On sentence structure

On sentence structure

Lingua 84 (199 1) 21 S-238. 215 North-Holland On sentence structure Tor A. Afarli* Department of Translation Studies, Agder College, Kristiansand...

1MB Sizes 1 Downloads 69 Views

Lingua

84 (199 1) 21 S-238.

215

North-Holland

On sentence structure Tor A. Afarli* Department of Translation Studies, Agder College, Kristiansand. Norway

1. Introduction In this article I discuss some questions pertaining to the analysis of sentence structure generally, and to the analysis of sentence structure in Norwegian specifically. In section 2 I discuss putative constraints on functional projections. Specifically, it is assumed that it is a constraint of X/-theory that no projection exists, unless it is the projection of some lexical or functional element at D-structure. In section 3 I discuss Norwegian syntax in the context of a ‘no functional element, no functional projection’ constraint. It is argued that one important consequence of the adoption of this constraint is that the verb must be assumed to acquire inflectional features (tense in Norwegian) by feature transfer via government, not as is often assumed, by adjunction via head to head movement. The analysis is also briefly extended to Icelandic syntax. Section 4 discusses the thesis that corresponding occurrences of different clause types partially show a common structure. It is argued that the phrase structure of a full clause is composed of a predicational or thematically relevant part, i.e. the lexical projection, and an inflectional part, i.e. the functional projection(s). The structure of a corresponding small clause is revealed by erasing the functional projection(s). Conversely, a full clause, say a tensed main clause, is argued to contain a small clause structure as a subpart. Section 5 concludes the paper.

*

Audiences

at workshops

in Lund and Oslo gave me valuable

this paper. Thank you. I would also like to thank Platzack, and Eirikur Riignvaldsson.

0024.3841/91/$03.50

0

1991 -

Lars Johnsen,

Elsevier Science Publishers

comments Torbjarn

to earlier versions Nordgard,

B.V. (North-Holland)

of

Christer

216

T.A. Afarli / On senkxce

2. Constraints on functional

structure

projections

Recently several linguists have assumed that inflectional features or affixes like tense and agreement, previously thought to occur under INFL, should exist as projections of their own, conforming to the principles of X-theory. These projections are often called functional projections as opposed to lexical projections. Several such functional projections have been proposed. Thus Pollock (1989) assumes the existence of CP, TP, AgrP and NegP. Others employ a subset of, a set including the subset of, or a superset of these projections, occasionally including Aspect Phrase (AspP) or Mood Phrase (MP), see for instance Ouhalla (1988), Chomsky (1989), or Laka (1989). The primitive features [ + / - N, + /-VI are usually assumed to define the lexical categories N, V, A, and P. Lexical categories are thus very constrained, in both their number and their internal composition. On the other hand, functional categories as currently employed seem to be quite unconstrained. We could imagine that functional categories, like lexical categories, were defined by a set of primitive features, constraining their number and internal composition, such that a set of primitive features would define a universal set of functional categories. However, here I want to propose and investigate another type of constraint, namely a constraint on the occurrence of a given functional projection in a construction type in a language, see (1): (1) No functional projection exists unless it is the projection of a functional

element at D-structure. The adoption of (1) has quite far-reaching consequences. For instance, (1) excludes the clausal phrase structure proposed in Chomsky (1986). There IP is the projection of tense and/or agreement and thus conforms to (1). But the C-projection (at least in main clauses) is not the projection of any functional (or other) element and does not conform to (1). At D-structure the Cprojection is empty and only provides a landing-site for verb movement. Also, (1) excludes the clausal phrase structure proposed for Mainland Scandinavian (including Norwegian) in Holmberg and Platzack (1988). In their analysis the tense element projects as a CP, which is OK, but the IP is generated as a projection devoid of any inflectional (or other) content, and it is therefore excluded by (1). The assumption of (1) can be defended in the following way. In a syntax employing (1) the functional phrase structure can be generated by letting X’theory apply to functional elements. The sequencing of functional projections

T.A. kfarli

/ On sentence structure

217

will be a consequence of the selectional properties of the functional elements occurring (cf. e.g. Ouhalla 1988). On the other hand, there seems to be no way to generate empty functional projections (like the main clause C-projection in Chomsky 1986 or the I-projection for Mainland Scandinavian in Holmberg and Platzack 1988) without the recourse to explicitly stated phrase structure rules. Thus the assumption of principle (1) furthers the aim of Stowell (1981) and related work, of abandoning explicitly stated phrase structure rules. In what follows, I want to investigate some consequences of the adoption of principle (1) in the syntax of Norwegian. In particular I will try to determine which functional projections are in fact employed in Norwegian clauses and whether different clause types differ as to the functional projections they employ.

3. Functional projections and Norwegian syntax I want first to consider Norwegian clauses with (1) in mind, then I will add a few remarks about Icelandic. 3.1. Main clauses One important criterion for postulating the existence of some particular functional projection in a given clause type in a given language, is that the verb shows an affix that can be related to an inflectional property. Beside that, various syntactic phenomena can be construed as effects of the presence of a given functional projection, or its absence. It has also been proposed that a negation element corresponding to ‘not’ heads a projection of its own, namely a NegP, even though it might not be a verb affix, cf. e.g. Pollock (1989). As regards Norwegian syntax, I know of no evidence that the adverb corresponding to ‘not’ heads a projection of its own which can be treated on a par with functional projections. Thus I want to adopt the more or less standard assumption that these adverbs are adjoined constituents. As for verb affixes, standard Norwegian only indicates one inflectional property, namely by means of its tense affix. There are no agreement, aspect, mood, etc. affixes:l 1 In some dialects passive participles show agreement affixes, see e.g. Christensen (1988). The finite auxiliary verb only shows a tense affix even in these cases.

and Taraldsen

T.A. Afarli / On sentence struc‘turr

218

(2) Johan kjor + te bil Johan drive + PAST car ‘Johan drove a car.’ Tentatively, I take this to suggest that Norwegian clauses have only one functional projection, namely the projection of the tense element. I call this projection the T-projection. I now want to consider whether or not this assumption is consistent with the general syntactic properties of Norwegian clauses. Temporarily assuming that the tense affix itself is generated under the Tprojection, the putative phrase structure of a sentence like (2) is indicated in (3) where the verb has moved to T and got adjoined to the tense affix. (3) [TP Johan However,

[T. [T kjor-i + te] ti bil ]]

we can also have the following

permutations:

(4a) Kjorte Johan bil? Drove Johan car ‘Did Johan drive a car?’ (4b) Bil kjorte Johan. Car drove Johan ‘A car Johan drove.’ On the assumption that the T-head selects a sentential subject + predicate structure, dubbed X in (5) these patterns can be explained as the effects of movements to T and/or to [Spec,TP]. (5) [TP [T, [T -te] [x Johan

kjsr- bill]]

Thus (2) is derived by moving the verb to the affix, and moving the subject to [Spec,TP]; (4a) is derived by moving only the verb; and (4b) is derived by moving the verb and the object. Indeed, Norwegian is a typical V2 language and this exemplifies the general pattern of derivation of main clauses in the language. One important consequence of the assumption that Norwegian employs only one functional projection is that the subject cannot be assumed to be base generated as the specifier of some functional projection, T in this instance. Assuming all categories to be endocentric (there is no exocentric S), the minimal assumption is that X in (5) is a projection of V, i.e. a VP. But

T.A. Afarli / On sentence structure

219

then (4a,b) show that the subject must be generated somewhere inside or ‘attached’ to this VP, because the subject occurs to the right of moved verb, which is situated in T. I tentatively assume that the subject is adjoined to VP, cf. e.g. Manzini (1988) or Hellan (this volume). (The position of the subject will be discussed in section 4.3.) In fact, this assumption is in line with the basic assumption made by many authors that Theta-roles are assigned within the projection of the Theta-role assigning item, see e.g. Hoekstra (1984). This assumption severely constrains the assignment of Theta-roles, and it independently excludes that the subject can be base generated in the specifier position of some functional category. We now have the following D-structure for main clauses in Norwegian (sub(ject) and obj(ect) positions are shown):

(6) LTPLT.LTafVLp sub LP LvT Lvverb1oh 11111 The VP-clause is the complement of T, the tense element heads the Tprojection, and T serves as the landing-site for verb movement. The [Spec,TP] position serves as a landing-site for movement of maximal projections. 3.2. Subordinate clauses Consider now the simple subordinate clause in (7): (7) Vi sag at Johan kjsrte bil.

‘We saw that Johan drove a car.’ In addition to having tense inflection on its verb, the subordinate clause is introduced by a complementizer. Observing principle (l), this fact might lead us to propose the following structure:

(8) LP [C L 4

[TP lT, LT

4

["P

Johan LP

he [V

kW1 bill 111111

Here two functional projections are postulated, one the projection of the complementizer, the other the projection of the tense affix. However, there is strong evidence that this structure cannot be correct. Consider the distribution of a sentence adverbial (SA) like heldigvis ‘luckily’ in a main clause as opposed to a subordinate clause (this is the main pattern;2 the distribution of SAs will also be discussed in section 4.3): *

The pattern

is somewhat

idealized.

Also, some SAs have idiosyncratic

distribution.

220

T.A. Afarli / On ,sentence .struc~ture

(9a) Johan kjorte heldigvis bil. Johan drove luckily car ‘Johan luckily drove a car.’ (9b) *Johan heldigvis kjorte bil. (SC) *Heldigvis Johan kjorte bil. (9d) *Johan kjsrte bil heldigvis. * .. ... ‘ (lob) ::: (10~) * . . . (10d) * . . . (10a)

trudde at Johan kjorte heldigvis bil. believed that Johan drove luckily car believed that Johan luckily drove a car.’ trudde at Johan heldigvis kjorte bil. trudde at heldigvis Johan kjorte bil. trudde at Johan kjorte bil heldigvis.

In (9) the distribution of the SA is explained on the assumption that it is leftadjoined to the (lower) VP, cf. (6). Both the subject and the verb must move out of VP, thus leaving the SA to their right. However, assuming that (8) is the correct structure of a subordinate clause like the one in (lob), and still assuming that the SA is left-adjoined to the lower VP, we would expect the SA to end up in the same position relative to the subject and the verb as in main clauses. That is, we would expect (10a) to be grammatical and (lOb,c,d) to be ungrammatical, contrary to the facts. If we want to maintain (8) as the structure for the subordinate clause, we would have to assume that the SA is left-adjoined to T’ in subordinate clauses. But then we have to argue why this is not allowed in main clauses, generating the ungrammatical (9b). In short, we would have to arbitrarily stipulate that the SA adjoins to the (lower) VP in main clauses and to T’ in subordinate clauses. I will follow Chomsky (1986) in assuming that adjunction is only to maximal projections. Furthermore, there is no reason to expect that an SA adjoins to different categories in main clauses and subordinate clauses. Specifically, I will assume that the SA adjoins to the lower VP both in main clauses and in subordinate clauses. On this assumption, the pattern in (10) shows that the verb does not move out of VP in subordinate clauses, because in the grammatical structure (lob) the verb is situated to the right of the SA. Leaving aside the question about tense for the moment, consider now the possibility that the complementizer heads the T-projection, i.e. that the complementizer fills the same head position at S-structure as the finite verb does in main clauses. This immediately explains the distribution of the SA in

main clauses versus subordinate clauses in (9) and (10). And indeed this is the standard view on V2 languages: a complementizer in a subordinate clause, and the finite verb in a main clause fill the same position, which may be considered the clause head. I adhere to this view, and I therefore conclude that (9a) and (lob) have the following structures: (11) [TP Johani

[TO[T kjorte,]

(12) [Tp [T, [= at] [vr Johan

[vp t, [vr heldigvis [vp heldigvis

[vp [v vk] bil . .

[vp [v kjorte] bil . . .

(11) and (12) show that the phrase structures of main and subordinate clauses are parallel, the main difference is related to movement. To explain the filling of the T-position, I want to adopt, without further motivation, the analysis defended in Holmberg (1986) that verb movement and complementizer insertion are triggered by the requirement that the clausal head must be either a predicate or an argument. 3.3. How does the verb acquire a tense a@x? Assuming (11) and (12) to be the correct structures of the two clause types in question, (12) poses a serious problem as regards tense. If the complementizer is inserted in the T position, the verb obviously cannot acquire its tense affix by movement to T. Moreover, if the tense affix itself is generated under the T-projection, we would expect the inserted complementizer to acquire the tense affix in subordinate clauses, contrary to the facts. Holmberg and Platzack (1988) and Pollock (1989) meet similar problems in their analyses, the first in the case of subordinate clauses in Mainland Scandinavian, the latter in the analysis of English. They propose to use ‘affix hopping’ or ‘lowering’ as a kind of ‘last resort’ device in the problematic cases of stem-affix merging, and of course that could be a possibility for the analysis presented here too. However, it seems to me that ‘lowering’ movement analyses should be excluded in principle because they generally imply a significant weakening of the constraints on movement. Instead I want to propose that the verbal stem and the tense affix are not merged via head to head movement in a subordinate clause like (12). Rather, I take the tense affix on the verb to be the spell-out of a tense feature assigned by a non-overt tense-assigning element of the category T to the verbal stem (much in the same way as a verb or preposition is standardly assumed to assign case to an NP, with the case affix the spell-out of the case features

222

T.A. &arli / On sentence smxiure

assigned). And because there is no evidence that the verb acquires its tense affix by different means in main and subordinate clauses, I will take the slightly more radical view that the verb acquires a tense affix in this way both in main and subordinate clauses. I formulate this assignment device as ‘feature transfer’ constrained by government, cf. Borer (1984: 15-27). 3 In (11) or (12) T governs V and the assignment of the tense feature can take place as expected (given an appropriate conception of the phrase structure of auxiliary verbs, the government requirement will secure that the tense affix can only occur on the first verb in a sequence of auxiliaries and main verb.) I also assume that Nominative Case is assigned by finite T via government, where T is finite when the tenseassigning element is specified as an assigner of past or present. 3.4. Selection The thesis that the same types of functional heads have different selectional properties in different languages has been explicitly defended by e.g. Ouhalla (1988) or Laka (1989). As argued by Laka (1989), the T-projection tends to be the highest functional projection, and in languages that employ two or more functional projections, this means that the T-head selects another functional projection which in turn selects the VP-clause. However, in Norwegian there is only one functional projection, the Tprojection, that selects the lexical VP-clause. It seems that T only selects VP; furthermore it seems that it does so obligatorily: (13) T obligatorily selects VP (in Norwegian) (13) means that if there is a T-head, then it must govern a VP. This explains several facts of Norwegian clause structure.

3 Once ‘feature transfer’ is adopted, explained in a uniform way. Consider

the traditional (i).

affix hopping

phenomena

in Aux can be

(i) Jens kan ha sett ein bjorn. ‘Jens may have seen a bear.’ As is very well known, different auxiliary verbs demand specific forms of the following verb. A modal verb demands that the following verb is in the infinitive, etc. Assume that each auxiliary verb heads its own V-projection.

I propose

that T governs

kan and assigns

tense, km governs

ha

and assigns a feature that is spelled out as infinitive, and ha governs serf and assigns a feature that is spelled out as past participle. This yields a very uniform analysis of tense and the other affix hopping

phenomena,

much in the spirit of the traditional

Standard

Theory

analysis,

223

T.A. &ark / On sentence structure

First, it explains why tensed clauses must contain a verb, i.e. why only verbs acquire tense affixes. This is so since tense originates in T and T must select VP, which has to be the projection of a verb. Assuming that declarative main clauses must be tensed, contrasts of the following kind are thus explained : (14a) Marit var irritert. ‘Marit was irritated.’ (14b) *Marit irritert. ‘Marit irritated.’ The presence of the copula is forced by (13) to provide a V-projection for T. Similar reasoning applies to the subordinate clauses in (15) : (15a) Eg sag at Marit var irritert. ‘I saw that Marit was irritated.’ (15b) *Eg sag at Marit irritert. ‘I saw that Marit irritated.’ In the analysis pursued, the association between the complementizer and a Tprojection is of the strongest possible kind: the complementizer is only inserted under a T-projection. Thus the presence of the complementizer implies the presence of a T-projection, which, by (13), implies the presence of a VP. Thus the ungrammaticality of (15b). (13) also explains why clauses in an AcI environment need not contain a verb : (16) Frank ser Marit irritert. ‘Frank sees Marit irritated.’ In section 4 I will argue that small clauses, including AcI small clauses as in (16), contain no functional projection at all, hence no T-projection. Thus, no VP is required. The perception verb selects a bare A-projection in this instance. 4

4 Note, however, that the perception verb may also select a VP (of course occurrence of a VP without a governing T head): (i) Frank ser Marit vere irritert. ‘Frank

sees Marit

be irritated



(13) allows

for the

The line of thought followed here also seems to indicate that the infinitival marker d, like the complementizer at, is an element inserted under the Tprojection. Simply put, the presence of the infinitival marker demands the presence of a VP, just like the presence of the complementizer does: (17a) Marit prover a vere irritert. ‘Marit tries to be irritated.’ (17b) *Marit prover a irritert. ‘Marit tries to irritated.’ On this analysis the infinitival marker is a complementizer and must be assumed to yield a non-finite T-projection, hence the absence of a phonetically expressed subject in (17). s Now consider a new set of data which pose a problem for (13). Norwegian allows main clause word order in that-clauses in restricted circumstances, i.e. when the clause is selected by a factive verb (see (18a, b)): (18a) Eg veit at bil kjorte Ola. I know that car drove Ola ‘I know that a car Ola drove.’ (18b) Eg veit at Ola kjorte heldigvis bil. I know that Ola drove luckily car ‘I know that Ola luckily drove a car.’ (18~) *Eg tvilar pa at bil kjorte Ola. I doubt on that car drove Ola ‘I doubt that a car Ola drove.’ (18d) *Eg spurte om bil kjorte Ola. I asked if car drove Ola (18d) ‘I asked if a car Ola drove.’

5 The existence of structures like (i) (beside structures like (ii)), seems to indicate that ci cannot generally be considered a T-element, assuming that a sentence adverb cannot adjoin to TP: (i)

Han vil love ikkje Q banne. ‘He will promise not to swear.’

(ii) Han vil love & ikkje banne. ‘He will promise to not swear.’ Tentatively, I might adopt an analysis where B is base-generated as adjoined to VP. Thus, d is adjoined to the lower VP in (i), whereas it is in the T-position in (ii). See also Pollock (1989: 375).

T.A. Afarli 1 On sentence structure

225

(13) stipulates that T obligatorily selects VP in Norwegian. But main clause word order in that-clauses seems to demand a T-head selecting a TP: (19) [TP [T atI LP b& L

k.i~rhJ [ Ola [Vvkl ti . . .

Thus, it seems that T sometimes selects VP and sometimes TP, falsifying (13). However, there appears to be a generalization underlying this seeming anomaly. In (19) [= at] governs a projection headed by a verb at S-structure (the lower TP), and it does so in ordinary that-clauses too (where it governs the VP). The generalization is that T selects a projection possibly headed by a verb. There are two projections a verb can occur in, namely a V-projection and a T-projection (the latter as a result of movement). Thus, if we instead of (13) stipulate that T obligatorily selects a projection possibly headed by a verb, rather than simply a VP, we predict that T may select VP or TP. I assume that it is the latter possibility that occurs in cases like (18). There is additional evidence that T must govern a verb rather than a VP. Consider VP topicalization as in (20): (20a) Kjorte bil er det sikkert at Johan gjorde. Drove car is it sure that Johan did ‘Drove a car it is sure that Johan did.’ (20b) *Kjorte bil er det sikkert at Johan. Drove car is it sure that Johan In (20b) [= at] presumably governs the empty VP. However, as shown by (20a), a dummy verb must be inserted. This may be interpreted in the following way. It is not sufficient that [= at] governs the category VP, it must govern a verb. The insertion of the dummy verb gjere is a consequence of the revised selectional properties of T. 3.5. Functional projections

in Icelandic

Facts from Icelandic seem to add support to the analysis defended here. In Icelandic the verb shows both tense morphology and agreement morphology.6 This suggests that Icelandic has two functional projections, a tense 6 In subjunctive clauses it also shows mood morphology, in such clauses. Here only indicative clauses are considered.

suggesting

the existence

of a MoodP

226

T.A. Afarli 1 On sentence sfruclure

projection (TP) and an only one, the TP. There is at least one opposed to Norwegian, scope is unimportant in Sigurasson 1989).

agreement projection

(AgrP), where Norwegian has

hard fact that strongly suggests that Icelandic, as employs two functional projections (their relative this context). Consider the data in (21) (taken from

(21a) Hefur Maria ekki sea Olaf. ‘Have M. not seen Olaf.’ (21b) Maria hefur ekki s&3 Olaf. ‘M. have not seen Olaf.’ (21~) *Maria ekki hefur sea Olaf. ‘M. not have seen Olaf.’ (21d) . . . aa Maria hefur ekki sea Olaf. that M. have not seen Olaf.’ (21e) *‘:I: a6 Maria ekki hefur s&5 Olaf. ‘ . . that M. not have seen Olaf,’ In Icelandic, the sentence adverbial must follow the finite verb both in main clauses and in subordinate clauses. Given that the D-structure positions of the subject (adjoined to VP) and the sentence adverbial (adjoined to lower VP) are identical in Icelandic and Norwegian, these data seem to show that verb movement takes place both in main clauses and subordinate clauses in Icelandic. Thus Icelandic is different from Norwegian, where verb movement only takes place in main clauses. This difference is easily explained if Icelandic has two functional projections, as is indicated by the fact that the verb shows both tense and agreement morphology.’ Consider (22), where the two functional projections are called X and Y (I assume that X is the clausal head where the verb is moved in main clauses and the complementizer is inserted in subordinate clauses):

(22) LP Lx,Lx1 [YP [Y,[y 1 [VP sub[VP SA [VP ...

’ It would also be explained on the assumption that the complementizer heads a functional projection of its own in Icelandic, as opposed to Norwegian, see the discussion in section 3.2. However,

since there

is independent

reason

to believe that

Icelandic

employs

projections, there is no reason to assume that Norwegian and Icelandic projection of their complementizers, to account for the pattern in (21).

differ

two functional regarding

the

T.A. .dfarli / On sentence structure

227

Still assuming feature transfer of inflectional properties, the verb has to move out of VP, to the Y-head, to come in a position where it can be assigned inflectional features from the X-head, whether or not this head position is occupied by the verb or the complementizer. This is due to the government requirement on feature transfer. Therefore, the sentence adverbial will always end up to the right of the verb. In contrast, where there is only one functional projection, as in Norwegian, we saw that the verb will remain in situ in subordinate clauses, accounting for the difference in distribution of the sentence adverbial in main and subordinate clauses in Norwegian. * We now briefly turn to the question of the identity of X and Y in (22) i.e. we must consider which values, T or Agr, the variables X and Y have in (22). Holmberg and Platzack (1988) assume (notational differences aside) that X=TandY=Agr: (23) [TP[T, [T 1 [spry [~pr, [Asr 1 [VP sub[VP SA [VP ... In an analysis in the spirit of Holmberg and Platzack (1988) this structure may be motivated because it facilitates a particular account of the null subject property in Icelandic. Icelandic has empty expletive subjects and that can be *

Icelandic

AcI-constructions

order of verb and sentence (ia) fig sa Mariu

show another adverbial.

pattern

than tensed clauses

(i) shows a typical

as regards

the relative

example:

ekki lesa bokina.

‘I saw M. not read the book.’ (ib) *fig sa Mariu

lesa ekki bokina.

‘I saw M. read not the book.’ Here the sentence adverbial must precede constructions, sentence

like their Norwegian

adverbial

adjoins

has no landing-site

adverbial.

very well be on the wrong

that

track,

that (ia) is derived

really originates sentence

adverbial,

are bare VP-clauses,

has pointed

however.

Sigurasson

(1989) gives data

Shift (see Holmberg not easily

adjoined

to the

out to me, this analysis

may

like the one in (ii), and

1986). so that the sentence

clause in cases like (ia). SigurBsson

AcI-

the verb

and the verb must then remain

Holmberg

are generally

if Icelandic

(see section 4). That is, the

As Anders

by Object

in the matrix adverbials

be explained

are bare VP-clauses

to the (lower) VP, but since AcI-clauses

to the left of the sentence

right of the sentence suggests

the verb. This might

counterparts,

adverbial

also gives other data that show

to the VP in AcI-constructions

in

Icelandic, (ii) *kg hafai

se8 Mariu

ekki lesa bokina.

‘I had seen M. not read the book.’ If this is the right interpretation of the data, the contrast in (i) of course cannot be used to motivate my analysis (even though it is compatible with it), and at the present moment I would have to stipulate

that sentence

adverbials

cannot

occur in Icelandic

AcI-constructions

at all.

accounted for on the assumption that Agr is nominal and can receive the Case assigned by T. This appears to require that T governs AgrP.9 Also, the structure in (23) is consistent with the assumption that T is the clausal head both in Norwegian and in Icelandic.lO In this analysis verb movement to Agr is obligatory because the tense feature must be assigned. However, verb movement to Agr also seems to be obligatory in non-tensed infinitival clauses with an infinitival marker. This is shown in (24k(25), where the first verb must occur to the left of the sentence adverbial, just as in tensed clauses.

(244

. ..aa ‘...to (24b) * . ..a3 ‘...to

lesa ekki read not ekki lesa not read

. ..aa ‘...to (25b) * . ..a6 ‘. . . to

hafa ekki lesia bokina. have not read the book.’ ekki hafa lesia bokina. not have read the book.’

(254

9

Roughly,

assign

their analysis

bokina. the book.’ bokina. the book.’

goes like this (translated

(or license) Nominative

this requires

that the subject

Being a nominal

element,

Case.

In Norwegian

position

to my terminology): or other

is filled. However,

Agr receives Nominative

Mainland

Icelandic

If T is finite, it must Scandinavian

clause structure

languages

contains

Case. The result is that the subject

Agr.

position

need not be filled in Icelandic. Thus, Icelandic may have empty expletive subjects. The analysis also provides for Quirky Case subjects in Icelandic. Note that Agr is outside the verbal projection.

Thus Agr cannot

that Icelandic

receive a Theta-role

does not allow ‘real’ pro-drop,

from the verb. It is therefore

i.e. pro-drop

correctly

of subjects with a Theta-role,

predicted as in e.g.

Italian. lo

Note

Icelandic.

that

there

is one general

In Icelandic

the order

problem

with

the assumption

of the tense and agreement

that

T governs

AGR

in

affixes is this:

(i) stem + tns + agr In an approach a subsequent before

that analyses adjunction,

it is adjoined

acquisition

this pattern

to the agreement

of verb affixes as movement clearly

means

of the stem to the affix with

that the stem is adjoined

affix. In other

words,

it clearly

to the tense affix

indicates

that

the Agr

projection is the highest projection in Icelandic (cf. the Mirror Principle of Baker 1988). In an approach, like mine, where acquisition of verb affixes is construed as feature transfer, nothing excludes the spell-out of features to be manifested on the verb stem each time, so that the affixes are successively added to the stem from the ‘inside’. The problem is that this manner of stem-affix merging is not forced. In the feature transfer approach, affixes could as well be added from the ‘outside’.

Thus,

approach. hierarchy

It makes no definite predictions of functional projections.

in its present

form,

my approach

makes

weaker

as to the succession

predictions

than

the first

of affixes on the verb and the

T.A. A,farli / On sentence slruciure

229

If we assume that the infinitival marker occurs in T, as in Norwegian, this word order may again be explained as obligatory movement to Agr. In this case, however, obligatory verb movement cannot be motivated by a requirement that the verb must receive tense. However, in section 3.4 I suggested that T, in Norwegian, must select a projection possibly headed by a verb. The data in (24) and (25) suggest that this requirement also applies in Icelandic, i.e. the requirement is fulfilled by verb movement to Agr, both in tensed and non-tensed clauses. This explains the data and adduces additional evidence for the suggestion from section 3.4. In my analysis I have tried to relate word order differences between Norwegian and Icelandic to differences in phrase structure, namely to the assumption that Icelandic contains an Agr projection, which Norwegian does not. This goes against the approach advocated in e.g. Kosmeijer (1986), Holmberg and Platzack (1988) and Siguri)sson (1989). All these linguists seem to consider it theoretically desirable that there should be one common phrase structure for clauses in Norwegian (Mainland Scandinavian) and Icelandic (Insular Scandinavian). As far as I can see such an assumption could either be motivated on empirical grounds or on a priori, universal grounds, i.e. to the extent that it is reasonable to assume that all languages of the world share one canonical clause structure. I find the idea that all languages of the world share one rigid clause structure extremely unlikely. Thus, if I am right, the motivation for a putative common clause structure for Norwegian and Icelandic can only be established on empirical grounds. However, as I have tried to argue, the empirical material suggests the existence of different clause structures for the two language groups. This is also suggested by some difficulties that the alternative approaches encounter. For example, Holmberg and Platzack (1988) must let verb movement in Mainland Scandinavian (including Norwegian) ‘skip’ the I-head, thereby, in essence, violating the Head Movement Constraint. Sigurbsson (1989) must invoke V/I reanalysis for Mainland Scandinavian. What this indicates is really that there is an inflectional head and its projection missing in Norwegian, as compared to Icelandic. In my analysis this missing projection is the Agr-projection. In fact, this is in line with Pollock’s remark that some Scandinavian languages may have got ‘rid of [Agr] entirely’ (Pollock 1989: 396). The wider implication is that an important part of syntactic variation between languages is due to differences as to which functional projections a language employs.

230

T.A. A,f’rli / On senfence structure

4.A common structure across clause types Here I explicitly argue for the thesis that tensed clauses in Norwegian are built around a small clause structure. A tensed clause is construed as a small clause with additional functional structure. 4.1.

UTAH

Consider

and its consequences the italicized

clauses in (26) and (27).

(26) Frank ser at ho kopierer biletet. ‘Frank sees that she copies the picture.’ (27) Frank ser henne kopiere biletet. ‘Frank sees her copy the picture.’ In some sense, at some rudimentary level, the underlined clauses in (26) and (27) seem to be occurrences of the same clause type. If we adhere to this intuition, the minimal assumption is that there is, at least partly, a common basic structure underlying the relevant clauses in each instance. Assuming that there is such common structure, it may be thought to reflect the thematic relationships or the predicational relationships of the clauses. The thematic relationship between ho and kopierer biletet in (26) is the same as the thematic relationship between henne and kopiere biletet in (27). In fact, if we adopt the Uniformity of Theta Assignment Hypothesis (UTAH) proposed in Baker (1988: 46), the assumption that ho kopierer biletet and henne kopiere biletet have a common basic structure seems inescapable. (28) UTAH: Identical thematic relationships between items are represented by identical structural relationships between those items at the level of D-structure. The two clauses under discussion show identical thematic relationships and, according to UTAH, they therefore show identical structural relationships at D-structure. I interpret identical structural relationships as identical phrase structure relationships. I call that part of the D-structure of a clause where the thematic relationships are structurally manifested the bare predicational structure of the clause. The common structure referred to above is now to be interpreted as a common bare predicational structure, i.e. the claim is that the clauses under

T.A. &ark / On sentence structure

231

discussion partially share a common bare predicational structure; nothing is so far claimed about whatever additional functional structure each of the clauses might or might not employ. 4.2. Small clauses What does the bare predicational structure underlying the underlined clauses in (26) and (27) look like? I want to claim that the AcI small clause complement of the perception verb in (27) exhibits this bare predicational structure in its pure form, i.e. the AcI complement contains nothing more than the bare VP. Two conspicuous facts suggest this. First, the AcI complement cannot contain a complementizer and, second, it never contains tense. Since both types of element originate in T, their absence suggests that the Tprojection is absent. Indeed, under (1) this conclusion is forced. The fact that there do not seem to exist landing-sites for w/z-movement in AcI complements of perception verbs is consistent with this conclusion too. (29) *Vi ser kva, henne kopiere e,. ‘We see what her copy.’ I.e. the absence of w/z-movement is compatible with the absence of a Tprojection. Consider also English. There too the AcI complement of a perception verb seems to consist of the bare predicational structure only. Consider the contrasts in (30) and (31): (30a) I heard her sing. (30b) *I caused her sing. (31a) *I heard her to sing. (3 1b) I caused her to sing. These examples show that certain clausal complements involving an infinitival must contain the infinitive marker to, whereas others cannot. The AcI complement of a perception verb never contains the infinitive marker. Since the infinitive marker is commonly assumed to be generated under some functional projection, its absence in the AcI complement of perception verbs strongly suggests that these complements do not contain this functional projection at all. I conclude that the AcI complement of perception verbs does not contain

232

T.A. Afarli / On sentence struclure

any functional projections, because there seems to be nothing that such a putative projection could be a projection of. This forces the conclusion that the subject in AcI small clause complements is ‘attached’ to the projection of the predicate, VP in this instance. This is precisely the conclusion we reached for tensed main and subordinate clauses in sections 3.1 and 3.2. Specifically, I assumed that the subject is adjoined to the projection of the predicate.ll Thus the D-structure position of the subject does not seem to differ in AcI small clauses and different types of ‘full’ clauses; all these clause types contain a ‘small’ VP-clause as a bare predicational subpart, whether this is a proper or non-proper subpart. In the case of a tensed clause this is a proper subpart, because a tensed clause contains additional functional structure, namely a Tprojection (see Chung and McCloskey 1987 for a similar analysis of Irish). I conclude that an AcI small clause of the type exemplified in (27) shows a structure like (32). A corresponding ‘full’ clause has the structure shown in (33). (32) . . . [VP NP [VP [v (33) [W 1~’ [T

. ..I NP 11

. ..I [VPNP [VP[v . ..I NP 1111

Note that the lexical part and the functional part of the phrase structure of the ‘full’ tensed clause exist as different ‘layers’. This accords quite well with the pattern of language learning, where something corresponding to the lexical layer is learnt first (the V-projection in (33)). Afterwards inflectional properties (the functional layer) of the clause are mastered (the T-projection in (33)). Note that my analysis makes very strong predictions. The occurrence of a tense affix is taken to imply the existence of a T-projection. But once a Tprojection exists, we predict that there exist verb movement and ‘topicalization’ (in main clauses), because it is the T-projection which provides for landing-sites in the case of such movements. Thus there is a close connection between the fact that a clause is tensed and the fact that there are landingsites available to the left of the bare VP-clause. In fact there is some support from language acquisition for this close connection. Lundin and Platzack (1988) show that the different word orders

I1

Given

predicate,

that

the small

clause

one more argument

subject

is ‘attached’

that small clauses

to the projection

are bare predicational

of the small

structures

clause

can be given.

The subject is obviously Case marked by the matrix verb. If there existed a projection and the small clause, that would block Case assigment.

between

it

T.A. &ark

connected

to the V2 phenomenon

233

/ On sentence structure

are acquired

after or at the same time as

tense is acquired, but never before. Children seem to learn the pure predicational structure of a clause first. Afterwards they learn the tense system and different word order variations. Thus, the acquisition of tense seems to be a precondition for the acquisition of V2 and different word order patterns connected to it, exactly as predicted by our analysis. The analysis of AcI small clauses presented here is easily extended to other small clauses. Consider (34), where (34a) contains a ‘prototypical’ small clause and (34b) an AcI small clause: (34a)

Frank ser henne irritert. ‘Frank sees her irritated.’ (34b) Frank ser henne vere irritert. ‘Frank sees her be irritated.’

The only difference between (34a) and (34b) is that the small clause in (34a) is the projection of A, whereas it is the projection of V in (34b). The structure of the small clause in (34a) is shown here: (35) . . . [AP [NP henne]

[AP [A irritert]]]

The small clause subject is adjoined to the projection of the predicate, as before. In a larger perspective, one important conclusion can be drawn from the preceding discussion. In section 2 I referred to work arguing that an important source of syntactic variation between languages can be traced back to variation in which functional projections are expanded in a given language. Now it seems that syntactic variation between clause types within a language can be captured in exactly the same way. Thus in the analysis developed here a substantial part of both inter- and intra-language variation between clauses may be traced 4.3.

back to absence/presence

The position

of functional

projections.

of the subject

In this subsection I want to take a closer look at the bare predicational structure. More specifically, I shall try to determine the position of the subject in a clausal structure. Up to now I have tentatively assumed that the subject is adjoined to the maximal projection of its predicate. However, no arguments were given for this. We can start by considering different answers to the following question: What possible subject position candidates exist?

234

T.A. kfarli

/ On sentence structure

The assumption that the subject is generated in a position outside the projection of the Theta-role assigning element, has already been excluded. Thus, the task reduces to finding out what position ‘attached’ to that projection the subject might be generated in. I want to consider three possibilities: the subject is generated in the [Spec, VP] position, or more generally in the [Spec, XP] position; it is adjoined to VP (or XP); or it is generated as specifier of some ‘super-projection’ of VP (or XP) (the possibility that the position of the subject varies with different predicates might also be considered, but here I will simply assume that it does not). Given the X-scheme with specifier, head, complement, the minimal assumption is that the subject position is the specifier position. (36) [VPsub [v, [v

. ..I . ..I1

The following might indicate that this assumption is viable. In analyses where the subject is generated outside the VP (as e.g. in Chomsky 1986), the [Spec, VP] position is systematically empty. Thus, it would be only a small extension of these analyses to generate the subject in the [Spec, VP] position, and then move it out of VP. However, there are several difficulties with this assumption. First, if the subject position is the [Spec, VP] position, in order to get the correct word order, we would have to assume that an SA left-adjoins to the intermediate level projection v’: (37) . . . at [vp Johan [v, aldri [v, [v ser] bjornen I]] ‘ . . . that Johan never sees the bear.’ However, this runs counter to the restriction that adjunction is only possible to maximal projections (cf. Chomsky 1986: 6). Second, a problem occurs with so-called predicate adverbials, even when adjunction is restricted to maximal projections. Assume that predicate adverbials right-adjoin to VP, as in (38a). This would lead to problems with explaining the possibility of VP movement in cases like (38b), since the moved part does not form a constituent. (38a) Marit veit at [vp [vp Johan [v, [v ser] bjornen]] i hagen] ‘Marit knows that Johan sees the bear in the garden.’ (38b) Ser bjornen i hagen veit Marit at Johan gjer. ‘Sees the bear in the garden Marit knows that Johan does.’

T.A. &ark 1 On sentence structure

235

The remaining possibility is to adjoin the predicate adverbial to V’, ,which again violates Chomsky’s assumption and in addition forces us to assume topicalization of intermediate level projections. Third, consider the following structures (see also Stowell 1983 : 311, n. 30). (39a) We made John Mary’s husband. (39b) We consider Peter very angry. It seems that Mary’s and very should be analysed as specifiers of XP here. If so, the small clause subject (John or Peter) must be generated in a position outside the specifier position. If we wanted to claim that the small clause subject is generated in the specifier position, we would have to motivate that Mary’s and very are generated in some position more deeply embedded in XP. l2 These problems disappear if we assume that the subject is generated in some position higher than the specifier position, where V? is either VP (which means that the subject is adjoined to VP), or some ‘super-projection’ of V. (40) . . . . [“? sub [vp .. . . ]] Now sentence adverbials may be left-adjoined to VP and predicate adverbials may be right-adjoined to VP. This yields a nice symmetry for the two types of adverbials, it obeys Chomsky’s restriction, and topicalization of a main verb together with its predicate adverbial would be unproblematic. This also provides for a solution to the problem posed by (39a,b): the ‘true’ specifier is now generated in the specifier position of XP and the subject is immediately dominated by X?. Note also that sentence adverbials sometimes may occur in pre subject position in Norwegian. l3 (41a)

. . . at Jon ikkje kan komme. that J. not can come (41b) ::: at ikkje Jon kan komme. . . . that not J. can come

I2

However,

the analysis

of NPs in terms of D-projections

framework for the solution of this problem. I3 Occurrence in post subject position is nevertheless

and N-projections

the preferred

option.

could

provide

a

236

(42a)

Kan Can (42b) Kan Can

T.A. kfarli / On sentence structure

Jon aldri komme? J. never come aldri Jon komme? never J. come

This can be explained

as adjunction

(43) X 1~7 SA [v~ sub [VP . .

to V?:

111

An initial characterization of the distribution of sentence adverbials is then given by the rule ‘Left-adjoin a sentence adverbial to VP or V?’ (where adjunction to VP is the preferred option). What identity does V? have? I do not have any conclusive answer to this question, but I will briefly consider the two possibilities which have been mentioned. Among others, Manzini (1988) and Hellan (this volume) assume that subjects are adjoined to their predicates, i.e. that V? = VP. (This is in line with the analysis suggested for small clauses in Chomsky 1986: 20-21, cf. also Stowell 1983.) Manzini shows how this assumption is consistent with the principles governing movement, and she even shows how it may facilitate a simplification of their definition. On the other hand, both authors point to the putative difficulty of accounting for the uniqueness of the subject in an adjunction strategy. Koopman and Sportiche (1988) propose that the subject is the specifier of a ‘superprojection’ of V (X, more generally), such that V? = V”: (44) . . . . [v” sub [vp . . . . ] ] V” is called the maximal projection of V and VP is called the phrasal projection of V. Their proposal amounts to adding a bar level to the Vprojection. Here the subject has a unique position as [Spec, X”], specifiers like Mary’s and very in (39) will occur in the [Spec, XP] position, and sentence adverbials are adjoined either to the maximal projection or the phrasal projection (with adjunction to the phrasal projection the preferred option). I do not want to settle for any particular one of the options V? = VP or V? = V”.

T.A. &a&

237

/ On sentence structure

5. Conclusion I started with considering one constraint on functional projections, namely the constraint that no functional projection exists unless it is the projection of a functional element at D-structure. It was argued on the basis of this constraint that Norwegian tensed clauses employ only one functional projection, the T-projection, and furthermore that several facts of Norwegian clause structure are compatible with this contention. It was also argued that the ‘only one functional projection’ assumption excludes merging of the verb stem and the tense affix by means of head to head movement (instead feature transfer was adopted). It also forces the assumption that the subject of a clause is ‘attached’ to the VP at D-structure, even in ordinary tensed clauses. Thus, it was argued that Norwegian tensed clauses and corresponding AcI small clause complements share a predicational structure, i.e. in the ‘bottom’ of every tensed clause, there is a small clause structure. The tensed clause only contains additional functional structure. The perspective that has emerged is this: the functional phrase structure of a clause is determined by the functional elements occurring in the clause. The functional elements that occur, or the fact that they do not occur, have important syntactic effects. The occurrence or non-occurrence of functional elements in some language accounts for a considerable degree of syntactic variation between languages (as exemplified by comparing Norwegian and Icelandic). But precisely the occurrence or non-occurrence of functional elements accounts for the bulk of the syntactic difference between clause types within a language too. In this way a very general analysis is attained.

References Baker,

M.C.,

Borer.

H., 1984. Parametric

Chomsky, Chomsky, Papers

1988. Incorporation. syntax.

Chicago,

IL: University

Dordrecht:

of Chicago

N., 1986. Barriers. Cambridge, MA: MIT Press. N., 1989. Some notes on economy of derivation in Linguistics

Christensen, agreement

K.K.

and

Press.

Foris. and representation.

MIT Working

IO. T. Taraldsen,

in Scandinavian

dialects.

1988. Ms.

Expletive

chain

formation

and

past

participle

T.A. &at-Ii / On sentence structure

238

Chung,

S. and J. McCloskey,

Linguistic

Inquiry

1987. Government,

barriers,

and small clauses

in Modern

Irish.

18, 173-237.

Hellan, L., this volume. A two-level X-bar system. Hoekstra, T., 1984. Transitivity. Dordrecht: Foris. Holmberg,

A.,

1986. Word

order

and

syntactic

features.

Doctoral

dissertation,

University

of

Stockholm. Holmberg,

A. and C. Platzack,

University

1988. On the role of inflection

Koopman,

H. and D. Sportiche,

Kosmeijer,

W.,

1988. Subjects.

1986. The status

University of Trondheim. Laka, I., 1989. Constraints

M.R.,

Constituent Ouhalla, J., London. Pollock,

on sentence

negation.

1988.

Constituent

structure

structure.

Dordrecht:

Foris.

1988. The syntax

J.-Y.,

Inquiry

syntax.

WPSS 42,

Ms., UCLA.

of the finite inflection

1989. Verb

of head

movement,

and

In:

Doctoral

grammar,

and

Papers

of verb of Lund.

locality.

movement. universal

in Icelandic

MIT Working

Lundin, B. and C. Platzack, 1988. The acquisition subordinate clauses in Swedish. WPSS 42, University Manzini,

in Scandinavian

of Lund. Swedish.

WPSS

in Linguistics

inflection,

verb

A. Cardinaletti

dissertation,

10. second,

and

et al. (eds.),

University

and the structure

26,

College

of IP. Linguistic

20, 365424.

SigurBsson,

H.A.,

1989. Verbal

syntax

and case in Icelandic.

Doctoral

dissertation,

Lund. Stowell,

T., 198 1. Origins

Stowell,

T., 1983. Subjects

of phrase across

structure.

categories.

Doctoral

dissertation,

The Linguistic

MIT.

Review 2, 285-312.

University

of