Eric Hoekstra¹
 
 

1. Checking assumptions

I want to propose an analysis in which the typical Verb-Raising configuration crucially involves a checking dependency relation between two heads. Recall that in the transformational framework of the seventies Verb-Raising was analysed as an adjunction structure not involving any checking dependencies. I will continue to analyse Verb-Raising as an adjunction structure, but I will require that checking must take place inside the adjunction structure. Thus I follow Chomsky (1992:16) in assuming that head movement is to a position in which inflectional features are checked. I will limit myself to clusters not containing to-infinitives (on those, see Lattewitz 1994), in dialects of Frisian and Dutch (on German, see e.g. Haftka 1991 and the references cited there).
Movement to Spec position is similarly movement into a position in which inflectional features are checked. However, it is unclear whether the same applies to adjunction to a maximal projection. Presumably, Chomsky has adjunction to a maximal projection in mind when he writes (p.17) "Recall that the checking domain is heterogenuous: it is the ‘elsewhere' set". While this statement is not a contradiction of his earlier claim that the checking domain is typically (but not exclusively) involved in checking inflectional features, it is obviously desirable if we could strengthen Chomsky's view of the checking domain, and claim that the checking domain is exclusively involved in checking inflectional features. As a result, the notion checking domain would no longer be heterogenuous, in the sense of lumping together non-checked adjuncts, on the one hand, and checked Specifiers and heads, on the other hand.

(1)
Checking domain: all movement into the checking domain must check inflectional features

Thus, both head-movement and XP-movement are triggered by inflectional features. The issue of whether the relevant movements are formally substitutions or adjunctions is irrelevant, since adjunction and substitution yield isomorphic structures on a minimalist view of phrase structure.
Below I will first discuss pre-Kaynian approaches to verb-clustering. The approaches that have been proposed conflict with assumptions.defended by Kayne (1993). Specifically, I adopt the following assumptions based on Kayne (1993):

(2)
a. Kayne's LCA applies uniformly in the derivation
b. Specifier positions are adjunctions (to either Xo or Xmax)²
c. Morphological checking applies only in spec-head configurations

The question arises which feature gets checked in V-Raising clusters. I propose (2d):

d. A semantic feature for verbal complements gets checked

This proposal immediately leads to an empirically interesting result: since a head never has more than one complement, it will never check more than one verb. This makes it unnecessary, in the domain of V-clusters, to stipulate that adjunction takes place only once.³
The main difference between Kayne's proposal and Chomsky's concerns the treatment of linear asymmetries. Kayne's proposal embodies an attempt to isolate and explain asymmetrical generalisations. Chomsky's proposal, as it stands, predicts symmetry, cross-linguistically. Thus, regardless of how asymmetrical generalisations must be explained, the presence or absence of those generalisation gives us a clue about the problems which we have to think about. If such generalisations exist, we must think about linearity. If they don't exist, we can conclude that left-right ordering is arbitrarily parametrically fixed.
 

2. Un-asymmetrical approaches

Let us consider first two examples which used to be analysed as Verb-Raising ever since Evers (1975):

(3)
a. Dat ik je daar wel eens zou hebben willen zien blijven staan
that I you there MP MP would have want-INF see stay stand
1 2 3 4 5 6
b. Omdat ik dy dêr wol ris stean bliuwen sjen wollen hawwe soe
that I you there MP MP stand stay see want-PfP have would
6 5 4 3 2 1

(3a) is from Dutch and (3b) is from Frisian.⁴ At first, it seems that here we have a mirror effect, as expected under a Chomskyan theory of UG not incorporating linear asymmetries. By the standard analysis of Evers, both (1a) and (1b) involve an SOV base followed by adjunction on the right in Dutch, and (string-vacuous) adjunction on the left in Frisian.⁵ Evers motivates head-adjunction on the basis of various transparency effects.
An Evers type of analysis has been worked out for Flemish by Haegeman (1992, 1994), except that Haegeman also allows VP-Raising in order to capture the well-known difference between Dutch/Frisian and Flemish, which is that Flemish verbal clusters may be broken up by NPs and PPs (particles will be discussed later). Thus Haegeman has a head-final base for Flemish, rightward movement of VP, and optional scrambling of non-verbal material to the left. For Dutch, Haegeman has rightward movement of Vo.
Den Besten & Broekhuis (1990) have attempted to eliminate the use of head-movement in the derivation of verb clusters in Dutch.⁶ They adopt an head-final base, with VP-Raising to the right, and leftward scrambling of all non-verbal material. Thus it seems as if the Dutch verb cluster consists of heads, but actually it involves maximal projections out of which all non-verbal material has been scrambled away obligatorily. Thus the Broekhuis & Den Besten analysis leads to an analysis which generalises over Dutch and Flemish. Haegeman (1994) presents a number of asymmetries indicating that this might be a wrong generalisation, since raised VPs display different behaviour from raised Verbs, suggesting that the latter cannot be analysed as VPs out of which all non-verbal material has been scrambled away.
These analytic differences between Den Besten & Broekhuis, on the one hand, and Haegeman, on the other hand, need not concern us further, since conceptually these theories are identical. They allow for six possibilities:

(4) Allowed for under pre-Kaynian analyses
111: Base 222: Direction of the movement 333: Adjunction site of moved element 444: Surface order.
a. 111: Head-initial 222: leftward 333: on the left 444: 21
b. 111: Head-initial 222: leftward 333: on the right 444: 12 (string-vacuous)
c. 111: Head-initial 222: none 333: none 444: 12
d. 111: Head-final 222: rightward 333: on the right 444: 12
e. 111: Head-final 222: rightward 333: on the left 444: 21 (string-vacuous)
f. 111: Head-final 222: none 333: none 444: 21

(5) Allowed for under Kaynian analyses
a. 111: Head-initial 222: leftward 333: on the left 444: 21
b. 111: Head-initial 222: none 333: none 444: 12

Chomsky's minimalist program, like the pre-Kaynian analyses, allows for the six logical possibilities given above, since Chomsky's program does not incorporate the LCA. Please note that under Kayne's proposal the head-final order is also the order universally found in LF. The Kaynian type of minimalist parametrisation thus involves whether or not (5a) takes place in overt syntax; it is anyhow the LF structure.
The Kaynian program rules out four of the six logical possibilities allowed for by Chomsky's program.Chomsky's program, like the older analyses, does not exclude the two possibilities allowed for by the Kaynian program. This means that the two alternatives we have to choose from entertain a subset relation. Suppose now that only the options allowed for by Kayne's program are found. In that case, the evidence will be compatible with both alternatives: in that case, adherents to Chomsky's program will not find out that the evidence actually supports a narrower hypothesis (exaggerating somewhat). Put differently, I claim that Kayne's hypothesis is to be preferred on methodological grounds. It is to be preferred (not as a scientific truth but as a working hypothesis) because it makes stronger predictions than the alternative. This also means that Kayne's proposal is more vulnerable. Thus, if we falsify Kayne's proposal, this does not mean that Chomsky's proposal "was right after all".⁷
I will now go on to present some asymmetrical generalisations, and I will attempt to derive them from Kayne's proposal. The presence of these asymmetrical generalisations supports a framework incorporating asymmetry.⁸
 

3. The head-final cluster generalisation: an argument for asymmetry

The two approaches differ with respect to the treatment of head-final clusters. For the asymmetrical approach outlined here, a head-final cluster can only be analysed in terms of leftward movement to Spec positions in overt syntax. Thus, the cluster in (3b) above would have the derived structure (6b):

(6)
a. Omdat ik dy dêr wolris stean bliuwen sjen wollen hawwe soe
that I you there MP stand stay see want-PfP have would
6 5 4 3 2 1
b. [[[[[ stean bliuwen] sjen] wollen] hawwe] soe]
c. stean is Spec to bliuwen, stean bliuwen is Spec to sjen, stean bliuwen sjen is Spec to wollen, etc.

According to Den Besten & Broekhuis (6b) is a base structure. It is understandable why they adopt that position. They want to keep the difference between D-structure and S-structure as small as possible, so that D-structure will be learnable from S-structure. As pointed out by Culicover & Rochemont (1994), if D-structure can be head-final and head-initial and if rightward and leftward movement are both allowed, then word order is not learnable without additional constraints, cf. (4) above. Notice that this issue does not arise under the Kaynian version of the Minimalist Program: D-structure is trivially learnable, since it is universally head-initial, not parametrised as it is for Den Besten & Broekhuis. They fail to explain, however, why the D-structure may not be affected by movement processes like PP-extraposition. The point is that the head-final order is extremely rigid.
This is not a generalisation specific to Frisian. Consider Dutch first. Dutch has a limited amount of head-final orders in the verb-cluster. Specifically, a head-final cluster may always be found with exactly two verbs. As noted by Bennis (1992), a head-final V-cluster may not be broken up in Dutch by a particle, whereas a non-head-final cluster may:

(7)
a. Omdat hij mij op wou bellen
because he me up wanted call
b. Omdat hij mij wou op bellen
because he me wanted up call
(8)
a. Omdat hij mij op bellen wou
because he me up call wanted
b. * Omdat hij mij bellen op wou
because he me call up wanted

Frisian only has head-final clusters, and this cluster is absolutely impenetrable for non-verbal material ("MP" stands for "particle with modal force"):

(9)
a. Omdat ik dy him wolris op beljen heare wollen hawwe soe
because I you him MP up call talk hear wanted have would (54321)
b. * Omdat ik dy him wolris beljen op heare wollen hawwe soe
c. * Omdat ik dy him wolris beljen heare op wollen hawwe soe
d. * Omdat ik dy him wolris beljen heare wollen op hawwe soe
e. * Omdat ik dy him wolris praten heare wollen hawwe op soe

These facts are given with particles; they can be reproduced with any sort of non-verbal material (verbal arguments, adverbials, PPs, etc.).The same facts can be found in Old English. Old English exhibits both head-initial and head-final order. Koopman (1990), quoted in Den Besten, points out that a head-final cluster (or subcluster) is never split up by non-verbal material, as shown below, where the dots indicate the places where non-verbal material may appear:

(10)
a. ... genumen beon wolde 321
called be wanted
b. ... wolde ... genumen beon 1...32
c. ... wolde ... beon ... genumen 1...2...3

The generalisation is the same as what we saw in the case of Frisian and Dutch. It is given in (11) below as the Head Final Cluster Generalisation:⁹

(11) HFCG: a head-final verbal cluster cannot be broken up by non-verbal material.

The rigidity of head-final clusters is not accounted for by Den Besten & Broekhuis. It is a mystery for any approach not incorporating asymmetry, including Chomsky's version of the minimalist program. The asymmetry approach accounts for head-final rigidity in general by supposing that a Specifier-Head relation is spelled out in overt syntax. Thus, if genumen is in the Spec of beon in overt syntax, then it follows that nothing can intervene between Spec and Head, or else the configuration of Spec-Head agreement is not met. The same applies to the relation between genumen beon (Spec) and wolde (Head). Thus, head-final structures are rigid because they realise a Spec-Head relation in overt syntax, hence Spec and Head must occur adjacent in overt syntax, with the Spec preceding the head. In the (b)-example, genumen beon only moves into the Spec of wolde in LF. Thus there will be no adjacency between Spec and Head in overt syntax. Genumen has moved into the Spec of beon in overt syntax: thus they are adjacent. In the (c)-example, all movements take place in LF, and thus no adjacency between the elements of the verb-cluster is required. In exactly the same way, the rigidity of head-final clusters in Dutch and Frisian is accommodated. These facts support assumption (2a): the LCA applies uniformly in the derivation.
 

4. The IPP-effect and the circumfix generalisation
 

4.1. Analysis

The IPP-effect is illustrated below:

(12)
a. Omdat Jan dat zo heeft gewild / *willen
because Jan that so has want-PfP / * want-INF
b. Omdat Jan dat zo heeft willen / * gewild doen
because Jan that so has want-INF / * want-PfP do

If the participle takes an infinitive, the participle itself shows up in the form of an infinitive (hence: Infinitivum-Pro-Participio effect). I will refer to infinitives "replacing" a perfect participle as ersatzinfinitives.
There is some discussion as to which phenomenon correlates with IPP. It has been suggested that the IPP effect is systematically absent in head-final Germanic dialects like Frisian (Weijnen 1966:320 among others). However, while this is true, the IPP-effect is also absent in Scandinavian languages, in Romance languages, in Japanese, in Turkish, etc. Hence, it is not very likely that there is a 1:1 correlation between absence of IPP and head-finality, or presence of IPP and head-initiality.
Lange (1981), cf. also the discussion in Hoeksema (1980), proposes that the presence of the perfective circumfix ge---d is the factor which correlates with the IPP.¹⁰ This is probably the right generalisation. The (strict) head-final dialects, represented here by Frisian, do not have IPP, but they don't have a perfective circumfix either (they just have a suffix). Hence, what needs to be explained is not the absence of IPP in those dialects but the absence of the perfect circumfix.
I will refer to the generalisation observed by Lange a.o. as the circumfix-generalisation:

(13) The circumfix generalisation
If the perfective participle is realised by a circumfix, then the IPP effect will be observed, and vice versa.

It is especially the first part of the generalisation which concerns us here. Vanden Wyngaerd (1994) proposes an extremely interesting analysis of the circumfix generalisation, which I take over here with minor modifications. I need to make one assumption about the prefix, which is perfectly reasonable:

(14)
The prefix ge- occupies the Xo Spec position of the verb which it is attached to.

We will assume that this same position is also the landing site of the infinitival main verb. Thus (15a) is correctly ruled out:

(15)
a. * Omdat Jan dat zo heeft gewild doen
because Jan that so has want-PfP do
b. Omdat Jan dat zo heeft willen doen
because Jan that so has want-INF do

Ge- in gewild blocks attachment of doen in LF, explaining the first half of the paradigm. Instead, another item from the lexicon is chosen, which does have a Spec position available and is minimally different: the infinitive. Thus gewild en willen are both in the lexicon as perfect participles, with willen having a verbal feature in its Spec position which must be eliminated by Vo-movement. Gewild is thus expected to allow the full range of complements of its stem (apart from verbal ones), whereas willen as a perfect participle is limited to verbs.

(16)
a. Hij heeft dat gewild / * willen
he has that want-PfP / * want-INF
b. Hij heeft naar Rotterdam toe gewild / * willen
he has to Rotterdam to want-PfP / * want-INF

This explains the other half of the paradigm. Gewild checks for non-verbal complements, and willen for verbal ones. (14) provides a rationale for the first half of the circumfix generalisation. Strictly speaking, no prediction is made for when there is no prefix.
What is important is that we have uncovered another left-right asymmetry. The perfective suffix doesn't block verb-clustering, as is clear from a language like Frisian. It is only the perfective prefix (part of the circumfix) which causes problems. In order to solve this problem, we assumed, following Vanden Wyngaerd, that the prefix blocks left- attachment of the complement verb. Interestingly, the effect shows up in head-initial languages like Dutch (in the verb cluster in overt syntax), which only move leftward in LF. The fact that the IPP effect shows up even though cluster formation is covert provides the strongest possible evidence for leftward movement in LF universally. In this way the left-right asymmetry existing in LF becomes visible: only prefixes (and circumfixes), not suffixes cause IPP. Thus we have a second major left-right asymmetry, in addition to the Head-Final Cluster Generalisation. Both the Circumfix Generalisation and the Head-Final Cluster Generalisation can be treated in an insightful way, given the Kaynian version of the Minimalist Program.
 

4.2. Support for the circumfix generalisation

An interesting testcase is provided by the Friso-Saxon dialect called Stellingwerfs, spoken on the border of the provinces of Frisia, Overijsel and Drente. This dialect is mainly head-final, but especially older people allow mixed order governed by verbs of perception. The dialect does not have a prefix. The prefix generalisation predicts absence of IPP, regardless of the order of the verbs. The Prefix Generalisation turns out to be correct (Bloemhoff 1994, and p.c., cf. Zwart 1995):

(17)
a. Ik heb ‘m heurd zingen
I have him heard-PfP sing
b. Da'k ‘m heurd heb zingen
that I him heard-PfP have sing

While present-day speakers find this word-order marginal, replacement of the participle with an infinitive is for all speakers considerably worse. Thus clusters with mixed head-final, head-initial order conform to the circumfix generalisation.
The case of head-final constructions and a prefix is provided by the Zaan dialect, spoken to the north of Amsterdam, along the river Zaan, in the province of North Holland. This dialect is mainly head-final in the verb cluster, and has a perfect prefix in the form of e-. A search was conducted through literature written in Zaans, and the dialect turned out to always exhibit the IPP-effect in the familiar configuration, as expected (Hoekstra 1994). Again, linear order of the verbs seemed irrelevant. What counts is the presence or absence of a prefix (actually a circumfix).
Nevertheless, IPP-sentences in Zaans exhibit a deviant word order: IPP sentences are neither purely head-initial nor head-final, a fact which needs to be explained.
 

4.3. Why IPP-sentences in the Zaan dialect exhibit deviant word order

I also found that IPP-sentences exhibited a deviation from pure head-final order which was otherwise quite rare. Thus, while a normal verb cluster is generally head-final (321), IPP-sentences systematically exhibit the order 312 or 132. Note that this only shows up in embedded clauses (if we limit ourselves to clusters of three verbs): after Verb-Second, the remaining cluster seems (superficially) to be purely head-final. Examples are given below (from Boekenoogen 1897, abbreviated as "B", and Woudt 1984, abbreviated as "W"):

(18)
a. As ie nog langer in die koledamp hadde zitte moeten zou je bezeeuwd hebbe
if he yet longer in that coalsmoke had sit-INF must would you fainted have / 1.3.2. / B72, PfP = emoete
b. Me vrouw heb welderes ezeid dat ik domenie had worre moete
my wife has MP said that I vicar had become-INF must 1.3.2. / W143
c. Behalleve den die in derloi automobiele hadde blaive moete
except then those that in their cars had stay-INF must 1.3.2. / W150
d. Meskien begraipe jollie nou dat ik puur wet knope heb deurhakke moete
maybe understand you MP that I quite some knots had cut must-INF 1.3.2. / W177
(19)
a. Ik zel et je wel zien hebbe late
I shall it you see have let 4.2.3. / B72
b. Nou wul ik helegaar niet zegge dat me de veroitgang teugehouwe hadde moete
now want I MP not say that we the progress stopped had must-INF 3.1.2. / W97
(20)
a. The IPP-examples involve 1.3.2. or 3.1.2
b. There is no IPP-example with order 3.2.1.
c. Non-IPP examples (so without have) are generally 3.2.1
d. Generalisation: HAVE (or BE) must precede the IPP-infinitive

Perfect participles move to Spec in overt syntax (they occur to the left of have) whereas ersatz-infinitives move to Spec in LF, occurring to the right of have in overt syntax. Why should this be so?
I will now show that the deviant word order of IPP-sentences in Zaan is due to language contact between the Zaan dialect and Standard Dutch.
The first step towards solving this puzzle is to note that the perfective prefix is still very young in Zaans. Boekenoogen (1897:LXX) presents four arguments in support of this claim. First, children's songs fail to feature the prefix. Second, older Zaans writers regularly fail to write the prefix. Third, older grammarians writing on Zaans perfect participles present forms without prefix. Fourth, archaic adjectives derived from participles don't have the prefix, e.g. grimmeld "spotted", hot "turned sour", etc. Thus it may reasonably be concluded that the prefix -e is a young development.
Our theory of IPP entails a close relation between the presence of the perfective prefix and the IPP-effect. Therefore, if we accept Boekenoogen's conclusion that the prefix is a recent development in Zaans then we must also conclude that the presence of the IPP-effect is a recent development in Zaans. This conclusion is reinforced by a dialect-geographical argument. Change generally obeys two dialectgeographical conditions. First, a change from A to B takes place on the border of an A-area and a B-area. If we move to the north to the West-Frisian dialect, which is otherwise closely related to Zaans, the IPP effect is virtually absent. And if we move further northward, we encounter the head-final dialects without IPP whatsoever of the provinces of Frisia and Groningen. If we move southward from the Zaan, all dialects exhibit IPP and head-final verb clusters are rare. Thus, Zaans was a dialect which was already on the border of an IPP area and a non-IPP area, fulfilling one of the conditions for change.
A second dialectgeographical condition is that change is geographically coherent, under the influence of a source. There is a very strong source of change in the form of the City of Amsterdam (an IPP area), from which the industrialisation of the Zaan took place in the 18th - 19 th centuries. Thus the two main dialectgeographical conditions for change were fulfilled: it affects the border area, under the influence of a strong source.
Note now, that speakers of Zaans had at least passive knowledge of Standard Dutch and Amsterdam Dutch (which coincide in having IPP). Thus they were familiar with two linguistic systems.Language contact with Standard Dutch led to knowledge of IPP, which in Standard Dutch always cooccurs with head-initial order. Clearly, the Zaan speakers did not acquire the IPP effect perfectly: there are word order differences between Standard Dutch and Zaans. I will now explain why imperfect acquisition of the IPP took the specific form it did.
I assumed that originally Zaans had head-final order and no IPP, like northern dialect generally. In order to acquire IPP perfectly, Zaan speakers had to learn the following three things:

(21) Step one. Auxiliaries
A. Acquiring the ersatzinfinitive: associating the infinitive with perfect participle semantics and with checking of the verbal complement's head.
B. Postulating a prefix for the perfect participle. ¹²

The morpho-lexical form of ersatzinfinitive and infinitive is the same, and checking Vo features involves a specification which infinitival auxiliaries already had and which they shared with perfect participles of auxiliaries. The postulation of a prefix for perfect participles of auxiliaries effectively makes IPP obligatory, and makes it impossible for perfect participles to check the infinitives they govern any longer. Note that step A is just a composite of ingredients which were already present in Zaans. Step one affects auxiliaries only, and these form a closed class.
The second step is to adopt the Standard Dutch word order effect associated with the ersatzinfinitives of auxiliaries:

(22) Step two: ersatzinfinitives are not checked in overt syntax

As a result, the ersatzinfinitive is found in the same position in which it occurs in Standard Dutch. Step two does not involve parameter resetting since ersatzinfinitives did not exist in older Zaans, and thus no parameter had to be reset. Step two involves rather parameter specifying for the newly created class of ersatzinfinitives. However, Zaan speakers did not take the third step:

(23) Step three: main verbs are not checked in overt syntax

Main verbs occur in head-final positions, even in IPP constructions. Thus we derive a system in which normal clusters are head-final whereas IPP clusters are partly head-initial (specifically, the ersatzinfinitive is not moved in overt syntax). Notice that step three involves resetting the parameter for all verbs (an open class) in the dialect. Pretheoretically, we can also imagine that step 2 was not taken whereas step 3 was. This would derive an IPP dialect in which the ersatzinfinitive precedes its governor, and in which the main verb follows the ersatzinfinitive and its governor.
Such a change did not take place, and there might be two reasons for this. A change involving a closed class of items (auxiliaries) can more easily take place than a change involving an open class of items (main verbs). More interestingly, I believe, step 3 would involve parameter resetting whereas step 2 just involved parameter adding. This is probably the crucial factor. After all, if mere bulk was a decisive factor, one would for example expect that main verbs which were frequent would have been subject to parameter resetting, but infrequent verbs would not. This is not the case.
Since the Zaan speakers intended to speak Zaans when they did, change had to operate below the level of consciousness. Thus, when the participial prefix is spelled-out, it does not have the Standard Dutch form ge- but it has the form of the most frequently occurring affix in Zaans, that is, the schwa.
Assuming a scenario like the preceding to be by and large correct, we draw the following conclusion. The fact that IPP sentences exhibit deviant word order in Zaans may plausibly be ascribed to the fact that IPP is easily learned in a dialect contact situation.¹³ Why should this be so? We have suggested that the reason for IPP's learnability is that it involves parameter adding rather than parameter resetting.¹⁴
 

4.4. The IPP-effect in West-Frisian

As noted by Willem Taanman in unpublished work, West-Frisian exhibits a highly limited IPP-effect. Examples from dialect literature are provided below:¹⁵

(24) IPP-forms for zitte "sit" and loupe "walk"
a. Ik hew loup-E te dromen (PfP=loupen)
I have walk-INF to dream
b. We hewwe zitt-E te klessen (PfP=zeten)
we have sit-INF to chat
(25) No IPP (with all other verbs)
a. Toe ben ik ok maar zitten bleven
then am I MP MP sit stay-PfP
b. Bleven is unambiguously PfP (INF= bloive)

West-Frisian seems to be a counterexample to the prefix generalisation, since it exhibits (a very limited amount of) IPP, and yet it does not have a perfect prefix. The West-Frisian data exhibit IPP only with a small set of aspectual verbs selecting to-infinitives. Note that in all examples of West-Frisian IPP the word orders of Dutch and West-Frisian coincide. West-Frisian differs even here from Standard Dutch, though, in featuring the infinitival marker te, like the dialects of Frisia and Groningen.
Now we could elaborate on the fact that the Kaynian approach only predicts one half of the circumfix generalisation, namely: if prefix then IPP. What we have here is a problem for the second half of the generalisation: no prefix, no IPP. Since the Kaynian approach does not support the second half of the generalisation we could simply conclude that the second half of the circumfix generalisation is wrong. Such a move would, perhaps unjustly, abstract away from the fact, pointed out in Hoekstra & Taanman (1995), that West-Frisian speakers are never monolingual speakers of West-Frisian. They always speak Standard Dutch as a first or second language. The question now is: does the circumfix generalisation apply to each subsystem of a bilingual speaker separately, or does it apply to the union of West-Frisian and Standard Dutch? Since the two subsystems interact, it is natural to suppose that the generalisation applies to the union of the two subsystems. This basically means that IPP can be present or absent: this is what we see. In some semantic contexts it is present, in others it is absent. The limited range of IPP-effects in West-Frisian betray the influencence of Standard Dutch on this dialect.¹⁶
 

5. Conclusion

Two linear asymmetries exist in the dialects of Dutch and Frisian. The head final cluster generalisation says that a head-final cluster is absolutely impenetrable for non-verbal material. The prefix generalisation says that IPP is only found in those dialects in which the perfect participle is marked with a prefix. Pre-Kaynian analyses were shown not be able to come up with an explanation of these generalisations. Both asymmetries provide direct evidence for a theory incorporating left-right asymmetries in the grammar, such as Kayne's (1993), according to whom Specifiers are on the left, complements on the right. The head-final verb cluster is built by head-movement to Xo Spec positions. This process takes place bottom up. If we vary the position of spell-out, we derive word order variations which are actually attested in the West-Germanic coastal dialects.
 

Notes

1. Large parts of this paper were presented at the Tilburg Conference on Rightward Movement under the title "Verb Raising and the Kaynian Program". I would like to thank Marcel den Dikken and Jan-Wouter Zwart for minimalist discussion, Jarich Hoekstra for sharing with me his immense knowledge of Frisian syntax, the audience at the Tilburg Conference on Rightward Movement for stimulating questions and discussion, and Riny Huybregts for extensive comments leading to a much shorter and more adequately focused paper.
2. Naturally, a Spec can be either a head or a maximal projection. The stipulation that a Spec must be a maximal projection is a relict of the late seventies. The arguments for that stipulation were valid at that time but now not anymore (Hoekstra 1991).
3. Recursive checking is possible only if the recurring element bears a non-semantic feature, following Den Dikken & Hoekstra (1995).
4. The dice are loaded: in other cases, such mirror image sentence are hard to find since Frisian auxiliaries are much more restricted semantically than their Dutch counterparts, see Dyk & Hoekstra (1987), Hoekstra & Tiersma (1994).
5. Ever's head-final examples are not from Frisian but from German.
6. Also Coppen & Klein (1992), except that Coppen & Klein have a head-initial base (for which no universality is claimed, though), superficially foreshadowing a minimalist analysis. Coppen & Klein criticise Den Besten & Broekhuis (1990) for postulating a 1:1 relation between V-Raising and IPP, seeing that VPR in West Flemish may also cooccur with IPP. This is correct, but Coppen & Klein's suggestion that IPP must be considered a peripheral lexically determined phenomenon can hardly be called illuminating.
7. The same fallacy could be observed in the connectedness/barriers debate. Connectedness was a much narrower hypothesis than the relatively unconstrained barriers framework. Hence it was easier to falsify connectedness.
8. I use "asymmetry" as a shorthand for "a mechanism predicting linear asymmetries such as proposed by Kayne."
9. Truckenbrodt (1994) refers to this generalisation as Haider's Generalisation.
10. Actually Lange's account predicts that a circumfix will always cause IPP regardless of word order, a prefix will cause IPP only in head-final clusters (* V3 prefix-V2), and a suffix will only cause IPP in head-initial clusters (* V2-suffix V3). Interestingly, there does not seem to be, as far as I know, a dialect in which the perfect participle bears a prefix, without also bearing a suffix. Vanden Wyngaerd's account relates IPP exclusively to the presence of a prefix. Thus, Vanden Wyngaerd predicts that "V2-suffix V3" should be grammatical. This prediction can perhaps be tested in the southern dialects of Niederdeutsch. Why is there so little linguistic literature on Niederdeutsch, and on dialects in general?
11. These modifications involve one point that needs to be mentioned. Vanden Wyngaerd's account crucially relies on the assumption that the infinitival ending and the perfect participial prefix have the same functional category, and that therefore an infinitive (category F) may not adjoin to the prefix (category F).
12. This prefix was spelled out as -e, the unmarked inflectional affix in Zaans. Borrowing the suffix ge- from Dutch was not an option, since people were conscious of the fact that this suffix was not Zaans. That is, change operated below the threshhold of consciousness on people who intended to speak Zaans. If they intended to speak Dutch, they would use the prefix ge-, but while speaking Dutch they would not (necessarily) adopt the required word order since this would be below the threshhold of consciousness again. There are well-known cases of (!) linguists reporting as Standard Dutch word order effects which are ungrammatical in Standard Dutch but not in those linguists' dialects.
13. IPP-sentences may also exhibit deviant word order in West Flemish (consider the example of Den Dikken 1994:83) and in Afrikaans (Den Besten 1989:161-162).
14. This whole section may be looked upon as a tentative first step in bringing together insights from language contact (Van Coetsem 1988) and insights from theoretical syntax.
15. My data are from Butter (1944), a very useful source. Corroborating observations come from unpublished work by Willem Taanman and myself.
16. It may be doubted, though, whether the development of IPP in West-Frisian will be like in Zaans. The reason is that the youngest generation(s) of speakers intend to speak Standard Dutch, not West-Frisian: thus change also takes place above the threshhold of consciousness. Furthermore, improved education and the modern communication media (neither of which was accessible to the Zaan speakers a century ago) guarantee a practically perfect acquisition of Standard Dutch.
 

References

Bennis, H. 1992. "Long Head Movement: the position of particles in the verbal cluster in Dutch". In R. Bok-Bennema & R. van Hout (eds), Linguistics in the Netherlands 1992. Amsterdam: John Benjamins, 37-47.
Besten, H. den 1989. Studies in West Germanic Syntax. Diss, University of Amsterdam. Amsterdam: Rodopi.
Besten, H. den, & H. Broekhuis 1990. "Woordvolgorde in de werkwoordelijke eindreeks". GLOT 12: 79-137.
Bloemhoff, H. 1994. Stellingwarfs Woordeboek F-K. Oosterwoolde: Stichting Stellingwarver Schrieversronte.
Boekenoogen, G.J. 1897. De Zaanse Volkstaal. Leiden: A.W. Sijthoff.
Butter, F. 1944. Trynke Boôd. De roman van een boeredochter. Hoorn: "West-Friesland".
Chomsky, N. 1992. "A Minimalist Program for Linguistic Theory". MIT Occasional Papers in Linguistics. Cambridge: MIT.
Coetsem, F. van 1988. Loan Phonology and the Two Transfer Types in Language Contact. Dordrecht: Foris
Coppen, A. & Klein, M. 1992. "Het einde van V-Raising". In M. Klein (ed.), Nieuwe Eskapades in de Neerlandistiek. Een bundel opstellen voor M.C. van den Toorn. Groningen: Wolters-Noordhoff, 32-46.
Culicover, P. & Rochemont, M. 1994. "Constraints on rightward and leftward movement". Talk presented at the Workshop on Rightward Movement. Tilburg: Tilburg University.
Dikken, M. den 1994. "Minimalist Verb (Projection) Raising". In C.J.-W. Zwart (ed.), "Minimalism and Kayne's Asymmetry Hypothesis." Groninger Arbeiten zur germanistischen Linguistik 37: 71-88.
Dikken, M. den & Hoekstra E. 1995. "Parasitic Participles". Ms. Amsterdam: HIL, Free University / Amsterdam: P.J. Meertens Institute, Royal Academy.
Dyk, S. & Hoekstra, J. 1987. "Oersjoch fan ‘e Stúdzje fan ‘e Fryske Syntaksis". In S. Dyk. and J. Hoekstra (eds), Ta de Fryske Syntaksis. Ljouwert: Fryske Akademy.
Evers, A. 1975. The Transformational Cycle in Dutch and German. Dissertation, Utrecht: University of Utrecht.
Haegeman, L. 1992. Theory and description in generative syntax. A case study in West Flemish.Cambridge: Cambridge University Press.
Haegeman, L. 1994. "Verb Raising as Verb Projection Raising: Some Empirical Problems". Linguistic Inquiry 25: 509-522.
Haftka, B. 1992. "Wann man angeblich soll dan finite Verb voranstellen müssen. Skizze zur Position infiniter Verbprojektionen ohne zu". In A. Steube (ed.), Zur Satzartigkeit von Infinitiven und Small Clauses. (=Linguistische Arbeiten 135). Tübingen: Niemeyer, 137-153.
Hoeksema, J. 1980. "Verbale verstrengeling ontstrengeld". Spektator 10: 221-249.
Hoekstra, E. 1991. "X'-theory and Licensing Mechanisms". The Linguistic Review 8: 47-73.
Hoekstra, E. 1994. "Woordvolgorde en het Infinitivus-pro-Participium Effect in het Zaans". Taal & Tongval 46: 132-141.
Hoekstra, E. & Taanman, W. 1995. "Een Westfriese gradatie van het Infinitivus-pro-Participium effect". Ms. Amsterdam: P.J. Meertens Instituut, Royal Academy.
Hoekstra, J. & Tiersma, P. 1994. Frisian. In J. van der Auwera & E. König eds. The Germanic Languages. London: Routledge.
Kayne, R. 1993. The Antisymmetry of Syntax. Ms., CUNY.
Koopman, W. 1990. Word Order in Old English. With special eference to the verb phrase. Dissertation, University of Amsterdam.
Lange, K.-P. 1981. "Warum Ersatsinfinitiv?" Groninger Arbeiten zur Germanistischen Linguistik 19: 62-81.
Lattewitz, K. 1994. "A Minimalist View on Verb Raising". Talk presented at the 9th Workshop on Comparative Germanic Syntax. Boston: Harvard University.
Taanman, W. 1994. "Enige opmerkingen m.b.t. een Westfries dialect." Ms.
Truckenbrodt, H. 1994. "Towards a Prosodic Theory of Extraposition". Talk presented at the Workshop on Rightward Movement. Tilburg: Tilburg University.
Weijnen, A. 1966. Nederlandse Dialectkunde. Assen: Van Gorcum.
Woudt, K. 1984. Deer hoor ik je. Gedachten over de Zaanse streektaal. Stichting Wormerveer: Uitgeverij Noord-Holland.
Wyngaerd, G. vanden 1994. "IPP and the Structure of Participles". In C.J.-W. Zwart (ed.), "Minimalism and Kayne's Asymmetry Hypothesis." Groninger Arbeiten zur germanistischen Linguistik 37: 265-276.
Zwart, C.J.-W. 1995. "A note on verb clusters in the Stellingwerf dialect". In M. den Dikken and K. Hengeveld (eds), Linguistics in the Netherlands 1995. Amsterdam: Benjamins, 215-226.