PP-Complements in Dutch Ad Neeleman Utrecht University/OTS

1. Introduction PPs can have different syntactic functions. One has to distinguish at least PPadjuncts, PP-arguments and PP-complements. Although demarcation is not always not trivial, clear examples of each of these functions can be given. Consider the following Dutch constructions: (1)

a. Dat Jan (op het laatste moment) geréd werd that John on the last moment saved was b. Dat Jan *(op de derde etáge) woont that John on the third floor lives c. Dat Jan *(op zijn gelúk) vertrouwt that John on his luck trusts

The PP in (1a) is uncontroversially analyzed as an adjunct. This classification is corroborated by the fact that it can be omitted without repercussions, and the fact that it does not attract sentence stress. Of course, the first argument is not conclusive: there are many optional arguments. The second is more persuasive. Abstracting away from focus, sentence stress in Dutch falls on the complement that immediately precedes the verb, or, in the absence of such a complement, on the verb itself (cf. Gussenhoven 1984). The indicated stress pattern therefore suggests that op het laatste moment is an adjunct, and not a complement. If this line of argumentation is correct, the PPs in (1b) and (1c) must both be analyzed as complements. As indicated in the examples, their presence is obligatory, a sign of argumenthood, and they attract sentence stress in the way arguments do. Still, the PPs in (1b) and (1c) present thematically different cases. The former can easily be analyzed as a true argument. That is, both the selectional relation between the PP and the verb and the semantics of the PP itself are straightforward. The PP denotes a location and the verb selects a locational object. For the purposes of this paper, a simple relation of 1-role assignment

will suffice to describe this state of affairs: (2)

V' V1

PP

PPs of the type in (1c), however, are much harder to analyze. It is unlikely that there is a thematic relation between the verb and the PP as a whole in constructions of this type. To say the least, the accurate description of what that relation would entail is not obvious. From a semantic point of view it would make more sense to say that, rather than the PP as a whole, the DP contained in it functions as an argument, to which the verb and the preposition jointly assign a 1-role. Since the preposition and the verb do not form a constituent, this insight cannot be expressed directly. Let me therefore assume - as a starting point of the discussion - that the DP is 1-marked by the verb, and that the preposition has an ill-understood mediating function: (3)

V' V1

PP P

DP

In contrast to PP-arguments, PP-complements pose a rather tricky problem for 1-theory. If the DP contained in such complements is thematically related to the verb, a type of 1-role assignment would have to be allowed for that is barred by standard assumptions. Let me briefly explain why this is the case. Williams (1980) observes that there is a c-command restriction on predication. That is, a predicate may not assign its 1-role to a DP that does not c-command it. So, the Dutch construction in (4a) is grammatical under either a subject-related or an object-related reading of the secondary predicate naakt `nude': both DPs c-command it. (4b) and (4c) only allow a subjectoriented reading, since the object-DP Marie does not c-command the predicate in these constructions. (4d) is ungrammatical under any reading, since there is no DP for the predicate to assign its 1-role to. For the purposes of this paper, the crucial example is (4c), in which it is the PP-node dominating Marie that blocks predication.1 (4)

a. Dat [Jani [Mariej [naakti/j ontmoette]]] that John Mary nude met

1 Even under a subject-oriented reading of the AP, this example is marginal. The reason is that scrambling of PPs is a marked option. Nevertheless, I believe the claim made here is valid. Predication into a PP is also ruled out if the PP follows the verb (cf. ia) or if the construction is nominalized (cf ib, and section 5). Examples like (ic) show that PPs also block predication in English. For these reasons I will ignore the marginality of PP scrambling.

(i)

a. Dat Jani naakti/*j kijkt [naar Mariej] that John nude looks at Mary b. *Het naakti kijken [naar Marie]i the nude looking at Mary c. Johni talked [to Maryj] nudei/*j

b. Dat [Jani [naakti/*j [Mariej ontmoette]]] that John nude Mary met c. Dat [Jani [[naar Mariej] [naakti/*j keek]]] that John at Mary nude looked d. *Dat [naakt [Jan [Marie ontmoette]]] that nude John Mary met There is reason to believe that Williams's c-command restriction (i.e. an argument must c-command the element that 1-marks it) does not hold only for predication, but in fact for all forms of 1-role discharge. In general, a head may not 1-mark a DP contained in its complement. So, although there might be exceptional case-marking (depending on one's analysis of ECMconstructions), there is no such thing as exceptional 1-marking. It is against this background that PP-complements pose a problem. As has been argued above, PP-complements differ from PP-arguments in that the verb 1-marks the DP contained in them. This, however, seems to violate the c-command restriction on 1-role assignment. There is no easy way out of this problem. For instance, it will not do to claim that the structure introduced by prepositions is not visible at the level of 1-role assignment, or to claim that prepositions are inserted after 1-role assignment has taken place. The observation barring such solutions is that in constructions like (4c) the PPstructure blocks 1-role assignment by the predicate naakt, but does not block 1-role assignment by the verb. Under the `invisible P hypothesis' one would now be forced to say that prepositional structure is invisible to some forms of 1-role assignment, but not to others, which is not much more than a restatement of the problem. In this paper, I will discuss the syntax of (Dutch) PP-complements. I propose a solution to the problem they pose in terms of LF incorporation of the preposition. For more discussion, I refer to Neeleman (1995). 2. Indirect 1-Role Assignment Marantz's (1984) theory of indirect 1-role assignment partly solves the problem outlined above. Although the intuition behind Marantz's analysis is an important one, it can be argued that the analysis itself is insufficient, particularly with respect to structural restrictions on the occurrence of PParguments. Marantz observes that not only verbs, but also prepositions are 1-role assigners, as is clear from simple PP-adverbials like during lunch : here the interpretation of the DP depends on the temporal nature of the preposition. In verbal syntax, a complement may only appear if it is assigned a 1-role by V. If prepositions, like verbs, are 1-role assigners, the null hypothesis would be that complements to prepositions must be assigned a 1-role by P. Marantz uses this relationship to explain the fact that the complement of P is interpreted as an argument of the verb. He claims that what is typical about PP complements is that the 1-role assigned by P is matched with a role in the 1-grid of the verb, as in (5). He further assumes that this kind of `indirect 1role assignment' is possible only if the verb's 1-role is lexically marked as having to be assigned through P.

(5)

V' V 1i

PP P 1i

DP

In my view, the main attraction of Marantz's analysis lies in the fact that it expresses the insight, already mentioned in section 2, that the verb and the preposition jointly 1-mark the DP-argument. However, the analysis is less appealing in two other respects. First, it enlarges the possibilities of 1-theory by introducing a process of 1-role matching that is essentially downward (contra the c-command restriction discussed above). One cannot explain away this undesirable property of the analysis by saying that a process of 1-role percolation from P to PP (or vice versa) mediates between P and V. 1-role matching involves the preposition's internal 1-role, and in general only the external 1-role of a head X may percolate to XP. Second, the theory fails to explain an important subject-object asymmetry with respect to PP-arguments. The point is that there are no PP-subjects in quite the same way that there are PP-complements. In the few cases where PP-subjects appear, the PP as a whole, rather than the DP contained in it, functions as an argument of the verb. In terms of the tripartition made at the outset, PP-subjects are always PP-arguments, as can easily be seen in an example like under the bed is a good hiding place. How can this be explained? If the 1-role of a PP-argument simply has to be matched with a 1-role of the verb, it is hard to see why the construction in (6), featuring a PP-subject, would not allow a matching of the 1-roles of P and V. The absence of PP-subjects in the relevant sense is unexpected. (6)

Vmax

* PP P 1i

V' DP

V 1i

Marantz's explanation for this asymmetry is based on the assumption that external 1-roles are not part of the verb's 1-grid. If it is further assumed that 1-roles must be lexically marked as `indirect' in order to be assigned through P, PP-subjects will be ruled out: a lexical marking of the external 1-role is impossible. The weakness of this line of argumentation lies in the claim that external 1-roles are not contained in the 1-grid of the verb. In my view there are strong arguments for including the external 1-role in the verb's 1-grid. For instance, it is argued by Levin & Rappaport (1988) that the English nominalizing affix -er binds the external argument of the verb it attaches to. This is hard to understand if the relevant 1-role is not included in the verb's lexical specification (here I adopt the lexicalist view of morphology). Similarly, the most natural way of describing verb classes seems to be in terms of external versus internal arguments: transitives have both an internal and an external 1-role, unergatives have only an external 1-role, and unaccusatives have only an internal 1-role. If external 1-roles are not part of the 1-grid of the verb, how can one distinguish transitives and unaccusatives?

In the following section I develop an analysis of PP-complements that adopts the basic insight Marantz's system, but that does not suffer from its deficiencies. 3. LF-Incorporation The resemblance between 1-role matching in the sense of Marantz and the system of 1-role identification developed by Higginbotham (1985) suggests that a unification of the two should be sought. Higginbotham argues that 1-roles can be saturated not only through assignment, but also through identification with another 1-role. The main empirical motivation for this process involves prenominal modifiers, although it has proven useful in other areas, too. Assuming that prenominal modifiers are predicates, the question arises how they satisfy their 1-role. There is no DP around that may function as the target for 1-role assignment. The answer to this question is based on the hypothesis that nouns have an external 1-role denoting their reference (the so-called R-role). This 1-role allows DP to function as a predicate, as in John is a man. According to Higginbotham, the problem of prenominal modification is solved if one allows the 1-role of a prenominal modifier to be discharged by being identified with the R-role of the nominal head: (7)

NP AP 1i green

N Ri door

As a result of 1-role identification, the semantics of the 1-role of the prenominal modifier are combined with the semantics of the R-role (in ways that are irrelevant here). Consequently, a green door denotes something that is green and that is a door. The similarities between 1-role identification and Marantz's process of 1role matching are straightforward. In both cases the 1-roles of two predicates are combined with the effect (i) that one of these 1-roles is saturated (in one case the 1-role of the verb, in the other the 1-role of the prenominal modifier), and (ii) that an element associated with one of the identified 1roles functions as the semantic argument of both. It would therefore be attractive to unify these processes. However, where 1-role identification is a strictly local process, 1-role matching is less local than expected, as pointed out in the preceding section. 1-theory in general, it Given that rather strict locality conditions hold of seems logical to try to reduce 1-role matching to 1-role identification. Suppose, then, that 1-role matching, like 1-role identification, can only take place in a strictly local domain. For the purposes of this paper, this domain can be defined in terms of mutual c-command, in accordance with, amongst other things, the c-command requirement on 1-role assignment. This definition might have to be adapted, depending on the analysis of, amongst other things, stacked prenominal modifiers, but I cannot go into this issue here. The unification of 1-role assignment and 1-role matching has the effect that the structure in (5), no longer allows for 1-role matching. I therefore propose that in order to facilitate this process, which is crucial in the analysis of PP-complements, the preposition incorporates into the verb, thus creating

the required local relation between P and V. The resulting structure is given in (8). (8)

V' V

PP

P 1i

V 1i

tP 1

DP

If it is assumed that 1-theory holds at LF (cf. Chomsky 1993), it follows that P-incorporation will be procratinated until the derivation has passed the spellout point. The relation of 1-role matching need not be established before LF, and hence nothing triggers overt incorporation. 4. Consequences Let me now turn to the empirical consequences of the P-incorporation analysis. I will consider four phenomena: (i) the absence of PP-subjects, (ii) the absence of predication into a PP-complement, (iii) the impossibility to coordinate a DP and a PP-complement, and (iv) the impossibility of having two PP-complements. A first argument for the P-incorporation analysis is that it explains why there are no PP-subjects (at least not in such a way that the DP-complement of P is interpreted as an argument of V, cf. section 2). In order for there to be such a subject, the preposition that heads it would have to incorporate into V, thus creating the proper environment for 1-role matching. However, as has been argued extensively by Baker (1988), incorporation out of a subject violates the ECP. The representation in (9) is therefore ruled ungrammatical. (9)

Vmax

* PP

V'

tP

DP

V P 1i

V 1i

Constructions in which a PP as a whole functions as the subject (e.g. under the bed is a good hiding place) are still allowed, since in such cases no 1-role matching, and therefore no incorporation, has to take place. A similar line of reasoning captures the observation that depictives cannot be related to the DP contained in a PP-complement. Consider the tree in (10), which corresponds (in part) to the Dutch example in (4c). Vmax

(10) DP

V' PP tP

V' DP

AP 1

V P 1i

V 1i

There is only one way in which the AP in this structure can discharge its 1role without violating the c-command restriction on predication, namely by assigning it to the subject Jan. Due to this restriction, it cannot assign its 1role to the DP contained in the PP-complement. It also cannot discharge it by matching it with that of the incorporated P. That would require a downward relation, again in violation of the c-command requirement. In fact, the only way a DP contained in a PP-complement could be thematically related to a secondary predicate would be if the preposition incorporated either into the predicate itself or into its head. Both options are ruled out on general grounds. The first would be a case of incorporation into a maximal projection, the second one of incorporation into a non-c-commanding head. In addition to this, incorporation into (the head of) a secondary predicate would make it impossible for the verb to discharge its internal 1-role. More generally, then, the proposed analysis explains why the DP in a PP complement can only be interpreted as an argument of V. A third argument for P-incorporation can be based on the behavior of verbs that select either a PP-complement or a DP. Examples are the Dutch verbs geloven (in) `to believe (in)' and vertrouwen (op) `to trust (on)'. As can be observed below, coordination of the two types of complements these verbs allow is impossible. (11)

a. Marie vertrouwt [PP op Jan's komst] Mary trusts on John's coming b. Marie vertrouwt [DP Jan's beloften] Mary trusts John's promises c. *Marie vertrouwt [[DP Jan's beloften] en [PP op zijn komst]] Mary trusts John's promises and on his coming

(12)

a. Marie gelooft [PP in Jan's komst] Mary believes in John's coming b. Marie gelooft [DP Jan's verhalen] Mary believes John's stories c. *Marie gelooft [[DP Jan's verhalen] en [PP in zijn komst]] Mary believes John's stories and in his coming

Note that the ungrammaticality of (11c) and (12c) cannot be attributed to a constraint of categorial identity. Since Sag et al. (1985), it is well-known that constituents with different categorial features may be coordinated, as in Pat is a republican and proud of it . This is also true in Dutch. (13) is an example in which a DP and a PP-argument (in the sense of section 1) are coordinated. (13)

Ik adviseer [of [DP die route] of [PP over Zwolle]] I advise either that route or through Zwolle

Abstract P-incorporation offers an simple explanation for the data in (11) and (12). The coordinate structure constraint demands that movement out of a coordination takes place across-the-board. In (11c) and (12c), however, Pincorporation only affects the PP-complement. The DP-conjunct neither allows nor requires incorporation. Therefore, the coordinate structure

constraint is violated:2 (14)

*

V' V

P

&P V

DP tP

&

PP

DP

The final argument I would like to give for abstract P-incorporation is based on a property of PP-complements that to the best of my knowledge has not yet been dealt with in a generative framework. As is observed in the reference grammar of Dutch, Geerts et al. (1984), verbs in Dutch can only take a single PP-complement. It is not the case, however, that PP-complements in Dutch generally exclude the presence of other arguments. There are many examples of verbs selecting a PP- and a DP-complement. Variants of the verb vragen `to ask' provide a striking example of the ungrammaticality of the presence of two PP-complements in the domain of a single verb. As the data in (15) show, vragen takes two internal arguments. Either one of these arguments can be a PP (under subtle shifts in meaning), but if both are PPs an ungrammatical construction obtains. (15)

a. iemand iets vragen someone something ask b. aan/van iemand iets vragen to/of someone something ask c. iemand naar iets vragen someone for something ask d. *aan/van iemand naar iets vragen to/of someone for something ask

In fact, (15) is a specific instance of a more general restriction: in most double object constructions the GOAL/EXPERIENCER-argument may show up in a PP headed by aan `to', but in case the THEME /PATIENT-argument is contained in a PP, this is impossible, as can be observed below.3

2 Often, there is a selectional relation between the verb and the preposition that heads a PP-complement. It is to count on and not to count to, etc. The ungrammaticality of (11c) and (12c) could therefore be attributed to a lack of parallelism: the verb has a idiomatic selectional relation with one conjunct, but not with the other. However, the selectional relation between P and V closely resembles that between A and N in structures of collocational prenominal modification. A Dutch example is een strenge winter `a strict winter' (a cold winter). Now, collocational prenominal modifiers can be coordinated with regular adjectives, showing that coordination is not restricted by the parallelism constraint, and hence that the reason PPcomplements resits coordination must be a different one.

(i)

Een [strenge en [relatief natte]] winter a strict and relatively wet winter

3 There are some apparent counterexamples to the claim made here. The construction in (ia), for instance, is grammatical. It can be argued, however, that at least the over phrase in (ib) is a PP-adjunct, not a PP-complement. It can be coordinated with an adverbial, in contrast to PP-complements (cf 1b,c). Other counterexamples can be explained in the same way: one of the PPs is either a PP-adjunct or a PP-argument in the sense of section 1.

(16)

a. (*aan) iemand van iets overtuigen to someone of something convince b. (*aan) iemand op iets antwoorden to someone to something answer c. (*aan) iemand van iets afbrengen to someone of something discourage

How can the observation that verbs take only a single PP-complement be explained? As far as I know, no account can be given under standard analyses of PP-complements. In Marantz's analysis, for instance, there is nothing that would make it impossible for a verb to select two PP-complements: what would rule out a verb with two 1-roles that are marked `indirect'? Under the P-incorporation analysis, however, a natural explanation is available. The idea would be that certain types of incorporation can take place only once, with the effect that, if two or more PP-complements are present, only one of them can be licensed. Recall that in constructions of PP-complementation, the verb discharges its internal 1-role by matching it with the 1-role of the incorporated preposition. Now, it is well-known that there are strict conditions on thematic relations in complex heads. In particular, a sub-zero argument must be adjacent to the head it is an argument of. This restriction was formulated first in Selkirk's (1982). According to her first order projection condition, the internal arguments of a head must be realized, in both morphology and syntax, within the first node dominating that head. I will not discuss the syntactic consequences of Selkirk's proposal (I believe that they are incorrect), but in morphology its effects are the following. In two-member compounds, the non-head may either be an adjunct, as in night flight, or an argument, as in truck driver. In more complex compounds, there may be both and adjunct and an argument, but only if the argument and the head are sisters. Finally, no compound may contain two arguments: (17)

a. b. c. d.

[X adjunct X] [X adjunct [X argument X]] *[X argument [X adjunct X]] *[X argument [X argument X]]

The crucial observation here is that compounds like (17d) are ungrammatical. That is, morphology differs from syntax in that it does not allow double object constructions. Compounds like *boy toy giver are generally ruled out. Given this restriction, it is to be expected that verbs may select only one PPcomplement. Since the verb satisfies one of its 1-roles after P-incorporation, incorporation of two prepositions would lead to a sub-zero double object construction: P2 in (18) does not have a local relation with V in the sense of the first order projection condition.

(i)

a. Dat Jan met Marie over de liefde sprak that John with Mary about the love spoke b. Dat Jan met Marie [onwillig en [over bijna niets]] sprak that John with Mary reluctantly and about almost nothing spoke c. *Dat Jan Marie [onwillig en [naar bijna niets]] vroeg that John Mary reluctantly and for almost nothing asked

(18)

*

V' V'

PP

V P2

PP V

P1

t2

t1 V

Summarizing the findings of this section, several properties of PPcomplements can be explained in terms of P-incorporation: (i) there are no PP-subjects, (ii) depictives cannot be related to the DP contained in a PPcomplement, (iii) PP-complements may not be coordinated with DP- or CPcomplements, and (iv) verbs may select only one PP-complement. A final argument for the proposed analysis will be presented in section 5. 5. PP-Complements to Nouns PP-complements to nouns differ in three ways from PP-complements to verbs: (i) in contrast to what is the case in the verbal domain, nouns can take more than one prepositional complement, (ii) PP-complements to nouns can function as subjects for secondary predicates, and (iii) in a nominal environment PP rather than DP is the unmarked option, that is, DPcomplements to nouns are generally ruled out. These three observations are illustrated with Dutch nominalizations below. For discussion of the internal organization of these nominalizations, I refer to Hoekstra (1986). (19)

a. Het geven van bloemen aan Marie the giving of flowers to Mary b. Het rauwi eten van vleesi the raw eating of meat c. Het kopen *(van) een huis the buying of a house

I believe that these facts support the proposed analysis of PP-complements to verbs. My basic claim in this section will be that the invisible P hypothesis mentioned in section 1 is correct for PP-complements to nouns, at least in the constructions at hand. If this is true, the contrasts between PP-complements to nouns and PP-complements to verbs show that the invisible P hypothesis must indeed be rejected for the verbal domain. The upshot of the preceding sections is that the prepositions that head PPcomplements of verbs are full lexical heads. A difference between nouns and verbs, however, is that only the latter assign case. This has the consequence that in the case of nouns a different type of preposition is available. Following a long generative tradition, I would like to claim that the preposition van `of' is a dummy preposition, inserted to provide the complement of a noun with case. I would further like to claim that van-insertion is comparable to doinsertion in English, in that it is a last resort operation. The analysis of nominal van as a dummy is supported by the fact that, in contrast to the prepositions that head PPs in verbal syntax, it never makes a contribution to the semantics of a DP. Consider, for instance, the aspectual properties of VPs and nominalizations are compared. It can be observed in

(20a,b) that the presence of a PP-complement yields a durative VP: PPs cannot `measure out an event'. PP- complements to nouns, however, do not force a durative interpretation. The nominalization in (20c) patterns with the VP in (20a), rather than with the one in (20b). It seems, then, that the preposition van does not make a contribution to the semantics of NP, in accordance with the claim that it is a dummy. (20)

a. Dat Jan ?urenlang/in een uur de krant leest that John for hours/in an hour the newspaper reads b. Dat Jan urenlang/?in een uur in de krant leest that John for hours/in an hour in the newspaper reads c. Het ?urenlang/in een uur lezen van de krant the for hours/in an hour reading of the newspaper

If van is indeed inserted to provide a complement with case, it follows that van-insertion will not be available in the verbal domain: verbs are able to assign case themselves, so that insertion of a dummy preposition would be the application of a last resort operation where regular licensing is available. Let me now turn to the claim that the invisible P hypothesis is correct for van-PPs in nominalizations. Given that van is a dummy, it follows from Williams's (1994) notion of relativized head, or Grimshaw's (1991) notion of extended projection, that the referential properties of DP are inherited by PPs headed by van. The preposition does not have semantic properties of its own that would block percolation of the relevant properties to PP. This can be formalized by saying that van does not have an index, so that the index of DP may move up, as in (21). (21)

PPi P van

DPi

As a result, it is now predicted that the structure introduced by van will be invisible. That is, the 1-role of the noun can be assigned to a van-PP with the effect that it is interpreted as being assigned to the DP contained in it: these two elements share their index (and van is semantically vacuous). For the same reason, van-PPs may function as the subject of a secondary predicate: in accordance with the c-command restriction on predication, the 1-role of the adjectival predicate in (19b) can simply be assigned to the van-PP as a whole, with the required interpretational effect. It turns out, then, that incorporation of P into V, which was required to license PP-complements in the verbal domain, does not (or not always) have a counterpart in the nominal domain. This in turn implies that the effects of Pincorporation will not show up with van-PPs. In particular, I argued that, due to limitations on P-incorporation, the presence of one PP-complement blocks the presence of another in the verbal domain. If incorporation is not required for van-PPs, it is predicted that they can freely be added to nouns that already select another PP-complement, a prediction borne out by (19a). The proposal that regular PP-complements and van-PPs have a different status is corroborated by a further inspection of the selectional properties of nouns. Although van insertion is excluded in the verbal domain, nothing so far excludes the occurrence of regular PP-complements in the nominal

domain. In other words, nothing prevents a noun from selecting a prepositional incorporee, and discharging its 1-role after head movement: (22)

N' N P 1i

PP N 1i

tP 1

DP

There is reason to believe that one should indeed allow for P-incorporation in NPs. It seems unlikely that there is more than one dummy preposition, inserted by a last resort operation. However, in addition to van several other prepositions may head PP-complements to nouns. If these prepositions are not dummies, the analysis presented in this paper predicts that P-incorporation will have to take place in order for the DP selected by P to be interpreted as an argument of the noun. The prediction that then follows is that PP-complements to nouns headed by prepositions other than van will behave much like PP-complements to verbs. This seems to be the case. (23a) shows that such PPs block predication, and (23b,c) show that they can be combined with a van-PP, which does not require incorporation of its head, but not with another regular PP. (23)

a. *Het naakti kijken naar Mariei the nude looking at Mary b. Het vragen van iets aan iemand the asking of something to someone c. *Het vragen naar iets aan iemand the asking for something to someone

As it turns out, then, PP-complements of nouns support the P-incorporation analysis in two ways. Some PP-complements must be analyzed on a par with the ones found in the verbal domain: they display the same behavior. Others, namely those headed by van `of', behave rather differently. The preposition that heads them seems to be a dummy, inserted for reasons of case. Consequently, the invisible P hypothesis briefly discussed in section 2 is correct for van-PPs, but crucially not for PP-complements of verbs. References Baker, Mark. 1988. Incorporation; A Theory of Grammatical Function Changing. University of Chicago Press. Chomsky, Noam. 1993. `A Minimalist Program for Linguistic Theory', in: K. Hale & S. J. Keyser (eds.) The View from Building 20 . Cambridge: MIT Press. Geerts, Guido, et al. 1984. Algemene Nederlandse Spraakkunst . GroningenLeuven: Wolters-Noordhoff. Gussenhoven, Carlos. 1984. On the Grammar and Semantics of Sentence Accents in Dutch. Dordrecht: Foris. Grimshaw, Jane. 1991. `Extended Projections'. Ms., Brandeis University. Higginbotham, James. 1985. `On Semantics'. LI 16, 547-539. Hoekstra, Teun. 1986. `Deverbalization and Inheritance'. Linguistics 24, 549584. Lapointe, Stephen. 1985. `Review of Alec P. Marantz, "On the Nature of

Grammatical Relations"'. NLLT 3, 341-370. Levin, Beth & Malka Rappaport. 1988l `Non Event - er Nominals: A Probe into Argument Structure'. Linguistics 26, 1067-1083. Marantz, Alec. 1984. On the Nature of Grammatical Relations. Cambridge: MIT Press. Neeleman, Ad. 1995. PP-Complements and LF 1-Role Discharge. Ms., Utrecht University/OTS. Sag, Ivan, et al. 1985. `Coordination and How to Distinguish Categories'. NLLT 3, 117-171. Selkirk, Elizabeth. 1982. The Syntax of Words. Cambridge: MIT Press. Williams, Edwin. 1980. `Predication'. LI 11, 203-308. Williams, Edwin. 1994. Thematic Structure in Syntax. Cambridge: MIT Press.