Discourse constraints on (non)extraposition from subject in English(*).

MILLER, PHILIP H.

Abstract

This paper analyzes the discourse conditions governing the choice between extraposition and nonextraposition of that-clause and infinitival-VP subjects. On the basis of a large corpus of naturally occurring data, it is shown that nonextraposition requires that the content of the subject be discourse-old or directly inferrable. If the content is discourse-new, then extraposition is necessary. The choice between extraposition and nonextraposition for discourse-old and inferrable subjects is examined and is shown to depend on the discourse status of the predicate and on whether it is the predicate or the subject that links to the following discourse. The paper ends with a discussion of the syntactic position of nonextraposed sentential subjects and concludes that it cannot be the same as that of fronted sentential complements. This means that the common discourse properties of fronting and nonextraposition must be linked to their common linear ordering properties, rather than to a common syntactic position.

1. Introduction

Extraposition of sentential subjects has been studied both in traditional grammars of English (e.g. Jespersen 1909-1949: vol. III, 2.1.3ff.; Quirk et al. 1985: 18.33ff.) and in the generative literature starting with Rosenbaum (1967) and Ross (1967).(1) However, there has been remarkably little work on the discourse conditioning of the choice between extraposed and nonextraposed variants, illustrated in (1a)-(1b) and (2a)-(2b).

(1) a. [[sub.s] That a bloodthirsty, cruel capitalist should be
 such a graceful fellow] was a shock to me. (Davis,
 The Iron Puddler)

 b. It was a shock to me [[sub.s] that a bloodthirsty,
 cruel capitalist should be such a graceful fellow].

(2) a. Yet [[sub.VP] to determine precisely to what extent
 and exactly in what ways any individual showed the
 effects of Christianity] would be impossible. (Brown
 Corpus, D140290)

 b. Yet it would be impossible [[sub.VP] to determine
 precisely to what extent and exactly in what ways any
 individual showed the effects of Christianity].

In these examples the (a) sentence has a sentential subject in the normal preverbal subject position. In the (b) sentence, the sentential subject is extraposed to sentence-final position. In the examples, the sentential subjects are italicized for clarity. Example 1 has a finite that clause as its sentential subject, example 2 has an infinitival VP. It is well known that NP subjects, on the other hand, cannot undergo extraposition.(2) This is illustrated in (3a)-(3b). Note that I am making the uncontroversial assumption that extraposition and right dislocation, as in (3c), are two entirely distinct constructions. They have many different properties, the most obvious of which is that they have different intonation patterns. (3c) is fine if it is pronounced with a sentence-final intonational contour on to me and a pause between to me and the following sentence, and the whole dislocated sentence is deaccented. The same type of intonation pattern would be inappropriate for (1b) and (2b), unless they were analyzed as instances of right dislocation rather than extraposition.

(3) a. [[sub.NP] The fact that a bloodthirsty, cruel capitalist
 should be such a graceful fellow] was a shock to me.

 b. *It was a shock to me [the fact that a bloodthirsty,
 cruel capitalist should be such a graceful fellow].

 c. Right dislocation:
 It was a shock to me, [the fact that a bloodthirsty, cruel
 capitalist should be such a graceful fellow].

Furthermore, it is systematically impossible in the case of extraposition to replace the impersonal it subject by another NP such as this, whereas this is possible (under appropriate discourse conditions) with right dislocation. This contrast is illustrated in (4a)-(4b).(3)

(4) a. Extraposition:
 *This was a shock to me [the fact that a bloodthirsty,
 cruel capitalist should be such a graceful fellow].

 b. Right dislocation:
 This was a shock to me, [the fact that a bloodthirsty,
 cruel capitalist should be such a graceful fellow].

Finally, the discourse conditions on the two constructions are different, contrary to what Lambrecht (1994: 203) suggests. Specifically, the content of a right-dislocated constituent must be discourse-old, as shown by Birner and Ward (1998: 145-150). This condition is in no way applicable to the content of the subject sentence in extraposition, as will be shown below.

The paper is organized as follows. Section 2 provides a summary of Bolinger's analysis of the discourse constraints on the choice between extraposition and nonextraposition and discusses initial counterexamples. Section 3 examines that-clause subjects and concludes with the central claim of the paper, namely that nonextraposition is only possible if the content of the clausal subject is discourse-old or inferrable. Furthermore, in that case, the choice between extraposition and nonextraposition is shown to depend on (i) the discourse status of the predicate and (ii) whether it is the predicate or the subject that provides the topic of the following discourse. Finally some comparison is provided with complex NP subjects of the type the fact that S and with fronted sentential objects. Section 4 shows that these conditions extend to the case of infinitival VP subjects. Finally, section 5 discusses the syntactic position of nonextraposed sentential subjects. The preceding analysis of the discourse conditions on nonextraposition make it possible to construct complex but plausible examples that call into question an analysis that has become more or less standard within transformational grammar, namely that of Koster (1978), according to whom sentential subjects are not in normal subject position in surface structure, but in a fronted position, similar to that of a fronted complement clause.

The paper is set within the framework developed by Prince, Ward, and Birner.(4) Following their practice, it is based on an examination of a corpus of naturally occurring examples.(5) This corpus-based approach is very useful because of its great heuristic value. It is unclear how the central thesis of the paper could have been discovered without the corpus. Furthermore, it is also useful because of the well-known difficulties involved in judging discourse conditions solely through intuition-based judgments. Of course, it is impossible -- and in fact undesirable -- to avoid all recourse to intuition. Specifically, I am assuming that within a given discourse context it is possible, to some extent at least, to give differential intuitive judgments about the appropriateness of variants, and that it is possible to judge certain variants as syntactically ill formed, such as (12b) and (13b) below.

2. Discourse conditions on extraposition and nonextraposition

While the syntax of extraposition has been widely discussed in the generative literature (see references in note 1), there has been remarkably little work on the discourse conditioning of the choice between the extraposed and nonextraposed variants. Lambrecht (1994: 203) assimilates extraposition with right dislocation. As pointed out above, it seems clear that this is an erroneous analysis. Birner and Ward (1998) do not discuss the question.

The most informative discussion of this topic is that of Bolinger (1977). He claims that the extraposed construction may be used when "the topic has been introduced" (1977: 68). Bolinger claims that the nonextraposed variant is impossible if the content of the sentential subject is topical and that sentences such as his [75] are impossible, as opposed to [76] and [77], because the anaphoric link entails topic status.

As for the extraposition of infinitives, while a pair such as

[73] To give in now would be fatal.

[74] It would be fatal to give in now.

are interchangeable in many contexts and look as if they might be in free variation, actually the it again relates to some kind of prior basis. Consider the following answers to the question What do you think of running him as a candidate?:

[75] *To do that would be a good idea.

[76] To run him as a candidate would be a good idea

[77] It would be a good idea to do that.

In [75], the use of that forces the anaphora -- the speaker has to be picking up the idea from his interlocutor and is therefore obliged to use it and the construction in [77]. But [76] is possible where the speaker is turning the question over in his mind and treating it as his own idea (Bolinger 1977: 72).

The corpus investigation on which this study is based proves that Bolinger's judgments here are not well founded. Specifically, numerous examples can be found of naturally occurring sentences with nonextraposed infinitival VPs containing anaphoric elements, as illustrated in (5).

(5) a. "So you get rid of that pistol right now, Mister McBride.
 You do that or take you out a permit right now." McBride
 couldn't do either, of course. Not immediately, as the
 deputy demanded. Not without a face-saving respite of at
 least a few minutes. To do so would make his job well-nigh
 impossible (Brown N09 1310).

 b. His [Faulkner's] denials of extensive reading
 notwithstanding, it is no doubt safe to assume that he
 has spent time schooling himself in Southern history
 and that he has gained some acquaintance with the chief
 literary authors who have lived in the South or have
 written about the South. To believe otherwise would be
 unrealistic (Brown G28 0660).

 c. Neither had a choice other than to accept the invitation.
 To have refused would have been political suicide (Brown
 B14 1400).

(5a) involves a case of do so anaphora; in (5b), otherwise is interpreted anaphorically with reference to the preceding context; and in (5c), we have a case of null-complement anaphora: the object of refused is interpreted as anaphoric with the invitation.(6)

Bolinger's claims about these examples derive from his analysis of it, which forms the topic of his chapter. His point is to refute any radical distinction between referential and impersonal uses of it, arguing that there is a continuum, and that what are usually analyzed as impersonal uses are in fact the most extreme cases of the vague situational uses of it. This explains his claim that "the it again relates to some kind of prior basis." The idea is that the it refers back to the previous discourse context. In our terms, Bolinger's conclusion can be expressed as (6).(7)

(6) Extraposition of infinitival VP sentential subjects is necessary
 if the reference of the VP is discourse-old or inferrable.

It should be noted that Bolinger is much more cautious regarding the extraposition of that clauses. For the latter case, he says, "So it appears that semantic weight, and not the `knownness' of the content of the clause, is what forces extraposition and with it the addition of it. Here we have the strongest case for it as a grammatically introduced particle" (Bolinger 1977: 74). Note however that in the discussion preceding this conclusion, Bolinger does claim that extraposition is better for that-clause sentential subjects if the that clause represents something more or less known.

The purpose of this paper is to show that Bolinger was mistaken in his analysis of the relevance of discourse status to the choice between extraposition and nonextraposition, for both that-clause and infinitival-VP subjects. On the basis of an examination of the corpus, it appears that, contrary to his claims, there is a very significant link between discourse status and nonextraposition. Specifically, the reference of the sentential or infinitival VP subject must be discourse-old or directly inferrable from the previous discourse context in order to remain in subject position. If this condition does not hold, extraposition is obligatory.

3. That-clause subjects

3.1. Nonextraposed that-clause subjects

Consider first the following example:

(7) [..] But we must never forget, most of the appropriate heroes and
 their legends were created overnight, to answer immediate needs.
 [...] Most of the legends that are created to fan the fires of
 patriotism are essentially propagandistic and are not folk
 legends at all. [...] Naturally, such scholarly facts are of
 little concern to the man trying to make money or fan patriotism
 by means of folklore. That much of what he calls folklore is the
 result of beliefs carefully sown among the people with the
 conscious aim of producing a desired mass emotional reaction to a
 particular situation or set of situations is irrelevant (Brown
 F19 0490-0870).

In this example, we have a very complex sentential subject followed by a very simple predicate. Such a sentence would be very awkward as the first sentence of a discourse. But here it is perfectly natural because the entire content of the sentential subject has been evoked in the previous discourse. In fact, a closer look at the whole text shows that the very purpose of the presence of this subject is to sum up the content of the preceding forty lines and thus make it available as a discourse referent for the judgment "is irrelevant."

This is an especially interesting example for a number of reasons. First, it shows that it cannot be true that the choice between extraposition and nonextraposition is made solely on grounds of syntactic complexity of the constituents involved.(8) Indeed, the sentential subject is far more syntactically complex than the predicate, and the extraposed variant would be perfectly well formed. We shall see in the next example that there are many cases where the nature of the predicate makes extraposition either impossible (this is actually rather rare) or stylistically awkward. However, even in such cases, the constraints on discourse status are still observed: syntactic or stylistic awkwardness of the extraposed variant does not licence nonextraposition in and of itself. If the sentential subject is not discourse-old or directly inferrable, another construction must be chosen.

(8) "He will be here before long now," said Van Helsing, who
 had been consulting his pocketbook. "Nota bene, in Madam's
 telegram he went south from Carfax. That means he went to
 cross the river, and he could only do so at slack of tide,
 which should be something before one o'clock. That he went
 south has a meaning for us. He is as yet only suspicious,
 and he went from Carfax first to the place where he would
 suspect interference least. You must have been at Bermondsey
 only a short time before him. That he is not here already
 shows that he went to Mile End next (Stoker, Dracula).

This passage contains two nonextraposed sentential subjects. The first, That he went south, is discourse-old, in fact it is a direct repetition. Note that the extraposed variant is possible here (It has a meaning for us that he went south), though maybe slightly awkward. In the second case, the content is directly inferrable from the previous discourse (specifically from the first line: He will be here before long now). In this case, the extraposed variant is not possible for syntactic reasons, namely because of the presence of a sentential object. Here is another example of the same type:

(9) The present Republican leadership as practiced by Mundt,
 Goldwater, Bridges, Dirksen, et al., is repeating the
 errors of the party leadership of the 1930s. In that decade
 the partisan zeal to defend Mr. Hoover, and the party's
 failure to anticipate or cope with the depression, caused
 a great majority of Americans to see the Republican party
 as cold and lacking in any sympathy for the problems of
 human beings caught up in the distress and suffering brought
 on by the economic crash. [...] One may be sure the present
 Republican congressional leadership hasn't meant to repeat
 this error. But it is in the process of so doing because
 it apparently gives priority to trying to downgrade
 John F. Kennedy. That this is not good politics is
 underscored by the latest poll figures which show
 that 72 per cent of the people like the way in which the
 new President is conducting the nation's business (Brown
 Corpus, B14 0960-1130).

This example (beyond showing that history seems to have an uncanny tendency to repeat itself) is another case where the content of the sentential subject is inferrable: the politics of the Republican party consists in trying to downgrade JFK; this is qualified as an instance of repeating an error, hence this policy is not good politics.

Overall, among the thirty naturally occurring examples of nonextraposed finite that-clause subjects examined, there are two cases where extraposition would be impossible for syntactic reasons, and ten where it would be more or less stylistically awkward. In the other cases, extraposition would be acceptable. In all cases, the nonextraposed sentential subject is either discourse-old or inferrable. On the other hand, it is easy to find extraposed sentential subjects that are discourse-new, discourse-old, or inferrable, the first case being illustrated in (10), which is the beginning of an article.

(10) European Central Bank Row Won't be Last

 PARIS -- It is astonishing that the real questions about
 Europe's new single currency, the euro, and about the new
 European Central Bank were never addressed during the 12-hour
 row among European governments that ended in Sunday's sad
 compromise on the new bank's president.

 Those questions are: Can this bank truly be independent?
 And, if it tried to be truly independent, could it survive?
 The answer to both clearly is "no" (Herald Tribune, 7 May 1998,
 first lines of the article).

Clearly, it would be infelicitous to have the nonextraposed variant of (10), given in (10'), as the initial sentence of the article. This corroborates the claim that only a discourse-old or inferrable sentential subject can appear in nonextraposed position.

(10') #That the real questions about Europe's new single currency,
 the euro, and about the new European Central Bank were never
 addressed during the 12-hour row among European governments
 that ended in Sunday's sad compromise on the new bank's
 president is astonishing.

3.2. Extraposition vs. nonextraposition for discourse-old that-clause subjects

Since both extraposition and nonextraposition are possible for discourse-old that-clause subjects, the factors that govern the choice in that case must be examined. It appears that the situation is similar to that described by Birner (1994) for inversion, that is, two factors are crucial. First, there is a strong tendency not to extrapose if the predicate is discourse-new. Second, it is important to consider the way the sentence connects with the following discourse. If the predicate provides the topic of the next sentence (or is directly linked to it), nonextraposition is preferred. On the other hand, if the sentential subject provides the topic of the next sentence, the opposite is true. For instance, in (8) above, in the first case of nonextraposition, the following sentence develops meaning, which is discourse-new; both factors thus weigh in favor of nonextraposition. On the other hand, in example (11), even though the content of the subject sentence is inferrable from the preceding elements Sunday and convey the household to the little church, the following sentence links with the subject clause (as shown by the anaphoric use of the fact), thereby favoring the choice of extraposition.

(11) The observance of Sunday at Bellomont was chiefly marked by
 the punctual appearance of the smart omnibus destined to convey
 the household to the little church at the gates. [...] It was
 Mrs. Trenor's theory that her daughters actually did go to
 church every Sunday; but their French governess's convictions
 calling her to the rival fane, and the fatigues of the week
 keeping their mother in her room till luncheon, there was seldom
 any one present to verify the fact (Wharton, House of Mirth).

3.3. The fact that S-type subjects

It is also interesting to contrast the use of finite that clauses and NPs of the type The fact that S as subjects of a sentence. What appears from an examination of twelve examples of the latter type is that in nine cases the content of the NP is new, vs. three where it is discourse-old or inferrable (in the latter cases a nonextraposed sentential subject would also be possible). Consider (12a) and (13a):

(12) a. For large letters, e.g. thermoformed of acrylic
 or butyrate, there are other techniques. For
 example, in a typical store installation, fifty 24"
 and six 36" red acrylic letters were mounted against
 a white painted wood background. The fact that even
 the larger letters weighed only 5 lb. each made it
 possible to secure the letters to the building
 through clear acrylic angle brackets cemented to the
 letters (Brown Corpus, E34 1590-1650).

 b. *It made it possible to secure the letters to the
 building through clear acrylic angle brackets cemented
 to the letters that even the larger letters weighed only
 5 lb. each.

(13) a. Apparently he was not a participant in the college or
 university theatricals, which he once attacked as utterly
 unworthy performances (see Apology, 3: 300); but even in
 that famous passage, Milton was aiming not at the
 theatricals as such but at their performance by "persons
 either enter'd, or presently to enter into the ministry."
 The fact that he nowhere mentioned theatrical performances
 as part of the activities of the boys later in his
 hypothetical academy (1644) should not be taken too
 seriously as evidence that he desired them to eschew such
 performances (Brown Corpus, G68.1720-1830).

 b. *It should not be taken too seriously as evidence that he
 desired them to eschew such performances that he nowhere
 mentioned theatrical performances as part of the
 activities of the boys later in his hypothetical academy
 (1644).

In both these examples, the preceding context (beyond what is cited here) makes it clear that the content of the italicized subject is discourse-new. Thus, it would not be possible, for discourse reasons, to have a simple sentential subject, without the introductory the fact. On the other hand, these two examples are interesting in that extraposition of the sentential subject obtained by deleting the fact, which would satisfy the discourse constraints, would lead to ungrammatical sentences, because of the presence of an infinitival VP object or sentential object. This is shown in (12b) and (13b). It thus appears that subjects of the type The fact that S can be used specifically to resolve the conflict arising in cases like (12a) and (13a) where the content is discourse-new (and nonextraposition is thus impossible), but where extraposition of the subject would be ungrammatical or stylistically awkward. Indeed, among the nine examples collected where the subject is discourse-new, there are seven where the extraposed variant would have produced either an ungrammatical or a stylistically awkward result, as illustrated in (12a) and (13a).

3.4. Fronted that-clause complements

It is also interesting to compare the status of nonextraposed finite that-clause sentential subjects with fronted that-clause complements. Consider the following example:

(14) "Yes, but they are both devoted to you, Tom," said the aged
 inventor. "But what is this you hinted at -- a silent motor you
 called it, I believe? Are you really serious in trying to invent
 one?"

 "Yes Dad, I am. I think there's a big field for an aeroplane
 that could travel along over the enemy's lines -- particularly
 at night -- and not be heard from below. Think of the scout work
 that could be done."

 "Well, yes, it could be done if you could get a silent motor, or
 propellers that made no noise, Tom. But I don't believe it can
 be done."

 "Well, maybe not, Dad. But I'm going to try!" and Tom, after
 a further talk with his father, began work in earnest on the big
 problem. That it was a big one Tom was not disposed to deny, and
 that it would be a valuable invention even his somewhat skeptical
 father admitted (Appleton, Tom Swift and his Airscout).

In accord with the results of Ward (1988 [1985]) and Birner and Ward (1998), the first fronted clause is discourse-old and the second is discourse-inferrable. It thus seems that keeping a sentential subject in subject position has a discourse effect similar to fronting a complement clause. This will be relevant to the discussion of the syntax of nonextraposed sentential subjects in section 5.

4. Infinitival VP subjects

4.1. Nonextraposed infinitival VP subjects

We find similar types of example to the ones mentioned above in the case of nonextraposed infinitival VP subjects. Let me begin though by returning to the examples in (5). These are clear counterexamples to Bolinger's claims concerning the effect of anaphora on extraposition. Among the sixty-one examples of nonextraposed infinitival VP subjects collected, four are of this type, three of which were given in (5).(9) Clearly in the first example the subject is discourse-old. In the second, believe is inferrable from assume, and the content of the anaphoric adverb otherwise is of course discourse-old. In the third to have refused is directly inferrable from the impossibility of choosing not to accept, and the null complement of refused is of course discourse-old.

The following example is similar to (7) above, in that the sentential subjects sum up the contents of the previous discussion to make it available for a judgment.

(15) TO WHAT extent and in what ways did Christianity affect the
 United States of America in the nineteenth century? How far and
 in what fashion did it modify the new nation which was emerging
 in the midst of the forces shaping the revolutionary age? To
 what extent did it mould the morals and the social, economic,
 and political life and institutions of the country? A complete
 picture is impossible -- partly because of the limitations of
 space, partly because for millions of individuals who professed
 allegiance to the Christian faith data are unobtainable. Even
 more of an obstacle is the difficulty of separating the
 influence of Christianity from other factors. Although a
 complete picture cannot be given, we can indicate some aspects
 of life into which the Christian faith entered as at least one
 creative factor. At times we can say that it was the major
 factor. What in some ways was the most important aspect was the
 impact individually on the millions who constituted the nation.
 As we have seen, a growing proportion, although in 1914 still a
 minority, were members of churches. [...] Upon most of these
 Christianity had left an impress and through them had had a
 share in making the individual what he was. Yet to determine
 precisely to what extent and exactly in what ways any individual
 showed the effects of Christianity would be impossible. At best
 only an approximation could be arrived at. To generalize for the
 entire nation would be absurd. For instance, we cannot know
 whether even for church members the degree of conformity to
 Christian standards of morality increased or declined as the
 proportion of church members in the population rose (Brown
 corpus, D14 0010-0360).

Note that extraposition of the subjects would raise absolutely no acceptability problem: the predicates are very light syntactically. On the other hand, the subjects are syntactically complex, especially the first. In both cases they are inferrable from the preceding discourse context. Note also that in each case, the discourse continues on the judgment expressed in the predicate. (16a), in which he should be interpreted as coreferent with Papa, provides one more typical example of a discourse-old nonextraposed infinitival VP subject. (16b) is an example with a discourse-old nonextraposed for-to infinitival clause.

(16) a. "Will he die?" "Everybody does." Ludie could be hateful.
 To speak of Papa dying was a sin (Brown corpus, K06
 1000-1010).

 b. Nor could the temptings of prudent counsel in his head
 induce him to run the risk of such a total turnover as the
 incurring of Laetitia's pity of himself by confiding in
 her. He checked that impulse also, and more sovereignly.
 For him to be pitied by Laetitia seemed an upsetting of
 the scheme of Providence
 (Meredith, The Egoist).

4.2. Extraposition vs. nonextraposition for discourse-old infinitival VP subjects

The situation is similar to that discussed in section 3.2 above. A discourse-new infinitival VP subject must be extraposed, but there is a choice if it is discourse-old. The discourse status of the predicate and whether it is the predicate or the subject that connects to the following discourse are relevant factors. For instance, in (17), the content of the infinitival subject is discourse-old, and it could be left in subject position. But the subsequent discourse continues on the subject.

(17) In a frenzy of excitement, he considered his plan. Beside his
 shorts, he would place something of hers. Instantaneously he
 would have won an immeasurable moral victory, for if she picked
 up, say, a pair of her panties, she might just as well lift his
 shorts lying alongside -- the expenditure of energy was almost
 the same. He felt that it would be a particular humiliation to
 Dolores to pick up her own underwear which he had laid on the
 floor. Furthermore, he could go on repeating the maneuver
 endlessly: every time he went in the bedroom, he could drop a
 slip or a brassiere, or maybe a girdle, next to his shorts.
 Sooner or later, Dolores would crack (Brown corpus, P28
 1270-1360).

On the other hand, in both cases of nonextraposition in (15) above, for instance, the following sentence connects to the predicate, explaining what is possible in the first case, and giving an example of why it would be absurd in the second.

5. The syntactic position of sentential subjects

In this section, it is shown that nonextraposed sentential subjects cannot be in the same syntactic position as fronted sentential complements (e.g. [14] above), contrary to what has often been claimed in the generative literature. The latter claim is crucially based on certain grammaticality judgments that can be shown to be false once one takes into account the discourse conditions on the constructions involved. This section thus shows how important it is for syntactic analysis to take into account the discourse conditions bearing on the constructions studied if one is to base one's theory on reliable data. Otherwise, one can be tempted to mark as ungrammatical a sentence that sounds impossible out of context but becomes perfectly fine within an appropriate discourse. This is of course especially the case with constructions that involve complex discourse conditions, such as fronting and nonextraposition.(10)

Koster (1978) proposes that subject sentences and infinitival VPs never appear in the usual position for NP subjects. Instead, they are base-generated in "satellite" position (i.e. as sisters of S'), binding an empty category in COMP position. What we have been calling the nonextraposed variant is thus analyzed in a way parallel to left dislocation, but with an empty element doubling the initial constituent. Variants of this analysis have since then been adopted in many studies within the generative framework (e.g. Haegeman and Gueron 1999: 114ff.). Other studies such as those by Iwakura (1991, 1994, 1995), following Emonds (1976: 121ff.) and Stowell (1981: 156), have maintained that the subject sentence is in normal subject position in deep structure, but that it cannot remain there because of, for example, the "case resistance principle" (Stowell 1981: 146). It must consequently be either adjoined to VP (leading to the extraposed variant) or moved to topic position. The latter case results in string-vacuous movement and provides the surface structure of the nonextraposed variant. What all these analyses have in common is that the nonextraposed sentential subject is in fact in a fronted position in surface structure. This analysis, for which the above authors argue on purely syntactic grounds, appears at first sight to be supported by the discourse function of the nonextraposed variant as discussed above. Indeed, it was shown that the constraints on nonextraposed sentential subjects were very similar to those on fronted sentential objects, as one would expect under such analyses.

However, there is a major syntactic problem with these analyses. Koster claims that it is impossible to have a fronted phrase in front of a nonextraposed sentential subject as in (18b). This impossibility is claimed to be an automatic consequence, given his analysis, of the fact that it is in general impossible to front two constituents in English. If such structures as (18b) were systematically ungrammatical, this would indeed provide strong evidence in favor of Koster's analysis and all those claiming that the nonextraposed subject sentence ends up in fronted position. However, it turns out that it is in fact possible to construct well-formed examples of the type illustrated in (18b). One example of this type, cited here as (18c), was provided by Kuno (1973).

(18) a. That he reads so much doesn't prove such things (Koster
 1978: 53).

 b. *Such things, that he reads so much doesn't prove (Koster
 1978: 53).

 c. To me, that the world is round is obvious (Kuno 1973:
 note 5).

Though he cites Kuno's paper, Koster does not mention this example, perhaps because he doubted that it was actually well formed. Even if it is well formed, however, it remains irrelevant, as pointed out to me by Betty Birner (personal communication). Indeed a PP like to me in (18c) can be interpreted as a free adjunct as well as a fronted complement. This is the only possible interpretation in an example like (19), where the verb does not subcategorize a PP[to].

(19) To me, chocolate is delicious.

The meaning difference between the putative interpretations of (18c) with the PP analyzed as a complement and as a free adjunct is tenuous, and it is unclear whether or not the fronted-complement analysis is in fact possible at all. Furthermore, we have shown in this paper that non-extraposed subject sentences are only possible in appropriate discourse contexts, so one would expect cases where there is a further fronted complement to be even more restricted, and to sound quite bad when given in isolation. Consider however the following examples.

(20) Descartes claimed that the two lines in figure C were parallel
 and provided a proof based on his second theorem. This proof was
 in fact mistaken. From his first theorem on the other hand, that
 the two lines are parallel certainly does follow, but
 remarkably, Descartes apparently never noticed this.

(21) a. Through a detailed observation of gulls, Lorenz thought he
 had shown that the image of the mother was acquired. This
 conclusion turned out to be based on a series of
 misinterpretations. *On the other hand, from his
 observations of ducklings, that the image of the mother is
 innate, we have since learned, though Lorenz himself never
 noticed this.

 b. Through a detailed observation of gulls, Lorenz thought he
 had shown that the image of the mother was acquired. This
 conclusion turned out to be based on a series of
 misinterpretations. On the other hand, from his
 observations of ducklings, we have since learned that the
 image of the mother is innate, though Lorenz himself
 never noticed this.

 c. Through a detailed observation of gulls, Lorenz thought he
 had shown that the image of the mother was acquired. This
 conclusion turned out to be based on a series of
 misinterpretations. On the other hand, that the image of
 the mother is innate, we have since learned from his
 observations of ducklings, though Lorenz himself never
 noticed this.

In these examples, the verbs follow and learn subcategorize a PP[from] complement. Crucially, this PP has no alternative interpretation as a free adjunct, so that the problem discussed above for (18c) does not arise here. In (20), follow takes a sentential subject. Furthermore, the preceding context is constructed so that both the PP[from] and the sentential subject are easily construed as discourse-old, so that the discourse conditions on fronting the PP and keeping the subject in nonextraposed position are satisfied. Under such conditions the final sentence in (20) seems perfectly well formed. In (21), on the other hand, learn takes a sentential object. The preceding context has again been constructed so as to satisfy the discourse conditions on fronting the PP[from] or the sentential object, and the examples in (21b) and (21c) show that either of these can be fronted in this context. On the other hand, the last sentence of (21a), where both constituents are fronted, is ungrammatical.

It thus appears that (20) and (21a) provide a minimal pair, between which the only relevant difference is whether the preverbal that clause is a subject or an object. If both these constituents are in the same nonsubject position, as hypothesized by Koster, then it is hard to see why one should be better than the other. On the other hand, if the sentential subject in (20) is in subject position, then these data are easily explained in terms of the general constraint that forbids fronting two complements in English.

The well-formedness of 20 thus casts strong doubt on Koster's analysis and suggests that the grammaticality judgment on (18b) is mistaken. Presumably, (18b) is grammatical, but unacceptable for performance reasons, which do not come into play when the semantic role of the fronted constituent is explicitly marked by a preposition as in (18c) or (20).(11)

This discussion of the position of nonextraposed sentential subjects thus leads us to a remarkable conclusion, namely that the common discourse properties between nonextraposed subjects and fronted complements are not due to identity in surface-structure position, but rather to the fact that they appear initially from a purely string-linear point of view. More generally, it underlines the importance of taking discourse factors into account in syntactic analysis. If one does not consider the discourse conditions on such cases as the final sentences in (20) and (21a), (21b), and (21c) above, it is impossible to see the difference in acceptability between them: they all seem impossible. However, once one understands the relevant discourse conditions, it is possible to take them into account when constructing the examples and to produce crucial minimal pairs such as (20) and (21a).

Universite Lille 3 UMR 8528 SILEX, CNRS

Received 30 August 2000 Revised version received 3 April 2001

Notes

(*) This material was presented at the Congres de la SAES in Rennes in 1998, at the 1999 LSA meeting in Los Angeles, and as an invited talk at the University of Durham in 1999. I would like to thank the audiences for their comments. I would also like to thank Peter Culicover, Liliane Haegeman, Ruth Huart, Jean-Charles Khalifa, Maarten Lemmens, Gregory Ward, Arnold Zwicky, and especially Betty Birner for discussion of various issues. Finally, I thank the two anonymous reviewers for their useful comments on the paper. Correspondence address: 15, Rue des Pretres, B-1000 Brussels, Belgium. E-mail: miller@univ.lille3.fr.

(1.) Recent studies include Bennis (1986), Iwakura (1991, 1994, 1995), Rothstein (1995), Stroik (1996), Van Eynde (1996), and Bouma (1996).

(2.) Except for what Huddleston (1984: 452) calls "concealed questions."

(3.) Gerund subjects usually pattern like NPs, as one would expect. This is illustrated in (i)-(ii). However, they can undergo extraposition with a restricted range of predicates as in (iii)-(iv). Very little appears to be known about the factors governing this contrast (see Larreya 1993 for some ideas).

(i) Yet [[sub.VP[ing]] determining precisely to what extent and
 exactly in what ways any individual showed the effects of
 Christianity] would be impossible.

(ii) *Yet it would be impossible [VP[ing] determining precisely to
 what extent and exactly in what ways any individual showed the
 effects of Christianity].

(iii) Yet [[sub.VP[ing]] determining precisely to what extent and
 exactly in what ways any individual showed the effects of
 Christianity] would be no problem.

(iv) Yet it would be no problem [[sub.VP[ing]] determining
 precisely to what extent and exactly in what ways any
 individual showed the effects of Christianity].

(4.) See e,g. Prince (1981, 1992), Ward (1988 [1985]), Birner and Ward (1998), Birner (1994). See also Vallduvi (1990).

(5.) The cited examples are from the Brown corpus (Francis and Kucera 1961) and from various novels available through http://www.gutenberg.net/. For the nonextraposed variants, the corpus contains 30 examples involving finite that clauses, eight with finite clauses with a wh complementizer, eight with nonfinite for NP to VP clauses, and 61 with infinitival VPs. For the extraposed variants, 17 finite that clauses, four finite clauses without a complementizer, two finite clauses with a wh complementizer, ten nonfinite for NP to VP clauses, and 18 infinitival VPs.

(6.) I am making no claims about the appropriate syntactic analyses for these cases, i.e. whether they involve some form of deletion or ellipsis. On null-complement anaphora, see e.g. Allerton (1975) and Fillmore (1986).

(7.) Using the terminology of Prince (1981, 1992) and of Birner and Ward (1998).

(8.) As shown by Erdmann (1988), there is a statistically significant tendency to choose the relative order between subject and predicate (i.e. extraposition vs. nonextraposition) that will lead to increasing syntactic weight. However, as he himself points out, "there are clear-cut differences between the use or non-use of (non-)extraposed sentences, which I cannot go into here" (Erdmann 1988: 332). Hawkins (1994), building on Erdmann's statistics, insists on the influence of relative weight but has nothing to say about examples like (7). It appears from our discussion that the relevance of relative weight to ordering is essentially a byproduct of informational status, namely there is an obvious statistical correlation between light and discourse-old and between heavy and discourse-new. When the two parameters do not point in the same direction, however, informational status wins out.

(9.) Souesme (1989: 121, 123) provides two further naturally occurring examples of this type: "Nakasone claims that on such visits, he never walks down the center of the traditional welcoming mat for heads of government, because to do so would display arrogance" (Time, 8/1/83); "Whatever happens next, the U.S. has little choice but to stay put in Lebanon for a while. To do otherwise would be to jeopardize the Gemayel government" (Time, 10/3/83).

(10.) See also Ward and Birner (1995) for a similar position with respect to the definiteness effect in existential sentences.

(11.) Koster (1978: 53, 61; cf. also Stowell 1981: 153) claims that it is impossible to have nonextraposed subject sentences in subordinate clauses. He argues that this is a consequence of the general impossibility of having fronting in subordinate clauses. Such analyses are problematic if one takes into account papers such as Hooper and Thompson (1973), who show that so-called root phenomena regularly occur in subordinate clauses if they are asserted. In any case, there are counterexamples, as in (ia) and (ib).

(i) a. Miss Knag of course replied, that to forget anything Madame
 Mantalini had directed, was a moral impossibility (Dickens,
 Nicolas Nickleby).

 b. I went down even into the vaults, where the dim light
 struggled, although to do so was a dread to my very soul
 (Stoker, Dracula).

Furthermore, in the case of sentences with identificational be as main verb, where extraposition is always impossible, there does not seem to be a problem with nonextraposition in subordinate contexts, as shown in (ii), even when the subordinate clause is not asserted, as in (iii).

(ii) he felt that to relinquish an inch of his baldness, an inch of
 his grey hair, an inch of his coat-skirt, an inch of his
 hat-brim, an inch of his walking-staff, would be to
 relinquish hundreds of pounds (Dickens, Our Mutual Friend).
 Contrast: *he felt that it would be to relinquish hundreds
 of pounds to relinquish an inch of ...

(iii) I was surprised/annoyed/sorry that (for them) to leave was the
 only solution. Contrast: *I was surprised/annoyed/sorry that
 it was the only solution (for them) to leave.

References

Allerton, D. J. (1975). Deletion and proform reduction. Journal of Linguistics 11, 213-237.

Bennis, Hans (1986). Gaps and Dummies. Dordrecht: Foris.

Birner, Betty (1994). Information status and word order: an analysis of English inversion. Language 70, 233-259.

--; and Ward, Gregory (1998). Information Status and Noncanonical Word Order in English. Amsterdam and Philadelphia: Benjamins.

Bolinger, Dwight (1977). Meaning and Form. Harlow: Longman.

Bouma, Gosse (1996). Complement clauses and expletives. In CLIN VI. Papers from the Sixth CLIN Meeting, G. Durieux, W. Daelemans, and S. Gillis (eds.), 1-17. Antwerp: University of Antwerp.

Emonds, Joseph E. (1976). A Transformational Approach to English Syntax. Root; Structure-Preserving, and Local Transformations. New York: Academic Press.

Erdmann, Peter (1988). On the principle of "weight" in English. In On Language: Rhetorica, Phonologica, Syntactica: A Festschrift for Robert P. Stockwell from his Friends and Colleagues, Caroline Duncan-Rose and Theo Vennemann (eds.), 325-339. London: Routledge.

Fillmore, Charles J. (1986). Pragmatically controlled zero anaphora. BLS 12, 95-107.

Francis, W. Nelson; and Kucera, Henry (1961). A Standard Corpus of Present-Day Edited American English. Providence, RI: Brown University.

Haegeman, Liliane; and Gueron, Jacqueline (1999). English Grammar: A Generative Perspective. Oxford: Blackwell.

Hawkins, John A. (1994). A Performance Theory of Order and Constituency. Cambridge: Cambridge University Press.

Hooper, Joan B.; and Thompson, Sandra A. (1973). On the applicability of root transformations. Linguistic Inquiry 4(4), 465-497.

Huddleston, Rodney (1984). Introduction to the Grammar of English. Cambridge: Cambridge University Press.

Iwakura, Kunihiro (1991). Expletive it and CP-trace. Linguistic Analysis 21, 97-115.

--(1994). The distribution of CP-trace and the expletive it. Linguistic Analysis 24, 122-141.

--(1995). The distribution of non-NP categories: a government approach. Linguistic Analysis 25, 78-94.

Jesperson, Otto (1909-1949). A Modern English Grammar on Historical Principles, volume III. London: George Allen and Unwin; Copenhagen: Ejnar Munskgaard.

Koster, Jan (1978). Why subject sentences don't exist. In Recent Transformational Studies in European Languages, S. J. Keyser (ed.). Cambridge, MA: MIT Press.

Kuno, Susumu (1973). Constraints on internal clauses and sentential subjects. Linguistic Inquiry 4(3), 363-385.

Lambrecht, Knud (1994). Information Structure and Sentence Form. Cambridge: Cambridge University Press.

Larreya, Paul (1993). Deixis, anaphore et cataphore dans les constructions extraposees. In L'ordre des mots II, domaine anglais, 253-279. Saint-Etienne: C.I.E.R.E.C.

Prince, Ellen F. (1981). Toward a taxonomy of given/new information. In Radical Pragmatics, P. Cole (ed.), 223-254. New York: Academic Press.

--(1992). The ZPG letter: subjects, definiteness, and information-status. In Discourse Description: Diverse Analyses of a Fund-Raising Text, S. Thomson and W. Mann (eds.), 295-325. Amsterdam: Benjamins.

Quirk, Randolf; Greenbaum, Sydney; Leech, Geoffrey; and Svartvik, Jan (1985). A Comprehensive Grammar of the English Language. London: Longman.

Rosenbaum, Peter S. (1967). The Grammar of English Predicate Complement Constructions. Cambridge, MA: MIT Press.

Ross, John R. (1967). Constraints on variables in syntax. Unpublished dissertation, MIT. (Reprinted [1986]. Infinite Syntax! Norwood, NJ: Ablex.)

Rothstein, Susan D. (1995). Pleonastics and the interpretation of pronouns. Linguistic Inquiry 26(3), 499-529.

Souesme, Jean-Claude (1989). DO, deux valeurs, une fonction. In Explorations en linguistique anglaise, Andre Gauthier (ed.), 91-151. Berne: Peter Lang.

Stowell, Timothy A. (1981). Origins of phrase structure. Unpublished doctoral dissertation, MIT.

Stroik, Thomas S. (1996). Extraposition and expletive movement: a minimalist account. Lingua 99, 237-251.

Vallduvi, Enric (1990). The informational component. Unpublished Ph.D. dissertation, University of Pennsylvania. (Distributed 1993 as IRCS Report 93-38.)

Van Eynde, Frank (1996). A monostratal treatment of it extraposition without lexical rules. In CLIN VI. Papers from the Sixth CLIN Meeting, G. Durieux, W. Daelemans, and S. Gillis (eds.), 231-248. Antwerp: University of Antwerp.

Ward, Gregory L. (1988 [1985]). The Semantics and Pragmatics of Preposing. New York: Garland.

--; and Birner, Betty (1995). Definiteness and the English existential. Language 71, 722-742.