Arizona State University

ellyvangelderen@asu.edu

DRAFT 1 July 2000

Giorgi & Pianesi (1997) argue that bare (eventive) infinitives without -e(n), as in English, have a perfective feature since otherwise they'd be no different from nouns. In Dutch and German, infinitives have an -e(n) ending and need no such feature. The empirical evidence Giorgi & Pianesi provide is two-fold. (a) Perception verb complements in English are perfective but they are not in Dutch and German, and (b) the simple present in English cannot be used to express present tense.

There are several problems, however, with this analysis. First, the infinitival ending in English is lost several centuries before the infinitive becomes perfective, as defined in G&P, and before the simple present ceases to be used. Second, eventive (bare) infinitives are not always interpreted as perfective in Modern English. They only are as complements to stative perception verbs, not as complements to modals, or non-stative perception verbs. This means stative perception verbs such as see are more like evidential modals whereas the non-stative variety in English and perception verbs in Dutch are `regular' verbs. Such an analysis accounts for a number of `old' problems, among others the difference in constituent structure between bare infinitives and -ing constructions (as discussed in Akmajian 1977), and restrictions in complementation to perception verbs. The structure I suggest also reflects the fact that in many languages evidentials and perfectives are related, in accordance with what Abraham (1998; 1999) and Comrie (1976) argue.

The main differences between Modern English on the one hand and similar languages on the other now emerge as (a) see is more grammaticalized in the former than in the latter, (b) the contents of ASP in Modern English depend on the affix, whereas in Germanic, ASP is ambiguous, and (c) as argued by traditional grammarians, the real changes are caused by the availability of -ing, and perhaps by the loss of aspectual markers, including -en. Around the same time that -en is lost, i.e. around 1400, to becomes reanalysed as I. Hence, tense starts to become more more prominent at the expense of aspect, but the effects of the aspectual system remain felt.

The outline is as follows. In 1, I sketch Giorgi & Pianesi's analysis of bare infinitives; and briefly describe the problems with this analysis. In 2, it is shown that complements to perception verbs continue to have non-perfective bare infinitives after the ending disappears. It is also shown that the use of the simple present continues long after. In 3, I argue that there are two kinds of see, an evidential and a full verb, each with their own structure, and in 4, I compare the structure of the evidential with that of the other modals.

Giorgi & Pianesi (hence, G&P, 1997: 163ff.) argue that English bare (eventive) infinitives carry a perfective feature. The reason they need this feature is that their morphology does not differentiate them from nouns. G&P derive two empirical differences between a language such as English, without an infinitival ending, and languages such as German and Dutch, with an infinitival ending. First, they (1997: 163ff.) argue that the difference between (1) and (2) is due to it:

1. I saw/*see him cross the street.

2. Ik zag/zie hem de straat oversteken.

It is well-known (e.g. Quirk et al. 1985: 1206) that Modern English bare infinitives differ from those in other Germanic languages in that the event referred to by the infinitive in (1) must be completed. In Dutch, on the other hand, the infinitive in (2) can refer to the action in progress (or to the completed action). If, as G&P argue, English eventive infinitives are [+perfective], (1) cannot have an imperfective meaning. Instead, to indicate an incomplete action in English, the progressive is used, as in (3) and (4), not the bare infinitive, as in (1):

3. I see him crossing the street.

4. I saw her reading the book for hours.

Languages such as Russian make similar morphological distinctions to indicate perfective, as in (5), or non-perfective complements, as in (6) (Buzarovska 2000):

5. Ja videla kak Bob pereshel ulicu

I saw if Bob cross-PF street

6. Ja videla kak Bob perehodil ulicu

I saw if Bob cross-impf street

There are other indications that the infinitive in (1) is really perfective: (7) is not that great since the `for hours' forces a durative reading, incompatible with the perfective, unlike its Dutch counterpart in (8):

7. ?I saw him read the book for hours.

8. Ik zag hem urenlang dat boek lezen.

As in other constructions, a perfective is typically triggered with definite objects, and imperfective with indefinite ones. Therefore, (9) should be worse in English than (10). It is, but only very marginally so, however:

9. ?I saw him read books.

10. I saw him read the book.

To jump ahead, I will agree with G&P that in (1), the complement is perfective but for a reason different from theirs. I will also argue that not all bare infinitives are perfective.

The second piece of evidence that G&P use is that eventives cannot occur in the simple present tense, as (11) shows, since they are perfectives and perfectives are bounded and the present is not. Instead, the progressive as in (12) is used:

11. *I eat right now.

12. I am eating right now.

The presence of [+perf] is compatible with the progressive which is bounded (Giorgi & Pianesi, p. 169). Stative verbs such as know and see are not associated with [+perf] since, like habituals, they are associated with a generic operator.

Thus, according to G&P, there are two reasons for assuming English bare infinitives are perfective: the interpretation of (1) and the ungrammaticality of (11). If this account were correct, languages without the infinitival ending would always be expected to be like English in these two respects, and English bare (eventive) infinitives would always be expected to be perfective. Neither of these two prove to be correct, as I show in the next two sections.

In this section, I will argue that interpretations as in (1) are not dependent on the infinitival endings, since the change to the Modern English interpretation of (1) does not coincide with the loss of the ending, and neither does the ungrammaticality of (11).

In Old and Middle English, the infinitive has an ending, just like in Modern Dutch, and the Middle English constructions in (15) and (16), with romen and smyte, have imperfective interpretations, as in (2) above:

15. Chaucer, I, 1098

The fairness of that lady that I see

Yond in the gardyn romen to and fro

Is cause of ... (Kerkhof 1966: 55)

16. Chaucer, I, 1220

The deeth he feeleth thurgh his herte smyte.

In Chaucer, both an -e and -en ending still occur, as they do in Modern Dutch where the -en ending is mainly pronounced -e. The ending is presumably pronounced in Chaucer, i.e. smyte in (16) rhymes with Arcite.

After Chaucer, the situation changes. Minkova (1991), citing a number of other scholars, argues that the ending is disappearing in the North from 1100 on but that "[b]y 1400 final unstressed -e had been abandoned in all parts of the country" (p. 30). Görlach (1990: 47) says that from 1400 onwards, the -e spelling is "arbitrary and optional" (see also Moore & Marckwardt 1951). Southworth (1947: 925) estimates that even in Chaucer the final infinitival -e is not pronounced in 82% of the cases. As mentioned in the introduction, this fits with the rise of I (cf. van Gelderen 1993), and means tense is `taking over' the aspectual system at least for non-finites. In the fifteenth century Paston Letters (hence, PL), the infinitival ending -en is fairly rare: apart from ben `to be', there are perhaps 20 in a large corpus. In the PL, neither bare infinitive nor -ing is popular. However, (17) occurs but it is not clear what the status of speke is, and adding more context does not help:

17. Paston #112 (1480)

he herd my moder speke it.

The PL show that, after the endings are lost, no immediate changes with perception verb complements occur.

In More's English (Visser 1946), from the beginning of the 16th century, i.e. a century or so after the loss of the infinitival ending, some infinitives have a perfective meaning, as in (18) with it, but not all. (19) emphasizes the action through the adverb and Modern English would use running. This text has many -ing forms too (Visser 1952: 810). Thus, even though the infinitive has no ending, it can be used as a non-perfective:

18. Apol, 489 E 8

She hard him boast it.

19. Rich 71 C 1

The fox ... saw him run so faste.

(Visser, 1952: 761-2)

An Early Modern English grammarian, Mulcaster, who writes an Elementarie in 1582, divides final -e into "soundeth or ... silent" (p. 111). The first category includes me, see, we, agree, yee, and e in Latin words, but the section is very short; the silent -e section is much longer (and talks about nouns as well as verbs). Silent -e is said to have an effect on the length of the vowel preceding it, as in made, cure, and is used in many other situations, e.g. in cause, excuse, deceiue, loue, moue. Thus, Mulcaster's description shows that Elizabethan English infinitival endings are not pronounced differently from Modern English. Franz (1909: 21) says the infinitival ending is used "ziemlich prinziplos" in Shakespeare's time, and -en is never used. This loss of -e(n), however, does not seem to force an increase in the use of -ing, since very few complements as in (20) and (21) occur. Checking all 852 instances of heare in the First Folio edition of 1623, only 5 occur with -ing complements, 2 in Hamlet, as in (20) and (21), 2 in King Lear, as in (22) and (23), and 1 in 1H4, as in (24). Notice that 4 of these have the same verb:

20. Hamlet, III, i, 55

I heare him comming.

21. Hamlet, III, iv, 7

I heare him coming.

22. Lear, II, i, 30

I heare my Father comming

23. Lear, IV, vi, 31

let me heare thee going

24. 1 Henry 4, II, ii, 103

I heare them comming.

Instances of bare infinitival complements after heare, as in (25) and (26), are very frequent. After heard, the only complements are bare infinitives, as in (27) and (28), participles, as in (29), and to-infinitives, as in (30). I come back to the latter below:

25. Tempest, II, ii, 20

And another storm brewing; I hear it sing i the wind.

26. Troilus & Cressida I, ii, 244

Har! do you not hear the people cry Troilus?

27. Macbeth, II, ii, 16

I heard the Owle schreame and the crickets cry

28. Macbeth II, ii, 35

Methought I heard a voyce cry `Sleep no more'.

29. Tempest I, i, 193

hear these matters denied.

30. CoE, V, i, 26

who heard me to deny it?

There are 2 -ing complements after see/saw, namely (31) and (32), but at least 10 bare infinitives, as in (33) to (35), and many past participles:

31. AYLI III, IV, 52

who you saw sitting by me on the turf.

32. Macbeth, V, v, 37

may you see it comming

33. AYLI V, ii, 23

see thee wear thy heart in a scarf

34. Tempest III, i, 12

She sees me work

35. LLL IV, iii, 181

see me write

Thus, in Shakespeare, there is no connection between the loss of -en and the perfective interpretation of the infinitive: the infinitive has lost its ending but it continues to be used as a non-perfective.

There is another frequent construction from later Middle English on that sheds light on the structure, namely (36), which seems especially frequent in the 16th and 17th centuries, as in (37) to (39), and (30) above:

36. Wyclif, Gen Vi 12

Whanne God had seen the erthe to be corrupt.

37. Bunyan, PP 144

I see it to be so.

38. Shak. Shrew I i 175

I saw her coral lips to move.

39. Lyly Euphues 397

Venus seeing him so earnestly to loue, and so effectually to pray, ...

In the early 15th century Pecock (Zickner 1900: 67), see is complemented by a bare infinitive only twice but by to 4 times and forto once. So, at the time the infinitival ending is disappearing, the to-infinitive takes over to express durativity for a while. This construction occurs with modals only if the modal is very far removed from the infinitive (Visser 1952: 590; 620).

Another reason that the loss of -en is not the cause for changes in (1) is (40), a construction in a contemporary of Chaucer:

40. Piers Plowman, B, 10.101-4

I haue yherde hiegh men . ettyng atte table

Carpen as thei clerkes were . of Cryste and of his mi3tes

And leuden fautes vppon the fader . that formed vs alle

And carpen a3eine clerkes . crabbed wordes (from Wandschneider 1887: 63)

Piers Plowman has infinitives with endings (mostly -e), -yng, as well as -ende, as in (41). Thus, it is not the demise of the infinitival ending that necessitates the introduction of -ing. The forms co-occur:

41. PP, B, 17.48

Thanne seye we a Samaritan . sittende on a mule,

Rydunge ful rapely ...

What is the reason for the changes, if not the infinitival ending? It has often been assumed that OE and OHG etc. display more aspect than their modern offspring. For instance, Streitberg (1891) assumes that the German prefixes are perfectivizing. Brinton (1988: 202 ff.) argues that OE verbal prefixes indicate telicity, but that ge- has become "seriously over-extended" (p. 212) by ME. Mustanoja (1960: 446) writes that ge- remains productive as a perfectivizer "down to the 13th century". Its disappearance is due to Norse influence and occurs first in the North according to Mustanoja, and Chaucer only uses a limited palatalized version in the poetry. For retains its productivity "down to the end of the ME period". The verbal prefixes are replaced by particles and phrasal verbs extend their domain in ME and become less marked (Brinton 1988: 226). Thus, if one wants another explanation for the change, it might be that with the loss of the aspectual prefixes, ASP is reanalysed as -ing.

In conclusion, the loss of -en cannot be shown to coincide with the bare infinitive becoming perfective. It is more likely that the infinitive remains ambiguous until gradually -ing is reanalyzed as imperfective marker.

There are two other problems for G&P's account, related to the use of the simple present. (a) The progressive is available in OE, i.e. is not introduced with the loss of -en: there is even no sudden increase in (12). (b) The use of the simple past remains frequent after the loss of the infinitival ending. Modern Afrikaans presents a counterexample as well. It is a language without infinitival endings (e.g. te drink, te se, te kom) but one in which eventives can be in the present tense (Nou dans die poppe, Ponelis 1991: 187; Paul Roberge p.c.).

The ultimate reason for the increase of the progressive is not known - it occurs especially in the 19th and 20th centuries, see Scheffer (1975: 110ff.) - but it does not coincide with the demise of the infinitival ending. Mosse (1938, II, 2ff.), based on Streitberg, attributes the popularity of (12) to the demise of the aspectual system from the 13th century on. In OE and Germanic, the simple verb is durative, whereas the verb composed of prefixes (he calls them `preverbes') is not. One of the problems with this theory is that already in OE the -ing/ende form is alive and well, as (42) shows, especially with verbs of dwelling and movement (see also Pessels 1896; Raith 1951: iii). It continues from then on, as (43) shows:

42. Beowulf 159

ac se æglæca ehtende wæs

but the monster pursuing was

43. Chaucer, I 929

We han ben waitynge

`We have been waiting'.

The additional problem for G&P is that long after the -en disappears, the simple present persists. I first provide some ME examples. The Middle English data in (44) and (45) are similar to those in other Germanic languages, e.g. Swedish, German, Dutch in (46), and OE in (47), since eventive presents occur. Chaucer, as shown above, still has an infinitival ending and hence G&P cannot be proved or disproved before 1400. Piers Plowman, also from before 1400, has three such forms of which (48) is one:

44. Chaucer, I, 3437

What do ye, maister Nicholay

45. Chaucer, I, 3763

He sharpeth shaar and kultour bisily

46. What doe je? Ik eet een appel.

`What are you doing? I am eating an apple'.

47. VP 11.6

nu ic arisu cwið drihten

`Now I rise up said the lord'. (Visser 663)

48. PP B Prol. 67

The most myschief on molde is mounting wel faste. (Scheffer 1975: 216)

In the 15th century Paston Letters, i.e. after the infinitival ending is lost, the special progressive is relatively rare (around 20 perhaps):

49. Paston #187 (1465)

þer ys comyng, ..., more than a thowsand.

50. Paston #40 (1452)

where the seid felechep is abydung.

51. Paston #336 (1469)

syche mony that he is owyng.

These occur with the same verbs as in OE. Most present participles head non-finite clauses without the auxiliary be. The present is expressed as in (52) to (54), with the latter possibly being fossilized. Most of those would not get -ing in Modern English, e.g. (50) and the frequent (51), and don't occur before 1450:

52. Paston #3 (1425)

I send you

53. Paston #4 (1426)

I make þis day a new apelle

54. Paston #3 (1425)

I recomaunde me to you.

By the time of More, i.e. the early part of the 16th century, the progressive is "employed rather sparingly" (Visser 1946: 248). In Shakespeare, -ing is used on occasion, e.g. (44) and (45) above and (55), cf. also Franz (1909: 514), but Visser (p. 662) says, about both More and Shakespeare that "at the time the choice between the two possibilities did not yet depend on any fixed principle":

55. MWW, III, ii, 36

Now she's going to my wife.

The simple present is used frequently, as in (57):

56. MWW, II, ii, 10

Whither go you?

57. MWW, I, i, 155

What say you, Scarlet and John?

58. MWW, II, ii, 75

But what says she to me?

Visser provides examples up to the present.

Concluding section 2.2, G&P's account for the differences between Dutch (46) and Modern English (11) by assigning [+perf] to the English eventive encounters empirical problems. Even as late as Shakespeare's time, long after the disappearance of -en, (56) to (58) are grammatical. I therefore suggest that this problem is independent of the ending, but depends on what is in ASP.

In this section, I show that bare (eventive) infinitives are not always perfective, and argue that the interpretation depends on the matrix verb instead.

As is well-known, perception verbs such as see are typically stative. As a result the simple present is used in (59) rather than the progressive in (60):

59. I see (the) mountains.

60. *I am seeing (the) mountains.

As seen in (1), infinitival complements to these stative verbs are perfective. Not discussed in this connection in the literature, as far as I know, but crucial for my analysis below, is the occurrence (61) to (64), where the bare infinitive expresses duration, and perfectivity is not connected to the infinitive:

61. Seeing her swim is exciting.

62. I made them watch Michael swim (for hours).

63. Martin took it, feeling himself surrender. (from Visser, 2251)

64. We'd be hearing him holler for mercy. (from Scheffer 1975: 68)

So, when the perception verb is not a state, and used in the progressive, as in (61) to (64), its bare infinitival complement need not be perfective. This is not explained in G&P's account.

I will argue that see is ambiguous between a non-stative reading, as in (61) to (64), and a stative reading, as in (1). In the latter case, it is an (evidential) modal. It is not unusual for verbs of perception to grammaticalize into evidentials. For instance, Gordon (1986) shows that in Maricopa, a Yuman language, see and hear can be either evidential or a full verb with a sentential complement.

Apart from (61) to (64), there are other cases where the infinitive need not be perfective, namely when the infinitive is dependent on a modal, as in (65), it is not perfective. I come back to (65) in section 4:

65. He must go.

Thus, there is some evidence that (1) is different from (2) even though both involve bare infinitival complements: an exclusively perfective reading in (1), an incompatibility with duration adverbs in (7), and a slightly marginal (9). However, this is only the case if the bare infinitive is embedded under a stative perception verb, as in (1). The (preliminary) structure I suggest for (1), (2), and (3) is (66) (Below, I suggest a slight modification of (66) and have see start out in v and move to ASP):

66. ASPP

. ASP'

ASP vP

. v'

v VP

. V'

saw V ...

-ing/-en cross

Assume that ASP indicates +boundedness/perfectivity, then Modern English has two options for ASP: either perfective saw or imperfective -ing. If see is stative, it originates in ASP, as in (1), and then cross cannot have an independent aspect but is dependent on see. This use of see is evidential and it is not surprising that it occurs only in the past. Abraham (1998) argues that "evidentiality is ... often triggered by the perfect or perfectiveness". Buzarovska (p.c.) reports that in Greek and Macedonian perception verbs in the past tense, i.e. as in (1), have a special infinitival complementation that makes them more grammaticalized into evidentials than the present tense ones.

See can also be non-stative, and then it is a full verb higher in the tree with cross checking either imperfective, as in (3), or perfective, with a tree as in (67), identical to (66) but with the higher VP showing:

67. ASPP

ASP VP

V ASPP

. ASP'

ASP vP

. v'

v VP

. V'

V ...

-ing see +pf him cross

Since -ing is not available in Dutch, the -en ending is ambiguous. Therefore, oversteken `cross' can have independent aspect which it checks in ASP.

The complement to a perception verb cannot be a stative, as (68) shows. Using Diesing (1992) and others to argue statives are IPs (see also Rochette 1988, and Higginbotham 1983: 118 for a different account), one can expect that IP complements such as in (68) will not occur since they do not `fit' in (66) and (67):

68. *I saw you be/being tall.

69. *I saw him know/knowing the answer.

The structure of these complements never allows auxiliary have or be (except passive) in either English or Dutch, as in (70) and (71) respectively:

70. *I saw him have crossed the street.

71. *Ik zag hem de straat zijn overgestoken.

This is again explained by the structure: since perfect have and progressive be result in states (see Vlach 1981: 287 and Comrie 1981), i.e. IPs, they cannot occur with perception verbs. Once have is used, as in (72), the structure changes into one where the -ing modifies the subject or object, and a comma intonation occurs between him and having:

72. I saw him having crossed the street.

Syntactically, this means that the complement is pretty reduced in structure, as shown in (66); not an IP or CP. The same is true in Dutch.

In 3.1, I argue that see can be stative or non-stative in English but not in Dutch. The complement to perception verbs is reduced in both langauges, i.e. ASPP not IP.

Evidence in favor of (66) comes from an old problem, namely the different constituent structures of (1), (2), and (3). Akmajian (1977) argues, on the basis of preposing and clefting, that the structures for (1) and (3) are quite different: in (1), the NP and infinitive are separate constituents; in (3), they are not. Thus, (73) is grammatical but (74) is not:

73. It was [the moon rising over the mountain] that we saw.

74. *It was [the moon rise over the mountain] that we saw.

Applying this to Dutch (75), the result is (76), where the infinitive patterns with the -ing in English:

75. We zagen de maan door de bomen schijnen

we saw the moon through the trees shine.

76. [De maan door de bomen schijnen] is wat we gisteren zagen

the moon through the trees shine is what we yesterday saw.

These differences between (1) and (2) should come out in the structure, and they can if one argues that see in Modern English is generated in a VP-shell, as in (77), a slight variation on (66). This is only possible if the embedded infinitive is bare and does not move to ASP, see being a closer head:

77. IP

. I'

I ASPP

. ASP'

ASP vP

. v'

pf v VP

^ she . V'

saw V NP

him cross the street

(Note that the subject receives theta-role, Experiencer, from see, but that him would be a Theme. I will not go into this further).

With -ing, as in (3), bare infinitives in (61) to (64), and in Dutch (2), the structure would be as in (78), similar to (66) and (67), i.e. with see less grammaticalized and the structure biclausal. In (78), cross moves to ASP and him to Spec ASP:

78. VP

V ASPP

saw . ASP'

ASP vP

ing . v'

v VP

. V'

V NP

him cross the street

The difference between (77) and (78) accounts for the data in (73) and (74). In a structure such as (77), the subject she would move to Spec IP and saw to ASP and to I, and therefore, the traces inside the ASPP in (74) would not be bound in preposing the ASPP. In (78), the ASPP would move and no trace of see or of the subject would be left un-c-commanded. Thus, the crucial difference is either having the subject of see and see inside or outside of ASPP.

If sentences such as (61) to (64) are different from (1), the preposing should be grammatical as well, and it is, as (79) and (80) show:

79. [The moon rise over the mountains] is what we missed seeing.

80. [The moon rise over the mountains] is what we wanted to see.

In (79) and (80), see is not stative and therefore, the complement can be preposed. Thus, the difference between (73) on the one hand and (79) and (80) on the other provides extra evidence for the correctness of (77) and (78).

There are some interesting data from code switching between Dutch and English that shed some light on the difference between (77) and (78) as well, even though a lot more research is necessary here. Sentence (81) is a possible sentence. The meaning of the sentence is that John was picking flowers, and have a structure as in (78). Thus, the English infinitive does not inherently have a perfective meaning or the CS should be ungrammatical:

81. Ik zag John pick some flowers.

I saw ..............

Sentence (82) is very marginal. Structure (78) explains this since Dutch infinitives need to move to ASP and in (81) this is impossible since saw would have done so:

82. ??I saw Jan bloemen plukken

.... John flowers pick

Before concluding this section, I'll return to the historical data of section 2. If the loss of -en is related to the loss of the aspect system, we expect ASP to become available for other elements, e.g. for see to move to. Sentences such as (36) and (37) show that (78) may be appropriate with to occupying ASP.

In conclusion, I have shown that bare infinitives are not always perfective and there is some evidence that see can occupy two different base positions. There are a number of empirical advantages to (77) and (78): (a) characterizing the perfectivity constraint in (1), (b) accounting for the difference in constituent structure between (1) and (3), (c) explaining (1) versus (61) to (64), both in terms of perfectivity and constituent structure, and (d) possibly explaining some code switching data. In Dutch and older English, zie would not have grammaticalized as far, since the -en would occupy ASP. Sentences such as (1) are grammaticalized: see behaves more like an auxiliary. I will not go into the question whether or not see and cross are one event at this point (the data using again/once more are very unclear).

This section shows that the assumption about eventive infinitives as perfective in Modern English does not work well for modals, as in (83). Linking the ending and perfective would also predict that with the loss of the infinitival ending, deontic modals would be lost (since their complements cannot be perfective), which is not the case. The account I pursue is that modal complements are sensitive to internal and external aspect. Finally, I compare `modal' see from section 3 to other modals.

Root or deontic modals take eventive verbs as in (83) (e.g. Abraham 1998), presumably because a change still has to take place and the deontic is an operator on that change (cf. Barbiers 1995), i.e. because they are bi-phasic (see Abraham (2000). If the English infinitive is [+perfective], as suggested by G&P, this change cannot occur anymore, and (83) should be ungrammatical:

83. He must/may read that letter (deontic reading and eventive verb).

Sentence (83) suggests a structure somewhat like (77), as I will argue later on in this section.

Epistemic verbs take stative complements, as in (84), since these verbs judge the possibility or necessity of a particular proposition, and would be in I, hence they can be complemented by a stative ASPP:

84. He must/may know that person (epistemic reading and stative verb).

An added problem for G&P is that the Dutch interpretation in (85) and (86) is similar to that in English, regardless of the difference in ending. Thus, it is the internal aspect (or Aktionsart) that is responsible for the data in (83) to (86), not the infinitival ending:

85. Hij moet die persoon wel kennen (epistemic).

86. Hij moet die brief lezen (deontic).

An eventive verb can combine with an epistemic modal in both Dutch and Modern English, if it is made stative through hebben and have, as in (87) and (88), respectively. Optionally, modal particles can be added:

87. Hij moet die brief (nu toch wel) gelezen hebben.

88. He must have read that letter (by now).

Another possibility in Modern English for an eventive to complement an epistemic is as in (89). The event denoted by the progressive is closed, and allows this reading:

89. He must be looking for that letter.

Again, as with deontics, I will provide a structure later on in this section.

Thus, it seems that the infinitival ending is not relevant to (83) to (89). If it was relevant, one would expect differences in modal complements as well. One might also expect deontics to disappear and epistemics to flourish as the ending disappears. This turns out not to be correct, as I now show.

As is very well-known (Goossens 1982, Denison 1993, Lightfoot 1979, Warner 1993, to name but a few), the modal comes to be used epistemically in ME. Traugott (1989: 41-2) claims that there are some OE examples of evidential sceolden, as in (90) and (91) (her (!23) and (!24)). None of the infinitives have lost their endings and hence, these should be grammatical (in (90), the infinitive is made perfective through ge). They are not interesting though since the ending is still there:

90. Chron E (Plummer) 1100:4

91. Bo. 35.98.25

These are very early examples and they prompt Warner (1993: 163) to say "that these examples ... are of rather indeterminate status between epistemic and dynamic". The first epistemic use of can listed in the OED has a stative infinitive, with -en, as well. Going through the uses of mot `must, may', cunnen `can, know' and mei `may, can' in Katerine, a thirteenth century text, most of the modals are deontic, their complements are eventive, and all have an infinitive with -en. Examples are (92) and (93), and the context is such that a deontic is the most likely:

92. Katherine 123-4

ah nes þer nan þ mahte neauer eanes wrenchen hire

`but not is there none that might entice her

93. Katherine 555

mahte ... habben aweld

might ... have compelled

So, before the loss of the infinitival ending, most modals are deontic. Therefore, the crucial change will be when the ending is lost and infinitives supposedly become inherently perfective. If the lack of ending indeed means that the infinitive is [+perf], deontics should not be able to have a complement without an ending, and one would expect deontics to die out. This is not the case, certainly not in British English (see Coates 1983: 25 for percentages). In American English, as Abraham 2000 argues for different reasons, the alternatives such as e.g. have to seem a lot more prevalent than e.g. must, but this is not related to the infinitive. In the 15th century Paston Letters, as in (83), deontics occur with an infinitive, as in (94) and (95), even though the lack of ending would mean having an incompatible perfective infinitive:

94. Paston #47 (1453 or after)

ye must remembre how that ...

`you must remember how ...'

95. Paston #89

he schall mow do myche good

`he shall have to do much good'.

Hence, there is no link between a lack of infinitival endings and loss of deontic modals.

A note on the use + perfective and + stative is in order. As mentioned above, epistemic modals take stative verbs, but non-stative verbs can be made stative by the use of have/be. This indicates that both internal (stative) and external aspect (have/be) are relevant. I am assuming terminative in Abraham's (96) (e.g. 1990; 1998; 1999), a table with restrictions for German complements to modals, implies both:

96. -terminative +terminative

EM + -

DM +/? +

Terminative in (96) must refer to the aspect of the entire complement, rather than just the infinitive. This needs to be expressed in the tree, and I will by assuming a deontic is in ASP and therefore in CD with have/be whereas an epistemic is in I, not in CD.

Again, as Abraham has shown in detail, a root/deontic modal in Modern English is in complementary distribution with a perfective or progressive auxiliary, as (97) and (98) show respectively:

97. *I can have read that book (root/deontic and perfective).

98. *I can be swimming (root/deontic and progressive).

So, for Modern English, the structures I assume are as in (99) for root/deontic modals and as in (100) for epistemics. The deontic in (99) is base generated in the head of the ASPP and moves to I, since it is well-known that root/deontic modals are lower in the tree than epistemics (Picallo 1984, 1990; Wurmbrand 1998; Avram 2000). This accounts for the complementary distribution and makes the root/deontic into an aspect marker, which given its history is not surprising. Thus, have/be and the root/deontic are in CD:

99. IP

. I'

I ASPP

. ASP'

ASP vP

. v'

v VP

. V'

V NP

can she read a book

100. IP

. I'

I AspP

. Asp'

Asp vP

. v'

v VP

. V'

V NP

might have she made you read a book

As in Thráinsson & Vikner (1995), I argue that epistemics are raising verbs. T&V are "non-committal" (p. 72) about the exact status. Roberts (1985: 37ff.) argues that modals are not raising verbs in ME but that the evidence is hard to come by.

The difference between a deontic modal and a perception verb is minimal, except that the latter is more verbal (therefore in v) and has its own subject. Deontic modals are said not to have their own agent-theta-role (e.g. Roberts 1985 calls them adjunct-theta-roles, and Wurmbrand 1998: 271; 292 argues mainly on expletive subjects that there is no theta-role). An epistemic modal selects a complement that is `complete' in terms of aspect, but cannot have its own subject since it is I.

Interesting with have is that unlike its use without a modal, the meaning is a real past tense without present relevance. Thus, an adverb as in (101) is perfectly grammatical, unlike in (102) when have is not the complement of a modal. This is evidence for (100) with have in ASP:

101. I may have made him ill yesterday.

102. *I have made him ill yesterday.

In (101), the present relevance of the `normal' present perfect is absent once the modal checks the tense in I. Have is just past since it checks the 0 in the ASPP. This provides evidence for structures such as (77), (78), (99) and (100) in which ASP features prominently.

In conclusion to 4, epistemic modals `originate' in I, deontics in ASP, and perception verbs in v (this should have repercussions for theta-theory which I don't explore). The most grammaticalized, i.e. epistemics, are highest in the structure; then deontics; then see.

In this paper, I shed doubt on the assumption that endingless eventive infinitives are perfectives. The reason for this doubt is that only certain bare infinitives are perfective and that at the time of the loss of the infinitive marker around 1400, there is no sudden change in either the interpretation of (1) or the grammaticality of (44).

In an attempt to provide an account why Modern English on the one hand and Dutch, Old and Middle English on the other differ, I argue that including an ASPP explains a number of structural facts and that see in (1) is an evidential modal: (a) the complementation similarities between deontics and perception verbs in Modern English stem from being in or being lower than ASP; epistemics can select an ASPP and are therefore higher, (b) structural differences, e.g. as between (73) and (74).

I will point to at least three areas that need further research.

The first point is an unaccounted observation that I haven't seen discussed much, namely that, early on (van Swaay 1901: 55-6), modals were often complemented by ge-infinitives, as in (103):

103. Beowulf 20

Swa sceal geong guma gode gewyrcean

(Visser 1586)

This suggests early modals had a structure as in (78), and were similar to see as in (36) to (39) except by then the aspect marker was different.

An interesting area I have not examined is event structure, and the similarities between verbs such as see and verbs like make (see Hale & Keyser 1993). In a structure with an embedded clause, there are obviously two events. The structures I have examined here are ones where bi-clausals become mono-clausal, and the two (original) events may be one. Therefore, it may be more accurate to say that there is one event in (1), rather than a seeing event and a perfective crossing, but two in (2). This may be a result of the grammaticalization. This remains for further research and tests such as the use of again/once more are very inconclusive.

A third question I have not considered is what kind of features occupying ASP are. If they are Uninterpretable (Chomsky 1995), overt movement would be necessary. In addition, one needs to know what the features of the word connected to the aspectual marker are.

Fischer (1992) argues for Middle English that to expresses tense independence

An interesting ending occurs in (104), which I will not go into further:

104. Chaucer F 1547

And told hym as ye han herd me sayd. (Visser 1952: 761)

??Austin, J.L. 1962. Sense and Sensibilia. Oxford: OUP.

Barbiers, Sjef 1995. The Syntax of Interpretation. Leiden dissertation.

Denison, David 1993. English Historical Syntax. Longmans.

Einenkel, Eugen 1884. Life of St Katherine. EETS

Gelderen 1993; 1997?

Giorgi, Alessandra & Fabio Pianesi 1997. Tense and Aspect. Oxford.

??Jelinek, Eloise 1995

Jespersen, Otto 1931. Modern English Grammar IV. London: Allen & Unwin Ltd.

Lightfoot, David 1979. Principles of Diachronic Syntax. CUP.

Mustanoja, Tauno 1960 A Middle English Syntax. Helsinki.

Ndayiragije, Juvénal 1999. "Checking Economy". Linguistic Inquiry 30.3: 399-444.

Palmer, F. 1986. Mood and Modality. Cambridge.

Parsons, Terence 1990. Events in the Semantics of English. Cambridge: MIT Press.

Picallo, Carmen. 1984. Opaque Domains. PhD.

-- 1990. "Modal Verbs in Catalan". NLLT 8: 285-312.

Ponelis, F.A. 1991. Afrikaanse Sintaksis. Pretoria: van Schaik.

Raith, Josef 1951. Untersuchungen zum englischen Aspekt. Munchen: Max Huber.

Southworth, James 1947. "Chaucer's final -e in Rhyme", PMLA 42: 910-935.

Warner, Anthony 1993. English Auxiliaries, CUP

Wurmbrand, Susi 1998. Infinitives. MIT PhD.