ON PIED-PIPED INFINITIVES IN CEKMAN RELATIVE CLAUSES*

tienk van Ric~llsdij k Tilburp. l l~~ ive r s i t y

Among t l~e more cunvus pl~e~io~ilelia in Gemian syntax. we find sen. tencts of the followins typc:

(1) a. Jetzt hat cr siclr cprrclliclr dcrr h'ogc.rr. [den zu kaufi~rl, cr now has Ilc t o - l~h lesc l f finally t l ic car wllich to buy h e

sick sclrc~rr larrgc~ ~vrgcrrorrrrrrt~rr Irattc. leistcrl kfinrrcn to-himself alrcady l ~ ) n g planncd l13d afford been-able-to

"Now Ilc has tin:~lly bcen ablc t o al'li)rtl tllc car wliicl~ hc had

planned l o buy for a long li~nc."

b. 6faran Iut Alaria, lrllir dcr k dell (irlo~rb ar f a l r ~ e n ] ~ ich one has MarU with wl~cim ill tltc vacalion to drivc I mcirrcvrr Lollc:4c.r I cprst mr :wi Tagc~rr crrr~~ji~ltl(~n lwt tc, gatcrrl to-my collcayc jtrsl a p l two days rccommcndcd hat yesterday abcrrd tot irr i1rrc.111 %irrrrrrttr arrjiCc./rrrrtlcrr evening d c ~ d in hcr roo111 found "Marb. wit 11 wl1~1111 I I ~ a d ~ I I S I t\vo L I ; I ~ S ago recolnn~cndcd t o

nly collcagus (o go 011 a V ~ C ; I ~ I ~ ) I I . was Ii)r~lld dead in l ~ c r room

last night ."

Thcse constrrlctions contain rclativc zlauscs. usually IIOII-restrictive oncs.

in which the rclativc prolloun is coi)t3irlcd in ; I I ~ infiiiitival coniplc~lient

clause. Wliat is renlarkablr' is tllat thc wliolc i~il'initival clause. a in the examples in ( I ) , is movcd under 1~11-11iovcn)cnt. Alongside ( I ) we alsv find

This aniclc is a somcwl~at rcviwd translation of tily 'Zum Rattcnfingcrcffckt bci

lnfinitiven in deutscllcn Rrhtivdt~en' . wllicl~ rppC;lrc.d in C'n>trCrgcr Arhcifcrr zitr Gennarrisfischen Litr~rrisrik Nr 21. 1982. T l ~ c stin~ulus fiir tlic paper wasa discussion during the Fust (and Iwprfr~lly not hst ) Vienna Rortnd Tablc on Comparativc Syntax. June 1981. I u,nr~kl kc to thank tllc participants of that \\.orksl~op, and in paicular the 'locale'. for I I ~ V I I I ~ cffccfivcly dispcllcd niy original sccpticisln towards the analysis which I prcwlr1 1lt.r~- in w c r i c ~ ~ ~ 2. 111 ~dd i~ io t i I would Iikc to thank Gugliclnio Cinque. Iluiwrt I l a d ~ r . Til11ia11 Iliil~k. ant1 Tltilo Taplic for their mni- mcnts on olrlicr vcriotrs ol' 1 l ~ i z art ~i'lc.

think about the positinti oTa itself. (3) does not specify whether a is in COMP or wliet her it in SOIIIC ot11cr way occupies a S C I ~ ~ C I ~ C C initial position. In view of the fact I11at w'lr-n~ove~i~cnl of an over1 wll?-phrase is always obligatory in German relative clauscs. it seems reasonable to assume that a in effect does urldcrpo ~ t ~ l ~ - n i o v c ~ ~ ~ e ~ ~ l 311~1 I I C I ~ T E is in COMP. This in- stance o f w l t - r n o v c o ~ ~ ~ l we will call ' C Z I C ' ~ I ~ ~ ~ ) ~ ~ I I - I I ~ O V C ~ I C I I ~ ' . Notc Ilia1 internal ic.11-nioven1c111 callrlot Ijc ~ I I V O ~ C ~ t o salisl'v tlie obligntoriness of wlr-movc~iient in rclativc clauses si~lcc i t cat1110t ~11ara11tce that a will cnd up in a sentence i~~ i t i a l ~losi t iol~. I t 1 vir\v of tllcsc considrrations wc will assunle without f l~r t l~cr disc~rssic~u tllat c* t c r~~a l wh-n~ovc~nent applies and that a is in COhll'."

2. THE ANSWER

If it can be shown t11at t11c relative pronoun illside a nlust be in COMP. in other words, that such sentences evidcncc wliat we m y call 'internal wh-movement'. t l \c~i it must a fortiori bc the casc that a = 3 . Most o f t h e arguments to be presented hclow havc illis st~uc.lt~rc.

2.1 Position

Hliile the linear ordcr o f c o ~ ~ s t i t u e ~ ~ t s is relatively free in Ccrnlan. there are nevertlleless a number ol' cascs wllcrc there is at least a strong pre- ference for one of tllc possihlc orders. This is parlic~~larly truc for un- stressed (i.c. clitic-like) prclnouos. as in (4).

(4) a. 1~11 IICIIIII~~ c3s trrIf i11t11 m1/' I takc it wit11 I~inl 111)

"I c11aIIc1ig~ I ~ i r t r " b.?*lcIr rrelrtnc trrir ilrtrr r s m~! '

In PPIs t l ~ c conlrast gocs ~ I I C ~ I I I P I . way:

( 5 ) a. ?* Afol~a~t~r~r t r l Ali 1st cirl Afurrrr. t s rlr i r drnr orcjirel~t~lol zrc Mohamnicd Ali is a nian it wit11 wlioni up takc t o wollcri, rcincr I~'u1111si11tr wire

want sl~ccr nladncss would-bc "M.A. is a oian, tu cl~allenge wllon~ would bc sheer madness"

b . MoIlot?rtnc.cl.4 li ist eitr Alarrtr . t ~ r it ( k s ~ r c,s orrf,zehmetr zu wollen, reirlcr kbl~t~sirrtr n ~ r c

The assumption Ilia1 internal ~ c l l r - ~ ~ ~ o v c t i i c ~ ~ t has taken place within the

(9) a .?* Wir habcn nrit clicsctn Bursclrctr dcn l f a r ~ s we have with this boy tho flans tatrcrr sclrrrr ttniissc~r~ dance sec Iii~d-to "We had t o scc Ilans dalicc wit11 this boy"

b . Wu lrabrrr dell //arts r r ~ i f ~ ~ C ' S L ' I I I Br4m(.I1i~tr tatr:en S C ~ C I I miissen

Again, t l ~ e contrast is invc~ lrd in t IIC 1'1'1 ~.onsI ruct ion.

(1 0) a . Dk 1,orrc isr cirl l t a ~ l t * I ~ c ~ ~ , rnir dcr unseren the Lotte is a girl with wllom o u r Soh11 arrsgchcrr :I( lassclrl rllir rrltrr ck1t.11 son go-out tu Icl t o - n ~ e erlipl~. particles ctwas U I I ~ ~ ~ S I I I I I crs~.lrcirr t somewl~at u ~ ~ r c c . o l ~ ~ ~ ~ ~ c r ~ d a h l s sccnls "Lottc is 3 girl. to let our so11 go out wit11 w11o11i really doesn't see111 rccor~~l~~cr~dabIC to IIIC"

b. *Die Lorre ist cirl ~ l l : d c h c ~ ~ . rrnsrrc~l Sohn mit der arrgehcn zu lasserr rnir rrurr tk)t.lr etrvas ut~ratsonr crsclrcir~t

( I 1) a. Dcr B~crsc.1~. , t i t c l rn~ den llar~s tarcerl sel,ert r u mli'sscri, the boy with wlionl the Hans dance see t o have-to rnir grossctr h'rmrmer hcrcitct hat. sclrcirlt ;ctzt eincn andcnr to-me great p i c f causcd 113s secr~ls now an otller Frcurld pc.firnllc~~ zrr habcrr freind found t o have "The boy t o llavc to sec Ilans dance with whom has caused me great grief, scelns 10 have found another boyfriend now"

b .*Dcr Bursclrc~. dc.rr Ifarrs nrit llc~rtr tan:or sclrar i t / mussor, r?lir grossctl Krrrnrrrcr hcrcitct Irat. sclrcint j~ptzt cit~crt andem Frcrrrrd gcfur~llco zu habcrt

The ungramlmticality 01' ( IOb and I I I h) is ayain a conscqucncc of tllc template ( 6 ) . And i! is ir~tcrr~:ll r c l t - r ~ i ~ ~ ~ c r ~ ~ c ~ ~ t ~ l1 ic11 111akcs i t possible for (IOa) and ( I l a ) t o es~. :~pc l.r1>111 I I I V I i lrcri~~g cf1bc.t 01. t l ~ c I C I I ~ ~ I : I I C . TIICSC facts t l ~ ~ r s S I ~ C I I ~ I ~ I ~ I I 0111 ; ~ \ s u ~ ~ i p t ~ o ~ ~ I I I ~ I I tile 1110vcd rvlr-pl~~asc. must be7.

The conclusi i )~~ d ~ a w i ~ I ' I ~ I I I 111e ~ ~ ) < . I I O I I ) ~ ' I I ; I ~ I I ~ ~ I I S S C C I i l l 2.1. viz. tj131 there is intcrnal rrlt-lllovc~llclll. 113s ( W o 1:lirI) ~ ~ : I I ~ - I L ' ~ L ~ ~ I ~ I I ~ furtl~cr consc-

I - r 5-1 mir, wohl kaum je in den Si~m kommen wurde

I I

hben konnte

COMP S

The iterative PPI effect can be seen in examples such as (15) and (1 6).

-

( 1 5 ) Noch so ein hlann, den zu Cbeneugen another such a man wl~orn to convince dir vonunehmen ich dir unbedingt to-you to-plan I to-you definitely empfehlen wiirde, ware der O~ornsky recommend would would-be the Chomsky "Another man. whom I would definitely recomnlend that you plan to convince, is Chomsky"

" I NP VP

I zu glauben COMP S

I

(1 6) Die Femsehkihe. die amusehen dir zu the TV-series which to-watch to-you t o empfehlen, mir kaunl einfallen wiirde, recommend to-me hardly occur would he@ ubrigens 'Dallas' is-called by-the-way 'Dallas' 'The TV-series, which it would hardly occur to me t o recommend that you watch. is called 'Dallas', by the way"

,:j t l [ i ? s L , 12 ptn

- I 2.3 me anti-crossover effect

There is a further argument t o the effect that a in ( I ) must be taken t o be 3, and this argument is independent of the question as t o whether. there is internal wh-movement. I t has to do with the so-called anti-cross- over effect.'

The well-known crossover situation can be characterized as follows: wh-phrases, and all their parts, behave with respect t o rules of (non-) coreference as if they had nor been moved. In example (IB), wessen and

- sie cannot be coreferential, a fact which corresponds directly t o the atuation in the parallel example (19) in which n o wh-movement has taken place.

(18) * [Wesseni WagenIj glmbt sie, duo er e. genommen h a ? whose car believes she that he ' taken has - 'Whose car does she believe that he has taken?"

(19) * G h b t siei, dap er Monikasi Wagen genommen hat? believes she that he Monika's car take? has "Does she believe that he has taken Monika's car?"

There is a systematic class of exceptions t o this principle which has been known for a long time but has remained unexplained. Examples of this class, which we may call the anti-crossover case, are o f the following type

(20) [ Welches Bild, das du von der Monikai gemalt which picture that you o f the Monika painted

ha~t,]j glmtbst du, dass siei schiin finder? have believe you that she beautiful finds 'Which picture that you painted of Monika do you believe that she fmds beautiful?"

In this sentence. Monika and sie can be interpreted as coreferential. even though Monika is inside the preposed wh-phrase and. contrary t o the general case, the unmoved variant does not allow coreferentiality.

(21) G h b s t du, dap siei das Bild, das believe you that she the picture that du von der Monika, gemalt hast , sclzon firuiet? you of the Monika painted have beautiful finds "Do you believe that she finds the picture that you painted of Monika beautiful?"

iOFi:,.l L,-L,I :! ~ I J ilr~e, O 12 ptn

- The anti-crossover effect yields yet another argument in favor of a = S. This argument derives from the contrast between (25) and (27).

(27) Dm ist die Frm, [mit der Herrn MiiUeri bekamt zu m a ~ h e n ] ~ ich ihm, niemals raten (= advise) wude

This contrast can be readily explained on the assumption that the PPI contains the PRQsubject of the infinitive, and hence cannot be VP. Under this assumption the contrast follows directly from the different control

- properties of the verbs versprechen "promise" and raten "advise". as is shown by the following representations.

(28) . . .[mit der PROj H.MI bekannt zu machenj, ichj ihmi niemals versprecllen wurde

(29) . . . [mir der PROi H-M .i bekannt zu machen]= ichj ihmi niemals mren W d e

Since mten is a verb of object control. PRO and Herm Muller are coin- dexed in (29) but not in (28). Therefore, the principles of (non-Norefer- ence1', will always predict that H e m Muller and ihm cannot be corefer- ential, regardless of the anti-crossover effect. If only the VP had been moved, the PRQsubject would have remained in its base position. In that case it would be quite obscure why the anti-crossover effect shows up in one case but not in the other.

On the basis of the arguments presented in this section we may regard two central properties of the PPI construction as firmly established. First. a-3, and second, inside a internal wh-movement applies.

3. FURTHER PROPERTIES

Let us now turn to some further characteristic properties of the PPI construction.

3.1 Infinitives only

It is remarkable thit an 3 can only be pied-piped if it is infinitival, i.e. when its inflection marker is [-tense] and hence when its main verb is an infinitive. In other words. sentences of the following type are sys- tematically impossible.

(30) a. *Das ist dcr Wagen, [dab er den gekaufi Iut]si that is the car that he whicli bought has

'The man. whom one needs a lot of luck to catch, has now also disconnected his phone"

Note that if we assume that the infinitival complementizer um is not a preposition but a COMP, despite Tappe (1983), the ungrarnmaticality of (32) will follow from our template (6). And to the extent that internal wh-movement has taken place, the DFC filter will yield the same result. Interesting confirmation of this analysis can be found in an example cited in Van de Velde (1977). In this case, an instance of the obligatory urn is deleted in a PPI-construction."

(33) a. Eine Musik, deretr teill~aftig zu werden man wolvalrren a music of-which partaking to become one pilgrim muke wie die Amerikancr t1acI1 Baymitll had-to like the Americans t o Bayreuth "A music which one had to undertake a pilgrimage like the Americans to Baureuth in order t o partake o f '

b. Man mupte walvahren wie die Amerikaner nach Boyreuth, um/*p dieser Musik teilhafiig zu rverden

3.2 Consequences of internal wh-movement

The necessary assumption of internal wh-movement is far from trivial from a theoretical perspective. There is a mostly implicit but nevertheless powerful principle floating around according to which an operator-like element (such as a relative pronoun o r a wh-phrase) is interpreted (i.e. gets scope) in the COMP-position which it occupies at s-structure. More generally speaking. many linguists assume that there is a virtual isomorphism between structures that result from move-wh at s-structure and the cor- responding operator-variable representations at LF.

Consider again the structure (3c) which was shown to give a correct characterization of German PPls. This structure is repeated in tree-form as (34).

-- - - - - - - Graphically. the following percolation possibilities result."

(36) a. COMP b. COMP c. COMP

Observe that (36) implies. so to speak. a type of recursion, in that (36a) or (37b) can be embedded under ( 3 6 ~ ) . as indicated by the [q, t rel] in the lowest COMP in (36c). Such a possibility is illustrated in (37).

(37) Der Mann, 6 6 mit [dessen ] baters Freudin ] ausgehen zu +re1

the man with whose father's girl-friend go-out to wonen,& ich mi& nie er-dreisten wtirde want I me never have-the-cheek would "The man, to want t o go out with whose father's girl-friend I would never have the cheek"

Thus, the percolation mechanism proposed provides a specific way of implementing the template (6) discussed earlier. We may now ask: does it also provide a way of uniting the template with the analogous problem in LF. Recall that !he main problem tliere was that some operator-like element, such as a relative pronoun, which is in some COMP can take its scope outside that COMP in some higher COMP which dominates a phrase or clause containing the lower COMP in question. We might now assume that the nodes along which the feature [+re11 percolates constitute a percolation chain in the obvious sense." Using this notion, we may formulate the following principle of LF.

(cf. Cinque (1981)). H'e will not pursue the issue any further in the present context.Ig

3.3 The contrast betwee11 German and Du tcl~

In Dutch. sentences with PPI are completely excluded. The Dutch version of (la), for example, is ungrammatical. -- - - . -

(40) *Nu heefi hij zich eindelijk de auto, die re kopen hij zifh now has he himself finally the car which to buy he himself al lang had voorgenomen kunnen permitteren already long had planned been-able-to afford "Now he has finally been able to afford the car which he had planned to buy for a long time"

I l e question arises. then, why these two languages, which are in many respects quite similar. differ so clearly in this respect. The question is all the more pressing since wh-phrases in Dutch have essentially the same structure as in German. In other words, they are subject to template (6), or, in terms of percolation. wh-features must percolate along left branches. Why, then. does Dutch allow (36a) and (36b) but exclude (36c)?

It would be interesting if this difference could be related to some in- dependently existing difference between the two languages. Such a differ- ence can indeed be found if we study the arrays of possible types of complement clauses in German and Dutch. In Dutch, an infinitival comple- ment clause can only be of one of two types: either it is extraposed or its verb undergoes verb raising, cf. Evers (1975). In the verb raising variant, the complement clause is to the left of the matrix verb, but its verb is extracted and adjoined to (the right of) the matrix verb. Under extra- position. the complenient clause as a whole stands to the right of the matrix verb. In German, on the other hand, there is, contrary to Dutch and contrary to Evers' claim, a third type in which the infinitival comple- ment clause as a whole is to the left of the matrix verb and in which verb raising does not apply to adjoin the complement verb to (the left of) the matrix verb.m

Schematically, we have the following situation:

(41) German Dutch

a. ...[ ...... Vk ... V t - -

b . . . [ ...... @...J + + (verb raising)

c. ...[Fj$. .v, + + (extraposition)

b. . .. weil er (uns) drohre, seinen RnaIcn zu toten (control interpretation)

(45) a. *seine Behemlzurtg, die zu verIierer1 er (uns) schon lange dmhte

b. ? sein Rivale, den zu toten er (uns) scl~on Iange drollte (control interpretation)

These arguments establish conclusively that PPI is incompatible with verb raising.

Let u s now push the analysis one stcp further. Verbs which permit the PPI construction are characteristically verbs which can occur in the form ...... V k X V. where X + $. In many cases. including many of the

examples given in the present article, X is an inherently reflexive pronoun, for example. Observe that in the non-moved variant the alternative order X [. .. . . .V]S V is generally also grammatical. Therefore it is not possible to es tabkh with certainty that in the PPl construction X intervened between the complement clause and the matrix verb. In other words it is difficult to choose between (46a) and (46b):

The weaker fact can be established. however: PPl cases are always such that (46b) can be their structure. This generalization is statistically con- firmed by the fact that the corpus of PPl cases which is presented in Van de Velde (1977) contains not a single example which is not amenable to an analysis of the form (46b). In view of this generalization it becomes attractive to speculate that this option, i.e. (46b), is exactly what Dutch lacks. In other words, we may hypothesize that the difference between Dutch and German given in (41) reduces t o the following difference: - .- - - - - .

(47) - German Dutch _S X V : OK * (where X # g)

In addition, we need a principle which excludes the corresponding structures in which the left-hand clausal complement is adjacent to the matrix verb unless verb raising has applied. say (48):

extraposed clauses appears to be quite plausibly excluded under the assumption of the principles of the government-binding theory, in particu- lar if (proper) government is assumed to be directional (cf. Stowell (1981), Kayne (1983). Hoekstra (1982)). In Dutch and German the verb governs leftward. In this view, the empty category principle (ECP) of Chomsky (1981) will block extraction of an ex t r aposed~ lause .~~

This completes the argument. ECP linlits PPI to lefthand complements in Dutch and German. The *W-filter limits PPI to non-adjacent lefthand complements. And according to (47) non-adjacent lefthand complement

- clauses can exist in Gernlan but not in Dutch. Hence German has PPI but Dutch doesn't. Recall that. unfortunately, this is only half an explanation. The absence of PPI in Dutch is shown to follow from the independent difference (47). However it remains utterly mysterious, at least to the present author. why the difference (47) obtains in the first place.

4. MORE QUESTIONS

This means that, as usual. a small step in the direction of understanding 'raises many more questions than it answers. In conclusion, let us list some of the more salient ones:

1. Why do Dutch and German differ as stated in (47)?

2. Why do some Romance languages which are otherwise similar t o Italian, French for example. lack PPI?

3. Why is internal wh-movement in Italian not only not necessary but virtually impossible?

4. Why should a principle llke the *VV-filter be operative in a variety of languages?

Future research will hopefully shed some light on these questions. But by way of speculation, let us briefly dwell on the last of these questions. Several proposals have recently been made which provide a more general rationale for the existence of such a filter. Kaynez has suggested that the core case of the case-filter is *N-NP. Kayne proposed that this core case could be made to follow from some otherwise mysterious principle which requires that predicates take arguments and that arguments be next to predicates. Such a principle. in turn, is close in spirit to Stowell's 'Case Resistance Principle' which requires that there is a bifurcation of cate- gories into case-asigning and case-receiving ones, and furthermore, that

NOTES ! :1-- -~ - - ~.

1. In most vilrictics of German 'long wh-movement' as in (2) is supposedly urn grammatical. The present article pre tn ts the facts from the perspective of the mostly southern variants in which long movement is not excluded. In addition, the degree of ungrammaticality which obtains in northern varieties is relatively low, as aompared to, say, island vio htions. Consequently, the relevant contrasts are assumed to hold for the northcrn variants as wcll, albeit at a somewhat lower level of accept- abitity. Cf. also footnote 5. 2. Cf. Nanni and Stilluigs ( 1 9 7 8 ) for discussion. 3. Many of these properties arc discussed, though somewhat implicitly, in the rich -- material on the issue of PPI constructions presented in Van de VeMe ( I 977). 4. I ignore here the possibility that the infinitival complement clause might be a VP-complement lacking a PRO-subject. I take such an option to be excluded on principled grounds, cf. Koster and May (1982) . More interestingly, the reasoning presented below provides an a foniori argument against such a possibility and in favor of PRO-subjects in infinitival clauses. 5. Tilman Hohle has pointed out to me that for him complement clauses can quite easily precede a personal pronoun subject, which is, again according to him, general- ly impossible for NPs. He cites the following as an exarnplc:

(i) Ich h b e schon rnehrfach betonl, dap sie mif diesem Kerl bekannt zu I have already often stressed that her with this guy aquanted to machen ich mich immet strikt geweigen hube make I myself always strictly refuscd have "1 have alrcady stressed sevcral times that I havc always strictly refused to introduce her to this guy"

I do not share Hohle's intuitions completcly. I find that NPs and clauses by and large behave alike in this respect. Cf.

(ii) ?*. . . weil diesen Kuchen ich nichr mag ... becausc this cake I not like "because I do not like this cake"

(iii) ?*.. . weil dies zu sagen ich nichr wiinsclre ... because this to s y 1 not wish "'ternuse I do not wish t o say this"

- -.- .- Alongside (i), which I would characterize with a question mark, we have (iv):

(iv) ? Ich hube schon melvfach beronr, dap eines sulcl~en L'or\vur/r ich rnicl~ I have already often stressed, that of-a such reproach l myself hrna srrikt erwehren wrirde a h a y s strictly defcnd would "I havc alrcady strcsscd rveral times that I would always strongly defend myself again51 such a reproach"

However, if Hdhle is right, little follows because my main argument has to do with the internal dnrcture of a. On the other hand r v e r d other problcms would arise if we were to assume that a is in sentence initial position by virtue of some process

- of the anti-croisover effect is roughly identical in German and English. on-sen: , tential right adjuncts in German, for examplc, produce the crossover effect contrary to Guiron's principle:

(ii) Mit welcher Frau vom Scheichi glau brt du schbft eri am licbsten? with which wife of-the sheikh believe yousleeps he the dearest 'With which wife of the sheikh do you believe he likes to sleep most?

10. As expected, the strong crossovu effect remains intact for the relative pronoun itself:

.-

(i) *Das ist die Frau, mit der Herrn AWller bekannt zu machen ich ihr that is the woman with whom Mr. M. aquanted to make I to-her nicht empfehlen wurd not recommend would 'That is the woman to whom I would never recommend her to introduce Mr. Miikr"

(ii) *Das irt die Frau. mit d m ich ihr nichr empfehlen wurde, that is the woman with whom I to-her not recommend would Herrn Miiller bekannt zu machen Mr. M. acquanted to make

. .

11. Such as the binding theory of Chomsky (1981 ). 12. There are wme dialects of German in which it is possible to violate the DFC f ~ e r . This is true in particular of Bavarian. The standard varieties of German dis- cussed in the text allow no such violations, however.

Cf. Baya (1983) and Lencn (1983). It should also be pointed out that there is one construction in which the obhatory presence of dap is overridden by the presence of a wh-phrase which takes scope in some higher COMP. as indicated by a soope marker (was). The following illustrates this:

(i) Was g h b r t du wen (*dap) Peter gesehen hat? what believe you whom (that) Peter secn has - . 'Whom do you believe Petcr has seen?"

See Van Riemsdijk (1983) for a brief discussion of this construction. See also example (39) bebw. 13. The t a m purposive is used here in a very loose and unsemantic way to identify the urn-zu infinitives under discussion. 14. (33a) corresponds to example (94) in Van de Velde (1977). IS. Rehtive pronouns in German can be either [+w]-pronouns, such as in ulomif "with what" or zu wekher ''to whom", or [+dl-pronouns, such as dem "whom" or dessen "Whoae". This is why I usc the neutral feature [+rel]. 16. Interestingly, percohtion from a true complement position appears to be im- possible, in view of the impossibility of piedpiping with postnominal, leftward governing, adjectives (cf. Van Riemdijk (1981)):

- (i) der Konb, Idem trculAp er eu-ig blieb

the king to-whom loyal he forever stayed

Finally. Thilo Tappe suggests that there is a correlation between clitic extraction and bng wh-movement. According to him, peoplc who do not accept long wh-movement rlso fmd (42a) awkward. If this correlation holds. it has interesting consequences which, unfortunately, 1 cannot pursue here. 23. Condition (i), which is thc mirror image of the corresponding condition for Italian, is taken to follow from the directionality of government. In German and Dutch, both SOV languages, the verb governs leftward, while in Italian the verb governs &htwatd since Italian is SVO. Condition (iii), which is also applicable in Italian, follows if participles arc neutralized categories, i.e. not nnalyzrble as J+V.-N]. 24. (50a) giver the Dutch surface order. The corresponding structure for Gennan would be as in (i):

(Sob), which abstracts away from reordering, is, of course, identical for the two h g u g e s . 25. It might, in fact, be more correct to say that government into A-positions is directional while government into A-positions works both ways to allow for long wh-movement out of extraposed complement clauses in languages like Dutch. This view may, however, be problematic in view of the fact that stranding by r-movement out of a PP to the right of the verb is excluded:

(0 H q heeft eri de delcr geverfi lei mecjpp he has it the door painted with

I t is generally nssumed that the r-position, h e n ei, is an x-position. If it is. ECP can- not be ured t o account for the ungrarnmaticality of (i) under the awmptions stated above. 26. Talk at the 1982 GLOW conferencc in Paris entitled 'Predicates and Arguments'. Cf. llso Hoekstra (1983) for an extension of Kayne's proposal.

REFERENCES

Aoun, Y. 1981. "Parts of speech, a cas: of redistribution."in A. BeUetti et al., eds.: Theory of Markedr~ess in Ger~craril*~ Gmn~mar. Proceedings of the 1979 GLOW Conference. Pis : Scuola Normale Supcriore.

Bayer, J. 1983. 'Towad an explanation of certain tl~ar-trace phenomena: the COMP node in Bavariam" Unpublished manuscript. Aachcn. Originally presented at the conference on rntential complementation, UFSAL, Brusselr

Besten, J.B. den. 1982. 'Some xcmarks on the ergative hypothesis." Gronitlger Arbeiten zur Germanistischen Lhtguisrl 2 1 : 6 1-82.

Chomsky, N. 1981. Lectures on Governmenr atrd Binding. Dordrecht: Foris. cinque, G. 1981. "On the theory of relative clauses and markedness." The Linguistic . Review 1 (1981182): 247-294.

Cinque, G. 1983. 'ronnedcdness in constructions with sentence peripheral phrases and move a," Unpublished manurript,Venice.

Evers, A. 1975. The Transformational Cycle in Dutch and German. Dissertation, Utrecht Univerdty;distributcd by the Indiana University Linguistics Club. -- . . . .