Printer Friendly

Subject extraction in English: some problems of interpretation.

1. Introduction (1)

In recent discussion of English syntax, the question of restrictions governing the extractability of the subject has generally been seen as a topic of great theoretical interest, and has therefore received a great deal of attention, especially in theoretically-oriented descriptions of the language. This work has concentrated on present-day English, and left the history of the phenomenon almost totally neglected, apart from the observation that Old English on this point of its structure was clearly different from present-day English (cf. Allen 1980, Pesetsky 1982).

In a full-length corpus study of the development of English extraction patterns from the earliest stages of the language down to the present day (Seppanen -- Bergh, forthcoming), we have charted the history of extractions, including subject extractions, and have offered a few preliminary results of the study in a recent article (Bergh -- Seppanen 1994). The main findings of our study as far as subject extraction is concerned can be briefly represented as follows.

Subject extraction has always been possible in English, and in Old English could indeed result in the that + gap sequence noted by Allen and Pesetsky, as illustrated in (1):
(1)a. Sio anlicnes waes gecueden oaet [DELTA] sceolde bion on oaes
 an illustration was said that [DELTA] should be on the
sacerdes hraegle oa readan appla ongemang oam bellum.
priest's robe the red apples among the bells
(Pastoral Care)

'It was said that as an illustration there were to be red apples on the
priest's robe among the bells'
b. Hu maegp yfel beonpte aelces monnes ingepanc weno
 how may that evil be that each man's mind thinks
 pte [DELTA] good sie (King Alfred's Boethius)
 that [DELTA] good be
'How can that be evil that cach man in his mind thinks is good?'

c. Ne bebode nan cristen man oorum buton paet he wille
 not ask any Christian man others but what he wants
 paet [DELTA] him sylfum geboden si (Wulfstan's Homilies)
 that [DELTA] him self askcd be

'Let no Christian man offer others anything but what he himself wants to
have offered to him'
Yet, during the whole of the Old English and Middle English period,
there was a growing tendency to avoid that sequence and to replace it by
one of two alternative forms. The more common of them was the zero+gap
pattern, as in (2):

(2)a. oaet me oinco [DELTA] sie p hehste good, paette mon
 that me seems [DELTA] be the highest good, that man
 ne ourfe nanes oores godes (King Alfred's Boethius)
 not needs no other good
'That it seems to me is highest good that no man needs any other

 good (besides it)'
b. oa getriewan friend ponne ic secgge [DELTA] sie oaet
 the true friend again I say [DELTA] be the
 deorwyroeste oing ealra pissa weoruldgesloa
 dearest thing of all this worldly prosperity
 (King Alfred's Boethius)
'The true friend I say again is the most precious of all these worldly

c. pus beo 3ewurood se man, pe se cynin3 wile
 thus be honoured the man, who the king wants
 [DELTA] wuroscipe haebbe (Assman's Homilies)
 [DELTA] honour have

'Let the man be thus honoured whom the king wants to have honour'

This structure was favoured even at a time when deletion of the complementizer was not normal outside of extractions, and it became the dominant pattern after 1300 and the only living form in the written standard after 1500 (cf. Bergh--Seppanen 1994:134-138).

The second way of avoiding the increasingly awkward that+gap structure was to use a resumptive pronoun in the position of the extracted element, as shown by the examples in (3):
(3)a. pa tunglu pu gedest pe gehyrsume, 7 pa sunnan pu
 the stars you make thee obedient and the sun you
 gedest p heo mid heore beorhtan sciman pa peostro
 make that she with her bright light the darkness
 adwaesco paere sweartan nihte (King Alfred's Boethius)
 dispell of-the black night
'The stars you make obedient to you, and the sun you cause to dispell
the darkness of the black night with her bright light'

b. Hwaeoer pe nu pince p se mon micelne anwald
 whether to-you now seems that the man much authority
 haebbe pe him selfum oinco p he naenne naebbe, ...?
 have who him self seems that he none not-have
 (King Alfred's Boethius)
'Does it now seem to you that that man has much authority who himself
feels that he has none?'

c. we ma3on cwepan paet se waeremid him sylfum, pe 3esaed is
 we may say that he were with him self, who said is
 on oam 3odspelle, paet he faeder forlete (Gregory's Dialogues)
 in the gospel, that he father abandoned

'We may say that he is alone who is said in the gospel to have left his

This device was brought in especially in the case of subject extraction, and less commonly in non-subject position, but it always remained a marginal pattern and fell into disuse in the standard language after 1500.

Taking as our starting-point the early history of the subject extraction patterns as summed up in the above remarks, we wish here to continue the discussion by addressing the more general issues connected with the development of the construction. More specifically, we shall here consider two directly interrelated questions: (i) Why was the that + gap sequence of (1) avoided and subsequently totally banned? (ii) What is the proper analysis of the structure of (2) that ultimately took its place?

2. Subject extraction and null subject/non-null subject languages

To view the history of the that + gap pattern against a wider background, consider the difference in the status of the following two sentences in English, illustrating the extraction of subjects and non-subjects, and comparing it with the lack of such a contrast in their Italian equivalents.

(4) a. What do you think that Mary has written [DELTA] to him?

b. *Who do you think that [DELTA] has written this letter?
(5) a. Che credi che Maria gli abbia scritto [DELTA]?
 what think-you that Maria to-him have written [DELTA]
 'What do you think that Maria has written to him?'
 b. Chi credi che [DELTA] abbia scritto questa lettera?
 who think-you that [DELTA] have written this letter
 'Who do you think has written this letter?'
The difference in subject extraction between the two languages reappears
in a similar contrast where Italian can dispense with a pronominal
subject (an anaphoric pronoun or a dummy) which must be overtly
expressed in English, as indicated by (6) and (7) respectively:

(6) a. Sandra non a potuto lavorare perche [empty set] e stata
 Sandra not has been-ablework because [empty set] is been ill
 b. Dove e Gianni? [DELTA] Non siamo dove [empty set] e.
 where is Gianni [DELTA] not know-we where [empty set] is
 c. [empty set] E pericoloso sporgersi.
 [empty set] is dangerous lean-out
 d. [empty set] Piove.
 [empty set] is-raining

(7) a. Sandra has not been able to work because she has been ill.

b. Where is Gianni? We don't know where he is.

c. It is dangerous to lean out.

d. It is raining.

Noting this correspondence, Perlmutter (1971) suggested a generalization which in some form or other has figured in much later work (cf. Chomsky - Lasnik 1977, Pesetsky 1982, Platzack 1985, Rizzi 1982, 1990): while languages which generally allow null subjects, like Italian and Spanish, also admit extraction of subjects out of complement clauses, those languages which generally require an overt subject in finite clauses, like English and French, are more restrictive, banning in particular the sequence complementizer+subject gap.

In connecting restrictions on subject extraction in English with the special status of the finite clause subject in the language, it is worth noting that there is nevertheless a genuine difference between the two types of structures compared. The point is that clauses exhibiting subject extraction, like (1) and (4b) above, are of course not strictly speaking subjectless, because the fronted element even in its unorthodox position still functions as the subject of the lower clause, as shown for instance by the subject-verb agreement and the active/passive relationship illustrated respectively in (8) and (9):

(8) a. [Which.sup.*] boys did you think [DELTA] was guilty?

b. Which boys did you think [DELTA] were guilty?

(9) a. Which of them did you say that Peter had invited [DELTA]?

b. Which of them did you say [DELTA] had been invited by Peter?

Nevertheless, with the subject moved into the higher clause, the that-clause contains a subject gap much like a truly subjectless clause, and therefore the ban on the that+gap sequence may well be a reflection of the general similarity of the subject extraction pattern to truly subjectless clauses.

An interesting piece of evidence of the connection between the two cases comes from a diachronic study of the corresponding phenomena in Swedish (Platzack 1985). Studying the disappearance of subjectless sentences and of the extraction pattern att-complementizer+subject gap, Platzack found that their development was largely the same, with both becoming gradually less common and disappearing more or less completely during the 17th century. Platzack assumed then that the Null Subject Parameter in Swedish changed its value from plus to minus during that century, and further strengthened this finding by the observation that some related phenomena, assumed to be linked to the minus value of the parameter, gradually came into use while the parameter was changing its value.

3. Null subjects in the history of English

From the facts quoted, it seems reasonable to hypothesize that in English too, the gradual loss of the that+gap pattern summarized above is derived from the general status of truly subjectless sentences. There are two kinds of data which can be adduced straightaway as arguments for the correctness of this hypothesis. First, we have the general development of genuinely subjectless sentences from Old English to Early Modern English. As commonly observed, Old English, though generally requiring a subject in finite clauses, admitted a number of exceptions to this main rule (cf. Closs Traugott 1992: 202, 213), but the general trend was for these exceptions to disappear gradually from the language. A few illustrations of this will be in order:
(10)a. Waere ou to-daeg on huntnooe? Ic waes, for-oaem sunnadaeg
 were you today in hunting? I was, because Sunday
 is. (AElfric's Colloquy, c. 980)
 'Were you hunting today? I was, because it is Sunday'

 b. Stodum ... esnaes ... aet gloedum foroon cold waes.
 stood ... retainers ... at the fire because cold was
 (Lindisfarne Gospels, c. 950)
 'There stood some retainers around the fire because
 it was cold'

(11)a. oa cwaep he me pyrst. (West Saxon Gospels, c. 1000)
 then said he I am-thirsty
 'Then he said: I am-thirsty'
b. Eet this when pe hrengrep. (Langland, c. 1377)
 eat this when you are-hungry
 'Eat this when you are hungry'

c. Me thynke noghte so as vn-to pe. (Hampole, c. 1340)
 me seems not so as unto you
 'It does not seem to me the same as to you'

(12) a. And as to the tytell that I have to the Lordship of Grettam, schal with in short tyme be knoweyn. (Paston Letters, 1422-1509)

b. As for these townes, wyll neuer tourne frenche. (Berners, 1525)

In the impersonal constructions of (10), the zero form was "extremely rare" even in Old English (Visser 1963: 4, Ogura 1986: 42-50), and normally replaced by a structure with hit as the subject. With the impersonal verbs of (11), the zero form similarly gave way to a construction with a clear NP subject during the Middle English period (cf. Fischer--van der Leek 1987, Denison 1990), and with the special "marked topic" constructions of (12), occurring commonly in late Middle English (cf. Johannesson 1985), the subjectless form survived into Early Modern English and disappeared in the 16th century (cf. Visser 1963: 10).

The picture that emerges from these typical examples is thus one of a situation where the originally subjectless constructions were gradually replaced by constructions with an overt subject, this tendency being clear even in earliest Old English and gradually growing during the following 700-800 years until sujbectless finite clauses were generally banned in late Middle English and/or Early Modern English. (2) This development, it will be noted, is exactly parallel to the history of the that+gap structure in extractions, from the growing avoidance of it during the Old English and Middle English periods until its total disappearance, at least in the standard language, in Early Modern English. The parallelism between the two cases, tallying very neatly with the findings of Platzack from Swedish, strongly supports the hypothesis that the history of the that-trace pattern can be seen as a reflection of the more general development of subjectless sentences.

The second point to note in this connection is the replacement of the that + gap sequence, in Old English and especially Middle English, by the that + resumptive structure, as given earlier in (3). To the extent that the introduction of the resumptive pronoun sufficed to remedy the that + gap structure, it directly shows that even though the that-clause had a displaced subject, the existence of the empty subject position was felt to be disturbing in much the same way as in genuinely subjectless sentences.

Having so far considered subject extractions and their historical development by viewing them as parallel to genuinely subjectless sentences, we must next note that the interpretation we have assumed is confronted with a major problem which, at least at first blush, makes the whole view look questionable. The point is that the form which ultimately replaced the that + gap sequence was the structure with a deleted complementizer, the zero + gap sequence, which in an obvious sense seems to be identical with the forbidden structure as far as the status of the subject is concerned. If it was the lack of an overt subject which led to the disappearance of the that + gap structure, how was it then possible for the zero + gap construction to survive? With the deletion of the complementizer, the complement clause loses its structural marker, and it might conceivably be assumed that when the clausal

status of the complement is not explicitly marked, it can more easily be accepted even in spite of its Jack of an overt subject. But if the status of that complement as a clause is not affected by the omission of the complementizer, it is difficult to see how the radical difference in grammaticality between *Who do you think that has done it? and Who do you think has done it? can be reconciled with the view that the former is ungrammatical because it has no overt subject.

The problem confronting us here has for a long time been one of the questions debated within the generative framework, and it will therefore be necessary for us to see how far the analyses offered there can contribute to the solution of our problem.

4. The that+gap/zero+gap structures in generative grammar

The discussion of the that+gap/zero+gap variation in generative syntax has often included related problems not directly connected with the task facing us. Attempting to disentangle from that discussion those aspects of the complex problem which are most directly relevant to our work, we can see two general approaches emerging, distinguished from each other in their view of the clausal status of the zero+gap pattern. The approaches are associated with different schools of generative grammar and are best discussed separately.

4.1. The GB analysis

Within the GB tradition, the approach to subjectless sentences proposed by Perlmutter has generally been replaced in later work by a treatment which distinguishes subject extraction from genuinely subjectless sentences. For the latter case, the analysis postulates the empty subject pro ("the small pro"), which may be an argument (as in Italian), or a mere dummy element (as generally in Old English), or may not at all be allowed (as in Modern English). Attempts to derive the use or non-use of pro in individual languages from some universal principle of licensing factors have not been altogether successful (3), but the explanation most commonly advanced, the licensing of pro by subject/verb agreement in richly inflected languages (cf. Platzack 1985, Rizzi 1986, Platzack 1987), might apply to English and explain the change from Old English to Modern English, as it has been claimed to explain the same change in mainland Scandinavian languages (cf. Platzack 1985, 1987).

Distinct from pro, the empty position resulting from subject extraction is occupied by an NP trace exactly as in cases of non-subject positions, and the general trend of GB work has been to show that the occurrence of subject gaps is governed by the same general principles that constrain the occurrence of non-subject gaps: all traces (empty nodes marked as extraction sites of moved elements) must, in the terse formulation of the Empty Category Principle (ECP), be "properly governed", and once this general requirement is met, there is no basic distinction in the extraction of subjects and non-subjects.

How the crucial term "proper government" is to be defined is a matter of dispute, and in current work different versions of the same basic view are found. We will begin our examination by considering the version proposed by Chomsky (1986: 47-48; cf. Haegeman 1991: 404-407), examining it in the light of the examples in (13) and their partial tree structure representation in (14):

(13) a. What do you think (that) he might want [DELTA]?

b. Who do you think (that) [DELTA] might want it?


a. What do you think t2 that he might want t1?

b. Who do you think t2 that t1 might want it?

In this version of the theory, proper government is either by theta government or by antecedent government. In (14a), the object trace t1 is theta-governed by the verb which governs it and theta-marks it (assigns a theta-role to it). In the case of the subject trace t1 in (14b), theta government is excluded, but there are three potential governors of the trace. First, the head of the IP, the I-constituent might, in-commands the trace (and assigns the nominative case to it, or to the chain who - t1), but not being coindexed with the trace, might is not a proper governor. As the trace is included within the IP-constituent and maximal projections are normally barriers to outside government, the trace can be properly governed only if it is stipulated that IP, though a maximal projection, is in fact not a barrier. With this stipulation, two potential governors are available from outside, the complementizer that and the intermediate trace t2. By the principle of strict minimality, the trace must then be governed (a ntecedent-governed) by the nearest governor, i.e. by the complementizer. But not being coreferential with t1, the complementizer is not a proper governor; t1 therefore is not properly governed and for this very reason the sentence is ungrammatical. With the complementizer deleted, the only governor of t1 that is left is t2, which is coreferential with t1, thus guaranteeing that t1 is properly governed and hence that the sentence is grammatical.

The machinery needed for the account is obviously on the heavy side, and contains at least two features which are in principle unattractive. First, as has been pointed out by Horrocks (1987: 141), it is by no means obvious that the two very different processes called theta-government and antecedent government should be thus combined, and secondly, the exceptional treatment of the IP as a non-barrier is evidently an ad hoc stipulation needed to make the system work. Even then, the result is clearly unsatisfactory for our purposes because the whole description is tailored to deal with the facts of present-day English. If the grammar treats Who do you think that [DELTA] might want it? as ill-formed, in terms of principles which are meant to be part of the innate universal grammar and hence applicable to all languages, that grammar is simply incapable of dealing with the facts of Old English and Middle English, and uninteresting for our purposes in the form in which it is standardly presented.

In this respect, the reformulation of the analysis proposed by Rizzi (1990) is in principle more promising, although it changes the theory only in a minor way. The crux of the new version of the ECP is that while traces must still be properly governed, this relationship now includes both the antecedent-government assumed on the standard view and a new type of government called head-government. According to this view, (13b) would be treated as follows (cf. (14)):

The intermediate trace t2 can antecedent-govern the trace t1 regardless of whether the complementizer position is empty or occupied by that (the principle of relativized minimality) and the difference between the grammatical and ungrammatical strings hinges on whether the .C, as head of the CP constituent, can head-govern the trace t1. Normally not a governor, the complementizer is made one if it agrees with the specifier, and the relationship between the specifier and the head can vary. In present-day English, the agreeing complementizer in complement sentences is zero, which by virtue of this agreement becomes a governor and then head-governs t1.

Rizzi does not explicitly deal with Old English and Middle English, but his reference to modern dialects which preserve the old that+gap structure shows how the description can be applied to historical material. As the possibility of a subject gap after that is dependent on the complementizer being made governor by its specifier-head agreement, the theory requires us to assume that the complementizer that did agree with its specifier (the intermediate trace t2) in Old and Middle English. In standard English, it was then gradually lost by about 1500, but was preserved in some dialects which therefore allow the that+gap pattern even in modern usage.

From this account, it is obvious that in one sense the version of ECP presented by Rizzi can deal with the historical data by the introduction of the [+ or -] agreement features for the complementizer. But taking a closer look at the suggested analysis, what does it mean to speak of agreement in this connection? In the case of the structure Who do you think t2 - t1 might want it? and its Old English and Middle English predecessors, the description makes the claim that a specifier-head agreement in terms of some unspecified categories (number, gender, etc.) holds between the specifier t2 and the zero complementizer, i.e. between two empty positions in the assumed tree representation of the sentence, and in the case of Who do you think t2 that t1 might want it? and its predecessors, such agreement similarly obtained between the empty position t2 and that in the older language but does not today. But is it meaningful to speak of agreement in the case of two empty positio ns, and in particular with reference to the complementizer position, which is empty even in a purely semantic sense? And in the case of the overt complementizer that, how do we know when it agreed with the postulated intermediate trace and when it ceased to do so when there is no sign of any variation of the item itself? As the [+ or -] Agr feature is postulated without being derived from anything else in the language, the only criterion available seems to be the possibility of subject extraction: if that+gap is possible, then that shows agreement, but not otherwise. But in such an analysis, explicitly suggested for the description of the modern dialectal use of that+gap structures (Rizzi 1990: 53), the postulated agreement, as far as we can understand, is nothing more than an ad hoc device for describing extractability: it has no explanatory power in the sense of deriving subject extractability from some other properties of the language. Would it therefore not be simpler, and hence preferable, to state it di rectly as a brute fact that the that+gap pattern was grammatical in Old English and Middle English, but is no longer so?

If Rizzi's account of subject extractions is criticized as relying on a purely technical device of agreement features, then the obvious way to make the account more plausible is to show that these features can be justified in terms of some linguistic patterning which affords independent evidence for the postulated features. This position is the starting-point for a study of the question by Platzack (1993) which is concerned with the equivalent of that+gap/zero+gap variation in modern Swedish, but which is in principle similarly applicable to English both in diachronic and synchronic terms. A brief examination of the study is in order here.

The concrete pattern dealt with by Platzack is the variation between the following two sentence types, of which the first is in common use in Swedish and the second is ungrammatical in Sweden but is a natural alternative to (a) in the variety of Swedish spoken in Finland: (4)
(15)a. vem tror du [DELTA] kommer?
 who think you [DELTA] comes
 'Who do you think is coming?'

 b. Vem tror du att kommer?
 who think you that comes
 'Who do you think is coming?'

Assuming the general approach to subject extraction proposed by Rizzi, the complementizer att and zero must bc able to identify the extracted subject vem by being coreferential with vem as indicated by the usual device of coindexing: (5)

(16) a. Vem tror du [DELTA]1 t1 kommer?

b. Vem tror du att1 t1 kommer?

Being coreferential with vem, the complementizer must obviously be nominal, and the acceptance of the (b)-example in Fenno-Swedish must thus imply that in this form of Swedish the complementizer att has nominal features which are not found in other varieties of the language. But nominality of the complementizer must in turn mean that the clause headed by it must in Fenno-Swedish have a distribution which is similar to the distribution of other nominal elements. Comparing the syntax of the two varieties of Swedish, Platzack next notes a difference between them in the use of prepositions, as in the following pair:
(17) a. Han underrattade henne om beslutet.
 a'. He informed her of the decision.
 b. Han underrattade henne (om) att han skulle resa.
 b'. He informed her (* of) that he would leave.

Before the NP beslutet, om is obligatory in both varieties of Swedish, but in front of the clausal complement (where it is obligatorily deleted in English), it can be optionally dropped in Sweden but not in Finland. By means of a test carried out among informants from both countries, Platzack shows that the difference is more generally found in the greater reluctance of Fenno-Swedes to drop a preposition in front of an att-clause in contexts where an NP obligatorily requires a preposition.

All in all, Platzack's basic idea of looking for concrete evidence for the assumed nominality of the complementizer att is laudable, and his discovery of a Fenno-Swedish reluctance to preposition deletion is of great interest. Nevertheless, before accepting these data as evidence for the subject extraction analysis assumed by Rizzi and Platzack, it seems necessary to view these data from a more general perspective. Above all, it is to be remembered that the most important distinction from the point of view of subject extraction is the contrast between the complementizer att and zero in that the zero+gap structure is perfectly acceptable to all speakers of Swedish whereas the att+gap structure represents a minority usage which is totally unacceptable to most speakers. If subject extraction is dependent on the nominality of the complementizer, then obviously the zero complementizer must be nominal in all varieties of Swedish, and the distribution of zero+clause must therefore be identical with the distribution of NPs, or at any rate closer to it than is the distribution of the that+clause pattern. What the facts of the case are we can examine in the light of the following sentences, illustrating the use of (att)-clauses in the function of subjects, objects and prepositional complements, and in apposition to an NP. The English sentences added are translations of the Swedish, but are at the same time treated as illustrations of the situation in English and are marked for grammaticality according to their status in this language:
(18)a. *(Att) Johan var narvarande pa motet var en overraskning
 a'. *(That) Johan was present at the meeting was a surprise

(19)a. Hon hanvisadetill *(att) taget hade varit forsenat.
 a'. She referred to *(*that) the train had been late
 b. Hon var glad over *(att) pojkarna hade kommit tillbaka.
 b'. She was happy about *(that) the boys had returned

(20)a. Hans forsta forslag. *(att) vi skulle stanna en vecka till,
 a'. His first suggestion, *(that) we should stay one weekmore,
 ar lattare att acceptera
 is easier to accept

(21)a. Han trodde (att) han hade ratt.
 a'. He thought (that) he was right
 b. Han trodde tydligen *(att) han hade ratt.
 b'. He though apparently *(that) he was right
 c. De kommer att foresl *(att) arendet skall bordlaggas.
 c'. They are going to propose ?(that) the question should be

As shown, the zero+S structure is totally impossible as subject, as prepositional complement and in apposition to an NP, and is acceptable in object function only with certain verbs (cf. (21a) vs. (21c)), and even there only under certain conditions (cf. (21a) vs. (21b)). Compared with the att+S pattern, the zero form is thus clearly less nominal in its distribution than the att+S structure, which is generally acceptable in all of the functions illustrated. If nominality of the complementizer is a precondition for subject extraction, we must then expect the zero+gap pattern to be generally unacceptable. Furthermore, since the distribution of the zero+S structure, as indicated in our translations, is as limited in English as it is in Swedish, the same conclusion must similarly apply to English. But since the zero+gap pattern is in fact the normal subject extraction structure in both languages, the whole idea of the nominal nature of the complementizer, crucial for the analysis assumed by Rizzi and Platzack, mu st be rejected at least as far as the zero complementizer is concerned.

Addressing ourselves next to the att-complementizer, let us consider Platzack's assumption that the non-deletability of the preposition in front of an att-clause is a sign of the nominal nature of the complementizer which then makes subject extraction possible. Since prepositions are (under certain conditions) deletable in front of att-clauses in all varieties of Swedish, and often also in English (thus about in (19b) above), it would seem that the complementizers att and that lack nominal features, in spite of the NP-like overall distribution of the clauses introduced by them, and that this lack provides the explanation for the ban on subject extraction. To test the plausibility of this hypothesis, let us for the moment leave Swedish and English, and consider thc situation in closely related languages. One interesting point of comparison is offered by Danish, which on this point differs from Swedish in its general ban on the kind of preposition deletion found in Swedish (and English). The following examples, quoted from a recent discussion of the question (Gotzsche 1994: 123-124), illustrate this contrast:
(22) a. De dr villige til at forhandle.
 b. De dr villiga att forhandla.
 they are willing to negotiate.
 'They are willing to negotiate'

(23) a. Lysten til at rejse var stor.
 b. Lusten att resa var stor.
 the desire to travel was great
 'The desire to travel was great'

(24) a. Lysten til ol er stor.
 b. Lusten efter ol ar stor.
 the desire for beer is great
 'The desire for beer is great'

While Swedish and English thus have a preposition before an NP, as in (24), but drop it in front of an att-clause, the Danish equivalents retain the preposition in both cases. The examples are from a Danish 19th century text (by H.C. Andersen) and its Swedish translation (from 1875), but the usage described is the same today, as shown in the following examples from Jacobsen (1986: 41-42), who uses them to illustrate the higher degree of nominality of Danish at-clauses compared with English that-clauses:
(25) a. De var enige om at det var en trist situation.
 a'. *The agreed on that it was a sad situation
 b. John er bange for at kore bilen.
 b'. *John is afraid of to drive the car
 c. Patienten blev saevkket af at laegerne gav ham
 c'. *The patient was weakened by that the doctors gave him
 den forkerte medicin.
 the wrong medicine.

All in all, it is thus clear that Danish at-clauses have a degree of nominality which goes far beyond the nominality of corresponding clauses in either Swedish or English. This leads us therefore to expect that in Danish, subject extraction from such clauses is fully natural in contrast to Swedish and English. However, the expectation turns Out to be totally erroneous, as can be seen from the ungrammaticality of the following kind of structures presented by Hansen (1967: 105; cf. Platzack 1993: 154-155): (6)
(26) a. Det syntes han (*at)var en drlig present.
 a'. That he thought (*that) was a foolish present
 b. Hvormange af dem sagde du (*at)havde set udstillingen?
 b'. Howmany of themdid you say (*that) had seen the exhibition?
 c. Det var Redsted som Jonas vidste (*at)havde vret rejsende
 c'. Det var Redsted who Jonas knew (*that) had been travelling
 for P.K. Hansen.
 for P.K. Hansen.

Since the nominality of Danish at is generally not enough to license subject extraction, it is difficult to accept the much weaker nominality of att in Fenno-Swedish as a valid explanation of the extraction of subjects in that language.

Another point of comparison is provided by German, where dass-clauses do not normally function as prepositional complements and where non-expression of a preposition in front of a dass-clause is common in contexts where a preposition would be obligatory in front of an NP complement. The situation is closely parallel to English, as is clear from the following examples (cf. Seppanen 1989):
(27) a. Der Lehrer freut sich (daruber), dass er ihnen die Grammatik
 the teacher make-happy himself (about) that he themthe grammar
 beigbracht hat.
 taught has
 'The teacher is happy that he has taught them the grammar'

 b. Er ist schuld (daran), dass alles missgluckt ist.
 he is to-blame (for-it) that everything failed has
 'He is blame for the fact that everything has failed'

 c. Ich bin mir (dessen) bewusst, dass es sonichtweitergehen kann.
 I am myself (of-it) aware that it so not continue can
 'I am aware of the fact that it cannot continue like that'

As the degree of nominality of dass and that appears to be about the same, we might expect them to have the same kind of effect on subject extraction. In reality, however, subject extraction is not always excluded in those varieties of German in which extraction in general is allowed, as in the following examples quoted from Andersson--Kvam (1984: 58-61):
(28) a. Welcher Zeitraum wurden Sie meinen, das [DELTA] dafur
 notig ware?
 what time would you think that [DELTA] for-it
 necessary were
 'What time would you think is necessary for it?'

 b. Was glaubst du, dass [DELTA] hier betrieben wird?
 what think you that [DELTA] here doneis
 'what do you think is being done here?'

Since the lack of nominal features of dass does not preclude subject extraction in German, there is again no reason to believe that the nature of that can explain the lack of that+subject gap structures in English.

To sum up our discussion, Platzack's attempt to seek independent evidence for the Rizzian analysis of subject extraction is in principle much to be recommended. However, when tested in the light of further data not considered by Platzack, the result is very different from that assumed by Platzack himself: whether it is the overt complementizer that or the empty complementizer position which is supposed to have nominal features which can identify the extracted subject, the predictions that follow from this assumed nominality are entirely false, and therefore force one to reject this version of the GB description of subject gaps as much as the two earlier ones. For our purposes, the most important aspect of the question is the status of the structure with a zero complementizer plus a subject gap, and its acceptability at a time when zero subjects in finite clauses were no longer allowed.

In the different formulations of the GB theory, it has always been taken for granted that the zero+gap structure has a clausal status with a subject gap, a zero complementizer before it, and an intermediate trace of the extracted subject in the Spec-C position. In view of the unsatisfactory character of the proposed analyses, it is natural to turn to a description where these basic assumptions are not taken as self-evident truths.

4.2. The GPSG account

Looking next at the description of null subjects within Generalized Phrase Structure Grammar (GPSG), we may first note that with its total rejection of all transformations, the theory deals with the general ban on null subjects simply by not admitting any PS rule which would generate a finite clause without an overt subject in a language like English. Extraction structures are generated by means of the rule XP+S/XP, where XP is equivalent to the fronted constituent and S/XP to the sentence with a missing XP constituent. The sentences in (29) and the tree structure in (30) will illustrate the normal situation where a gap appears in the expected position of the fronted constituent:

(29)a. That boy we believed that Peter had met [DELTA].

b. I don't know what they thought that Alex should buy [DELTA].

c. That is not tire case that we hoped that you would handle [DELTA].

As regards subject extraction, the theory treats it as in one sense radically different from non-subject extraction. When a non-subject XP is extracted, the VP constituent is left but contains a gap (VP/XP, where the XP can be an object NP, predicative NP or AdjP, etc.); but when the subject XP is extracted, the subject constituent as such disappears and the S/XP is rewritten as a mere VP. The examples in (31) and the tree structure in (32) will illustrate the situation:

(31)a. The Wilsons I thought [vp would stay at home.]

b. Which of them did you think [vp was the best of the lot?]

c. I didn't much care for those books which Martha said [vp were so good.]

Compared with the GB analysis, this description obviously offers a very different structural interpretation of the data. On the other hand, the structure assumed is not altogether novel, being identical with the surface structure postulated by Perlmutter (1971) on the basis of a tree-pruning rule originally proposed by Ross (1967: 24-64). The essence of the rule is the pruning of any non-branching S-node. Thus, in the structure the woman that he said that bought the rutabaga, the tree after the relativization is as follows (cf. Perlmutter 1971: 112):

Here the lower is a branching node by virtue of the retained complementizer, and thus ungrammatical as a finite clause lacking a subject. If the complementizer itself is zero, then S does not branch, and must be pruned so that bought the rutabaga comes to be a mere VP rather than a clause, and thus does not violate the subject requirement.

This earlier description of the mere-VP analysis of the subject extraction pattern is of interest, but the crucial question to ask is of course whether this analysis can be justified in terms of concrete data which are not dependent on any particular view of transformational operations. One argument for its correctness, derived from coordination, is illustrated by the following examples (Gazdar et al. 1985: 178):

(34)a. I know a man who Bill saw and Mary liked.

b. I know a man who saw Bill and liked Mary.

c. *I know a man who Bill saw and liked Mary.

d. I know a man who Mary likes and hopes will win.

In (34a-b), the sentences contain a straightforward coordination of like categories, two S/NPs and two VPs, respectively. Different from this, (34c) fails because it attempts to coordinate two different categories, S/NP and VP, whereas (34d) again is grammatical because the coordinated elements are both VP/NPs. The interesting case here is (34c), showing the critical role of the distinction between S/NP and VP in making correct predictions about grammatical and ungrammatical sentences.

Another argument to the same effect comes from the facts of auxiliary reduction, discussed by Schachter (1984). Consider the following set of examples, with the structures analyzed in terms of GPSG:

(35)a. How much {milk has milk's} been spilt?

b. How much {milk has [milk's.sup.*]} Ben spilt?

c. How much {milk has [milk's.sup.*]} John said's been spilt?

(36)a. [S [NP How much milk] [VP has been spilt?]]

b. [S [NP How much milk] [S/NP has Ben spilt [DELTA]?]]

c. [S [NP How much milk] [S/NP has John said [has been spilt?]]]

It seems thus that auxiliary contraction is blocked by a sentence boundary, as in (35b-c), but is possible when no sentence boundary intervenes, as in (35a).

Next, consider the contractions in the following sentences:

(37)a. What do you think's happening?

b. What do you think's been happening?

c. How much milk has John said's been spilt?

In each case, the subject of the lower clause is extracted, and the possibility of contraction shows that the residue is treated as a mere VP rather than an S/NP. (7)

One further indication of the mere VP status of the residue is the following contrast obtaining in some varieties of modern colloquial English, as noted by Kroch (1981: 127). The examples are quoted from his account, with his judgements of acceptability:

(38)a. There was one guy who I didn't even think that he would come.

b. [There.sup.*] was one guy who I didn't even think he would come.

If, after the extraction of the subject, the that-clause has the full clausal structure (that) - [DELTA] - VP, then there is no reason to expect the resumptive pronoun to be any better in the that+gap structure than in the zero+gap structure, whereas the bare VP, not containing any empty subject position, will automatically exclude the resumptive pronoun. The contrast is not all that clear in many, perhaps most, varieties of English, but in Swedish, where the same VP analysis seems to apply, the contrast is very clear because the complementizer+resumptive pattern is fully grammatical, as shown by Engdahl (1985: 40):
(39)a. Kalle kan jag sl vad om att han kommer att klara sig.
 a. *Kalle can I bet about that he is going to succeeed
 b. *Kalle kan jag sl vad om han kommer att klara sig.
 b. Kalle can I bet about he is going to succeed.

Thus, there is some concrete evidence for the view that in cases of subject extraction, the defective structure that remains is in fact not a clause in present-day English but a mere VP. Accepting that analysis, we must note, however, that the GPSG description of the subject extraction pattern is not generally adequate for our historical study. The crux of the matter is that the "missing subject" construction of (32) is generated by means of a metarule, which as such can apply only to lexical rules, and hence not to the rules which expand S[COMP that], which are not lexical rules. The effect of all this is that the grammar can generate a sentence like The man who I think chased Fido returned, but not *The man who I think that chased Fido returned (Gazdar et al. 1985: 161-162). But while this works for present-day English, it is obviously inadequate as a general approach to subject extraction, and even as a framework for describing the history of that structure. Th remedy this shortcoming, the system would obv iously have to be modified, but although different alternatives have been considered as possible avenues of approach (Gazdar, personal communication), so far at any rate none of them has been worked out.

To sum up, it seems then that the current version of GPSG is inadequate as an overall framework for a description of the history of extractions in English. In our discussion of this history (in the next section), we shall assume that the GPSG view of subject extraction structures as bare VP complements is correct, but will not make any specific assumptions about the way this structure is generated: in terms of the GB theory, by developing a way of generating the bare VP structure by means of tree-pruning or trace-deletion(8), or in terms of GPSG, by modifying the description so as to enable it to handle the that+gap sequence.

5. Syntactic reanalysis in subject extraction structures

After our long discussion of subject extraction patterns in present-day English, it is now time to return to the history of these structures. Sketching the development of this area of English syntax, we saw how the admission of null subjects was severely limited even in earliest Old English, and came to be more and more restricted during the Old English and Middle English periods till their total demise in standard usage in Early Modern English. This development may be described in terms of a general rule of obligatory subjects and of a gradual elimination of exceptions to the rule. In his discussion of null subjects, Perlmutter (1971: 100) formulated the rule as a surface constraint, but we can reformulate it without his transformational implications and ignoring details (imperatives, etc.) not directly relevant here:

(40) A finite clause must have an overt subject.

The rule is clearly similar to the Extended Projection Principle (EPP) of more recent GB syntax which requires every sentence to have a subject but allows for the possibility of non-overt subjects (cf. Haegeman 1991: 54, 240-243), and may indeed be taken to be a special case of the EPP. The imposition of the subject rule was bound to have an effect on the subject extraction patterns, and was directly reflected in two developments which we have noted above: the decline of the that + gap pattern and its total abandonment around 1500, and the use of the that + resumptive structure as one way of replacing the disfavoured sequence. The question that this situation raises is the status of what we have been calling the zero + gap pattern: if the development on this point was generally a consequence of the overt subject rule, how was it possible for this pattern to survive and ultimately become the only acceptable form in standard English? After our discussion of the structure of this sentence pattern in present-day English, we can now suggest that the pattern survived because it was reanalyzed so that it was no longer an exception to the usual subject requirement. In its original form, the subject extraction structure was a sentence with a zero complementizer and a subject gap, to be represented in the GB and GPSG systems as follows:

(41)a. Johni we though [CP ti - [IP ti [VP was the best candidate]]]

b. Johni we thought [S' [S [DELTA]i [VP was the best candidate]]]

After the reanalysis, the structure is to be represented as a mere VP, as in (42):

(42) John we thought [VP was the best candidate]

The reanalysis involves changes which are recognized from other syntactic developments in English: the loss of clause boundary (cf. the rise of I believe John to be guilty, Lightfoot 1981, van Kemenade 1987: 227, Fischer 1988: 69-72, 1989: 161) and the loss of an empty category (for several examples and a general discussion, cf. van der Wurff 1990: 35-58). The special combination of these component changes, we believe, is due to the similarity between the subject extraction structure and a structure involving a parenthetical comment clause, as illustrated in (43):

(43)a. Johni we thought [DELTA]i was the best candidate.

b. John, we thought, was the best candidate.

While the two structures may be clearly distinguished from each other, for example by verbal inflection (subjunctive vs. indicative), by the presence of the complementizer, by a different placement of the comment clause (cf. John was, we thought, the best candidate or John was the best candidate, we thought), or by a difference in the word order (main clauses vs. dependent clauses), often no unambiguous indicator of this kind is present and the sentences are in fact quite commonly confused in written English (cf. Seppanen -- Bergh, forthcoming). It is this influence from the parenthetical construction that we suggest lies first at the root of the early omission of the complementizer in cases of subject extraction, and then in the reanalysis of the S/NP structure with a subject gap as a mere VP. Considering the very slow replacement of that + gap by zero + gap, the reanalysis may also have taken a long time, but was probably completed by the beginning of the Early Modern English period, when subjectless senten ces in general disappeared from the language.

One indicator of this syntactic change, we believe, is the demise of the zero + resumptive pattern noted above: as long as the structure was interpreted as involving a zero complementizer and a subject gap, the gap could also be filled by a resumptive pronoun, but in its new structural interpretation it did not have a subject node to be filled by a resumptive pronoun.

Finally, the reanalysis seems to provide an explanation for the increase in the percentage of subject extractions among all extraction patterns noted after 1500 (27 per cent before 1500, 38 per cent after this date). As is well known, extractions of elements Out of a finite clause are often subject to restrictions, and are indeed in many languages only very marginal, for example in German (cf. Andersson -- Kvam 1984), or virtually non-existent in the written language, for example in Russian (cf. Pesetsky 1982: 298-299, 315-321) and Finnish (cf. Seppanen et al. 1988: 96-99). Even in languages which are relatively free in this respect, like English, they seem to be subject to stylistic restrictions which often vary between different individual writers (cf. Seppanen -- Bergh, forthcoming). It seems that subject extraction in its reanalyzed form, as it does not leave behind a truncated clause S/NP, is less bound by such restrictions, and being in this sense more readily available in different contexts, could incr ease its frequency in relation to other types of extraction.

(1.) We are grateful to Olga Fischer, Gerald Gazdar, Lilo Moessner, Christer Platzack, Robert Stockwell, Tarald Taraldsen, Johan van der Auwera, Ans van Kemenade, and Wim van der Wurff for comments on an earlier version of the paper, and to Hans Gotzsche for help with the Danish material. Needless to say, none of thc persons mentioned are responsible for any of the views expressed in our article.

(2.) Many of the instances of subjectless sentences recorded after 1500 seem to reflect a conservative usage that is idiosyncratic with some individual writers (cf. Visser 1963: 35).

(3.) For a summary of the attempts to derive the use of pro from such a universal principle, see Haegeman (1991: 418-419).

(4.) In his discussion of the contrast, Platzack describes the zero form as though it was not used in Fenno-Swedish, but in actual fact, the (a)-variant is as natural in Finland as it is in other varieties of Swedish.

(5.) Following Platzack's non-technical account, we leave out here the intermediate trace before the complementizer, and replace Platzack's dialectal example by a simpler sentence (cf. Platzack 1993: 162-163).

(6.) The examples from Hansen are here given in a slightly simplified and modernized form.

(7.) It should be added that the force of this argument is weakened by the fact that individual judgements are not altogether uniform on these and equivalent structures.

(8.) Platzack (personal communication), accepting the bare VP analysis as a possible interpretation of the data, applicable even to the description of Modern Swedish, offers some remarks on how the resulting structure could be dealt with within the GB theory.


Adamson, Sylvia -- Vivien Law -- Nigel Vincent -- Susan Wright (eds.)

1990 Papers from the 5th International Conference on English Historical Linguistics. Amsterdam: Benjamins.

Allen, Cynthia L.

1980 "Movement and deletion in Old English", Linguistic Inquiry 11: 261-324.

Allen Sture -- Lars-Gunnar Andersson -- Jonas Lofstrom -- Kerstin Nordenstam -- Bo Ralph (eds.)

1985 Svenskans Beskrivning 15. Goteborg Goteborgs Universitet

Andersson, Sven-Gunnar -- Sigmund Kvam

1984 Satzverschrankung im heutigen Deutsch. Tubingen: Gunter Narr.

Baker, Carl Lee -- John McCarthy (eds.)

1981 The logical problem of language acquisition. Cambridge, Mass.: MIT Press.

Bergh, Gunnar -- Aimo Seppanen

1994 "Subject extraction in English: The use of the that-complementizer", in: Francisco Fernandez -- Miguel Fuster -- Juan Jose Calvo (eds.), 131-143.

Chomsky, Noam

1986 Barriers. Cambridge, Mass.: MIT Press.

Chomsky, Noam -- Howard Lasnik

1977 "Filters and control", Linguistic Inquiry 8: 425-504.

Closs Thaugott, Elisabeth

1992 "Syntax", in: Richard M. Hogg (ed.), 168-289.

Denison, David

1990 "The Old English impersonal reviewed", in: Sylvia Adamson -- Vivien Law -- Nigel Vincent Susan Wright (eds.), 111-140.

Engdahl, Elisabeth

1985 "Parasitic gaps, resumptive pronouns, and subject extractions", Linguistics 23: 3-44.

Fernandez, Francisco -- Miguel Fuster -- Juan Jose Calvo (eds.)

1994 English historical linguistics 1992: Papers from the 7th International Conference on English Historical Linguistics. Amsterdam: Benjamins.

Fischer, Olga

1988 "The rise of the for NP to V construction", in: Graham Nixon -- John Honey (eds.), 67-88.

1989 "The origin and spread of the accusative and infinitive construction in English", Folia Linguistica Historica 8: 143-217.

Fischer, Olga -- Frederike van der Leek

1987 "A 'case' for the Old English impersonal", in: Willem Koopman -- Frederike van der Leek -- Olga Fischer -- Roger Eaton (eds.), 79-120.

Gazdar, Gerald -- Ewan Klein -- Geoffrey Pullum -- Ivan Sag

1985 Generalized phrase structure grammar, Oxford: Blackwell.

Gotzche, Hans

1994 Derivational syntactic structures: A contrastive linguistic study in the syntax of Danish and Swedish. Goteborgs Universitet: Institutionen for Svenska Spraket.

Haegeman, Liliane

1991 Introduction to government and binding theory. Oxford: Blackwell.

Hall, Robert A (ed.)

1985 The Eleventh Lacus Forum 1984. Columbia, S.C.: Hornbeam Press.

Hansen, Aage

1967 Modeme Dansk Bind I: Analyse. Kobenhavn: Grafisk Forlag.

Hogg, Richard M. (ed.)

1992 The Cambridge history of the English language. Vol. 1. Cambridge: Cambridge University Press

Horrocks, Geoffrey

1987 Generative grammar. London: Longman.

Ivars, Ann-Marie -- et al. (eds.)

1993 Sprak och Social Kontext. Helsingfors: Helsingfors Universitet.

Jacobsen, Bent

1986 Modern transformational grammar. Amsterdam: North Holland.

Johannesson, Nils-Lennart

1985 "Oh gracyous god my harte it burstes -- On the historical development of unmarked and marked topics in English", in: Robert A. Hall (ed.), 432-446.

Koopman Willem -- Frederike van der Leek -- Olga Fischer -- Roger Eaton (eds.)

1987 Explanation and linguistic change. Amsterdam: Benjamins.

Kroch, Anthony

1981 "On the role of resumptive pronouns in amnestying island constraint violations", Chicago Linguistic Society Papers 17: 125-135.

Lightfoot, David

1981 "The history of NP movement", in: Carl Lee Baker -- John McCarthy (eds.), 86-119.

Nixon, Graham -- John Honey (eds.)

1988 An historic tongue Studies in English linguistics in memory of Barbara Strang, London: Routledge.

Ogura, Michiko

1986 Old English impersonal' verbs and expressions. Copenhagen: Rosenkilde and Bagger.

Perlmutter, David

1971 Deep and surface constraints in syntax. New York: Holt, Rinehart and Winston.

Pesetsky, David

1982 "Complementizer-trace phenomena and the Nominative Island Condition", The Linguistic Review 1: 297-343.

Platzack, Christer

1985 "Syntaktiska forandringar i svenskan under 1600-talet", in: Sture Allen -- Lars-Gunnar Andersson -- Jonas Lofstrom -- Kerstin Nordenstam -- Bo Ralph (eds.), 410-417.

1987 "The Scandinavian languages and the Null-Subject Parameter", Natural Languages and Linguistic Theory 5: 373-402.

1993 "Satsflata med att-sats-subjekt", in: Ann-Marie Ivars -- Hanna Lehti-Ekland -- Pirkko Lilius -- Anne-Marie Londen -- Helena Solstrand-Pipping (eds.), 153-167.

Rizzi, Luigi

1982 Issues in Italian syntax. Dordrecht: Eons.

1986 "Null objects in Italian and the theory of pro", Linguistic Inquiry 17: 501-558.

1990 Relativized minimality. Cambridge, Mass.: MIT Press.

Ross, John R.

1967 Constraints on variables in syntax. Bloomington: Indiana University Linguistics Club.

Schachter, Paul

1984 "Auxiliary reduction: An argument for GPSG", Linguistic Inquiry 15: 514-523.

Seppanen, Aimo

1989 Inflexional and prepositional marking in complementation. [unpublished MS.]

Seppanen, Aimo -- Clacs Goran Engstrom -- Ruth Seppanen

1988 "On the Complex Noun Phrase Constraint", Studia Linguistica 42: 93-114.

Seppanen, Aimo -- Gunnar Bergh forthcom. Extractions in English syntax: A synchronic and diachronic study. van der Wurff, Wim

1990 Diffusion and Reanalysis in Syntax. [Unpublished Ph.D. dissertation, University of Amsterdam.]

van Kemenade, Ans

1987 Syntactic case and morphological case in the history of English. Dordrecht: Foris.

Visser, Frans Theodor

1963 A historical syntax of the English language Vol. 1. Leiden: Brill.
COPYRIGHT 1996 Adam Mickiewicz University Press
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 1996, Gale Group. All rights reserved. Gale Group is a Thomson Corporation Company.

Article Details
Printer friendly Cite/link Email Feedback
Author:Bergh, Gunnar
Publication:Studia Anglica Posnaniensia: international review of English Studies
Geographic Code:1USA
Date:Jan 1, 1996
Previous Article:From Domesday Book to Lay Subsidy Rolls: place-names as informants of linguistic change.
Next Article:The temporary merger of OE scitan and scyttan, or: a case of harmless homophony.

Related Articles
LNG Trade - The Legal Challenges.
Use, usage or both in english language teacher training programmes? (Language Teaching & Learning).
The acquisition of psychological-verb alternations in Spanish: two teaching approaches. (Language Teaching & Learning).
English resultant, affected and eventive objects in Polish grammar.
The interpretation of English reflexive pronouns by Polish learners of English. (Linguistics).
Successful experiences competition: highlights problem-solving teamwork: good ideas are sometimes their own reward, but for three Latin American pulp...
Reading for translation: investigating the process of foreign language text comprehension from an information processing perspective.
Let and the "bare infinitive": an exploratory exercise in traditional (notional) grammar.

Terms of use | Privacy policy | Copyright © 2021 Farlex, Inc. | Feedback | For webmasters |