Spektator. Jaargang 4

[p. 161]

On Extraposition^*
Ger J. De Haan

Details aside, two fundamentally different theories, mainly based on English. have been proposed with respect to the extraposition phenomenon, that of Rosenbaum (1967) and of Emonds (1970). In this paper, we will compare the two alternative theories and their applicability in Dutch syntax. It turns out to be the case that both theories are inadequate in some respects and we present a new solution, which is some sort of a compromise between the points of view of Rosenbaum and Emonds.

* This paper is part of a larger study that will be published elsewhere. It is not checked by a person who is familiar with the English language. Therefore, no claim is made with respect to its readability.

1.0. Earlier theories on extraposition

1.1. Rosenbaum (1967).

Sentential complements can function as subject, compare,

(1)	dat de kraai dood is, ergert mij
(2)	dat Karel gek is, is voor iedereen duidelijk

According to Rosenbaum (1967), we must relate these sentences to

(3)	het ergert mij, dat de kraai dood is
(4)	het is voor iedereen duidelijk, dat Karel gek is

Rosenbaum expresses this relation in the following way: he postulates for these sentences underlying structures like

He derives now from these structures the sentences (3) and (4) by a rule called Extraposition which moves the embedded sentence to the end of the matrix sentence. Assuming that Extraposition is optional in this case, Rosenbaum derives the sentences (1) and (2) by a rule called It Deletion which we relabel here as Pronoun Deletion.

Extraposition also applies to direct object sentences, cf.,

(7)	Karel waardeerde het in jou, dat je altijd op tijd was

The situation is about the same with regard to prepositional object sentences, although an additional rule is required in order to derive correct surface forms.

Copyright © 1974 by Ger J. de Haan

[p. 162]

Compare the underlying structure¹,

If we apply Extraposition to (8) - and of course a verb second rule - we yield the ungrammatical sentence

(9)	* Kareltje rekende op het, dat Henk zou komen

The derivation also ends up in a bad result, if we apply Pronoun Deletion together with Extraposition:

(10)	* Kareltje rekende op, dat Henk zou komen

We need a rule which changes op het into erop: we can represent the effect of this rule as²

We call this rule Pro Shift. After Pro Shift, we get a correct surface form:

(12)	Kareltje rekende erop, dat Henk zou komen

One of the big problems, Rosenbaum encounters in his treatment of the extraposition phenomenon, is the question how to account for the distributional peculiarities of sentential complements. In general, embedded sentences occur obligatory at the end of their matrix sentences, compare

1 We assume that Dutch is underlying SOV: this is irrelevant, however, for the matters underdiscussion.

2 Compare for some problems in connection with this rule, De Haan/ Koefoed/ Des Tombe (1974), 132-134.

[p. 163]

(13)	a	Karel betreurde het erg, dat hij niet kon komen
	b	* Karel betreurde, dat hij niet kon komen, erg
	c	ik wist dat Karel het betreurde, dat hij niet kon komen
	d	* ik wist dat Karel, dat hij niet kon komen, betreurde

There is, however, an exeption to this general statement: if the matrix sentence of a subject sentence is an embedded clause, then this subject sentence must occur in clause-final position, cf.

(14)	a	de dokter zegt dat het duidelijk is, dat Karel ziek is
	b	de dokter zegt dat, dat Karel ziek is, duidelijk is

- so this distribution still falls within the scope of the general statement -, but if the matrix sentence of a subject is a main clause (or a root sentence), then this subject sentence may occur in clause-initial position as well as in clause-final position, cf.

(15)	a	het is duidelijk dat Karel ziek is
	b	dat Karel ziek is, is duidelijk

Rosenbaum accounts for these distributional facts by making Extraposition optional for subject sentences in main clauses, and obligatory for the other cases. In a slightly different approach, Ross (1968) drops the OB/OP condition on Extraposition: he assumes that Extraposition is optional and he claims that sentences like (13)b-d and (14)b are filtered out as a consequence of a more general principle. the Internal S Condition:

(16)	Internal S Condition
	‘Grammatical sentences containing an internal NP which exhaustively dominates S are unacceptable’ (Ross (1968), 331.

Ross uses the following definition of the notion ‘internal NP’ ‘an NP is internal to a sentence if it is both preceded and followed by non-null parts of that sentence’ (Ross (1968), 59 fn. 8).

1.2. Emonds (1970).

A completely different approach to the extraposition phenomenon is outlined in Emonds (1970). One of the main reasons for Emonds' dissatisfaction with Rosenbaum's proposal is, that Rosenbaum is forced to formulate an ad hoc OB/OP condition on Extraposition. This becomes even worse in the light of sentences with so-called bisentential predicates, verbs which can have simultaneously sentential complements as their subject and (direct or prepositional) object, cf.

(17)	a	dat Karel een rode neus had, overtuigde mij ervan, dat hij dronken was
	b	dat stenen kunnen denken, impliceert, dat god bestaat

In these cases, Extraposition may not apply to the subject sentences:

(18)	a	* het overtuigde mij ervan, dat hij dronken was, dat Karel een rode neus had
	b	* het impliceert dat god bestaat, dat stenen kunnen denken

[p. 164]

If we embed sentences with bisentential predicates, we get bad results with and without application of Extraposition:

(19)	a	* Henk dacht, dat, dat Karel een rode neus had, mij ervan overtuigde, dat hij dronken was
	b	* Henk dacht, dat het mij ervan overtuigde, dat hij dronken was, dat hij een rode neus had
	c	* Henk dacht, dat, dat stenen konden denken, impliceerde, dat god bestond
	d	* Henk dacht, dat het impliceerde, dat god bestond, dat stenen konden denken

Rosenbaum will have to modify his Extraposition rule in order to exclude sentences like (18)a and b. (19)b and d. It is not clear that this can be done in a non- ad hoc way.

Emonds also does not accept Ross' alternative solution: he argues that there is no independent motivation for the existence of the Internal S Condition as a more general principle than just a restriction on Extraposition. The only example Emonds could think of as an argument for the more general nature of this restriction, i.e. headless relative clauses, appears to be a counterexample to the Internal S Condition³. If we assume that headless relative clauses are S's exhaustively dominated by NP in surface structure, then the Internal S Condition would wrongly predict that the following sentences are unacceptable:

(20)	a	Hans veronderstelt dat wie eens steelt altijd een dief zal blijven
	b	ik hoop dat wat Karel gezegd heeft jou ervan zal overtuigen, dat hij niet dronken is

In his theory on extraposition, Emonds avoids the OB/OP problems by dropping the rule of Extraposition. He proposes that all sentential complements are generated as S. not dominated by NP, in deep structure in ‘extraposed’ position, i.e. at the end of the VP. The underlying structure of (21)

(21)	ik betreurde het dat Karel mijn fiets gestolen heeft

will be (22)

The sentences

(23)	het is duidelijk dat Karel ziek is
(24)	dat Karel ziek is, is duidelijk

have the same underlying structure (25)

3 Higgins (1973) and Satyanarayana (1973) attempt to show that Emonds is wrong in claiming that headless relative clauses constitute counterexamples to the Internal S Condition, by assuming that such clauses do have pronominal heads. So, the Internal S Condition would not filter out internal headless relative clauses. Higgins and Satyanarayana fail to give arguments for their position; in particular, they will have to show that headless relative clauses are complex NP's in surface structure. But even under the assumption that they are right, their remarks bear no relevance to Emonds' claim that there is no evidence for a principle like the Internal S Condition which is more general than a rule-specific constraint

[p. 165]

The observation that sentential complements which are intuïvely interpreted as direct and prepositional object, can never occur in direct and prepositional object position, but always appear in ‘extraposed position’ is automatically accounted for, within such an analysis. The fact that sentential complements can not occur in subject position in non-root sentences, whereas they can in root sentences. Emonds explains by the postulation of a rule of Subject Replacement, which optionally replaces, only in root sentences, a pronoun in subject position by a sentential complement. So, sentence (24) is derived from (25) with application of Subject Replacement.

Emonds accounts for the distribution of the pronominal element het roughly in the same way as Rosenbaum does: he assumes that het is generated in deep structure in the position where Rosenbaum generates the configuration [het - S]_NP and a subsequent rule of Pronoun Deletion must give the correct distribution. In order to make Pronoun Deletion and also Subject Replacement work, Emonds will have to formulate both rules in terms of coreferential indices⁴.

How does Emonds solve the distributional problems, related to bisentential verbs? Emonds has introduced, for independent reasons, the concept of doubly filled node: two constituents of the same category can occupy one phrase structure position, provided that only one such constituent is present in that position at the level of surface structure. Sentences like

(26)	dat stenen kunnen denken, impliceert, dat god bestaat

have underlying structures of the following type

Such an underlying structure results only in a well formed surface structure if one of the sentential complements is removed, which is the case if Subject Replacement applies. Subject Replacement, however, can only apply in root sentences; so, Emonds correctly predicts that sentences with bisentential verbs are only well formed if they are root sentences in which Subject Replacement has applied, whereas all the other ungrammatical sentences need not be excluded by an ad hoc

4 Emonds must account for the interpretations of the sentential complements as subject, direct and prepositional object whereas they occupy all the same position in deep structure. The assumption of an underlying [het]_NP in the indicated positions together with the use of coreferential indices provides a solution to this problem. See Emonds (1970), 92

[p. 166]

condition on Subject Replacement, but are filtered out as a consequence of independent factors in the grammar.

Summarizing, Emonds claims that his proposal is superior to Rosenbaum's theory on extraposed sentences, in that it avoids an ad hoc treatment of distributional properties of these sentential complements: in particular, he has no OB/OP condition on Extraposition, nor an ad hoc constraint like the Internal S Condition. He achieves this result by replacing Extraposition with the optional root transformation Subject Replacement and by making certain assumptions about the underlying structure of sentential complements⁵.

5 As far as the underlying structure of sentential complements is concerned, Emonds claims that his proposal is also in another respect more adequate than Rosenbaum's. Within Rosenbaum's framework, sentential complements are dominated by NP in underlying structure, but Emonds argues that sentential complements in no way act like real noun phrases. So. according to him, there is no evidence that sentential complements are dominated by NP in some stage of the derivation. We return to this point in section 2.2. Compare Higgins (1973) for relevant discussion

1.3. Objections to Emonds' theory on extraposed sentences.

Although it is clear that Emonds' theory on extraposed sentences has some advantages over Rosenbaum's proposal, several objections against his treatment can be raised.

1.3.1.

Emonds' account of the distribution of sentential complements relies heavily upon an extensive use of a supposed coreferentiality relation between pronominal traces and sentential complements. These coreferentiality relations differ from other such relations. Some conditions on coreferentiality patterns can be stated in terms of primacy relations. One of the implications of these conditions is, that if the antecedent and the anaphoric element command each other, the antecedent must be to the left of the anaphoric element. It is easy to see that the coreferentiality relations between pronominal traces and sentential complements as described by Emonds, violate these condition. Such coreferentiality relations are totally unrelated to other coreferentiality relations and they must be accounted for by an entirely new mechanisme which seems to be ad hoc.

1.3.2.

The way Emonds makes use of coreferentiality relations between pronominal traces and sentential complements is also unsatisfactory in another respect: one of the consequences of Emonds' theory is that at least some syntactic rules must be formulated in terms of coreferential indices. As Emonds points out himself, Subject Replacement must be formulated as a rule which substitutes an S for a pronominal element in subject position, with which the S is coreferential. The use of coreferential indices is also essential in Pronoun Deletion: in order to assure that only pronominal elements are deleted which are coreferential to sentential complements in the tree, it is also necessary for Emonds to formulate Pronoun Deletion in terms of coreferential indices. There is not much evidence however that transformations must be able to refer to such indices. The rule Emonds cites as support for his approach is Equi NP-deletion, but it is not obvious at all that this rule must be stated in such a way⁶I claim that Emonds' theory is suspect in that it forces him to postulate such rules⁷⁸

6 For example, Chomsky (1971) replaces Equi NP Deletion by a general rule of Pro Deletion, and as far as I understand him, this rule need not refer to coreferential indices.

7 Emonds points out elsewhere that the (Extended) Standard Theory allows transformations making use of information about coreference (Emonds (1973)).

8 Higgins (1973) argues that the difficulties with Subject Replacement - granted that transformations may mention coreferential indices - are so serious that Rosenbaum's Extraposition should be readopted. He does so partly on the basis of facts from English which correspond more or less to the facts from Dutch, I will present later on.

1.3.3.

The fact that Emonds can only account for the distribution of sentential complements, connected to bisentential verbs, on the basis of the concept ‘doubly filled

[p. 167]

node', makes his theory dubious since the introduction of such a concept requires a drastic change in linguistic theory, whereas the amount of evidence for the necessity of this concept is nearly nihil⁹

9 I won't repeat Emonds' argument for the introduction of doubly filled nodes, since it requires the exposition of a lot of irrelevant matters. I refer the reader to Emonds (1970), 64-65, and leave it to his own judgement whether my statement above is fair, or not.

1.3.4.

Emonds notices that Subject Replacement fits into a class of preposing root transformations. He observes that there is a restriction on the number of preposing root rules that can apply in one root S: no more than one such rule can apply in a root. For example, the rules Negated Constituent Preposing and Topicalization are preposing root transformations. Compare the following sentences

(28)	a	zij zou die auto nooit gekocht hebben
	b	nooit zou zij die auto gekocht hebben
	c	die auto zou zij nooit gekocht hebben

Negated Constituent Preposing and Topicalization have applied in the derivation of (28)b and c, respectively. If we apply both rules in the same derivation, we get bad results, compare,

(29)	a	*nooit zou die auto zij gekocht hebben
	b	*die auto zou nooit zij gekocht hebben^10

Subject Replacement belongs to the class of preposing root rules which are subjected to the preposing restriction. Compare,

(30)	a	het heeft de jongen nooit geërgerd, dat hij de vaat moest doen
	b	dat hij de vaat moest doen, heeft de jongen nooit geërgerd (Subject Replacement)
	c	*nooit heeft, dat hij de vaat moest doen, de jongen geërgerd (Subject Replacement and Negated Constituent Proposing)
	d	*de jongen heeft, dat hij de vaat moest doen, nooit geërgerd (Subject Replacement and Topicalization)^11

Emonds notices that there is a weak point in his statement of the restriction. Wh Movement also belongs to the class of rules which are subject to the condition, but this rule is not a root transformation. Emonds points out that he does not attempt to give a principled explanation for the restriction on preposing rules. If it turns out to be impossible, however, to modify his theory in such a way as to provide such an explanation, it is inadequate at this point.

The fact that Wh Movement also undergoes the preposing restriction, indicates that it is not a constraint on root transformations. If we want to find what is going on here, we will have to look for some property, the relevant rules share. For nearly all the rules, Emonds lists, the assumption can be made, that they move elements into complementizer position. If these rules are complementizer insertion rules, it is only possible to move an element into a complementizer position which is already filled, by admitting that more than one constituent can occupy the same phrase structure position. Since, to my mind, Emonds' evidence for the necessity of allowing doubly filled nodes in certain stages of a derivation (not in surface structure) is rather weak, I would propose a new condition which claims just the opposite, prohibiting that one phrase structure position can be occupied

10 We have applied Verb Second in these derivations in such a way that the verb is placed between the two fronted constituents, otherwise even worse sentences would result:

(i)	*nooit, die auto zou zij gekocht hebben
(ii)	*die auto, nooit zou zij gekocht hebben

11 Cf. Emonds(970), 83-85 for an exhaustive list of relevant preposing rules and some more relevant (English) examples.

[p. 168]

by more than one constituent. Such a condition - I will refer to it as the Condition on Doubly Filled Nodes seems quite natural to me, and it could account for the restriction on preposing rules mentioned above¹²¹³

This explanation for the restriction on preposing rules¹⁴ can not be given within Emonds' framework without certain modifications, for I have stated earlier that nearly all the rules listed by Emonds can be regarded as complementizer movements. There is one exception, i.e. Subject Replacement which moves an element into subject position. So, the explanation for the preposing restriction, we suggest here, can only be given in a framework in which an analysis making use of Subject Replacement is replaced by an analysis making use of a complementizer movement rule.

12 Cf. Higgins (1973), 152 fn. 5, for similar remarks.

13 Note that, if true, such a Condition on Doubly Filled Nodes would rule out Emonds' account for the distribution of sentential complements connected to bisentential verbs.

14 If these rules turn out to be complementizer substitution rules instead of insertion rules, the explanation runs as follows: it is impossible to move a constituent into a complementizer node which is already filled without violation of an independently motivated condition, i.e. the Condition of Recoverability of Deletion (see Chomsky (1964)). It makes no difference for the matters under duscussion whether these rules are insertion or substitution rules; what is crucial, is, that they are complementizer movement rules.

1.3.5.

The acceptance of Subject Replacement also prevents other generalizations. Since Emonds generates sentential complements at the end of the VP, he needs Subject Replacement in order to derive sentences like

(31)	a	dat Maarten niet van koekjes houdt, ergert Sies
	b	dat Nixon fraude gepleegd heeft, is voor iedereen duidelijk

It is by no means clear, however, that these preposings should be accomplished by a rule which specifically preposes sentential subject complements, since sentential direct object complements can also be preposed, compare,

(32)	a	Karel betreurde het, dat Marietje ziek was
	b	dat Marietje ziek was, betreurde Karel
(33)	a	Karel was het vergeten, zijn moeder te bellen
	b	zijn moeder te bellen was Karel vergeten

I assume that the sentences (32)-(33) should be accounted for along the same lines. Within Emonds' framework, we need therefore a more general rule of complement preposing. If such a complement preposing is a general version of Subject Replacement, then this rule would also be a pronoun substitution rule, automatically accounting for the fact that preposing of complements is incompatible with the occurrence of pronominal forms, cf.

(34)	a	*dat Maarten niet van koekjes houdt, ergert het Sies
	b	*dat Nixon fraude gepleegd heeft, is het voor iedereen duidelijk
	c	*dat Marietje ziek was, betreurde Karel het
	d	*zijn moeder te bellen, was Karel het vergeten^15

For sentences like (31)-(33), I see then two possible derivations from underlying structures with sentential complements at the end of the VP:

(i) - application of generalized Subject Replacement

- application of Topicalization of the sentential complement

Under these assumptions, we get derivations like,

(35)	a	Karel het betreurde, dat Manetje ziek was
			⇓generalized Subject Replacement

15 I fail to see how these facts can be explained in a non-ad hoc way under the assumption that the generalized complement preposing rule is not a pronoun substitution rule. Calling upon Pronoun Deletion is of no help, compare also Higgins (1973), 158.

[p. 169]

	b	Karel, dat Marietje ziek was, betreurde
			⇓Topicalization + Verb Second
	c	dat Marietje ziek was, betreurde Karel

After application of generalized Subject Replacement, Topicalization must apply Obligatory, otherwise we could derive from underlying structures like,

(36)	Karel het was vergeten, zijn moeder te bellen

sentences like

(37)	* Karel was, zijn moeder te bellen, vergeten

Since Topicalization is optional in all the other cases, we must express this restriction on the application of these rules by means of a global rule. There is no convincing evidence, however¹⁶, that linguistic theory must be extended in such a way as to incorporate global devices of this kind: therefore, this solution must be rejected.

(ii)- application of a rule which moves pronominal elements to the front of a sentence

- application of generalized Subject Replacement¹⁷

Under these assumptions, we get derivations like.

(38)	a	Karel het betreurde, dat Marietje ziek was
			⇓fronting of the pronominal element
	b	het Karel betreurde, dat Marietje ziek was
			⇓generalized Subject Replacement + Verb Second
	c	dat Marietje ziek was, betreurde Karel

The problem with this alternative is that there is no independent motivation for the existence of a rule which preposes pronominal elements. One could suggest that the rule which takes care of movements of the desired kind, is the independently needed rule of Topicalization. Unfortunately, Topicalization does never prepose pronominal elements like het:

(39)	a	Karel heeft het altijd al gezegd
	b	*het heeft Karel altijd al gezegd

Therefore, this solution should also be rejected¹⁸

16 Cf. Chomsky (1970), Baker & Brame (1972) and Emonds (1973), for relevant discussion.

17 Compare for discussion of this alternative, Higgins (1973).

18 This conclusion only holds if the restriction which takes care of the ungrammaticality of (39)b is linked up with Topicalization and should not be regarded as a surface structure constraint. Our conclusion roust be seen in the light of the lack of evidence for such a constraint.

1.3.6.

My conclusion of this section is then that a treatment of S complements in a framework which generates such complements at the end of the VP leads to an unsupported broadening of the definition of a transformational rule, and furthermore that such an approach misses certain significant generalizations about S preposings.

2.0. A new proposal.

My claim is that all the objections to Emonds' analysis of extraposition

[p. 170]

mentioned in 1.3. can he traced back to his assumption that sentential complements should he generated at the end of the VP in deep structure. We will show in this section that such objections cannot be raised against a framework in which we maintain - with certain modifications - Rosenbaum's original analysis, i.e. if we readopt a rule of Extraposition. Of course, this proposal must meet the following problems in a non-ad hoc way:

(i)	distributional properties of complement sentences
(ii)	an account of the restriction on complementizer movement rules
(iii)	the NP or non-NP character of sentential complements
(iv)	the treatment of pronominal elements, related to sentential complements
(v)	the (non) use of information about coreference in transformational rules

Then, there is at least one additional problem connected to a proposal making use of a rule of Extraposition:

(vi)	Emonds has developed his theory on extraposed sentences against the back ground of an attempt to constrain the choice of possible transformations. He proposes that there are three kinds of transformational rules:root transformations, structure preserving transformations and minor movement rules. Root transformations attach constituents only to root S's (Emonds (1970),
24).	Transformations which are structure preserving have output structures which satisfy the phrase structure rules: they preserve the structure provided for by phrase structure rules. We disregard minor movements, here.

I have not brought the Structure Preserving Hypothesis up earlier, since Emonds claims that the controversy between Rosenbaum and him about extraposed sentences is not crucial for his hypothesis: according to Emonds, Subject Replacement is a root transformation and Rosenbaum's Extraposition is a structure preserving transformation. Hence, both approaches fit into the structure preserving framework. Higgins (1973), however, points out that it is not at all obvious that both approaches are in general compatible with the Structure Preserving Hypothesis. More specifically. he argues on the basis of facts of German syntax that it is very unlikely that a transformational approach to extraposed sentences ‘can be fitted into the structure preserving framework’ (Higgins 1973), 183).His argument goes roughly as follows: Emonds has shown that one of the consequences of the Structure Preserving Hypothesis is, that languages like German (and by the same token Dutch) have underlying verb-final. A rule of Extraposition will move a sentential complement to the end of the VP; so, the effect of Extraposition can be schematically represented as:

[p. 171]

According to Higgins, the S node at the end of the VP is not independently motivated which means that there are no independent grounds for the postulation of a phrase structure rule of the form

(41)

VP⇢

...V S

Hence, Extraposition cannot be shown to be structure preserving, and since it must a structure preserving rule according to the Structure Preserving Hypothesis, the rule :s a counterexample to this hypothesis. Since Higgins' argument can be easily duplicated for Dutch, we will pay some attention to the relation between Extraposition and the Structure Preserving Hypothesis.

2.1. The readoption of Extraposition.

We account for the distribution of sentential complements by generating these complements in subject or object position in deep structure¹⁹and readoption of Extraposition. Compared to Rosenbaum (1967), there is one difference: we assume that Extraposition is obligatory, in all cases. We derive the problematic non-extraposed sentences - Emonds' Subject Replacement cases - by the more general rule of Topicalization. This rule enables us to give a natural explanation for the restriction on certain preposing transformations, since Topicalization can be assumed to be a complementizer movement rule. Consequently, if we derive sentences like

(42)	a	dat Karel niet gekomen is, verbaast me
	b	dat Marietje jongens plaagt, is bizonder vervelend

with the help of Topicalization, we can explain why clause-initial position of sentential complements is incompatible with complementizer movements like Wh Movement, by calling upon the Condition on Doubly Filled Nodes. The fact that Topicalization of sentential direct objects is also mutually exclusive with the other complementizer movement rules, compare,

(43)	a	Karel heeft deze jongen beloofd, dat hij de Notre Dame gaat schilderen
	b	wie heeft Karel beloofd, dat hij de Notre Dame gaat schilderen (Wh Movement)
	c	cal hij de Notre Dame gaat schilderen, heeft Karel de jongen beloofd (Topicalization)
	d	* wie heeft, dat hij de Notre Dame gaat schilderen, Karel beloofd (Wh Movement - Topicalization)
	e	* dat hij de Notre Dame gaat schilderen, heeft wie Karel beloofd (Topicalization Wh Movement)

confirms this proposal²⁰

We have seen in section 1.3.5. that preposing of sentential complements is incompatible with the presence of pronominal elements related to these complements. This means in our proposal that only complement sentences in non-extraposed position may be topicalized. This can be achieved by ordering Topicalization before Extraposition. But there is a problem, here. Since we have proposed to make Extraposition obligatory, Extraposition would apply after Topicalization, and we would never be able to derive sentences like

(44)	da: Harry getrouwd is, verbaast me

19 Compare however section 2.3. and fn. 31.

20 An additional piece of evidence for an approach which selects Topicalization over Subject Replacement is that Topicalization is a rule which makes no use of coreferential indices. Our conclusion that an analysis which uses a rule, not referring to such indices, must be preferred over an analysis which permits this device, seems to be quite natural in the light of our claim that analysis of the last kind are generally suspect.

[p. 172]

This problem can be overcome by making the very natural assumption that Extraposition of topicalized complements must be prevented. We can achieve this, if we formulate Extraposition as follows:

(45)	Extraposition (preliminary version)
	COMP	X	S	Y	condition: obligatory
	1	2	3	4
				====⇒
	1	2	Ø	4+3

Disregarding irrelevant details, we derive now(44) through application of Topicalization, without subsequent Extraposition, from an underlying structure like

(46)	COMP [dat Harry getrouwd is]S verbaast me

I note, that although the surface structure of sentences like (44) looks a lot like their deep structure, our description does justice to the intuition that such sentences are stilistically marked, by deriving them with Topicalization - a rule which always has a stilistic effect.

2.2. The underlying structure of sentential complements under a transformational analysis of extraposed sentences.

Rosenbaum (1967) assumes that every sentential complement has the underlying structure [het - S]_NP: so, he generates pronominal elements related to sentential complements in the base, and provides for the correct distribution by Pronoun Deletion. I am not aware of any convincing evidence for such an analysis; on the contrary, it is an implausible analysis, if one thinks of what the internal structure of [het - S]_NP would look like. One may think of structure like [[het]_NPS]_NP, a structure quite generally proposed for noun phrases with relative clauses, or [[het]_NS]_NP, a structure sometimes postulated for noun phrases with nouns like feit, bewering as heads followed by a sentential complement. Under both analyses, it is assumed that the pronoun is the antecedent of the complement sentence, but as is pointed out by Williams (1972), 386, legitimate antecedents generally cannot be deleted, and it is therefore very unlikely that such pronominal elements which can be deleted and sometimes must be absent, should be regarded as such

An alternative approach could be, that pronominal elements are inserted after the application of Extraposition into the position originally occupied by the sentential complement. In this approach, the pronoun can be viewed of as a certain kind of place-marker. The reason why a Pronoun Deletion analysis is favored over a Pronoun Insertion analysis by some linguists, for instance Lakoff (1968), not withstanding the lack of evidence for an underlying configuration like [het - S]_NP, is that the proper environments for the deletion of pronominal elements can be quite easily formulated, whereas it is not clear at all that this can be done with respect to insertion of such pronouns (cf. Lakoff (1968), 17). Since Lakoff does not really try to formulate a rule of Pronoun Insertion, as to make his objections against such a rule explicit, his statement is without content and does not

[p. 173]

reach beyond the level of suggestion. There are as far as I know no arguments against a rule of Pronoun Insertion.

If we assume that sentential complements are underlying [S]_NP, then we are confronted with an empty NP after the application of Extraposition, since Extraposition moves only the S constituent to the end of the sentences and leaves the NP node behind. Pronoun Insertion can be regarded now as a rule which fills such empty NP nodes.

As we have already noted, the occurrence of pronominal elements is obligatory for extraposed sentential subjects and prepositional objects, and optional in connection with extraposed sentential direct objects. I suggest that this obligatory- optional condition can be related to the notion obligatory-optional node, introduced by Emonds (1970). According to Emonds, an obligatory node differs from an optional one in that it must be chosen in deep structure. His definition of the dichotomy obligatory-optional node intends to imply that a subject NP is an obligatory node, a direct object NP an optional one, and although the phrase structure choice of the node PP is in general optional, the object NP of certain prepositions is again an obligatory node. Pronoun Insertion can then be formulated as follows:

(47)	Pronoun Insertion^21
	X	[△]NP Y			condition: α OB, if NP is
	1	2	3		a α OB node
				====>
	1	het	3

We have given a simplified version of the occurrence of pronominal elements connected with direct object sentences. Not only there are verbs which can get sentential direct objects optionally accompanied by pronominal elements, but there are also verbs, like beweren, geloven, zeggen etc., which can get sentential direct objects normally never accompanied by pronominal elements. The analysis we are defending takes the occurrence of a pronoun het, connected to a sentential complement, as an indication that such a complement is dominated by NP in deep structure. Consequently. we must assume that sentential complements to verbs like beweren, are not dominated by NP in deep structure. Our approach differs essentially from Emonds': he claims that there is no evidence for the hypothesis that sentential complements are NP in some stage of the derivation. I claim that the fact that some sentential complements are accompanied by a pronominal element, indicates that those complements are underlying NP, whereas complements never connected with such a pronoun are not NP's, but just plain S's in deep structure. So we subcategorize verbs like beweren for a plain S complement: Extraposition applies to such complements without leaving behind an empty NP, therefore Pronoun Insertion is not applicable, and the absence of a pronoun het is explained²² We regard the optional or obligatory presence of a pronominal element as a first piece of evidence for the NP status of some sentential complements.

A second argument for a distinction between non-NP and NP sentential complements can be found with respect to topicalized sentences. It is generally assumed that Topicalization applies to PP and NP constituents and not to S constituents²³

Complements which can be accompanied by a pronominal element can also be

21 Emonds allows the occurrence of empty nodes. ‘Rather, I require only that a node dominates terminal elements at some point in a transformational derivation’ (Emonds (1970), 28). It is possible now, that we can drop the condition on Pronoun Insertion by adding to Emonds' restriction on empty nodes a surface structure condition, certainly language-specific, to the effect that obligatory (optional) nodes must (may) be filled in surface structure.

22 Compare however section 2.3. and fn. 31.

23 Assuming that Topicalization also applies to PP's, we are confronted with a problem, since only Topicalization of subject and direct object NY's dominating S is allowed. Topicalization of prepositional objects of the same nature must be precluded, as the following example illustrates:

(i)

op dat jij zou komen, heb ik gerekend
I have no explanation for the ungrammaticality of these sentences, and I propose for the moment to put a condition on the rule in order to block such derivations. Topicalization can then be formulated as

(ii)

Topicalization

COMP	X	{NP}	Y	conditions:
COMP	X	{PP}	Y	(i) optional
1	2	3	4	(ii) PP ≠ Prep - S
			=====>
1<3	2	0	4

Condition (ii) is ad hoc and this may indicate that this approach is somewhere wrong. This has no effect, however, on our evaluation of some of the ideas of Emonds, since as far as I can see he has to formulate an ad hoc condition too, for the same cases. Note, that given (ii) it is also possible to topicalize [S]_NP out of the configuration [Prep - [S]_NP]_PP. We would derive then: (iii)^*dat jij zou komen, heb ik op gerekend The ungrammaticality of (iii) can be accounted for in a less ad hoc way by means of a constraint which is not exclusively tighed up to Topicalization, and which prohibits under certain conditions dangling prepositions in Dutch, compare De Haan/ Koefoed/Des Tombe (1974), 131-134 and 137. Cf. also Ross (1968), for relevant discussion of such language-specific preposition constraints.

* This paper is part of a larger study that will be published elsewhere. It is not checked by a person who is familiar with the English language. Therefore, no claim is made with respect to its readability.

[p. 174]

topicalized; complements which cannot be accompanied by a pronominal element, cannot be topicalized, cf.

(48)	a	Karel betreurde (het), dat Marietje ziek was
	b	dat Marietje ziek was, betreurde Karel
(49)	a	Jan was (het) vergeten, dat hij zijn moeder moest bellen
	b	dat hij zijn moede: moes: beilen, was Jan vergeten
(50)	a	Maarten beweerde (*het), dat Sies gek was
	b	*dat Sies gek was, beweerde Maarten
(51)	a	Alice gelooft (*het), dat stenen kunnen praten
	b	* dat stenen kunnen praten, gelooft Alice^24

This approach also explains why Topicalization of sentential complements is incompatible with the occurrence of pronominal forms related to these complements. If we topicalize a sentential complement, dominated by NP, the whole NP constituent is preposed. Since subsequent Extraposition is precluded, there is no empty NP node left, and Pronoun Insertion cannot apply. This explains why sentences like

(52)	a	*dat Karel gek geworden was, betreurde ik het
	b	*dat Karel niet gekomen is, is het heel goed mogelijk

are bad, and why sentences like

(53)	a	dat Karel gek geworden was, betreurde ik
	b	dat Karel niet gekomen is, is heel goed mogelijk

are grammatical.

A third argument has to do with the occurrences of the pronoun er and passive sentences. Compare the following sentences:

(54)	a	er liep een jongen op straat
	b	* er liep de jongen op straat
(55)	a	er weerklonken gisteren bij de Amerikaanse ambassade anti-CIA leuzen
	b	*er weerklonken gisteren bij de Amerikaanse ambassade de anti-CIA leuzen

Following Emonds (1970), 35-36, we can say that a rule which moves an indefinite subject NP to the right, has applied in the derivation of (54)a and (55)a; the b examples illustrate the necessity of the indefiniteness restriction. Apparently, this rule of Indefinite Subject NP Movement leaves a pronominal trace, er, behind. We have also the following sentences in Dutch:

(56)	a	de hond beet de man
	b	de man werd door de hond gebeten
	c	* er werd de man door de hond gebeten
	d	er werd door de hond gebenten
(57)	a	de kinderen dansten op het plein
	b	er werd door de kinderen op het plein gedanst

24 Complements to verbs like geloven and beweren can be topicalized, if the clause in which these verbs occur, receives heavy stress, compare

(i)

dat Sies gek was, bewéérde Maarten

This fact does not constitute counter-evidence,since under this condition such complements can Also be accompanied by a pronominal element.

(ii)

Maarten bewéé het. dat Sies gek was

Cf also for such sentences V d Hoek (1971)

[p. 175]

Observe that the pronoun er can also occur in some passive sentences whereas these occurrences of er don't result from the application of Indefinite Subject NP Movement, since all the NP's in the examples above are definite. Assuming that the passivization phenomenon can be decribed with the rules of Agent Postponing and Direct Object Preposing²⁵ which apply only to NP's, we note that Agent Postponing has applied in the derivation of (56)d and (57)b, without subsequent Direct Object Preposing, and that the sentences (56)b and c are derived with the application of both rules. It seems to be the case that the occurrence of er in passives is only compatible with the application of Agent Postponing, and not with both Agent Postponing and Direct Object Preposing. If we assume, following again essentially Emonds (1970), that Direct Object Preposing moves an object NP into subject position, we can attempt to generalize over the two cases of er-occurrences discussed sofar in the following way: the element er occurs in subject position whenever a subject NP is postponed without leaving behind an empty NP node, and this subject position is, not re-filled again.

Corresponding to the active sentences

(58)	a	iedereen beweerde, dat Karel ziek was
	b	iedereen betreurde (het), dat Karel ziek was

we have the following passive forms

(59)	a	er werd door iedereen beweerd, dat Karel ziek was
	b	het werd door iedereen betreurd, dat Karel ziek was

but not

(60)	a	‘het werd door iedereen beweerd, dat Karel ziek was
	b	‘er werd door iedereen betreurd, dat Karel ziek was

The sentences with the verb betreuren can be easily accounted for within our proposal, since complements to such verbs, accompanied (optionally) by het, are dominated by NP in deep structure. We derive (59)b along the following lines:

(61)	a	iedereen [[dat Karel ziek was]S]NP betreurde
			⇓ Agent Postponing + Direct Object Preposing
	b	[[dat Karel ziek was]S]NP werd door iedereen betreurd
			⇓ Extraposition - Pronoun Insertion
	c	het werd door iedereen betreurd, dat Karel ziek was

Direct Object Preposing applies in this derivation, so there is no way in which the element er cart occur, and correspondingly, we never generate sentences like (60)b.

The occurrence of er in passives like (59)a can be explained in the same way as the occurrences of er, mentioned above, if we can make it plausible that a subject constituent is moved to the right in this derivation and this position is not filled again. We can achieve this by making use of the assumption that Direct Object Preposing only applies to NP constituents and that complements of verbs like beweren are not dominated by NP in deep structure; this implies that the derivation of (59)a will proceed as follows without application of Direct Object Preposing:

25 As is suggested in Chomsky (1967).

[p. 176]

(62)	a	iedereen [dat Karel ziek was]S beweerde
			⇓ Agent Postponing
	b	[dat Karel ziek was]S werd door iedereen beweerd
			⇓ filling of an empty subject position with er
	c	er [dat Karel ziek was]S werd door iedereen beweerd
			⇓ Extraposition^26
	d	er werd door iedereen beweerd, dat Karel ziek was

Since the complement of beweren is not dominated by NP. it is impossible to derive a sentence in which the pronoun het is left as a place-marker after the application of Extraposition: under these assumptions, we will never derive a sentence like (60)a. This result is correct. Hence, the assumptions that certain sentential complements are dominated by NP in deep structure, while others are not, enables us to explain the differences between sentences like (58)-(60) in a way which fits quite natural in a general approach to the distribution of the pronoun er²⁷

26 Compare section 2.3. and fn. 31.

27 V.d. Hoek (1971) tries to account for the differences between sentences like (58)-(60) by assigning to sentential complements a feature + or - definite. He assumes then, that Indefinite Subject NP Movement has applied in the derivation of (59)a, as opposed to the derivation of (58)a. Since 1 have shown that the occurrence of er in passives has not necessarily something to do with the indefinite status of the constituents involved, his approach is inadequate. Or to put it differently, V.d. Hoek does not relate the occurrences of er in (56), (57) to the occurrences of er in (54), (59) and (60), therefore his approach is not general enough.

2.3. Extraposition and the Structure Preserving Hypothesis.

Until sofar, we have neglected a problem with the formulation of the rule of Extraposition. We have assumed that Extraposition applies to all sentential complements, functioning as subject or object regardless whether they are dominated by NP or not. In this section, we will argue that there is some evidence that this assumption is not true.

We have proposed Extraposition of plain S and Extraposition of S from under the domination of NP. We have to formulate the structural description of Extraposition in such a way as to cover both cases. It appears at first sight that this can be done without difficulties by the formulation of Extraposition we have given under (45).

In addition to this rule of Extraposition, there is also a rule called Extraposition from NP which postpones sentential complements to noun phrases. This rule relates sentences as the following:

(63)	a	Jan heeft altijd al van meisjes die rood haar hebben gehouden
	b	Jan heeft altijd al van meisjes gehouden, die rood haar hebben
(64)	a	de jongen zal het verhaal dat jij mij verteld hebt, wel ontkennen
	b	de jongen zal het verhaal wel ontkennen, dat jij mij verteld hebt

Without investigating this phenomenon in any detail, we can easily see that the two extraposition rules should not be collapsed. Observe for instance, that Extraposition from NP is optional, while Extraposition is obligatory. Furthermore, Extraposition from NP is highly restricted in Dutch in a way which Extraposition is not. Some of the restrictions on Extraposition from NP can be demonstrated by the following examples

(65)	a	de jongen die rood haar had, was erg ziek
	b	*de jongen was erg ziek, die rood haar had
(66)	a	ik heb het boek aan het meisje dat aan de overkant woont gegeven
	b	ik heb het boek aan het meisje gegeven, dat aan de overkant woont

[p. 177]

(67)	a	ik heb het meisje dat aan de overkant woont, dit boek gegeven
	b	*ik heb het meisje dit boek gegeven, dat aan de overkant woont

The problem with the way, we have formulated Extraposition under (45), is that the structural description makes the rule applicable on Extraposition from NP- configuratuion, obligatorily postponing such complements. This leads to wrong results: we derive sentences like (65)b and (67)b, and we are no longer able to derive sentences like (65)a, (66)a and (67)a. We must therefore modify the structural description of Extraposition in such a way, that the scope of the rule will exactly cover the cases, it is introduced for.

If we assume that sentential complements, which are not dominated by NP, are generated at the end of the VP in deep structure, then Extraposition only applies to sentential complements exhaustively dominated by NP. The structural description of Extraposition would then be something like, X - [S]_NP - Y, and this formulation would suffice to discriminate between Extraposition and Extraposition from NP. We express the difference between verbs which get sentential complements at the end of the VP, and verbs which do not, with the help of subcategorization features. Verbs like beweren, zeggen etc. are subcategorized as [+ - S] and verbs like betreuren as [-NP-].

This partial readoption of Extraposition would still solve the problem we have encountered in Emonds' treatment, since it were exactly the sentential complements with obligatory or optional pronominal traces in surface structure - or, in our terminology, the sentences by NP in deep structure - which required the readoption of Extraposition. So we change (45) into

(68)	Extraposition
	COMP	X	[S]NP	Y	condition: obligatory
	1	2	3	4 =====>
	1	2	Ø	4+3

It will be clear, that we need now a phrase structure rule of the form

(69)

VP⇢...V

(S)^28

The distribution of sentential complements to bisentenuial predicates follows now as a natural consequence from the approach outlined above, and not only that, but also some of the interactions between Extraposition and Extraposition from NP can be explained in a very natural way. Since there is only one S node at the end of the VP available according to(69),it follows from the Structure Preserving Hypothesis that only one sentential complement can extrapose, unless one assumes the possibility of doubly filled nodes. We have already claimed that this last possibility must be ruled out: the structure preserving status of extraposition rules together with the Condition on Doubly Filled Nodes restricts the number of possible applications of extraposition rules to exactly one. Compare the following underlying structure:

(70)	COMP [[dat Jan zij neus rood was] S]NP mij [[van]Prep[[dat hij dronken was]S]NP]PPovertuigde

28 If this argument can be duplicated for German, Higgins' argument against the structure preserving character of Extraposition in German no longer holds.

[p. 178]

If Topicalization applies, Extraposition can apply only once, and we will derive correctly,

(71)	dat Jan zijn neus rood was, overtuigde mij ervan, dat hij dronken was

If Topicalization does not apply, Extraposition must apply twice: since we have established that meta-conditions restrict the number of possible applications of extraposition rules to one, the derivation will block, because of an independently motivated filter which says that if an obligatory rule cannot apply because of a meta-condition, the derivation will block²⁹ We predict correctly that the sentence

(72)	*het overtuigde mij ervan, dat hij dronken was, dat Jan zijn neus rood was

is ungrammatical. Hence, derivations of sentences with bisentential predicates only result in wellformed surface structures, if one of the complements can, and is, topicalized. Since Topicalization is a root transformation, we predict again correctly that sentences with bisentential predicates cannot occur as non-root S's: compare,

(73)	a	*Karel dacht dat da: Jan zijn neus rood was, mij ervan overtuigde, dat hij dronken was
	b	*Karel dacht, dat het mij ervan overtuigde dat hij dronken was, dat Jan zijn neus rood was

Since Topicalization is imcompatible with the rule which provides for direct questions - cf.

(74)	a	slachtte de boer het varken?
	b	*het varken slachtte de boer?
	c	*slachtte het varken de boer?^30

we predict that sentences with bisentential predicates cannot occur as direct questions, which is correct:

(75)	a	*overtuigde, dat Jan een rode neus had, mij ervan, dat hij dronken was?
	b	*overtuigde het mij ervan, dat hij dronken was, dat Jan een rode neus had?

Extraposition and Extraposition from NP interact along the following lines: compare the following underlying structure:

(76)	COMP dit verhaal [[het meisje]NP[dat rood haar had]S]NP heeft verbaasd

Extraposition from NP can apply to this structure, resulting after Verb Second in

(77)	dit verhaal heeft het meisje verbaasd, dat rood haar had

The verb verbazen can also have a sentential subject: suppose, we have the following underlying structure:

(78)	COMP[[dat jij ziek was]S]NP[[het meisje]NP[dat rood haar had]S]NP heeft verbaasd

If Topicalization does not apply, Extraposition must apply to the subject sentence, and consequently the position of the S at the end of the VP is filled.

29 Cf. Perlmutter (1971).

30 Since Direct Question Formation which is a verb fronting rule in Dutch, is incompatible with all complementizer movement rules, it is very attractive to assume that it is also a complementizer movement rule, so that it falls under the scope of the explanation we have given above.

[p. 179]

According to our conditions, Extraposition from NP is ruled out now, which is indeed correct as the following ungrammatical sentence shows:

(79)	*het heeft het meisje verbaasd dat jij ziek was, dat rood haar had

Since Extraposition from NP is an optional rule, we get a correct result, if we choose not to apply this rule, compare,

(80)	het heeft het meisje dat rood haar had, verbaasd, dat jij ziek was

In this light, it will become clear that the attractiveness of our solution to the problem, how to prevent Extraposition from applying to Extraposition from NP- cases, is that it lends some support to the Structure Preserving Hypothesis, especially with regard to the rules of extraposition³¹ It is precisely the structure preserving character of these extraposition rules which enables us to account for the distribution of sentential complements to bisentential predicates, and also of sentential complements to noun phrases, without calling upon such ad hoc constraints as the Internal S Condition, or the concept of doubly filled nodes.

31 If the discrimination between Extraposition and Extraposition from NP - cases follows as a natural consequence from independent factors in the grammar (for example, independently motivated conditions on transformations as discussed in Chomsky (1971) and De Haan (1972)), then nearly all the claims of this section are unsupported. Perhaps, we will have to call upon some sort of a surface structure condition in order to explain the restrictions on extraposition rules discussed above.

3.0. Summary.

We have compared the theories of Rosenbaum (1967) and Emonds (1970) on extraposed sentences. Since both theories are inadequate, we have outlined a new approach. This proposal includes the following points:

- an account of the distributional aspects of sentential complements, consisting of

(i)	an obligatory structure preserving rule of Extraposition,
(ii)	a rule of Topicalization of sentential complements, providing for the occur rence of sentential complements in the clause-initial position of root S's,
(iii)	a general Condition on Doubly Filled Nodes, taking care of a restriction on a certain class of preposing rules, and - in corporation with the Structure Preserving Hypothesis - of the fact that only one S can extrapose in a given clause,
(iv)	a subcategorization of verbs for (non) NP status of sentential complements, also covering a distinction between VP-final and VP-nonfinal position of such complements

- an account of the distribution of pronominal elements related to sentential complements, consisting of

(v)	a Pronoun Insertion rule which is closely tighed up with a hypothesis about the underlying (non) NP status of sentential complements and the notion obligatory - optional node,
(vi)	a rule order Topicalization - Extraposition^32

I think that it is fair to conclude that our proposal meets all objections, Emonds has raised against Rosenbaum's (and partly Ross') theory on extraposed sentences: in particular, we don't make use of ad hoc mechanismes like OB/OP conditions on Extraposition, or the Internal S Condition. Furthermore, our theory is clearly superior to Emonds' proposal in that we can account for the distri-

32 A theory incorporating a rule of Extraposition always run into the problem whether this rule is cyclic or not. Although I don't want to go into this matter here, I would like to stress that the assumption that Topicalization must be ordered before Extrapositiondoes not imply that Extraposition is not cyclic: even if we abandon - which I do - the possibility that Topicalization is a last- cyclic rule, a cyclic nature of Extraposition can be defended by assuming that Topicalization is successively cyclic in the sense of Chomsky (1971).

[p. 180]

bution of sentential complements without calling upon ad hoc devices like doubly filled nodes; that the transformations we have postulated don't make use of information about coreference; that we don't have to make strange assumptions about coreferentiality between pronominal elements and sentential complements, and that we can give a natural explanation for the restriction on a certain class of preposing rules.

[p. 182]

Bibliography

Baker, C.L. and Brame, M.K. (1972):‘Global Rules’: a Rejoinder’. In Language XLVIII, 51-75.

Chomsky, N. (1964): Current Issues in Linguistic Theory, Den Haag.

Chomsky, N. (1970): Some Empirical Issues in the Theory of Transformational Grammar. IULC paper.

Chomsky, N. (1971): Conditions on Transformations. IULC paper.

Emonds, J. (1970): Root and Structure-Preserving Transformations. IULC dissertation.

Emonds, J, (1973): ‘Alternatives to Global Constraints’. In Glossa VII, 39-62.

Haan, G.J. de (1972): ‘On Transparancy’. In A. Kraak (ed), Linguistics in the Netherlands, Assen, to appear.

Haan, G.J. de/ Koefoed, G.A.T./ Tombe, A.L. des (1974): Basiskursus Algemene Taalwelenschap, Assen.

Hoek, Th van den (1971): ‘Opmerkingen over zinscomplementatie’. In Studia Neerlandica 7, 189-215.

Higgins, F.R. (1973): ‘On J. Emonds's Analysis of Extraposition’. In J.P. Kimball (ed.),.Synlax and Semantics, Volume2, New York.

Lakoff, G. (1968): Deep and Surface Grammar. IULC paper.

Perlmutter, D.M. (1971): Deep and Surface Structure Constraints in Syntax, New York.

Rosenbaum, P.S. ( 1967): The Grammar of English Predicate Complement Constructions, Cambridge (Mass.).

Ross, J.R. (1968): Constraints on Variables in Syntax. IULC dissertation.

[p. 183]

Satyanarayana, P. (1973): ‘Why Saying that Sentences Like This Are Unacceptable Is Wrong’. In C. Corum, T. Smith-Stark, and A. Weiser (eds.), Papers from the Ninth Regional Meeting of the Chicago Linguistic Society, Chicago, Illinois, 568-576.

Williams, G. (1972): ‘The Internal Structure of Antecedents’. In P. Peranteau, J. Levi, and G. Phares (eds.), Papersfrom the Eigth Regional Meeting of the Chicago Linguistic Society, Chicago, Illinois.

* This paper is part of a larger study that will be published elsewhere. It is not checked by a person who is familiar with the English language. Therefore, no claim is made with respect to its readability.