Linguisties 35 (1997),1057-1089 0024-3949/97/0035-1057 Walter de Gruyter

MARCEL DEN DIKKEN and ERIC HOEKSTRA

Abstract
Frisian exhibits the curious phenomenon of parasitic participles.- one occurrence of the perfect auxiliary is able to license more than one perfect participle. Recursive checking (a one-many relation) is possible just in case the feature that recurs is not the one that is semantically interpreted. Thus the feature on the perfect auxiliary is the one that is semantically interpreted. Central to our analysis is a relativization of the head-movement constraint: movement of X across Y is illegitimate if Y possesses features relevant to X.

1. Introduction
Combinations of a main verb, a modal verb, an auxiliary of the perfect, and an element expressing irrealis come in two surface forms in Dutch, given in (1). These two sentences are not semantically equivalent - they differ with respect to the scope relationship between the modal willen and the auxiliary of the perfect: in (la) the modal takes scope over the auxiliary (which we shall annotate as M > H) while in (lb) the modal falls within the scope of the auxiliary of the perfect (H > M). [note 1]
(1) a.willen > hebben (M > H)
hij zou het willen hebben gedaan [note 2]
he would it want-INF have-INF do-PTC
he would like to have done it'
b. hebben > willen (H > M)
hij zou het hebben willen doen
he would it have-INF want-IPP do-INF
he would have liked to do it'

[1058 M. den Dikken and E. Hoekstra]
In the Dutch H > M construction (lb) not a single past participle shows up, in spite of the presence of the auxiliary of the perfect hebben. This underrepresentation of participial forms in multiverb clusters in Dutch is weil known from the Germanic literatura; it is commonly referred to as the Infinitivus-pro-Participio (IPP) effect. Frisian lacks the IPP effect. Interestingly, though, Frisian has something like the inverse of Dutch IPP, to which we may refer as the Participium-pro-Infinitivo (PPI) effect. As the (b) examples in (2) and (3) show, the Frisian counterparts of the Dutch sentences in (1) may feature two participles in the presence of only a single instance of the auxiliary of the perfect ha.
(2) (M > H)
a. hy soe it dien ha wolle
he would it do-PTC have-INF want-INF
b. hy soe it dien ha wollen
he would it do-PTC have-INF want-PTC
'he would like to have done it'
(3) (H > M)
a. hy soe it dwaan wollen ha
he would it do-INF want-PTC have-INF
b. hy soe it dien wollen ha
he would it do-PTC want-PTC have-INF 'he would have liked to do it'
We shall refer to the italicized participles in (2b) and (3b), which are not directly selected by a token of the auxiliary of the perfect, as parasitic participles - they parasitize on the presence of the "real" (i.e. ha- complemented) participle. PPI constructions of the type in (2b) and (3b) are ungrammatical in Standard Dutch (though they are found in north- eastern dialects; Bloemhoff 1979). [note 3]
West Germanic participial constructions in general, and Frisian "parasitic" participle constructions like (2b) and (3b) in particular, pose several questions:
(4)
a. How and where are participles licensed?
b. How do "parasitic" participles come about? How are their features licensed?
c. Why do Dutch and Frisian differ regarding the possibility of "parasitic" participles?
To answer these, we develop an account of the licensing of participles in a minimalist and antisymmetric theory of syntax (Chomsky 1993, 1995a, 1995b; Kayne 1994). In sections 3 and 4 we argue that VPs in the complement of an auxiliary selecting a participial phrase must move into

[Parasitic participles 1059]
the specifier position of a functional head in whose checking domain the participial features are licensed (also cf. Zwart 1994, 1995).
Our account of Frisian participle constructions is modelled on Chomsky's (1993) analysis of agreement, according to which a functional projection mediates agreement between an NP argument and the verb. The nature of the functional projection (and of the checking process obtaining in its domain) is determined by the nature of the elements that move there. The relationship between an auxiliary and the participle that it combines with may be looked upon as a kind of agreement or case-checking relationship similar to the one obtaining between a subject or object and an inflected verb. lf this proves a viable perspective, our analysis contributes toward extending the minimalist approach beyond the domain of subject/object agreement and case-feature checking, for which it was originally developed.
Two other theoretical consequences of the analysis are also worth highlighting at the outset. One is the conclusion that multiple feature checking must be allowed, a possibility that can be shown to fit naturally into Chomsky's (1995a, 1995b) outlook on features and their checking. Our analysis of the Frisian facts also gives occasion for a reevaluation of the locality condition on head movement (Travis's 1984 head-movement constraint), such that an accurately delimited class of "head-skipping" head-movement operations is ruled in. In section 5 we shall address the consequences of our relativized head-movement constraint (RHMC). We shall argue that the RHMC is an accurate restrictor of head movement, and that its adoption is necessary in the minimalist theory. The paper closes with a summary of our main findings and some concluding remarks in section 6.

2. On the semantics of Frisian ("parasitic") participle constructions

The two types of PPI construction, (2b) and (3b), share three necessary ingredients: (i) a token of have, (ii) a modal (kinne 'can', moatte 'must', wolle ‘want’, sille ‘shall/will', meie 'may', hoeve ‘need’ doare 'dare'; cf. Hoekstra [1990: 62, 90] for the same observation made for the modal- contraction construction mentioned in note 3), and (iii) a counterfactual interpretation. [note 4]
But beyond this correspondence between (2b) and (3b) there is an important semantic difference between the two types of “parasitic” constructions. We hinted at this in the introduction, where we pointed out that in (2) the modal takes scope over the auxiliary of the perfect, while in (3) it is the auxiliary of the perfect that takes the modal

[1060 M. den Dikken and E. Hoekstra]
in its scope. This difference between the two constructions is particularly evident in the following context. The Dutch verb cluster kunnen uitstaan ‘can stand’ is an idiomatic verb-raising collocation that may not he broken up by an intervening occurrence of have, as is shown in the examples in (5) and (6):
(5) (H > M)
niemand zou dat hebben kunnen uitstaan
nobody would that have can-IPP stand-INF
‘nobody would have been able to stand that'
(6) (M > H)
*niemand zou dat kunnen hebben uitgestaan
nobody would that can-INF have stand-PTC
The Frisian sentences in (7) and (8) show that the two types of PPI construction do not behave in the same way with respect to this idiom test:
(7) (H > M PPI)
nimmen soe dat útstienlferneard kinnen ha
nobody would that stand-PTC can-PTC have
(8) (M > H PPI)
*nimmen soe dat útstien/ferneard ha kinnen
nobody would that stand-PTC have can-PTC
The PPI construction in (8) patterns with the Dutch example in (6), in which the auxiliary of the perfect interrupts the contiguity of the modal and the main verb. The modal in (8) takes the auxiliary's projection as its complement and cannot idiomatically combine with the main verb.
lt is worth pointing out here that with respect to the idiom test (and other diagnostics that space does not allow us to go into) the PPI constructions in (7) and (8) are completely on a par with their non-PPI counterparts, given in (9) and (10):
(9) (H > M)
nimmen soe dat útstean/ferneare kinnen ha
nobody would that stand-INF can-PTC have
(10) (M > H)
*nimmen soe dat útstien/ferneard ha kinne
nobody would that stand-PTC have can-INF
Apparently, the presence or absence of a "parasitic" participle is inconse- quential; what matters is that ha must not occur between the modal and the main verb. The facts in (7)-(10) thus show that "parasitic" participle constructions have precisely the same distribution as their nonparasitic

[Parasitic participles 1061]
counterparts with the same relative order of ha and the nonfinite modal. What counts is word order, not form.

3. Participles: their placement and checking

Now that we have established the primacy of word order and the truly parasitic nature of the extra participle in double-participle constructions, we can start developing our analysis of double-participle constructions in Frisian. Before returning to examples of the type in (2b) and (3b), we should first of all address the question of where and how participles in single-participle constructions are licensed. We shall argue on the basis of an investigation of Frisian hoeve te 'need to' constructions that selected participial VPs are XP-moved to specifier position. [note 5]
Participles in Frisian embedded clauses always find themselves to the left of the auxiliarles selecting them, as shown in (11):
(11)
a. omdat hy it dien hie
because he it do-PTC had
b. * omdat hy it hie dien
Following Kayne (1994) in assuming a strictly head-initial underlying structure for all languages of the world, we can envisage at least two ways of deriving this word order - we couid incorporate the participle into the auxiliary, left-adjoining it (in keeping with Kayne 1994), as in (12a), or we could have the participle's projection undergo XP-movement into the specffler position of a functional projection (labeled FP) to the left of the auxiliary, as in (12b). These two scenarios are completely on a par with the two theoretical possibilities of objective case checking: noun incorporation, (13a), and object shift, (13b). [note 6]
(12)
a. [FP Spec [F’ [Vptci+Vauxlj + F [VP tj [VP ... ti ... ]]]]
b. [FP [VP ... Vptc ... ]i [F’ Vauxj + F [VP tj ti ]]]
(13)
a. [AgOP Spec [AgO’ [Ni+Vlj + AgO [VP tj [NP ... ti ... ]]]]
b. [AgOP [NP ... N ... ]i [AgO’ Vj + AgO [VP tj ti ]]]
Just looking at (1 l), we cannot make out which of the two strategies in (12) Frisian employs.
But we do believe that there is a way of settling this issue. We believe that the examples in (14) supply evidence in favor of the second approach (i.e. leftward XP-movement of the participle's projection). The Frisian verb hoeve 'need' in (14) embeds a passive to infinitival clause (kontro- learre te wurden); the participle consistently shows up all the way to the left of the verbal cluster, but the other two verbal members of the cluster

[1062 M. den Dikken and E. Hoekstra]
apparently freely change places, with the verb linearly following the infinitival marker showing up with the -n affix typical of te-infinitival verbs. [note 7]
(14)
a. ... kontrolearre hoeve te wurden
checked need to become + n
b. ... kontrolearre wurde te hoeven
checked become to need + n

How can the pair in (14) be accommodated? Consider first of all the underlying, strictly head-initial structure of the examples, which we assume reads as in (15). Here te is assumed to head a projection of its own; as in our initial structures in (12), an additional functional pro- jection, FP, is included outside the verbal cluster [note 8] - this FP, whose label presumably is AspP or (as in Fujita 1994, 1996: 170-171) VoiceP, serves as the locus for the checking of participial morphology (see below). [note 9]

(15) [Spec:zero [F:zero [V:hoeve [?:te [V:wurde [V:kontrolearre ... ]]]]]]

The derivation of (14a) can now be analyzed in terms of leftward move- ment of the participe kontrolearre into the checking domain of F; the other verbs will not vacate their initial positions prior to SPELL-OUT in (14a).
In order to end up to the left of hoeve te wurden in (14a), the participe could - given the structure in (15) - in principie get itself involved in three types of movement, summed up in (16). We will be led to adopt option (16c).

[Parasitic participles 1063]
(16)
a. kontrolearre head-moves to F in one fell swoop, skipping the intermediate heads;
b. kontrolearre head-moves to F via all intermediate heads, left- adjoining to each higher head and excorporating from it under subsequent movement (in conformity with Kayne 1994: 17);
e. kontrolearres VP is XP-moved to SpecFP.
Evaluation of the feasibility of the first of these a priori possible scenarios presupposes the construction of a theory of head movement and its locality. Head-skipping head movement is standardly deemed impossible as a consequence of Travis's (1984) head-movement constraint, which categorically prohibits all head movement across c-commanding heads. While we believe that the original head-movement constraint is too strong, we do want to continue to restrict the possibilities of nonlocal head movement. In our discussion of the syntax of double-participle constructions to be developed in section 4, we shall present evidence supporting the following revision of the HMC (to which we shall return in section 5, below, where we shall also address "long head movement" constructions):
(17) Relativized head-movement constraint:
Movement of a head X across a c-commanding head Y is illegiti-mate if X and Y share features and is forced otherwise.
This feature-oriented reinterpretation turns the old, rigid HMC into a minimality condition of the Rizzian type (Rizzi 1990) - a moving head is sensitive only to intervening heads that have some relevant property in common with the moving head (for feature-relativized minimality approaches along similar lines, cf. also Ferguson and Groat 1994; Kitahara 1994). Our approach to the HMC is thus close in spirit to Roberts's (1991) outlook on the locality of head movement, which relativized the HMC with regard to the A/A' status of the heads involved. We shall see later on that (17) yields us the desired results in the domain of double-participle constructions. Given (17), for which we believe there is solid evidence, we can now exclude option (16a), on the obvious assumption that wurde shares a feature with the participle - wurde, which is an auxiliary selecting a participial complement, has a participial feature (to be checked against the participle), just as a transitive verb possesses an objective case feature (to he checked against the object). The auxiliary wurde is hence an unskippable head under movement of the participle kontrolearre.
So suppose instead that kontrolearre climbs up to F undergoing strictly local, excorporating (Roberts 1991) head movement (cf. [16b]). From

[1064 M. den Dikken and E. Hoekstra]
the perspective of such a scenario, one would expect the participle to minimally be able to adjoin to the left of the auxiliary wurde - a step that it must take prior to hopping further up in the tree. lt turns out, however, that a word order in which the participle is sandwiches by te and wurde, which would arise if participle adjunction to wurde were a legitimate option in Frisian, is ungrammatical: *hoeve te kontro- learre wurde(n). From this we conclude that Frisian participles do not adjoin to the auxiliaries selecting them. With this said, the excorporation scenario in (16b) is straightforwardly ruled out on account of the fact that it wouid include unattested participle adjunction to the selecting auxiliary as an intermediate step in the derivation.[note 10]
This then leaves us with the third scenario, (16c) - the participle's projection undergoes XP-movement to SpecFP in the structure in (15) in the course of the derivation of (14a), as depicted in (18a). With kontrolearre's VP moving to SpecFP, the relativized head-movement constraint in (15), which is defined on heads, does not come into play. The analysis of (14a) is now essentially complete, as far as overt syntax is concerned - the remaining heads retain their underlying order and do not undergo overt syntactic movement.
(18)
a. [fp [vp kontrolearre ]i [f’ F [vp hoeve [tep te [vp wurde + n [vp ti ]]]]]]
b. [fp [vp kontrolearre ]i [f’ F [vp [v [te wurdej+te ]k + hoeve + n
[tep tk [vp tj [vp ti ]]]]]]
In (14b), by contrast, we assume that wurde incorporates into (i.e. left-adjoins to) [note 11] te, and wurde te subsequently incorporates into (i.e. left- adjoins to) hoeve, yielding wurde te hoeven. [note 12] This is the present-day incarnation of the classic verb-shuffling rule of verb raising - strictly left-adjoining verb movement in a consistently head-initial tree (cf. also Zwart 1995). The derivation of (14b) is shown in (18b). [note 13]
By eliminating all other theoretical options with regard to the placement of participles, then, we have now construed an empirically based argument for leftward XP-movement of participial VPs in Frisian. This is an important result. lt ties in neatly with what seems to be an inescap- able conclusion for Dutch multiverb constructions featuring a main-verb participle all the way to the left of the verb cluster such as (19): [note14]
(19)
(4-1-2-3)
dat hij het gedaan zou kunnen hebben
that he it do-PTC would can-INF have
'that he could have done it'

[Parasitic Participles 1065 ]
lt seems likeiy, then, that Continental West-Germanic participles (can) have their projections undergo XP-movement to a specifier position to the left periphery of the verbal cluster. Frisian is no exception in this regard. We have identified the position in question as SpecFP - the specifier position of the functional projection in whose checking domain the participial feature is checked. With this in mind, let us turn to the analysis of double-participle constructions in Frisian, the focus of our paper.

4. Double-participle constructions: analysis
4.1. The options
A priori, several analitical options come to mind when it comes to
dealing with PPI constructions. We have listed some in (20):
(20)
a. Rules of rightward/leftward spreading of participial morphology (-EN):
-(i) Vmain+EN - have - Vmodal --> Vmain+EN - have - Vmodal+EN
-(ii) Vmain - Vmodal+EN - have --> Vmain+EN - Vmodal+EN - have
b. Phonological assimilation ("participle harmony") "
c. Have" deletion
d. Syntactic movement
Of these options, (20a) is unilluminating. The main problem with such an approach is conceptual. The aim has heen, ever since the 1970s, to eliminate transformations since they are too strong, and hard to learn. The phonological assimilation approach in (20b), while perhaps justifiable in the case of the H > M PPI construction, in which the two participles are linearly adjacent, is difficult to maintain for M > H PPI constructions, in which ha intervenes between the "harmonizing" participles. And have deletion, (20c), though a feasible theoretical option in principie, is not an option for the Standard Frisian PPI constructions - neither of them has the semantics of a double have construction like I would have liked to have done it; and even if one of them did, it would still be impossible to derive the other in terms of have deletion as well. We are left, then, with option (20d), which we shall develop in detail.

[1066 M. den Dikken and E. Hoekstra]

4.2. Double participle constructions I: the H>M construction
As noted before, double participle constructions in Standard Frisian come in two kinds, differing with respect to the word order inside the verbal cluster, and concomitantly, with respect to the relative scope of the nonfinite modal and the auxiliary of the perfect. We shall discuss the two types one by one, starting with the construction in which the nonfinite modal is in the scope of the auxiliary ha. A relevant example is presented in (21). As (21b) shows, this PPI construction is grammatical in both root and embedded clauses. On the plausible assumption that the VPs in (21) are hierarchically arranged according to their scope properties, and on the further assumption (cf. note 8) that the F-projection in whose domain participial morphology is checked finds itself outside the projections of all verbs in Frisian, we end up with the initial structure for the construction in (21) given in (22).
(21)
ha>kinnen (H>M)
a. hy soe it dien kinnen ha
he would it done-PTC can-PTC have
b. omdat hy it dien kinnen ha soe
because he t done-PTC can-PTC have would
'(because) he would have heen able to do it'
(22)
[ Spec [ F [ V:soe [ V:ha [ V:kinnen [ V:dien NP ]]]]]]
The bold VP is moved to SpecFP.
Following up on our discussion of the Frisian examples in (14) in section 3, let us now first of all assume that the following constraint holds:
(23)
Frisian VPs in the complement of auxiliaries selecting participles
are moved overtly into the checking domain of F.

[Parasitic participles 1067]
In (22), then, the VP dominating kinnen is moved to SpecFP, prior to SPELL-OUT. [note 15] Recall that movement of the PROJECTIONof the participe is not subject to the RHMC by definition.
But movement to SpecFP of the bold VP in (22) in itself is not enough to guarantee the appropriate surface word-order pattern in (21). Several things happen in addition to movement to SpecFP. We must ensure that dien comes to linearly precede kinnen, that both these verbs can check their participial features, and that (in the embedded clause, in which soe undergoes no movement to verb-second position) ha comes to precede soe - see (2lb).
We assume that dien ends up to the left of kinnen by incorporating into (i.e. left-adjoining to) kinnen. [note 16] We assume that by adjoining to kinnen, which is the head of the VP in the checking domain of the participle-checking head F, dien is licensed as a participe (technically speaking, dien's participial feature can be checked). The "parasitic" participe dien in (21) is thus the result of a theoretical possibility which Frisian exploits to the full:
F can license or check multiple participes in its checking domain.
Multiple checking of the features of an F-head (cf. also Den Dikken 1994a) is a theoretical possibility that Chomsky (1995a: 432) appeals to in order to allow for "multiple assigmnent of Case and agreement from the same head" in multiple-specifier structures (cf. also Chomsky 1995b: 286). And for the specific case of F in our structure of PPI constructions, multiple checking of the participial feature contributed by the auxiliary can be made proper sense of from the perspectiva of Chomsky's (1995b: section 4.5) discussion of the difference between [+Interpretable] and [-Interpretable] features with respect to elimination under checking. Chomsky suggests that only [- Interpretable] features are deleted and erased; [+ Interpretable] features survive at LF and receive an interpretation there. Though he further argues that the features of the checker must necessarily be [- Interpretable] (which for our present case might have the unfortunate consequence that F's features could NOT be multiply checked), this conclusion can easily be seen to have been arrived at rather too quickly. The rationale behind the claim that features of the checker must be [-Interpretable] is that, if a constituent with a [+ Interpretablel feature F moved into the checking domain of a checker H with a likewise [+Interpretable] token of F, the movement operation would be "locally superfluous" - neither feature seeks elimination (each being [+Interpretable], hence not subject to deletion). This rationale is sound; the conclusion to be drawn from it is that the feature of at least one of the members of a checker/checked pair must be [-Interpretable] - but in principie that member can be EITHER the checker OR the checked; the choice depends on the case considered. [note 17]

[1068 M. den Dikken and E. Hoekstra]
With this in mind, consider now the case of the feature [+participial]. This feature is borne by participles (the "checked") and by auxiliarles selecting participles (the "checkers"). The feature [+participial] has a discrete semantic (aspectual) contribution to make. lt is plausible, there- fore, that one of the members of the checker/checked pair will possess a
[+ Interpretable] token of the participial feature. The question is which Now we happen to be in the privileged position of already having presented the evidence on which the answer to this question can be based - the facts of the Frisian PPI construction. As noted above, PPI constructions of the type in (2b) and (3b) contain two participles, one of which (the "parasitic" participle) is NOT interpreted as if it were a parti- ciple at all. lt cannot be the case, then, that the participial features of the PARTICIPLES contribute to the interpretation of participial construc- tions; instead, it must be the CHECKER (i.e. the auxiliary selecting the "true" participle) that is equipped with the [+Interpretable] token of the feature [+ participial]. [note 18] With this conclusion drawn, we may now understand how multiple checking of participial features is possible in Frisian PPI constructions:
The checker's [+participiall feature is [+Interpretablel (in the sense of Chomsky 1995b), hence is not deleted under checking, hence can be multiply checked by participles in the checker's checking domain.
Our appeal to multiple checking of the [+participial] feature under F in (22) is strongly confirmed by the behavior of PPI constructions in Standard Frisian. Suppose we adorn the example in (21a) with one more modal underneath the auxiliary of the perfect, wollen in (24). The multiple-checking approach now leads us to expect that it should be possible for ALL the verbs embedded under ha to show up in participial form. This prediction is borne out - (24c), with THREE participles (two of which are "parasitic"), is grammatical.
(24)
a. hy soe it dwaan kinne wollen ha
he would it do-INF can-INF want-PTC have have
b. hy soe it dwaan kinnen wollen ha
he would it do-INF can-PTC want-PTC have
c. hy soe it dien kinnen wollen ha
he would it do-PTC can-PTC want-PTC have
d. ??hy soe it dien kinne wollen ha
he would it do-PTC can-INF want-PTC have
'he would have liked to be able to do it'
Of course, nothing forces the verbs below the modal in ha's complement to be participial - kinne and dwaan may be included in the numeration

[Parasitic participles 1069]
unadorned with participial features, so long as the head of the VP moved into the checking domain of F (wollen in [24]) possesses a participial feature. Thus, alongside (24c) the variants in (24a) and (24b) are also perfectly acceptable. The status of (24d) is less clear, though. Our infor- mants tell us that it is highly marginal, as is indicated by the double question mark. lt is not quite clear what to conclude from this; but let us assume that (24d) is syntactically ill formed. The cause of this may then he sought in a "connected paths" approach to "parasitic" participles, á la Kayne (1984). The verbal cluster in the head of the VP raised to SpecFP is created by (i) left-adjunction of dien 'done' to kinne, and subsequent left-adjunction of this cluster to wollen, and thus has the intemal structure [[dien kinne] wollen]. In this cluster there is no connected path leading from dien to the participial feature under F, given that nonparticipial kinne intervenes. We thus assume that, like the better-known parasitic gap construction, the parasitic participle construction is subject to a connectedness requirement of sorts, violation of which is responsable for the deviance of (24d).
With (24d) thus set aside, we can return to the other examples in (24) and reiterate the important point that our account of the parasitic participle construction predicts all three surface outputs: one in which only wollen is participial, (24a); one in which wollen and the modal in its complement, kinnen, are participial, (24b); and one in which two "parasitic" participles are found, (24c). Especially the grammaticality of (24c) constitutes strong confirmation of the multiple feature-checking approach taken here. [note 19]
To handle the placement of ha relative to soe in the embedded clause in (21 b), we once again exploit the parallelism between case-feature checking and the checking of the feature [+participial]. Frisian is an object-shift language, so it picks option (13b) as far as case-feature checking is concerned - the object raises to SpecAgrOP, and its objective case feature is checked against the verb in AgrO (presumably directly, without any mediation by Agr0; cf. Chomsky 1995b: 255). Similarly, Frisian is a VP-movement language as regards the checking of the [ + participial ] feature of a selected participial VP - VP moves to SpecFP; and in order to render a local feature-checking relationship possible, the auxiliary ha (the "checker") must raise to F.
How does ha get to F? Does it adjoin to soe, with subsequent movement of ha + soe to F, or does ha move straight to F, skipping the finite modal intervening between it and its landing site? lf ha were to first adjoin to soe, a complex head would be formed headed by the modal; the modal being nonparticipial, there is no participial feature present on this complex head. Hence subsequent movement of ha + soe to F would be fruitless - ha would not he able to establish a checking relationship with the participial VP. What happens instead is the following: [note 20]

[1070 M. den Dikken and £ Hoekstra]
ha moves to F DIRECTLY, skipping soe.
This nonlocal head movement is legitimate from the perspective of our relativized head-movement constraint in (17) since ha and soe have no checking features in common [note 21] - soe is altogether invisible to the participially featured ha moving to the participial checking position F. In effect, ha is not just allowed but even FORCED to skip soe (as desired). By overtly moving to F across soe, ha will rightly end up to the left of soe in the linear string of the embedded clause in (21b). In the main clause in (2l a), soe will undergo additional movement to the verb-second position, details of which are immaterial in the present context.

4.3. Double-participle constructions II: the M > H construction
Now that we have an account of the H > M PPI construction in (21), let us move on to consider the derivation of the M > H PPI construction in (25), which, as (25b) shows, is ungrammatical in embedded contexts. [note 22]
(25) kinnen > ha (M > H)
a. hy soe it dien ha kinnen
he would it done-PTC have can-PTC
'he would be able to have done it'
b. *omdat hy it dien ha kinnen soe
because he it done-PTC have can-PTC would
'because he would be able to have done it'
This construction differs from (21) in that the scope relationship between ha and the nonfinite modal kinnen is reversed. This means that the projections of these verbs should change places in the inicial structural representation - in the structure of (25), ha's projection should be located in the complement of kinnen, as depicted in (26):
(26)
[ Spec [ F [ V:soe [ V:kinnen [ V:ha [ V:dien NP ]]]]]]
The bold VP is moved to SpecFP.

[Parasitic participles 1071]
One thing that will certainly happen in (26) is that the VP in the complement of ha (i.e. the VP headed by dien) moves to SpecFP in overt syntax. This will ensure that the main verb surfaces up front in the verbal cluster. But that is really the only easy part of the analysis of (25). What now remains to he accounted for is that (25b) is ungrammatical, that in the word order of the main clause in (25a) (in which soe is moved to the verb-second position) ha and kinnen change places, and that the modal kinnen can get its participial morphology licensed. Theoretically, we should ensure that ha, whose participial feature should be checked against the dien VP in SpecFP and also against the modal participle kinnen, ends up in F (the locus of [+participial] feature checking) in the course of the overt syntactic derivation.
Just as in the case of (22), we should primarily ask how ha moves to F. Can ha move straight to F in (26), just as in (22)? Recall that in (22) the head intervening between ha's base position and its landing site has no features that ha is itself specified for. The relativized HMC in (17) thus led us to conclude that ha can - in fact, must - move to F directly, without adjoining to soe first. Things are different in this regard in the structure in (26). lf ha, which possesses a participial feature, were to move straight to F, it would skip a head specified for a feature that ha is equipped with - the participle kinnen. Skipping kinnen is thus not allowed by the RHMC in (17). So what happens instead is that
ha's first movement step involves left-adjunction to kinnen, forming the cluster ha kinnen.
This cluster is specified for the participial feature (both members of the cluster possessing it); and since the modal kinnen is the (right-hand) head of the cluster, it is also specified for the feature [+modal].
Let us at this point pause for a second and ask what the conclusion that ha must left-adjoin to kinnen means from a minimalist perspective. This application of move does not itself seem to be triggered by myopic feature-checking considerations - ha does not check any feature via adjunction to the higher, "parasitic" participle. But adjunction of ha to kinnen does nevertheless qualify as a necessary step toward convergence. lf ha did not adjoin to kinnen, getting a free ride on the latter's back on the way to F, the auxiliary could never make it to F to get its [+participial] feature checked. And notice that, since ha does indeed get a free ride to F as a result of kinnen movement (after adjunction of ha to the "parasitic" participle), the derivation that we are led to is in fact not more costly than the alternative in which the auxiliary and the “parasitic" participle each move to F on their own - both derivations involve two head-movement steps: ha-to-F and kinnen-to-F on the (illegit

[1072 M. den Dikken and E. Hoekstra]
imate) "independent movement" scenario, and ha-to-kinnen and [ha + kinnenl-to-F in the "free ride" derivation. The choice between the two scenarios is unambiguously determined, in an appropriately "local" fashion, by the demands imposed by the RHMC. [note 23]
As already intimated in the previous paragraph, after adjunction of ha to kinnen the entire cluster ha kinnen moves on toward F, to establish a configuration in which the [+participial] features of the auxiliary of the perfect and the two participles can be checked, as desired. [note 24] But note that ha kinnen is still separated from F by the finite modal soe. Again, therefore, a question of locality arises: how can ha kinnen (the cluster that is specified [+participiall) move on to F? Suppose first that soe is in place, as in embedded clauses (in which the modal undergoes no movement, in any event not in overt syntax). Now recall our relativized head-movement constraint in (17). The cluster ha kinnen is a complex modal, soe is a modal as well. Hence,
Movement of the ha kinnen cluster across the modal soe is prohibited by the relativized HMC since soe possesses a feature relevant to the moving ha kinnen cluster.
So the only way for ha kinnen to continue its upward journey in embedded clauses is for it to first left-adjoin to soe, yielding ha kinnen soe. Still excluding the possibility of excorporation for Frisian (cf. notes 10 and 24), we reach the conclusion that subsequent upward head movement will affect the entire cluster ha kinnen soe.
Now notice that while this cluster is specified for the feature [+modal], it is NOT specified for the feature [+participial] - the head of the cluster does not possess this feature (instead, it is specified [+ finite]). This means that moving ha kinnen soe to F will NOT establish a checking relationship with the participial feature of the VP in SpecFP. The conclusion that ensues, then, is that:
There is no grammatical derivation for (25b).
This is in perfect conformity with the empirical facts, (25b) being ungrammatical.
What remains to be answered is why the root-clause counterpart of (25b), given in (25a), is fully acceptable. The difference between (25a) and (25b) lies in the fact that in the former, the finite modal soe undergoes movement to the verb-second position. Let us assume that:
Movement of soe to verb-second position leaves behind a trace lacking modal features. [note 25]

[Parasitic participles 1073]
T'he [+modal] feature is hence lacking on the highest Y-node in (26) in the root-clause example in (25a). This being said, movement of ha kinnen, specified [+modal], across the TRACE of soe, which is unspecified, is perfectly legitimate with respect to our relativized HMC in (17). The root/nonroot asymmetry in the M > H PPI construction in (25) now follows: strictly local pied-piping head movement, the only possibility in the nonroot construction, leads to failure of feature checking in FP; in the root clause, the ha kinnen cluster, possessing the requisite participial feature, can move to F across the TRACE of soe. [note 26]
The story of (25) is almost complete now. We have accounted for the word order found in (25a) - dien up front, in SpecFP, and ha kinnen following it, in F. We have also furnished an explanation for the root/nonroot asymmetry in (25a) and (25b). And we have accommodated the participial form of the main verb, dien. One remaining question concerns the form of the nonfinite modal, kinnen - a "parasitic" participle. This is now straightforward. In the derivation of (25a), ha kinnen ends up in an adjunction position to F - that is, in a position in the CHECKING domain of F. In conformity with Chomsky's theory of checking domains, then, F can check its participial feature against that of the modal under F. Recall that we have argued that multiple checking of the participial feature under F is possible. And again we find confirmation of this assumption in the domain of PPI constructions. Just as in the H > M construction, TRIPLE participles (two of which are "parasitic") are possible in multiple modal constructions of the type in (27):
(27) hy soe it dien ha kinnen wollen
he would it do-PTC have can-PTC want-PTC
'he would like to he able to have done it'
We now have a full account of the examples in (25), which differ from the H > M constructions in (21) in that the projection of the "parasitically" participial modal (kinnen) dominates that of the auxiliary of the perfect and forces a local, pied-piping derivation upon ha movement to F. The account of the root/nonroot contrast in (25) offered here has leaned heavily on the relativized head-movement constraint in (17), which bars movement of ha across kinnen in (26) since kinnen, like ha, is in the possession of the [+ participial] feature. This account makes the predic- tion that "regular" (i.e. non-PPI) M > H constructions, in which the nonfinite modal is infinitival rather than participial, should be fine in both root and nonroot clauses. This prediction is borne out, as (28) shows:

[1074 M. den Dikken and E. Hoekstra]
(28) (M > H non-PPI)
a. hy soe it dien ha kinne
he would it done-PTC have can-INF
'he would be able to have done it'
b. omdat hy it dien ha kinne soe
because he it done-PTC have can-INF would
'because he would be able to have done it'
Since in (28) the modal governing ha does not possess a participial feature, movement of ha across it is not barred by (17) the way it was in (26). And since ha has no relevant feature in common with soe either, ha can move straight to F in the derivation of (28). The contrast between (28b) and (25b) is thus readily accounted for given our analysis. [note 27]

4.4. Why Dutch has no PPI constructions
We now have an account for the two types of Frisian PPI constructions. As we noted in the introduction, both these constructions are absent from the grammar of Dutch. Why? The obvious thing to do in our search for an answer to this question is to relate this difference between Frisian and Dutch to another robust difference between the two languages - word order in the verbal cluster. In Frisian, the finite verb always comes last in the linear sequence of verbs in embedded clauses; the most deeply embedded verb comes first. In Dutch multiverb clusters, this is precisely the other way around. What does this mean from a theoretical perspective? Basing ourselves, as before, on a strictly head-initial initial structure for both Dutch and Frisian, we can say that in Dutch, verbs do not cluster. In Frisian, on the other hand, verbs do undergo overt clustering. lt is to this verb clustering that the difference with respect to the distribution of PPI between Dutch and Frisian can be related.
In the derivation of the H >M PPI construction in (21), dien left-adjoins to kinnen, which heads a VP in the checking domain of F. In the derivation of the M > H PPI construction in (25), ha left-adjoins to kinnen, after which the complex ha + kinnen moves to an adjunction position in the checking domain of F. In both cases, the "parasitic" participle (dien in (21), kinnen in (25) ends UP, AS A RESULT OF VERB CLUSTERING, in a position in the checking domain of the F-head checking the participial feature. This is how "parasitic" participles are licensed:
"Parasitic" participles are licensed in head-adjunction positions in F's checking domain.

[Parasitic participles 1075]
In Dutch, since no cluster formation takes place, the configuration licensing "parasitic" participles never arises, which is why PPI constructions are ungrammatical in Dutch. [note 28]
So we see that the difference between Dutch and Frisian with respect to the distribution of PPI constructions is reducible to the independent difference with respect to verb raising between the two Germanic dialects.

5. On the locality of head movement, and "move" versus "attract"
One of the major theoretical claims of this paper lies in our revision of Travis's (1984) head-movement constraint along the lines of (17), repeated here:
(17) Relativized head-movement constraint.,
Movement of a head X across a c-commanding head Y is illegitimate if X and Y share features and is forced otherwise.
In the cases discussed above, this RHMC correctly ensures that modals cannot move across modals and that participles may not head-move across auxiliaries of the perfect and vice versa. [note 29] At the same time it enables the auxiliary of the perfect to move across nonparticipial modals: moving heads may skip any c-commanding heads with which they share no features.
But as things stand, it may seem that our RHMC is both too lenient and too strict. lt would seem to allow for ungrammatical examples of the sort in (29b), while it disqualifies an approach to sentences of the type in (30b) in terms of long head movement, as in the work of Rivero and her colleagues (cf. Rivero 1991 and references cited there).
(29)
a. will John buy that?
b. *buy John will that?
(30) (Serbo-Croatian)
a. Jovan je istukao Petra
Jovan is beaten Petar
'Jovan beat Petar'
b. istukao je Petra
beaten is Petar
'(he) heat Petar’
The problem is this: in (29b) there is nothing in the feature specifications of the moving infinitive and the finite modal that would make our RHMC prevent head-skipping head movement to Comp, yet (29b) is ungrammatical; and in (30b) the RHMC does supply a straightforward ban on

[1076 M. den Dikken and E.. Hoekstra]
participle movement across the auxiliary, yet long participle movement, to Comp on Rivero's assumptions, is apparently grammatical.
Tuming to the examples in (29) first, we have seen that our RHMC does not rule (29b) out. Does this mean we cannot accommodate (29)? The answer is "no," for there is a straightforward way of preventing sentences of the type in (29b) from ever being derived, as Zwart (1993: 19) convincingly argues (cf. also Boskovic 1994: 12-13). On minimalist assumptions, all movement is feature-driven. The verb moving to C in the yes/no questions in (29) hence must he forced to move there for feature-checking purposes. Now, it is well known that Y-movement to C, also in the absence of head-skipping configurations, is restricted to FINITE verbs - see, for example, the Dutch examples in (31):
(31)
a. zal Jan dat kopen?
will Jan that buy
b. Jan dat kopen? ( - vergeet 't maar!)
Jan that buy ( - forget it)
e. *kopen Jan dat? vergeet 't maar!)
buy Jan that ( - forget it)
The finite verb zal in (3la) readily moves to C in questions; the infinitive in the reduced question in (31b) and (31c) cannot, on the other hand. C, then, welcomes finite verbs and bans nonfinite verbs - put differently, C possesses a finiteness feature that attracts finite verbs. Now let us return to the illegitimate head-skipping example in (29b). The deviance of this example can now be readily understood: there is no motive for nonfinite verb movement to C across the finite verb. Thus, even though the RHMC in principle allows for movement of buy across will in (29b), there are independent considerations that ensure that constructions of the type in (29b) will never arise.
By the same token, and in agreement with the conclusion drawn by Boskovic (1994: 12-13), we are led to reject a Rivero-type movement- to-Comp approach to the Serbo-Croatian example in (30b) (cf. also Cavar and Wilder (1992), Wilder and Cavar (1993) for a different variant of the movement to C account). Since Comp arguably possesses no feature that could ever attract a nonfinite verb, there is no motive for long participle movement to Comp, hence such movement is illicit. That our RHMC rules out such a scenario for (30b) thus does not turn out to be a dramatic defect. On the contrary, Boskovic (1994) has recently argued on both theoretical and empirical grounds that the participle does not move to C in (30b) at all. He assumes instead that the participle in (30b) left-adjoins to the auxiliary (following Wilder and Cavar (1993) in assuming that this adjunction is triggered by the feature [+Aux]) in a

[Parasitic participles 1077]
position lower than C); alternatively, one might assume that the parti- ciple’s PROJECTION is raised to the specifier of the functional head in whose checking domain the participial feature is checked. [note 30] lt may in fact be that Serbo-Croatian employs both options -just as Frisian does (cf. our account of double-participle constructions above: one participle has its projection undergo movement to Spec, the other is part of an X-zero cluster). The basis for this suspicion lies in the following observation about Serbo-Croatian past perfectives (from Boskovic 1994: section 2.1). Serbo-Croatian past perfectives feature two tokens of the auxiliary of the perfect (biti 'be'), one finite and one participial, and a participial main verb, as (32a) shows. Interestingly, now, in past perfective constructions lacking an overt subject (the context in which participle fronting to a position to the left of the auxiliary obtains), two surface outputs are possible, given in (32b) and (32c):
(32) (Serbo-Croatian)
a. vas dvoje ste bili ccekali Marijinu prijateljicu
you two were been waited Marija's friend
‘you two had been waiting for Marija's friend'
b. bili ste ccekali Marijinu prijateljicu
been were waited Marija's friend
e. ccekali ste bili Marijinu prijateljicu
waited were been Marija's friend
'(you) had been waiting for Marija's friend'
Of the two alternatives in (32b) and (32c), the former can be derived by left-adjoining the participle bili to the finite auxiliary ste (in confor- mity with Kayne's (1994) antisymmetry theory), with subsequent move- ment of the complex head to F, while the latter can be taken to involve movement of the projection of the participial main verb ccekali to SpecFP. (lt suffices in overt syntax to move ONE participle into the checking domain of F; whether or not the nonovertly moved participle raises at LF is immaterial here.) Note that the derivation of (32c) cannot, on our assumptions, involve movement of the participial main verb ccekali across bili - this is blocked by the RHMC. But it can legitimately front to a position to the left of the finite auxiliary by having its projection move to SpecFP. [note 31]
Leaving further details aside, we conclude that there is evidence, quite independently of the discussion in this paper, that (30b) involves no nonlocal movement of the participial head across the auxiliary of the perfect. This eliminates the second potential empirical threat to the RHMC in (17).

[1078 M. den Dikken and E. Hoekstra]
So far we have been concerned in this section with potenciai problems for our RHMC. At least as interesting, however, would be a theoretical argument - alongside the empirical one that the Frisian facts discussed in this article constitute - that we cannot do without the RHMC. We believe that such an argument can indeed be constructed. In particular, we believe that there is an indication that the minimalist theory should adopt a movement approach (constrained by the RHMC) rather than Chomsky's (1995b: chapter 4) "attract" perspective.
Before tuming to the actual argument, however, let us step back for a moment and consider the status of the RHMC. Clearly, it can function only in the "move" approach of all principles-and-parameters, work up to Chomsky (1995b: section 4.5.6); from the "attract" perspective one might think that the effects of the RHMC fall out immediately, given that a functional head will always attract the ciosest feature bundle. But "attract" does not yield the right result in the case of the M > H PPI construction in (26). lt would trigger movement of kinnen to F and of the dien VP to SpecFP, and the participial feature of ha would never get checked. While this problem might perhaps be circumvented, there is a stronger, more general reason why the "attract" scenario incurs problems - problems that do not arise under the movement-cum-RHMC approach.
Consider again the structures in (13), repeated here:
(13)
a. [AgOP Spec [AgO’ [Ni+Vlj + AgO [VP tj [NP ... ti ... ]]]]
b. [AgOP [NP ... N ... ]i [AgO’ Vj + AgO [VP tj ti ]]]
These structures represent the two alternative ways of checking the objective case feature of the object and the transitive verb: noun incorporation (as in [13a]) or object shift (as in [13b]). Tertium non datur - that is, there is no such thing as a derivation of the type in (13c), where the object's head moves straight to AgrO and the remnant VP is raised to SpecAgrOP:
c. * [AgOP [VP V [NP ... N ... ]]i [AgO’ Nj + AgO [VP ti ]]]
Technically, in the configuration in (13c) it is perfectly possible to check the object's case feature against that of the verb (which is projected up to VP). In effect, (13c) is the reverse of (13b) with regard to which of the two - the projection of the object or that of the verb - undergoes movement to SpecAgrOP. lf a feature-checking relationship between V and N can be established in (13b), then there is no a priori reason to expect checking to fail in (13c). Nonetheless (13c) does not seem to be attested. The question is why not.
An uninteresting and problematic answer would he that checking rela- tionships are asymmetric in a relevant sense - even though it may well

[Parasitic participles 1079 ]
be true that such relationships are intuitively asymmetric, there is no obvious reason to expect checking to be asymmetric given a Spec-Head agreement approach, and what is more, the "direction" of the asymmetry seems to be different from one type of checking relationship to another (cf. phi-feature agreement vs. case), and it would thus be very difficult to formalize this intuitive asymmetry in an adequate fashion (cf. Chomsky 1995b: 259).
Adhering to the theoretically simpler view that checking relationships are symmetric, we cannot rule (13c) out on account of alleged failure of feature checking. What does Chomsky's (1995b: chapter 4) "attract" approach to movement have to offer, then, to block (13c)? Nothing, it seems. AgrO attracts the closest case-feature and Y-feature bearer to its specifier (viz. VP), and the head of the object is attracted to AgrO in order that a case-feature checking relationship between Y and N can be
established within AgrOP. [note 32] Nothing seems to go awry in this derivation.
By contrast, (13c) is straightforwardly ruled out given a movement
approach assuming the RHMC. Movement of N to AgrO across Y is a case of head movement across a head that shares a feature with the moved head - a case of head-skipping that is filtered out by the RHMC. The movement cum RHMC approach thus manages to block the illegitimate derivation in (13c), which the "attract" scenario of Chonisky (1995b: chapter 4) does not seem to do. This, we believe, is a theoretical argument in favor of the approach taken here.
In sum, we believe that the locality of head movement is curbed by one overarching factor - movement is triggered by feature-checking considerations only; movement is oblivious to structurally intervening but featurally irrelevant positions. This outlook on head movement is entirely minimalist. In fact, Chomsky (1995b: 307) notes that "there is nothing to prevent alfa from skipping some head gamma that offers no features to be checked." Yet, while leading him to retreat from his strict interpretation of the original HMC in Chomsky (1995a: 433), his reinterpretation of all LF movement as - often nonlocal - FEATURE movement to Xo (cf. Chomsky 1995b: chapter 4) does not quite win him over to the RHMC "camp" as yet." [note 33] Chomsky asks (1995b: 307), "Can HMC fall within the framework just outlined?" and answers "That seems doubtful, continuing to leave the status of the HMC in the bare theory unclear (also cf. Chomsky 1995b: 328). lt is our conviction that the HMC can plausibly be relativized along the lines of (17), fully in keeping with minimalist assumptions. lt remains to be investigated in detail whether the general principies of the theory are sufficiently developed to accurately restrict the distribution of head-skipping head movement.
 

[1080 M. den Dikken and E. Hoekstra]

6. Concluding remarks
Frisian allows sentences with only a single instance of the auxiliary of
the perfect (and with "monoperfective" semantics) to contain multiple participles as a result of.a conspiracy of the theoretical option of multiple checking of the participial features of the auxiliary of the perfect (I) and the empirical fact that Frisian features left-adjoining verb raising (II), which, in the cases discussed, creates complex Xo categories whose members all end up in the checking domain of the functional head managing the checking of the feature [+participial]. Double-participle constructions in Standard Frisian come in two forms, differing with respect to word order and, concomitantly, with respect to the relative scope of the auxiliary of the perfect and the nonfinite modal. The syntax of these constructions can he analyzed on the assumptions in (III) and (IV):
I Multiple checking of the participial feature is provided by the auxiliary of the perfect and is checked in the checking domain of a functional head F (Asp/Voice) outside VP.
II Multiple checking is facilitated by left-adjoining verb raising, which Frisian has but Dutch lacks.
III Participial VPs raise overtly to the specifier position of FP in Standard Frisian.
IV Nonlocal head movement is not categorically excluded; the rela- tivized head-movement constraint curbs the distribution of head- skipping head movement.
Let us end this discussion of severaL types of auxiliary verb construc-
tions on a theoretical note. We started out by noting that the two Standard Frisian types of PPI constructions, in (2b) and (3b), above, are not semantically equivalent - in the former the modal takes scope over the auxiliary of the perfect, in the latter these scope relations are precisely reversed. We took this to mean that the initial structures of the two construction types are different. Scope, then, can he read off the initial representation of these sentences.
But in the course of the derivation, all sorts of things happen in Frisian PPI constructions - participial VPs move to SpecFP; verbs move up, skipping intervening verbs with which they have nothing in common; and verbs cluster via left-adjunction. The results of these movements apparently do not affect scope relationships. How can this be?
The fact that the initial scope relationships between the verbs in Frisian PPI constructions are unaffected by the various overt syntactic movement operations that the verbs and verbal projections in our structures of PPI constructions undergo can be accounted for if we start out from

[Parasitic participles 1081]
Chomsky's (1993) view of auxiliaries and their LF role. [note 34] Chomsky asserts that auxiliaries are LF-invisible altogether, hence will not receive an interpretation. Nonetheless there is no denying that the presence of a modal or aspectual auxiliary in a sentence does contribute to the semantics of that sentence. How should we conceive of the semantic effect that modals and aspectuals have? In line with Kratzer (1977), McDowell (1987), Brennan (1993), and others, we can assume that auxiliares make a QUANTIFICATIONAL contribution to semantics (see also Bennis et al., this issue). More specifically, what we may assume is that this quantificational contribution is made via operators located at the periphery of the clause (in the A-bar-domain). We may now conjecture that the relative ordering of these operators, to which the scopes of the various auxiliarles are imputed, matches the syntactic hierarchization of the projections of the auxiliarles (a conjecture that is plausible from a "crossing-rather-than-nesting dependencias" point of view; cf. Chomsky 1993). In the structure in (22), then, the modal and aspectual operators associated with soe, ha, and kinnen, are stacked in such a way that the first has scope over the other two and the second has scope over the third; in (26) the relative order of the operators associated to ha and kinnen is reversed.
This outlook on the scope relationships among auxiliary verbs (or some- thing along similar lines) seems independently necessary if Chomsky's (1993) "LF-invisibility" thesis is to be embraced. With it we account for the fact that scope interpretation is oblivious to the various reorderings of the verbs and verbal projections that we evidenced in our account of Frisian PPI constructions.

Received 7 November 1995
Revised version received 12 June 1996
Free University Amsterdam, P. J. Meertens Institute, Amsterdam

Notes
We would like to thank, for comments and discussion, Frits Beukema, Jarich Hoekstra, Jan-Wouter Zwart, the guest editors of this volume, two anonymous reviewers, and the audiences at the Taalkundich Wurkferbán of the Fryske Akademy, Ljouwert (19 March 1994), the TABU-dag, Groningen (24 May 1994), and the 10th Comparative Germanie Syntax Workshop, Brusseis (18 January 1995). Correspondence addresses: M. den Dikken, Vakgroep Taalkunde (ATW), Vrije Universiteit Amsterdam (HIL), De Boelelaan 1105, 1081 HV Amsterdam, The Netherlands; E-mail dikken@jet.let.vu.nl; E. Hoekstra, Koninklijke Akademie van Wetenschappen, Fryske Akademy, P.O. Box 54, NL 8900 AB The Netherlands, e-mail ehoekstra@fa.knaw.nl

[1082 M. den Dikken and E Hoekstra]
1. The fodowing abbreviations are used in the glosses:
INF infinitive ,
IPP lnfinitivus-pro-Participio: infinitive where past participle would be expected to be found
PTC past participle (in italics: "parasitic" past participle - PPI, Participium- pro-Infinitivo)
2. Matrix participles in Standard Dutch may either precede or follow the verb cluster (cf. also note [19], below). Especially in more complex verb clusters, many speakers prefer ordering the participle hefore the cluster, though for a substantial number of speakers it is also possible to position the participle after the verb cluster (as in (1a); cf. the map in Stroop (1970:142). There does not seem to be a meaning difference associated with this variation in the positioning of the participle.
3. Alongside (3a) and (3b), a third way of expressing 'he would have liked to do it' exists in Frisian exemplified in (i). Thís example, like (3b), again features a participial main verb in spite of its nonperfective interpretation; it differs from (3b) in that the modal willen is CONTRACTED with the finite irrealis marker soe, yielding the suppletive lexicalization woe 'would'. This construction (which is studied in detail in Hoekstra 1990) was quite common in nineteenth-century written Frisian, although it is now essentially obsolete. Space restrictions prevent us from discussing this very interesting construction here; we hope to retum to it in detail in future work.
(i) hy woe it dien ha
he would it do-PTC have
4. To our knowiedge there are no modals that do not take part in the PPI construction. The counterfactuality requirement (whatever its root) is responsable for the ungrammaticality of putative PPI examples like *hy hat it dien wollen 'he has it do-PTC want-PTC'.
5. Zwart (1995) argues essentially the same point, based on West Fiemish facts originally reported in Den Dikken (1994b); see Haegeman (1995) for more recent discussion. Given a strictly head-initial underlying structure (Kayne 1994) and an analysis of verb- projection raising along the lines of Zwart (1993) and Den Dikken (1994b), (i) can only be derived by moving the VP in the complement of the auxiliary of the perfect eet 'has', i.e. willen dienen boek kuopen, to a specifier position to the left of eet.
(i) (West Flemish)
da Valére willen dienen boek kuopen eet
that Valére want-IPP that book buy-INF has
'that Valére wanted to buy that book'
Notice also that, as Zwart notes, a word-order pattern of the type in (i) is impossible if eet is replaced with zou; cf. (ii). XP-movement to Spec is thus impossible for complements of auxiliaries not selecting a participle.
(ii) *da Valére willen dienen. boek kuopen zou
that Valere want-INF that book buy-INF would
6. In (13) we adopt the Agr-based structure of Chomsky (1993, 1995a), not the Agr-less alternative proposed in Chomsky (1995b: section 4.10), because the former expresses the parallel between (12) and case-checking constellations more adequately.
7. An empirical note on the distribution of the infinitival marker te in these examples: te can be omitted in (14b) but not in (14a).

[Parasitic participles 1083]
8. We assume that FP is located entirely outside the verbal cluster in Frisian since Frisian is not a verb-projection raising language; cf. Den Dikken (1994b: section 5) for the view that VPR dialects of West Germanic differ from non-VPR dialects in allowing certain functional projections to be interspersed with V-projections.
9. The nature of the te phrase in (15) (TP or "something else") is unclear to us; while leaving this question open, we do emphasize that (14) strongly suggests that te must be an independent head in Frisian and cannot be taken to be a base affix of the te-infinitival verb: it behaves independently from the verb.
In (15) and all other structures to fodow, all SpecVP positions are ignored since they do not play a role. Movement of the participial VP to SpecFP across potential intervening specifier positions should not lead to locality problems. This can plausibly be ensured by stating the locality condition on XP-movement in a way similar to our relativized head-movement constraint in (17), below, such that movement of VP to SpecFP is not sensitive to the intervention of NP-specifier positions (of VP or teP in [15] and the structures to follow). We shall not work this out in any detail, the locality of XP-movement being tangential to our concerns in this paper.
10. In a restrictive theory assuming head-skipping head movement such as ours (which incorporates (17)) there is presumably no room for excorporation to begin with. Given that the two are essentially two alternative ways of accommodating long-distance head movement, it is reasonable to require of a theory settling on one of the two that it avoid making simultaneous use of the other. We shall assume that excorporation is not a feasible option for Frisian (which, we argue, employs head-skipping head movement); but we should expect the theory as it stands to accurately curb the applicability of head-skipping head movement and excorporation in the appropriate way, so that there should be no need to categorically eradicate excorporation by brute force. A minimalist theory decrees that excorporation be a TRIGGERED operation (cf. Watanabe 1993). From this perspective, too, an excorporation approach to (14a) does not seem to make much sense. In (14a), the participle seeks to check its participial feature against that of the auxiliary governing the participle. Such checking wouid he accomplished by adjunction of the participle to wurde (if not immediately, then certainly after subsequent movement of the verb complex to a functional head); any further movement of the participle is then barred since it has no trigger. Adjunction of kontrolearre to wurde would hence effectively bleed participle movement into the checking domain of F; the participle would be prevented from reaching its desired surface position to the left of the rest of the verbal cluster. This reinforces our text conclusion that excorporation is not a feasible option to derive (14a).
Chomsky (1995b: 285) briefly discusses the possibility of excorporation in the hare theory. On the assumption that incorporation of an element alfa into beta is triggered by a morphological feature [affix], which is presumably [-Interpretable], "excorporation of alfa will he impossible because the feature [of alfa] will have been erased and will thus be unavailable for further checking." As a categorical statement about excorporation, this does not seem entirely correct. Excorporation of alfa from beta might still be triggered in principle by some other feature of alfa's that has not yet heen checked by incorporation into beta. Be that as it may, the bare theory does not seem to leave much room for excorporation, certainly not in the case of (14a). We therefore set (16b) aside.
11. In conformity with Kayne (1994).
12. On standard assumptions, (14a) presumably undergoes at LF the head-movement operations that (14b) instantiates in overt syntax. But see Hornstein (1995: 9lff.) for an interesting argument against the LF application of V-movement, based on the

[1084 M. den Dikken and E. Hoekstra]
cross-linguistic distribution of ACD constructions. We have no insights to offer with respect to the apparent optionaiity of the verb incorporation process; nor is it clear to us what triggers these head adjunctions.
13. Apart from word order, the two members of the pair in (12) also differ with regard to the surface realization of the verb following te and the (non)deletabiiity of te (on the latter, cf. note 7, above). These are extremely complex issues, which we shall not attempt to solve here; see Visser (1989) for data and discussion.
14. See Zwart's (1994) assumption for Dutch and German that "past participles have a feature which must be licensed in the specifier of an agreement phrase." Also cf. Zwart (1995), where it is assumed, however, that participial phrases move to the specifier of the auxiliary's projection itself; we must adopt the FP approach (in line with Zwart 1994) in our account of "parasitic" participles in Frisian.
15. A reviewer asks how this VP-movement approach to SOY order could capture the order Y-Aux-CP. lf it is assumed that CP remains in its base position inside VP (a "stranding" approach á la Kayne 1994), this word-order pattern seems impossible to derive on our analysis. Koster (1995) argues against this stranding approach, however. Instead, he presents an account of CP "extraposition" in terms of asyndetic coordination of a CP with an empty NP in the verb's complement. A sentence like Dutch (i) would then he analyzed as in (ii) (where ":" marks the head of the "colon phrase" of asyndetic coordinations). We will adopt Koster's (1995) analysis.
(i) dat Jan beredeneerd heeft dat de strandinganalyse niet deugt
that Jan argued has that the standing analysis not works
(ii) dat Jan [vp beredeneerd [np zero]] heeft [:[cp dat de strandinganalyse niet deugt]]
16. Recall from our discussion of (14b) above that Frisian features left-adjoining verb raising; our adjunction of dien to the left of kinnen is hence not a construction-specific artefact of the analysis (although it continues to be unclear what triggers "verb raising," as noted earlier).
17. By tacitly assuming that whenever the feature of one of the pair's members is [+Interpretable], the member in question will be the checked, not the checker, Chomsky reaches the more specific conclusion that the checker must of necessity have [- Interpretable] features only. In practice, though, Chomsky's implementation of this conclusion is not entirely consistent. Thus, in the case of multiple case-feature checking in multiple-specifier structures, he assumes that "a [-Interpretable] feature F of H is not necessarily erased when checked and deleted, a parameterized property" (Chomsky 1995b: 286); and in sections 4.10 and 4.11, Chomsky dees assume that functional heads such as D and T have "features that survive through the derivation and appear at the interfaces, where they are interpreted" (1995b: 378).
18. Note that it cannot he assumed (as one reviewer suggests) that of the two participles in the Frisian double-participle construction one is [+Interpretable] and the other is [-Interpretable]. Apart from the fact that such an approach does not seem very principled, the Dutch Infinitivus-pro-Participio (IPP) effect (cf. [ la], above) shows that it cannot be correct. In IPP constructions, no morphological participle shows up, but the aspectual interpretation of the construction is unaffected - it must be the auxiliary that contribuyes this interpretation.
19. Last resort is not violated by the OPTIONALITY of multiple checking. Frisian F is set for multiple checking, which means that it can check either one or more participial features. F's participial feature must be checked, but the number of participles checking it is

[Parasitic participles 1085]
inconsequential. Regardless of whether there are three, two, or one participle(s) taken from the lexicon, they CAN AND MUST be checked against F, unproblematically. We avoid problems with the last-resort principle by building in the optionality that the data reflect into the parameter itself.
20. In the discussion of (14b) we have seen that modal verbs can attract the infinitival head of their complement and have it left-adjoin to them (cf. [18b]). This is the antisymmetric incarnation of verb raising. But finite modais do not (obligatorily) trigger left-adjoining verb raising. lf they did, they would never be able to raise up to Comp in verb-second clauses, for instance (since excorporation of hosts is illegitimate; Kayne 1994: 17). That ha skips soe is thus not problematic. Notice further that the infinitival form of the verb is arguably a default form (involving no morphological feature checking in a functional projection of sorts) - cf. the Infinitivus-pro-Participio effect in standard Duteh, where verbs receive infinitival morphology by default, whenever assigning them the feature f+participial] is blocked.
21. On infinitive "selection," see the closing lines of note 20, above. A reviewer points out that in tense-chain theories (such as the one elaborated in Guéron and Hoekstra 1988) there is feature sharing through chains of auxiliarles. This does not affect our point, though. Whatever the precise nature of the features shared through T-chains, these will of course not be formal features like [± finite].
22. The Frisian M > H PPI construction is on a par in this regard with the double-participle construction that Bloemhoff (1979) reports for the Stellingwerf dialect (spoken in the border arca of Friesland, Drente, and Overijssel, and bordering on southern Frisian dialects): in the Stellingwerf dialect the PPI construction is also ungrammatical in embedded contexts. See Zwart (1995) for interesting discussion, restricted to the Stellingwerf type of PPI construction; Standard Frisian is richer (in having two types of double-participle construction) and is not amenable in general to an approach along the lines of Zwart (1995), as far as we have heen able to establish.
23. This passage has a bearing on Chomsky's (1995) discussion of the three interpretations of last resort given in his (20) (1995b: 257). Our case seems to be one of intermediate movement that facilitates feature checking later in the derivation - ha adjoins to kinnen on the way up to F, with a view to getting the raised VP's participial feature checked by the end of the derivation. lt thus looks like Chomsky's (20c): "the operation is a necessary step toward some later operation in which a feature of alfae [or, note 35, of the ultimate target] will be checked." Chomsky (1995b) rejects this option; it seems that we must allow for it (for otherwise any derivation built on the structure in [26] would crash). In this connection, consider the discussion in Chomsky (1995b: 345-348): "we select Attract/Move [rather than Merge) even violating Procrastinate if that is necessary for convergence." The need to arrive at a converging derivation will always have the final say, also in the text example.
24. The alternative derivation involving excorporation of ha from the ha kinnen cluster formed by left-adjunction (all in accordance with Kayne 1994, in principle) can readily be discarded. Empirically, Frisian modais must surface to the right of the verbs they embed in the surface string of subordinate clauses: omdat hy dat dwaan soel*soe dwaan 'because he that do would/*would do'. Arguably, then, some sort of leftward movement will have to affect kinnen. Theoretically, kinnen must make it to F to get its [+participial] feature checked. And clearly, it is more economical to select the entire ha kinnen cluster for movement to F than to move ha and kinnen separately.
25. A reviewer objects to this claim with reference to the copy theory of movement. lt is not obvious, however, that lower copies (i.e. traces) must of necessity (continue to)

[1086 M. den Dikken and E. Hoekstra]
possess all the features of their antecedent. Chomsky's (1995b: 303) view of the matter is that "formal features of trace are deleted (hence erased) if they are not necessary for the formation of legitimate LF objects that satisfy FI." Since we believe that [+ modal] is such a feature (cf. the discussion in the final paragraphs of section 6, below), we find the text claim viable.
26. H. Entjes's observation (quoted in Bloemhoff 1979: 37, note 1) that (i), a variant of our ungrammatical Standard Frisian example in (25b) with - presumably - parallel M > H semanties, is widely accepted in the dialcets of Overijssel (south of Friesland) might he accommodated along similar lines, if it is assumed that in the structure of this example our FP is generated BELOW the finite modal zol ‘will'. Unfortunately, we have no further data to underpin this suggestion.
(i) (Overijssel dialects)
umdat hi'j dat zol (e)daon hebben (e)kund
because he that will done-PTC have could-PTC
27. In the linear string of the embedded elause, kinne will change places with soe by left-adjoining to soe ; this is verb raising in the head-initial era, as noted hefore. Notice further that the fact that ha and kinne in (28) are both infinitivaldoes NOT interfere with movement to F of the former across the latter. As was pointed out in note 20, we do not view the infinitival morpheme as the instantiation of a morphological feature [+infinitive].
28. Some care is needed here; we must somehow ensure that cluster formation does not obtain at LF either, or, alternatively, that "parasitic" participles must he licensed in overt syntax. (The latter of course recalls the famous adage "parasitic gaps are licensed at S-structure, not at LF." Here we might endorse Hornstein's (1995: 9lff.) argument against the LF application of head movement.
In our analysis, head-final ordering is a necessary condition, but not a sufficient one, for "parasitic" participle licensing; Standard German, for example, has head-final orderings without "parasitic" participles. lt is possible that we unjustly abstracted away from a very obvious difference between Frisian, on the one hand, and Dutch and German, on the other: the latter two languages mark their past participles not only with a suffix but also with a prefix (cf. Vanden Wyngaerd 1994, i.p.; and others).
A further caveat concerns Scandinavian double supines (Swedish [i]; cf. Lindqvist 1944; Ljunggren 1934). Whether the analysis of examples such as these should he assimilated to that of Frisian "parasitic" participle constructions is a moot point; Frisian has no grammatical double participial counterpart of the Swedish example in (i), which shows that the context in which Frisian may employ double participles is not identical with that in which Scandinavian exhibits double supines.
(i) (Swedish double-supine construction)
Jag har inte kunnat sovit
1 have not can-SUPINE sleep-SUPINE '
I have not heen able to sleep'
29. Our RHMC is a relativized minimality condition in two respects: it is relativized with regard to feature content and bar level. lt also does not state that the c-commanding head must necessarily be capable of checking the features of the moved head. In both respeets there is a difference between our proposal and that of Ferguson and Groat (1994), according to which "a category X hearing an unchecked feature beta may move to the Checking Domain of the CLOSEST VISIBLE CATEGORY Y capable of checking beta". Our approach allows VPs projected by participles to skip the auxiliary of the perfect

[Parasitic participles 1087]
on their way up to SpecFP - an empirically desirable result. This echoes Rizzi's (1990) insight that heads interfere with movement of heads, and maximal projections mingle with movement of other maximal projections. From the perspective of the "bare theory" of phrase structure (Chomsky 1995a, 1995b) this insight may be incorporated into a feature-relativized núnimality framework on the assumption that [+ minimal] and [± maximal] are features that the minimality condition is sensitive to.
30. The object, Petra, is moved out of the participle's projection prior to VP-movement to SpecFP, by something like object shift; see Boskovic (1994) for evidence for the existence of object shift in Serbo-Croatian. After VP-movement to SpecFP the participle might raise out of VP and adjoin to the left of the auxiliary (but this is irrelevant from our perspective).
31. The account suggested here is simpler than Boskovic's (1994). He suggests that (32b) involves left-adjunction of bili to ste followed by right-adjunction of ccekali to ste, while (32c) features right-adjunction of bili to ste followed by left-adjunction of ccekali to ste. This analysis is incompatible with Kayne (1994), whose antisymmetry theory blocks mixed left/right adjunction; Boskovic (note 18) in effect explicitly assumes the direction of adjunction to be free, contra Kayne.
32. Notice that there is no point in trying to invoke "rule ordering" to rule (13c) out as a violation of "attract the closest element." Nor does a statement to the effect that N/D-features and Y-features are different with regard to whether they attract an Xo category or a maximal projection seem very insightful: after all, the N/D-feature of AgrO seems unselective in that it can attract both maximal projections (object shift) and heads (cliticization), and this paper has argued extensively that VPs can be attracted to specifier positions to check what would appear to be a Y-feature of F.
33. Chomsky (1995b: 264) in effect exempts covert movement from the minimal-link condition (assuming that principles like the MLC only hold for whole categories), so that violations of the MLC under covert feature movement will never arise on his view.
34. The suggestions to follow are tentative. Another a priori conceivable approach to the problem, built on the hypothesis that the movement operations in question obligatorily "reconstruct" or (in copy-theory terms) obligatorily have their INITIAL copies retained and interpreted at LF, seems unfeasible. Hornstein (1995) argues, on the basis of his account of the cross-linguistic distribution of antecedent-contained deletion, that in a head chain the head of the chain never deletes. lf Hornstein's account of the ACD facts holds water, it cannot be the case that verb movement (or head movement in general) obligatorily reconstructs.

References
-Bloemhoff, Henk (1979). Heranalyse van een Stellingwerver oppervlaktestructuur. Us Wurk 28: Scripta Frisica. Tinkbondel foar Anne Spenter (special issue), H. Arhammar and T. Hoekema (eds.), 31-38.
-Bolkovié, keljko (1994). Participle movement, object shift, and second position cliticization in Serbo-Croatian. Unpublished manuscript, University of Connecticut at Storrs.
-Brennan, Virginia (1993). Root and epistemic modas auxiliary verhs. Unpublished dissertation, University of Massachusetts.
-Cavar, Damir; and Wilder, Chris (1992). Long Head Movement: Verb Movement and Clitization in Croatian. Sprachwissenschaft in Frankfurt Arbeitspapier 7. Frankfurt: University of Frankfurt.

[1088 M. den Dikken and E. Hoekstra
-Chomsky, Noam (1993). A minimalist program for linguistie theory. In The View from Building 20, K. Hale and S. J. Keyser (eds.), 1-52. Cambridge, MA: MIT Press.
-Chomsky, Noam (1995a). Bare phrase structure. In Government and Binding Theory and the Minimalist Program, G. Webelhuth (ed.), 385-439. Oxford: Blackwell.
-Chomsky, Noam (1995b). The Minimalist Program. Cambridge, MA: MIT Press.
-den Dikken, Marcel (1994a). Auxiliaries and participles. In NELS 24, 65-79. Amherst, MA: GLSA.
-den Dikken, Marcel (1994b). Minirnalist verb (projection) raising. Groninger Arbeiten zur germanistischen Linguistik 37, 71-88.
Ferguson, Scott; and Groat, Erich (1994). Defining "Shortest move." GLOW Newsletter 32, 24-25.
-Fujita, Koji (1994). Middle, ergative and passive in English - a minimalist perspective. MIT Working Papers in Linguistics 22, 71-90.
-Fujita, Koji (1996). Double objects, causatives, and derivational economy. Linguistic Inquiry 27, 146-173.
- Guéron, Jacqueline; and Hoekstra, Teun (1988). T-chains and the constituent structure of auxiliaries. In Constituent Structure: Papers from the 1987 GLOW Conference, Anna Cardinaletti, Guglielmo Cinque, and Giuliana Giusti (eds.). Dordrecht: Foris.
- Haegeman, Liliane (1995). IPP constructions and Y-movement in West Flenúsh. Unpublished manuscript, Université de Genéve.
- Hoekstra, Jarich (1990). Trijetiidwurdkonstruksjes. lt Beaken 52, 59-95.
- Hornstein, Norbert (1995). LogicalForm. From GB to Minimalism. Oxford: Blackwell.
- Kayne, Richard (1984). Connectedness and Binary Branching. Dordrecht: Foris.
- Kayne, Richard (1994). The Antisymmetry of Syntax. Cambridge, MA: MIT Press.
- Kitahara, Hisatsugu (1994). Target alfa: a unified theory of movement and structure building. Unpublished dissertation, Harvard University.
- Koster, Jan (1995). Extraposition as coordination. Unpublished manuscript, University of Groningen.
- Kratzer, Angelika (1977). What "must" and "can" must and can mean. Linguistics and Philosophy 1, 337-355.
- Lindqvist, A. (1944). Über einen Fall von syntaktischer Assimilation. Studia Neophilologica 16, 277-285.
- Ljunggren, R. (1934). Supinum och dubbelsupinum. Unpublished dissertation, Uppsala University.
- McDowell, J. (1997). Assertion and modality. Unpublished dissertation, University of Southern California.
- Rivero, Maria-Luisa (1991). Long head movement and negation: Serbo-Croatian vs. Slovak and Czech. Linguistic Review 8, 319-351.
- Rizzi, Luigi (1990). Relativized Minimality. Cambridge, MA: MlT Press.
- Roberts, Ian (1991). Excorporation and minimality. Linguistic lnquiry 22, 209-218.
- Stroop, Jan (1970). Systeem in gesproken werkwoordsgroepen. Taal en tongval 22, 128-147.
- Travis, Lisa (1984). Parameters and effeets of word order variation. Unpublished dissertation, MIT.
- Vanden Wyngaerd, Guido (1994). IPP and the structure of participles. In Minimalism and Kayne's Asymmetry Hypothesis, C. J. W. Zwart (ed.), 265-276. Groninger Arbeiten zur germanistischen Linguistik 37. Groningen: University of Groningen.
- Vanden Wyngaerd, Guido (1994) Participles and bare argument structure. In Minimal Ideas, W. Abraham, S. Epstein, H. Thráinsson, and C. J. W. Zwart (eds.). Amsterdam: Benjamins.
- Visser, Willem (1989). Ta de nominale ynfinityf yn it Frysk. In Philologia Frisica anno 1988, 51-74. Ljouwert: Fryske Akademy.

[Parasitic participles 1089]
- Watanabe, Akira (1993). Agr-based case theory and its interaction with the A-bar system. Unpublished dissertation, MIT.
- Wilder, Chris; and Cavar, Damir (1993). Word Order Variation, Verb Movement, and Economy Principles. Sprachwissenschaft in Frankfurt Arbeitspapier 10. Frankfurt: University of Frankfurt.
- Zwart, C. Jan-Wouter (1993). Dutch syntax. A minimalist approach. Unpublished dissertation, University of Groningen.
- Zwart, C. Jan-Wouter (1994). Dutch is head-initial. Linguistic Review 11, 377-46.
- Zwart, C. Jan-Wouter (1995). A note on verb clusters in the Stellingwerf dialect. In Linguistics in the Netherlands 1995, M. den Dikken and K. Hengeveld (eds.). Amsterdam: Benjamins.