On the Parametrisation of Functional Projections in CP

Eric Hoekstra, Royal Dutch Academy of Sciences1

(A.J. Schafer (ed) 23th Proceedings of the North-East Linguistic Society 1992. U. of Massachusetts, Amherst, (1993), 191-204. )
 
 

1. Introduction

Consider the sentences in (1) below:

(1) a. Hij weet welke jongen of dat je gezien hebt
he knows which boy C2 C3 you seen have
b. Dat is niet zo gek als of dat hij gedacht had
that is not as crazy C1 C2 C3 he thought had

It can be noticed that there can be as much as three complementisers in varieties of substandard and dialectal Dutch, such as Frisian Dutch and Flemish (De Rooy 1965) where English only has one complementiser, as in the translation of (1b), or none, as in the translation of (1a).2 To account for these facts, two hypotheses can be drawn up, as in (2):

(2)
a. H1: three complementisers ===> three syntactic heads (words) ===> three maximal projections
b. H2: three complementisers ===> one syntactic head (word) ===> one maximal projection

It is an empirical question which hypothesis is correct. In section 2 it is shown that facts of conjunction reduction support H1.

In section 3 I show that each of the three postulated syntactic heads has its own semantic character. Combining this structure (three heads and, possibly, three specifiers) with the format of parametrisation proposed in the Minimalist Program (Chomsky 1992) yields a a number of possibilities for parametric variation. Some of these are actually attested in Germanic dialects such as (i) lack of Verb-Second in a specific semantic environment in Standard Dutch, hitherto unaccounted for (section 4) (ii) optionality of Verb-Second in West Flemish and French Flemish after (-Wh) topics (section 5). In addition, the system directly accommodates a number of asymmetries between topic-movement and Wh-movement in Dutch (section 6). Complementiser strengthening under subject extraction is the basis for a fresh analysis of that-t effects (section 7). Finally (section 8), it is argued that topics in Dutch are not licensed by Spec-Head agreement but as adjuncts. Hence topics may only adjoin to non-argument types such as ungoverned CPs, explaining the asymmetry between root sentences and non-root sentences.

2. Conjunction reduction supports H1

The morphological hypothesis in (2b) predicts that the sequence of complementisers cannot be broken up by conjunction reduction. From the point of view of the syntactic hypothesis in (2a), there is no reason why conjunction reduction could not split up the sequence of complementisers. Consider the following sentences (cf. De Rooy 1978):

(3)
a. * [in-[dien hij koning is] en [-dien zij koningin is]]
if he king is and -if she queen is
b. * [om- [dat het nu regent] en [-dat het straks gaat sneeuwen]
because it now rains and -because it later will snow
===> a syntactic head cannot be split up by conjunction
(4)
a. [Als [of [ [dat hij koning is] en [dat zij koningin is]]]]
C1 C2 C3 he king is and C3 she queen is
b. Ik vraag me af [of [dat Ajax de volgende ronde haalt]
I wonder C2 C3 Ajax the next round reaches
en [dat Celtic verslagen kan worden]]
and C3 Celtic beaten can be
===> Conjunction of C3P: C2 (C3P & C3P)
(5)
a. Niet zo snel [als [of dat hij hoopte] en [of dat zij verwachtte]]]
not as fast C1 C2 C3 he hoped and C2 C3 she expected
===> Conjunction of C2P: C1 (C2P en C2P)
b. [Als [of dat hij koning is] en [of dat zij koningin is]]]
C1 C2 C3 he king is and C2 C3 she queen is

Consider first the sentences in (3). These are ungrammatical and they show that conjunction reduction cannot split up a word. Conversely, (4) and (5) show that conjunction reduction may split up a sequence of three complementisers. Specifically, (4) involves a conjunction of maximal projections of C3's, and (5) involves a conjunction of two maximal projections of C2's. This indicates that the sequences of complementisers must not be analysed as one word, but as three words, and this implies that there are three syntactic heads.

3. Semantic Nature of Each Head

Each of the three functional heads has its own semantic character, as illustrated below with facts from Standard Dutch:

(6)
a. Het feit dat/ *of/ *als hij op tijd is:
the fact C3 *C2 *C1 he on time is
===> C3 in affirmative contexts (C3P = TopP)
b. De vraag of/ *dat/*als hij op tijd is
the question C2 *C3 *C1 he on time is
===> C2 in question (Wh) contexts (C2P = WhP)
c. Niet zo gek als/*of/ *dat ik gedacht had
not as crazy C1 *C2 *C3 I thought had
===> C1 in comparative contexts (C1P)

(6a) shows that C3 occurs in affirmative contexts. For reasons that will become clear in a moment, I will refer to the projection of C3 as a topic projection, and to the head C3 as Top. (6b) shows that C2 occurs in Wh-contexts. Hence I refer to the projection of C2 as a Wh projection and to the head C2 as Wh. C1 is associated with what I have tentatively dubbed comparative contexts.

4. Parametrisation of Functional Projections

The structure in (7) outlines my proposal:

(7)
a. C1P - WhP - TopP - AgP
C1 Wh Top correspond to (als) (of) (dat)
b. Three heads ===> three potential landing sites for V/2 & ===> three potential Spec-positions

Each functional head may in principle have a Spec. The presence of the Spec does not depend on X'-theory, but on whether it is licensed by the presence of appropriate functional features, e.g. the Wh-feature in the case of WhP (Hoekstra 1991). Furthermore, each functional head may or may not trigger Verb-movement before PF (cf. Den Besten 1989), depending on whether the licensing feature is strong or not (Chomsky 1992). Let us see how the three functional projections behave with respect to licensing Spec and attracting the verb:

(8)
a. Jan kent hij niet
Jan knows he not
b. * Jan hij kent niet
Jan he knows not
===> TopP triggers V/2
(9)
a. Wie kent hij
who knows he
b. * Wie hij kent
who he knows
===> WhP triggers V/2
(10)
a. * [Hoe sneller je loopt] ben je des te eerder thuis
the faster you walk are you the sooner at home
b. [Hoe sneller je loopt], des te eerder ben je thuis
the faster you walk the sooner are you at home
===> C1 does not trigger V/2

Consider (8) first. It shows that TopP (C3P) licenses a topic as its Spec, and that it attracts the verb. This is one half of the generalisation that Dutch is a Verb-Second language. Consider next (9). It shows that the WhP (C2P) licenses a Spec and that it attracts the verb. This is the other half of the Verb-Second generalisation.

(10) illustrates that there is also a functional projection which does not attract the verb, although its Spec may be filled. I would like to suggest that this is C1, the functional head for comparative contexts. The head of this projection is weak; hence it does not trigger V/2. There is independent evidence for a C1 projection. Suppose there is a language in which C1 is strong. Hence C1 would normally host a complementiser or a verb. In embedded Wh-contexts, this complementiser would PRECEDE the Wh-phrase rather than follow it. This is exactly the case in Hungarian (Marácz 1989):

(11)
a. Nem tudom hogy kivel találkozott János
not know-1S C1 who-INS met-3S John
I don't know who John met
b. Az a kérdés, hogy mit látott Péter érdekes
it the question C1 what-ACC saw-3SG Peter interesting
"the question what Peter saw is interesting

These sentences are directly accommodated in the universal structure postulated for CP on the basis of facts from Germanic.

5. A discovery: Flemish dialects with optional V/2 after topics

My analysis entails that Verb-Second is not a unified phenomenon. There is ample evidence for this claim. It is well-known to historical linguists (e.g. Van Kemenade 1987) that throughout the history of the germanic languages Verb-Second for Wh-words has been more stable than Verb-Second for topics. Throughout the history of English, V/2 always occurred after Wh-phrases, but not always after topics.The dichotomy between the two is predicted by my analysis, taken in conjunction with a minimalist view of parameters.
There are also dialects of Dutch in which topics (but not Wh-words) fail to trigger Verb-Second (Vanacker 1968, Debrabandere 1976):3

(12)
a. West Flemish (AND)
[Als de vogels een sperwer zien] ze hebben angst
if the birds a sparrow-hawk see they have fear
b. St. Sylvestre-Kappel, French-Flemish (RND 6: N105)
[Bleke viezen] me zieni fele gimeer uliern
pale calves we see-not many anymore here
c. Kaasteren, French-Flemish (Winkler 1874:393)
[Over vele jaren] der wos a boer
many years ago there was a farmer
d. Kortrijk, French Flemish (DeBrabandere 1976:88)
Met da slich were mijn haar krul were
with that bad weather my hair curls again

This is the same asymmetry we find in English. It is accommodated by saying that the head of the Top-projection is weak in those dialects, as it is in English.

Here I want to address a general problem. As it turns out, Verb-Second in Flemish dialects is optional. This gives rise to the following problem:

(13) Question: how can V/2 be optional, given economy? Answer: both parameter values have been acquired.

In general, economy forbids all optionality. If a parametric specification triggers Verb-Second then it will be obligatory. If a parametric specification does not trigger V/2 movement then it will be ruled out by economy. So how can it be optional? Here I want to offer the general suggestion that two contrary parameter specifications can be acquired by the language learner. After all, nothing in the theory rules out that both parametric values are available. The advantages are tremendous. It not only solves the problem of the Flemish facts. It also makes possible language change. Furthermore, there is evidence from language acquisition that children start out with all parameter values available (Verrips 1992). If parameter values are available simultaneously, we can also account for the fact that speakers command not just a dialect but a continuum between two dialects. For example, a speaker from the East will command a continuum from the Eastern dialect to Standard Dutch. In addition, if both values of a parameter are available (optionality), all kinds of sociolinguistic factors may come into play. This parametric picture also makes clear that people's judgments are less homogenuous than we expect, and many a battle over grammaticality judgments may be explained in this way (see Hoekstra 1992 for more extensive discussion). Thus, appraisal of and insight into the actual speaker and the actual speech community will help to form a more adequate picture of the ideal speaker and the ideal speech community.

6. Three arguments for TopP and WhP

Even in standard Dutch, we find an asymmetry between topics and Wh-words supporting the proposed view of functional projections in CP:

(14)
a. Wie denk je [of ik gezien heb]
who think you Wh I seen have
b. * Jan denk ik [of ik gezien heb]
Jan think I Wh I seen have

The (a)-sentence shows that long Wh-movement is compatible with the complementiser "of", the complementiser for the Wh-projection. The (b)-sentence shows that long topic-movement is incompatible with this complementiser. This follows, since the topic, lacking Wh-features, may not go through the Spec of Wh. Not being eliminated, the Wh-features in the Spec of WhP will survive into PF, causing a violation of Full Interpretation. Both WhP and ToP are active in Dutch. Assuming that either of these projections can be used in principle for relativisation, we account for the fact that there are two types of relativisation in Dutch (Weynen 1966):

(15)
a. Het huis wat ik zag (rel. on WhP)
the house which I saw
b. Het huisje dat ik zag (rel. on TopP)
the house Top I saw

One type employs Wh-words and the other type employs D-words. Wh-relativisation is analysed as involving the Spec or Head of WhP, whereas D-relativisation involves the head of TopP. The Spec of TopP may not be involved, for reasons given in section 8. While this morphological correlation need not be perfect in every dialect, complementiser doubling provides a clear indication of the type of relativisation that is involved.
We now expect that the Wh-type may be followed by a Wh- complementiser but the ToP type not. This is correct:

(16)
Cooccurrence with Wh-complementizer.
a. ? Het huisje wat of ik gezien heb
the house which Wh I seen have
b. * Het huisje dat of ik gezien heb
the house that Wh I seen have

Relativisation on ToP cannot cooccur with a following Wh-complementiser like "of", whereas relativisation on (the Spec of) WhP is perfectly compatible with the Wh-complementiser, as expected

7. A fresh analysis of the that-t effect

The head of the topic projection may be filled, but this is not obligatory if the head of WhP is filled. Interestingly, there is a light that-t effect here, as shown below:

(17)
a. ? Wie denk je [of hem gezien heeft]
who think you Wh him seen have
b. Wie denk je [of hij gezien heeft]
who think you Wh he seen have
c. Wie denk je [of dat hem gezien heeft]
who think you Wh To him seen have

(17a) is worse than (17b) for some speakers. Those speakers who do not like to use "of" in the first place do not have a difference in grammaticality. There are no speakers who prefer (17a) to (17b). Furthermore, a subject extraction increases in grammaticality if you make the topic head lexical as in the (c)-sentence. The effect is too marginal to be attributed to the ECP. Independent evidence against an ECP-account is provided by Culicover (1992).

If not the ECP, then what is the explanation for the that-t effect? Gerken (1990) shows that subject-drop in child language is due to metrical reasons. This account, contrary to a pro-drop account, also explains why subject determiners are more often dropped in child language than object determiners. Furthermore, given that there is both AgO and AgS, the pro-drop account predicts that objects and subjects will show symmetric behaviour, in the earliest stages. But children drop subjects rather than objects.

Gerken's analysis can serve as the inspiration for a fresh analysis of the that-t effect as well. Suppose that a trace is organised into a metrical foot with the preceding constituent. Suppose further that the combination of a trace with a weakly stressed syllable yields the marginal unacceptability which characterises that-t sentences. This account makes the following predictions:

(18) Only languages with weakly stressed complementiser exhibit a that-t effect.
(19) Complementiser deletion or complementiser strengthening will obviate the that-t effect.
(20) An adverbial which may precede the subject will obviate the that-t effect.
(21) The that-t effect will show up with the lowest trace only

(18) is confirmed by the fact that English has a that-t effect, seeing that the English complementiser is weakly stressed. Conversely, Dutch hardly has a that-t effect because the Dutch complementiser "dat" is not weakly stressed. French is a paradigm case. "que" is weak, and the que-to-qui rule is typically an example of phonological strengthening, revealing the involvement of stress. The other way to satisfy metrical requirements is to delete the complementiser altogether, as in English (also supporting (19)). Culicover's (1992) facts bear out (20), as the paradigm below illustrates:

(21)
a. * Who do you think that t was the mayor of the city
b. Who do you think that for all intents and purposes t was
the mayor of the city

The sentential adverb prevents the generation of a metrical foot consisting of a complementiser and a trace. Finally, prediction (20) is borne out by the paradigm which is given below:

(22)
a. * Who do you think that John said that came
b. Who do you think that John said came

The that-t effect shows up on the lowest complementiser only. This follows, since this is the only complementiser left-adjacent to a trace. The Dutch facts in (17) similarly support this analysis. "Of" is phonologically weaker than "dat". Hence strengthening of "of" to "of dat" is preferred.

Interestingly, the Wh-complementiser "of" can only show up with verbs like "think" or "say":

(23) a. Wie bekende Jan ?dat/*of dat hij gezien had
who confessed Jan Top/Wh Top he seen had
b. Wie betreurt Jan ?dat/*of dat hij op moet bellen
who regrets Jan Top/Wh Top he up must call

These same verbs of (23) do not allow embedded topicalisation in English (Watanabe 1992), supporting his suggestion that there is a factive operator (-Wh), which we would situate in the Spec of WhP, from where it would block the presence of the <+Wh> complementiser "of" in Dutch. It would also force WH-movement to take place in one swoop, which explains why Wh-movement is worse with factives than with verbs like "think" or "say".

8. On the Licensing of Topics

With so many projections there is some overgeneration. One specific problem is why topics only occurs in main clauses and not in embedded clauses, as shown below:

(24)
a. * Ze dachten Jan dat ik gezien had
they thought Jan Top I seen had
b. * Ik weet waarom of Jan dat ze belden
Ik know why Wh Jan Top they phoned

To solve this problem, let us investigate our minimalist roots. Notice first that Spec and Head of an arbitrary functional projection need not both be strong or weak. That is, the Spec of a functional projection may be strong without the head being strong and vice versa. This implies that Spec-Head agreement does not require agreement with respect to strength. This claim is independently supported by Chomsky's (1992) analysis of Arabic subject agreement (but cf. Huybregts 1991).

The question can be asked: is the Spec of TopP licensed by Spec-Head agreement? If it were, then the topic would be found both in main and embedded clauses. Hence we conclude that the Spec of TopP is NOT licensed by Spec-Head agreement. Circumstantial evidence for this conclusion comes from the fact (25):

(25) Fact: there is no Germanic dialect, now or in the past, which ever morphologically marked topics

This holds true even for those dialects which retain a reasonably elaborate system of morphological Case like Icelandic or German. Notice I am now assuming that morphological marking of topics takes place under Spec-Head agreement with the verb (at LF, see Law 1987). The difference between Germanic, on the one hand, and, say, Tagalog, on the other hand, would be that topics in Germanic do not participate in Spec-Head agreement. The similarity is that for both the topic feature in the Spec of TopP is strong.

The next question to ask is: if topics are not licensed by Spec-Head agreement in Germanic, then how are they licensed? A reasonable answer is that they get licensed as adjuncts (Hoekstra 1991). Adjuncts are licensed if they adjoin to a non-argument type (Chomsky 1986). Main clause CP's are non-arguments, since there is no predicate which takes a main clause CP as an argument. Embedded CP's are argument-types. It follows that topics cannot be licensed if they are adjoined to an embedded CP. By parity of reasoning, a topic cannot be adjoined to a TopP if that TopP is the argument of a higher head (Wh). This derives the fact that topics only occur in main clauses.4

So-called "CP-recursion" provides another piece of circumstantial evidence. There is a licensing asymmetry between verbs and expletives in Top with respect to topics:

(26)
a. Ik denk dat [Jan dat ik die niet ga feliciteren]
I think that Jan that I him not go congratulate
b. Ik ga [Jan ga ik niet feliciteren]
I go Jan go I not congratulate
(27)
a. * Ik denk [dat Jan dat ik niet ga feliciteren]
I think that Jan that I not go congratulate
b. * Ik ga [Jan ga ik hem niet feliciteren]
I go Jan go I him not congratulate

The topic adjacent to the verb gets licensed as an adjunct. The topic adjacent to "dat" cannot be licensed as an adjunct. Instead, it must be doubled by a pronoun, indicating that it gets licensed as a left-dislocated element. This correlates with the fact that quantifiers are excluded in the position adjacent to the expletive "dat", whereas they are allowed in the position adjacent to the verb:

(28)
a. * Niemand die ken ik
nobody him know I
b. * Ik denk dat niemand dat hij die kent
I think that nobody that he him knows
(29)
a. Niemand ken ik
nobody know I
b. Ik ga niemand ga ik vandaag feliciteren
I go nobody go I today congratulate

These facts indicate that a topic in an embedded clause is not normally allowed; it can only be licensed as a left-dislocated element. A topic in a main clause is unproblematic. This follows from the idea that topics in Germanic move to a Spec position which is not licensed by Spec-Head agreement and which subsequently needs to be licensed as an adjunct position.5

9. Conclusion

The evidence for splitting up CP into three projections is overwhelming. Comparative contexts without V/2 can now be accommodated, as well as the various asymmetries between topics and Wh-words. Combining this structure with the Minimalist view of parametrisation, we predict cross-linguistic variation in V/2 which is actually attested in Germanic dialects.

Acknowledgements

I would like to thank Werner Abraham, Peter Ackema, Josef Bayer, Hans den Besten, Marinel Gerritsen, Jarich Hoekstra, Wim Kosmeyer, Ad Neeleman, Eric Reuland, Berber Voortman, Jan-Wouter Zwart, the participants of the conference "Dialektsyntax in den germanischen Sprachen", the participants of the TABU-day (Groningen) and the participants of NELS 23 for useful discussion and comments.

References

Besten, H den (1989) Studies in West Germanic Syntax. Dissertation, University of Amsterdam.
Blancquaert E. & W. Pée (1930-1982) (eds) Reeks Nederlands Dialect-atlassen (RND). Antwerpen.
Brachin, P. (1977) "Of + inversie (II)". De Nieuwe Taalgids 70: 422.
Chomsky, N. (1986) Barriers. MIT Press, Cambridge.
Chomsky, N. (1992) "A Minimalist Program for Linguistic Theory". MIT Occasional Papers in Linguistics 1. Massachusetts Institute of Technology, Cambridge (USA).
Culicover, P. (1992) "The Adverb Effect: evidence against ECP accounts of the that-t effect". Talk presented at NELS 23, Ottawa.
Debrabandere, F. (1976) "De SVf-woordorde in Zinnen met een Aanloop". In Handelingen van de Koninklijke Commissie voor Toponomie en Dialectologie, 87-97. Drukkerij G. Michiels, Tongeren.
Den Dikken, M. (1992) Particles. Dissertation, University of Leiden.
Gerken, L. (1990) "A Metrical Account of Children's Subjectless Sentences". In J. Carter, R.-M. Déchaine, B. Philip & T. Sherer (eds) NELS 20, 121-134.
Gerritsen, M. (1991) Atlas van de Nederlandse Dialectsyntaxis (AND). P.J. Meertens Instituut, Amsterdam.
Hoekstra, E. (1991) Licensing Conditions on Phrase Structure. Dissertation, University of Groningen.
Hoekstra, E. (1992) "Generative Grammar meets Dialectal Variation", in D. Gilberts & S. Looyenga (in press) Language and Cognition II, University of Groningen.
Hoekstra, J. (1987) "'Verb-Second' en de 'Imperativus Pro Infinitivo' in het Fries". Taalkundig Bulletin (TABU) 17: 96-121.
Hoekstra, J. & L. Marácz (1989) "Some Implications of I-to-C Movement in Frisian". In H. Bennis & A. van Kemenade (eds) Linguistics in the Netherlands 1989, 81-90. Foris, Dordrecht.
Huybregts, R. (1991) "Allosteric Agreement in VSO Languages", in F. Drijkoningen & A. van Kemenade (eds) Linguistics in the Netherlands 1991, 81-90, Benjamins, Amsterdam.
Law, P. (1987) Effects of Head-Movement on Theories of Subjacency and Proper Government. Dissertation, MIT.
Platzack, C. & A. Holmberg (1989) "The Role of AGR and Finiteness in Germanic VO-languages". GLOW Newsletter 22, 53-54.
Rooy, J. de (1965) Als-Of-Dat. Van Gorcum, Assen.
Rooy, J. de (1978) "Het ene 'of' is het andere niet". De Nieuwe Taalgids 71, 146-148.
Safir, K. & D. Pesetsky (1981) "Inflection, Inversion and Subject Clitics". Proceedings of North-East Linguistic Society (NELS) 11: 331-344.
Vanacker, V.F. (1968) "Syntaktische Daten aus Französisch-Flämischen Tonband Aufnahmen". Verhandlungen des 2e Internationalen Dialektologencongresses, Beiheft 4 des Zeitschrift für Mundartforschung, 844-855. Franz Steiner Verlag, Wiesbaden..
Verrips, M. (1992) "Learnability meets development: the case of pro-drop", Ms, University of Amsterdam.
Watanabe, A. (1992) "Larsonian CP-Recursion, Factive Predicates and Selection". Talk presented at NELS 23, Ottawa.
Weijnen, A. (1966) Nederlandse Dialectkunde. Van Gorcum, Assen.
Winkler, J. (1874) Algemeen Nederduitsch en Friesch Dialecticon. Martinus Nijhoff, 's Gravenhage (reprinted (1972) Dr. Martin Sändig oHG, Wiesbaden).
Zwart, J.-W. (1992) "Verb-Movement and Complementizer Agreement". GLOW Newsletter 26, 58-59. Ms, University of Groningen.
 

1 ehoekstra@fa.knaw.nl

2 At the conference "Dialektsyntax in den germanischen Sprachen", it became clear that the phenomenon of triple complementisers is also found in Bavarian.

3 A note on sources is in order here. AND (Gerritsen 1991) stands for "Atlas van de Nederlandse Dialectsyntaxis" ("Atlas of Dutch Dialect Syntax"), an overview of the geographical variation in The Netherlands and Flanders with respect to a number of syntactic parameters. This overview was based on written inquiries sent out by the Meertens Institute to its 1000 informants all over the country. RND (Blancquaert & Pée 1930-1982) stands for "Reeks Nederlandse Dialectatlassen" ("Series of Dutch dialect atlases"), a systematic investigation in which dialect speakers translate a specific set of sentences from Dutch or French into their dialect. Winkler (1874) is a dialecticon in which dialect speakers retell the biblical story of the lost son.

4 The distribution of verbs and expletives still remains to be explained but could conceivably be related to the argumenthood of CP as well, in the spirit of Safir & Pesetsky (1981), cf. Platzack & Holmberg (1989). Cases of Verb-First following an expletive (see e.g. Brachin 1977, J. Hoekstra 1987, and others) must be ascribed to the (lexically exceptional) failure of the expletive to bind lower functional heads. In the normal case, an expletive obviates V-movement by binding lower functional heads (e.g. Zwart 1992, cf. J. Hoekstra & Marácz 1989).

5 CP-recursion in English seems to be different from CP-recursion in Dutch. In English, it is conditioned by semantic factors (Watanabe 1992), whereas in Dutch there do not seem to be such restrictions. The non-parametrisability of LF entails that "CP-recursion" is a term used to describe two unrelated phenomena. It should also be kept in mind that as a theoretical solution CP-recursion is unattractive, because of its stipulative nature. This is also true of Larsonian VP-recursion. It does not hold of the extensive VP-internal structures proposed in, for example, Den Dikken (1992), since those (small clause) structures are semantically motivated, as well. One would expect that the CP-recursion facts can also be explained once the specific structure of CP has been properly fleshed out.