I and C on the one hand and AGR on the other because the former but not the latter are lexicalized in many languages, for instance, the, to and that respectively (cf. van Gelderen 1996a).
As a summary I provide an instance of how the features in a Minimalist system work. In (9), an NP (with Interpretable categorial and phi-features, but with non-Interpretable Case) moves to Spec IP overtly to check the strong categorial feature ([D]) in I. The phi-features and Case move along and are checked in due course. For English, assuming the, non-listed, categorial V-features are weak, this is the only overt movement necessary. The phi-features of the verb or Auxiliary are attracted to I at LF. The features of the object NP move to the specifier of v and the main verb adjoins to v. The result is that all the non-Interpretable features are checked:
[D] . .
[Case] v VP
[D] NP V'
[Case] [phi] [phi]
[phi] [Case] [Case]
In (9), the Interpretable features are the phi-features of the NPs; the non-Interpretable ones are the features of the targets (C, T and v), the Case features of the NPs and verbs, and the phi-features of the verb.
There are a number of matters not discussed in Chomsky (1995). Even though categorial features are introduced, it is only the category of the noun, i.e. the D-feature, that is addressed. Since categorial features are the only ones that can be strong and trigger overt movement, I argue that V-features must be present as well. Otherwise, overt Verb-movement would not take place, as it does in a number of languages. The status of weak categorial features is not clear. Since categorial features are Interpretable, they need not be checked. Thus, weak categorial features might as well not exist since they will never be checked, nor will they trigger movement.
There is a certain redundancy about the features. In Chomsky (1981), NPs move to satisfy the Case Filter; in Chomsky (1986a), they must be Case marked in order to be visible for theta-marking; in Chomsky (1995), NPs move to check the categorial features, but the Case features must still be checked. Only in the case of expletives (as explained above) is there a reason to postulate both categorial and Case features in I. When features appear in `early' Minimalism, they trigger movement. In the present model, most of the features are abstract and most movement takes place at LF. The child acquiring her or his grammar might not find much direct evidence and this is problematic.
0.3 Lack of Agreement
In this section, I discuss the reasons given in this book for the `breakdown' of agreement. The book is not organized according to these four. Instead, each chapter examines a different set of constructions.
A Lack of Spec-Head agreement. It has been argued that `older' languages lack (some) functional categories (cf. Kornfilt 1991 for Old Turkic; Lenerz 1985 for Old High German; Abraham 1993 and 1995 for Modern German; Kiparsky 1995 for Old English and van Gelderen 1993 for Old, Early Middle English and Dutch). If this is true, checking of agreement and Case cannot be under Spec-Head agreement but must take place under government related to theta-marking. The question arises immediately as to why `older' languages display more Case and agreement. For Case, this can be accounted for by assuming Case is very much related to the thematic structure of the verb. For instance, a Goal may get dative and a Theme accusative. There is evidence that some Case in Old English is assigned/checked in this fashion (Mitchell & Robinson 1986: 105ff; van Gelderen 1986b), i.e. that Case is inherent rather than structural. This means government is relevant rather than Spec-Head agreement (cf. den Besten 1983 for Dutch). I argue that the agreement, i.e. phi, features are also checked under government.
In `modern' languages such as Dutch and Middle English, Case assignment under government is not as suited for the transmission of agreement features and a `breakdown' of agreement is often the result. In (10), from the 15th century, the singular verb precedes the plural subject and in Dutch, Verb-Subject structures also display less agreement. In (11), the second person -t is missing, which can be seen by comparing (11) to a Subject-Verb structure as in (12):
10. Mandeville's Travels, 71-18
In that cytee was the sittynges of the .xij. tribes of Israel.
11. Veeg jij de vloer even,
Wipe-S you the floor PART
`Would you wipe the floor'.
12. Jij veegt de vloer vaak,
You wipe-2S the floor often
`You often wipe the floor'.
This absence of Spec-Head agreement will be discussed in chapters 3, 4 and 5. It is also relevant to Case checking since, in many languages, Case is checked under government, as in (13), where the subject deze boeken `these books' remains inside the VP:
13. dat mijn oom gisteren deze boeken toegestuurd zijn,
that my uncle-DAT yesterday these books-NOM sent are
`that these books were sent to my uncle yesterday'.
I will not examine the oldest forms in detail but will focus instead on when the switch occurs from a system where government plays a prominent role to one in which Spec-Head becomes operative. Kiparsky (1994) calls this change "the rise of positional licensing". I claim this happens when expletives are introduced. After the introduction of functional categories e.g. in English at the time of Middle English (cf. van Gelderen 1993), the possibility of Spec-Head agreement arises or increases, since it is available in Universal Grammar, in addition to the head-complement relationship. There are still some remnants of Case assignment under government. These are often instances where agreement is `incomplete'.
There are some interesting constructions in Diné, O'odham and Hopi where number is checked under Spec-Head agreement as well as under government. These will be discussed in chapter 10, as will some cases of wh-checking under government in chapter 4.
B C/overt movement and expletives. A language where government is said to bring about different agreement from Spec-Head relationships is Arabic. Koopman and Sportiche (1991) discuss cases where a subject and a verb do not agree (completely) in a VS order as (14) shows, as opposed to the VS-construction in (15):
14. Darab-at/*-na l-banaat-u Zayd-an,
hit-FS the girls Zayd
`The girls hit Zayd'.
15. al-banaat-u Darab-na/*-at Zayd-an,
the girls hit-FP Zayd
It is, however, hard for Koopman & Sportiche to account for the fact that the verb agrees in gender but not in number. Therefore, rather than arguing that the difference in Arabic is caused by the difference between government by the V of the subject and a Spec-Head relationship between the subject and the verb, I will argue that agreement can be `deficient' if NP-movement is covert because the checking of the overtly moved verb will be with an expletive (inserted to check the strong categorial features).
Throughout the book, I argue that expletives vary from language to language. Hence, the term `expletive' is a cover-term for an element `deficient' in features. For instance, in Arabic, expletives have categorial and number features (cf. chapter 1); in English, there has categorial features (chapter 6) whereas it has categorial as well as person and number features (chapter 7); in Dutch, het `it' is unspecified for number but is specified for person and gender (chapter 8); and in Middle English, as in (16), an it unspecified for person features occurs:
16. Chaucer, The Miller's Tale I, 3766
What, who artow? It am I Absalon.
In chapter 2, object expletives are argued to occur in e.g. Swedish (18) but not in (17) where the NP moves to a preverbal position:
17. ?Det blev tre bilder målade,
There were three pictures painted-P
18. Det blev målat tre bilder i söndags,
There were painted-S three pictures on Sunday
In French, object expletives have categorial as well person and number phi-features. As a result of the varying nature of the expletives, the agreement will appear `aberrant' but is in fact regular.
C Impact from grammaticalizing processes. Many of the constructions that I examine contain as their subjects, or as part of their subjects, elements that have undergone a process resembling grammaticalization (cf. Bopp 1816; von der Gabelentz 1891; and more recently Heine 1984; Lehmann 1985 etc.), e.g. that in chapter 4, there in chapters 5 and 6, men in chapter 9 and and and with in chapter 10. These elements start out as lexical items, specific in meaning, but acquire a much more general lexical meaning and/or a more grammatical function. One could say they lose phi-features and change categorial features. In van Gelderen (1993), I suggest that some grammaticalization results in a reanalysis of these elements as Functional Categories (see also Roberts 1993 and Haspelmath 1993). For instance, to is exclusively a preposition of location in Old English but, I argue, becomes an auxiliary by Middle English. In Minimalism, the emphasis is on feature checking and grammaticalization can be seen as an anchoring of categorial features onto one lexical item (cf. van Gelderen 1996a). The features that are anchored with the categorial ones are the verbal (non-)tense features in Middle English that become associated with to. This is true with other verbal categories: have is associated with verbal perfect features and be with passive and progressive ones. When nominal elements such as there and that grammaticalize, they lose phi-features but not the categorial D-features.
In some of the chapters below, I examine ways in which nominals (men), pronominals (there), demonstratives (that, there) and relative elements (that, there) grammaticalize and how they are reanalyzed. For instance, there starts out as a demonstrative and becomes a relative and indefinite pronoun. I will argue (in chapters 4 and 5) that features play an important role. When a lexical item becomes less lexical and more `grammatical', it loses most of its phi-features. For instance, in (19), that brings about third person singular on loveth, but in the more modern (20), this is not the case (sentence (20) is slightly awkward because pronouns are not modified by restrictive relatives, but the agreement is clearly first singular):
19. Chaucer, Knight's Tale 1736-7
and it am I That loveth so hote Emelye the brighte.
20. I that am/*is going away still need to do a lot of work.
The grammaticalization of men is unlike that of other elements in that it goes back and forth between singular indefinite as in (22) and plural lexical item as in (21):
21. Layamon's Brut, Caligula 6869
for men hit sæiden wel iwhær,
because men it said-P nearly everywhere
`because it was said almost everywhere'.
22. Idem, Otho
for men hit saide wel i-war,
because men it said-S nearly everywhere
D Structural Configurations. Coordinate constructions are intransparent (or opaque) for agreement and Case because of the complexity of the structure. I indicate how, in an asymmetrical structure as in Munn (1992), one of the two NPs as in (23) is in a privileged position and its Case and phi-features are attracted to I(nflection). In (23), the Diuell's Case and phi-features are attracted to C and the verb agrees with this NP rather than with thee (which is accusative):
23. Shakespeare Henry IV, 1 1.2.126 (Jespersen 1913: 175)
How agrees the Diuell and thee about thy soule?
This accounts for `breakdowns' in Case and agreement. Since in English and renders an NP plural (in the same way a negative element renders an NP negative), this asymmetry has no consequences for agreement. In other languages, it does. Coordinate structures are very often ambiguous between and, with and after either being a preposition or a coordinating conjunction or a subordinating conjunction. If they are coordinating conjunctions, the agreement on the verb will be plural; otherwise, it will be singular.
In this section, I discuss issues relating to features and outline the changes I propose in the theory. The present study indicates that Case and agreement features are separate (cf. also Hulk and van Kemenade 1991; and Kosmeijer 1993), and that agreement, i.e. the phi-features, is (at least) divided into person and number.
(I) I will be assuming, following Chomsky (1992; 1995), that items are selected from the lexicon with their morphological features fully spelled out. (II) I will be making a number of proposals concerning the representation of features. Chomsky (1992; 1995) speaks of phi-features that a verb checks. I will argue that in fact person and number must be checked separately. I also assume that gender can be part of the phi-features (in Arabic, French and Dutch) but since phi-features are Interpretable, they need not be checked. However, not every feature (person, number, gender) has its own Functional Category (cf. van Gelderen 1993 for a similar argument concerning tense and agreement) but bundles of features are grouped together. In addition, languages differ as to the (phi)-features that are used. For instance, English does not have dual or gender. (III) Since languages differ as to the realization of features, I propose a number of parameters.
I Lexical or derivational. One may argue that items appear unspecified and get their specifications in a particular position. These specification are then spelled out later. This fits Baker's (1988: 13) Mirror Principle which says that "[m]orphological derivations must directly reflect syntactic derivations (and vice versa)". Similarly, Lumsden (1987: 28) argues that in the "underlying representation, syntactic features are underspecified". An early version of underspecification is assumed in van Riemsdijk (1983: 239ff.): "[t]hus a matrix of the type [+N, -V] is gradually built up to a full morphosyntactic syntax" and the morphological features are contained in "a kind of generalized position" (p. 240) AG(reement) which may be a sister to an N. Similarly, Lumsden (1987: 18) argues that "since the affixes of inflection signal syntactic features, they must be the heads of independent phrases in the underlying representation". This might mean that each feature F has its own projection as in (24):
N (cf. Lumsden 1987: 19)
Zwarts (1992: 33) puts it this way: "In the ideal case, every functional head has one characteristic feature". In (24), the N must move rightward to each of these heads and adjoin to the left of the affix (Lumsden 1987: 19). Lumsden argues that the morphologically complex word is built up through a picking up of the affixes. Movement is to the right since the affixes appear on the right side. Of course, this could be reformulated as movement to the left with left-adjunction of the head to F, and thus, by limiting movement to the left, constraining Universal Grammar considerably.
In Chomsky (1992; 1995), as mentioned, a lexicalist position is assumed. The head of a Functional Category contains categorial and Case features and the NP and V (taken fully inflected from the lexicon) check these features. If the categorial D-features are strong, the NP moves (or is attracted) into the Specifier position of the Functional Projection and the verb adjoins to the Head position. The features causing movement are abstract: strong does not mean that the element is overtly marked morphologically. I will assume that the insight that movement occurs because of having to `pick' up features is basically correct and is, in earlier work, seen as the determining factor behind Verb-second, but overt morphology is not linked to strong features (since languages such as Swedish and Afrikaans show minimal verbal agreement but display overt Verb-second). The most important aspect of features for Chomsky (1992) as well as (1995) is to regulate movement, i.e. movement is necessary when the features are strong. The non-Interpretable features of V and NP also need to be checked explicitly.
II Inventory of Phi-features. In earlier work (e.g. Chomsky 1992), V-features trigger movement and the checking of a complex of person, number and gender features as well as finite, tense, aspect and passive ones (to name but a few). One might wonder whether or not they should each have a structural position, e.g. should there be a PersonP, Num(ber)P, GenderP, T(ense)P, ASP(ect)P, Pass(ive)P? For some languages, there might be evidence for some of these, but there would have to be evidence (cf, Iatridou 1990). In a pre-checking framework, there could be separate positions since an item moves to `pick' up inflection and Case. I previously have argued (van Gelderen 1993) that agreement and tense features can occupy the same position and hence, I continue to assume that there is no necessity to provide each feature with a Functional projection of its own. In a checking framework, a lexical item is selected from the lexicon fully inflected and hence, separate positions are not necessary. Based on evidence from Arabic, I will argue that the phi-features related to agreement are divided up into at least gender, person and number, each of which can be strong or weak, but they are clustered in one Functional Head. Belfast English shows that number and person features are separate. In chapter 7, I will briefly examine some of the other features that trigger verbs to move, e.g. tense, aspect, perfect and passive. Chomsky (1995) does not discuss categorial V-features but I assume that tense, perfect, passive features trigger V-movement and are therefore best regarded as the categorial V-features.
Forchheimer (1953), in a different framework, surveys languages and observes that almost all languages have person, but not all have number and gender. There is unclarity as to how many features English has: just number; or person and/or gender as well. Modern English shows number (singular and plural, but not dual) and person (mainly third person -s and first person with the verb to be), but not gender. Kayne (1991b) argues that "[i]f one takes you to always be grammatically plural, somewhat as French vous, despite sometimes referring to a singular, then, if one takes I to be non-singular (there is clearly nothing that I is a true singular of), English -s can be considered to be a pure indication of number (+singular), rather than involving person in any way". Historically, you is certainly plural (accusative plural) whereas thou is nominative singular; thee accusative singular and yee nominative plural. In many languages, the first and second person singular are indistinguishable from the first and second person plural. Kayne's proposal would reduce the features necessary to account for verbal agreement and would mean that in English there is only one feature: number, with I am as the exception. I do not assume Kayne's proposal because there is evidence presented in chapter 4 that both person and number are necessary.
III Feature Parameters. I assume that a derivation starts with the selection from the lexicon of an NP or NPs and a verb or verbs. Some of their features will be intrinsic; others will be optional. Since some of the features are Non-Interpretable, these must be checked. This much is part of Universal Grammar of which (25) is a principle. Functional categories such as I are selected (at least in English but as will be shown in chapter 3, not necessarily in other languages) and merged with the NPs and Vs. The parameters to be provided by Universal Grammar are (26) to (30). These differ in a number of ways from Chomsky (1995):
25. Non-Interpretable features (Case and verbal phi-features) must be checked.
26. Select Functional Categories, with +strong Categorial features: I, AGRo etc.
27. Checking is through: Spec-Head, Head-Head and/or government
28. Select Categorial V-features: tense, perfect, progressive, passive, etc.
29. phi-features are: +person, +number, +gender, etc.
30. Feature x is: +independent.
Parameter (26) is the spelled out version of what appears in Chomsky (1995) but will be revised in later chapters by extending it to all features, not just categorial ones. (27) allows government as a checking relation, unlike regular Minimalism. Parameter (28) ensures that tense and perfect are connected to a position with V-features in a particular language. For instance, in English, perfect is connected with have, and be is with progressive or passive. Parameter (29) reflects that phi-features differ among languages and (30) forces a language learner to look for evidence for where they are placed, i.e. how many Functional Projections there are. In English, tense and the agreement complex are placed independently in a position usually referred to as I or T; in Dutch, tense is arguably placed in C (cf. e.g. den Besten 1983). The task will be to restrict the number of features.
0.5 Book Outline
In chapters 1 to 7, I indicate that Verb-Subject (hence VS), Verb-Object (hence VO) and V-wh structures often display a `deficient' agreement and that in some languages this is caused by the categorial features in the head of the relevant Functional Category being strong. Hence, empty and non-empty expletives are inserted to check the categorial features but they check other features at the same time. Depending on whether the verb moves overtly (Arabic and earlier English) or covertly, the phi-features of the Functional Head will be checked with the expletive or with the postverbal NP respectively. In chapter 1, I examine cases of VS that have often been seen as instances where the verb governs the Subject. I argue that empty expletives are present. The same is true in chapter 2 when expletives occur in object position. In chapter 3, I indicate that the loss of agreement can occur because of the checking of Case-features under government and the checking of phi-features against C. This happens in Dutch and earlier forms of English. In chapter 4, I argue phi-features are also present in C in certain dialects of English. In chapters 5 and 6, I discuss overt expletives such as there in various stages of English. Sometimes, there is an expletive with fewer phi-features; other times, it is an argument with complete phi-features. In chapter 7, I argue that some instances of it in English are expletives; and, in chapter 8, that other its are arguments. The agreement in the former case can be dealt with if it originates in Spec CP as in Stroik (1991); in the latter case, it has features of its own which determine agreement.
Overt wh-movement brings about agreement more often than does wh-in-situ in English and Chamorro as I show in chapter 4. I try to account for these in the same way as for overt versus covert NP-movement. The wh-constructions in certain relative clauses show that Case and agreement (phi-features) need not both be checked in a Spec-Head agreement relationship but that one of them may be checked under government. This confirms some of the results of chapter 3. I also demonstrate that relative pronouns in English derive from demonstratives and that as a result they initially have features of their own. The demonstrative that undergoes grammaticalization and gradually loses its features. Thus, differences in agreement can be accounted for in that way.
In chapter 7, apart from showing that overt movement results in `more' agreement, I explore the question of whether it is possible to analyse Extraposition as lack of movement as in Kayne (1994). The reason for such an analysis is that eliminating rightward movement simplifies Universal Grammar.
In chapter 9, I discuss the implications of checking for those arguments that can be both plural and singular. The agreement is with the subject but since the subject loses and gains phi-features over time, there is no `breakdown'. Chapter 10 is a discussion of agreement `breakdowns' in coordinate constructions. I assume that coordinate structures are asymmetric and that, in some languages, the phi-features of the first conjunct are attracted to I. This results in `disagreement'.
My major conclusions are: (a) verbal agreement differs in languages depending on whether they select Spec-Head agreement or government for checking purposes, (b) phi-features must be divided into person, number etc., (c) expletives differ in feature content across languages (d) c/overt movement (VS versus SV and VO versus OV) is relevant because the features may have been checked by the `wrong' element (e.g. the expletive), (e) through grammaticalization elements gain and lose features relevant to agreement and may cause structural ambiguity (e.g. is with a preposition or a conjunction?), (f) coordinate structures are asymmetric and this has consequences for agreement. In chapter 11, I outline additional lines of research related to the four causes of breakdown in agreement listed in 0.3.
 From now on, I use agreement rather than concord to indicate the marking on the Verb caused by the `subject'. The reason for this is that it is presently fashionable to do so. No principle is involved. In the German literature, Kongruenz is used; in the French literature, accord; in Spanish, concordancia, etc.
 For Inuktitut, see Bok-Bennema (1991); for Dyirbal, see Dixon (1972); for Urdu/Hindi, see Mahajan (1990) and Butt (1993); for Georgian, see Harris (1981).
 Cf. Young and Morgan (1987).
 This term is used for identifying the model of Chomsky (1981). It is, as Chomsky often notes, an unfortunate habit to name a framework for some of its technical characteristics. I will nevertheless.
 Cf. also Abraham (1993; 1995); Thráinsson (1996), etc.
 As in Chomsky (1995: 349), I assume that Functional Categories do not have phi-features but that I has Case (when finite).
In (9), Chomsky allows for the subject being in Spec vP rather than being in Spec VP.
 In Chomsky (1995), inherent Case is seen as an interpretable feature which does not need to be checked. It is not further worked out, however.
 However, in van Gelderen (1992b and 1996b), I have indicated that the opposite might be argued: OV occurs if Case is overtly marked. If one equates morphological strength with strong (functional) features, some of the puzzles regarding VS structures are solved. Pollock (1993: 35) also argues that only morphologically identified functional heads can be checked overtly.
 Chinese Pidgin English may be an exception, but according to Forchheimer, the data is unreliable.