Subjects, nominative case, agreement and functional heads

Subjects, nominative case, agreement and functional heads

Lingua 89 (1993) 181-215. 181 North-Holland Subjects, nominative case, agreement and functional heads* Aafke Hulk” and Ans van Kemenadeb a Departm...

2MB Sizes 0 Downloads 25 Views

Lingua

89 (1993) 181-215.

181

North-Holland

Subjects, nominative case, agreement and functional heads* Aafke Hulk” and Ans van Kemenadeb a Department of French, University of Amsterdam. Spuistraat 134. 1012 VB Amsterdam, The Netherlands b Department

of English, Free

University. De Boelelaan

1105.

1081 HV

Amsterdam,

The

Netherlands

1. Introduction The aim of this article is twofold: first we want to develop a theory for the licensing of the functional properties case and agreement in governmentbinding theory; secondly, we want to show that this theory specifically accounts for a wide range of phenomena in the Germanic and Romance languages, both modern and historical, synchronic and diachronic. Our starting point is that the functional features case and agreement of any category, lexical or functional, need to be licensed. We will discuss the principles involved in detail, following up Chomsky (1986, 1988) Pollock (1989) and subsequent work. The theoretical proposal that we make elaborates three issues: first, we shall suggest a distinction between C-oriented and I-oriented languages with concomitant typological distinctions. Second, we shall eliminate the option of spec-head agreement from current definitions of nominative case marking and suggest that nominative case can be assigned exclusively by C or I under government. Third, we will propose that ‘subjectverb agreement’ is a separate licensing operation, which constitutes one of the options for licensing INFL features for person, number and gender. The empirical result of our proposal is a unified treatment of the interaction between the V2 phenomenon, various types of case properties and

* We thank Jaqueline Gueron and Barbara Vance for helpful comments on an earlier draft, Halldor Sigurasson for a discussion of Icelandic and for his native speaker judgements, and Martin Everaert for helping us to find the literature we needed on Icelandic.

0024-3841/93/$06.00

0

1993 -

Elsevier Science Publishers

B.V. All rights reserved

182

A. Hulk, A. van Kemenade / Subjects and agreemenf

various types languages.

of pro-drop

and

agreement

in the Germanic

and

Romance

2. Verb second and C versus I-orientation In this section, we suggest a typology of languages, in which they can be either C-oriented or I-oriented. One of the functional projections C or I is dominant. We propose (1) as a licensing condition for functional heads. l (1) The dominant

functional

head must be lexicalized.

Let us first consider how we view V2. We assume, following current work, that V2 takes place regardless of the basic sentence structure, and fronts the verb (Vf) to presentential position in all types of root clause: declarative, interrogative, imperative, as in (2): (2) Gisteren heeft Jan hem het boek gegeven yesterday has John him the book given ‘Yesterday John gave him the book’

(Dutch)

Dutch is a V-final language, witness the position of the participle in (2). In root clauses such as (2) some XP is moved to a sentence-initial satellite position, followed by the preposed Vf. We analyse this as in the structure (3) adapted on the basis of Chomsky (1986) and Sportiche (1988), modulo VP internal and IP internal word order: 2

1

Sometimes

restricted

a non-root

however,

C position

to environments

can remain

lexically

null in C-dominant

where the complementizer

languages.

is selected by the matrix

This is verb, cf.

Rizzi and Roberts (1989). Also, in indirect questions, the C position may count as lexical if spec. C is filled by a wh constituent, as in I wns wondering /r,,ec,C. who Lc e Irp u,e should turn mjjj. Presumably, spec,C’ and C are coindexed here, in the vein of Pesetsky (1982). 2 (3) represents I-medial order and VO order in the VP. Presumably in Dutch and V-final VP.

we find l-final IP

A. Hulk, A. van Kemenade / Subjects and agreement

183

(3) A specC A C

I I, spec1’ A 1[

I

A

specV ext.arg.

V/I to c

A V

. ..

v to I

One of the core characteristics of V2 in this sense is the asymmetry between root and non-root clauses with respect to verb-fronting. Vf/I is fronted to C in root clauses only, cf. early work on V2 in Den Besten (1989). This is the main reason why V2 is widely analysed as movement of Vf/I to C, which is blocked in non-root clauses by the presence of a base-generated complementizer. We assume that the external argument is base-generated in spec,VP and may move to spec,IP under conditions to be defined below. Thus, in V2 languages, Vf in non-root clauses can be moved to I, but never to C. Evidence for the root/non-root asymmetry is the fact that apparent embedded V2 with topicalization takes place only after bridge verbs, as exemplified for modern Dutch in (4): (4a) Hij zei hij heeft hem gisteren gezien he said he has him yesterday seen (4b) Hij zei gisteren heeft hij hem gezien he said yesterday has he him seen (4c) Hij zei dat hij hem gisteren gezien heeft he said that he him yesterday seen has (4d) Hij zei dat gisteren heeft hij hem gezien he said that yesterday has he him seen (4e) *Hij zei dat heeft hij hem gisteren gezien he said that has he him yesterday seen ‘he said he saw him yesterday’

A. Hulk, A. van Kemenade / Subjects and agreement

184

After bridge verbs, embedded clauses. For further discussion,

clauses as in (4) are to be analysed see e.g. Weerman (1989).

as root

In terms of the structure (3), V2 languages (the continental Scandinavian languages, Modern German, Modern Dutch, Old and Early Middle English, Old French) are analysed as follows: V2 languages have movement of Vf to I, and subsequent movement of V/I to C in root clauses, subject to the Head Movement Constraint and triggered by the necessity for C to be lexically realized, by principle (1). In root clauses C is filled by moving V/I; in nonroot clauses by a base-generated complementizer. The notion that C needs to be lexically realized is found in various guises in recent proposals for the analysis of V2. 3 We shall be referring to V2 languages of the sort discussed so far as CV2 languages. Crucial for this notion is a root/non-root asymmetry for v2. Non-V2 languages (ModE and ModFr) are characterized by movement of Vf to I (or vice versa4), and no further movement to C, except under restricted circumstances such as in ModE interrogative contexts and ModFr subject-clitic-inversion and complex inversion constructions. Rizzi (1990) calls these ‘residual V2’ as opposed to ‘full V2’. Strikingly, V/I to C movement in these contexts exhibits the same root/non-root asymmetry as verb movement in CV2 languages. We come back to this below. In languages that are not CV2, we still find V2 word orders. On the assumption that in the structure (3) above, the spec, I’ position can be an XP position rather than an NP (subject) position as in the standard analysis, an initial XP position followed by Vf may reflect an IP rather than a CP. This is discussed in Santorini (1989) and Diesing (1990) with respect to Yiddish, and similar proposals have been made with respect to Modern Icelandic (most recently Rognvaldsson and Thrainsson 1990). Crucial is though, that in such instances there is no root/non-root asymmetry. We shall be referring to this type of language as an IV2 language. In such languages, non-interrogative XP-Vf-subject sequences occur in both root and non-root contexts. How-

3

For instance,

Evers (1981, 1982) proposes

that TENSE

is an operator

be lexicalized; Koopman (1984) proposes that lexicalization is triggered assignment; De Haan and Weerman (1986) suggest that COMP (containing

in COMP

and needs to

for nominative case AGR) has categorial

features and must therefore be lexical; Weerman (1989) assumes that C assigns a modal role to the proposition, and that C needs to be lexical to be visible as an assigner of modal role. 4 For the moment we do not wish to go into the details of the relation between V and I in these languages. At least for ModE there is some evidence, e.g. from Neg placement facts (see Pollock 1989, Roberts that in ModE

1985) that V does not move to I. These facts also underlie the pre-Barriers I is lowered to V, see e.g. Lasnik (1981). We will come back to this below.

idea

A. Hulk, A. van Kemenade / Subjects and agreement

185

ever, these are to be analysed as instances of V to I. The asymmetry between root and non-root contexts, which reflects movement of V/I to C, can only emerge in a CV2 language (or in w&contexts as we shall see below). The term V2 as applied to IV2 languages is to some extent misleading, since it wrongly suggests that V2 applies as an absolute word order constraint. This does not seem to be the case in IV2 languages. Consider the following examples from Icelandic, where the word order XP-adverb-Vfis found in both root and nonroot clauses : (5) Icelandic

(a) Hann einfaldlega kann ekkert (root; Thrainsson 1986) he simply knows nothing ‘he simply knows nothing’ (b) hegar Cg loksins fann lagib (adverbial clause; Maling 1990) when I finally found tune ‘when I finally found the tune’ (c) hegar mer allt i einu datt i hug aa . . . (ibid.) when me suddenly occurred that ‘when it suddenly occurred to me that . . .’ Thus, although many instances of the word order XP-V’ . . . occur in IV2 languages, the more remarkable property of these languages is not this word order, since other constituents can intervene, but rather the fact that spec,IP is available for constituents other than the subject. We will come back to the status of the subject position in these languages below. From the point of view of lexicalization of functional heads, the essential difference between CV2 (CP) languages on the one hand and non-V2 and IV2 (IP) languages on the other then is that in CV2 languages COMP is always lexically realized, whereas this is not the case in IP languages. We take it that this is a reflection of the fact that in CV2 languages C has a number of crucial head properties that in non-V2 language are characteristic of I. For instance, in a number of CV2 languages we find morphological agreement between COMP and the finite verb. s Also, in West-Flemish, COMP is a head that

5 Consider for instance the following Southern Dutch sentences (from Den Besten 1989). For related facts in German, see Tomaselli (1990). (ia) date Jan en Kees morgen komen zullen that+pl John and Bill tomorrow shall +pl come ‘that John and Bill will come tomorrow’

A. Hulk, A. van Kemenade / Subjects and agreemew

186

attracts clitics, cf. Bennis and Haegeman (1984).6 In our view, the distinction between C and I languages as drawn above reflects a deeper distinction between these languages. In CV2 languages, C is in some sense a more prominent head than I, whereas in non-V2 languages the reverse seems to be the case. We will henceforth call this C-orientation (CV2 languages) and I-orientation (non-V2 languages, and IV2 languages, i.e. languages that have no root/non-root asymmetry for V2) (cf. also Hulk and Van Kemenade 1988; Tomaselli 1990).7 We suggest the licensing condition (1) above for functional projections. (1) is repeated here for convenience: (1) The dominant

functional

head must be lexicalized.

Lexicalization can proceed through V movment to I in I-oriented languages; in C-oriented languages C is lexicalized by V/I to C movement in root clauses, and by a base-generated complementizer in non-root clauses.s It seems plausible that (1) is the motivating force behind movement of V and I in the way indicated. This approach subsumes a variety of motivations that have been suggested in the literature as the trigger for lexicalization of C, cf. note 2. We remain neutral on the question whether it is any particular functional feature of C that must be lexicalized; tense, AGR, assignment of nominative case, etc. and simply assume that the dominant head must be lexical in order to be visible as a head bearing functional features.

(ib)

dat(*e)

Jan

morgen

zal

komen

that(* + pl) John tomorrow shall -pi ‘that John will come tomorrow’ These examples 6

show that C has number

Bennis and Haegeman

(1984) propose

On the basis of this, an anonymous role in our treatment

compelling, see e.g. Zwart (1992). ’ The distinction we make between distinction

between

strong

and weak

agreement

with Vf.

that subject

reviewer

here. However,

come

pronouns

suggested

the analysis C-oriented functional

cliticized

that W-Flemish

of W-Flemish and

l-oriented

projections

to C license pro-drop. ought

as pro-drop may

as drawn

to play a central seems in no way

seem

similar

in Chomsky

to

the

(1988) and

Pollock (1989). This is an independent phenomenon though; to illustrate, Modern French and Modern English are, in our view, both I-oriented languages, as neither of them have I to C movement in declaratives. French however has a strong AGR (V to AGR), whereas English has a weak AGR

(AGR

to V).

* Or in non-root clauses with a selected complementizer, being coindexed with a wh-element in spec,CP, cf. note

C can remain

1.

empty,

or be lexical by

A. Hulk, A. van Kemenade / Subjects and agreement

187

2. I. Mixed languages

In the previous section, a distinction between C-orientation and I-orientation was suggested. In this typology, Modern English and Modern French are I-oriented languages, as they do not have I to C movement in declaratives. This suggests that under these conditions, the C-level is not involved. Interestingly, in precisely those contexts where the C-level is involved, viz. in interrogative contexts, and Modern English constructions with a preposed negative constituent, we find typical CV2 properties; I to C movement with do-support in English and complex inversion and subject-clitic inversion in French, both crucially with asymmetry between root and non-root clauses. It seems then, that once the CP-level is activated, i.e. when there is an operator in specC’, C must be lexicalized.g We will come back to this below. A different kind of mixing is found in Modern Icelandic. Icelandic has been widely analysed as a CV2 language to the extent that XP-Vf-subject orders are found that seem to be characteristic of CV2 properties. However, on closer inspection, it turns out that matters seem to be more complicated. According to Thrainsson (1986) there is a clear root-non-root asymmetry with respect to the position of Vf only in wh-contexts. (7a) Hvern hefur Maria kysst whom (A) has Mary (N) kissed (7b) Eg veit ekki hvern Maria hefur kysst I know not whom (A) Mary (N) has kissed (7~) *Eg veit ekki hvern hefur Maria kysst I know not whom (A) has Mary (N) kysst ‘I don’t know whom Mary has kissed’ Riignvaldsson and Thrainsson (1990) give examples of embedded topicalization after non-bridge verbs: (8a) Jon efast urn ab a morgun fari Maria snemma a faetur up John doubts that tomorrow get Mary early ‘John doubts that Mary will get up early tomorrow’ (8b) Jon harmar aa hessa bok skuli eg hafa lesib John regrets that this book shall I have read ‘John regrets that I have read this book’ p Unless the operator is filled by the subject. of this that is compatible with ours.

See Koopman

(1984); Rizzi (1990) for a treatment

188

A. Hulk, A. van Kemenade 1 Subjects and agreement

To add to this, Sigur6sson (1989: 12) observes: ‘However, adapting GCA [Generalized Comp Analysis, i.e. the idea of analysing Icelandic as CV2 in our sense (H/vK)] to Icelandic is not as simple and straightforward as all this might seem to indicate. The principal reason for that is that Icelandic is generally “V2” not only in main clauses but also in subordinate clauses (with the exception of w/z-clauses)‘. A core characteristic of CV2 languages is an asymmetry between root and non-root clauses with respect to the relative position of Vf and the negative element, and this asymmetry, too, is lacking in Icelandic, as the following examples from Sigurbsson (1989) show: (9a) Maria Zes ekki baekur (Icelandic) Mary reads not books (9b) . . . aa Maria les ekki baekur that Mary reads not books (10a) Maria laser inte backer (Swedish) Mary reads not books backer (lob) . . . att Maria inte her that Mary not reads books Examples like (10) are usually adduced to support the root/non-root asymmetry for CV2 languages like Swedish. Note however, that the asymmetry is lacking in Icelandic. We therefore suggest that Icelandic is, like Yiddish, an IV2 language, but a CV2 language in w/r-contexts. The difference between such languages and French and English is that for some reason, SpecI’ is an XP position in Icelandic and Yiddish, whereas it is obligatorily the position for the subject NP in Modern English and French. We will suggest an explanation for this distinction below. Icelandic is like Modern English and Modern French in that in interrogative contexts, i.e. where there is an operator in specC’, the CV2 characteristic is invoked, leading unequivocally to I to C movement. whith a concomitant root/non-root asymmetry. So far, we have not discussed why it should be that in I-oriented languages like English, French and Icelandic, an operator-type of first constituent, such as a wh-element or (in English) a negative constituent, should invoke the COMP level. A possible answer to this question is suggested by the w+zcriterion as formulated in Rizzi (1990) to account for ‘residual’ V2 phenomena such as ModE subject-verb inversion and ModFr complex inversion and subject-clitic inversion :

A. Hulk, A. van Kemenade / Subjects and agreement

189

(11) wh-criterion: Principle A: Each [ + wh] X must be in a Specifier-Head relation with a wh-phrase. Principle B: Each wh-phrase must be in a Specifier-Head relation with a [+wh] X. Once a wh-phrase moves to specC’, the head C must be lexicalized in order to satisfy principle B of the wh-criterion. It must be noted that French differs from English (and Icelandic, as we see it) in that I to C movement is not obligatory when spec,CP is filled in the syntax, cf. the grammaticality of (12) with non-inversion : (12) A qui tu as parlt to who you have spoken ‘who did you speak to?’ This is a problem for any account that seeks to derive I to C movement from independent principles. Rizzi (1990) suggests that French has an additional way of licensing a [ + wh] specification under agreement in CP. We leave this for further research.

3. Principles of licensing In this section we develop a theory for the licensing of functional properties. By this we mean: the identification of functional heads such as C and I; case-marking properties of both lexical and functional heads; subject-verb agreement. Apart from condition (4), which requires lexicalization of the dominant functional head, we assume the following condition for the functional properties of both lexical and functional heads: (13) Functional features must be licensed. Functional features such as case and agreement must be identified structurally or morphologically. lo Let us consider in some more detail how (13) is instantiated.

lo

We leave tense out of consideration

here.

A. Hulk, A. van Kemenade / Subjects and agreement

190

3.1. NP licensing; case

NP is typically licensed by being case-marked. Case on NP is realized either syntactically, i.e. structurally, or morphologically. We concentrate on structural case (nominative and objective, i.e. accusative). We assume the configuration (14) for structural case assignment: (14) Structural case XP spec X’

In (14), a case-marking head X assigns structural case under government to the dependent NP, either to the left or to the right. In this article we are primarily concerned with nominative case. Nominative case is considered a structural case in all the languages under discussion, and is thus an instantiation of the structure (14).l’ In what follows, we want to defend the definition (15) for nominative case marking: (15) NP + nominative iff it is governed by the dominant head C or I l2 where government is defined under strict c-command. As a result, there are two ways of assigning nominative case in the structure (2): C assigns nominative case to the right to NP in spec,IP, or I assigns nominative case to the right (or to the left, in I-final Dutch), to the subject base-generated in spec,VP. Observe that a consequence of (15) is that in ‘ordinary’ declarative sentences with SVO surface order in a non-V2 language, both the subject and its nominative case are chain-related with the spec,VP position, as in (16): I1

Nominative

inherently

case on French

nominative.

subject

We might assume,

clitics may be an exception, following

Rizzi and Roberts

as the clitic seems to be (1989)

that a subject clitic

is associated with nominative case by means of its incorporation in I”. I2 Up to Chomsky (1981), it was assumed that the feature [ +TNS] constitutes the element assigning nominative case under government. In Chomsky (1981) [+TNS] was abandoned in favour of AGR as nominative marker. In the literature it has been pointed out frequently that either option is problematic for some languages. Roberts (1990) suggests that the choice of the head relevant for nominative case assignment may be a parameter: in some languages it is TNS, in others AGR (where both head a functional projection). Since it is not crucial for our argument, we leave this question

open for further

research.

A. Hulk, A. van Kemenade / Subjects and agreement

191

(164 Jean frappe Marie (16WLP Jean, [IW frappel[VP6 [vk[NPMarie1111 ‘Jean hits Marie’ The NP Jean is chain-related with the base-generated position in spec,VP, and thus chain-associated with nominative case. We come back to this below. The definition (15) differs from current ones. For instance, Sportiche (1988) suggests a definition where (15) is an option besides spec-head agreement between NP in spec,IP and I. In other current work, (15) is assumed with government defined under m-command, which comes down to the same thing as Sportiche’s definition. There is good reason to assume that at least one of the options for nominative case assignment is under government, contra recent proposals in terms of functional projections. l3 Below, we will discuss a number of instances of nominative case assignment to VP-internal arguments that cannot possibly be viewed as spec-head agreement, but can be reasonably analysed as an extension of nominative case marking under government. Since this option is needed anyway, it is worth exploring whether the option of spec-head agreement can be eliminated. In all current definitions, nominative case coincides with ‘subject-verb agreement’. We consider this inappropriate. We will provide arguments for this in the next subsection. Once we have established that nominative case does not always coincide with verbal agreement, we will show that all relevant nominative marking can be handled under (15) with the addition of a mechanism of chain government for nominative objects. Subsequently, we will motivate that subject-verb agreement results from a separate principle requiring the licensing of the functional (phi-) features of I. This principle will eliminate the need for a tripartite definition of nominative case marking, and thus make this definition more restrictive. 3.1.1. Nominative case is not an instantiation of verbal agreement Nominative case is usually characteristic of what is called the subject of a sentence and coincides to a very large extent with the element that shows agreement with the verb. In most Germanic and Romance languages, it seems to be the case that agreement is with a nominative NP. There are, however, I3 In proposals in terms of functional projections, would head a functional projection, e.g. AGRSP. spec,AGRSP

and receive nominative

Nominative The external

case under spec-head

case as a morphological feature argument would then move to

agreement.

192

A. Hulk, A. van Kemenade / Subjects and agreement

some facts that suggest that the assimilation of nominative case with agreement is not compelling: it is possible to have a nominative that does not agree with the verb, as the following sentence from Fiorentino dialect) shows (example from Brandi and Cordin 1989):

(a Modern

Italian

(17) gl’ ha telefonata le tu’ sorelle there has (3sg) telephoned the your sisters (N,pl) ‘your sisters have telephoned’ A similar situation appears to hold true for Icelandic: tive does not necessarily have agreement:

(postverbal)

nomina-

(18) heim likaai ekki pessar athugasemdir them (D) liked (3sg) not these comments (N,pl) ‘they didn’t like these comments’ On the basis of these facts, we propose that the notions nominative case and subject-verb agreement should be divorced. Facts from Arabic discussed in Huybregts (1991) show even more clearly that nominative case does not necessarily go together with agreement or with tense. It must be noted, however, that in all the languages under discussion here the generalization holds that, if there is agreement, it is with a nominative NP. The fact that nominative can exist without agreement suggests that the two do not necessarily go together, but nominative case does seem to be a prerequisite for agreement. We suggest that this is because nominative case and subject-verb agreement are both functional properties associated with the domains of C and/or I. Therefore they appear to go together, although the facts above show that this is not always the case. We will propose below that agreement is a licensing condition separate from nominative case. The data from Fiorentino and Icelandic, however, raise the question whether nominative case is really ‘assigned’ by some head, or whether it is a default case, as suggested by Koster (1986); Everaert (1990) among others. If nominative is a true default case, one would expect that it can be mapped arbitrarily to whatever NP in the sentence is in need of case, in order to satisfy the demands of the case filter. However, we do not find the notion of default status of nominative very revealing. It appears that in those cases where nominative is not on the canonical ‘subject’ (to be understood as the element in spec,IP or spec,VP), it is always within the domain of I. l4 I4 An exception

to this are left-dislocated

elements.

Everaert

(1990) hypothesizes

that the case

193

A. Hulk, A. van Kemenade / Subjects and agreement

As far as we know, there is only one set of syntactic counterexamples to this: instances of nominative objects in passives, ergatives and impersonals in some Germanic and Romance languages. We will discuss this in the next subsection and come to the conclusion that nominative, when assigned, is always in the domain of C or I, and we suggest a principled extension of our definition of nominative case marking (15). This forms a strong counterargument to any theory in which nominative is in some sense a default case.15 3.1.1.1. Quirky subjects and nominative objects. In Modern Dutch, Modern German, Modern Icelandic, we find NP-marked nominative in the domain of a verb in constructions with ergative verbs, psych verbs, as in (19); and quirky passives of ditransitive verbs (20): jouw verhalen niet bevielen (Dutch) that my brother (D) your stories (N, pl.) not pleased (pl) (19b)dass meinem Bruder (D) deine Geschichten (N,pl.) nicht gefielen (~1.) (German equivalent of a) i (194 haa hiifi5u sennilega sokkia einhverjir batar there had probably sunk some boats (N) in firainum (Icelandic, from Sigurasson 199 1) the bay ‘some boats had probably sunk in the bay’ dit boek toegestuurd is (Dutch) (204 dat mijn oom that my uncle (D) this book (N) sent is ‘that this book was sent to my uncle’ GObI dass meinem Onkel (D) dieses Buch (N) zugeschickt worden ist (German) gefnir peningarnir (Icelandic) WC) honum voru him (D) were (pl) given (N,pl) the money (N,pl) ‘he was given the money’

VW dat mijn broer

on left-dislocated

elements

in particular

Thus, the German

sentence

(i) with a left-dislocated

a default

case, therefore

languages

represents nominative,

the default

case of that language.

would suggest that nominative

not ‘assigned’:

den kenn (i) dieser Mann, this(N) man (N), that (A) know

ich nicht I

not

‘This man, I don’t know him’ Also, in many languages, nominatives arc used in isolation. It is not clear to us that such nominatives should be treated as the same thing as nominative NF’ in dependent position in the sentence. I5 This includes Koster (1986) and Everaert (1990).

is

194

A. Hulk. A. van Kemenade / Subjects and agreemenl

Examples like (19) and (20) constitute apparent counterevidence to the claim that nominative case can be assigned by C or I only, as it can be demonstrated that the nominative element is in the complement domain of V, cf. Den Besten (1989); Zaenen et al. (1990). Let us first consider the status of the preposed dative. In Dutch and German, the dative is not really ‘preposed’; Den Besten argues convincingly that the dative remains in VP. Thus the peculiarity of the construction is that the object gets assigned nominative and has agreement in situ. For Icelandic, Andrews (1982, 1990) and Zaenen et al. show in LFG terms that in the case of Icelandic ‘quirky’ sentences like (18) above, quirky subjects are genuine subjects in that they behave as such with respect to subjecthood tests. Notably they can undergo raising to object; they act as antecedent for reflexives; they can undergo subject-verb inversion; they occur in ‘subject position’ in indirect questions, comparatives, relatives; they can undergo indefinite subject postposing and subject ellipsis; controlled PRO can be quirky. l6 In such constructions then, the nominative is a true object, i.e. a complement of V. The notion of subjecthood assumed in the abovementioned LFG analyses is the grammatical function Subject. It may be useful to compare this with the notion of subject used in GB. In GB, subject is a derived notion. Thematically speaking, the subject is the agent or external argument base-generated in spec,VP; syntactically speaking, there is general consensus that the subject is the nominative NP agreeing with the finite verb. Neither notion is fully compatible with the LFG notion of subjecthood as applied to the Icelandic facts. Though the (LFG) subjecthood characteristics of Icelandic quirky subjects are undeniable, we feel that they are not external arguments or syntactic subjects. The constructions featuring quirky subjects in Icelandic involve what one might call verbs of low transitivity. These do not take agent subjects, and we therefore suggest that both arguments are base-generated as verbal complements, and none of them as external arguments in spec,VP. Since quirky subjects are non-nominative and non-agreeing, they are not syntactic subjects either. From a syntactic point of view, they are topics in Spec,IP and fit our analysis of Icelandic as an IV2 language; the spec,IP position can be occupied by an XP, including NP that is not case-dependent upon I. The subject characteristics of quirky subjects are to be derived in a different way, e.g. in terms of linking theory as in Kiparsky (1990).

I6 This presupposes a notion of case-marked worked out in detail in SiguriSsson (1991).

PRO,

against

standard

assumptions.

This is

A. Hulk, A. van Kemenade / Subjects and agreement

195

This does not explain, however, how V comes to assign nominative in these constructions. To see this, observe that the verbs in the constructions in question are verbs that do not assign structural objective case; ergatives, psych verbs, passives. Thus, they are not verbs that have a structural case of their own to assign, only in some instances a lexical case (psych verbs and passives of ditransitives in the relevant languages). Observe furthermore that I cannot assign nominative case directly, since there is no external argument NP to assign it to; if there is an NP in spec,VP, it bears lexical case. We follow Den Besten (1989), who hypothesizes that it is in precisely those environments that I can transmit the structural case feature (i.e. nominative or accusative) to V. He formulates this as chain government (21a); case marking then proceeds as in (21b): (21a) a chain-governs p iff a governs yl, y1 governs y2, . . . , y”-l governs y”, and y” governs l3 (n > 1). (21 b) If NP, is governed by a category a which cannot or may not assign structural (addition ours H/vK) case, NP, will acquire its case from the first structural case-assigner up by which it is chain-governed. We differ from Den Besten in that nominative case marking in Dutch is from C, not from I since Durch is a CV2 language, cf. the definition of nominative case marking (13). With respect to chain government, this does not alter anything; since both the nominative object and the ‘preposed’ dative are in VP, and C chain-governs the nominative NP according to (18a) (chain government is recursive), we may assume that according to (18b), NP, may acquire its case from one step further up, as there is no other case-assigner or case-assignee intervening. With respect to nominative case marking in (19) and (20), the result of this is that in those cases where spec,VP (or spec,IP) is occupied by NP bearing lexical case, and if V has a complement to which it cannot, assign structural objective case, I (or C via I) transmits its nominative feature to V through chain government. Chain government makes the right predictions for Icelandic.17 Consider the following examples (from Sigurasson 1991): (22) involving an ergative verb; (23) a passive. The (a) sentences give a main clause; the (b) sentences

I’ Sigurasson (199 1) analyses nominative as a ‘clausal case’ in such constructions here. This appears to be along the same lines as our analysis here. An alternative the same empirical effects is Yip et al. (1987).

as we discuss proposal with

196

A. Hulk, A. van Kemenade

/ Subjects

and agreement

show what happens when the relevant verb is embedded under an ECM verb like believe; the (c) sentences show what happens when believe is passivized. ha6 hiif& sennilega sokkia einhverjir batar i there had probably sunk some boats (N) in firBinum (cf. (19c)) the bay ‘some boats had probably sunk in the bay’ (22b) Hann taldi i firainum hafa sokkia einhverja bata he(N) believed have sunk some boats (A) in the bay ‘he believed some boats to have sunk in the bay’ hafa sokkia einhverjir batar (22~) haa voru taldir there were believed have sunk some boats (N) ‘there were believed to have sunk some boats’ (22a)

(23a)

haa hafa veria keyptir einhverjir batar there have been bought some boats (N) ‘some boats have been bought’ hafa veri6 keypta einhverja bata (23b) Hun taldi she believed have been bought some boats (A) ‘she believed some boats to have been bought’ hafa verib keyptir einhverjir b&tar (23~) haa voru taldir there were believed have been bought some boats (N) ‘there were believed to have been bought some boats’ In the (a) sentences, the NP some boats has nominative through chain government from I. In the (b) sentences the NP is in a clause embedded under an ECM verb, and instead accusative case is transmitted through chain government from the matrix verb believe. In the (c) sentences, believe is passivized and cannot assign accusative, and thus the nominative case of the matrix I is transmitted through chain government. The chain government cases discussed here are the only ones we are aware of (in NOM/ACC languages) where nominative, when assigned, is not straightforwardly in the domain of I or C. And the notion of chain government is well motivated outside the domain in which we now employ it; Broekhuis (1987) shows that chain government plays an important role in an array of Dutch and German constructions involving binding and tense linking. Therefore, we feel that chain government offers a more restrictive way of dealing with these constructions than simply calling nominative case

197

A. Hulk, A. van Kemenade / Subjects and agreement

‘default’ in some sense, cf. footnote 16. The latter notion incorrectly reduces nominative case to a completely arbitrary notion. Observe that the instances of nominative objects dealt with here cannot possibly be treated as instances of nominative case assigned under spec-head agreement in IP, since the nominative is in the VP. Finally, recall the consequence of our restricted definition of nominative case marking that in ‘ordinary’ declarative sentences with SVO surface order in a non-V2 language, derived from a structure like (3) above, both the subject and its nominative case are chain-related with the spec,VP position, as in (16) repeated for convenience: (16a) Jean frappe Marie (16b) LIPJean, h fowl

Uvp ti (N) Wk LWMarie1111

The NP Jean is chain-related with the base-generated position in spec,VP, and thus chain-associated with nominative case. Thus, given our theory, the movement from spec,VP to spec,IP can never be triggered by the need for nominative case assignment. The fact that the NP in spec,IP is very often a nominative is, from the point of view of case theory, a coincidence. The element in spec,IP can equally be a topicalized object or adverbial in IV2 languages like Icelandic or Yiddish, as shown above. The crucial question then is why languages like Modern English and Modern French always have the nominative subject in spec,IP position. In the next section we will suggest an explanation for this in terms of the licensing of functional features of I.

4. Licensing of IP and VP Beside the requirement of lexicalization where relevant imposed by condition (l), (13) above demands identification of the functional features of I(P). Functional features include the agreement features for person, number and gender, which we shall henceforth call phi-features. Phi-features identify the content of I. Furthermore, it has been argued by Rizzi (1982, 1986) specifically in relation to pro-drop phenomena, that in some languages I (or another head) is a designated licensing head. l* Since we build crucially on these I8

As yet it is unclear

what the crucial

property

is that makes

I in any particular

language

a

designated licensing head; it does not necessarily coincide with overt rich morphology. Roberts (1990) adopts a different metaphor (+ / - m) for the same abstract property. It seems that overt rich morphology is a necessary, but not a sufficient condition for morphological I-licensing. Some other property

is needed.

We come back to this below.

198

A. Hulk, A. van Kemenade / Subjects and agreemenr

properties, which have been related in the literature we will first outline an analysis of pro-drop. 4.1. Functional

properties

to pro-drop

phenomena,

and null subjects

In this section we shall propose an analysis of the pro-drop phenomenon that gives a principled characterization of both full pro-drop (omission of argument pro) and expletive pro-drop (omission of expletive pro). Full prodrop appears in languages such as Modern Italian, where the subject pronoun can remain lexically null, even when it has a thematic role. This is exemplified in (24) with some sentences from Italian, but the phenomenon also occurs in Spanish, Roumanian, Old French (in a more restricted form), among others. (24a)

(24b)

hanno parlato troppo have talked-3pl too much ‘they talked too much’ ti conosco you know- 1sg ‘I know you’

We define full pro-drop then, as ‘omission’ of a subject that has a thematic role, as opposed to expletive pro-drop, which involves ‘omission’ of a nonthematic subject, as in impersonal contexts. Expletive pro-drop is found in a number of Germanic languages. Examples from Modern Dutch and Old and Middle English are those in (25): (25a)

op het plein werd gedanst in the square was danced ‘there was dancing in the square’ &em (25b) 1.. aaette forby to ungemetlice ne sie gliaod that therefore too greatly not be mitigated the scyldgan guilty ‘that therefore it must not be mitigated too greatly to the guilty’ (Cura Pastoralis, ed. Sweet, p. 15 1, 1.2) (25~) henne scheomeh me (obj) perwih then shames me with that (3. Marherete, ed. Mack, p. 34, 1.30) ‘then I am ashamed of that’

A. Hulk, A. van Kemenade / Subjects and agreement

199

Similar examples can be adduced from Middle Dutch, Modern and historical German, Old and Modern Icelandic, but space prevents full discussion. We now suggest an analysis for the various types of pro-drop. Our analysis of pro-drop follows up the classical analysis of full pro-drop as in Italian (Taraldsen 1978, Chomsky 1982, Rizzi 1982,1986) and of expletive pro-drop in the Scandinavian languages as analysed by Platzack (1987). Following Rizzi (1982,1986) we assume that a pro-subject must satisfy two conditions in order to be identifiable: it must be formally licensed, and its content (i.e. phi-features) must be identified. We assume the conditions formulated by Rizzi (1986) : (26) pro-licensing: (a) Pro is formally licensed under government by a designated caseassigning head. (b) The content of pro (phi-features) must be identified. Rizzi (1986) considers that the licensing conditions for pro (26) should be satisfied by one and the same head. We assume that each language has an inventory of designated heads for the formal licensing of pro, and that it is also possible that languages lack such designated heads entirely. There is general crosslinguistic evidence that (26b) presupposes (26a), but not the reverse. That is, if a language has the means of identifying the content of pro, it necessarily has a designated head formally licensing pro. The reverse does not hold: if a language has designated heads for the formal licensing of pro, it does not necessarily have a way of identifying the content. With respect to the formal licensing of null subjects, we propose that the designated case-assigning head (DCH) in (26a) is crucially the dominant head for C/I orientation in terms of our theory. In C-oriented languages, C assigns nominative and can be the DCH. lg In I-oriented languages, I assigns nominative and can be the DCH. This is typically the case in Modern Spanish and Modern Italian. Phi-features, features for person, number and gender, constitute the ‘content’ of pro. These are always generated under I and must be identified as some kind of syntactic or morphological agreement. We assume a definition of government where government is under strict ccommand. This predicts that formal licensing under (26a) is such that in an Ioriented language, a DCH I licenses pro in spec,VP; in a C-oriented language a DCH C licenses pro in spec,IP. I9

This is our rendering

of an insight due to Platzack

(1987)

200

A. Hulk, A. van Kemenade 1 Subjects and agreemrnr

(26), with the addition of our distinction of C/I orientation yields a restrictive view of the distinction between referential pro-drop and expletive pro-drop: in both cases the null subject is formally licensed by a DCH; in the case of referential pro-drop, the phi-features of pro are licensed morphologically by rich I morphology, which distinguishes person, number and gender morphologically; in the case of expletive pro-drop, the content of pro need not be licensed, since there is no content in the form of a theta-role, phifeatures, etc. Thus, referential drop presupposes expletive drop, but not the reverse. This yields the following typology of pro-drop (the specification for DCH is given under the dominant head for every type):

(27) Dutch Cont. Scandinavian Italian French/English Icelandic Old French

COMP DCH DCH

INFL DCH DCH -

+ +

phi phi phi phi phi phi

expletive pro-drop no pro-drop full pro-drop no pro-drop expletive pro-drop full pro-drop in I to C contexts

In languages of the Modern Dutch type (including German, Old English and Middle English), C is the dominant head and the DCH; since there are no overt phi-features, expletive drop is formally licensed by C. Continental Scandinavian has no DCH for subject pro and no overt phi-features and thus has lexical subjects only, cf. Platzack (1987). In the Italian type (including Spanish and Roumanian), I is the DCH and has overt phi-features, hence I licenses full pro-drop. French and English have no DCH and no overt phifeatures, hence have lexical subjects only. Icelandic is I-oriented and I is a DCH. There are no overt phi-features, hence I licenses expletive drop. Old French is a particularly interesting case not discussed so far: it is a CV2 language and the work of Adams (1987) and further work discussed below has shown that it has full pro-drop in V2 contexts and expletive drop in both V2 and non-V2 contexts. We suggest that as a C-oriented language, C is the DCH which formally licenses pro. I has overt phi-features, and full pro-drop is licensed only in those cases where I incorporates into C, i.e. V2 contexts, so that the V/I/C amalgam satisfies both licensing conditions for referential pro. We come back to Old French below.

A. Hulk, A. van Kemenade / Subjects and agreement

201

4.2. Licensing of I

We saw above how the various instantiations of the pro-drop phenomenon are related to functional properties of I and/or C. Crucial for the formal licensing of pro is that the dominant functional head be a DCH. The presence of overt phi-features then further determines the possibility of full pro-drop. The properties of I (and C) thus allow or disallow some form of pro-drop. In this section we argue that the reverse also holds true: we claim that the functional features of I need to be licensed and this is crucially determined by the same properties as those licensing pro-drop. The property of being a DCH acts as a formal licenser for I; and the phi-features of I must be visible as some form of syntactic or morphological agreement. With respect to formal licensing there are three types of I: (28a) I is a DCH. (28b) I is not a DCH but can be related to C which is a DCH. (28~) I is not a DCH. The typology in (28) is complemented with the option for I to have morphologically overt phi-features. If I is a DCH ((28a)) and has overt phifeatures, it is formally licensed and its phi-features are inherently identified. If I is not a DCH, has overt phi-features and is related to a DCH C (28b)), it is formally licensed and has its phi-features inherently identified in those contexts where I incorporates into C, i.e. in root contexts. In non-root contexts, it can be chain-related to a DCH C. If I is not a DCH and has overt phi-features, but is not related to a DCH C, there is no way of formally licensing I, and I will always need the local licensing of agreement with NP in spec,IP. There is a fourth way of licensing that we call default licensing. This is explicated below. There are thus four ways of identifying the phi-features of I, if you will, four types of agreement relationship: (29a) syntactic: SPEC-HEAD Agreement with NP in spec,IP (I is not a DCH and not related to a DCH C) (29b) AGR-chain: I forms part of an agreement chain between a DCH and NP marked nominative (I is DCH or related to a DCH C) (29~) morphological: phi-features are morphologically realized (I is a DCH) (29d) default: impersonal form

202

A. Hulk. A. van Kemenade / Subjects and agreement

(29a) represents the ‘ordinary’ case of agreement between a preverbal subject and I, such as John hits Mary, where John is in spec,IP and is a syntactic licenser of the phi-features of I. This is further discussed in 4.2.1. (29b) represents agreement between V/I and NP in spec,VP or NP in a position that is chain-governed by I (or by C via I). (29~) is the case of languages where phi-features are realized as I morphology, also the case of full pro-drop. Observe that this overlaps partly with languages covered by (29b), since I in the relevant languages is a DCH. The cases (29b-c) are cases where one of the functional heads C or I is a DCH, hence an agreement chain can be formed with a postverbal nominative. This will be discussed further in 4.2.2. (29d) represents cases of impersonal sentences where I has no agreement features at all: it is in the default form 3rd person singular. This is discussed further in 4.2.3. If I is formally licensed and its phi-features are identified, it is what we call ‘fully licensed’. We claim that the licensing of I is crucially intertwined with that of VP. Thus, only a fully licensed I can license the functional properties of V; i.e. its structural case-marking properties. We therefore propose the following condition for VP licensing: (30) V in VP can assign structural I.

case iff it is governed

by a fully licensed

(29d) then represents cases where I is not fully licensed and therefore V cannot assign structural case. It follows that V in such cases can only take complements which are licensed independently, i.e. by morphological case. 4.2. I. Syntactic licensing In this subsection we motivate (29a). (29a) complements our definition of nominative case marking (15). Current definitions of nominative case marking subsume SPEC-HEAD Agreement (henceforth SHAGR) in IP. It was shown above that it is possible to handle nominative marking under a more restrictive definition (15). We propose that the relation between the nominative NP in spec,IP and I is to be handled by a separate licensing mechanism: the identification of phi-features for person, number and gender under I. This mechanism is visible as subject-verb agreement, and is a prerequisite for the full licensing of I. (29a) constitutes a structural option: the phi-features of I are identified by means of local agreement with NP in Spec,IP position. We claim that this structural option for the licensing of I constitutes the trigger for movement of the subject from its base-generated position in spec,VP to spec,IP.

A. Hulk, A. van Kemenade / Subjects and agreement

203

There are two types of language to be considered here: C-oriented languages like Modem Dutch, Modern German, continental Scandinavian languages; 2o I-oriented languages like Modern English, Modem French. In both types I is not a DCH and does not have inherent phi-features. Consequently, I must be licensed by SHAGR with NP in spec,IP according to (29a). This licensing requirement constitutes the trigger for movement of the ‘subject’ NP from spec,VP to spec,IP:

(314 Jean frappe Marie (3lb)

(French) John hits Mary LIPJew LIvJfrwel [W 4 Iv, LNP Marie1111

SHAGR (3lc) Dat Jan Marie slaat (Dutch) that John Mary hits ‘that John hits Mary’ (31d) [CPdat [IP Jani [VPei L- Marie qll L/WsWl1

The difference between the two types is that, while both have I that is not a DCH, in the Dutch but not in the French/English type, I can be related to a DCH C, and can in some contexts form part of an agreement chain in the sense of (29b). This is discussed in the next subsection. 4.2.2. Agreement chains In this subsection we elaborate (29b). Let us take as a starting point the situation where I is DCH and is thus formally licensed (i.e. the Italian and Icelandic types). We propose that by virtue of being formally licensed, I can but need not obey the very strict local licensing of SHAGR in IP, and has the option of making its phi-features visible by forming an agreement chain with a nominative NP to its right. Interestingly, the position of the nominative subject in languages of this sort is rather free: in Italian it may freely appear in the right periphery, as in (32a). In Icelandic too, nominatives may appear preverbally and agree obligatorily, but also appear in immediate postverbal position, presumably in spec,VP; and in the right periphery in chain government constructions as *O We will consider Old English and the more complicated case of Old French below.

A. Hulk, A. van Kemenade 1 Subjects and agreement

204

discussed above.21 It is attractive to relate the relatively free positioning of the nominative subject in these languages to the DCH status of the functional head I.22 In an I-oriented language where I is a DCH, I is formally licensed and can therefore form an agreement chain with nominative NP to the right: with a nominative external argument in spec,VP ((32a)); with a nominative verbal complement in a chain government construction ((32b)) thematic subject in the right periphery, as in (32~ and d).23

or with

a

hef ’ aldrei hitt 1 chain leg Mary (A) have( lsg) I (N) never met ‘Mary I have never met’ i sennilega sokkia einheverjir bcitar Wb) baa hoifbu chain I I boats (N) in there had(3pl) probably sunk some firbinum (Icelandic, from Sigurasson 199 1) the bay ‘some boats had probably sunk in the bay’ mangia (3sg) Gianni (32~) chain Gianni eats mangiato Francesco (32d) una pizza ha chain I I Francesco (N) a pizza has(3sg) eaten ‘a pizza Francesco ate’

(324 Mariu

The form of the chain is thus determined by the mode of nominative assignment. If I assigns nominative to spec,VP, it will agree with. NP in spec,VP. If I assigns nominative to a complement of V under chain government,

it will agree with that NP.

21 Agreement some impersonal

between

finite V and a postverbal

verbs like lika, as we are informed

nomintive

is not absolutely

by an anonymous

reviewer,

obligatory

with

cf. example

(I 8)

above. We analyse these as instances of default licensing as discussed in 4.2.4. This was adduced above as evidence for a separation of nominative case and verbal agreement. 22 This is our rendering of the insight in Rizzi (1982) that there is a typological between null subjects in Italian and what he calls ‘free inversion’. 23

With respect to the Italian

postverbal

Rizzi 1982), we feel tempted to analyse right-hand periphery of VP. However, leave it as a tentative

suggestion.

subject

in ‘free inversion’

(according

relationship

to the analysis

in

the position of the postverbal subject as a specifier in the since the consequences of this idea are not yet clear, we

A. Hulk, A. van Kemenade / Subjects and agreement

205

As a result of this freer form of I-licensing, the phi-features of I are licensed by means of agreement with an element to its right. Observe that this commits us to the assumption that in the Italian examples (32~4) I has no inherent phi-features as was assumed above for sentences with null subjects, or I would not need to form an agreement chain. We suggest therefore, following Rizzi (1982), that overt phi-features are optional in Italian, and that referential pro is only possible when they are overt. When they are not overt, I forms a chain with a postverbal nominative as in (32cd), or has local SHAGR as in Gianni mangia (3sg) ‘Gianni eats’. The possibility of agreement with nominative NP to the right leaves the spec,IP position free to be occupied by a constituent other than the nominative subject, as in (32a) and (32d). Note that this is what we analysed in section 2 as the IV2 characteristic. We suggested an IV2 analysis for Icelandic, following Rognvaldsson and Thriinsson (1990); and for Yiddish, following Diesing (1990). This idea is new for Italian; constructions like (32d) were analysed in Rizzi (1982) as adjunction to S. 24 Observe that our proposal of this section explains why the IV2 characteristic obtains in these languages: the DCH status of I opens the option of forming an agreement chain between I and a postverbal nominative NP, thus leaving spec,IP free for another constituent to move to, be it an accusative object as in (32a and d), or a quirky object as discussed in the previous sections, or some other element. Let us now consider C-oriented languages where I is not a DCH, hence not formally licensed, but where C is a DCH, i.e. Dutch, German. 25 Observe that in these languages I is non-pronominal and must therefore be licensed by SHAGR in IP as discussed and illustrated in the previous subsection for Dutch. But in all of the relevant languages, there are contexts where there is agreement between V/I and a nominative internal argument in VP: the chain government contexts discussed above, examples of which are repeated here for convenience : jouw verhalen niet bevielen (Dutch) (33a) dat mijn broer that my brother (D) your stories (N, pl.) not pleased (pl) (33b) dass meinem Bruder (D) deine Geschichten (N,pl.) nicht gefielen (~1.) (German) ‘that my brother was not pleased by your stories’ *4 Recall our observation in section 2 that the term IV2 should not be taken as a strict word order constraint. 25 To these may be added Old English and Old French. These will be discussed below.

206

A. Hulk, A. van Kemenade / Subjects and agreement

We analyse this as follows: C assigns nominative case in these contexts under chain government, as discussed. C is a DCH and, in the same way as I does in I-oriented languages, forms a chain with the nominative NP in VP. Observe that while I is not the head of this chain as in I-oriented languages, it does form part of the chain between C and the nominative NP and thus satisfies (29b). In C-oriented languages where C is a DCH, agreement chains of the sort discussed here occur only in chain government contexts - in nonchain government contexts syntactic licensing according to (29a) is obligatory. This allows a very precise characterization of the spec,IP position in these languages: in all constructions with an external argument, spec,IP is the position for the nominative subject since I is not formally licensed and hence must be licensed by SHAGR; in ergatives, passives, impersonals and the like, chain government allows nominatives in the VP, and a chain can be formed between a DCH C and the nominative in the VP. Only in these contexts, therefore, the spec,IP position is not obligatorily occupied by the nominative subject, since the phi-features of I are licensed via an agreement chain. Observe that the present analysis establishes a typological correlation between several seemingly unrelated properties in a large number of languages under one parameter: the DCH character of the dominant functional head. This DCH character formally licenses null subjects locally and allows agreement chain formation between the dominant functional head and a nominative NP to the right, thus allowing agreement with nominatives in the VP. There is a typological relationship between these properties and the existence of chain government since chain government creates nominatives in VP. One might even take this correlation one step further and hypothesize that chain government is a general option, i.e. not a language-specific choice, that is constrained in many languages/contexts by the obligatoriness of local agreement under SHAGR in IP. 4.2.3. Morphological licensing In this section we discuss languages where I has morphologically phi-features. There are two possible options: 26

realized

(34a) I is a DCH and has overt phi-features. (34b) I is not a DCH and has overt phi-features and can be related to a DCH C. 26 Logically there is a third option: I is not a DCH, has overt phi features and cannot be related to a DCH C. It was observed above, however, that the presence of overt phi-features presupposes formal

licensing,

but not the other way round.

This excludes

this third option.

A. Hulk, A. van Kemenade / Subjects and agreement

207

Recall from our analysis of null subjects in section 4.1 that in Italian I is a DCH which ensures the possibility of an agreement chain to the right. I also optionally has morphologically realized phi-features, which, if present, license the content of pro. This is the situation that obtains in (34a). We claim that the same properties fully license I inherently. As a result I with overt phifeatures has inherent agreement and does not need to form a chain with another NP in the sentence. A more complex case is (34b). C is a DCH formally licensing I. I is not a DCH and has inherent phi-features, hence I is not formally licensed. In this CV2 language there are two ways in which I can be related to the DCH C: one is by movement of V/I to C in root contexts; the other is when I forms part of a chain between C and a nominative in the VP. Old French is a case in point and will be discussed more extensively below. 4.2.4. Default licensing Default licensing (29d) is a special case of licensing. Default licensing typically presents itself in languages where I is a DCH or forms part of a chain between a DCH C and a nominative NP, but cannot have any person features or phi-features. Thus, default licensing typically represents the case of impersonal sentences: there is no nominative subject or there is no agreement with a nominative subject; there is no verbal agreement as INFL assumes a default form third person singular; thus I is not fully licensed and it follows from condition (30) above that V in such cases cannot assign structural case. V can therefore not take NP complements unless these are licensed in a different way, e.g. by morphological case or by clitics. Let us illustrate this with a few examples: (35a) op het plein werd gedanst (Dutch) in the square was danced ‘there was dancing in the square’ (35b) *op het plein werd appels gegeten (Dutch) in the square was (3sg) apples (pl) eaten on bisse worulde (OE) (35~) hine (A) nanes hinges (G) ne lyste nothing him not pleased in this world ‘nothing in this world pleased him’ (Boeth., 102, 9) (35d) 11 a CtC mange beaucoup de pommes (French) of apples (pl) there has (3sg) been eaten many ‘Many apples have been eaten’

208

A. Hulk. A. van Kemenade

/ Subjects and agreemenr

The Modern Dutch sentence (35a) is an impersonal passive. Thus, there is no thematic subject, and the finite verb werd has the default third person singular form. I is then not fully licensed, as there is no structural or morphological licenser. Therefore V gedunst cannot assign structural case and thus cannot have any direct object, as (35b) shows. Hence, only a PP op de tufel can complement it, in the absence of morphological case in Dutch, as a potential morphological licenser for NP. In contrast, the Old English sentence (35~) does have verbal complements, but note that these are identified by overt morphological NP marking. Another contrast forms the French impersonal passive (35d), which is derived as ergatives and intransitives above: it has been suggested by Hulk (1989) that the subject clitic il fully licenses I, hence I licenses V, therefore the passive V can assign objective case to the object beacoup de pommes. Similar to these impersonal passives are Icelandic impersonal sentences with a nominative and without agreement; (18) above is repeated here for convenience : likaai (18) heim ekki hessar athugasemdir them (D) liked (3sg) not these comments (N,pl) ‘they didnt like these comments’ There is no agreement in (18), hence I is not fully licensed and cannot license V, and V can have NP complements because these are licensed by morphological case.

5. Language change In this section we will show that the approach advocated in this article provides an interesting perspective on the history of V2 and pro-drop in French and English. We will do this briefly here; the data and analysis are spelled out in more detail in Hulk and Van Kemenade (1990). The facts concerning French are taken from Adams (1987) Hirschbiihler (1990) Vance (1989), Dupuis and Lemieux (1990); those concerning English from Van Kemenade (1987) and textual sources. Cited references are taken over from the relevant secondary sources. Old French (OF) and Old English (OE) are both CV2 languages; there is, as far as is known, a root/non-root asymmetry with respect to V2. Some examples of V2 in root clauses :

A. Hulk, A. van Kemenade 1 Subjects and agreement

209

(36a) Ensi fut Joseph perdus une grant piece (OF) Thus was Joseph lost for a long while ‘Thus Joseph was lost for a long time’ (Roman de Graal, 27) (36b) for haes wintres cyle nolde se asolcena erian (OE) for the winter’s cold not-wanted the layabout plough ‘the layabout didn’t want to plough because of the cold’ (Blfric’s Homilies, ed. Godden, 17, 1.116) OF has full pro-drop, but interestingly here too, there is a root/non-root asymmetry; (37a) is a root V2 context, and so is the first part of (37b). (37b) is particularly interesting, since it has a null subject in the root clause and a lexical subject in the embedded clause: (37a) Or fait senblant con s’ ele plore as if she cries now makes (she) seem ‘now she pretends to cry’ (Tristan, 1. 8) (37b) Einsi corurent par mer tant que il vindrent a Cademelee thus ran (they) by sea until they came to Cadmee ‘thus they ran by the sea until they got to C.’ (Cl. XXV) OE has expletive pro-drop in both root and non-root clauses: has ilcan (38a) honne ofpynch him he he zr forbar then regrets him(D) the same(G) that he before endured ‘then he regrets what he endured before’ (Cura Pastor&s, ed. Sweet, p. 225, 1.19) swa miclum sniwde swelce mice1 flys feolle (38b) and and 0 so heavily snowed as if much fleece fell ‘and it snowed so heavily, as if a lot of fleece were falling’ (Epist. Alex. 159, 538, from WahlCn 1925) OE and OF both have chain government:27 en fu coverte la vraie semblance (OF) (39a) aincois lor rather to-them of-it was covered the true appearance (N) ‘rather, the true nature of it was hidden from them’ (Queste 16) 27 For a detailed discussion of OF in these terns, see Vance (1989).

210

A. Hulk, A. van Kemenade / Subjects and agreement

(39b)

cele nuit furent servit et aiesie li compaignon that night were served and satisfied the companions(N) that night the companions were served and satisfied with . . . (Queste 27) (39~) Gif hwam se0 lar oflicige, . . . WI if anyone (D) the doctrine (N) dislike (subjunctive), ... ‘if the doctrine should be displeasing to anybody, . ..’ (39d) gif aam gifran angemetlicu sprat ne eglde if the greedy (D) eloquent speech (N) not afflicted ‘if the greedy are not afflicted by loquacity’ (Cura Pastoralis, ed. Sweet, p. 309, 1. 3)

de . . . with

We analyse both the similarities and differences between OF and OE as follows: as indicated above, both are CV2 languages with V movement to I and V/I movement to C. C is the dominant head, hence nominative is assigned by C under condition (15) and C is a DCH that formally licenses pro under (26a). In both languages spec,IP is the position for the nominative subject, except in ergatives, passives etc. : chain government contexts, in which the phi-features of I can be licensed through the agreement chain to the right. The difference is that I in OF has overt phi-features that license referential pro-drop under (26b), whereas I in OE lacks these features. Therefore OE has no way of identifying referential pro. OF has a root/non-root asymmetry for pro-drop because it is only in root contexts that the V/I/C amalgam can satisfy both conditions for licensing pro (26). In the transition to Middle French (MiF), the root/non-root asymmetry disappeared both for V2 and for pro-drop (we refer to Hulk and Van Kemenade (1990) for motivations for these changes), and we find a situation very similar to that in modern standard Italian: (40a)

(40b)

que envers moi a commis . . . . . . comment celle faulte this mistake that against me have committed, .. how ‘how this mistake that (you) have committed against me, . . (Vigneulles 094044) quant temps fut de partir prindrent congie et s’en allerent when time was to leave took leave and went ‘when (it) was time to leave, (they) took leave and went’ ( Vigneulles 00 10 18)

The examples

(40) illustrate

embedded

V2 with pro-drop.

We analyse

this

A. Hulk. A. van Kemenade / Subjects and agreement

211

generalization of V2 and pro-drop from root contexts to both root and nonroot contexts, as a change from a CV2 to an IV2 language. Recall that the IV2 characteristic is possible only in languages that have a DCH I, since spec,IP can be an XP position only if I can form an agreement chain to the right. Given the DCH I, IV2 languages have some form of pro-drop, modulo the question whether they have inherent phi-features or not. Middle French does, hence it is an IV2 language with referential pro-drop. French lost its IV2 character later on; because the DCH I character and inherent phifeatures were lost, a nominative NP in spec,IP became obligatory. We predict that at this stage, chain government was lost too. We have no information on this, and leave it for further research. Let us now consider the further history of English. Recall that OE is CV2 with expletive pro-drop. This situation continued through the Middle English period : (41a) henne scheomeh me (obj) herwih then shames me with that ‘then I am ashamed of that’ (St. Marherete, ed. Mack, p. 34, 1, 30) (41b) him (obj) wile sone longe harafter him wile soon long after that ‘he will soon long for that’ (Trinity Homilies, p. 148, 1. 19) At the end of the ME period, V2 became restricted to the contexts where it is still used (cf. Van Kemenade 1987), and expletive pro-drop disappeared. We interpret this as a change from CV2 with a DCH C to I-oriented with non DCH I (the motivations for this are spelled out in Hulk and Van Kemenade 1990). The result of this is that the licensing of the phi-features of I according to (29) came to be restricted to exclusively syntactic contexts, and a nominative NP in spec,IP became obligatory in all contexts. 28 We predict that at this stage chain government was lost too. It is rather problematic to work out clear cases of chain government in ME, since distinctive case endings have declined and as a result we cannot discriminate between dative and nominative for non-pronomitral NP, and for pronominal NP there are other complications. Summarizing, our theory provides an interesting perspective for the diverging yet similar developmental histories of French and English with respect to 28 Observe that there is no a priori reason why English did not pass through an intermediate IV2 stage of the Icelandic type. We can only say that there is no evidence for this. We leave this for further research.

A. Hulk, A. van Kemenade 1 Subjects and agreement

212

V2 and pro-drop. The key to the CV2-IV2-non-V2 development of French is to be sought in a development from OF DCH C and I with inherent phifeatures, to MiF DCH I with inherent phi-features (which were later lost) that permitted an intermediate IV2 stage. For English there is a two-step development because I in English always lacked this complete set of specifications.

6. Conclusion In this article, we have developed some parameters deriving verb placement phenomena, nominative case marking and I-licensing and interactions between them. In section 2, a distinction was made between C- and I-orientation of languages on the basis of V2 phenomena resulting in a condition requiring that the dominant head be licensed. In section 3 we elaborated a licensing condition for the functional feature nominative case for NP. In section 4 we motivated a new theory for the licensing of the phi-features of I, which, interacting with our view of head dominance, results in a typology of four types of subject-verb agreement. The licensing of I in turn determines whether the dependent V can assign structural case. Finally, it was shown that this theory casts an interesting light on the divergent histories of English and French with respect to V2 and pro-drop.

References Adams,

Marianne,

Linguistic Andrews,

Avery,

The mental Andrews, Annie Diego, Andrews,

1987. From

Theory

Avery,

of grammatical

of prodrop.

of case in Modern

Icelandic.

relations,

Cambridge,

1990a. The VP-complement (eds.), Syntax

and semantics,

CA: Academic Press. Avery, 1990b. Case structures

Annie Zaenen

to the theory

1982. On the representation

representation

Zaenen

Old French

Natural

Language

and

In: Joan Bresnan

(ed.),

5, l-32.

(eds.), Syntax

and

and semantics,

analysis

427-503.

in Modern

Vol. 24: Modern control

Icelandic.

Icelandic

in Modern

Vol. 24: Modem

MA: MIT Press.

Icelandic. Icelandic

In: Joan

syntax,

In: Joan

syntax,

Maling,

165-187.

San

Maling,

187-235.

San

Diego, CA: Academic Press. Bennis, Hans and Liliane Haegeman, 1983. On the status of agreement and relative clauses in West-Flemish. In: W. de Geest, Y. Putseys (eds.), Sentential complementation, 33-55. Dordrecht: Foris. Besten, Hans den, 1989. Studies in West Germanic syntax. Amsterdam: Rodopi. Brandi, Luciana and Patrizia Cordin, 1989. Two Italian dialects and the null subject parameter. In: Osvaldo Kluwer.

Jaeggli,

Kenneth

Safir (eds.), The null subject

parameter,

111-143.

Dordrecht:

213

A. Hulk, A. van Kemenade / Subjecrs and agreement

Broekhuis,

Hans,

Chomsky,

Noam,

Chomsky,

Noam,

binding.

1987. Chain government. 1981. Lectures

1982. Some concepts

Cambridge,

The Linguistic

on government

and consequences

Chomsky,

Noam,

1986. Barriers.

Noam,

1988. Some notes on economy

Molly,

Linguistic Dupuis,

Cambridge,

1990. Verb movement

Theory

Femande

and

Monique

in Old and Middle

Conference,

York,

Martin,

Abraham,

in 1992.)

Foris.

of the theory

of government

and

MA: MIT Press. of representation.

and the subject position

Unpublished

in Yiddish.

ms., MIT.

Natural

Language

and

8(l), 41-8 1.

phenomena Everaert,

(Appeared

Dordrecht:

MA: MIT Press.

Chomsky, Diesing,

Review 6,297-374.

and binding.

April

Lemieux, French.

1990. Arguments

for an IP analysis

Talk given at the First Generative

of the V2

Diachronic

Syntax

1990.

1990. Nominative

Wim Kosmeijer,

anaphors

Eric Reuland

in Icelandic:

Morphology

(eds.), Issues in Germanic

or syntax?

syntax,

In: Werner

277-307.

Berlin: de

Gruyter. Evers, Arnold, Evers, Arnold,

198 1. Verb second movement rules. Wiener Linguistische 1982. Twee functionele principes voor de regel ‘verschuif

Gazette 26, 15-34. het werkwoord’. Glot

1,

1l-30. Godden,

Malcolm

Gutron,

Jaqueline,

(ed.), 1979. Elfric’s

Smith (eds.), Going Haan,

Germen

Haider,

Romance

1986, 83-107.

de and Fred Weerman,

Martin

Prinzhom

Dordrecht:

Foris.

Hirschbiihler,

Paul,

(eds.),

Hulk,

Aafke,

1990. VI, V2 and V>2

Hulk,

Syntax

and verb fronting

In: Hubert

languages,

in Philippe

de Vigneulles

clauses

Conference,

and the acquisition

in Frisian.

in Germanic

et la structure

1988. Nominative

Bill Dotson

York, April

identification

Aafke Hulk (eds.), Linguistics

CNN.

1990.

de la phrase.

of word order

77-110.

Recherches

in L2 French.

in Germanic

in the Netherlands

Second

and Romance 1988. 69979.

Foris.

Aafke

and Ans van Kemenade,

change.

To appear

in Adrian

Oxford

University

Press.

Huybregts,

Riny, 1991. Allosteric

Kemenade

(eds.), Linguistics

Kemenade, Ans van, Dordrecht: Foris.

1990. V2, pro-drop,

Battye,

Ian Roberts

agreement

in VSO languages.

in the Netherlands

1987. Syntactic

Hilda,

Koster, Jan, Groningen.

1984. The syntax

1986. On the relation

functional

(eds.), Language

case and morphological

of verbs. Dordrecht between

projections change

In: Frank

1991, 81-90.

Kiparsky, Paul, 1990. Grammaticalization vs. reanalysis: the Eleventh Groningen Grammar Talks. Koopman,

Peter Coopmans,

ICG.

phenomena

embedded

Diachronic

setting

In: Peter Coopmans,

Dordrecht:

Dordrecht:

1989. La construction impersonelle de Vincennes 18, 59-8 I. 1991. Parameter

EETS, series II s.s.5. London.

Bordelois,

1986. Finiteness

Language Research 7( 1). l-34. Hulk, Aafke and Ans van Kemenade, languages

Homilies.

Verb second

Talk given at the First Generative Hulk, Aafke, Linguistiques

Catholic

1986. Le verbe avoir. In: Yvonne

and language

and verbal systems.

Drijkoningen,

Ans van

Berlin: de Gruyter. case in the history

of English.

The case of lexical case. Paper given at : Foris.

V2, prodrop

and

scrambling.

Unpublished

ms.,

Lasnik, Howard, 1981. Restricting the theory of transformations: A case study. In: Norbert Homstein, David Lightfoot (eds.), Explanation in linguistics, 152-174. London: Longman.

214

A. Hulk, A. van Kemenade / Subjects and agreement

Mack,

Frances

Maling,

Joan,

Zaenen

(ed.), 19.58. Seinte Marherete. 1990. Inversion

(eds.), Syntax

CA: Academic Pesetsky, David, The Linguistic Platzack, Pollock, Hans

and semantics,

clauses in Modem Vol. 24: Modern

Icelandic.

Icelandic

In: Joan Maling,

syntax,

71-95.

Annie

San Diego,

Press. 1982. Complementizer-trace Review 1, 297-345.

Christer,

Language

EETS, OS 193.

in embedded

phenomena

1987. The Scandinavian

and Linguistic

Jean-Yves,

Theory

languages

5, 37740

Jean-Yves

Pollock

and the null subject

island condition. parameter.

Natural

1.

1984. On case and the syntax

Obenauer,

and the nominative

of infinitives

(eds.), Grammatical

in French.

In: Jaqueline

representation,

Gutron,

Dordrecht :

293-326.

Foris. Pollock,

Jean-Yves,

1986. Sur la syntaxe

Daniel Couquaux

(eds.), La grammaire

modulaire,

1989. Verb movement,

universal

Pollock,

Jean-Yves,

Inquiry 20, 365425. Rizzi, Luigi, 1982. Issues in Italian

syntax.

Rizzi, Luigi,

1986. Null objects

Rizzi, Luigi,

1990. Some speculations

Nespor

(eds.),

Grammar

Natural Roberts,

Ian,

in Italian

Language

parameters

and Linguistic

Ian, 1990. Verbs and diachronic

Ronat,

de Minuit. of IP. Linguistic

Foris. of pro. Linguistic

essays

inversion

for

Inquiry

In: Juan

Henk

in French.

van

17, 501-559.

Mascaro,

Riemsdijk,

Probus

and the development

Theory

Editions

and the structure

V2 phenomena.

GLOW

1989. Complex

1985. Agreement

Paris:

grammar,

Dordrecht:

on residual

du sujet nul. In: Mitsou

21 l-246.

and the theory

in progress:

Dordrecht: Foris. Rizzi, Luigi and Ian Roberts, Roberts,

de en et le parametre

l(l),

of English

Marina 375-387.

l-30. modal

auxiliaries.

3, 21-58. syntax.

Unpublished

ms., Geneva.

Rdgnvaldsson, Eirikur and Hiiskuldur Thriinsson, 1990. On Icelandic word order once more. In: Joan Maling, Annie Zaenen (eds.), Syntax and semantics, Vol. 24: Modern Icelandic syntax, 341. San Diego, CA: Academic Santorini,

Beatrice,

Yiddish.

Unpublished

Sigurasson,

Press.

1989. The generalization

Halldor,

Ph.D.

dissertation,

1989. Verbal

syntax

tion, University of Lund. Sigurasson, Halldor, 1991. Icelandic Natural Sportiche,

Language Dominique,

Theory

1988. A theory

second

in the history

of

Unpublished

PRO and the licensing

doctoral

disserta-

of lexical A-positions.

9(2), 327-363.

of floating

19, 425451. King Alfred’s

constraint

of Pennsylvania.

and case in Icelandic.

case-marked

and Linguistic

structure. Linguistic Inquiry Sweet, Henry (ed.), 1871-72.

of the verb University

quantifiers

West-Saxon

and its corrollaries version

of Gregory’s

for constituent Pastoral

Care.

the that-trace

filter.

EETS, Vols. 45 and 50, London. Taraldsen,

Knut

Tarald,

1978. On

the NIC,

Unpublished ms., MIT. Thorpe, Benjamin (ed.), 1864. The homilies

vacuous

application

of the Anglo-Saxon

and Church.

The Aelfric

Society,

London. Thriinsson, Hoskuldur, 1986. Vl, V2, V3 in Icelandic. In: Hubert Haider, Martin Prinzhorn (eds.), Verb second phenomena in Germanic languages, 169-195. Dordrecht: Foris. Tomaselli, Alessandra, 1990. COMP as a licensing head - An argument based on cliticization. Juan Mascara, Marina Nespor (eds.), Grammar in progress: GLOW essays for Henk Riemsdijk,

433447.

Dordrecht:

Foris.

In: van

A. Hulk, A. van Kemenade / Subjects and agreement

215

Vance, Barbara, 1989. Null subjects and syntactic change in Medieval French. Unpublished dissertation, Cornell University. Wahltn, N., 1925. The Old English impersonalia, Part I. Giiteborg. Weerman, Fred, 1989. The V2 conspiracy. Dordrecht: Foris. Yip, Moira, Ray Jackendoff and Joan Maling, 1987. Case in tiers. Language 63, 217-250. Zaenen, A., J. Maling and H. Thrainsson, 1990. Case and grammatical functions: The Icelandic passive. In: Joan Maling, Annie Zaenen (eds.), Syntax and semantics, Vol. 24: Modem Icelandic syntax, 95-137. San Diego, CA: Academic Press. Zwart, J.W., 1992. Verb movement and complementizer agreement. Paper given at the GLOW colloquium, Lisbon.