Professional Documents
Culture Documents
Dutch has a type of complement clause that is transparent for extraction. The
construction does not have A’-like characteristics and is therefore better
analysed as instantiating A-movement. This is at odds with the intuition that
cross-clausal A-movement is last resort. The facts are accounted for if the
complement clause is not a phase in the sense of Chomsky (2001).
1. Introduction
Dutch has three ways of forming an infinitival complement clause. The first
way is traditionally analysed as involving rightward movement of the
embedded verb of a preverbal complement clause (Evers 1975); hence the term
Verb-Raising construction (1a). The second construction has been analysed as
involving rightward extraposition of the preverbal complement clause; hence
the term extraposition construction (1b). The third way is dubbed the third
construction in Den Besten et al. (1988) and differs from the extraposition
construction in (1b) in that the internal argument of the complement clause has
moved into the matrix clause (1c):
*
In Martin Salzmann and Luis Vicente (eds.) Leiden Papers in Linguistics 2.3 (2005), (1-16)
http:/www/ulcl.leidenuniv.nl ISSN 1574-4728
The choice of the type of complement is largely determined by the matrix verb.
Verb-Raising is found with a restricted set of verbs including perception verbs,
modals, perception verbs, aspectual verbs and raising verbs (Evers 1975;
Rutten 1991). 1 This construction will not be discussed in this paper. When the
Verb-Raising construction is not possible, the extraposition construction or the
third construction is used. Many verbs allow both construction types, but there
are no verbs which allow the third construction but not the extraposition
construction.2 Finally, a small set of verbs allows all three complement types.
In those cases, the selecting verb can be shown to have different properties
depending on the type of complement it selects.
In this paper, I characterise the third construction. I demonstrate that the
movement of the embedded internal argument in (1c) does not show A’-like
properties, and is therefore more likely to be analysed as involving A-
movement. This result is at odds with the observation that cross-clausal A-
movement is generally impossible unless the clause in which the argument
originates is in some sense deficient, as recently put forth in Wurmbrand
(2001). However, the transparency of Dutch complement clauses might be
taken as support for a phase-based derivation, as proposed in Chomsky (2001).
The crucial difference between (1b) and (1c) will be shown to be the presence
of a CP-layer. Being a phase, CP triggers Spell Out of its domain, which
contains the embedded internal argument. Cross-clausal movement is therefore
impossible. Third construction complements are TPs, hence not phases, and as
a consequence, material contained in the complement clause is accessible to the
matrix vP.
2. Previous analyses
Since Den Besten et al. (1988) it is held that the third construction (2a) is
qualitatively different from the Verb-Raising construction (2b), even though it
may seem to produce the same word order:
1
I use the terms “Verb Raising” and “Extraposition” as descriptive labels. “Verb Raising” in
infinitival clauses should not be confused with the verb movement that takes place in finite main
clauses. The extraposition construction in this paper denotes a clausal complement that surfaces to
the right of the verb it depends on. Following Zwart (1993), I assume that complement clauses are
actually base generated in this position.
2
It is not clear what determines whether a verb can select for a third construction complement.
Den Besten et al. (1988) notice that non-bridge verbs, particle verbs and reflexive verbs generally
do not allow the third construction, but there are exceptions to this generalisation. As far as I can
tell, the exceptions regarding reflexive verbs are those verbs which are not factive. Factive verbs
are incompatible with the third construction. I do not have an account for the fact that particle verbs
are incompatible with the third construction.
Transparency in Dutch TP complements 3
(2) a. dat zij de vergadering heeft besloten te verzetten
that she the meeting has decided to reschedule
‘That she decided to reschedule the meeting.’
b. dat zij de vergadering heeft moeten verzetten
that she the meeting has must reschedule
‘That she has had to reschedule the meeting.’
The examples in (2a) and (2b) differ in several respects. First, Verb-Raising
(3a), but not the third construction (3b), is characterised by the IPP-effect, in
which the matrix verb takes on the infinitival form when selected by the perfect
tense auxiliary, instead of the participial form:
Second, the perfect tense auxiliary and the matrix verb are in a fixed order in
Verb-Raising constructions (4), but not in the third construction (5):
Third, an argument of the embedded clause may not intervene between the
matrix verb and the embedded verb in the Verb-Raising construction in
standard Dutch, but this is possible in the third construction:
4 Janneke ter Beek
(6) a. dat ze het boek (aan Piet) moest (*aan Piet) geven
that she the book to Piet had-to to Piet give
‘That she had to give the book to Piet.’
b. dat ze het boek (aan Piet) besloot (?aan Piet) te geven
that she the book to Piet decided to Piet to give
‘That she decided to give the book to Piet.’
(7) a. dat Jan heeft *beloven/beloofd om het boek aan Piet te geven
that Jan has promise/ promised for the book to Piet to give
‘That Jan has promised to give the book to Piet.’
b. dat Jan beloofd heeft om het boek aan Piet te geven
that Jan promised has for the book to Piet to give
‘That Jan has promised to give the book to Piet.’
For these reasons, Den Besten et al. (1988) suggest that the extraposition
construction and the third construction are related, while the Verb-Raising
construction represents a different type of complement. Specifically, they
propose that the Verb-Raising construction involves a VP-complement, the
verb of which undergoes rightward movement. The other two constructions
involve sentential complements. The extraposition construction allows a
complementiser, which suggests that the complement in these constructions is a
CP. In the third construction, however, a complementiser is impossible,
suggesting that these structures involve IP-complements, or TPs, in more recent
terminology. According to Den Besten et al. (1988), the internal argument of
the embedded clause undergoes A’-movement into the matrix clause:
The structures in (8) have gained general acceptance among linguists working
on the topic.3 However, the trigger for movement of the object has not received
3
See among others, Den Besten & Rutten (1989), Rutten (1991), Broekhuis (1992), Broekhuis
et al. (1995), and more recently, Wurmbrand (2001). IJbema (2002) deviates from the
Transparency in Dutch TP complements 5
4
the attention I think it deserves. In part, this is due to historical factors. In the
framework that these authors worked in, the Dutch internal argument was
assumed to be base generated to the left of the verb, and accusative case was
assumed to be assigned in-situ; therefore the possibility that the movement in
(8a) might be A-movement seemed unlikely. Furthermore, movement into the
matrix clause is not obligatory, as examples like (7b) are also acceptable when
the complementiser is dropped. Moreover, movement is not restricted to
elements that need case, as can be seen in (6b), in which a prepositional phrase
moves into the matrix clause. Hence the movement of the object in the third
construction was straightforwardly identified as A’-movement.
However, if the Dutch verb phrase is analysed as in Zwart (1993), with the
preverbal position of the object resulting from movement from the postverbal
base position into the position that licences the internal argument as the direct
object, then the possibility arises that the third construction may actually
instantiate A-movement.5 In the next section I demonstrate that this is indeed
the correct scenario.
As a starting point, let us assume that the embedded clause is not deficient in
licensing its arguments, that is, that the A’-movement analysis of Den Besten et
al. (1988) is correct. The internal argument is licensed in the embedded clause,
after which it moves into the matrix clause. What could trigger such
movement? If licensing is not the trigger, we might assume that this movement
has interpretational effects. For instance, we might expect that an indefinite
argument that undergoes movement acquires a strong, specific reading, as in
simple sentences (cf. De Hoop 1992).6 This does not seem to be the case,
however:
aforementioned authors in that she judges examples like (8b) as marginal, but she reports that
judgements are subject to speaker variation.
4
Wurmbrand (2001) is exceptional in making a principled distinction between A-movement
and A’-movement in such constructions. On the basis of the semantic properties of the matrix verb
she predicts that the Dutch third construction involves A’-movement rather than A-movement. We
will see that the prediction is not borne out.
5
Zwart (1993) proposes that the embedded clause in the third construction does not have an
AgrOP to check the accusative case feature, but he does not take a stand on the categorial status of
the complement clause.
6
I follow the observations made in De Hoop (1992). However, I do not assume that the object
in (9a) comes to precede the adverb through movement out of its licensing position, cf. Zwart
(1993).
6 Janneke ter Beek
(9) a. dat Jan een beroemde acteur vaak wil ontmoeten
that Jan a famous actor often wants meet
‘That Jan wants to meet a certain famous actor often.’
b. dat Jan een beroemde acteur beweert te kennen
that Jan a famous actor claims to know
‘That Jan claims to know a famous actor.’
Whereas the indefinite object een beroemde acteur ‘a famous actor’ must be
interpreted as specific in (9a), this interpretation is not forced in (9b).7
Furthermore, the third construction does not seem to yield distinctive
informational properties. Examples like (10a) are appropriate in out of the blue
contexts, which suggests that they represent a neutral information structure (cf.
Wöllstein-Leisten 2001 for German). Note that scrambling in simple sentences
like (10b) does affect the informational properties of the sentence:
Thirdly, quite a few idiomatic expressions are quite natural in the third
construction, which is unexpected if the third construction yields
interpretational effects. In contrast, scrambling of part of an idiomatic
expression in simple sentences results in loss of the idiomatic interpretation:8
9
German not only has contexts for movement to object position, but also for movement into
subject position. It is generally assumed that Dutch does not have this option (Rutten 1991;
Broekhuis 1992, among others). I have to assume that these structures are excluded for independent
reasons, if they are indeed impossible, but the issue requires careful investigation.
8 Janneke ter Beek
In (13a), the main clause and the embedded clause can both be modified with a
temporal adverb.10 In the third construction, this is not possible; the adverb that
modifies the embedded clause in (13a) is illicit in (13b). However, if one of the
adverbs in (13b) is dropped, the sentence becomes grammatical:
10
One might object that the embedded clause in the third construction is degraded in the
presence of any adverb. Some speakers perceive a contrast between examples like (13b) and
examples in which the embedded clause contains a manner adverb, but the majority seem to find
both types of adverbs equally unacceptable. Wöllstein-Leisten (2001) provides grammatical
German examples of the third construction in which the embedded clause contains a (low) adverb,
which suggests that Wurmbrand’s (2001) conclusion that the ungrammaticality of examples like
(13b) in German is due to temporal modification is valid.
Transparency in Dutch TP complements 9
when the adverb is in the main clause. This suggests that the ungrammaticality
of (13b) may be due to processing difficulties.11
Another way to investigate whether the event described in the main clause
and the event in the embedded clause are independently situated in time is to
study the use of temporal auxiliaries. If it is possible to construe a main clause
in the present tense with an embedded clause that is set in the past, as indicated
by the use of the temporal auxiliary ‘hebben’, we have evidence that the two
clauses are independently situated in time. Indeed, the third construction is
compatible with a perfect tense auxiliary:
?
(15) dat Jan zijn huiswerk gisteren beweert te hebben gemaakt
that Jan his homework yesterday claims to have made
‘That Jan claims to have done his homework yesterday.’
(16) a. Iki heb de studentj een goed cijfer beloofd [zonder PROi/*j
I have the student a good grade promised without
mei/ *zichj ervoor te schamen]
myself/ himself for-it to shame
‘I promised the student a good grade without being ashamed of it.’
b. Iki heb de studentj dit artikel aangeraden [zonder PROi/*j
I have the student this article recommended without
mei/ *zichj ervoor te schamen]
myself/ himself for-it to shame
‘I recommended the student this article without being ashamed of it.’
The adjunct clause contains an anaphor which is bound by PRO. If PRO can
only be controlled by the matrix subject, it follows that the anaphor cannot be
coreferent with the matrix indirect object in (16). Coreference between the
matrix indirect object and PRO in the adjunct clause is possible, however,
when it is mediated by a second PRO. If a complement clause is added to a
sentence like (16), then PRO in the adjunct clause can be controlled by PRO in
the complement clause. If the matrix indirect object controls PRO in the
complement clause, coreference between the matrix indirect object and PRO in
the adjunct is possible:
11
I do not have an explanation for the presence of the adverb in the matrix clause. For the sake
of simplicity, I assume without argumentation that adjuncts can be base generated in the matrix
clause in the third construction.
10 Janneke ter Beek
(17) dat iki de studentj aanraadde [PRO*i/j het artikel te lezen] [zonder
that I the student recommended the article to read without
PROi/j mei/ zichj ervoor te schamen]
myself/ himself for-it to shame
‘That I recommended to the student to read the article without being
ashamed of it.’
(18) dat iki de studentj het artikel aanraadde [PRO*i/j te lezen] [zonder
that I the student the article recommended to read without
PROi/j mei/ zichj ervoor te schamen]
myself/himself for-it to shame
‘That I recommended to the student to read the article without being
ashamed of it.’
In (18), the matrix indirect object can be coreferent with PRO in the adjunct,
which can be taken as evidence for the presence of PRO in the complement
clause.
One might object that the constrast between (16) on the one hand, and (17)
and (18) on the other can be explained by a difference in the c-command
relations between PRO and its controller, in which case the coreference
possibilities do not tell us anything about the presence of PRO in the
complement clause. The adjunct in (16) has to be attached to the matrix clause.
I follow Chomsky (1995) in assuming that it is attached to VP, but one might
argue that the adjunct is actually attached somewhere lower than the subject,
but higher than the indirect object. In this analysis the indirect object is not a
possible controller because it does not c-command PRO.
In (17) and (18), however, there are two possible attachment sites for the
adjunct clause. If the adjunct is attached in the matrix clause, again between the
position the subject and the indirect object, PRO in the adjunct can only be
controlled by the matrix subject, as this is the only DP that c-commands it. If
the adjunct is attached in the complement clause, however, then the matrix
subject and the matrix indirect object both c-command it, and as a
consequence, both qualify as a controller for its PRO-subject, regardless of
whether the complement clause has a PRO-subject. If this should be true,
however, we predict that the matrix indirect object should be a possible
controller for PRO in the adjunct clause even when it is not the controller of
PRO in the complement clause. The prediction is not borne out:
Transparency in Dutch TP complements 11
(19) dat iki de studentj beloofde [PROi/*j het artikel te lezen] [zonder
that I the student promised the article to read without
PROi/*j mei/ *zichj ervoor te schamen]
myself/ himself for-it to shame
‘That I promised the student to read the article without being ashamed of
it.’
I take the fact that coreference between the matrix indirect object and PRO in
an adjunct clause introduced by zonder ‘without’ is only possible in the context
of an infinitival complement clause to an object control verb, to mean that PRO
in such adjunct clauses is truly subject-oriented. The fact that coreference
between the matrix indirect object and PRO in such an adjunct is possible with
object control verbs in the third construction (18) is then most naturally
explained by assuming that third construction complements may have a PRO-
subject.
Another indication that the complement clause in the third construction has
a PRO subject can be found in the interpretation of PRO. It has been noticed
that in certain cases, PRO and its controller are not identical in reference (cf.
Landau 2000). Certain predicates allow for partial control, in which the
controller represents only a subset of the referents associated with PRO. It
seems that when this is possible, the option is retained in the third construction:
12
Examples like (20) should be distinguished from sentences like (ia), which do not
necessarily represent partial control, as the controller is not among those individuals who actually
carry out the activity, but rather represents the group as a whole:
i) a. Balkenende besloot naar Irak te gaan ‘Balkenende decided to go to Iraq.’
b. Balkenende ging naar Irak ‘Balkenende went to Iraq.’
Another indication that (i) does not represent partial control is the observation that the monoclausal
sentence (ib) is acceptable in the interpretation of a collective going to Iraq, despite the fact that the
only DP in the sentence is singular. I thank Jan-Wouter Zwart for pointing out the examples in (i).
12 Janneke ter Beek
proposed by Wurmbrand (2001). One might conclude that the Dutch third
construction therefore cannot reflect A-movement.
Wurmbrand (2001) argues that several verbs, most notably irrealis verbs,
are incompatible with A-movement into the matrix clause. Focus scrambling is
possible in those cases, however. It is therefore worth investigating whether the
Dutch third construction might be the result of focus scrambling. Even though
we have discarded a trigger in terms of interpretive effects in section 3, the
possibility of focus scrambling should be investigated, especially since this is
actually what Wurmbrand (2001) suggests might be the correct analysis for
Dutch, based on the fact that the class of verbs that allow the third construction
in Dutch roughly corresponds to the class of verbs that allow focus scrambling
but not A-movement in German.13
The data clearly demonstrate that we can exclude focus scrambling,
however. Firstly, as pointed out in Neeleman (1994), focus scrambling may
alter the order of the arguments (21a), but this is not true for the third
construction (21b):
(21) a. dat zulke boeken zelfs Jan niet koopt (Neeleman 1994; 84)
that such books even Jan not buys
‘That such books, even Jan does not buy.’
b. * dat deze wetenschappers Jan beweert te kennen
that these scientists Jan claims to know
‘That Jan claims to know these scientists.’
c. dat zulke wetenschappers zelfs Jan beweert te bewonderen
that such scientists even Jan claims to admire
‘That such scientists, even Jan claims to admire.’
The example in (21b) is ungrammatical with neutral intonation. Notice that the
point is not that focus scrambling is impossible in the third construction, but
that the third construction cannot be reduced to focus scrambling. The example
in (21c) demonstrates that focus scrambling is possible in the third construction
(if pronounced with the distinctive intonation pattern associated with it).
Secondly, focus scrambling is incompatible with wide focus (22), but the
third construction is not, as was shown in (10a):
Thirdly, focus scrambling does not apply to parts of idioms (23), unlike the
third construction, as demonstrated in (11a) above:
13
Actually, the class is a bit larger than the class of verbs allowing focus scrambling in
German. Wurmbrand (2001) demonstrates that German propositional verbs are incompatible with
focus scrambling., but the third construction in Dutch is possible with several propositional verbs.
Transparency in Dutch TP complements 13
These contrasts are hard to explain if the third construction is derived by focus
scrambling. Therefore, I take (21)-(23) to show that focus scrambling is not
involved. However, if focus scrambling is not at issue, and we maintain that the
embedded clause in the third construction is not deficient in any obvious way,
then we are faced with a paradox. On the one hand, the third construction
behaves like a deficient clause in that it fails to license its internal argument,
but on the other hand, no related properties of deficiency can be identified,
other than the absence of a CP-layer. The next section explores a line of
analysis that might reconcile these contradictory properties.
If one assumes that derivation proceeds by phase, with phases being vP and CP,
as Chomsky (2001) proposes, then a situation like the one we find in Dutch is
not unexpected. If TP is not a phase, then we predict that such clauses display
transparency. The Phase Impenetrability Condition (PIC) restricts access of the
matrix v to its domain , but it has no access to material lower than the phase
head of the phase dominated by vP. The reason for this is that Spell out
proceeds by phase: when the head of a new phase is merged, the domain of the
lower phase (but not the edge) undergoes Spell Out.
Normally, clausal complements are CPs. Thus, a probe in the matrix vP has
access to C and the edge of the CP-phase, but the structure below C is
inaccessible, having already undergone Spell Out. The analyis accounts for the
fact that A-movement out of CP is impossible, regardless of whether CP is
finite or non-finite:
If an Agr in the matrix vP-phase attracts the embedded goal argument, the
matrix clause must have access to the material inside the complement clause.
From this perspective, it is unclear why the theme is not also attracted.
I speculate that this problem is avoided if we adopt a scenario along the
lines of Zwart (2001). Let us assume that contrary to what we assumed above,
the embedded clause in a third construction complement does not license its
internal arguments, that is, it does not project the relevant AgrPs. If the
complement clause were a CP, this would be fatal: only the edge of CP is
accessible to the matrix vP phase, so the matrix v, or more precisely, the
relevant Agr(s) in that phase cannot license the internal argument(s) in the
embedded clause. However, if the complement clause is a TP, then the AgrP(s)
in the matrix vP can access the embedded clause and attract an argument.
Now, if we assume that the transparency of the embedded clause allows for
the generation of the AgrP that licenses the internal argument in the functional
domain of the matrix clause, then we can understand how (25) comes about.
The embedded clause has to project two AgrPs, but because there is no CP-
layer to prevent it, one of them is generated in the matrix clause in (25).15
14
Note that I ignore a serious locality issue here. Unless we want to assume that a probe in the
matrix vP-phase attracts PRO before it attracts the object, we have to account for the fact that PRO
fails to yield the predicted intervention effect.
15
If the position in which functional heads are generated is flexible within the phase, we might
wonder why we cannot generate the Dutch equivalent of [CP that [AgrOP himj [TP Ii [vP ti [VP like tj ]]]]].
I cannot answer this question at this point. I speculate that if locality should be understood along
the lines of Fox & Pesetsky (2004), we may exclude such cases on the basis of ordering principles,
but clearly more research has to done to support such a view. An advantage of such reasoning is
that since the ordering principles are assumed to hold for elements with phonological features, the
locality issues concerning PRO in footnote 15 vanish.
Transparency in Dutch TP complements 15
Note that this scenario is independently needed to account for the fact that
the matrix clause in (25) is capable of licensing an indirect object in the first
place. The verb besluiten ‘decide’ can take a nominal direct object in certain
cases, but it cannot take a nominal indirect object (except in the third
construction, as was above):
Given that most of the matrix verbs that allow for a third construction
complement never take two nominal internal arguments, it is not immediately
clear that the matrix clause should be able to license two nominal internal
objects. I therefore follow Zwart (2001) in assuming that the embedded internal
arguments are not licensed by independently available functional structure of
the matrix clause, but rather move into the functional projections associated
with the embedded clause, which for whichever reason are generated in the
matrix clause. However, a more careful study of the issues awaits further
research.
6. Conclusion
I would like to thank the participants of the First Syntax AiO Meeting and the
2005 TIN-Dag, as well as the department of General Linguistics of the
University of Groningen for useful comments and suggestions. Special thanks
go to Jan-Wouter Zwart for helpful remarks on an earlier version of this paper.
Needless to say, all remaining errors are mine.
References
Besten, H. den, J. Rutten, T. Veenstra & J. Veld (1988). Verb raising, extrapositie en de derde
constructie. Ms., University of Amsterdam.
Besten, H. den & J. Rutten (1989). On verb raising and free word order in Dutch. Jaspers, D. , W.
Klooster, Y. Putseys & P. Seuren (eds.), Sentential complementation and the lexicon:
Studies in honour of W. de Geest. Foris, Dordrecht, pp. 41-56.
Broekhuis, H. (1992). Chain-government: Issues in Dutch syntax. Diss., University of Amsterdam.
Broekhuis, H., H. den Besten & K. Hoekstra (1995). Infinitival complementation in Dutch: On
remnant extraposition. The Linguistic Review 12, pp. 93-122.
Chomsky, N. (1995). The minimalist program. MIT Press, Cambridge, MA.
Chomsky, N. (2001). Derivation by phase. Kenstowicz, M. (ed.), Ken Hale: A life in language.
MIT Press, Cambridge, MA.
Evers, A. (1975). The transformational cycle in Dutch and German. Diss., University of Utrecht.
Fox, D. & D. Pesetsky (to appear). Cyclic linearization of syntactic structure. É. Kiss, K. (ed.)
Theoretical linguistics.
Haaften, T. van (1991). De interpretatie van verzwegen subjecten. I.C.G. Printing, Dordrecht.
Hoop, H. de (1992). Case configuration and noun phrase interpretation. Diss., University of
Groningen.
IJbema, A. (2002). Grammaticalization and infinitival complements in Dutch. Diss., University of
Leiden.
Landau, I. (2000). Elements of control: Structure and meaning in infinitival constructions. Kluwer
Academic Publishers, Dordrecht/Boston/London.
Neeleman, A. (1994). Complex Predicates. Diss., University of Utrecht.
Rutten, J. (1991). Infinitival complements and auxiliaries. Diss., University of Amsterdam.
Wöllstein-Leisten, A. (2001). Die Syntax der dritten Konstruktion: Eine repräsentationelle Analyse
zur Monosententialität von ‘zu’-Infinitiven im Deutschen. Stauffenburg Verlag, Tübingen.
Wurmbrand, S. (2001). Infinitives: Restructuring and clause structure. Mouton de Gruyter ,
Berlin/New York.
Zwart, C.J.W. (1993). Dutch syntax: A minimalist approach. Diss., University of Groningen.
Zwart, C.J.W. (1995). Word-order, intonation, and noun phrase interpretation in Dutch. Samiian,
V. & J. Schaeffer (eds.), Proceedings of the Western Conference on Linguistics 7. pp 279-
289.
Zwart, C.J.W. (2001). Object shift with raising verbs. Linguistic Inquiry 32:3, pp. 547-554.