The internal syntax of the Dutch extended adjectival projection 1

This paper is concerned with the phrase structural and word order properties of the (extended) adjectival projection, a phrase structural domain which has received relatively little attention in the generative literature. Focusing on the internal syntax of Dutch adjective phrases, I will come to the following conclusions. First, there is a strong empirical (and theoretical) basis for extending the functional head hypothesis to the adjectival system (i.e. for adopting the DegP-hypothesis). Secondly, a distinction should be made between two types of functional degree categories: Deg(P) and Q(P). This split is represented structurally, with Deg selecting QP and Q selecting AP (the split degree system hypothesis). Thirdly, there is empirical support for the existence of a third functional projection, AgrP, within the adjectival domain. Fourthly, as regards directionality of headedness within the Dutch functional system, it is concluded that Deg and Q take their complements to the right, whereas Agr takes its complement to the left. It is proposed that this asymmetry of headedness within the functional structure of the adjectival projection relates to the nominal orientation of Deg and Q and the verbal orientation of Agr. Finally, three movement operations will be identified within the Dutch adjectival system: A-to-Q raising, A-to-Agr raising and leftward scrambling. The latter two are at the basis of the word order variation which is found within the Dutch adjectival system.

Introduction

An important theme in current syntactic theory is the projection of functional categories in phrasal structure. Traditionally, most functional categories are analyzed as occupying the specifier position of some lexical head (Jackendoff, 1977). Under such a phrase structure analysis, the functional projection is embedded within the lexical domain. Recent phrase structure research, however, has reinterpreted the structural relation between the functional domain and the lexical domain. The functional category is conceived of as heading a phrasal functional projection and taking a lexical phrase as its complement (Abney 1987; Fukui and Speas, 1986). The two analyses are given schematically in (1), (1a) representing the traditional lexical head hypothesis, (1b) representing the

functional head hypothesis (Grimshaw, 1991). (‘L’ stands for Lexical, ‘F’ stands for Functional).

(1)

a. [_LP FP [_L' L XP]]

b. [_FP Spec [_F' F LP]]

The functional head hypothesis, represented in (1b), has been succesfully applied in recent years to the verbal and nominal domain. As for the verbal system, it has been proposed that the lexical VP-projection is included within a functional IP-projection, plausibly further split up into constituent components such as AgrP, TP etc., which is itself contained within CP (cf. Chomsky, 1986; Pollock, 1989; Belletti, 1990). The nominal system has been reanalyzed as being a projection of the D(eterminer), which takes a lexical NP as its complement (Brame, 1981; Abney, 1987; Fukui and Speas, 1986; Longobardi, 1994). 2 This more articulated conception of verbal and nominal phrase structure has proven quite fruitful in finding an account for various ordering effects within these syntactic domains.

A syntactic domain which has received much less attention in recent years, both from the perspective of phrasal structure and word order, is the adjectival system. The purpose of this paper is to deepen our insight into the adjectival system by closely examining phrase structural and word order properties of the Dutch adjective phrase. 3 In what follows, I will briefly discuss the major issues which will be dealt with in this paper.

The first major issue concerns the question whether the functional head hypothesis (i.e. (1b)) can be extended to the adjectival system. That is, is there any evidence for interpreting degree words - which are traditionally analyzed as occupying the specifier position of AP (Bowers, 1975; Jackendoff, 1977) - as heading a functional Degree Phrase (DegP), taking the lexical projection AP as its complement (the so-called Degree Phrase Hypothesis (Abney, 1987; Corver, 1991))? On the basis of a variety of empirical arguments it will be concluded in Section 2 that the DegP-hypothesis is preferable to the traditional AP-structure.

The second major issue concerns the question whether there is a uniform system of functional degree words consisting of such items as in (2) (cf.

Jackendoff, 1977) or whether a further distinction should be made within the class of functional degree items, for example between those which are quantifier-like (e.g. meer, ‘more’, minder ‘less’, genoeg ‘enough’) and those which are not (e.g. te, ‘too’, hoe, ‘how’).

(2)

even, zo, te, hoe, meer, minder, genoeg, …

as, so, too, how, more, less, enough, …

In Section 3, it will be shown that the the quantifier-like degree items behave differently from the other degree words in various respects, leading to the conclusion that besides the functional DegP-projection a functional QP-projection should be distinguished within the functional domain of the extended adjectival projection. In Section 4, I will show that the head of this QP can function as a landing site for gradable adjectival predicates.

The third major issue to be addressed concerns the directionality of headedness withm the adjective phrase. If degree words (Deg°) and quantifiers (Q°) are functional heads, the question arises whether they take their complement to the right or to the left. This question is pertinent to Dutch in view of its often assumed mixed branching nature (Koster, 1987). After having examined both the head-initial and head-final hypothesis, I will come to the conclusion in Section 5 that the Dutch DegP (and QP) should be analyzed as being head-initial. This may lead to the assumption that Dutch adjectival phrase structure is head-initial throughout. However, in Section 6, evidence will be given for the existence of a head-final functional node Agr (heading AgrP) which can function as a landing site for adjectival heads that are moved rightward (the fourth major issue discussed in this paper). This raises the question what accounts for the mixed headedness in the functional system of the extended adjectival projection: Deg° and Q° being head-initial and Agr° being head-final (the fifth major issue). In Section 7, the mixed headedness of the functional structure will be interpreted as a reflection of the categorial feature definition of the syntactic category ‘adjective’. Adjectives, being defined as [+N, +V] (Chomsky, 1970), have both nominal and verbal properties. The idea will be that Deg° and Q° are more nominal in nature and analogously to nominal heads within the nominal extended projection (i.e. DP) take their complements to the right. The functional head Agr°, on the other hand, is more verbally oriented and, like the Infl-node within the verbal extended projection, takes its complement to the left.

The sixth major issue will be dealt with throughout the paper and concerns the word order variation within the Dutch adjective phrase. It will be shown that this variation is due to reordering operations within

the adjectival domain, such as leftward scrambling of maximal categories, on the one hand, (cf. Section 5) and rightward A°-to-Agr° raising, on the other hand (cf. Section 6).

2. In Support of the DEGP-Hypothesis

This section addresses the question how functional categories of the adjectival system are projected in syntactic structure. If we adopt the standard assumption that degree words as in (2) are the function words of the adjectival system, we can formulate two answers to this question. One answer would be to say that degree words head a Degree Phrase which is located in the [Spec, AP] position (Jackendoff, 1977). This view will be referred to as the lexical head hypothesis: the lexical head A° is the head of the adjective phrase. According to the alternative answer, which represents the functional head hypothesis, the whole adjectival construction is conceived of as coinciding with the maximal category DegP (i.e. Degree Phrase) and of AP as the complement of the degree word (cf. Abney, 1987; Bowers, 1987; Corver, 1990, 1991). Both views are schematically represented in (3):

(3)

a. [_AP [_DegP Spec [_Deg' Deg]] [_A' A XP]] (lexical head hypothesis)

b. [_DegP Spec [_Deg' Deg [_AP A XP]]] (functional head hypothesis)

In both analyses it is assumed that the specifier position of DegP can be occupied by various elements qualifying the degree word (Jackendoff, 1977; Abney, 1987). Some examples are given in (4).

(4)

a. twee centimeters te lang

two centimeters too tall

b. veel, minder lang dan Peter

much less tall than Peter

From a conceptual point of view, the DegP-hypothesis in (3b) is the null hypothesis: if the lexical domain is closed off by the functional projec tion in the case of the nominal and verbal system, one would, for reasons of cross-categorial correspondence, expect the same to hold for the adjectival system (Abney, 1987; Grimshaw, 1991).

At the empirical level, it has been noted in Abney (1987) that under a Degree Phrase-hypothesis it is possible to accomodate the variety of adjectival specifiers under a two-bar X-bar theory. As shown, for example, by the English sentences in (5), degree words like how and so can co-occur with other specifying elements like very and utterly. This co-occurrence is

also found in the Dutch example (6), where the degree word zo and the modifying phrase heel erg appear simultaneously within the adjective phrase.

(5)

a. Fred was [so utterly confused that he fell off the podium]

b. [How very long] he can stay under water!

(6)

[Zo heel erg slim] is deze jongen anders niet!

So quite very smart is this boy however not

This boy is not all that smart after all!

Under Jackendoff's traditional AP-analysis, the co-occurrence of these items is unexpected, since functional degree words and adverbial degree modifiers are assumed to be located in one and the same structural position, namely [Spec, AP]. Abney notes that under a structure as in (3b), on the other hand, the two specifying items can be accomodated, one of them (so/how in (5)) occuring in the head position of the degree phrase, the other (utterly/very) in the [Spec, AP] position. 4

The question, of course, arises whether there is a broader empirical basis for adopting the so-called DegP-hypothesis for the adjectival system. In the following subsections I will discuss a number of phenomena from Dutch which give support to extending the functional head hypothesis to the adjectival system.

2.1. Head-to-Head Movement

A first argument in support of the functional head hypothesis (i.e. structure (3b)) comes from the formation of analytic comparative adjective phrases as in (7), where we find the bound comparative morpheme -er attached to the adjectival stem. 5 This morpheme alternates with the free comparative morpheme meer (‘more’), which occurs in periphrastic comparative adjective phrases as in (8).

(7)

a. [Sterker dan Karel] leek Jan me

Stronger than Karel seemed Jan to-me

b. [Langer dan Karel] leek Jan me

Taller than Karel seemed Jan to-me

(8)

a. [Meer gebrand op revanche dan Karel] leek Jan me

More keen on revenge than Karel] seemed Jan to-me

b. [Meer ingenomen daarmee dan Karel] leek Jan me

More pleased with-it than Karel] seemed Jan to-me

Although, in general, adjectives either take the analytic option or the periphrastic option for comparative formation, there are a number of adjectives which permit both options. 6

(9)

a. Jan is [veel meer vatbaar voor de griep

Jan is [much more susceptible of the influenza

dan Karel]

than Karel

b. Jan is [veel vatbaar-der voor de griep

Jan is [much susceptible-COMPAR of the influenza

dan Karel]

than Karel

(10)

a. Jan was [meer benieuwd naar de voetbaluitslagen

Jan was more curious about the soccer-results

dan Karel]

than Karel

b. Jan was [benieuwder naar de voetbaluitslagen

Jan was curious-COMPAR about the soccer-results

dan Karel]

than Karel

In a standard AP-analysis, it is generally assumed that the bound comparative morpheme -er is base-generated in the spec-position of AP (cf. Emonds (1976) for English), that is, the position that is also occupied by meer. The complementary distribution of -er and meer, as exemplified in (11), is in line with such an assumption. 7

(11)

a. *Jan is [meer vatbaar-der voor de griep dan Karel]

b. *Jan was [meer benieuw-der naar de voetbaluitslagen dan Karel]

If the bound morpheme occupies [Spec, AP], there are two options for deriving the analytic comparative forms in (7), (9b) and (10b) under the lexical head hypothesis: either by moving the bound morpheme rightward to the adjectival head or by moving the adjectival head leftward to the specifier position and adjoining it to -er (cf. (12)). Clearly, the two movement patterns violate the ban against movement to a non-c-commanding position. 8

(12)

[_AP [_DegP -er] [_A' [_A vatbaar]]]

Under the DegP-hypothesis, on the contrary, the comparative forms (e.g. vatbaar-der) can be straightforwardly derived by head-to-head movement, a general movement operation which also applies in other syntactic contexts (cf. Baker 1988). As is illustrated in (13), the adjectival head is adjoined to the c-commanding functional head:

(13)

[_DegP [_Deg A°_i[_Deg -er]] [_AP t_i XP]]

The above argument in support of the DegP-hypothesis is reproducable under an analysis in which the analytic comparative adjective is inserted as a lexical item in syntax and requires licensing/checking of the comparative degree feature by the functional head Deg. 9 If checking involves movement of the comparative adjective to Deg, the appropriate structural configuration is the one in (3b) since it involves head movement to (more specifically, substitution into) a c-commanding position. Schematically:

(14)

[_DegP [_Deg e] [_AP A + er XP]]

In the rest of this paper, I will assume that the bound comparative morpheme is part of the adjectival word in the lexicon and that the comparative adjective must raise to the functional head Deg in order to to satisfy license requirements on the comparative degree feature (cf. Sections 3.2. and 4.1. for further discussion).

2.2. Left Branch Extraction Effects

A second argument in support of extending the functional head hypothesis to the adjectival system is based on such left branch extraction facts as in (15) (cf. Ross, 1967; Corver, 1990):

(15)

a. *Hoe_i is Jan [t_i verslaafd aan slaappillen]?

How is Jan addicted to sleeping pills

How much addicted to sleeping pills is Jan?

b. *Hoe_i is die man [t_i behaard]?

How is that man hairy

How hairy is that man?

These ill-formed clauses show that subextraction of a left branch degree word out of an adjective phrase is impossible. As shown by (16), pied piping of the rest of the adjective phrase is required.

(16)

a. Hoe verslaafd aan slaappillen is Jan?

How addicted to sleeping pills is Jan

b. Hoe behaard is die man?

How hairy is that man

The ungrammaticality of the strings in (15) can be easily accounted for under the functional head hypothesis (3b). Movement of the interrogative degree word involves extraction of a zero-level category to the specifier position of CP and hence violates the structure preservation requirement on substitution operations (Chomsky, 1986). Schematically:

(17)

[_CP Hoe_i [_C' [_IP … [_DegP [_Deg' [_Deg t_i] [_AP behaard]]]…]]]

Under the traditional AP-analysis represented in (18), on the other hand, the ungrammaticality of (15) is hard to account for since now it is a maximal projection, namely the entire Degree Phrase in [Spec, AP], which is input to the movement rule (see (18)). Movement of this phrasal category into [Spec, CP] would not yield a violation of the structure preservation requirement on substitution. In short, the ill-formedness of the left branch extractions in (15) remains unaccounted for under a lexical head hypothesis.

(18)

[_CP Hoe_i [_C' [_IP … [_AP [_DegP t_i] [_A'. verstandig]]…]]]

It should be noted at this point that removal of left branch maximal categories out of adjective phrases is allowed in Dutch. This is shown, for example, by the following examples, which differ minimally from those in (15): 10

(19)

a. Hoe erg_i is Jan [t_i verslaafd aan slaappillen]?

How much is Jan - addicted to sleeping pills

How much is Jan addicted to sleeping pills?

b. Hoe zwaar_i denk je dat die man [t_i behaard] is?

How heavily think you that that man - hairy is

How hairy do you think that man is?

In these examples, the degree word hoe combines with the adjectives erg and zwaar; the resulting forms hoe erg and hoe zwaar function as modifiers of the adjectival heads verslaafd and behaard, respectively.

Under the functional head hypothesis, the contrast between (15a,b) on the one hand and (19a,b) on the other hand can be related to a difference in the categorial status of the extracted left branch constituent. Extraction of hoe is X°-movement (i.e. Deg°), whereas extraction of hoe erg/hoe zwaar involves XP-movement (i.e. removal of a left branch DegP-adjunct). 11 Under a lexical head hypothesis, the contrast between the left branch extractions in (15) and those in (19) remains unexplained: both involve removal of a left branch maximal category.

2.3. Extraction from [Spec, DegP]

Another argument in support of the DegP-hypothesis is based on the Dutch extraction phenomena in (20) and (21), involving wh-movement of a measure phrase contained in the specifier position of the Degree Phrase.

(20)

a. Hoeveel cm te klein_i denk je dat ze t_i was?

How many cm too small think you that she - was

b. *Hoeveel cm te denk je dat ze klein was?

c. Hoeveel cm_i, denk je dat ze [t_i te klein] was?

(21)

a. Hoeveel cm minder lang_i, denk je dat Jan daardoor

How many cm less tall think you that Jan because-of-that

t_i is geworden?

- has become

b. *Hoeveel cm minder denk je dat Jan daardoor lang is geworden?

c. Hoeveel cm_i, denk je dat Jan daardoor [t_i minder lang] is geworden?

The c-examples show that movement of only the measure phrase is permitted. 12

The a-examples illustrate that the entire adjectival phrase can be pied piped. The relevant examples which favor the functional head hypothesis are the b-examples. Here the sequence ‘measure phrase -degree word’ is extracted, yielding an ungrammatical sentence. Under a phrase structural analysis as in (23), the ungrammaticality is directly explained by the fact that non-constituents cannot be input to wh-movement. 13 Under the lexical hypothesis, represented in (22), the ill-formedness of the (20b, 21b) remains a mystery. If you can move the lower measure phrase and if you can pied pipe the entire adjective phrase, why should movement of the entire Degree Phrase (a maximal category) out of [Spec, AP] be blocked?

(22)

a. [_AP [_DegP hoeveel cm te] [_A' klein]]

b. [_AP [_DegP hoeveel cm minder] [_A' lang]]

(23)

a. [_DegP [hoeveel cm] [_Deg' te [_AP klein]]]

b. [_DegP [hoeveel cm] [_Deg' minder [_AP lang]]]

2.4. Distribution of ‘Free Adverbs’

A fourth argument in support of the DegP-hypothesis comes from the distribution of so-called ‘free adverbs’ (e.g. adverbs like ongeveer ‘approximately’ and precies ‘precisely’), which exhibit a rather free distribution with respect to the wh-phrase they modify. These adverbs appear either left-adjoined or right-adjoined to the interrogative maximal category they modify. 14 This is illustrated below for Dutch:

(24)

a. [[Ongeveer hoe diep] onder de grond] ligt het lijk?

Approximately how deep under the ground lies the body

a'. [[Hoe diep ongeveer] onder de grond] ligt het lijk?

b. [Ongeveer waar] heb je dat boek gevonden?

Approximately where have you that book found

b'. [Waar ongeveer] heb je dat boek gevonden?

c. [[Ongeveer hoe goed] daartegen bestand] is Marie?

Approximately how well there-to resistant is Marie

c'. [[Hoe goed ongeveer] daartegen bestand] is Marie?

In (24a,a'), the free adverb either occurs to the left or to the right of the wh-phrase hoe diep, which is contained within the PP. In (24b,b'), the free adverb is left- or right-adjoined to the wh-phrase waar, which occupies the [Spec, CP] position. (24c,c') exemplifies the possibility of having a free adverb either to the left or to the right of the left branch interrogative DegP-adjunct hoe goed modifying the adjective bestand.

Now if these free adverbs can occur on both sides of a left branch wh-phrase, the same would be expected to be possible under a traditional lexical head analysis with the bare interrogative degree element hoe

(‘how’), which is analyzed as a maximal category (DegP) occupying the [Spec, AP] position. Schematically, we would have the following structures:

(25)

a. [_AP [_DegP (Ongeveer) hoe (*ongeveer)] [_A' lang]] is Bill?

(Approximately) how (approximately) tall is Bill

b. [_AP [_DegP (Ongeveer) hoe (*ongeveer)] [_A' afhankelijk

(Approximately) how (approximately) dependent

daarvan]] is Bill?

on-it is Bill

As shown by these examples, however, it is impossible for the free adverb to occur to the right of the interrogative element hoe, separating the latter from the adjectival head. It is unclear under the lexical head hypothesis why this is so: The free adverb would simply be right-adjoined to a maximal category (viz. DegP), on a par with the prime-examples in (24). 15

Under the functional head hypothesis, the ill-formedness of the string hoe ongeveer lang is directly accounted for. The constituent lang is the AP-complement selected by the degree word how. Given the sisterhood requirement on head-complement relations (Chomsky 1986), the AP lang should occur as a sister of the selecting functional head Deg°. However, in the sequence hoe ongeveer lang the sisterhood requirement is violated: Lang is not a sister of Deg due to the intervention of the free adverb ongeveer that is right-adjoined to the maximal category DegP. 16

As expected, besides the sequences ongeveer hoe lang in (25a) and ongeveer hoe afhankelijk daarvan in (25b), the strings hoe lang ongeveer and hoe afhankelijk daarvan ongeveer are quite acceptable. Under a DegP-hypothesis, the former sequences involve left adjunction of the free adverb to DegP, the latter right adjunction to DegP. Schematically:

(26)

a. [_DegP ongeveer [_DegP hoe [_AP lang]]]

b. [_DegP [_DegP hoe [_AP lang]] ongeveer]

2.5. Negative Polarity

A negative polarity item is licensed when it occurs in the c-command domain of certain scope bearing elements, such as negation or negative quantifiers (Klima 1964; Hoekstra 1991). 17 This is illustrated by the following examples from Dutch (niets = ‘nothing’; ook maar iets = ‘anything’):

(27)

a. *Jan is [banger voor ook maar iets anders]

Jan is more-afraid of anything else

b. Niemand was [banger voor ook maar iets anders

Noone was more-afraid of anything else

dan deze spinnen]

than these spiders

Interpretation: These spiders scared everyone more than anything else

c. Niemand was [ook maar iets banger voor spinnen

No one was anything more-afraid of spiders

dan Karel]

than Karel

No one was anymore afraid of spiders than Karel

d. *Jan was [ook maar iets banger voor niets anders]

Jan was anything more-afraid of nothing else

*Jan was anymore afraid of nothing else

Sentence (27a) is ungrammatical because there is no c-commanding

negative item in the clause, which could license the polarity item ook maar iets anders. (27b) is well-formed, since the negative subject has c-command over the polarity item contained in the PP-complement of the adjective. In (27c) a negative subject c-commands and hence licenses the polarity item occupying the specifier position of the Degree Phrase. (27d) is ruled out, since the polarity item, now contained in the specifier position of the Degree Phrase, is not c-commanded by the negative item that is part of the PP-complement.

Consider next the following examples in which the negative element is contained in the specifier position of the Degree Phrase and the polarity item is part of the PP-complement of the adjective (see (28a)) and of the prepositional dan-phrase (see (28b)). 18

(28)

a. Jan is [niets banger voor ook maar iets anders]

Jan is nothing more-afraid of anything else

Jan is no more afraid of anything else

b. Jan is [niets dommer [dan ook maar iemand

Jan is nothing more stupid than anyone

in z'n klas]]

in his class

Jan is no more stupid than anyone in his class

The traditional AP-analysis and the DegP-hypothesis make different predictions with respect to the well-formedness of these sentences. The former (i.e. the lexical head hypothesis) assigns the structures (29a) and (29b), respectively, to these sentences. 19 In these structures, the negative

item has no c-command over the polarity item, and the sentences are incorrectly predicted to be out. Under the DegP-hypothesis (i.e. the functional head hypothesis), the sentences (28a and b) are assigned the representations (30a,b) respectively. In these representations, the negative element c-commands the polarity item. Hence, the sentences are correctly predicted to be well-formed. 20

(29)

a. [_AP [_DegP niets -er] [_A' bang [voor ook maar iets anders]]]

b. [_AP [_AP [_DegP niets -er t_j] [_A' dom]] [dan ook maar iemand in z'n klas]_j]

(30)

a. [_DegP niets [_Deg' -er [_AP bang [voor ook maar iets anders]]]]

b. [_DegP niets [_Deg' [_Deg' -er [_AP dom]] [dan ook maar iemand in z'n klas]]]

This concludes my discussion of the empirical arguments in support of extending the functional head hypothesis to the adjectival system. All in all, there seems to be a sufficient empirical basis for adopting the DegP hypothesis.

3. Towards a Split Degree System

In the previous section, it was assumed that the class of functional degree words heading the DegP consists of such items as in (31):

(31)

zo, te, hoe, even, meer, minder, genoeg

zo, too, how, as, more, less, enough

The hypothesis that all items in (31) belong to one and the same class of function words, viz. the class of degree words (Deg°), will be refered to as the uniform degree system hypothesis (cf. Jackendoff 1977). In this section, I will argue against a uniform treatment of the degree words in (31) and propose that a distinction should be made between two types of

functional degree words: Deg° and Q°. 21 The former consist of such items as in (32a) and the latter of such elements as in (32b): 22

(32)

a. zo, te, hoe, even (Deg)

b. meer, minder, genoeg (Q)

I further propose that this split in the functional degree system is reflected in phrasal structure. More specifically, I will assume that adjectival structures introduced by Deg° (e.g. te lang; ‘too tall’) have a structure like (33a) and that those introduced by Q° (e.g. minder lang; ‘less tall’) have a structure like (33b): 23, 24

(33)

a. [_DegP te [_QP e [_AP [_A' lang]]]

b. [_QP minder [_AP [_A' lang ]]

Lexical items of the categorial type Deg or Q carry the semantic content of specifying the degree or extent of the property denoted by the adjectival predicate. This degree can be interpreted as a realization of a property along a scalar dimension of comparison (cf. Zwarts 1992). If the degree specification is realized by Q°, the property denoted by the adjective is determined quantificationally, i.e. in terms of the extent to which a property is present. The property of being tall, for example, can be manifested in different degrees of tallness. In comparative forms the degree either exceeds some point on the tallness-scale (as in langer dan Jan ‘taller than Jan’) or is lower than some point on the scale of degrees (as in minder lang dan Jan ‘less tall than Jan’).

In the case of degree specification by Deg°, the property denoted by the adjective is realized in a more identificational way. This is most clearly illustrated in (34a), where the demonstrative degree word zo identifies a point on the scale of degrees. 25 The degree word even in (34b) identifies a point on the scale of degree of tallness which equals 2 meters. In (34c), finally, the interrogative degree word asks for a point on the scale of degrees of tallness.

(34)

a. Jan is 2 meter lang. Niemand anders in mijn klas is zo lang

Jan is 2 meters tall. No one else in my class is that tall

b. Jan is 2 meter lang. Dan is hij even lang als Karel

Jan is 2 meters tall. Than is he as tall as Karel

c. Hoe oud is Jan precies? Hij is 2 jaar en 3 dagen oud

How old is Jan exactly? He is 2 years and 3 days old

This identificational, referential function of Deg, exemplified in (34), is reminiscent of the referential role of determiners within the DP-projection, whence Bresnan's (1973) characterization of such items as being ‘determiner-like’.

Another phenomenon which is at least suggestive of the determiner-like nature of Deg° is the occurrence of the clitic definite article in superlative forms (cf. (35)). 26 I assume that this clitic occupies the Deg° position and that superlative morphology is associated with the Q° position: [_DegP 't [_QP -st [_AP lang]]]. 27 Semantically, this clitic has the same function as the definite article in nominal predicative phrases like (36) (cf. Stowell, 1991):

(35)

Marie is ['t domst]

Marie is the-NEUT stupid-SUPERL

Marie is the stupidest

(36)

Marie is ['t lievelingetje van de leraar]

Marie is the darling-DIMIN of the teacher

Presence of the definite article in the noun phrase has the effect that the nominal predicate designates a property of uniqueness. In the same way, presence of the definite determiner in superlative adjective phrases has

the semantic effect of assigning a property of uniqueness to the degree of stupidity denoted by the adjectival expression. 28

A full discussion of the (different) semantics of determiner-like degree words and quantifier-like, degree words is beyond the scope of this article and definitely af topic of further research. 29 As will become clear in the next section, there are also syntactic asymmetries between Deg and Q in the adjectival system, which are quite similar to certain asymmetries found between Det and Q in the nominal domain. This further strengthens the idea of drawing a parallel between Det and Deg, on the one hand, and adjectival Q and nominal Q, on the other hand.

3.1. Syntactic Asymmetries Between Deg and Q

What evidence is there in support of the split degree system hypothesis and against the uniform degree system hypothesis? Notice that under the latter hypothesis one would expect a symmetric behavior of the items belonging to the degree system. It turns out, however, that the quantifier-like items (Q°) and the determiner-like degree items (Deg°) behave differently in various ways, suggesting that they should not be treated on a par.

A first asymmetry suggesting a split in the class of functional degree words relates to adjectival structures in which part of the adjective phrase has been pronominalized. Before turning to such examples, consider first the examples in (37), which illustrate that the entire adjective phrase can be replaced by a pro-form (cf. Ross, 1969):

(37)

a. Bang voor honden, Jan is 't gelukkig nooit geweest

Afraid of dogs, Jan has it fortunately never been

Fortunately, Jan has never been afraid of dogs

b. Jan is nogal gevoelig voor kritiek, wat ik Marie overigens

Jan is rather sensitive to criticism, what I Marie by-the-way

ook vind

also consider

Jan is rather sensitive to criticism, which I believe Mary is too

c. Jan was bekend met die problematiek. Piet leek

Jan was acquainted with these problems. Piet seemed

me dat toendertijd niet.

to-me that then not.

Jan was acquainted with these problems. At the time, Piet didn't seem to be so.

In the Hanging Topic Left Dislocation construction (37a), the neuter clitic pro-form 't substitutes for the predicative adjective phrase bang voor honden. In (37b), the pronominal relativizer wat of the appositive relative clause takes the adjective phrase nogal gevoelig voor kritiek as its antecedent. In (37c), finally, the demonstrative pronoun dat refers to the adjective phrase bekend met die problematiek of the previous utterance.

Consider next the following examples, in which these adjectival pro-forms replace part of the adjective phrase: 30

(38)

a. Bang voor honden, Jan is 't_i tegenwoordig gelukkig

Afraid of dogs, Jan is it at-present fortunately

[_QP een stuk [_Q' minder [_AP t_i]] dan vroeger]

a lot less than in-the-past

b. Jan is gevoelig voor kritiek, wat_i ik Marie

Jan is sensitive to criticism, what I Marie

overigens [_QP een stuk [_Q' minder [_AP t_i]]] vind

by-the-way a lot less consider

c. A: Is Marie bekend met die problematiek?

A: Is Marie acquainted with these problems

B: Ze leek 't_i me in ieder geval [_QP meer [_AP t_i] dan

B: She seemed it to-me in any case more than

haar man]

her husband

d. Begerig naar nieuwe ontdekkingen, Jan lijkt me dat_i

Eager for new discoveries, Jan seemed to-me that

[genoeg t_i om een groot wetenschapper te worden]

enough for a great scientist to become

The part which is not pronominalized in these adjectival projections is the quantifier-like degree expression and, if present, the comparative dan-phrase and the expression modifying the quantifier (e.g. een stuk in (38a)). As indicated by the labelled bracketings, I assume that the pro-forms

replace the lexical phrase AP and are reordered out of the extended adjectival projection QP, stranding the quantifier and, possibly, material accompanying it.

As illustrated in (39), adjective phrases introduced by a functional degree word of class (32a) do not permit partial replacement by a pronominal form:

(39)

a. *Bang voor honden, Jan is 't_i helaas [even t_i als Piet

Afraid of dogs, Jan is it unfortunately as as Piet

b. *Jan is gevoelig voor kritiek, wat_i ik Marie overigens ook

Jan is sensitive to criticism, what I Marie by-the-way also

[veel te t_i] vind

much too consider

c. A: Is Marie bekend met die problematiek?

A: Is Marie acquainted with these problems

B: *Ze is 't_i [zo t_i dat ik haar niet hoefde in te lichten]

B: She is it [so that I her not needed PRT to inform

d. A: Is Jan begerig naar nieuwe ontdekkingen?

A: Is Jan eager for new discoveries?

B: Ja hoor, hij is [xxx begerig ernaar]

B: Yes, he is [<unintelligible speech> eager for-it

A: *Hij is 't_i [HOE t_i]?

A: He is it HOW

He is HOW eager for new discoveries?

So, there is a clear asymmetry between the functional degree words in (32a) and those in (32b) with respect to the phenomenon of partial ‘pronominalization’ of the adjective phrase. The former, which I have characterized as being ‘determiner-like’, block partial pronominalization, whereas the latter, i.e. the quantifier-like ones, permit it.

Interestingly, a similar contrast is found within the nominal domain. That is, partial pronominalization by the demonstrative die (that/those) is permitted with noun phrases introduced by a quantifier (Q) but not with those introduced by a determiner (Det). This contrast is illustrated in (40)-(41):

(40)

a. Auto's, Jan bezit die_i [een stuk minder t_i dan Piet]

Cars, Jan owns those a lot less than Piet

b. Postzegels uit China, ik geloof dat je die_i nu [meer t_i]

Stamps from China I believe that you those now more

hebt dan ik

have than I

(41)

a. ?*Auto's, Jan bezit die_i [deze t_i]

Cars, Jan owns those these

b. *Postzegels uit China, ik geloof dat je die_i nu

Stamps from China I believe that you those now

[al de t_i] hebt

all the have

I believe that you have all stamps from China now

These facts point out to a certain parallelism between Det and Deg, on the one hand, and adjectival Q and nominal Q, on the other hand.

A second asymmetry supporting the split degree system hypothesis comes from the phenomenon of split topicalization with Dutch adjectival phrases. The phenomenon of split topicalization has mainly been discussed for the nominal system and refers to those constructions in which part of the noun phrase has been topicalized, leaving behind some specifying element (Van Riemsdijk, 1989). As shown by the examples in (42) and (43), there is an asymmetry in the strandability of determiners and quantifiers in split topicalization contexts. In Dutch, quantifiers can be stranded, but bare determiners can not.

(42)

a. Boeken_i denk ik dat hij [meer t_i dan Piet] heeft

Books think I that he more - than Pete has

I think he owns more books than Pete does

b. Boeken_i heeft hij [(meer dan) genoeg t_i]

Books has he (more than) enough -

He has (more than) enough books

(43)

a. *Boeken_i heeft ie [de t_i]

Books has he the

b. *Boeken_i heeft ie [die t_i]

Books has he those

A similar contrast is found within the adjectival system. The examples

in (44) show that part of the adjective phrase can be fronted if a quantifier-like degree element is stranded. As is indicated by the ill-formedness of the examples in (45), split topicalization is not permitted if the element left behind is a determiner-like degree item (Deg°) like te, zo etc. 31 In short, we have another clear asymmetry between determiner-like degree words (i.e. Deg°) on the one hand and quantifier-like elements (i.e. Q°) on the other, suggesting a non-uniform treatment of the two types of degree elements. 32

(44)

a. Bang voor honden_i denk ik dat hij [een stuk minder t_i

Afraid of dogs think I that he a lot less -

dan Piet] is

than Piet is

b. Gebrand op revanche_i leek Jan mij toen [meer t_i

Keen on revenge seemed Jan to-me then more -

dan Piet]

than Piet

(45)

a. *Bang voor honden_i is hij [veel te t_i]

Afraid of dogs is he much too -

b. *Gebrand op revanche_i was Jan [zo t_i dat het een

Keen on revenge was Jan so - that it an

obsessie werd]

obsession became

To conclude, I have provided empirical justification for distinguishing two types of functional degree words, viz. Deg and Q. The former category turned out to behave similarly in certain respects to the functional category. Det in the nominal system, whence its characterization as a determiner-like functional degree word; the latter category displayed properties similar to those of (weak) quantifiers within the nominal system.

3.2. Dummies and degrees

Having provided empirical support for distinguishing two classes of functional degree words, I will return to the structures in (33a,b), repeated here as (46a,b). 33

(46)

a. [_DegP te [_QP e [_AP [_A' lang]]]]

b. [_QP minder [_AP [_A' lang]]]

(46a) structurally represents the assumption that the category Deg selects a QP, whose head in turn selects AP. In (46b), which represents adjective phrases mtroduced by elements of the class in (33b), the Deg-projection is assumed to be missing.

What evidence is there for the structure in (46a)? And how do we account for the ill-formedness of such sequences as in (47), where the adjectival projection contains both a lexical item of the category Deg and one of the category Q?

(47)

a. *[_DegP te [_QP minder [_AP lang]]]

too less tall

b. *[_DegP even [_QP meer [_AP begaan met ons lot]]]

as more feeling-sorry-for our destiny

In what follows, I will provide empirical support for the articulated structure in (46a) and account for the ill-formedness of the sequences in (47).

The idea that Deg selects a quantifier-like element (which in turn selects an AP) goes back to the traditional assumption that a phrase like too tall (or its Dutch equivalent te lang) represents the underlying concept x-much tall, where x represents a degree of tallness, expressed here by too/te (cf. Bresnan, 1973; Creswell, 1976). In Bresnan (1973), it is proposed that this quantifier much is always present underlyingly and undergoes an obligatory rule of much-deletion when much is immediately followed by the adjectival head. Thus, the string too tall transformationally derives from too much tall.

Rather than assuming that much (or its Dutch equivalent veel) is always present underlyingly, I propose that the quantifier much/veel acts as a dummy adjectival element, which is only inserted in the Q-position of the extended adjectival projection as a last resort, i.e. ‘to save’ an underlying adjectival structure yielding no output (cf. Corver (1997)). I will further assume much/veel-insertion is a language-particular rule whose application is more costly (i.e. less economical) than that of a UG-operation (e.g. head movement). 34 Hence, if an adjectival structure can be saved by the application of some UG-operation Z or the language-particular rule of much-insertion, the former is to be prefered since it bears a smaller cost.

Let us see now in what adjectival structures the rule of much/veel-insertion applies and how it provides evidence for the articulated structure in (46a). For the sake of illustration and given the limited contexts in which much/veel-insertion applies, I will draw examples both from English and Dutch.

For English, the last resort nature of much-insertion is nicely illustrated by the examples in (48). In (48a), where we have the adjectival head fond, the dummy much must be absent. In (48b), however, where the projection AP has been substituted for by the pro-form so, the dummy much must be present. The ill-formed (48c), finally, shows that presence of the dummy much is dependent on the presence of a Deg-element like too.

(48)

a. John is [too (*much) fond of Mary]

b. John is fond of Mary. Maybe he is [too *(much) so]

c. *John is [much so]

What is important is that the string too much so in (48b) provides direct evidence for the articulated structure in (46a):

(49)

[_DegP too [_QP much [_AP so]]]

The facts in (48) are highly reminiscent of the do-support phenomenon in the extended verbal domain. The do-support strategy is not resorted to in declarative clauses when a main verb is present (cf. (50a)). When the VP-complement is pronominalized by so, as in (50b), the dummy element do must appear.

(50)

a. *John did kiss her

b. John kissed her, and Bill did so too

The appearance of the verbal element do and the adjectival element much in so-pronominalization contexts suggests a similar function of the two elements. In Chomsky (1991), it is claimed that the do-support strategy is a language-particular rule and, as such, is more costly than a UG-operation like V-to-I head movement. That is, if tense features can be checked off by raising a finite verb to Tense, then this computational operation is to be prefered over Tense-feature checking by the less economical, language-particular strategy of do-support. If V-to-I raising cannot apply (e.g. because the VP has been pronominalized by so), the do-support strategy is resorted to in order to save the extended verbal structure.

Along the same lines, I will assume that much-support is a language-particular rule which only operates in an adjectival structure if that structure cannot be saved by the application of some universal and therefore more economical computational operation. The question then is: Which UG-process blocks the application of much-support in such adjectival contexts as (48a)? I propose that this process is the A°-to-Q° raising operation, a substitution operation which raises the adjectival head into the empty functional Q-position. Schematically:

(51)

[_DegP too [_QP [e] [_AP fond of Mary]]]

In adjectival structures in which the AP-complement is substituted for by the pro-form so, there is no adjectival predicate available which can be input to the A-to-Q raising operation. In such a case, resort must be taken to the rule of much-support, yielding a structure as in (49).

The next question to ask is: What forces the application of A-to-Q raising and the rule of much-support? I propose that these operations take place in order to overcome a violation of the principles of thematic discharge (Higginbotham 1985), which ultimately fall under the overarching condition of Full Interpretation. One of the processes of thematic discharge is theta-binding. This process relates the open referential argument position of a lexical predicate to an operator, this way restricting the predicate's denotation. Drawing an analogy with the verbal and nominal system (cf. Higginbotham 1985, Williams 1981), where the functional heads T(ense) and D° are considered operators which bind the ‘referential’ argument positions E(vent) and R, respectively, of the thematic grids associated with V and N, I assume that Deg° functions as an operator which must theta-bind a referential argument position of the thematic grid associated with an adjectival predicate. With Zwarts (1992), I assume that this is an argument position over degrees, which will be refered to by G(rade). Thus, the gradable adjective fond has a lexical entry like (52):

(52)

fond, +V +N, <1, 2, G>

This thematic grid contains three argument positions: the thematic argument positions (1 and 2), and the referential argument position G. Since the referential argument is open in (52), the adjective denotes each of the degrees of ‘tallness’ (i.e. a set of degrees). The reference of the gradable adjectival predicate is restricted when it is theta-bound by a functional head which acts as a binder of the referential argument position. Theta-binding of the G-variable by Deg realizes, as it were, the property denoted by the adjectival predicate along a scalar dimension of degrees (cf. Zwarts 1992).

Thus far, I have proposed that A°-to-Q° raising is enforced by requirements of thematic discharge (i.e. theta-binding of the referential G-argument). With Higginbotham (1985), I assume that theta-binding is a local licensing relation between a functional degree operator and the referential argument G, which is part of the gradable adjective's thematic grid. I propose that the locality of the theta-binding relation is expressed in terms of the local head-head relation: A functional head Deg° can only theta-bind the referential argument of the gradable adjectival predicate, if the latter heads the minimal complement of the functional head Deg.

(53)

[_DegP too_i [_QP fond_<Gi>k [_AP t_k of Mary]]]

In short, A°-to-Q° raising creates the appropriate configuration for theta-binding of the degree argument G (i.e. the variable) by the Deg-operator (i.e. the theta-binder).

In what way does much-support save an adjectival expression like [_DegP too [_QP e [_AP so]]]? I propose that analogously to the dummy verb do, whose traditional interpretation is that of a substitute for the main verb (cf. Chomsky, 1955; Pollock, 1989), the dummy quantifier much functions as a substitute for the adjectival predicate. More specifically, the dummy adjectival quantifier copies the referential degree argument position G associated with the pro-form so. This way, much-support rescues the structure [_DegP too [_QP e [_AP so]]], because it enables theta-binding of the referential degree argument by the Deg-operator: the dummy element, carrying the copied degree argument, enters into a local head-head relation with the c-commanding Deg-operator. Schematically: 35

(54)

[_DegP too_i [_QP much_<Gi> [_AP so_<1,2,Gi>]]]

Thus, in (54), the referential argument G associated with the pro-form so is bound via its copy which is carried by the dummy much. Much-support creates the appropriate local configuration for theta-binding: the Deg-operator enters into a local head-head relation with the dummy quantifier much. 36

Having justified the articulated structure in (46a), I will next consider a structure like (46b), which is repeated here as (55) for English:

(55)

[_QP less_i [_AP fond_<1,2,Gi> of Mary]]

As indicated by the coindexation, I assume that the quantifier less also functions as an operator theta-binding the degree-argument G. Notice that (55) has the right local relationship for establishing a theta-binding relationship between the Q-operator and the adjectival predicate carrying

G in its thematic grid. This local relation is also found in such adjectival expressions as in (56), where the quantifier less takes a so-complement. 37

(56)

John is fond of Mary, but he is [less_i so_<1,2,Gi> than Bill]

Consider now the following ill-formed structures in which the adjectival projection contains both a (non-dummy) lexical item of the category Q and a lexical item of the category Deg (cf. also the Dutch structures in (47)).

(57)

a. *John is [_DegP too [_QP less_i [_AP fond_<1,2,Gi> of Mary]]]

b. *John is [_DegP too [_QP less_i [_AP so_<1,2,Gi>]]]

The ill-formedness of these adjectival expressions is accounted for straight-forwardly. These expressions contain two potential theta-binders for the degree-argument G. The degree word less in Q stands in a local head-head relation with fond/so and therefore is able to theta-bind the degree variable G. Since G is already bound by less, the Deg-operator too remains vacuous and, hence, forms an illegitimate object at LF. 38

Our discussion thus far has shown that at least in English there is overt evidence for the co-occurrence of Deg and Q in the extended adjectival projection. Since this paper focuses on the internal syntax of Dutch adjective phrases, the question arises whether independent support can be given for the presence of a QP-projection in adjectival structures introduced by a

Deg-operator. Support for this assumption might come from the following facts: 39

(58)

a. Ik vond Jan [_DegP iets [_Deg' te [_QP *(veel)

I considered Jan somewhat too (much)

[_AP daarvan afhankelijk]]]]

on-it dependent

I considered Jan a bit too much dependent on that

b. Ik vond Jan [_QP (*veel) [_AP daarvan afhankelijk]]

I considered Jan (much) that-on dependent

(58a) shows that presence of veel is required in such sequences as ‘te + PP-complement + A’. (58b), furthermore, suggests that this quantifier veel is a dummy element. Its presence in the adjectival structure is dependent on the presence of the Deg-head te. When the extended adjectival projection contains no Deg-projection, veel does not show up.

I assume that the dummy element veel in (58a) fulfills the same function as the English dummy much: it copies the referential degree argument G associated with the predicate of the lexical projection AP, this way creating the right local configuration for establishing a theta-binding relation between the Deg-operator te and the degree variable G. Schematically:

(59)

[_DegP iets [_Deg' te_i [_QP veel_<Gi> [_AP daarvan afhankelijk_<1,2,Gi>]]]]

Interestingly, veel-support can be absent when the order of the PP-complement daarvan and the adjective afhankelijk is reverse:

(60)

iets te afhankelijk daarvan

Let us assume for the moment that in a string like (60), the adjectival predicate has undergone A°-to-Q° raising. 40 The derived structure provides the appropriate local configuration for theta-binding of G by the Deg-operator te.

(61)

[_DegP iets [_Deg' te_i [_QP afhankelijk_<Gi>k[_AP daarvan t_k]]]]

If, however, the string in (60) is derived in the way depicted in (61), then the question arises why the language particular and hence more costly rule of veel-support in (58a) is not blocked; application of A-to-Q raising is cheaper than insertion of the dummy element veel. In Section 7.2, I will come back to this matter and show that different derivations underlie the adjectival expressions in (58a) and (61). What is important right now is that (58a) provides evidence for the possible co-occurrence of DegP and QP within the Dutch extended adjectival projection.

3.3. Summary

To summarize Section 3, I have provided empirical evidence for the distinction between two types of functional degree words: words of the category Deg° and those of the category Q°. Both Deg and Q (modulo dummy much/veel) function as operators which must theta-bind a referential degree argument G contained within the thematic grid of the gradable adjectival predicate. It was further observed that Deg and (adjectival) Q display certain grammatical properties which are quite similar to those of the nominal functional categories Det and Q, respectively. The language-particular rule of much/veel-support provided overt evidence for the co-existence of the two functional projections DegP and QP within the extended adjectival projection. Insertion of the dummy much/veel enables the Deg-operator to enter into a local (i.e. head-head) theta-binding relation with the degree argument G. If the local relation can be created in a more economical way, e.g. via raising of A to Q, then this is to be prefered. Let me, finally, point out that it also seems conceptually attractive to adopt the more articulated structure [_DegP Deg [_QP Q [_AP A]]]. In many recent studies on the syntax of noun phrases (e.g. Abney, 1987; Lobeck, 1991; Ritter, 1991; Watanabe, 1991; Giusti, 1991; Shlonsky, 1991), a quite similar structure has been proposed for the nominal system, viz. one in which, besides the topmost DP-projection, there is a separate QP-projection which is headed by a quantifying element, i.e. [_DP D [_QP Q [_NP]]] (e.g. [_DP those [_QP three [_NP cars]]]). Hence, the articulated adjectival structure proposed in this section strengthens the phrase structural parallelism between the adjective phrase and the noun phrase.

4. A°-to-Q° Raising

In the previous section, I defended the view that in a string like [_DegP te [_QP e [_AP [_A' lang]]] (‘too tall’), the gradable adjective substitutes for Q so that its degree argument G can be theta-bound by the Deg-operator te. Clearly, such a movement operation applies in a string vacuous way. Since the head raising operation is not visible by itself, the question arises whether there is any independent and preferably overt evidence for the existence of A-to-Q raising in Dutch. In this section, I will discuss three phenomena suggesting the existence of this movement operation.

4.1. P-stranding

A first piece of evidence for the existence of an A°-to-Q° movement operation comes from P(reposition)-stranding effects in Dutch with PP-complements of adjectives. As shown by the following examples, PP-complements can either occur to the left or to the right of a positive adjective: 41

(62)

a. Jan is volgens mij [(daarop) verliefd (daarop)]

Jan has according to-me (there-with) in-love (there-with)

geweest

been

Jan has been in love with her/him

b. Jan is [(daarvoor) gevoelig (daarvoor)] gebleken

Jan has (there-to) sensitive (there-to) turned-out

Jan turned out to be sensitive to that

Pre- and post-adjectival occurrence of the PP-complement is also found with analytic comparative adjectives:

(63)

a. Jan is volgens mij [(daarop) verliefder (daarop)]

Jan has according to me (there-with) more-in-love (there-with)

geweest dan Sue

been than Sue

Jan has been more in love with her/him than Sue has

b. Jan is [(daarvoor) gevoeliger (daarvoor)] gebleken

Jan has (there-to) more-sensitive (there-to) turned-out

dan Sue

than Sue

Jan turned out to be more sensitive to it than Sue is

P-stranding is always possible when the PP-complement follows the adjectival head, no matter whether the adjective has the positive form (gevoelig/verliefd) or the comparative form (gevoeliger/verliefder). This suggests that in such cases the PP-complement occupies its L-marked base position. 42, 43

(64)

a. Waar_i is Jan volgens jou

Where has Jan according-to you

[verliefd(er) [_PP t'_i [_P' op t_i]]] geweest?

in-love(-er) - with - been

Who has Jan been more in love with according to you?

b. Waar_i is Jan [gevoelig(er) [_PP t'_i [_P' voor t_i]]] gebleken?

Where has Jan sensitive(-er) - to - been

What did Jan turn out to be more sensitive to?

As shown by the examples in (65), extraction from pre-adjectival PP-complements is more restricted. More specifically, extraction is blocked if the adjectival head has the comparative form:

(65)

a. Waar_i is Jan volgens jou [[t_i op] verliefd/?*verliefder] geweest

b. Waar_i is Jan [[t_i voor] gevoelig/??gevoeliger] gebleken

The possiblity of P-stranding with pre-adjectival complements of positive adjectives suggests that the PP-complement occupies its AP-internal L-

marked base-position. Let us assume, for the sake of the argument, that PP-complements of adjectives are either base-generated to the left or to the right of the adjective (see, however, Section 6). With Zwarts (1992), I will further make the assumption that a bare positive adjective like verliefd in Jan is verliefd op Marie (Jan is in-love with Marie) does not denote a degree (e.g. a degree of being in love with), but rather a property (e.g. the property of being in love with). 44 Since verliefd is potentially gradable, as is clear from its comparative form, I will assume that in the lexicon the thematic grid of verliefd contains an optional degree argument G. Hence, we have something like <1, 2, (G)>, where 1, 2 are the thematic arguments and G the referential degree argument. Under a gradable reading, the G-argument must be discharged in syntactic structure; under a property reading, the G-argument is absent in syntax. I will further assume that the functional structure which is involved in the licensing of G (i.e. Deg(P) and Q(P)) is absent if the adjective has the property-denoting reading. This has the obvious consequence that there can be no A-to-Q raising in adjective phrases headed by adjectives lacking a degree argument. Consequently, the PP preceding the bare positive adjective in (65a) does not occupy a position adjoined to QP, but simply remains in its base position.

The impossibility of extracting from pre-adjectival complements of comparative adjectives suggests that the PP-complement no longer occupies its AP-internal L-marked base position. As a matter of fact, this would follow directly if the comparative adjectival head has been moved to Q° in the course of the derivation. Recall from Section 2.1, that I assume that comparative morphology is attached to the adjectival stem in the lexicon. Since comparative adjectival forms are plainly gradable, the question arises how the degree argument G is discharged in syntax. It is self-evident that the way the degree argument is discharged in analytic comparative forms should be similar to the way in which G is discharged in periphrastic comparative forms. In periphrastic forms, the functional Q-head meer (‘more’) theta-binds G contained in the adjective's thematic grid (cf. (55) for the relevant configuration). For analytic comparative adjectives a similar theta-binding configuration can be obtained if it is assumed that movement involves copying (cf. Chomsky 1992). For a comparative adjective like verliefder this means that raising to Q (involving

substitution) creates a structure like (66a). 45 For convergence at LF, we must have an operator-variable (i.e. theta-binding) structure. Such a structure is obtained if the comparative morpheme -er (i.e. the operator) is the only survivor in the operator position (i.e. Q) and the gradable adjectival stem verliefd (i.e. the non-operator part) survives in the trace position. In the spirit of Chomsky's (1993) approach towards the LF-interpretation of pied-piped wh-phrases, I will assume that at LF, the operator part of the raised comparative adjective (i.e. the comparative morpheme -er) is extracted out of the adjectival head and attached to it, yielding a structure like (66b). A correct theta-binding structure is obtained if in the operator position Q, everything but the operator phrase is deleted (i.e. the gradable adjectival stem) and if in the trace position the adjoined comparative operator is deleted. The resulting operator-variable structure is (66c).

(66)

a. [_QP verliefder_<G> [_AP verliefder_<G> waarop]]

b. [_QP [_A [-er]_i [_A verliefd_<G> t_i]] [_AP [_A [-er]_i [_A verliefd_<G> t_i]] waarop]]

c. [_QP [-er]_i [_AP verliefd_<Gi> waarop]]

If raising of the comparative adjective to Q applies in overt syntax, the only way a PP-complement of an adjective can show up to the left of the adjectival comparative head is by leftward movement of the PP within the adjective phrase. Such a leftward shift will be interpreted in Section 5 as scrambling within the adjective phrase. If scrambling involves adjunction, we end up with the following structure for analytic comparative adjectives with pre-adjectival PP-complements.

(67)

[_QP [waarop]_i [_QP verliefder_j [_AP t_j t_i]

After scrambling, the PP is no longer in a L-marked position and hence forms a barrier for extraction, which explains the contrast with pre-adjectival PP-complements of positive adjectives (cf. (65)). 46

To conclude this subsection, consider also the facts in (68), which provide independent evidence for the fact that the PP-complement preceding the analytic comparative adjective is not in its base position and has been reordered as a result of scrambling within the adjective phrase. What these examples show is that when the PP-complement is in a position preceding the comparative adjective, it can only occur to the left of the modifier veel, which has been argued to occupy the specifier position of QP (cf. Section 2). 47

(68)

a. Jan, [veel gevoelig-er daarvoor], verliet de zaal

Jan, much sensitive-COMPAR to-it, left the room

Jan, who was much more sensitive to this, left the room.

b. *Jan, [veel daarvoor gevoeliger], verliet de zaal

c. Jan, [daarvoor veel gevoeliger], verliet de zaal

4.2. Modifier-Head Agreement

A second empirical argument for locating adjectival degree modifiers in Spec, QP comes from the phenomenon of (optional) agreement (overtly expressed by the presence of -e) between an adjectival degree modifier and the adjectival head in Dutch attributive adjective phrases. 48 This phenomenon is exemplified in (69):

(69)

a. een [erg(e) dure] fiets

a very-(INFL) expensive-INFL bike

a very expensive bike

b. een [ontzettend(e) interessante] opmerking

an extreme-(INFL) interesting-INFL remark

an extremely interesting remark

c. een [belachelijk(e) dure] fiets

a ridiculous-(INFL) expensive-INFL bike

a ridiculously expensive bike

In these adjectival structures, the inflectional morpheme -e is obligatory for the attributive adjectival head but optional for the adjectival modifier. This suggests that it is the former adjectival element which determines the agreement relation.

If agreement is the reflection of a spec-head relation (cf. Koopman

1987; Chomsky 1993), then the agreement pattern in (69) is suggestive of an analysis in which the adjectival degree modifier occupies a specifier position within the extended adjectival projection. I propose that this position is [Spec, QP] and that a proper agreement configuration is established after the inflected adjectival head has raised into the empty Q-slot (cf. (70a)). 49 As for the modification structure lacking agreement (e. g. erg dure in (69a)), I assume that the adjectival degree modifier is not in Spec, QP, but rather in a position adjoined to QP (cf. (70b)). 50

(70)

a. [_QP erge [_Q' dure_i [_AP t_i]]

b. [_QP erg [_QP [_Q' dure_i, [_AP t_i]]]]

Adjectival modifiers that are adjoined to QP are not close enough to the inflected adjectival head to enter into an agreement relation with it. 51 This is also suggested by the examples in (71), where the modifier does not express degree but rather modality (71a), evaluation (71b) or temporality (71c):

(71)

a. een [waarschijnlijk(*-e) dure] fiets

a probable(-INFL) expensive-INFL bike

a probably expensive bike

b. een [gelukkig(*-e) goedkope] fiets

a fortunate(-INFL) cheap-INFL bike

a fortunately cheap bike

c. een [tijdelijk(*-e) goedkope] fiets

a temporary-(INFL) cheap-INFL bike

a temporarily cheap bike

That these modifying elements occupy a position quite high in the extended adjectival projection is clear from such examples as in (72), where they appear in left-peripheral position preceding the functional heads Q and Deg, and the modifier of Q, veel. 52

(72)

a. een [_DegP waarschijnlijk [_DegP veel [_Deg' te dure]]] fiets

a probably much too expensive bike

b. een [_QP [_QP tijdelijk [_QP minder [_Q' [_AP goedkope]]]]

a temporarily less cheap

fiets

bike

Thus, agreement between an adjectival degree modifier and an attributive adjective is only manifested when the two elements stand in a Spec-head relation (with A raised to Q).

What evidence do we have for the assumption that the Spec-head relation is established in overt syntax, for such relations might also be established by raising the inflected attributive adjective in covert syntax? That A-to-Q raising applies overtly is suggested by the word order patterns in table (73), which compares an attributive adjective phrase displaying adjectival agreement with its counterpart not displaying spec-head agreement.

(73) 53.

The third column shows that the adjectival agreement pattern requires adjacency of the two inflected adjectival elements. That is, intervention of a PP-complement (cf. the topmost cell of the ‘agreement’ column) or of a stranded preposition (cf. the middle cell) blocks agreement. 54 The only well-formed, agreement pattern is the one in the lowest cell of the third column, where the two adjectival elements are adjacent. As will be shown in Section 5.2, the PP-complement ends up in a pre-modifier position as a result of leftward scrambling within the adjective phrase. Obviously, the adjacency effect depicted in the third column directly follows from an analysis in which overt agreement is a reflection of a spec-head relation created by overt raising of the attributive adjective to Q.

As shown by the ‘no agreement’-column, there is no adjacency requirement between the non-inflected modifying adjective and the inflected adjective afhankelijke. For the moment, I will confine myself to remarking that this non-agreement pattern suggests that the modifier and the attributive adjective are not in a spec-head relation in overt syntax. I will return to these word order effects in (non)-agreement contexts in Section 7.2.

4.3. genoeg-inversion

The phenomena discussed in Sections 4.1 and 4.2 show in an indirect way that the gradable adjective undergoes overt A-to-Q raising in Dutch, namely by the impossibility of P-stranding out of PP-complements preceding analytic comparative adjectives and by the adjacency requirement on the agreement relation between the adjectival degree modifier and the attributive adjective head. The application of overt raising in these adjectival structures is not visible from a reordering of the raised adjectival predicate with respect to the functional head Q, simply because the raised adjective substitutes for Q. Only when the adjectival head would (left-)adjoin to Q, such a reordering would be visible. A case in point might be the phenomenon of genoeg-inversion (cf. also Hoekstra, 1984), which is illustrated in (74b).

(74)

a. *Jan is [genoeg bang daarvoor]

Jan is enough afraid there-of

Jan is afraid enough of it

b. Jan is [bang genoeg daarvoor]

This inversion pattern has sometimes been interpreted as resulting from a rightward shift of the quantifier, placing it between the quantifier and the complement of the adjective (see e.g. Maling (1983) for Swedish). Such an operation involves lowering, i.e. movement to a non-c-commanding position, and hence should be rejected in view of such requirements as the Proper Binding Condition or the ECP.

Under a functional head analysis, the phenomenon of genoeg-inversion

may be reanalyzed as a leftward head movement operation left-adjoining the adjective to the quantifier: 55

(75)

[_QP bang_i + genoeg [_AP t_i daarvoor]]

This raising operation overtly shows the existence of A-to-Q raising in Dutch. Notice that the adjunction operation does not violate the c-command requirement on the antecedent-trace relation. A question which remains unanswered, though, concerns the trigger for A°-to-Q° raising in this syntactic context: The Q°-position is lexically filled by genoeg, which should be able to function as a local theta-binder for the degree-argument G associated with the adjectival predicate (analogously to a structure like minder bang daarvoor (‘less afraid of it’)). Thus, one would expect adjective raising not to be required, contrary to fact. By lack of any deep explanation of this deviant behavior of the quantifier genoeg, I tentatively propose that this idiosyncratic property of genoeg is encoded in its lexical entry.

5. Directionality of Headedness

Having come to an empirically motivated, more articulated structure of the adjectival system, I will now address the question as to how the headedness parameter is set within the functional system of the Dutch adjective phrase. For the sake of simplicity, I will temporarily abstract away from the distinction between QP and DegP. Hence, the main question will be: Does the degree element Deg° precede or follow the AP-complement within the Degree Projection (DegP)? Although this question would receive a straightforward answer in rigidly head-initial languages, like English, its answer is much less obvious in a language like Dutch in view of the fact that heads do not take their complements in a uniform direction. Certain heads take their complement to the right (e.g. COMP. Det, N), others are often assumed to take their complement to the left (e.g. V, I) (cf. Koster 1987). 56 In what follows, I will explore the two

logical hypotheses, the Deg-initial hypothesis and the Deg-final hypothesis. I will start with the former and come to the conclusion that the head initial analysis is to be prefered.

5.1. Against the Deg-final Hypothesis

According to the head-final hypothesis, Deg° is a head taking its AP-complement to the left in Dutch. The surface order ‘Deg - Adj’ (e.g. te lang (‘too tall’)) is derived by incorporating A° into Deg°, as shown in (76b) (cf. Bennis, 1991; Hoekstra, 1991).

(76)

a. … dat Jan [_DegP [_AP lang] te] is (D-structure)

… that Jan tall too is

b. … dat Jan [_DegP [_Deg' [_AP t_i] te lang_i]] is (S-structure)

Rightward incorporation of the adjective into Deg° also directly explains the adjacency effect which holds between the degree word te and the adjective in Dutch (Hoekstra, 1991). This effect is exemplified in (77) (cf. also Section 3.2.). 57

(77)

a.   dat Jan [daarvan te afhankelijk] is

… that Jan thereon too dependent is

b. *… dat Jan [te daarvan afhankelijk] is

c. … dat Jan [te afhankelijk daarvan] is

The PP-complement daarvan ‘on it’ either precedes or follows the string te afhankelijk. As shown by the ill-formedness of (77b), it cannot intervene between the degree word and the adjective. Under a Deg-final analysis, the sequence in (77a) is simply derived by rightward incorporation of the

adjective into Deg°; the complement daarvan remains in its base position. 58 Schematically:

(78)

[_DegP [_Deg' [_AP [_PP daarvan] t_i] [te afhankelijk_i]]]

The string in (77b) will not be derived because of the restriction that a maximal projection like PP cannot be incorporated into (i.e. adjoined to) the head (i.e. structure preservingness). Notice further that in order to derive the sequence in (77c), DegP-internal PP-extraposition must have taken place besides rightward adjective incorporation.

Further support for the head final analysis of the Degree Phrase appears to come from complex adjectival constructions in which two indirect objects occur simultaneously, one being selected by the adjectival head, the other by the degree word (cf. Bennis 1991). This construction is exemplified in (79).

(79)

a. … dat zij [Jan gehoorzaam] is

… that she Jan (IO) obedient is

… that she is obedient to Jan

b. … dat zij [mij Jan te gehoorzaam] is

… that she me (IO) Jan (IO) too obedient is

… that for me she is too obedient to Jan

c. een [mij Jan te gehoorzame] jongen

a me (IO) Jan (IO) too obedient boy

In the simplex construction (79a), the adjective gehoorzaam takes the indirect object complement Jan. In (79b), the complex Degree Phrase contains two indirect object DPs: Jan is the indirect object of the adjective gehoorzaam, and the pronoun mij is selected by the degree word te. So, the string exhibits two crossing dependencies. A similar crossing pattern is visible in (79c), where the adjectival projection is in attributive position.

Under a Deg-final hypothesis, the ordering of the two indirect objects in (79b,c) can be straightforwardly derived by rightward incorporation of

the adjective into the right branch degree head (Deg°). The two indirect objects simply remain in their base positions. Schematically: 59

(80)

[_DegP [_Deg' mij [_Deg' [_[AP [_A' Jan [_A gehoorzaam]]] [_Deg te]]]]

On closer examination of the data, however, it turns out that something else underlies the observed crossing pattern, namely leftward scrambling of the two indirect objects. This becomes clear when we look at complex structures in which a measure phrase is present in the specifier position of the degree phrase. Consider the following examples:

(81)

a. … dat zij [mij Jan te gehoorzaam] is

… that she me Jan too obedient is

… that for me she is too obedient to Jan

b. … dat ze [veel te gehoorzaam] is

… that she much too obedient is

(82)

a. *… dat ze [veel mij Jan te gehoorzaam] is

b. *… dat ze [mij veel Jan te gehoorzaam] is

c. *… dat ze [Jan veel mij te gehoorzaam] is

d. … dat ze [mij Jan veel te gehoorzaam] is

e. een [mij Jan veel te gehoorzaam] meisje

a me Jan much too obedient girl

As is clear from (81b), where the indirect objects are absent, a measure phrase can occur in the specifier position of the degree word. Look now at the examples in (82) and notice what happens when the indirect objects are added to the structure. The ill-formed (82a) shows that the indirect objects cannot occur in between the measure phrase in [Spec, DegP] and the degree word te. This is totally unexpected under a head final analysis for the degree phrase, in which this surface order would simply be derived by rightward incorporation of the adjective into the degree word. (82b) and (82c) show that removal of one of the two indirect objects does not make the structure any better. In fact, both objects must surface to the

left of the measure phrase in [Spec, DegP], yielding a pattern of two crossing dependencies. This exemplified in (82d) for a predicative adjective phrase and in (82e) for an attributive one.

But if multiple scrambling is involved in deriving the adjectival structures in (82d,e), there is no compelling reason anymore for adopting the Deg-final hypothesis for such surface structures as (79b,c). That is, the serialization of the two indirect objects in these examples rather seems to be the result of scrambling of the indirect objects to the left periphery of the adjectival structure. In that case, we can just as easily adopt a head initial analysis for the Degree Phrase.

Furthermore, there are two additional problems for the head final analysis. First of all, it remains unclear why P(reposition)-stranding is not possible from PPs in a pre-degree word position (cf. (83a)), since the PP would simply occupy its base (theta-marked) position under a Deg-final structure. The only reordered element is the adjectival head, which has been incorporated into the right branch Deg° (cf. (84)).

(83)

a. Ik wist dat hij toendertijd [daarvan te afhankelijk] was

I knew that he then there-on too dependent was

b. *Ik wist [waar_i, hij toen [[t_i, van] te afhankelijk] was

I knew where he then on too dependent was

(84)

[_DegP [_Deg' [_AP [_PP daarvan] t_i [te afhankelijk_i]]]

Secondly, a Deg-final analysis incorrectly predicts that preposition stranding is not allowed with PP-complements that occur to the right of the adjective, as is the case in (85a). Under a Deg-final analysis, these PPs occur in this position as the result of DegP-internal extraposition, as represented in (85b).

(85)

a. Waar_i is Jan [te afhankelijk [t_i van]] geweest

Where has Jan too dependent on been

b. Jan is [_DegP [_DegP [_AP t_j t_i] te afhankelijk_i] daarvan_j]

Jan has - - too dependent there-on

geweest

been

Normally, extraposition of a PP bleeds preposition stranding. This is shown, for example, by the contrast between the well-formed (86a) and the ill-formed (86b). In (86a), P-stranding has applied to a preverbal PP-complement. The ill-formedness of (86b) shows that P-stranding is not permitted with extraposed PPs.

(86)

a. Waar_i, heeft Jan [t_i op] gerekend?

where has Jan on counted

What did Jan count on?

b. *Waar_i, heeft Jan t_j gerekend [t_i op]_j?

This freezing effect of extraposition can also be illustrated within the adjectival domain itself. Consider the following examples, in which the adjective phrase has been topicalized to [Spec, CP].

(87)

a. [Daar_i een stuk minder afhankelijk [t_i van] dan Sue] was

There a lot less dependent on than Sue had

Jan geweest

Jan been

b. ?*[Daar_i een stuk minder afhankelijk t_j dan Sue [t_i van]_j] was Jan geweest

In (87a), the preposition van has been stranded as a result of leftward scrambling of the pronominal element daar within the adjective phrase. P-stranding is permitted in this sentence, since the PP-complement occupies its L-marked base position. In (87b), on the other hand, the stranded preposition occupies a position to the right of the dan-phrase, which is located within the higher DegP-projection. This suggests that the PP-complement headed by van is no longer in its L-marked base position, which explains why subextraction of daar is blocked.

On the basis of the above considerations I reject the Deg-final analysis and adopt the Deg-initial structure for Dutch. The question which then arises is: How do we account for the word order phenomena in (79) and (82), in which the degree word is preceded by two nominal indirect objects, one of which is selected by the adjectival predicate. The answer to this question, already hinted at in this section, is the existence of leftward scrambling within the Dutch adjective phrase. In the next section, the presence of this movement process within the Dutch adjectival projection will be further examined. The issue about the adjacency effect (cf. (77)) will be taken up again in Section 7.2.

5.2. Scrambling and the Deg-initial Hypothesis

My answer to the word order phenomenon in (79) and (82) is that this complex serialization within the adjectival system is the result of multiple

indirect object scrambling within the adjective phrase. More specifically, in line with them Uniformity of Theta Assignment Hypothesis (cf. Baker 1988) I will assume that two nominal indirect objects find their origin in the same structural positions as their prepositional counterparts at D-structure. 60 I will further assume that a complement to an adjective is always-base-generated to the right of it (cf. also Hoekstra 1984). That is, the base order is ‘A + complement’.

Consider the examples in (88)-(90). (88a) shows that the presence of the indirect object is optional. As is illustrated by (88b,c), the adjective gehoorzaam takes its prepositional indirect object in a postadjectival, AP-internal position, but requires its nominal counterpart to occur in pre-adjectival position. 61 As exemplified in (89), the same pattern is found with those structures in which the indirect object is selected by the degree word. (89a) illustrates the optionality of the indirect object. (89b,c) show that the prepositional form headed by voor (‘for’) occupies a postadjectival position, whereas the nominal indirect object obligatorily precedes the degree word. 62. Sentence (90), in which the two nominal objects precede the degree word, is derived by applying twice leftward IO-scrambling within the adjective phrase. 63.

(88)

a. … dat zij erg gehoorzaam was

… that she very obedient was

b. … dat zij [erg gehoorzaam aan jou] was

… that she very obedient to you was

(IO-PP of A)

c. … dat zij [jou_i erg gehoorzaam t_i] was

… that she you very obedient was

(scrambled IO-DP)

(89)

a. … dat ze [te gehoorzaam] was

… that she too obedient was

b. … dat ze [te gehoorzaam voor mij] was

… that she too obedient for me was

(IO-PP of Deg)

c. … dat ze [mij_k te gehoorzaam t_k] was

… that she (for) me too obedient was

(scrambled IO-DP)

(90)

… dat ze [mij_k jou_i [te [gehoorzaam t_i] t_k]] was

… that she (for)me (to)you too obedient was

(2x IO-DP)

that she was too obedient to you for me

Besides scrambling of DPs, so-called R-pronouns (e.g daar, er (‘there’)), which are selected by prepositions, and PPs can undergo leftward scrambling within the adjectival domain. Some examples are given in (91).

(91)

a. … dat ik toendertijd [_SC niemand [daar_i te bang [t_i

… that I at-the-time no one there too afraid

voor]]] achtte

of consider

… that I considered no one too afraid of it at the time

b. [_CP [daarvan_i veel te afhankelijk t_i] [_C' leek Jan

thereon much too dependent seemed Jan

mij toendertijd]]!

to-me at-the time

Jan seemed to me much too dependent on it at the time

In (91a), the R-pronoun daar has been scrambled out of the PP-complement of the adjective. It has landed in a position to the right of the indefinite subject niemand, which is the subject of the. adjectival small clause. In (91b), the PP daarvan has been moved to a position to the left of the measure phrase which occupies [Spec, DegP]. The landing site of the PP.-complement must be internal to the adjective phrase, as the adjective phrase occupies [Spec, CP]. 64

These leftward, reordering operations yield a certain amount of word order freedom within the adjectival domain. 65 The phenomenon of variable word order has been examined extensively for the extended verbal projection and has raised various important theoretical issues, e.g. whether scrambled structures are derived by movement or not, and if they are, whether scrambling creates A-chains or A'-chains. 66 A full discussion of these theoretical issues in relation to the adjectival system is certainly beyond the scope of this paper. I will take the position here that scrambling within the extended adjectival projection involves adjunction to a maximal projection (i.e. movement of the A'-type). In (92a), for example an R-

pronoun has been adjoined to QP, and in (92b) it has been adjoined to DegP. 67

(92)

a. Jan, [_QP daar_i [_QP veel [_Q' minder [_AP gevoelig [_AP [t_i

Jan, that much less sensitive

voor]]]]]], betrad vol vertrouwen het podium

to, mounted full(of) confidence the platform

Jan, who was much less sensitive to it, mounted the platform full of confidence

b. Jan, [_DegP daar_i [_DegP net zo [_QP bang [t_i voor]] als ik verliet

Jan, that just as afraid of as I, left

meteen de zaal

immediately the room

Jan, who was just as afraid of it as I was, left the room imediately

In what follows, I will show that some of the argumentation that has been put forward in support of a movement approach towards scrambling within the extended verbal projection extends to the adjectival domain (cf. Webelhuth, 1987; Corver and Van Riemsdijk, 1994).

A first argument for interpreting the scrambled structure as a movement-derived structure is its sensitivity to island constraints. This is shown, for example, by the ill-formed example (93a), in which the R-pronoun er has been subextracted out of one of the adjectival conjuncts, in violation of the Coordinate Structure Constraint (Ross 1967). As expected, across-the-board movement of the R-pronoun yields a well-formed adjectival structure (cf. (93b)). 68

(93)

a. *[Er_i zowel [verliefd [op Sue]] als [afhankelijk [t_i van]]]_j

there both in-love with Sue and dependent on

was Jan t_j geweest

had Jan been

b. [Er_i zowel [verliefd [t_i op]] als [afhankelijk [t_i van]]]_j was

there both in-love with and dependent on had

Jan t_j geweest

Jan been

Jan was both in love with her and dependent on her

A second phenomenon suggesting that scrambled structures within the adjectival domain are derived by movement is parasitic gap licensing, a property characteristic of (A'-)movement-derived structures (cf. Bennis and Hoekstra (1984)). The relevant configuration is provided by such complex adjectival structures as in (94a), in which an infinitival clause appears that is selected by the degree item voldoende (‘sufficiently’). As shown by the ungrammaticality of (94b), scrambling of an R-pronoun to a position external to the infinitival clause is not allowed. However, a gap can appear within the PP of the embedded clause, if an R-pronoun has been scrambled out of the PP-complement of the adjectival head dol. The occurrence of this gap in the infinitival clause is clearly parasitic on the scrambling of the R-pronoun out of the PP headed by op. As is shown by (94d), the (parasitic) gap is impossible if the R-pronoun er remains within its PP.

(94)

a. Ik acht Jan [voldoende [dol [op Sue]] [om met haar

I consider Jan sufficiently fond of Sue for with her

te trouwen]]

to marry

b. *[Er_i voldoende dol [op Sue] [om [t_i mee] te trouwen]]

there sufficiently fond of Sue for with to marry

acht ik Jan

consider I Jan

c. [Er_i voldoende dol [t_i op] [om [e mee] te trouwen]]

there sufficiently fond of for with to marry

acht ik Jan

consider I Jan

I consider Jan sufficiently fond of her to marry her

d. *?[Voldoende dol [erop] [om [e mee] te trouwen]] acht ik Jan

A third phenomenon that can be directly explained under a movement analysis towards scrambling is illustrated in (95) and (96). What we find here is an asymmetry in the preposition stranding behavior of postadjectival PP-complements on the one hand and those PP-complements which uccur in a pre-degree word position on the other hand. The former PPs occur in a theta-marked position and hence permit preposition stranding. The latter, on the other hand, exhibit the well-known freezing effect of moved PPs. The PP-complement, which ends up adjoined to DegP in (95b) and adjoined to QP in (96b), is no longer within the L-marking domain of the adjective. Preposition stranding is therefore prohibited. 69

(95)

a. Ik wist [waar_i hij toendertijd [veel te afhankelijk [t_i

I knew where he then much too dependent

van]] was]

on was

I knew what he was too dependent on at the time

b. *Ik wist [waar_i hij toendertijd [[t_i van]_j veel te afhankelijk t_j]

was]

(96)

a. Ik wist [waar_i hij toendertijd [minder bang [t_i voor]] was]

I knew where he then less afraid of was

I knew what he was less afraid of at the time

b. *Ik wist [waar_i hij toendertijd [[t_i voor]_j minder bang t_j was]

Finally, the following coordination facts are also directly accounted for under a movement approach to scrambling within the adjectival system:

(97)

a. Ik vond toendertijd [_SC niemand [die kinderen_i [en

I considered then no one these children both

[(veel) te gehoorzaam t_i] en [(veel) te behulpzaam t_i]]]]

(much) too obedient and (much) too helpful

At that time I considered no one both much too obedient to the children and much too helpful to these children

b. [_DP een [_NP [mij_i haar_j [niet alleen [te gehoorzaam t_j t_i]

a me her not only too obedient

maar ook [te trouw t_j t_i]]] iemand]]

but also too faithful person

a person who, for me, is not only too obedient to her but also too faithful to her

In these sentences, a scrambled noun phrase (i.e. DP) appears external to a coordiated structure, whose conjuncts are introduced by so-called initial conjunction words (en … en; zowel … als). A property of these conjunction words is that they can only conjoin maximal categories (cf. Neijt (1979)). In (97), the two conjuncts are DegPs. In (97a), the DP die kinderen, which functions as an object of the adjectives gehoorzaam and behulpzaam, occupies a position external to the coordinated DegP-structure. This word order pattern is directly explained under a movement analysis: the DP has been moved in an ATB-fashion out of the two conjoined DegPs and has subsequently been adjoined to the coordinated DegP. The word order pattern in the attributive adjective phrase in (97b) can be explained along the same lines. In this example the DPs mij and haar, which function respectively as object selected by Deg° and as objeet selected by A°, occur externally to the coordinated DegP-structure. Also this complex word order pattern follows from an analysis in which the objects have been moved in an ATB-fashion out of the coordinated phrase.

On the basis of the above considerations, I conclude that the variable word order within the Dutch adjective phrase is the result of leftward scrambling within this syntactic domain.

6. Head-Final Agrp and A-to-Agr Raising

The head-initial character of the projections AP and DegP (and, by hypothesis, QP) may lead one to assume that the adjectival phrase structure is head-initial throughout. In this section, however, which deals wiih the identification of an adjectival AGR(eement)-node within the Dutch adjectival system, I will discuss phenomena which are suggestive for

the conclusion that Dutch has a head-final inflectional node (AGR). 70

Adjectival inflection is very poor in Dutch. Predicative adjectives never show overt agreement with the subject (cf. (98a)). As for the (prenominal) attributive adjectives, agreement is only spelled out overtly in the form of -e (cf. (98b)). As shown by (98c), -e does not appear on adjectives modifying indefinite neuter singulars. These also take the zero-morpheme.

(98)

a. Het boek is moeilijk-Ø

The book is difficult

b. Het moeilijk-e boek

The difficult-e book

c. een moeilijk-Ø boek

a difficult book

If a separate adjectival AGR-projection is adopted for the adjectival system, the question should be asked what evidence there is for this additional functional projection. At the conceptual level, the existence of such a level is defendible in view of the fact that inflectional features within the clausal domain, such as Tense and Agreement, are associated with distinct syntactic positions as well. Furthermore, if the external argument of an adjectival predicate is assumed to originate in the specifier of AP (the XP-internal subject hypothesis), then there must be a syntactic position to which the subject can be moved in (absolute) small clause structures like (99a). Notice that in this example the noun phrase Romário occurs to the left of the the Deg°-head zo and the nominal measure phrase 2 keer, which occupies [Spec, DegP]. 71 This position is plausibly (Spec, AgrP]. The structure we get then is the one in (99b). 72, 73

(99)

a. [Met [Romário 2 keer zo gevaarlijk als Bergkamp]] moest

with Romario two times as dangerous as Bergkamp had to

het Nederlandse voetbalelftal het onderspit delven tegen

the Dutch soccer team the subsoil dig against

Brazilië

Brazil

… the Dutch soccer team had to taste defeat at the hands of Brazil

b. [_PP Met [_AgrP Romário_i [_DegP 2 keer zo [_AP gevaarlijk] als Bergkamp]]]

At the empirical level, an argument in support of AgrP can be built on the distribution of PP-complements within the Dutch adjectival system. As indicated by the paradigm in (100), PP-complements exhibit a rather free distribution within the adjectival system.

(100)

a. … dat we waarschijnlijk [nauw verwant daaraan] waren

… that we probably closely related there-to were

(MOD A PP)

… that we were probably closely related to it

b. … dat we waarschijnlijk [nauw daaraan verwant] waren

(MOD PP A)

c. … dat we waarschijnlijk [daaraan nauw verwant] waren

(PP MOD A)

d. … dat we daaraan waarschijnlijk [- nauw verwant] waren

In (100a), the PP-complement is assumed to be in its post-adjectival base position. In (100b) it occurs to the immediate left of the adjectival head and is preceded by the modifier nauw. In (100c), it occurs internal to the adjective phrase but to the left of the modifier. In (100d), for the sake of

completeness, the PP-complement of the adjective shows up in a scrambled position within the VP.

A question which arises is: How do we account for the rather free distribution of the PP-complement within the adjective phrase? In Section 5, we have already noticed that leftward scrambling within the adjectival domain is an option available for PPs. So we might hypothesize that ‘APP’ is the base order and that those strings in which the PP-complement precedes the adjective (cf. (100b,c)) are derived by scrambling within the adjective phrase, yielding the derived structure [PP_i (ZP) A° t_i] (where ZP is a potentially intervening modifier).

Such a uniform scrambling approach faces a number of problems, however. First of all, it would be unable to explain the asymmetrie P-stranding behavior of the two leftward scrambled PPs in (101b) and (101c). That is, one would expect a freezing effect for both scrambled orders. However, this is not what happens, as is illustrated in (101).

(101)

a. Ik weet waar_i we toen [nauw verwant [t_i aan]] waren

I know where we then closely related - to were

I know what we were closely related to at the time

b. Ik weet waar_i we toen [nauw [t_i aan] verwant] waren

c. *Ik weet waar_i we toen [[t_i aan] nauw verwant] waren

d. *Ik weet waar_i we [t_i aan]_j toen [t_j nauw verwant] waren

(101a) shows that P-stranding is permitted from a post-adjectival PP-complement. In (101b), P-stranding is permitted as well. So, this pre-adjectival complement behaves similarly with respect to P-stranding as the post-adjectival PP. (101c), however, shows that P-stranding is blocked when the PP-complement occupies a position to the left of the modifier nauw. (101d), for the sake of completeness, illustrates the impossibility of P-stranding with those PP-complements that have been scrambled into the verbal domain.

What is important is that there is an extraction asymmetry between (101b) and (101c). This is quite unexpected, since under a uniform scrambling analysis, both PPs no longer occur in their L-marked base position. Thus, one (incorrectly) would expect the same island behavior.

Another argument against a uniform scrambling analysis for those sequences in which the PP-complement precedes the adjective comes from ATB-extractions from PP-complements. The often held descriptive generalization is that ATB-extraction from two phrases requires parallelism in

the structural position of these two phrases. Now if one adopts the analysis under discussion, one would expect that P-stranding is not allowed from a coordinated adjectival structure where the PP-complement occurs pre-adjectivally in one conjunct but post-adjectivally in the other. It turns out, however, that ATB-extractions from such configurations are possible, as is illustrated in (102) and (103).

(102)

Waar_i denk je dat Jan …

Where think you that Jan …

Who do you think that Jan …

a. [goed bevriend [t_i mee]] en [financieel afhankelijk

well friendly - with and financially dependent

[t_i van]] is?

- on is

was very friendly with and financially dependent upon?

b. [goed [t_i mee] bevriend] en [financieel [t_i van] afhankelijk] is?

c. [goed bevriend [t_i mee]] en [financieel [t_i van] afhankelijk] is?

d. [goed [t_i mee] bevriend] en [financieel afhankelijk [t_i van]] is?

(103)

a. Het meisje waar_i Jan [niet alleen [erg gesteld [t_i op]] maar

The girl which Jan not only very keen on but

ook [goed opgewassen [t_i tegen]]] leek werd later

also well equal to seemed became later

zijn vrouw

his wife

The girl who Jan was not only very keen on but also well matched with later became his wife

b. … waar_i … [niet alleen [erg [t_i op] gesteld] maar ook [goed [t_i tegen] opgewassen]]

…

c. … waar_i … [niet alleen [erg gesteld [t_i, op]] maar ook [goed [t_i tegen] opgewassen]]

…

d. … waar_i … [niet alleen [erg [t_i op] gesteld] maar ook [goed opgewassen [t_i tegen]]]

…

In (102/103a,b), the PP-complements in the two conjuncts are located on the same side of the adjectival heads. In (102/103c,d), however, the PP-complements are ordered differently with respect to the adjectival heads. In (102/103c), the PP in the left conjunct follows A°, whereas the one in the right conjunct precedes A°. In (102/103d), we find the reverse pattern. What is important is that in both (102/103c) and (102/103d), ATB-extraction is permitted in spite of the apparent absence of parallelism in the structural position of the two PP-complements.

In conclusion, a uniform leftward scrambling analysis for such sequences as ‘MOD PP A’ (cf. (100b)) and ‘PP MOD A’ (cf. (100c)) faces some problems. An alternative analysis which takes the order ‘PP + A’ as the base order and the order ‘A + PP’ as being derived by extraposition basically faces the same sort of problems as the uniform leftward PP-scrambling analysis. One of the questions is, for example, why a freezing effect does not occur after extraposition of the PP. As is illustrated by the contrast between (104a) and (104b), scrambling of R-pronouns is blocked when the PP-complement occurs in an extraposed (i.e. non-base) position.

(104)

a. [Daar_i net zo verliefd [t_i op] als Piet]_k zei Jan dat ie t_k

there just as in-love - with as Piet said Jan that he

was geweest.

had been

Jan said that he had been as much in love with that/her as Piet

b.?*[Daar_i net zo verliefd t_j als Piet [t_i op]_j]_k zei Jan dat

there just as in-love als Piet with said Jan that

ie t_k was geweest.

he had been

So, the question remains how to derive the word order ‘MOD PP A’ in such sequences as (100b). 74

What I would like to propose is that those sequences in which the PP-complement intervenes between the modifier and the adjectival head are

derived by rightward movement of the adjectival predicate into a higher right branch Agr-node. Schematically: 75

(105)

a. [_DPeen [_NP [_AgrP PRO [_Agr' [nauw t_i daaraan]

a closely there-to

[_Agr verwante_i]]] [_NP man]]]

related-infl man

a man closely related to it

b. [_AgrP [_Agr' [nauw t_i daaraan] [_Agr verwant-Ø_i]]] leek

closely there-to related seemed

Jan me niet

Jan to-me not

Jan didn't seem to be closely related to it

What is crucial in this analysis is that the PP-complement (daaraan) remains in its base position.

Before discussing the merits of this analysis, I should point out that within the attributive adjective phrase in (105a), overt raising of the adjective to Agr takes place obligatorily. Hence, a sequence like een [nauw verwante daaraan] man (a closely related-infl there-to man), in which the PP-complement intervenes between the adjective and the modified noun, is ill-formed. In the predicative adjective phrase (105b), on the other hand, the adjective (carrying the zero-morpheme Ø) can remain in its base position, yielding the word order nauw verwant-Ø daaraan (closely related Ø there-to).

One might hypothesize that the obligatoriness of A-to-Agr raising in attributive contexts is due to the presence of the overt adjectival inflection -e, which must be licensed in Agr. However, A-to-Agr raising is also required when the adjective carries the zero-morpheme (Ø)):

(106)

a. *een [nauw verwant-Ø daaraan] persoon

a closely related-Ø there-to person

b. [_DP een [_NP [_AgrP PRO [_Agr' [nauw t_i daaraan]

a closely there-to

[_Agr verwant-Ø_i]]] [_NP persoon]]

related-Ø man

a person closely related to it

If overt raising of A-to-Agr in attributive adjective phrases would obligatorily take place for reasons of morphological feature checking (independently of whether the constellation of phi-features is morphologically expressed (-e) or not (Ø)), the question immediately arises why such overt raising is not required for ‘predicative’ adjectives, which also carry a zero-morpheme Ø in need of checking. In other words, the obligatoriness of A-to-Agr raising in (105a) and (106b) does not seem to be related to morphological feature checking requirements. Instead, I tentatively propose that overt A-to-Agr movement takes place obligatorily in (105a/106b) in order to cirumvent a violation of Williams' (1981) Head Final Filter. This surface filter on prenominal modifiers prohibits such elements from terminating in anything other than their heads.

What are the merits of the analysis depicted in (105), in which the adjective raises to Agr? First of all, the P-stranding facts in (101) are accounted for. In the sequence ‘MOD A PP’ (cf. (101a)), P-stranding is permitted, since the PP-complement occupies its L-marked base position. Consider next the relevant example (101b), representing the surface order ‘MOD PP A’. Under a rightward A°-movement analysis, P-stranding is predicted to yield a well-formed sentence, since the PP-complement still occupies its L-marked base position. The only element which has been moved (to the right) is the adjectival head. Extraction out of the sequence ‘PP MOD A’ (cf. (101c)), finally, is ruled out; after scrambling of the PP-complement, the PP no longer occupies its L-marked base position and therefore blocks P-stranding.

Secondly, the ATB-extraction from apparently non-parallel PP-complements (cf. (102/103)) is predicted to be permitted. Under a rightward A°-movement analysis, the PP-complements in such sequences as ‘MOD A PP & MOD PP A’ and ‘MOD PP A & MOD A PP’ occupy parallel structural positions. The only difference between the two adjectival conjuncts is that in one conjunct rightward adjective movement to AGR° has applied. What is important is that the PP-complements in the two conjuncts occupy parallel, L-marked positions.

A third potential argument in favor of the rightward adjective movement approach might come from such coordination facts as in (107). 76

(107)

a. een [mij goed maar jou slecht gehoorzame] dienaar

a me well but you badly obedient servant

a servant who is very obedient to me but hardly obedient to you

b. een [noch goed met mij noch goed met jou bevriende]

a neither well with me nor well with you friendly

jongen

boy

a boy who is neither friendly with me nor with you

What we see here is coordination of what appear to be non-constituents. Take for example (107b), where we appear to have a coordination of two conjuncts consisting of an adverbial modifier and a PP-complement. The adjectival head is lacking in the two conjuncts. In view of the generalization that only constituents can be input to coordinations, these coordinations might be interpreted as providing evidence in support of a rightward adjectival head movement. That is, the adjectival head has undergone ATB rightward movement to the higher right branch Agr. 77 Schematically:

(108)

een [_AgrP [noch [_AP goed - met mij] noch [_AP goed - met jou]] bevriend-e] jongen

In brief, conceptual and empirical considerations lead to the conclusion that the Dutch adjectival system contains a right-headed functional projection AgrP.

7. Mixed Headedness

If there is a separate functional projection AgrP within the Dutch adjectival system, we finally end up with the following fully articulated structure: 78

(109) [_AgrP [_Agr' [_DegP [_Deg' Deg [_QP [_Q' Q [_AP DP [_x' A XP]]]]]] Agr]]

A question which arises is: What explains the mixed directionality of the functional system, i.e. Deg° and Q° taking their complements to the right and Agr° taking its complement to the left?

7.1. Nominal Orientation Versus Verbal Orientation

A possible approach would be to relate the mixed directionality of the functional heads to the categorial feature definition of adjectives. In Remarks on Nominalizations, Chomsky (1970) defines adjectives in terms of the atomic features +N, +V, which implies that adjectives have both nominal (substantive) and verbal (predicative) properties. Suppose now that the verbal and nominal properties of the adjective are reflected in the adjectival functional system. The intuitive idea would be now that determiner-like degree words (Deg) and quantifier-like degree words (more, less, etc.), are more nominal in nature and take the headedness of the Dutch nominal system, whereas the adjectival Agr-node is more verbal in nature and hence takes the headedness of the Dutch verbal system. As is well-known, the nominal system (i.e. DP) is consistently right-branching in Dutch. Consequently, degree words and quantifiers take their complement to the right within the adjectival system. The Dutch VP and its inflectional projection are standardly considered to be head final (cf. among others Bennis and Hoekstra, 1984; Koster, 1987). 79 So, What we get is mixed headedness within the functional system of the adjective phrase.

The question, of course, arises whether there is any justification for this dichotomy between nominally oriented and verbally oriented functional projections within the adjectival system. As a matter of fact, there are certain phenomena suggesting this distinction. The nominal orientation of Deg° and Q° has already been discussed in Section 3. It was observed that quantified noun phrases and quantified adjective phrases exhibit parallel behavior in certain respects (cf. split topicalization, partial pronominalization). The functional category Deg° turned out to display grammatical properties characteristic of the nominal determiner system: anaphoricity, d-linking, the uniqueness interpretation of the clitic definite article in superlatives, the absence of split topicalization.

The assumption that adjectival Agr is more verbal in nature also receives

empirical support. The verbal orientation is first of all suggested by the possibility of having adjectival verbs (participles) carrying adjectival inflection in attributive position (see (110a,b)).

(110)

a. de mij hatend-e vrouw (present participle)

the me hating-AGR woman

b. de (door mij) gehat-e vrouw (passive participle)

the (by me) hated-AGR woman

The verbal orientation of adjectival AGR is further suggested by a split within the class of adjectives selecting a PP-complement. As was observed in Section 6, an adjective like verwant (‘related’) allows its prepositional complement to occur both to its immediate right ((MOD) A PP) and to its immediate left ((MOD) PP A) (cf. (100a,b)). It was argued that this second order is derived by moving the adjectival head to a right branch Agr-node. It turns out now that not all adjectives selecting a PP allow this pattern of word order variation. This is illustrated by the following table which in each of the rows compares the word order possibilities of two adjectives selecting a PP headed by the same prepositional element.

(111)

This table shows that the adjectives in the middle column permit both word order patterns, whereas the ones in the rightmost column only permit the order in which the adjective precedes the PP-complement. 80 The word order (MOD) + A + PP is permitted in all cases, which suggests that A + PP is in fact the base order.

Careful study of the class of adjectives shows that only deverbal adjectives allow a PP-complement in immediate pre- and post-adjectival position Deverbal adjectives are of two types: First, those adjectives which exhibit participial morphology and as such are formally indistinguishable from verbal forms (cf. Den Besten 1981) (see table (112)). 81 Second, those adjectives that are derivationally related to a verb (cf. (113)).

(112) Deverbal adjectives exhibiting participial morphology

(113) adjectives derivationally related to V

The contrast noted in (111) between the b-examples in the middle column and the b-examples in the rightmost column suggests that only the more verb-like adjectives permit overt A-to-Agr movement. Those which, in view of the N-V dichotomy, could be characterized as ‘nominally oriented’ are not able to move overtly to the verbally oriented Agr-head.

Let us explore this distinction within the class of adjectives and try to come to a more refined characterization of what it means for an adjective to be verbally oriented or nominally oriented (cf. also Wetzer (1992)). I propose that the dichotomy within the class of adjectival predicates can be defined in terms of categorial feature dominance. In nominally oriented adjectives, the categorial feature +N is dominant ([+N, +v]); in verbally oriented adjectives the categorial feature +V is dominant ([+n, +V]). I will further assume that N-dominance of the adjectival predicate may overtly activate the N-oriented (i.e. N-dominant) part of the functional structure of the adjective phrase, whereas V-dominance of the adjective may overtly activate the verbally oriented (i.e. V-dominant) part. More specifically, if the adjective is N-dominant, it may raise overtly into the N-oriented functional head Q, whereas a V-dominant adjective may raise overtly into the V-oriented functional head Agr. Thus, an α-dominant adjective may only raise overtly into an α-dominant functional head F°. This can be interpreted as a sort of structure preservingness effect: substitution of an adjectival head into a functional slot requires identity of the lexical categorial feature make up of the adjective and the functional head substituted for.

Those grammatical properties of the adjectival head that require licensing by the non-dominant functional head in the extended adjectival projec-

tion are licensed in covert syntax. More specifically, if the thematic feature G (i.e. the referential degree argument) is licensed (e.g. by theta-binding) after an N-dominant adjective (i.e. [+N, +v]°) has raised overtly to the N-dominant functional head Q, then the φ-features of the N-dominant adjective are licensed at LF by raising them to the adjectival Agr-node, where they enter into a spec-head relation with the subject in Spec, AgrP. 82

(114)

On the other hand, if the φ-features are licensed after a V-dominant adjective (i.e. [+n, +V]°) has raised overtly to the V-dominant Agr-node, then the degree argument G gets theta-bound at LF. Schematically:

(115)

It turns out that many deverbal adjectives exhibit ambiguous behavior, in the sense that, besides permitting overt A-to-Agr movement (involving substitution of a V-dominant predicate into a V-dominant functional slot), they also allow overt A-to-Q movement (involving substitution of a N-dominant adjective into an N-dominant functional slot). In the next section, this will be exemplified by the behavior of the adjective afhankelijk (‘dependent’). I propose that within a numeration, such ambiguous adjectival predicates either take the N-dominant option or the V-dominant option. As a consequence, simultaneous overt activation of the N-dominant functional system and the V-dominant functional system is excluded. That is, successive application of overt A-to-Q raising and overt A-to-Agr raising is impossible. This again might follow from the structure preservingness requirement on substitution discussed above: if dominance of a categorial feature is a property of the categorial feature make up of adjectives and if this property is relevant to the categorial matching

requirement on substitution, then the successive cyclic head-movement of a [+N, +V]-dominant adjective (i.e. [+N, +V]) will never take place: [+N, +V] is not identical to Q (i.e. [+N, +v]) nor to Agr (i.e. [+n, +V]).

7.2. Some Word Order> (A)symmetries

With the above assumptions in mind, consider next the table in (116), which illustrates a number of interesting word order (a)symmetries between adjective phrases headed by the gradable adjective trots (‘proud’), on the one hand, and the gradable adjective afhankelijk (‘dependent’), on the other hand. The former is N-dominant and therefore allows overt activation of the N-dominant functional head (manifested by overt A-to-Q raising); the latter, as we will see, has the property of either being N-dominant or V-dominant. That is, if the +N-feature of afhankelijk is taken as dominant, the adjective can overtly raise to a N-dominant functional head like Q, but if the +V-feature is dominant, the adjective can only undergo overt head raising to the V-dominant functional head Agr.

(116)

Word order (a)symmetries between trots ([+N, +v]) and afhankelijk ([+N, +v] or [+n, +V])

Let us now consider each of the pairs x-X and see how we can account for their (a)symmetric behavior. 83 Consider first the symmetrie behavior of the pair (a-A). In each of the two examples, the analytic comparativc adjective has raised overtly to Q (cf. Section 4.1. for evidence). Raising to this N-dominant functional head is possible if the adjectival predicate is +N-dominant. 84

In (b-B), A-to-Q raising does not take place since the quantifier minder directly theta-binds the degree argument G of the gradable adjective heading AP.

The contrast in (c-C) relates to the possibility of having overt A-to-Agr raising. Under the V-dominance option, the adjective afhankelijk can overtly raise to the right branch Agr-node, resulting into a word order pattern in which the PP-complement linearly intervenes between the adjective and the functional head Q (cf. 117C). 85 The ill-formed pattern (c) is ruled out, because an N-dominant adjective cannot raise overtly to a V oriented functional head Agr (cf. (117c)).

(117)

c. *[_AgrP PRO [_Agr' [_QP veel [_Q' minder [_AP t_i daarop]]] trotse_i]]

C. [_AgrP PRO [_Agr' [_QP veel [_Q' minder [_AP t_i, daarvan]]] afhankelijke_i

The contrast between (d) and (D) is explained along the same lines as the previous pair (c-C). That is, (d) is out because the N-dominant adjective trots cannot overtly raise to the V-dominant right branch Agr-node. The pair (d-D) differs from the pair (c-C) in having an additional reordering operation within the adjective phrase, viz. scrambling of the R. pronoun daar from within the PP-complement to a position adjoined to QP. The derived structures are schematically represented in (118).

(118)

d. *[_AgrP PRO [_Agr'[_QP daar_j [_QP veel [_Q' minder [_AP t_i [t_j op]]]] trotse_i]]

D. [_AgrP PRO [_Agr' [_QP daar_j [_QP veel [_Q' minder [_AP t_i [t_j van]]]]] afhankelijke_i]]

The pair (e-E) does not exhibit any contrast in grammaticality. The quantifier minder heads QP and theta-binds the degree-argument G of the adjectival predicate. The PP-complement has been scrambled and adjoined to QP. 86 Notice that the sequence in (E) may be derived both under the +N-dominant option (cf. (119E)) and the +V-dominant option (cf. (119E')).

(119)

e. [_AgrP PRO [_Agr' [_QP daarop_j [_QP veel [_Q' minder [_AP trotse t_j]]]] Agr]]

E. [_AgrP PRO [_Agr' [_QP daarvan_j [_QP veel [_Q' minder [_AP afhankelijke t_j]]] Agr]]

E'. [_AgrP PRO [_Agr' [_QP daarvan_j [_QP veel [_Q' minder [_AP t_i t_j]]]] afhankelijke_i]]

The pair (f-F) illustrates an adjectival word order pattern involving the adjectival degree modifier erg(e). In Section 4.2, it was argued that the agreeing adjectival degree modifier occupies Spec, QP and exhibits overt agreement (-e) with the head A° of the extended adjectival projection, after A° has raised to Q°. Hence, agreement between the adjectival modifier and A° implies the application of overt A-to-Q raising. It was further proposed in Section 4.2. that in the non-agreement pattern, the degree modifier is adjoined to QP rather than in Spec, QP (i.e. sister to Q'). Even though A raises overtly to Q, no agreement relation is established since the two adjectival elements do not stand in the required spec-head relation to each other.

Let us first consider the ill-formed pattern (116f). Although the first head movement step in the derivation (A-to-Q) is legitimate since trots is + N-dominant, the subsequent raising of the adjective to Agr is illegitimate, since an N-dominant adjective cannot raise overtly to a V-dominant functional head (cf. (120f, f'))- Turning to the patterns in (116F), we notice that only the pattern with overt agreement is impossible. Recall

that agreement pattern is allowed if the two adjectival elements stand in a local spec-head relation in overt syntax, which implies that the gradable adjective has raised overtly to Q. Head movement of A to Q implies that the ‘N-oriented track’ has been taken and that the categorial feature +N of afhankelijk is dominant in that structure. But, if +N is dominant, the adjectival predicate can never raise overtly to the verbally oriented Agr-node, with the Result that such patterns as ‘MOD-e PP-complement A-e’ are never found. What about the non-agreement pattern in (116F)? In that adjectival structure, afhankelijk has a V-dominant interpretation. Overt A-to-Agr can apply, yielding the well-formed pattern erg daarvan afhankelijk. The degree argument G of the adjectival predicate can be licensed at LF via theta-modification by erg, after the property G has raised to Q.

(120)

f. *[_AgrP PRO [_Agr' [_QP erg [_QP [_Q' t'_i [_AP t_i daarop]]]] trotse_i]]

f'. *_[AgrP PRO [_Agr' [_QP erge [_Q' t'_i [_AP t_i daarop]]]] trotse_i]]

(121)

F. [_AgrP PRO [_Agr' [_QP erg [_QP [_Q' Q [_AP t_i daarvan]]]] afhankelijke_i]]

F'.*[_AgrP PRO [_Agr' [_QP erge [_Q' t'_i [_AP t_i daarvan]]]] afhankelijke_i]]

In (g-G), we find the same contrasts as in (f-F), the only difference being that the R-pronoun daar has been scrambled out of the PP-complement and adjoined to QP. The ill-formed examples are not out because of illegitimate scrambling; scrambling out of the PP-complement is permitted because the PP simply occupies its base-generated, L-marked position. The ill-formed patterns are out because of illegitimate overt movement of a +N-dominant adjective to the +V-dominant Agr-node.

In the well-formed patterns in (h-H), the PP-complement has been scrambled and adjoined to QP. The agreement patterns in (h-H), involve overt raising of the gradable adjective to Q, whose Spec is occupied by the adjectival modifier. The scrambled PP is adjoined to QP (cf. (122h'), (123H')). The non-agreement pattern in (116h) is derived in the same way as the agreement pattern, with the only difference that the adjectival modifier is adjoined to QP rather than in Spec, QP (cf. (122h)). Finally, the non-agreement pattern in (116H) can be derived in two ways. If afhankelijk is +N-dominant, it has the same derivation as the agreement pattern (again the only difference being that the modifier is now adjoined to QP); but if afhankelijk is + V-dominant, the adjective raises to Agr and not to Q (cf. 123H).

(122)

h. [_AgrP PRO [_Agr' [_QP daarop_j [_QP erg [_QP [_Q' trotse_i [_AP t_i [t_j]]]]]] Agr]]

h'. [_AgrP PRO [_Agr' [_QP daarop_j [_QP erge [_Q' trotse_i [_AP t_i [t_i]]]]] Agr]]

(123)

H. [_AgrP PRO [_Agr'[_QP daarvan_j [_QP erg [_QP [_Q' Q [_AP t_i [t_j]]]]] afhankelijke_i]]

H'. [_AgrP PRO [_Agr' [_QP daarvan_j [_QP erge [_Q' afhankelijke_i [_AP t_i [t_j]]]]] Agr]]

Consider, finally, the patterns (i, j, k) and (I, J, K). In these examples, the adjectival projection contains the functional DegP-layer (headed by te ‘too’). The lexical item veel (‘much’) in brackets is the dummy element inserted in Q, which copies the degree argument G of the adjectival predicate, this way creating the appropriate configuration for theta-binding between the Deg-operator and the degree argument (cf. Section 3.2.).

Let us first consider the structures (i, j, k) lacking the dummy veel. In these examples, the N-dominant adjective trots raises to Q, where the degree argument G of the adjectival predicate is close enough to Deg to be bound by this operator. After having raised to the nominally oriented Q-head, the N-dominant gradable adjective cannot move on to the verbally oriented Agr-node. In the ill-formed examples, this second head movement step has been illegitimately applied, creating the following structures:

(124)

i. *[_AgrP PRO [_Agr' [_DegP veel [_Deg' te [_QP [_Q'. t'_i [_AP t_i daarop]]]]] trotse_i]]

j. *[_AgrP PRO [_Agr' [_DegP daar_j [_DegP veel [_Deg' te [_QP [_Q' t'_i [_AP t_i [t_j op]]]]]]] trotse_i]]

As shown by (124j), scrambling has applied to the R-pronoun daar in (116j). However, it is not this scrambling operation which is illegitimate; the R-pronoun has been removed out of an L-marked PP-complement.

The sequence daarop veel te trotse in (116k) is well-formed. Besides A-to-Q raising of trots, the PP-complement has been scrambled and adjoined to DegP. What is important is that no overt raising of the N-dominant gradable adjective to the verbally oriented Agr has taken place. The relevant structure is the one in (125k).

(125)

k. [_AgrP PRO [_Agr' [_DegP daarop_j [_DegP veel [_Deg' te [_QP [_Q' trotse_i [_AP t_i [t_j]]]]]]] Agr]]

The sequences in which the dummy element veel is present in (i, j, k) are all ill-formed. The unavailability of these structures may be explained

in terms of economy. Since the N-dominant adjective may undergo A-to-Q raising, a head movement operation made available by UG, this rule will always block application of the language-specific process of veel-support (cf. Section 3.2. for dicussion).

Let us next consider the complex structures in (I, J, K). The patterns (I, J), which lack the dummy veel, are out for the same reason as the dummy-less examples in (i, j): afhankelijke raises overtly to the N-dominant Q (cf. 3.2.). This is only possible if the adjective has taken the N-dominant option (i.e. [+N, +v]). Being N-dominant, afhankelijke cannot move on to the V-dominant functional head Agr. As a result, the dummy-less patterns (I, J) will never be derived. The dummy-less pattern in (116K), however, is well-formed. The reason is that in this example, afhankelijke has taken the + N-dominant option; its right-peripheral occurrence in the adjectival string (in line with the Head Final Filter) is the result of leftward scrambling of the PP-complement to a position adjoined to DegP (analogous to (125k)).

As regards the patterns containing the dummy veel, recall that the dummy copies the degree argument G of the gradable adjectival predicate, this way enabling the Deg-operator te to locally theta-bind G. Clearly, afhankelijk in these dummy-contexts is not N-dominant, since if it were, we would expect A-to-Q raising to have operated rather than the more costly rule of veel-support. Hence, afhankelijke being V-dominant, raises directly to the V-dominant Agr-node, ending up in a position e-commanding the functional heads Deg and Q. This, however, has no consequences for the theta-binding requirement on the Deg operator te the inserted dummy element veel has copied the degree argument G from the raised adjectival predicate and creates the proper configuration for theta-binding to take place. Notice, finally, that in (116I), the PP-complement occurs in its base position, that in (116J) the R-pronoun has been scrambled out of the PP, and that in (116K) the entire PP-complement has undergone leftward scrambling.

Summarizing, I have studied various word order (a)symmetries between two types of adjective phrases, one headed by the N-dominant trots, the other by the adjective afhankelijk, which is optionally N-dominant or V-dominant. Under the assumption that an α-dominant adjective can only substitute for an α-dominant functional head, a kind of structure preservingness requirement, various intricate word order phenomena could be explained. Of course, in view of the limited sample of adjectives discussed here, a fuller investigation of the relation between categorial feature dominance and word order possibilities is definitely required.

8. Conclusion

The purpose of this article was to get insight into the phrase structural and word order properties of the extended adjectival projection, a phrase structural domain which has received relatively little attention in the generative literature. Focusing on the internal syntax of Dutch adjective phrases, I have come to the following conclusions. First of all, there is a strong empirical (and theoretical) basis for extending the functional head hypothesis to the adjectival system (i.e. for adopting the DegP-hypothesis). Secondly, a distinction should be made between two types of functional degree categories: Deg(P) and Q(P). This split is represented structurally, with Deg selecting QP and Q selecting AP (the split degree system hypothesis). Thirdly, besides DegP and QP a third functional projection is found in the extended adjectival projection, viz. AgrP. Fourthly, as regards directionality of headedness within the functional system, it was concluded that Deg and Q take their complements to the right, whereas Agr takes its complement to the left. This asymmetry of headedness within the adjectival system was assumed to be related to the nominal orientation of the Deg and Q heads and the verbal orientation of Agr. Finally, three movement operations have been identified within the Dutch adjectival system: A-to-Q raising, A-to-Agr raising and leftward scrambling. The latter two, especially, are at the basis of the word order variation which is found within the Dutch adjectival system.

References

Received: 25 June 1996

Revised: 20 January 1997

Grammar Models/Center for Language Studies (CLS)

Tilburg University

P.O. Box 90153

5000 LE Tilburg

The Netherlands

E-mail: N.F.M.corver@kub.nl