<<

TOPICALIZATION IN GERMAN: MOVEMENT TO COMP OR BASE-GENERATION IN TOP?

Anna Cardinaletti üniversitä di Venezia

1. in German is usually referred to as Vorfeldbesetzung: the topicalized constituent (not necessarily a subject) appears su­ perficially in the so-called Vorfeld, namely to the left of the fi­ nite . A topicalized structure in German is thus an independent sentence with the finite verb in second position (v/2) at S-structure, as shown in (l):

(l) a. diesen Mann habe ich lange nicht mehr gesehen "this man have I long no more seen" b. ich habe diesen Mann lange nicht mehr gesehen

Two different approaches to German Topicalization are found in the literature. Let us consider first the proposal originally made by Thiersch (1978), who formulates the rule system in (2) and the sen­ tence structure in (3):

(2) R, : ...V r „. -, —> V... ' 1 (+tensej Thiersch (1978:38) 2 (-verb J

(3)

COMP Thiersch (1978:164)

den Mann

The rule R in (2) moves any X constituent to the COMP position on the left of the COMP position, to v/hich the finite verb has already been moved (from its underlying sentence-final position)2. This sug­ gestion iè based on the fact that Topicalization in German is a much more frequent and therefore less marked construction type than, for _ 20-3 _

instance, in languages like English. Furthermore, it can involve 3 almost any type of item with very few exceptions . Topicalization can therefore be accounted for in German by a general movement rule. Let us refer to this analysis as the movement analysis of Topicaliz­ ation. This analysis has, however, many important consequences which should receive more attention than is usually paid to them. The top­ icalization rule in German should be on a par with wh-movement in languages like English, for it moves something to COMP, even if it appears to move also non-wh-words as well as wh-words. This means, however, that COMP cannot be marked £+whJ in German, as appears to be the case for other languages, but it must receive different fea­ tures. Furthermore, since the trace which is left by the rule is in an A-position, is Case-marked and locally A-bound, it must be re- garded as a variable (that is, corresponding to a wh-trace in Eng­ lish, even if it is not always a wh-trace). But there is some evi­ dence in Italian, that it is not sufficient for a variable to be X- bound, as Cinque (1985) has pointed out; the notion of operator-bin- ding must be added to the definition of variable . Sentence (4) is in fact impossible: (4) * LN_ GianniJ ho visto L ej Cinque (1985:5) i i "Gianni (i) have seen"

The NP Gianni in TOP, even if it A-binds the empty category J_eJ, cannot identify it as a variable, because it does not qualify as an operator. Since the empty category does not qualify as any other type of empty category either, the sentence is correctly ruled out. Since a sentence like (l) is possible in German, we must assume that in German, differently from other languages, any X (also a non-wh-word) can qualify as an operator for its own trace. Furthermore, since some parasitic gap constructions seem to be possible in German with a topicalized item , this would mean that -201 -

a full NP might qualify as an operator for a parasitic gap:

(5) diesem Kann, folgte ich t. ohne e. zu vertrauen "this man followed I without trusting"

The other current analysis of German Topicalization is that first made by Koster (1978a,b) (cf. also Thiersch (1982)). He suggests that the topicalized item is base-generated in the sentence-initial n position in which it appears superficially, that is, m the X po- — 7 sition, sister-node of S, in the following structure : (6)

Koster (1878b:201)

This proposal is based on the similarity of the constructions of Left Dislocation (Linksversetzung) and Topicalization, which differ in Dutch, and in German, in the presence or absence (respectively) o£ the so-called resumptive d-word:

(7) a. die man (die) ken ik Koster (1978b:200) "that man that know I" b. diesen Kann (den) kenne ich

A topicalized structure is then to be derived from a left-dislocation structure through deletion of the d-pronoun by the rule of free de­ letion in COMP: the deletion leaves an empty category with feature

[+whjf which binds an empty category in the S-internal A-position (cf. Koster (1978b:204)). This gives plausibility to the wh-analysis of Topicalization, first proposed by Chomsky (1977) for (English) Topicalization, also for Dutch and German. Note that the fact that Topicalization is a very freauent phenom- _ 205 _

enon in German is not an argument against this base-generation anal­ ysis. In fact, a language can use a syntactic configuration in a v/ay different from its use in another language, by involving more types of items or by using it in a greater number of situations. Last but not least, Koster's analysis has been vised extensively in the case of sentential subjects, and it seems to us that it v/ould be an interesting generalization to assume it for all types of top- icalized items.

We conclude then that from a theoretical point of view the base- generation of the topicalized phrase in TOP is less costly than the movement analysis: it has the advantage of accounting for the sim­ ilarity of Topicalization and Left Dislocation in languages like German and Dutch, of preserving a generalization with other langua­ ges with regard to the sentence structure and the specification of o the COMP positions and of avoiding the parameter concerning the class of possible operators in German, which, as noted above, would be needed in an analysis like Thiersch's. 2. We v/ill now try to give some empirical evidence in favour of the base-generation analysis. We want to show that the COMP position is not open to just any type of constituents, as it is implied by the movement analysis. In order to do that, let us consider some left-dislocation struc­ tures. In the sentences (8), (9) and (lo), there is simtactic con­ nectedness (in the sense of Vat (1981)) between the phrase in TOP and the S-internal A-position: the phrase in TOP, even if in senten­ ce-initial position, behaves w.r.t. various grammatical principles as though it actually occupied the S-internal position where the d-pronoun is base-generated. For example, in (8) the dislocated phrase contains the lexical anaphor sich which is bound by the HP Franz, even if sich is outside of the governing category which con- - 206 -

tains its antecedent Franz (informally, it is on the left of its antecedent). In (9) and (10), the connectedness involves a pronoun, seinen, which can be bound by a quantifier on its right, that is, a pronoun which is outside of the c-command domain of its binder. These properties guarantee that the following sentences are to be regarded as instances of Left Dislocation , that is to say, as sentence- constructions, and not as some kind of discourse- grammar phenomena :

(8) a. [einem Anschlag auf sich.J, I demj ist Franz, t. gestern L i j J knapp entgangenJ "an attack on himself that is F. yesterday with great difficulty escaped" b. | L einem Anschlag auf sich.J, Franz. ist t. [ demj gestern

knapp entgangen!

c. [einem Anschlag auf sich.J, gestern, ist Franz. LpdemJ t. knapp entgangen] (9) a. ("„seinen. Wagen], [ den] wäscht jeder, t. einmal im Monat NP 1 NF -> 1 J "his car it washes everyone once a month" b. [seinen. WagenJ, gestern hat [ den]jeder, t t. gewaschen 3 X 3 "his car yesterday has it everyone washed"

(10) a. ich glaube,[= (Npseineni WagenJ, 1^ denj hat jeder, gestern t. gewaschen! J -1 "I think ..." b. ich glaube,[= [seinen. WagenJ, gestern hat [ denj jeder, t t. gewaschen]

Why are sentences (8b,c), (9b), (10b) ungrammatical? Note that Thiersch's analysis would incorrectly rule them in as grammatical: in fact, the left-dislocated element would be base-generated in TOP and any other element of the sentence could have been moved to COMP 13 2 - 207. -

In order to account for the ungrammatical status of the starred sentences, we could examine the possibility of setting up a condi­ tion on a left-dislocation structure to the effect that the d-­ noun must be obligatorily moved to COMP in order to be adjacent to the coreferential element in TOP. But this does not seem to be the right way to rule them out, since sentences like the following, 14 where the d-pronoun remains in the so-called Mittelfeld , are gram­ matical:

(11) einem Anschlag auf sich., wer. ist dem gestern knapp entgangen? "who" (12) seinen. Wagen, wer. hat den gestern nicht gewaschen? "his car, who has it yesterday not washed" (13) a. den Artikel über sich., wenn Hans, den doch ironisch auf- li fassen könnte! "the article about himself, if H. it ironically understand could" b. den Artikel über sich., könnte Hans, den doch ironisch l l auffassen! In (11) and (12) a w(h)-pronoun has been moved to COMP instead of the d-pronoun; in (13), sentences with the^verb respectively in end position (V/E) and in first position (v/l), the d-pronoun cannot be moved to COMP for independent reasons. As a second hypothesis, we could appeal to the weight of the el­ ement which is moved to COMP . One could imagine that full NPs such as Franz in (8b) or adverbs such as gestern in (8c), (9b), (10b) are heavier than the d-pronoun, which therefore must be priorily moved to COMP, if present (sentences (11) and (12) would be possi­ ble because a w(h)-pronoun would be as heavy as a d-pronoun). But why are sentences (14b) and (14c) impossible, where a personal pro­ noun, er and ihm respectively, has been moved to COMP instead of the d-pronoun? These pronouns should be as heavy as a d-pronoun: - 208 -

(14) a. ich glaube, |= [„„seinen. WagenJ, /„_ den hat Fritz, seinem S NP l J NP. J i Vater t. geliehen] u "I think, his car it has F. to his father lent" b. ich glaube,[= seinen. Wagen, er. hat t. den seinem Vater

geliehen] „he„ c. ich glaube,[= seinen. Wagen, ihm. hat den Fritz, t. geliehen/ S 1 J 1 j -J "to him" The only possible explanation must rely on the particular status of the d-pronouns and the w(h)-pronouns (the only elements that can be moved to COMP in these constructions) in contrast to all the oth­ er elements. They can be S-structure operators, in the sense that they can enter an operator/variable configuration already at S-struc- ture, whereas all other lexical elements cannot. If we assume that COMP is an operator position, a landing-site for operators only, the ungrammaticality of (8b,c), (9b), (10b), (l4b,c) follows straight- forwardly 15 This step is justified by the fact that COMP is marked in German with the feature £dj in addition to the feature [w(h)J . Relative d-pronouns and relative and interrogative w(h)-pronouns form the class of designated elements that can appear in COMP in German and therefore undergo movement to COMP. The d-pronouns which occur in left-dislocation structures must also be associated to this class 17 As a consequence, we would expect that the syntactic behavior of the d-pronouns in these constructions is different from that of the demonstrative d-pronouns, which must be regarded as R-expressions. 18 This is in fact the case; consider the sentences (15) and (16) : (15) ihn. (selbst), den. glaubt Hans, bespitzelt die Polizei e. "him self him thinks H. tails the police " ace ace nom r nom (16) den Hans.^ glaube ich den. mag Julia nicht e. "the H. think I him likes J. not" r ace ace nom - 209 -

In these sentences the d-pronoun is not subject to principle (c) of Binding theory. On the contrary the demonstrative d-pronoun usually 19 is ; the sentences of (17) are in fact ungrammatical:

(17) a. den. glaubt Hans, bespitzelt die Polizei e. x b. Hans, glaubt, den. mag Julia nicht e.

If the d-pronoun belongs to the same class of the w(h)-pronouns, we can make the right prediction that, if in a sentence there is a d-pronoun and a w(h)-pronoun, either of them can be moved, as in the case of two w(h)-pronouns; cf. (18) and (19) respectively (see also (11) and (12) above):

(18) a. den Hans, den hat wer gesehen? "the H. him has who seen" b. den Hans, wer hat den gesehen? (19) a. wer hat was gesehen? "who has what seen" b. was hat wer gesehen?

Consider finally sentences like (20), which seem to be counter­ examples to what we have been saying so far, in that they show a full NP instead of a d-pronoun after the leftmost item:

(20) a. den Hans, [diesen IdiotenJ mag ich nicht "the H. this idiot like I not" b. ich glaube, den Hans, [diesen IdiotenJ mag sie nicht "I think, the H. this idiot likes she not"

Since sentences (21) are also possible,

(21) a. den Hans, [diesen IdiotenJ, den mag ich nicht b. ich glaube, den Hans, [diesen Idioten], den mag sie nicht where again there is a d-pronoun, we will regard the bracketed NP in (20) and (21) as in apposition to the leftmost NP and therefore 20 (20) as an instance of Topicalization and not of Left Dislocation - 2110 -

The grammatically of the sentence is lowered when the bracketed NP can less easily be interpreted as being in apposition:

(22) den Hans, [diesen Kann], (den) mag ich nicht

To summarize: If COMP must be specified as an operator position, it follows that a topicalized item must be base-generated in TOP position and that it cannot be base-generated in S-internal posi­ tion and then moved to COMP position. This means that the base-gen­ eration analysis of German Topicalization is superior to the move­ ment analysis: the second cannot explain why sentences (8b,c), (9b), (10b) and (14b,c) are ungrammatical, because it makes no distinction 21 between wh- and non-wh-phrases with regard to the movement to COMP

3. The German sentence structure that we assume is then, in agreement with what has been recently proposed by Chomsky (Chomsky (forthcom­ ing)), that shown in (23):

(23)

TOP

specCOMP (=C0MP ) C (=S')

COMP (=C0MP ) Ï ( = s)

NP Ï

VP

A question arises with regard to the status of the S-node (in the S-system), which we have tentatively called C', pointing out that it must be regarded as a further projection of COMP and not as a projection of TOP. In fact, C'is also intuitively a sentence and not a projection of the element which appears in TOP. Further­ more, if a C' appears in TOP itself, the subcategorization selection - 211 _ holds between the matrix verb and the head of C', that is the com­ plementizer in its COMP; the two elements appear underlined in the following examples:

(24) a. |= i= dem Hans, dass du dem hilfst! 1= (das) ist allen C C ~^~~~ "^ L» ~~—— bekann tll "the H.. that you him help that is to everybody well-known" b. Is,Is, dem Hans, dass du dem hilfstj [= (das) glaube ich nicht]] "the H. that you him help that think I not" c. [=, l=, dem Hans, ob du dem hilfstj [= (das) frage ich mich C C ^ iJ ständig] "the H. whether you him help that wonder I all the time" In the case of a left-dislocated C', the resumptive pronoun is that of a sentence, das, and not that of the phrase which appears in the TOP position of C*; the sentence (25) is in fact impossible: (25) dem Hans, dass du dem hilfst, dem ist allen bekannt

The case of a relativized NP in TOP contrasts with (25); in this case, the d-pronoun "resumes" that NP:

(26) dem Kerl, dem du hilfst, dem vertraue ich nicht "the guy that you help him trust I not"

There does not seem to be, then, any structural parallelism between a topicalized structure and a relativized NP. If C' is a further projection of COMP, it may be possible to as­ sume that C' acts a$maximal projection when present, but we will not pursue this issue here. C' must be regarded, however, as an optional projection of COMP (cf. also Chomsky (19 81:140)). In German, then, the C' projection will correspond not only to left-dislocation structures like (27): (27) a. ^einem Anschlag auf sich, dem ist Franz gestern knapp ent­ gangen ( = 8a) - 212 -

b. den Artikel über sich, verin Hans den doch ironisch auf­ fassen könnte! (=13a) c. den Artikel Über sich, könnte Hans den doch ironisch auffassen! (=13b) but also to a topicalization structure, including the case of a main sentence with subject first. This amounts to saying that a topic­ al i zed structure in German has in some senses one projection more than the w(h3-sentences and the embedded sentences, which correspond to C (indeed in all languages). However, a parameter can easily be set up to the effect that the 5' option is expjM$^J.n German in 22 '- /:» ---«•--— more cases than m other languages . :|U A C' can also occur embedded: a) when it isfitself dislocated or topicalized (see above (24)); b) in the context of like glauben, denken, etc., which allow C to be a v/2-structure:

(28) ich glaube, [p, dem Hans, L= dem hilfst du schon J] "I think, the H. him help you really"

The sentences in (29) are on the contrary ungrammatical:

(29) a. ich glaube, dem Hans, dass du dem hilfst b. ich frage mich ständig, dem Hans, ob du dem hilfst c. ich frage mich standig, dem Hans, wer dem helfen kann "I wonder all the time, the H. who him help can"

It can be supposed that in (29) the C' projection blocks the govern­ ment of the embedded sentence by the matrix verb; this is not the case in (28), where the embedded sentence, even if it appears to occupy a usually subcategorized position, may be supposed to be neither subcategorized nor governed by the matrix verb (cf. Reis (1985)). It has in fact the structure of a main clause, it is name­ ly a V/2 sentence at S-structure.

4. Let us consider now Koster's analysis in more detail. According to that analysis, deletion of the d-word leaves an open sentence, which, much like relative clauses, is satisfied by the XP in TOP _ 213 _

position. The open sentence functions almost as a complement to the topicalized item. Hence, it follows that only constituents that can have complements in general, can be topic(alized), that is, they can appear in unstressed sentence-initial position. If the item in TOP is stressed, it is a focus and therefore does not undergo the 23 analysis at issue . If we adopt the same analysis for German Top- icalization, we note that the quasi-semantic notions of 'topic', 'open sentence', 'possibility of having a complement' are not rele- 24 vant there . What we want to suggest here is a strict and uniform syntactic analysis of Topicalization. The wh-movement and deletion analysis, even if interpreted as purely syntactic, is, however, no longer available, first, because it implies the generation of a topicalized structure in different ways: either by deletion of the d-pronoun where the left-dislocation structure is possible, or di­ rectly when, as in the case of sentential adverbs, overt anaphors, quantifiers, indefinite pronouns, etc., the left-dislocation struc- 25 ture is not possible ; second, because of the tendency, in works subsequent to Chomsky (1977), to dispense with the rule of wh-dele- tion in COMP in those cases accounted-for there by this rule. Thus, rather than deriving Topicalization from :Left Dislocation, it is more natural to regard them as two independently generated constructions. The topicalized item is base-generated in TOP position, and an empty category, presumably with feature [djt is generated in the following Ï. The difference between Left Dislocation and Topic­ alization is reduced then to the optional base-generation of either 26 a d-pronoun or an empty category As in the case of Left Dislocation, there is obligatory connected­ ness between the phrase in TOP and the S-internal empty category. The phrase in TOP behaves w.r.t. various grammatical principles as if it actually occupies the S-internal position where the empty cat­ egory is %base-generated. The sentences in (30) show some cases of 27 connectedness - 214 -

(30) a. Lproit dem HansJ habe ich darüber schon [ej gesprochen "with the H. have I about-it already spoken"

b. [„„den Gar aus J will er mir LT„eJ machen NP J TtfP -" (he wants to do me in) c. (Vogel, glaubt), l= [Zeitungsberichte über sich.J mag JC C NP —j 1 [= Willi Brandt. Fe] nicht]! "V. thinks newspaper-reports about himself likes W.B. not"

d. (Vogel, glaubt), [=f f Zeitungsberichte über ihn ,."] mag ft w-B-i W nicht3 e. L seinen. WagenJ wäscht jeder, einmal im Monat CpeJ

In order to account for the connectedness phenomena, we need a rule of (re)construction which allows the element in TOP to be in­ terpreted in S-internal position when the relevant rules and prin­ ciples of the grammar apply. (Note that the element in specCOMP does not count for the reconstruction effects.) Since the element which appears in TOP has never been moved from the S-internal position, we cannot use the usual reconstruction rule, which, for the purposes of interpretation, 'brings' a moved wh-phrase back to its original .. . 28 position We will tentatively account here for the reconstruction effects by means of a rule of co-superscripting which applies at D-structure (cf. Cinque (1983)) and which coindexes the element in TOP with the empty category in the S-internal A-position. In the case of Left Dislocation, we have assumed the same coindexing procedure between the phrase in TOP and the d-pronoun (cf. Cardinaletti (1986)). This type of analysis has the added advantage of assuming the same kind of reconstruction in the case of Topicalization and in the case of Left Dislocation, while, in the movement analysis of Topicalization according to Thiersch, two different kinds of reconstruction should 29 be assumed The two co-superscripted elements form an Ä-chain, the head of -215 -

the chain being in TOP, an A-position. Following Rizzi (1982), we assume the G-criterion to be a wellformedness condition on chains, according to which the chain of co-superscripted elements, the phrase in TOP and the resumptive empty category, must contain only one ar- 30 gument, namely the phrase in TOP . As the d-pronoun in Left Disloc­ ation, the empty category must count as a non-argument, in spite of the fact that it occupies an A-position at D-structure. The relation­ ship between the phrase in TOP and this empty category is reminis­ cent of that between a post-verbal NP or clause and the expletive element with which they are co-superscripted. The G-criterion is not violated since the G-role which is assigned to the position oc­ cupied by the empty category is shared by the argument in TOP, a ex­ position. A chain can in fact be regarded as a sequence of 'connec­ ted' elements which count as a single element even if they occupy more structural positions. Through a condition similar to that for­ mulated by Belletti and Rizzi (19 81:140), the rules of the grammar are applied to the chain and not to a single structural position. In the case of Topicalization, this amounts to saying that the element in TOP shares the properties of the A-position which is the 'foot' of the chain even if it occurs outside of that position. Like the d-pronoun, the empty category undergoes movement to specCOMP and binds its trace in A-position. The S-structure of a topicalized sentence will be something like (31): (31) fg. to? den Han^k % ei k kennej k ich *i *M. Following Cinque's (1984) analysis of Italian Topicalization, we consider the empty category e in (31) as pro, undergoing wh-movement 31 (exactly, d-movement to specCOMP) and t as its own trace The empty category in A-position enters two different chains:

(32) Cr = (den Hans, t)

l C± = (e, t)

It must be marked with the features {-pronominal;-anaphorj. In the - 216 -

chain of co-superscripted elements, it does not qualify as a varia­ ble, because the phrase in TOP does not qualify as an operator for it. In this chain it must count as non-argument. In the chain of co-subscripted elements, the empty category t qualifies as a varia­ ble bound by the operator e in specCOMP and counts as argument. Note that pro, in its movement to specCOMP, has not retained the super­ scripted index. In fact, superscripting must be regarded as indexing of positions, and the lexical items bear the superscripted index only when they fill them. Note also that the subscripted index of the XP in TOP cannot be identical to that of pro and its trace be­ cause of cases such as:

(33) sich hat er rasiert "himself has he shaved"

If the rule of coindexing applied at S-structure, the sentence would be ruled out by Binding theory because the variable t would be bound in its governing category:

(34) sich. e. hat er. t. rasiert

11 11 Since (33) is possible, the indexing must look like (35): r r (35) sich. e. hat er. t. rasiert The coindexing of the XP in TOP and of pro and its trace is accoun­ ted for at LF-structure by a rule of construal. Let us now consider the Identification of pro. pro in specCOMP is Identified by the phrase of same category in TOP, its local X-ante- 32 cedent, which governs it and is co-superscripted with its trace Both requirements must be met with regard to the Identification of the feature content of pro.'The absence of one of them leads to un- grammaticality. Let us consider first the possibility of Left Dislocation but not of Topicalization in the sentences with finite verb in sentence-final position or in sentence-initial position at S-structure: _ 217 _

r r (36) a. den Hans , wenn ich den nur treffen könnte! (V/E) "the H. if I him just meet could" b. den Hans , wenn ich e nur treffen könnte! (V/E) Y» Y» (37) a. den Hans , könnte ich den nur treffen! (v/1) 3£ r r b. den Hans , könnte ich e nur treffen! (v/l) The d-pronoun and pro cannot be moved to specCOMP in v/E and v/l sentences for independent reasons. The same facts are present when a w(h)-pronoun is moved to specCOMP instead of the d-pronoun: (3 8) a. den Hans , wer hat den getroffen? "the H. who has him met" je r r b. den Hans , wer hat e getroffen? From sentences (36b)-(38b) it is clear that it is not sufficient that pro be governed by a lexical category. They are ruled out be­ cause pro, even if co-superscripted with the antecedent in TOP, is 33 - not governed by it (nor can the NP in TOP A-bind it as an operator binds a variable, since it does not qualify as an operator). The movement of pro to specCOMP ensures then that it is governed by its antecedent in order to be Identified by it. This requirement is not operative in the case of the lexical pronoun in (36a)-(38a). The opposite case concerns the left-dislocation structures in which the rule of co-superscripting, which we assume to be optional, does not apply. This causes the absence of connectedness between the phrase in TOP and the resumptive d-pronoun. The former is, for instance, not assigned the same Case as the d-pronoun, and can only be a nominative NP, being the nominative the unmarked Case, assigned to an independent NP: (39) dieser Mann, den. kenne ich t. "this man him know I" nom

The corresponding topicalization structure, with the base-generation of pro instead of the d-pronoun, is not possible:

(40) dieser Mann e. kenne ich t. l l - 218 -

In fact, the feature content of pro cannot be determined; at S-struc- ture pro is governed by the NP in TOP, but it is neither co-subscrip­ ted with the NP in TOP nor its trace co-superscripted with it.

\ - 2]9 -

FOOTNOTES

A preliminary version of these remarks was presented at the Com­ parative Fiesta on 24-8-1985, during the 4th International Summer School in Linguistics, Salzburg. I am indebt­ ed to many of the participants to the Summer School for helpful comments on that earlier version and for the judgements about many of the sentences here discussed. 1. Cf. Engel (1970). Olsen (1982) and Thiersch (1982) point out the interesting overlapping of the German sentence structure elab­ orated in the framework of the 'Stellungsfeldertheorie' and in the framework of the X-bar theory. 2. Many proposals have been made about the status of the COMP and COMP positions. Thiersch (1978), Den Besten (1981), Lenerz (1981) regard them as base-generated positions; Safir (1982) consider the rule R as an adjunction to S, while Scherpenisse (1985) and Haider (1984) propose R as chomsky-adjunction to S'. We will not pursue this issue here and will discuss Thiersch's analysis as representing an analysis involving COMP in contrast with Koster's analysis (see below). 3. These exceptions are the finite verb, modal and degree particles, es and some clitics in particular conditions which are at present poorly understood. Cf. Lenerz (1977). 4. See Chomsky (1981:185,275). 5. Cinque (1985:6) gives the following definition of variable:

(i) Vbl = ( e) in A-position, operator-bound and locally A-bound. For the class of operators, see Chomsky (1981:115) and (1982:lOlf) and Cinque (1985:7). 6. Speaker judgements on parasitic gap constructions vary greatly. Nevertheless it seems that sentence (5) is better than (i) and (ii), where the NP which binds the parasitic gap is not topicalized: - 220 -

?? (i) ich folgte diesem Mann, ohne e. zu vertrauen ?? 1 1 (ii) es ist bedauerlich, dass sie diesem Mann, ohne e. zu ver- 11 trauen folgte "it is regrettable that she this man without trusting followed"

Following Bennis and Hoekstra (1984/85) and Felix (1985)» we may assume in (i) and (ii) that the NP diesem Mann has adjoined to some position under S in order to yield a possible X-binder for its trace and the parasitic gap. This is however a much more marked option than Topicalization, as it is shown by the different judge­ ments of (i)-(ii) and of (5) in the text. = n 7. The E node corresponds to the S projection and the X node im­ mediately dominated by E corresponds to the TOP position in the phrase structure rules proposed by Chomsky (1977:91): (i) I ^TOP S

The X position dominated by S corresponds to COMP and the position for the finite verb (Vp) corresponds to COMP in the sentence struc­ ture of Thiersch (1978), 8. See below, where it will be concluded that COMP must be spec­ ified as an operator position and COMP as a position for inflec­ ted elements, and where the sentence structure (23), corresponding to that proposed by Chomsky (forthcoming), will be assumed for Ger­ man. 9. For a discussion of the notions 'bound', 'governing category' and in general of Binding theory, see Chomsky (l981:Chapt. 3.2.3.). Note that the same pattern of judgements is obtained in the case of other categories, too. Let us consider the case of an anaphor inside an AP:

(i) stolz auf sich., das ist Franz. nie gewesen (cf. (8a)) "proud of himself that is F. never been" (ii) stolz auf sich., Franz, ist das nie gewesen (cf. (8b)) _ 22U -

(iii) stolz auf sich., wer. ist das nie gewesen? (cf. (11)) (iv) ich glaube, stolz auf sich., Franz. ist das nie gewesen (cf. (10b)) 10. The sentences of (8), (9) and (10) show other connectedness properties; for instance, the element ;in TOP meets the subcategor- ization requirements of the following verb, bears the e-role as­ signed by it to the d-pronoun and is marked with the same Case as­ signed to the d-pronoun by the verb governing it. Note that also part of VP can be dislocated and topicalized, as shown in:

(i) das Buch gelesen (das) habe ich schon "the book read that have I already" (ii) geliebt (das) wurde sie von Karl sehr "loved that was she by K. much"

11. A left-dislocation construction has the following intonative contour, which is important to keep in mind: the phrase in TOP has a progressive intonation; it always bears a (almost primary) stress; between it and the following S there is a hardly perceptible pause or none at all. A left-dislocation construction is used to fix as 'theme' the phrase which occurs in leftmost position. 12. Note that the starred sentences in (8) and (9) are indeed pos­ sible with a different intonation pattern (usually, with raising intonation on the phrase in first position and with a clear pause between it and the following sentence) and a different pragmatics (the element in first position has usually occured in the immediate­ ly foregoing discourse and is therefore taken from the context), which show that they are to be interpreted as discourse-grammar phe­ nomena. This amounts to saying that the phrase in first position and the following sentence belong to two different syntactic units. This also explains why a sentence such as (10b) cannot be rescued by means bf a different intonation. In this case, the S occurs in an embedded context, i.e., inside a syntactic unit (this possibil­ ity is available only with verbs like glauben, denken, sagen, etc.). 13. Note that also an adjunction analysis as in Haider (1984) would wrongly predict that (8b,c), (9b), (10b) are grammatical. According to that analysis, any item can be chomsky-adjoined to the COMP position when the finite verb has already been moved there; such a configuration can also occur embedded in a topic construction, that is to say, as a sister of TOP (cf. Haider (1984:30ff)). 14. That is to say, the d-pronoun has either remained in situ or has been adjoined to some element under S. The Mittelfeld corre­ sponds to that part of the sentence dominated by S (in the X-bar system) which precedes the position of the verb in embedded sen­ tences. 15. One last possibility in order to rescue Thiersch's analysis is to assume that a w(h)-/d-pronoun must be obligatorily moved to COMP if present and if COMP is available for movement. So, (8b,c), (9b), (10b), (14b,c) would be ungrammatical because the d-pronoun has not moved to COMP . But the existence of echo-questions con­ tradicts this claim, in that they show a w(h)-pronoun in situ and something else fronted:

(i) Karl hat wen angerufen? "K. has whom phoned"

(According to our analysis (see below), the structure of an echo- question is something like the following:

(ii) Karl e. hat t. wen angerufen

An empty category has moved to COMP instead of the w(h)-pronoun.)

16. Cf. among others Koster (l978b:200f), Den Besten (l981:i). 17. Note, however, that the d-pronoun in Left Dislocation cannot be regarded as a relative pronoun (as for example in Fanselow (1982)); first, it cannot be substituted by the other German relative pro­ noun: welch-:

(i) den Mann, welchen kenne ich "the man whom know I" - 223 -

Second, it triggers verb-second, contrary to the relative d-pronoun; third, it may be stressed, whereas the relative pronoun cannot. But none of these facts contradicts our claim; simply, the two d-pro- nouns enter different structures, both being able, however, to act as an operator. Note also that a w(h)-pronoun can be stressed, too. Tappe (1981:209) points out the complementary distribution of the d- and w(h)-pronouns in their use as operators. The former can be coindexed with a c-commanding category; they appear therefore in Left Dislocation and relative clauses, buttriot in free relatives:

(ii) Paul lädt ein, den er will "P. invites him he wants"

The latter cannot be coindexed with a c-commanding phrase; they ap­ pear in interrogative clauses and free relatives. Under our analysis, w(h)- and d-pronouns have also a symmetrical distribution with regard to root and embedded sentences. 18. Note that in (16) the d-pronoun is not moved to the higher COMP , as it usually is:

(i) den Hans, den glaube ich mag Julia nicht

Since a sentence such as (ii) is also possible,

(ii) den Hans., den. glaube ich den. mag Julia nicht where there are two coindexed d-pronouns, it is natural to regard these constructions as somehow similar to those constructions with scope-marking w(h)-phrases in the higher COMPs, as in (iii) (cf. Van Riemsdijk (1982:12f) from which the example is taken):

(iii) was glaubst du, was Peter meint, was Hans sagt, was Klaus behauptet, mit wem Maria gesprochen hat? "what think you, what P. means, what H. says, what K. affirms, wjth whom M. spoken has"

In sentence (16), there must be an empty operator in the highest - 224 _

COMP ,coindexed with the d-pronoun:

(iv) den Hans., e. glaube ich den. mag Julia nicht

19. Principle (C) of Binding theory requires that an R-expression (a potentially referential phrase, cf. Chomsky (1981:102)) must be free (cf. Chomsky (1981:188)), that is to say, not coindexed with some other element. Note that in (17) Principle (C) applies to the demonstrative d- pronoun as if it actually occupied the position of the empty cate­ gory e, where it is A-bound by the NP Hans. The same happens in the grammatical (15), where there is no violation of Principle (C), even if the NP Hans seems to be in the c-command domain of ihn (the pronoun ihn is however in an A-position). In (15), the pronoun ihn can be A-bound by the NP Hans, according to Principle (B) of Bin­ ding theory (see Chomsky (1981:188)) and indeed must be A-bound by it when selbst is also present. 20. Haider (1984:31) assumes, incorrectly for us, that a sentence such as (20) is an instance of Left Dislocation: in fact, he con­ siders it as being on a par with a left-dislocation construction with the d-pronoun, cf. his examples:

(i) die Callas, diese Sängerin bewundert er immer "the C. this singer admires he always" (ii) die , die kann ihm gestohlen werden "the syntax that can from him stolen be"

Under our analysis, examples such as (i) and (20) are no evidence against the claim that only a d-pronoun can occur in COMP . 21. There are in principle two more possibilities to obtain Topic- alization. First, the topicalized phrase could be moved directly to TOP, from where it would bind its trace. This derivation must be excluded, however, both for theoretical and empirical reasons. In fact, the position which usually serves as an escape hatch for the movement out of S is COMP, not TOP. Furthermore, this rule would give empirically wrong results. Let us consider (ii), a pos- - 225 -

sible answer to (i), in which it is natural to assume a (discourse- linked) empty operator in COMP which binds the empty category in the S-internal A-position (cf. also Bayer (1983/84:fn. 4)):

(i) hast du Hans gesehen? "have you H. seen"

(ii) r [ l habe ich s COMP2 °i^ COMP1 jl ^ astern e. getroffen t.] "have I yesterday met"

If a constituent is topicalized, the sentence is ungrammatical, even if this movement does not alter the configuration inside S:

(iii)a. gestern habe ich getroffen b » T^ L^-r.gestern.1 l[- L„ , O.J l[-_._ habe J l /„ich t, e. S TOP lc S TCOMrTP IJ COMP J S kl getroffen t .Jj|

Analogously, a sentence such as:

(iv) L«„den Hans J I- L^ I [_„.„ hast J L du t. gesehen t .1 v ' TOP l lS COMp-w ^ LCOMP jJ 'S l a jJJ should be good, since the requirement for v/l sentences that COMP be empty is satisfied (cf. also Haider (1984)). But only a sentence such as (v) is possible:

(v) den Hans, hast du den/ihn gesehen? "the H. have you him seen"

The contrast between (iv) and (v) simply suggests that the item in TOP cannot function as an operator for its trace, hence the need of a resumptive element: den/ihn, and, consequently, of a different analysis. The second possibility, that the topicalized item is moved to COMP and then to TOP, must be excluded as well: it seems that there is no reason why it sh^ould be moved further from COMP to TOP. (Note that this derivation has been suggested by Vat (1981) for Left Dis­ location,' but here also it meets with many problems; for example, it cannot derive sentences like (ll)-(l3) in the text). _ 226 _

22. Another possibility could be the following structure:

(i) 6.

TOP ^-operators] COMP Ï

Such a structure would have the advantage of being the same, pre­ sumably, for all German sentence types, the difference between the non-w(h)- and the w(h)- v/2 sentences lying in the realization or not, of the TCP position. But we will not discuss this possibil­ ity further. 23. For instance, idioms, that have NPs which are incompatible with complements, can appear in Dutch only as focus. Also object clitics, which cannot have complements, cannot be topicalized or left dislocated. 24. In German, the possibilities for an item to be a topic or a focus seem to be exactly the same. Besides, idioms can be topic­ alized and left dislocated (cf. (i)) and the same holds true for some object personal pronouns in particular conditions (see foot­ notes 2. and 26.):

(i) den Garaus (den) will er mir machen (he wants to do me in)

25. It is obviously not desirable to mark the deletion as obliga­ tory in these cases. 26. The case of those elements which cannot be left dislocated can be now accounted for in another way than Koster's. A possible motivation which can be valid for lexical anaphors, quantifiers and indefinite pronouns, including the pronoun man, is that the d-pronoun might be possible only with more referential, referen- tially autonomous items for feature matching requirement. So, for speakers who cannot dislocate parts of idioms, we may suppose that - 227 -

they cannot any longer split up the idiom into its parts in order to get a quasi-referential reading for them. 27. The connectedness properties that we have not spoken about yet, concern the possibility of topicalizing part of an idiom (as shown in (30b)) and the regular application of the principles of Binding theory (see (30c) and (30d) for Principles (A) and (B) respectively, where the relevant governing category is that of the resumptive emp­ ty category (here Ï)). 28. See Chomsky (1977:83ff); see also Chomsky (1981:345,fn. 10) for a similar problem and Vat (1981) for the discussion of possible reconstruction mechanisms. 29. In the case of Topicalization, reconstruction would have the form of the usual reconstruction rule discussed in Chomsky (1977) because of its nature of movement-construction, whereas in the case of Left Dislocation the reconstruction mechanism proposed by us in the text would be necessary. 30. This step is justified by the fact that the chain does not result from movement, but from a free mechanism of chain formation, which links two elements of same category and with the same features. It can happen that there are two possible candidates for the coin- dexing with the phrase in TOP, as is shown by the D-structure in (i): (i) f d Han er d enn T0P fNPp-masculmr " ?. eI t ff fNPi+masculmer ?. nl NPrf+masculm"^ J^e l [+singular J L+singularJ [_+singular J

We can exclude the coindexing of the dislocated NP and of er by saying that d-/e is not an argument, so that the ©-criterion is satisfied only when the NP in TOP is coindexed with it. The alternative would be to mark the coindexing procedure to the effect that the phrase in TOP is coindexed with a phrase with feature (+dj(a step, however, which does not seem to be very desir­ able). In such a case, the argument could be either d-/e_ (and the phrase in TOP would be its lexical content, something like what is - 228 "

proposed in Koster (1982)) or both the phrase in TOP and d-/e, with the 6-criterion slightly modified to the extent that in a chain there cannot be two elements both in G-positions. 31. The structure under C can be considered the same as in the cases of discourse-linked empty operator in specCOMP discussed in footnote 21. (see also Huang (1984)); the content of the empty cat­ egory is determined through some element in the context, whereas in Topicalization its content is identified by the phrase in TOP. 32. In the case of Topicalization, the XP in TOP qualifies as the antecedent of pro at S-structure, even if it is not directly coin- dexed with it at this level, for a sort of transitivity operates in the coindexation of pro and its own trace and of the latter and the phrase in TOP. For an extensive discussion of the licensing condition for pro, cf. Rizzi (1985). 33. The requirement on Identification of pro seems to be stronger than ECP, in that it consists in government only by a coindexed category. - 2?9 *

BIBLIOGRAPHY

Bayer, Josef (1983/84), "COMP in Bavarian syntax", The Linguistic Review 3.3, 209-274. Belletti, Adriana and Luigi Rizzi (1981), "The syntax of "ne": Some theoretical implications", The Linguistic Review 1.2, 117-154. Bennis, Hans and Teun Koekstra (1984/85), "Gaps and parasitic gaps", The Linguistic Review 4.1, 29-87. Cardinaletti, Anna (1986), "Lo status dei pronomi d- e la Ricostru- zione nella 'Dislocazione a sinistra' in tedesco", Rivista di Grammatica Generativa 8, 111-125. Chomsky, Noam (1977), "On wh-movement", in: P. Culicover, T. Wasow, A. Akmajian (eds) Formal Syntax, New York, 71-132. - (1981), Lectures on Government and Binding, Dordrecht, Foris. - (1982), Some Concepts and Consequences of the Theory of Government and Binding, Cambridge (Mass.), The MIT Press. - (forthcoming), Barriers, The MIT Press. Cinque, Guglielmo (1983), "Topic constructions in some European lan­ guages and 'Connectedness'", in: K. Ehlich, H. Van Riemsdjik (eds) Connectedness in Sentence, Discourse and Text, Tilburg, 7-41. - (1984), "X-bound pro vs. variable", mimeographed, University of Venice. (1985), "Bare quantifiers, quantified NPs and the notion of operator at S-structure", mimeographed, Univer­ sity of Venice (to appear in Proceedings of the Troms^ Work­ shop on Romance Syntax, T. Taraldsen (ed)). Den Besten, Hans (1981), "On the interaction of root transformations and lexical deletive rules", Groninger Arbeiten zur germanisti­ schen Linguistik 20, i-iii, 1-78. Engel, Ulrich (1970), "Regeln zur Wortstellung", in: Forschungs­ berichte des Instituts für deutsche Sprache 5, Mannheim, 3-148. Fanselow, Gisbert (1982), "Universale Grammatik und deutsche Syntax. Eine Anwendung der "Government-and-Binding"-Theorie Noam Chomskys auf die Syntax des Deutschen", mimeographed, Univer­ sity of Konstanz. Felix, Sascha (1985), "Parasitic gaps in German", in: W. Abraham (ed) Erklärende Syntax des Deutschen, Tübingen, G. Narr. - 230 -

Haider, Hubert (1984), "Topic, focus and V-second", Groninger Ar­ beiten zur germanistichen Linguistik 25, 1-48. Huang, C.T.J. (1984), "On the distribution and reference of empty pronouns", Linguistic Inquiry 15» 531-574. koster, Jan (1978a), "Why subject sentences don't exist", in: S.J. Keyser (ed) Recent Transformational Studies in European Lan­ guages, Cambridge (Mass.), 53-64. (1978b), Locality Principles in Syntax, Dordrecht, Foris. (1982), "Enthalten syntaktische Repräsentationen Varia­ blen?", Teil 1., Linguistische Berichte 80, 70-100 (Teil 2., Linguistische Berichte 83 (1983), 36-60). Lenerz, Jürgen (1977), Zur Abfolge nominaler Satzglieder im Deutschen, Tübingen, G. Narr. (1981), "Zur Generierung der satzeinleitenden Posi­ tionen im Deutschen", in: M. Kohrt, J. Lenerz (eds) Sprache: Formen und strukturen, Tübingen, Niemeyer, 171-182. Olsen, Susan (1982), "On the syntactic description of German: Topo­ logical fields vs. X-theory", in: D. Weite (ed) Sprachtheorie und angewandte Linguistik, Tübingen, G. Narr, 29-45. Reis, Marga (1985), "Satzeinleitende strukturen im Deutschen, über COMP, Haupt- und Nebensatze, w-Bewegung und die Doppelkopf­ analyse", in: W. Abraham (ed) Erklärende Syntax des Deutschen, Tübingen, G. Narr, 271-311. Rizzi, Luigi (1982), "On chain formation", mimeographed, University of Calabria (published in Modèles Linquistiques VII (l). (1985), 119-157» with the title: "Conditions de bonne formation sur les chaines"). (1985), "Null objects in Italian and the theory of pro", mimeographed,MIT. Safir, Ken (1982), "Inflection-Government and Inversion", The Lin­ guistic Review 1.4, 417-467. Scherpenisse, Wim (1985), "Die Satzstruktur des Deutschen und des Niederlandischen im Rahmen der GB-Theorie. Eine Reaktion auf Marga Reis' Doppelkopfkritik", in: W. Abraham (ed) Erklärende Syntax des Deutschen, Tübingen, G. Narr, 313-334. Tappe, Hans Thilo (1981), "Wer glaubst du hat recht? Einige Bemerk­ ungen zur COMP-COMP-Bewegung im Deutschen", in: M. Kohrt, J. Lenerz (eds) Sprache: Formen und Strukturen, Tübingen, Niemeyer, 203-212. Thiersch, Craig (1978), Topics in German Syntax, MIT Phd diss. - 23ï: -

Thiersch, Craig (1982), "A note on 'Scrambling' and the existence of VP", Wiener Linguistische Gazette 27-28, 83-95. Van Riemsdjik, Henlc (1982), "Correspondence effects and the empty category principle", Tilburg Papers in Language and Litera­ ture 12, 1-20. Vat, Jan (19 81), "Left dislocation, connectedness and reconstruction", Groninger Arbeiten zur germanistischen Linguistik 20, 80-103.

ADDRESS:

Anna Cardinaletti Seminario di Linguistica e didattica delle lingue Universitä di Venezia S. Marco 3417 1-30100 Venezia/ Italy