Arizona State University



To not or not to: Optionality in Syntax?

Elly van Gelderen

Arizona State University

ellyvangelderen@asu.edu

public.asu.edu/~gelderen/elly.htm

LASSO, 4-6 Oct 2002, Pasadena, CA

As is well-known, Modern English is the only Germanic language where the infinitival marker need not be adjacent to its verb (the status of the Scandinavian variants of att being unclear). The so-called split infinitive construction has existed in English since the 14th century, as (1) shows, but has met with a lot of prescriptive opposition:

1. fo[r] to londes seche

`for to countries seek'. (Layamon Brut Otho, 6915)

Data from different corpora show that the split infinitive is relatively infrequent with negatives. For instance, in the spoken CSE[i], there are 381 instances of not preceding infinitival to, as in (2), and 59 of to not, as in (3), indicating that only 13.4% of negative infinitives are split. This is quite a bit compared to the BNC (the written and spoken part) where the figures are 17381 vs 93, which would mean .54% is split:

2. I was trying not to be specific (CSE-COMM697)

3. ... they chose to not use calculators (CSE-COMM8A97)

There are accounts for the variation in (2) and (3), sometimes phrased in terms of a difference in meaning: (2) is said to be sentential negation and (3) phrasal. Even though, differences in negative scope may sometimes be relevant, in (2) and (3), the meaning seems very similar, and (4) has the same verb as (2):

4. ... is to try to not necessarily set those problems up (CSE-COMM797)

In this paper, I add data obtained from large corpora to the debate and show that to in (2) occupies a different position from to in (3): split constructions such as (3) are only possible when a C-position is available. One piece of evidence is that when C is occupied by for, as in (5), the sequence is not to:

5. it would be inconsistent for Thatcher not to do this (BNC-AA9 753)

The data with adverbs other than not are different, since they occupy different positions depending on whether they are CP, IP, or VP-adverbs (to use their traditional names). However, even with `high' adverbs, to precedes the adverb frequently, and this strengthens the conclusions reached for not, namely that there is a position for to relatively high (C) and one below it (M), but that the one below it is preferred in negatives, due to the status of the negative element.

In 1, I describe data that show that to can be in C, and in 2, I give evidence for what its lower position is. In 3, modals are briefly discussed since they are in complementary distribution with to. In 4, I discuss other adverbs, and in 5, an alternative analysis. In 6, the theoretical implications and conclusions are given.

1 The higher to

There are, as noticed in Beukema & den Dikken (1989) and Bernstein (1994)[ii], reasons to assume a second position, which is C. Using data from extensive corpora, I provide additional arguments that there are two positions, and that the `high' position of to is C. First, I examine the structure of the the CP.

1.1 The CP

To determine where high to is, we need to consider if the infinitival CP is split, as suggested by Rizzi (1997) for Italian (but the same would have to be determined in Cinque's 1999 framework). In English, finite complementizers that and for precede topics, which shows they are in the head of Force Phrase, as in (8) (note that for is hardly ever used this way):

6. She said that prickly pear jelly she sort of likes t.

7. ... for that jelly I sort of like.

8. ForceP

. Force'

Force TopicP

. Topic'

Topic FinP

. Fin'

Fin IP

that jelly she ...

This means that and finite for are in Force not in Fin[iii]. Historically, this has probably always been the case for finite for, which is first attested as a finite complementizer around 1200, as in (9). That is now in Force but in Fin probably till about 1500, as (10) shows:

9. For frenshippe we haue foune. (York, 10,12)

10. She loved Arcite so

That when that he was absent any throwe,

Anon her thoghte her herte brast a-two (Chaucer Benson 377: 93-5)

The infinitival complementizer for, as in (11), cannot occur with a topic. The reason is that for needs to be adjacent to the subject for Case reasons. However, a topic cannot precede it either, as the ungrammaticality of (12) shows. This is unexpected if for is in Fin:

11. *I expect for [her homework] her to do.

12. *I expected [her homework] for her to do

If we test the position with other non-finite complementizers, the result is the same, i.e. topics cannot precede them, as in (13) and (14), or follow them, as in (15). Unlike in (11) and (12), Case to the subject is not an issue, so these constructions are perhaps more reliable indicators than (11) and (12):

13. *He swam the chanel i[to England] in order to get ti.

14. *I wonder i[to Alaska] whether to go ti.

15. *I wonder whether i[to Alaska] to go ti.

Thus, non-finite clauses do not project a split CP. This eliminates a number of possibilities for the position of to. If for is present, I will argue below, to is not in C. Now, I'll provide some examples of the to not construction.

1.2 To not

Since a number of linguists (e.g. Bernstein 1994; Kayne 2000) argue that to not involves phrasal negation, I'll provide some instances that it isn't always. It occurs as complement to adjectives, as in (16), to nouns, as in (17), to verbs, as in (18), as adverbial clauses, as in (19):

16. It would be unrealistic to not expect to pay higher royalties (BNC-CSS 245)

17. the freedom to not attend lessons (BNC-AHG 1258)

18. He professes to not be ready for that (BNC-CGB 1649)

19. to train the dog to not be afraid of people (BNC-K54 6582)

Almost all of these have not to counterparts, some minimal pairs:

20. It would be unrealistic not to show them to be human (BNC-CBF 14312)

21. one of the best advertisements not to have an animal (BNC-HV0 52)

22. He professes not to want the job (BNC-ABJ 970)

23. We'll train you how not to `blow it' (BNC-CFV 2052)

1.3 C

There are a number of arguments that to in (3) is in C. One of the most convincing ones is mentioned in the introduction, namely that to not does not occur with the complementizer for. For instance, in the BNC, (24) occurs 233 times (with pronoun and two-word subjects) but for followed by to not occurs only once in a passage with a number of peculiarities, namely (25):

24. she prefers for me not to stay on the phone for very long she does.(BNC-KPY 150)

25. it was an hour late, which it was late enough for us to not to know if it was coming, so we had to go and get another PA, otherwise we'd have gone to get, there was no other option. (BNC-J92 528)

Below, I'll argue that the repetition of to in (25) indicates that to moves from M to T, but results in an uncommon sentence. This suggests to is in C or M but not in T.

In the (smaller) American corpus (CSE), there are ten sentences with for followed by a subject followed by not to, as in (26):

26. if you as a parent choose for your child not to participate, you may opt out (CSE-COMR6B97)

In this corpus, there are, however, also four apparent counterexamples. (27) is an instance of an ECM verb and it is often argued that the object moves higher to check Case (in AGRo). So (27) is not decisive. Checking want in the BNC, it occurs 26 times in a construction as (27) but always with not to. So, (27) may be an exception. The other three possible counterexamples are as in (28), and involve not only/just ... but constructions where not is part of a VP conjunction, and phrasal nagation is indeed at stake:

27. They don't want them to not get them before their science and social studies. (CSE-COMR6A97)

28. You can look for the president to not only talk about legislative initiatives ... but also to tie those to ... (CSE-WH94)

So, to when it precedes not is (generally) in complementary distribution with for. This means that in that position, it is in C.

A second argument that the to of to not is in C involves verbs such as seem. It is known from independent tests that seem does not have a CP complement (raising and lack of overt complementizer). If they are C-less, they should not have the to not sequence, and this is indeed the case. For instance, in the BNC, forms of seem with not to, as in (29), occur frequently (namely 249 times), whereas there is one hesitant to not, as in (30):

29. At first, the darkening official mood seems not to have troubled Prokofiev (BNC-ABJ 524)

30. pay schemes ... seem to not y'know deal with that very effectively (BNC-JTO 120)

Verbs other than seem are less clear. For instance, with complements to try and want, it is often assumed that there is an empty complementizer in sentences such as (31). If that is the case, does to occupy this C? The numbers are as follows: 782 forms of try have not to, and only 3 have to not, of which one is (32) with a double to:

31. You tried to go.

32. I mean I don't try to not to take a lot of notice of it (BNC-G5R 32)

This suggests that the C in complements to control verbs is an empty C, as argued e.g. in van Gelderen (1993: chap 5), indicating a future/unrealized tense. Unfortunately, control verbs such as try cannot be compared with ECM-verbs, generally believed not to have a C, such as believe, guess, suspect, imagine, know, since none of the latter occurs as an ECM verb in the BNC or CSE with an adverb. Native speakers prefer (33) over (34), but that may be because of prescriptive reasons:

33. I believe him not to be a good guy.

34. I believe him to not be a good guy.

A third argument involves that sentences often end in not to, as in (35), but never in to not, as in (36). This can be explained if to is in C in (36):

35. because they desperately wanted not to realize it. (BNC-A69 1473)

36. *because they desperately wanted to not realize it.

The notion of `phase', as in Chomsky (2001a) is helpful in explaining this since CP and vP are phases that can be deleted but TP (and ASPP) cannot. This means that in (35) either CP or vP is deleted. CP is unlikely since some remnant of the embedded clause remains. Therefore, vP is deleted and to is in T, or in M, or in ASP. The reason (36) is ungrammatical is that to is in C, and ASPP or TP is deleted. In the next section, I will examine which position the lower to is in.

A last argument that there are two positions is that in a number of modern corpora, (37) and (38) occur. This is less an argument about where to is than that there are 2 tos and that movement takes place:

37. - as a request to not to -. (CSE-WH97A)

38. This is to try to not to overturn the ... (CSE-WH97A)

In (37) and (38), assuming the copy theory of movement as in e.g. Chomsky (1995; 2001ab), the copy of to is not deleted. This construction is relatively frequent, e.g. in the BNC, it occurs 27 times, and native speakers find it reasonably acceptable. It is too consistent to be a `performance' error that is independent of the actual structure.

2 The lower to

The evidence for a position below the C needs less justification. The lower position has always been the focus of attention. In the early 1980s (e.g. Akmajian et al. 1979), to comes to be argued to be I (AUX in that work). In this early framework, there are only two Functional Categories, namely I and C, and arguments that to (and modals) are in I are therefore simple:

(a) They are not part of the VP since VP-deletion and VP-preposing occur, leaving both modals and to, as in (39) and (40). They can also be split off from the VP, as in (41). Hence, they are both in I:

39. He wants to read and she wants to.

40. He can read and she can.

41. I've had to actually drag it out of you. (BNC-KRP 278)

(b) Infinitival to and modals are in complementary distribution, explained if both occupy the same functional position. Both modals and to indicate future, unreal actions, so they should both be in the same position. Using these arguments, one of the positions that to can occupy is I (or a variant thereof in a split IP framework[iv]). (c) They are not in C in (42) since an overt complementizer occurs together with to, as in (42):

42. I expected for them to do their HW.

Thus, in a framework with just I and C, to in (42), (39) and (41) is in I. However, in the late 1980s, the I is split, and more possibilities become available. An extreme version of the split IP is Cinque (1999) who provides an incredibly expanded functional structure with over 30 functional categories that are included in every sentence of every language. In such a framework, it is much harder to determine where to is. Cinque does not discuss to, except in a very brief note (p. 189) referring to the work by Bernstein (1994) mentioned above. I'll try to determine what Cinque's system would say. The evidence for Cinque's system comes from adverbs in the specifiers of the positions. In (43), I list some of the `TP' categories, with the adverbs listed underneath:

43. Modepist Tpast Tfut Moodirr Modnecc Modposs Modvol Modability/perm

probably once then perhaps necessarily possibly intentionally [NA]

Cinque (1999: 79; 81; 106)

However, many of these adverbs never cooccur. For instance, in the 100 million word BNC, there are no instances of perhaps and possibly next to each other; in the 45 million word (British and American) Cobuild Corpus, there is one of the reverse order possibly perhaps, but with a pause in between[v]. Considering this, I will argue that (43) should be seen in terms of clusters, such as CP, TP, MP, and ASPP, that are sometimes expanded. Thus adapting Cinque's system, I argue that to is in M. I will now give arguments for that claim.

In relation to (35) above, lower to is argued to be either ASP, M, or, T. ASP can be ruled out since to can precede aspectual material in English, as in (44), and hence T or M are left as possibilities:

44. It is a pleasure not to be going there.

Historically, this was not always so (see also Abraham 2002 for a lower position for Dutch and German `to'). Only after 1500 does to start to cooccur with aspectual auxiliaries. There is some additional evidence that to is in ASP in earlier stages, namely the occurrence of (45) where to indicates progressive aspect:

45. I saw her corall lips to moue (Shakespeare, Shrew I, i, 75).

Having eliminated ASP, there are two arguments in favor of to being in M, as in (46), over it being in T. One is that, as mentioned, sequences such as to not to occur regularly (in Cobuild 6 times, in BNC 7 times, and in CSE twice) but not of to to not in the BNC, Cobuild, CSE, and HC. If to were ever in T, the latter would be expected:

46. CP

. C'

C TP

to . T'

T NegP

. Neg'

Neg MP

not . M'

M ASPP

to . ASP'

ASP vP[vi]

The second reason is that sentences such as (25) above are very rare, and the typical sequence is as in (26) above, expected if to is in M. In (46), I have included T to clarify the possibilities but the fact that to to not never occurs points to to not stopping in T on its way to C. Hence, the correct structure is (47) without the TP (see also Stowell 1982):

47. CP

. C'

C

to NegP

. Neg'

Neg MP

not . M'

M ASPP

to . ASP'

ASP vP

Many people have commented on the difference in to: to in complements to control verbs, such as try, is irrealis, but it is dependent on the tense of the main verb with verbs such as believe. Stowell (1982) attributes it to the presence or absence of C, as I have mentioned above. IJbema (2002: 105ff) argues for Dutch there is a realis te in T (with an independent tense that is not unrealized or future) and an irrealis one in M. The latter is the case, I argue, in English, where to is not in T.

Since the preferred position of to is after not, there are several possible accounts but the one I argue for is where to moves to C in (2), leaves a copy in (37) and (38), and is ungrammatical in cases of TP-deletion as in (36). In section 6, I discuss what causes the movement of to from M to C (if anything), but first I discuss modals, often seen as in CD with to.

3 Modals

Since modals do not occur together with to, it is expected that they at some point occupy M. Based on Abraham's work on modals in Germanic (e.g. 1998), van Gelderen (2002) argues that deontic modals in English originate in ASP whereas epistemic modals are in a higher position and select ASP Phrases as their complements. This accounts for a number of facts, the main one that epistemics but not deontics can have stative complements with auxiliaries such as have and be. For instance, in (48) and (49), must is epistemic and cannot be deontic. This is accounted for if epistemics select an ASPP, but if deontics are in ASP in complementary distribution with other aspectuals:

48. He must have read that letter.

49. He must be looking for that letter.

As mentioned, there is some evidence in the history of English that (for a short time) to occupies ASP, but since to coocurs with have and be in Modern English, it is like the epistemic modal, i.e. in the higher position.

Unlike in the case of to, Cinque's (43) is relevant to modals, which occur in the heads of projections that have the adverbs in the specifier. (43) would predict a preference for an epistemic modal being followed by an irrealis or possibility adverb, whereas a deontic modal might be preceded by such an adverb. Modals too have positions that are relatively fluid, like to, but they end up in T. Thus, there should be no difference. Checking two corpora, the 100 million word BNC and the smaller CSE, there is no difference between the types of modals and adverbials. This is predicted if they are both in T. In fact, both are typically followed by the adverb[vii]. To account for these data, Cinque would have to argue that all modals move to or via T (which he seems to do, e.g. p. 87), but that takes the sting out of the proposal. A second argument that all modals end up in T, even though they do not seem to be checking tense and agreement there, is the cliticization with the negative n't. If the NegP is below TP (see Ouhalla 1990), the modal moves to the head of Neg on its way to T, just like the auxiliaries have, be, and do that move to check their features.

4 Adverbs

The relative numbers of (2) and (3) are very different for non-negative adverbs, e.g. always, actually. The typical pattern is the split infinitive with many, as in (50), and not the non-split one, as in (51):

50. You're going to always need more (CSE-COMR797)

51. So our view is always to find outstanding individuals (CSE-WH96A)

There are some exceptions, as shown in Table 1. Lower adverbs such as completely are expected to always show a split infinitive, since to is in M and the adverb closer to VP. This is indeed the case, and whether or not to moves to C wouldn't make a difference. The higher adverbs would be above M and would be expected not to be split. However, TP-adverbs such as probably when split show that to has moved to C:

___________________________________________________________________________

BNC Cobuild CSE

split non-spl split non-spl split non-spl

_______________________________________________________________

CP-adverbs

frankly 1 0 0 0 0 3

TP

probably 7 18 19 17 8 0

perhaps 37 45 17 37 6 5

possibly 5 3 4 3 1 1

v/VP

always 35 26 40+ 18 8 3

completely 3 0 6 0 1 0

___________________________________________________________________________

Table 1: split vs non-split in the spoken corpora

In the case of to not sequences, there is evidence that to is in C, and that's true for other split constructions with TP-adverbs as well. In the entire BNC, for does not occur with to probably/to allegedly. As in the case of not, constructions with double to are found, namely (52) to (55):

52. the thing to do would be to probably to separate the six ... (BNC-JS8 368)

53. To possibly to help to get those people around (Cobuild)

54. Okay time to perhaps to relax (BNC-F88 295)

55. I should be able to just to deal with this (linguist in e-mail message)

Again, as with the to not to, these `errors' occur in a specific pattern and are therefore not `just' performance errors.

Comparing the movement of to to C in negatives to that in non-negatives shows that the latter is more frequent. There is, however, much variation in the corpora: the BNC and Cobuild selection being only British and the CSE American. For instance, in the case of probably, to moves to C in 28% of the cases in the BNC, but in the Cobuild, this is almost 50%, and in the CSE a 100%. This may also have to do with not every speaker placing the adverbs as high.

In conclusion, not to is preferred over to not but with non-negative TP-adverbs the pattern is different. The VP adverbs are expected to be split, and the CP-ones expected not to be split (and this is so in the case of CSE frankly, but not the BNC).

5 Alternative approaches

Kayne (2000: 301-3) provides an analysis of (2) and (3) by using a difference in meaning. According to him, the split infinitive in (3) involves constituent negation, whereas the non-split one in (2) involves sentential negation (Bernstein 1994: 17; 20 argues the same). The derivations of (3) would be as in (56), and that of (2) as in (57):

56. WP[ [trying tj]i tok ZP[ [not be specific]j tk [ti]].

57. WP[ [trying not tj]i tok ZP[ [be specific]j tk [ti]].

The derivation starts with trying not be specific and to gets merged to that. After this, either not be specific, as in (56), or be specific, as in (57), move to the Spec of to. Then, to moves to the head of WP and the VP to the Spec of WP. The difference between the sentences is whether or not not is stranded. Even though Kayne's account explains the VP-deletion data in (35) and (36) above, there are two problems with this account. One is that Kayne predicts a sharp difference between (58) and (59) on the one hand and (60) on the other. Only (60) is an impossible derivation since both the sentential and phrasal not would be preposed with promised:

58. John promised to not not do it.

59. John promised not to not do it.

60. John promised not not to do it. (Kayne 2000: 303)

With the right intonation, some native speakers accept all but some accept none. The BNC and Cobuild only have instances of (59) and in most of them there is doubling to not to not to, as in (61):

61. The aim is not to not to gain weight. (BNC-FL6 74)

The second problem is that multiple to, as in (38) above, cannot be derived using (56) or (57). Only (62) and (63) can be:

62. WP[ [try tj]i tok ZP[ [not overturn the decision]j to [ti]].

63. WP[ [try not tj]i tok ZP[ [overturn the decision]j to [ti]].

Kayne's framework provides a way to first merge the lexical material and after that the functional material. Theoretically, this is very desirable, but does not account for (38) and others.

6 Theoretical implications, conclusion, and further research

The conclusion so far is that the behavior of not and TP-adverbs in relation with to shows that to moves from M to C on occasion. It can only do so when the C is present but not filled by for. When C is empty, however, it is not clear why it moves on one occasion but not on the other, as in (64) and (65), and other minimal pairs listed above:

64. I'll be trying to perhaps get invited to the meeting (BNC-KS1 236)

65. I'd just like perhaps to seek a little clarification from Mr Curtis (BNC-HVJ 169)

Taking a strictly Minimalist view (Chomsky 1995 etc), one would not expect to to move unless it is actually checking something. Moreover, we'd expect at some point to have `late merge' (Kayne 2001) to avoid having to merge as well as move. The problem for the `merge over move' principle is to explain why to would not always be in the higher position.

There are two possible reasons. One is that the rule against splitting infinitives is still very strong in standard English (see Fowler 1926 [1950]: 558; Quirk & Greenbaum 1973: 312). This is shown in the difference between CSE and BNC. The second is that typically not is a head (see 3.2 above) and that in order to allow to to move across not, as in (3), not is forced to be a specifier. Unlike modals and auxiliaries, to doesn't cliticize to the negation. This means the negation is a specifier of NegP in (3), but possibly a head in (2), the way it is in the case of modals, blocking movement to C. This is still possible in Modern English since nothing else is in the Specifier position (due to the powerful, prescriptive prohibition on double negation).

The situation sketched above can be seen in terms of grammaticalization of an element from a lower to a higher Functional Category. In present-day English, both M and C are used but in earlier forms to was `lower' (in ASP), as were the modals. It is expected that further changes would continue the development and have to become a C element. I will not go into do here, which like modals is in CD with to, but it fits the same picture. Like deontic modals, do is originally aspectual, before the ing takes this over. For instance, the 1688 grammar by Miege lists as the present tense I love or I do love. The 1700 grammar by Lane comments: "[t]he Auxiliaries of the Present Tense, are, do, dost, ...; am, art, ...; as I do call ...; I am calling" (p. 44). So it has grammaticalized the same way.

References

Abraham, Werner. 1998. "The Morphological and semantic classification of `evidentials' and modal verbs in German: the perfect(ive) catalyst". Groninger Arbeiten zur germanistischen Linguistik 42: 192-206.

-- 2002. "The Grammaticalization of the infinitival preposition". MS.

Akmajian, A., S. Steele & T. Wasow 1979. "The Category of AUX in Universal Grammar". Linguistic Inquiry 10.1. 1-64.

Baltin, Mark 1995. "Floating Quantifiers, PRO, and Predication". Linguistic Inquiry 26.2: 199-248.

Bernstein, Michael 1994. "Some Notes on Infinitives and Negation in English". Cornell WPL 12: 1-25.

Beukema, Frits & Marcel den Dikken 1989. "The position of the infinitival marker in the Germanic languages", in Sentential Complementation and the lexicon: 57-75. Dordrecht: Foris.

Chomsky, Noam 1995. The Minimalist Program. Cambridge: MIT Press.

-- 2001a. "Derivation by phase". Kenstowicz, Michael (ed.) Ken Hale: A Life in Language: 1-52. Cambridge: MIT Press. [References still to the ms form]

-- 2001b. "Beyond Explanatory Adequacy". ms.

Cinque, Guglielmo Adverbs and Functional Heads. OUP.

Denison, David 1993. English Historical Syntax. Longman.

Faarlund, Jan Terje 2002. talk at NRG II.

Fowler, H. W. 1926 [1950]. A Dictionary of Modern English Usage. Oxford: Clarendon.

Gelderen, Elly van 1993. The Rise of Functional Categories. Amsterdam: Benjamins.

-- 2002. "ASP(ext) and English Modals". MS.

IJbema, Aniek 2002. Grammaticalization and Infinitival Complements in Dutch. Leiden.

Kayne, Richard 2000. Parameters and Universals. OUP.

-- 2001. "Prepositions as Probes", ms.

Miege, Guy 1688. The English Grammar. Scolar Press.

Ouhalla, Jamal 1990. "Sentential Negation, Relativised Minimality and the Aspectual status of auxiliaries". The Linguistic Review 7: 183-231.

Quirk, Randolph et al. 1973. A University Grammar of English. London: Longman.

Rizzi, Luigi 1997. "The Fine Structure of the Left Periphery", in Elements of Grammar. Liliane Haegeman (ed.): 281-337. Dordrecht: Kluwer.

Stowell, Tim 1982. "The Tense of Infinitives". Linquistic Inquiry 13. 560-70.

Notes

-----------------------

[i]. The corpora used in this paper are the 100-million word British National Corpus (or BNC, see thetis.bl.uk), the Helsinki Corpus (or HC), the Cobuild Corpus (see titania.cobuild.collins.co.uk/form.html), and the 2-million word Corpus of Professional Spoken American English (or CSE, see ). The latter contains transcriptions of commerce dept meetings (COMM), faculty meetings (FACMT), and White House briefings (WH).

[ii]. The problem with Beukema & den Dikken's account is that the discrepancy between not and other adverbs is not explained, since other adverbs adjoin to AGR presumably. Also, the behavior of not as a clitic in finite clauses, such as (i) would be unexplained, since it wouldn't be adjoined to AGR:

i. ... that I didn't want food.

Bernstein (1994) suggests to (obligatorily) moves to C, but this runs into problems accounting for the not to sequence of (2).

[iii]. Notice that, unlike topicalization in (3), left dislocation is only possible in a main clause as in (i) and not in a subordinate as in (ii):

i. Those books, I read them.

ii. *I know that those books I read them.

[iv]. T in van Gelderen (1993); AGR in Beukema & den Dikken (1989).

[v]. The same is true for some of the CP-adverbs (not shown in (43)). According to Cinque (e.g. p. 106), frankly, fortunately, allegedly, probably, in this exact order, are all in specifiers of separate functional categories. Yet, frankly and fortunately never cooccur; neither do fortunately and allegedly; or allegedly and probably (even though the adverbs occur frequently by themselves: the most frequent, probably, occurs 27304 times).

[vi]. Note that Baltin (1995) provides a neat argument that the PRO subject stays internal to the VP (based on the position of quantifiers).

[vii]. The numbers in the BNC with perhaps are: 139 can perhaps, 7 perhaps can(not), 73 might perhaps, and 9 perhaps might. With possibly, the numbers are: 76 might possibly, 3 possibly might, 147 can possibly, and 134 possibly can. The reason the latter is so high is unknown but they are all fixed elliptical expressions after comparatives in as NP possibly can, and in conditionals if NP possibly can. It may be that these deontics don't move from ASP to I since these are irrealis. I'll not go into this more.

-----------------------

3

................
................

In order to avoid copyright disputes, this page is only a partial summary.

Google Online Preview   Download