7/28/2019 p.11

1/31

GLossoLoGA 17 (26) tt-41'

A FRE-CoCE TEM HDDEN N VERBALMoRPIoLoGYSBN ATRDOU*

This paper is couched rvithin the hypothesis that verbal mohoogy can provide a deter-miner on the event description. t explores a particuar construction in Greek and arguesthat it contains a free choice indefinite description. n the course of the investigation, thefollowing topics are addressed as well: existential constructions, interval semantics, themeaning of Universal Perfect, temporal modifiers, raising.1. ntroductionThis paper is an investigation within the working hypothesis that verbal mohol-ogy provides a determiner on the description of the (VP)-event (see Tsoulas1994, Portner t995, Baker and Travis 1997, Beghelli 1998). According to thestrongest version of this hypothesis, we have different types of verbal moholo-gy because v/e have different types of determiners. I will not attempt to argue forsuch a strong statement. will restrict myself to arguing that Some types of verbalmohology differ on providing different types of determiners for the event-de-scription. This paper is in certain ways a companion to latridou (2003). Wherevernecessary, will summarize Some of its relevant points.n this paper focus on sentences like (1) and argue that we can detect a Free-Choice (FC) event descriptin in the extended projection of the verb:

1. Echo pende chrnia na dho tonhave.lsg five years NA see thelit: have five years N see Mano* For helpful comments, would like to thanktopoulou, Noam Chomsky, Kai von Fintel,Mose Roumi Pancheva, David Pesetsky.

't has been five years since salv Mano' (to be refined)ManoMano

the following people: Elena nagnos-rene eim, Winnie Lechne malia

7/28/2019 p.11

2/31

t2 Sabine atridou2. Comparing three constructions

2.. The pu- and since-constructionsConsider the sentences in (2), which will be referring to as the "pu-construction"because of the factive complementizer pu and in analogy to what referred to asthe since{onstruction in atridou (203), vhich is also the English translation of(1)r:1. There re some variations on the pu-construction ( is the one in the text):(i) :echi/ ine pende chronia pu ton idhe teleftea forahas/ is five years PU him sa\r last timeB: echi/ ine pende chronia apo tote pu ton idhe teleftea forahas/ is five years from then PU him saw last tirneC: echun perasi pende chrnia *(apo tote) pu ton idhe teleftea forahave passed five years from then PU him sav/ last timeThere appears to be a dialectal split for the possibility of subject agreement on the ma-trix verb in /B vhen the verb is VE (i.e. for the verb to not be in the impersonal 3rdsingular). When the verb is BE, ageement is never possible, as far as can tell.l1 three variations show the now familiar pattern with RB/LB behavior (see atridou,naostopoulou and zvorski 2001, von Fintel and atridou 205).Variation B seems the partner of the following English phrase:(ii) t has been five years from/ since the time/ day that saw him lastVariation C is most similar to the following English construction:(iii) a. Five years have passed since sav him lastb. Five years have passed from/ since the time/ day that saw him lastVariation C and English (iii) seem to me to be garden-variety pefects, not temporalexistentials, though nothing depends on this decision.One important difference between the versions and B/C regards the possibilities forlong distance readings in the clause that describes the eventuality. n the C versions orerreadings are predictably available, given the -bar operator involved in the relativeclause:(iv) Echun perasi pende chronia apohave passed five years fromoti irthe o Kostasthdt came the ostas'Maria believes that Kostas arrived five years ago'The B variation permits a lower reading at least for some speakers, myself included:(v) Echi pende chronia apo tote pu nomizi i Maria oti irthe o Kostashas five years from then PU believes the Maria that came the Kostas'Maria believes that Kostas arrived five years ago'On the other hand, variation A does not permit long-distance readings:(vi) Echi pende chronia pu nomizi i Maria oti irthe o ostashas five yers U beliees the Maria that came the ostas(only reading: Maria believes something for 5 years)

tote puthen PU i Maria nomizithe Maria believes

7/28/2019 p.11

3/31

2. (the pu-construction)Echi/ine pende chronia pu pethane oA free-choice item hidden in verba morphoogy

thios tu3

has/ is five years PU died the uncle his't has been five years since his uncle died' (the since-construction)The light verb can be either B or HAVE. The pu-construction behaves in allrelevant respects like the since-construction; will not duplicate the tests here2.This difference between and the B/C versions is significant in that it puts more sup-port behind the possibility that pu in the version is a factive complementizer and notpa of a relative clause. f it had been the latter we rould have expected long distancereadings.with the since-construction, have found conflicting judgments on vrhether long-dis_tance readings are possible:(vii) othas been five years since Maria believes that Peter (has) left.2. The only interesting difference that have found peains to their licensing of NPs.The since-construction licenses NPs in the since-clause:(i) t has been five years since he budged an inch/ met anyone decent/ etc.But the pu-construction does not:(ii) a. *echi pende chronia pu to kunisa ap'edohas five years U it moved from hereb. *echi pende chronia pu piga puthenahas five years PU went anywhereWithin the theory according to vhich NPs are licensed in Downward Entailing Envi-ronments (see Fauconnier 1974, Ladusaw t977), we would have to start by showing thatthe sce-construction provides a Downward Entailing Environment.n order for he since-clause to be Downward Entaiing, the truth of (iiia) would have toentail the truth of (iiib):(iii) a. t has been o months since had (any) dessert

b. t has been two months since had baklavaThis does not appea to be the case. Certainly (iiia) can be true vithout (iiib) being true,as for example in the case vhere t'uro months ago did not have baklava but had tiramisu.This means that the relevant environments are not Downrvard Entailing and yet they li-cense Ns. von Fintel (1999) noticed problems of this sort and irgues for the followingamendment io the Downvard Entailint theory of licensing.(iv) "We can define a notion of entailment that will only check whether an inference istruth-preserving under the assumption that all the conventional implicatures and presup-positions of premises and conclusions are satisfied" (von Fintel t999).von Fintel calls this type of entailment "Strawson Entailment." So here is what vrehave. s we saw, (iiia) does not entail (iiib). oweve recal that the since constructinhas as presupposition that the eventuality in the embedded clause has, in fact, occurred, asit iS built in as its LB. To check rheher there is Strawson-Entailment, we need to Satisfythe presuppositions of the relevant sentences. That is what is done in (v), below. (va) pro-vides the environment whose Strawson-Down'ward-Entailing properties \e are checking.

7/28/2019 p.11

4/31

14 Sine atridoun a nutshell, a pu- and since- construction (in the Present Tense) like that in (2)mean that there are five years between the event of his uncle dying and now:n our calculations ''lre should assume the truth of (vb) along vyith that of (va). Now rve canask the question rvhether (vc) is entailed and the answer is that it is.(v) a. t's been five yers since had dessertb. Five years ago had baklavac. -> t's been five years since had baklavaSo von Fintet provides us ryith a way to apply the Dovrnward Entailing theory to cap-ture the licensing of NPs in the since-construction.But what about the pu-construction? Shouldn't it be able to license Ns in exactly theSame way? magine that Dovnvard ntailment is a necessary but not sufficient conditionfor N licensing. Recall Linebarger's (1987) mmediate Scope Constraint, according tovrhich an NP can be licensed by negation only if there is no operator that intervenes be-tween negation and that NP. This can be accommodated in the Down'urard Entailing ac-count as well. This is not to say that the environment vould cease to be DE. That is, the in-tervention effect of the intervening operators would not be due to the operators breakingor undoing the DE properties of the environment in question. The mmediate Scope Con-straint wuld be an additional, independent condition (to be attributed to Some other con-straint, possibly locality-related). With respect to the pu-construction, we can apply thevon Fintel-augmented DE approach and still find it DE. oweve the presence of the fac-tive complentizer pu is what causes the problem. n other words, a factive-marked com-pement cannot contain NPs, even if it is DE. There are other arguments for this:(vi) Nobody thinks that John ate anhing(vii) *Nobody found out that John ate anythingThe NP is not licensed in (vii), yet, it is easy to see that the environment remainsStrawson-Downward-Entailing. The intervening factivity could be the problem in (vii).Consider also:(viii)t has been five years since met anybody that liked(ix) *t has been five years since the time that met anybody that likedOnce we put an overt definite in (ix), the NP is not licensed. nd if the position thatfactivity should be assimilated to definiteness is correct (Melvold 1991; Baker and Travist997),then the unacceptability of (ix) is akin that of NPs in the pu-construction.But this conclusion is in conflict vith the tendency that we have found that the pu-con-struction arld the since-construction behave alike. There are several .'ways to go from here.We could say tha the since-construction and the pu-construction are not rftar alike andleave it at that. Alternatively, we could say that even though the truth of the since-clause ispresupposed, the since-clause is not a grarnmaticalization of a factive description of theeventuality. As a result, there is no intervention effect. lternatively, we could Say thatthe licensing conditions for any, and other English NPs, are somewha different from thatof Greek NPs, as has been argued by Giannakidou (1997). Howeve if we said that ng-lish any is different from Greek NPs in being licensed in factive environments, we wouldwrongly predict that (vii) is fine too. So perhaps the best way to go is to say that the since-clause is not grammaticalized as a factive clause.

7/28/2019 p.11

5/31

A ftee-choice item hidden in verba morphoogy3. five stEl.. .......Nos/L(eft) B(oundary) R(ight) B(oundary)The meaning of the pu/since-constructions is roughly this:4.1t [5 years (t) & t the TS & (LB (the TS)= the t' [his uncle died at t' ]) &RB (the TS)=91n prose, there are five years in the Time Span rhose Left Boundary is theevent of his uncle dying and rhose Right Boundary is the Time of Utterance. Iwould like to refer the reader to atridou (2OO3) for more details3. The conclusionthat we take from that paper is that the event in pu/since+lauses behaves like adefinite description. That is, it has an existential presupposition (the event musthave taken place) and a uniqueness presupposition (there should be only oneevent that fits the event description).The existential presupposition is intuitively easy to accept: if his uncle has notdied, then (2), or its negation, suffers from presupposition failure. Moreove it isclear that the existence f this event is not an assertion of these sentences. The as-sertion is about the "amount of time" between the events and the Time of Utter-ance. gain, see atridou (2003) for more details on the exact nature of this pre-supposition.The uniqueness presupposition on the event in the pu-clause can be illustratedas follovs:5. o ostas pandreftike stin eklisia ton Agion Apostolonthe ostas was married in the church of the oly postles'Kostas got married in the church of the oly Apostles'

Echi/ine pende chronia pu pandreftikehas/ is five years PU got married't has been five years since he got ma:ied'6. o Kostas pandreftike tris fores stin eklisia ton Agionpostolontheostas wasmarried three times inthe church of the Holy postles'Kostas got married three times in the church of the Holy Apostles'#*Echi/ine pende chronia pu pandreftikehas/ is five years PU goi married't has been five years since he got married'

3. n atridou (2003), mention was made of the "Perfect Time Span" instead of "TimeSpan," as that paper \ryas cast in the context of a discussion about the Perfect. That neednot preoccupy us here.

t5

7/28/2019 p.11

6/31

r6 Sabine atridouThe second sentence in (6) suffers from the fact that there is no unique marry-ing event that the definite description in the pu-construction can pick out. Thesentence becomes fine once ve add "ya teleftea fora" ('for the last time'), but bydoing that we have again created a uniquely referring event description. More ar-guments for the uniqueness presupposition can be found in atridou (203).n short, the pu-construction contains a definite event description of the eventin the pu-clause, since it has both an existential and a uniqueness presupposition.From the point of view of what is to come, it is importantto reaize that the timespan in (3) is empty of events of the type of his uncle dying and it is also impor-tant to realize ho this is derived in the pu/since-constructions: because thepu/since

7/28/2019 p.11

7/31

A free-choice item hidden in verba morphoogy 17both cses the event of his uncle dying and of seeing Mano must have taken place,or the constructions are infelicitous. We will return to the question of the unique-ness presupposition shortly.As with the pu-constructions, the temporal adverbial is obligatory in the na-constructiona:

8. xEcho na ton dhohave N him seeOn the surface, the na-construction differs from the pu-construction on a varietyof moho-syntactic points; for example, agreement on the verb (to vhich ve willreturn) and the fact that the pu-construction can contain either HAVE or BE, lvhilethe na-construction is restricted to }vE. But there are important differences be-tween them in their conditions of use and we can learn a lot from these differences.Consider what happens vrhen the two constructions are combined with unique, non-recurring events (for reasons of space, the since-construction lvill appear as thetranslation of the pu-construction, though it is itself a member of the paradigm).9. Echi/ine pente chronia pu pethane ihas/ is five years PU died the't has been five years since his cat died'10. *# i gata tu echi pende chroniathe cat his has five years

gata tucat hisna pethaniNA die

t is obvious vhat is wrong with (10) once we consider what ve need to accom-modate to make it good: (10) conveys that his cat is able to die more than once.4. The sentence is grammatical on the relevant reading if one makes a paicular gesturewith one's hand (circling clockvise at the elbow) which indicates 'a long time'. other_wise, it is close to a sentence that is interetable as a modal:(i) echo na dho ton yatro stis 5have.1sg N see the doctor at 5' have to See the doctor at 5'oweve the Greek construction is not identica to the English have ro translation. Forexample, consider the follo'ring, where in English have to can appear in place of musf butin Greek it cannot:(ii) a. ' mustave to see John in order to understand'urhat he has'b. prepi/ *echo na dho ton Yani ya na katalavo ti echimust/ *have.lsg na see the Yani in orderto understand rvhat hasc. f you want good baklava, you must/have to to to Chatzi'sd. n thes kalo baklava prepi/ *echis na pas stu Chatziif v/ant.2sg good baklava, (you) must/ have na go to Chatzi's

7/28/2019 p.11

8/31

18 Sabine atridoun this, (10) feels akin to the English negated perfect:

11. #His cat hasn't died in five yearsHoweve there is an important difference betveen the na-construction and the

English negated perfect: the na-construction, like the pu/since{onstructions,clearly has an existential presupposition on the event. That is, (10) conveys thatthe cat died (at least) one time (at least) five years ago. On the other hand, thenegated Perfect in English has no such presupposition:L2. A: Has the patient had a stroke?B: The patient hasn't had a stroke in five years/ in the five years that havebeen working here. don't know about before that.Going back to the contrast betvreen (9) and (10), what do these facts tell us?Unique events are fine in the pu-construction. n fact, in atridou (2003), arguedthat they re necessary for the since-construction and therefore by extension tothe pu-construction. Howeve for the na-construction to be felicitous, thereshould exist the possibility of more than one occurence of the event-type de-scribed in the na-clause. That is, the na-construction is best vhen in principe,more than one event fits the description of the LB-eventuality. will call this the"plurality presupposition" of the na-construction but it should be understood asmeaning "possibility for plurality." t follows that the na-construction is infelici-tous vith event descriptions that refer to events that by their nature are unique.This is of course the exact opposite of what we have seen in the pu/since-con-structions.This difference beteen them makes us also understand better rhen each of the

u- or na-constructions is used. When an event either happened only once or canbe uniquely described (when we have a uniquely referring LB-event description),the pu-construction is used. When we have more than one event that fits the LB-description, the na-construction is used. Compare the contrast in (14) - (15), in thecontext of (13):

13. o Yanis travmatistikethe Yanis was injured dheka chronia,ten years,prin apo pende chronia ke prin apo ena minabefore from five years and before from one month'Yanis was injured ten years ago, five years ago and one month ago'

The unmodified (i.e., without "for the last time") pu-construction is not possi-ble:

prin apobefore from

7/28/2019 p.11

9/31

A free-choice item hidden in verba morphoogy 1914. Me ala loyia, echi ena mina pu travmatistike *# (ya teleftea fora)with other words, has one month PU he was injured *# (for the last time)

But the na-construction is just fine:15. Me ala loyia, echi ena mina na travnatistiwith other words, has one month N ras injuredThe status of (14) (without the expansion) is due to the infelicitous use of a def-inite description rvhen there is more than one individual that fits the description.s lvith (6), the sentence with the expansion is fine because then we createuniquely referring event-descriptions. On the other hand, the na-construction isperfectly fine as its plurality presupposition is satisfied in the context of (13).We can understand this pattern once we make the assumption that in contrastto the pu/since-constructions, which contain a definite description of the LB-eventuality, the fla-construction contains an indefinite description. t is not possi-ble to describe unique events vith indefinites. The unacceptability of (14) is akinto the unacceptability resulting from an indefinite determiner on NPs referring to

things which are presuppsekno}vn to be unique. When a definite description israrranted, a definite description must be used and an indefinite description is in-felicitous (awkins 1978, eim 199|, and others). The possibility of plurality isrequired for the felicitous use of indefinites:L6 a. ratched a soccer game last night.Thel*a1*one referee v/s very unfair.b. vatched a married couple play chess. hel*a1*one man had noendgame.When the context does not entail uniqueness, an indefinite must be used.17 a. watched several games over the lveekend. /one/*the referee }as veryunfaif .b. lvatched many married couples play chess yesterdaY. /one/*the manhad no endgame whatsoever.The incompatibility of unique events vith the indefinite LB-event descriptionof the na-construction aso predicts the following contrast:

Pu/sin ce-construction :l8.Echi/ine dhio chronia pu ton ifta ya teleftealproti/triti/ etc. foraHas/is two years PU him saw.lsg for last/ first/ third/etc. time't has been two years since saw him for the last/first/third/etc. time'5. Unless it is clear that all the games had the Sme referee.

7/28/2019 p.11

10/31

Sabine atridouna-construction:

19. *Echo dhio chronia nahave.lsg two years N tritithirdyafortonhim dhosee teleftea/protillast/ first/ foratimeWe sav earlier how modifiers |ike first, Iast, third etc. can be used in the

pu/since+onstructions to yield a unique event description. This explains the ac-ceptable status of (18). on the other hand, the na-construction in (17), s an indef-inite, has a plurality presupposition and (19) suffers from vhat *a first/ast/thirdperson in the rov suffer from. n other ords, we See behavior that supports thefollowing difference between the pu- and na-constructions (though we will seemore differences later on):2O. pu/since-constructions :...since (the time of6) tre event of VP (my seeing him)2'1,. na-construction:...since (the time of) NDEF (an(y)) event of VP (my seeing him)

3. A free-choice itemn the previous Section 'we Sa}v that there is good reason to believe that tbe na-construction contains an indefinite event description. n this section we will ad-dress the question of the more exact nature of NDEF.n intuitive description of the constructions ve have been ooking at is thatthere is a (temporal) zone that is empty of events of a certain type. We saw howwe derived this "emptiness" in the case of the pu/since-constructions: as the eventin the pu/sincelause is unique, there simply is no other such to be found in thetime Span beteen LB and RB. We will see later hov to derive the emptiness ofthe time span in the na-construction. n addition we saw that in the na-construc-tion the event has to be something that is in principle repeatable (the pluralitypresupposition). For this reson, will run the discussion in this section in parallelto that of a close topological analogue suggested to me by rene eim (p.c.)(where NDEF can rewrite as a or any in English):

22.We are five miles (away) from NDEF gas StationThe "gas station sentence" appeqrs similar to the na-construction in that it alsocreates aone empty of a certain type of individual. To highlight the parallel, con-sider the following alternative paraphrase for the na-construction:

6. See Glasbey (1992) for arguments that "a temporal entity from [an] event entity" (p.289) is not all that straightforward and that ve should maintain a distinction betweenevents and times.

7/28/2019 p.11

11/31

A free-choice item hidden in verba] morphoogy 2|23. We are five years (away) from NDEF event of my seeing himFor both the na-construction and the gas station sentence, the question is whatprecisely the nature of NDEF is. The gas station sentence has its own intrinsic in-terest but here ve will be focusing only on those of its properties that provide a

bridge to understanding the na-construction.The first option to consider for NDEF in (22) - (23) can be ruled out quickly,namely that of a specific indefiniteT. The gas station sentence (with expansion a,no any for NDEF) can certainly be understood as containing a specific indefi-nite, as for example when }e are ignorant about any other gas station closer by,or when we cannot exclude that there are others closer than five miles to us, oreven when we knorv that there are. So the reading with the specific indefinitecould be paraphrased as follovs:Z4.There is a gas station that we are (exactly) five miles away fromCrucially, (z4) can be truthfully uttered also rhen there is another gas stationthat is closer by than the one the speaker has in mind.oweve the na-construction cannot have this interetation at al. That is, ifsentence (25) is true, there is no other event of my seeing him that is closer thanfive years to us.25. Echo akrivos pende chronia na ton dhohave.lsg exactly five years NA him see't has been exactly five years since saw him'This is a crucia difference between the gas station sentence and the na-con-struction and it tells us that even if NDEF can be a specific indefinite in the gas

station sentence, NDEF cannot be a specific indefinite in the na-construction.We conclude then that the na-construction does not contain a specific indefinite8.s ve are interested in the gas station sentence only insofar as it can help us un-derstand the na-construction, ill not be discussing the specific indefinite inter-pretation of the gas station sentence any fuher.The next option to consider for NDEF is that of an existentia quantifier. t7. owever one implements specific indfinites.8. Another rgument supporting this S{me conclusion is that in the interetation of N-DF as a specific indefinite in the gas station sentence, i iS possible to have a pronomi-nal refer back to it:(i) We are five miles from a gas station but it is a very expensive one.

Such pronominal back reference is not possible with the na-construction, though thereare means to refer back to events in Greek (see atridou and Embick t997).

7/28/2019 p.11

12/31

Sabine atridouseems hard to make this work in our cases. These sentences would be predicted tomean that we are five years from some event of the relevant type and therefore itwould again be wrongly predicted that the na-construction would be true if there}vere closer events. n fact, in the absence of negation it is hard to imagine hovthe right meaning of the na-construction vould be derived if NDEF rvere an exis-tential quantifier.The final option for NDEF that we will consider is a free-choice indefinite (thiswas proposed for na-clauses inside relative clauses in Begheli 1998). I will followa common practice and assume that FC indefinites are vide_scope universals,augmented by whatever properties FC items differ in from other universa quanti-fiers. will not go into vhat all FC adds to the universalforce; there is ample liter-ature on this topic.Let's start with the gas station sentence. f we are dealing with a Free-Choice(wide-scope) Universal then the gas station sentence would mean the following:

26. We're five miles from every+FC gas station27.ery+FC statin is such that v/e re five miles from itone might think that (26) _ (7) would force us to a reading where ve are atthe epicenter of a circle of a five-mile radius, on the periphery of which are gasstations-and this is not necessarily the meaning of the gas station sentence. Butthis is not a problem: if five mies stands for at east five mies, we get the correctmeaning of the gas station sentence:28. Every+FC station is such that we are at least five miles from itSimilarly, on this view of NDEF,the na-construction vould look as follows:29.ery+FC event of us seeing him is such that we are at least five years from itOne further argument in favor of (29) can be found in ho we derive the resultthat a certain time span is empty of events of a certain sort in each of the threeconstructions that we have been looking at. n he pu/since-constructions we gotthis result by having the definite event description picking out the unique (oruniquely salient) eventuality that fit the description of the pu/since-clause. Perforce, the time span had no other events of this sort. But as v/e also saw, the na-construction has partially overlapping truth conditions in that the time span inthe na-constructin is also empty of events of the relevant sort.How can we derive the desired result for the na-construction? Here is a firstapproximation of the answer. f ve are five years away from any event of therelevant type then \ve are five years away from the most recent event of therelevant type - the way that being five miles away from any gas station entails

7/28/2019 p.11

13/31

A free-choice item hidden in verba morphoogy 23/e are five miles away from the closest gas station. will conclude then that afree-choice indefinite (event) description in the na-clause prduces the rightresults with respect to this important interetive feature of the na-construc-tion9'10.

9. This is not to say that a na-c|ases are necessarily free choice indefinites. There aredifferent types of na-clauses and the cases would have to be looked at individually.10.one might raise the following two objections to the position that NDEF in the gas sta-tion and na-construction is FC.The first possible objection has to do vith the fact that FC any has been claimed tostate a policy (Dayal 1998). Consider the following:(i) This store accepts checks from eerylany bankn the expansion lith any, the sentence is said to convey the existence of a policy thatif a ne'w bank is built, this store will accept its checks. This is said to be due to anyranging over possible banks as well as actua ones. No such policy is entailed withevery. our gas station sentence, howeve states no policy. f another gas station isbuit, it is not claimed that it will be at least 5 miles from rvhere we are. Similarly, thereis no matter of policy in the na-construction. s this an argument that any in the na-construction and the gas station is not FC (but NP, for example)? There are languageswhere the FC items are not homophonous with Ns. Greek is such a language. nGreek (talian and other languages), the gas station Sentence cntains a FC item (andcannot contain a NP):(ii) maste pende milia apo opiodhipote venzinadhikowe-are five miles from whatever-FC gas stationThis is expected given everything that we have said so far. But v/hat about the issue ofpolicy implications? There is no policy about vhere gas stations can appear in theGreek tas station sentence any more than there is in the English one. One might re-spond by saying that Greek FCs never implicate a policy. But this is not so. n otherpolicy-testing environments, Greek behaves the same way as English with respect topolicy implications, aS long as the environment contains mperfective spect:(iii) fto to magazi dhextan/ *dhextike epitayes apo opiadhipote trapezathis the store accepted.Mr7 *ppa checks from FC bankn short, in Greek, a policy-implication is a possible but not a necessary correlate ofovert FC items (possibly a correlate of the mperfective). This means that the absenceof policy implications with the covert FC item am postulating in the na-constructionis not an argument against the propsal that there is a FC description in the na-con-struction. Given the absence of policy implications in the English gs-Station sentencewith any, will ssume that for English as rvell, policy implications are not a necessarycorrelate and willinteret von Fintel (2000) as essentially arguing that arbitrariness iswhat is involved with FCs and policy is one possible source for that but not the onlyone.

7/28/2019 p.11

14/31

24 Sabine atridou4. The structures of tbe pu/since- and na-constructinsWe have seen how the meanings of the pu/since- and na-constructions differ. Buthow do their structures differ? Consider the following examples of na- and pu-constructions, vhere the temporal pivot is in square brackets:

30. Echo [pende chronia] na dho ton Manohave five years N see the Mano31. echi/ine [pende chronia] pu idha ton Manohas/is five years U saw.lsg the ManoThe na-construction permits a greater variation in the choice of temporal piv-ot than the pu-construction. The na-construction permits amount pivots as rellas apo ('from/since')-adverbials. The pu-construction permits only the former:32. Echo na ton dho a. pende chroniahave.lsg NA him see five yearsb. apo to t99ll apo totelfrom 1991,1 from then/'since L991' 'since then'c. apo tote pu fagame mazifrom then PU ate.lpl together'since \re ate together'The second possible objection to NDEF being FC has to do with a fact that v/e Sawearlie namely, the LB-eventuality is presupposed to have occurred. This is what wehad called the existential presupposition of the na-construction. The question isrhether we can postulate a FC item in a contet with existential presuppositions.Howeve the fact is that the Greek gis Station Sentence (and an talian version),rhich unambiguously contains a free-choice item, also has an existential presupposi-tion, as shown in (ivb), and as expected the English gas station does too(iv) There are no unicorns...a... # Therefore, 'we are five miles from any unicorn.b... # ]imaste pende milia apo opiondhipote monokero

\rye are five miles from FC unicornContrast the unacceptability of (iv) vith:(v) There re no unicorns. Therefore there are no unicorns in our parking lot.So here is no way around the fact that at least some FC behave as if they are compat- \ible with existential presuppositions. opefully future vrork will permit us to under-stand the environments under rryhich FC items appe{r with existential presuppositions. conclude that these two possible objections to NDEF being FC have been overcomeand that v/e can maintain the hypothesis that the na-construction contains a FC indefinite description f the LB eventuality.

7/28/2019 p.11

15/31

A free-choice item hidden in verba morphology 2533a. ine/ echi [pende chronia] pu to idhais/ has five years PU him saw't has been five years since I saw him'b. *ine/ echi [apo to 1991] pu ton idhais/ have [since the 1991] PU him saw

c. *ine/ echi pu ton idha [apo to 1991Jis/ has PU him saw [since the 1991JThere is no reson to believe that in the pu/since-construction the temporalpivot is anything other than a temporal amount, specificaly, the amount phrasewhich specifies how much time fits in the time span between the LB-event and theRB. n the na-construction, howeve the fact that the temporal amount pivot canbe substituted by apo ('from/since')-adverbials should make us pause. Such adver-bias are typical LB-adverbials in the regular erfect (just like since):Existential Perfect:34. po totel apo to 1991', echo pai tris fores sto Londhinosince then/ since 1,99t, have.lsg gone three times to the London'Since then/ since |991, have gone to London three times'35. po tote pu efiges, echo pai tris fores sto Lndhinosince PU left.2sg have.lsg gone three times to LondonUniversal Perfect:36. fto to vivlio o Yanis to grafi apo to 1990this the book the Yanis it writes since the 1990'Yanis has been writing this book since 1,990''Sentence (36) is the equivalent in meaning of the English Universal Perfect, as

its translation indicates. Greek (among other languages) cannot express themeaning of the U-Perfect with the Perfect mohosyntix (see atridou, Anagnos-topoulou and zvorski 2001 for the reason vhy).To convey the relevant meaningthe simple tenses are used, but with the exact same set of adverbials that permitor force the U-Perfect reading in English. will still be referring to the meaning of(36) a.s the "U-Perfect reading" but the reader should keep in mind that in Greekthe actua mohosynt( for this reading is not the perfect.So one class of temporal pivots of the na{onstruction are actually LB-adver-bials. The other class, ts we sav, seemed identical to the temporal pivot in the pu-construction, namely, temporal amounts like five years.What we see then, is this: in the pu/since+onstruction, the LB of the time spanis the event in the pu/since+lause. The temporal amount pivot describes the sizeof the time span.n the na-construction, on the other hand, we have the event of the na-clause

7/28/2019 p.11

16/31

26 Sabine atridouand one of a temporal amount pivot or an LB adverbiat. The obvious questionthat arises is the following: vhen we have an LB adverbial, what role does theevent play? t cannot be the LB of the time span, unlike the event in the pu/since-construction, vhich is clearly an LB. Here, schematically, is the problem:

37.

oveve it turns out that there is homophony in Greek that has fooled us. InEnglish, we have the temporal amount phrase ten years, vrhich appears in thesince-construction, mong other places. We also have for ten years, which is anLB-adverbial, which appears in fte has been writing this book for ten years (seeatridou, nagnostopoulou and zvorski 2001 and references therein).In Greek the equivalent of English erfect-lee| for ten years does not have thepreposition, i.e., it is plain dheka chronia ('ten years')l1:

38. ser ton osta dheka chronia (equivalent to the English U-Perfect)have knorn.lsg the Kosta for ten yearsThis means that it is not possible to distinguish, just from the form, vhether atemporal phrase like pende chronia ('five years') is a temporal amount argument(as argued it is in the pu-construction) or an LB-adverbial.f we take this homophony into account, we can unify the class of temporalpivots that appears in the na

7/28/2019 p.11

17/31

A free-choice item hidden in verba morphoogy 27This, in turn, means that the role of the pivot in the na-construction is entirelydifferent from that in the pu-construction:39.

Event i pu/since-or naclause Temporal aountargument LB-adverbialPu/since- con-struction Sets Left Bound-ary of time span Specifies size oftime span Not possibleNa-construc-tion Role as yet un-clear Not possible Sets Left Bound-ary of time spanWhat does all this tell us about the syntax of the t'wo constructions?n the pu-construction, the temporal measure (five years) is the first argumentin an existentia and the pu-clause is the LB-adverbial. That is, the representationof a sentence ike (40) is that in (41):4O.ine/echi pende chronia pu pethane o thios tuisas five years U died the uncle his't has been five years since his uncle died'4t.ight verb"*irrential [five years]u [in the time span vhose LB is the event ofhis uncle dying and whose RB is NOWIs(40141) are akin to (aua!:42.Thercare five liters of water in the balloon43. Iight verb""i"lentiar [five liters of water]u [in the balloon]oThe representation in (41) captures the fact that in the pu/since+onstructionthe main assertion is the size of the time span. The boundaries of the time span(and therefore its very existence) are presupposed. (41) also makes clear whyonly amount arguments can be the temporal pivot of the pu/since+onstruction.On the other hand, according to (39), the na-construction is being argued to lookas in (44), with [five years] not being the first argument of an existential but anLB adverbial, just like slncelauses:(ii) *echi/ine edh ke dheka chronia pu ton idha (pu-construction)has/ is here and ten yeirs U him saw.lsgThis is also an LB-adverbial (and therefore it is predicted to occur with the na-construc-tion but not the pu-construction):(iii)grafi afto to vivlio edho ke dheka chroniawrite.3sg this the book here and ten yeirs

7/28/2019 p.11

18/31

45. P

28 Sabine latridou44. echo na ton dho [apo to 19901 pende chronia]have.lsg N him see [since the 1990/ five years]The difference betreen (a1) and (44) captures the fact that the tempora pivotplays a totally different role in the two constructions. For the Pu/since+onstruc-

tions, one can now transfer one's favorite representation of existential sentences.For the na-construction, what we have is the following:

NP

ton dhohim see.lsgThe representation in (45) captures the fact that the temporal pivot in the na-construction is always and only an LB-adverbial.n the na-construction, the amjunt pivot pende chronia appeirs in the c- \cusative. According to (45), this Accusative is not associated with the matrix verbhave. This appears coTect: ccusative is the Case that al temporal adjuncts ap-pear in. This is supported by the fact that the temporal adjunct appears in the Ac-cusative no matter vhat the matrix verb is:

['

^.,/-\./\VP ?!1 chronialapo to 1991\' (for) five years/ since the 1991

VP

7/28/2019 p.11

19/31

A free-choice item hidden in verbal morphoogy 2946.Perpata ena chronowalks one year.cc'He has been walking for one year'47.ne arostos ena chronos sick one year.Acc

'He has been sick for one year'n the pu-construction, vhere the temporal pivot is an argument of the matrix(existential) verb, the Case n the temporal pivot is predicted to depend on theverb. Recall that the pu-construction has a choice between be and have. When theverb is have the temporal pivot appears in the Accusative, when the verb is be,the temporal pivot appears in the Nominative:48.Echi enan chrono pu ton idhahas one year.cc PU him saw.lsg49. ne ens chronos pu ton idhais one year.Nom PU him saw.lsgThese facts support the position that the temporal pivot is an adjunct in the na-construction but an argument in the pu-construction. We will return to more as-pects of the structure in (45) shortly.

5. First attempt at composing the meaningFinally v/e come to the question of hor to derive the meaning of the na- andpu-constructions. have argued that the possible temporal pivots of the na-con-struction are all LB-adverbials. Specifically, we Saw that these include apo-ader-

bials and (for)-five-years-adverbials. The latter, as LB-adverbias, {re compatibleony with the Universal Perfect (see atridou, Anagnostopoulou and zvrski2t for arguments). Now, recall that in Greek, the meaning of the Universal Per-fect is done with the simple (imperfective) tenses. f we combine these facts andlook again at representation (44) (repeated below), we see that the na-construc-tion looks exactly like other U-"perfects" in Greek:44.Echo na ton dho [pende chronia/apo to 1990]have.lsg NA him see [five' years/ since the 1990]50.sero ton Kosta [pende chronia/apo to 1990]know.lsg the Kostas [five years/ since the 1990]' have known osta for five years/since 1990' (the "U-Perfect" in English)51.ne arostos [pende chronia/apo o 1990]is sick [five years/ since the 1990]

7/28/2019 p.11

20/31

30 Sabine atridou'He has been sick for five years/ since 1990' (the "U-Perfect" in English)

n (50) - (51) it is clear lvhat is going on: the predicates ksero ton osta and lnearostoshold throughout (since it is a U-Perfect) the interval whose LB is five yearsago or in 1990 and rhose RB is the moment of utterance. Similarly' (44) conveysthat the predicate echo na ton dfto holds throughout an equivalent interval.The next question, of course, is to determine what echo na ton dho means andhor it yields the meaning of the na-construction. There is a modal constructionthat has a similar look (see also footnote 4):

52.echo na dho ton yatrohave.lsg NA see the doctor' am scheduled to see the doctor'Since \ve see the string echo na ton dho also in this modal construction, oneonders whether the na-construction shares part of its meaning with it. But themeaning of the modal is quite different from what we have in the na-construction,hich lacks any meaning of obligation or schedulel3. Fuhermore, certain typeof subjects cannot appear in (52) for obvious reasons, while they have no prob-lem in the na-construction:53.Echi na weksi pende vdhomadheshas N rains five weeks54. *Echi na vreksihas N rainsattempted: 't is scheduled to rain' conclude that (52) is not part of the meaning of the na-construction. Rather

than taking you through everhing that echo na ton dho does nof mean in the na-construction, let me take you directly to what think is going on. n order to dothis, tve vrill first need to sholv that the subject of the na-constructin is derived.n other words, we need to show that there is indeed -movement of the subject,as depicted in (45).6. A_moement in the na-construction

n this section will argue that the matrix subject of the na-construction origi-nates inside the loer clause (tbe na-clause) and that (44)' in fact, is better repre-sented as (55), already foreshadowed in (a5):13. For more on the modal construction in (60), see von Fintel and atridou (205).

7/28/2019 p.11

21/31

55.DPk echoDP have.1sg

56 a. (emis) echume pende(e) have.I fiveb. (esis) echete pende(you.pl) have.2pl five

A free-choice item hidden in verba morphoogy 3tna ton dhol [apo to l99Dlpende chronia]NA him seel [since the 1990/ five years]ItIt

There are several reasons to suspect this is on the right track. For one, the lightverb has the Person and Number features of the embedded subject:chronia nayears Nchronia nayears NA

ton dhumehim see.lplton dhitehim see.2plOne might counter that the na-construction cannot possibly contain Raisingsince the verb of the na-clause has overt agreement features as well. This shouldnot vioTy us, however. Determining finiteness in the Balkan languages is wellknown as a difficult issue. There is good reason to believe that even PRo co-e-ists with agreement on the verb in these languages. t has been argued that the testfor "non-finiteness" is not the presence/absence of agreement but thepresence/absence of a temporal domain that is.independent from the higherclause (atridou 1988, Varlokosta 1,994'Terzi L993, and many others; see Philip-paki-Warburton & Catsima|i t999 for an opposing view; see lexiadou andnagnostopoulou 2000 for an overview of the debate on this issue). will ssume the following description of the general environment from latri-dou (1988; see also Varlokosta 1994,Terzi t993, Alexiadou and nagnostopoulou20): if the na-clause can be put in the Past Tense, its subject can appear in theNominative (in addition to the option of pro, Greek being a pro-drop language).f the na-clause cannot be put in the Past Tense, the subject cannot appear overtlyin the Nominative. t must either remain as PRo o possibly in some restricted

cses, be assigned ccusative through EcM. n addition to this correlation, whenthe na-clause can be put in the Past Tense, its empty category subject can referoutside the clause, hence supporting its status as pro. On the other hand, when thena-clause cannot be put in the Past Tense, its empty category subject cannot referutside the clause, Supporting its status as Ro. n short:

S7.ossibility for Past Tensela in a na-c|alse correlates with14. While accepting the verall proposa| of atridou (1988)' Varlokosta (1994) suggestedthat the notion of "Past Tense" in (32) should not just include simple past tense, as had in 1988 but any periphrastic tense that conveys "past," specifically also thePerfect. She suggested this because of examples like (i), which contain Nominative forthe embedded subject but which cannot, according to Varlokosta, contain simple past:(i) *poli thelo na efije o Petrosmuch want.lsg N left the Peter

7/28/2019 p.11

22/31

32 Sabine atridou- possibility for Nominative on the subject- vhen covert, the subject can refer outside the sentence (i.e., it is pro)

58. mpossibility for Past Tense in a na-c|ause colTelates rvith- no possibility for Nominative on the subject- when covert, the subject cannot refer outside the sentence (i.e., it is PRO)

From the above, atridou (1988) concluded that the source for Nominative inGreek is +Tense and not +GR, as all na-clauses shov agreement but only somecan provide Nominative for their subject.Now'what about the na-clause in the na-construction that re re investigating?Can it be put in the Past Tense? The ansv/er is no:59.Echo pende chronia na ton fto/ *idha/ *echo dhihave.lsg fiv years N him see.RS/ *sar.PST/ have seenThe subject cannot appear in the Nominative:60.Ego echo pende chronia (*ego) na ton dho have.lsg five years (*) na him seeThe embedded subject cannot be distinct from the matrix subject, indicatingthat the embedded subiect is not pro:61. *Ego echo pende chronia na ton dhi have.lsg five years NA him see.3sg conclude, therefore, that the na-clause of the na-construction is as close asGreek has to a non-finite clause and that (as in the context of the aforemen-

tioned discussion) the presence of agreement on the embedded verb is not an ob-stacle in the path to accepting the presence of Raising in the na-construction.The reader is referred to lexiadou and nagnostopoulou (1999) for tests forRaising in Greek.A final-test that is applicable to the construction at hand and that diagnosesRaising, is the following. t has been observed that there is no PRo expletive(Jaeggli and Safir 1989 and references therein). The grammaticality of the follov-ing then indicates that the EC subject of the na-clause cannot be PRO but must bean A-trace (the example is chosen to satisfy the semantics and felicity conditions *of the a-construction):(ii) poli thelo na echi fiii o Petrosmuch vant.lsg NA has left the eter' vant himer to have left'

7/28/2019 p.11

23/31

A free-choice item hidden in verbal morphology 3362.chi5O chronia tc na anakinothi oti metapsimchothike enas Lama]has 50 years IEC N be announcedthat was reincarnated a Lama]'t has been 50 years since it was announced that a Lama has been reincar-nated' conclude then that the na-construction contains Raising. We can now go backto composing its meaning.

7. Second attempt at the meaning of the na-constructionNow that we have established the existence of Raising, we can go back to therepresenation in (a5):45.

N '

^, VP i!q chronia/apo to 1991' (for) five years/ since the 1991Ech+ave+

VPton dhohim see.lsg

Recall that this is in a sense a U-Perfect in that the matrix predicate is assertedto hold throughout the time span the LB of vhich is one of the LB-adverbials and

ra

7/28/2019 p.11

24/31

34 Sabine atridourhose RB is the Time of Utterance (when the na-construction is in the presenttense). fter the discussion about -movement, we aso knovr that the predicatethat is asserted to hold throughout the time span is Lght Verb [DP na VP].Theexistence of A-movement permits us to recognize the matrix Light Verb as an im-personal verb, i.e. a verb without a thematic subject. Since this verb is HVE andsince the existential verb in Greek is, in fact, HVE, argue that the matrix verbin the a-construction is an existential verb.n short, the matrix predicate in the na-construction is an existential construc-tion. The fact that the matrix predicate is also a U-Perfect means that the exis-tence is asserted to hold throughout the time span. What is it that the matrixpredicate asserts the existence of? argue that the matrix predicate asserts the ex_istence of a time span between two points. This means that there are two timespans involved in the na-construction. On the one hand we have the matrix timeSpan, whose LB is the LB-adverbia, and whose RB is set by Tense. This is theTime Span v/e are already familiar vith; will be referring to its LB and RB asLBm and RBm, where 'm' stands for 'matrix.' On the other hand, we have theembedded time span, vhich is the time span the existence of which is the contentof the main assertion of the na-construction. s all time spans, the embeddedtime span also has an LB and an RB (LBe and RBe; 'e' for embedded). We willreturn shortly to what these are. So fa then, we have the following for the mean-ing of the na-construction:

63. Throughout/ for every point in the time span between LBm and RBm thereexists a time span between LBe and RBeRBm is a variable, set by matrix Tense (i.e., it is the Time of Utterance in theresent tense, an interval in the Past, with Past tense etc. (see atridou, Anagnos-topoulou and zvorski 201; von Fintel and atridou 205). LBm is set by the

LB-adverbial. What are LBe and RBe? will argue that LBe is the event descrip-tion provided by the na-clause of the na-construction (the one we say contained aFC event description). RBe is a temporal variable, just like any RB of the timespans we have been looking at. argue that RBe covaries with time points of thematrix time span, universal quantification over vhich yields the U-Tense/Perfectreading, so that (63) is effectively (64):64.For every point in the time.span between LBm and RBm there exists atime span between LBe and it.Consider example (65); lvhat re have said about the na-construction is repre-sented in (66) - (69), with (69) being the composite meaning:65. Echo na dho ton Mano pende chronia/ apo to 1990have.lsg NA see the Mano five years / since the 1990 etc.

7/28/2019 p.11

25/31

A free-choce item hidden in verba morphology 3566. MatrixTimeSpan:LBm is set by LB-adverbial (pende chronia,'for five years' or apo to 1990,'since 1990')RBm is set by Tense67 . t (t C MatrixTimeSpan + p holds at t)68.P= a Time Span (EmbeddedTime Span), between FC any event of the na-clause type (LBe) and t (RBe)The na-construction:69. Vt (tXMTS * "FC e (na-clause (e) + I Time Span between e and t)1n (69) we can also see the role that the Free-Choice description of the eventplays.Let's say that (7) represents the MatrixTime Span, which is composed byLBm (temporal pivt) and RBm (matrix Tense, Present Tense in the particularexample):70. Matrix Time Span: t..... ....t.. ......1t99o No\/What (69) says is that for every f in the Matrix Time Span there will be a "non-null distance" (i.e. a time span, specifically, the Embedded Time Span) betveen fand any event of the type of the na-clause. n this Way, the entire span will be freeof events of the type of the na-clause. This is how we get an empty time span and

a more direct assertion about it than in the pu- and srnce-constructions.How do rve get the Matrix Time Span to be empty of events of the relevanttype? f there v/ere an event of the relevant sort in the time span, let's say at t':.71. Matrix Time Span: t...... ........t' ........1t99o Event No/then t' would falsify (69) in that there wouldn't be a non-null time Span betveenan event of type e and it (t'). What (69) in effect achieves is that every subintervalof the matrix time span is "free" of events of type e, namely: "t (tXMTS) -$na(e) at t (where na(e) stands for event of the type of the na-clause).n contrast, the meaning of the Pu/since

7/28/2019 p.11

26/31

36 Sabine atrido8. ConclusionWe have seen that all three of the pu- and ra-constructions place RB, the Utter-ance Time if the Tense is Present, at a point that is at a distance from the closestevent of the relevant sort. But they do this in different viays. The since- and pu-constructions place the unique (or uniquely relevant) event at LB and therebyconvey that there is no other such event in the time span between LB and us.On the other hand, the na-construction makes a direct assertion about theevent-emptiness of the time Span. t does this by combining a variety of ingredi_ents: a free-choice description of an event, a matrix predicate that is an existen-tial assertion over time spans and that is, moreove a Universal Perfect.

Sabine atridouM..T.e -mai : iat ridou@ mit. e du

7/28/2019 p.11

27/31

A free-choice item hidden in verba morphoogy 37AppendixTsoulas (1,994) argues that na-clauses are indefinites (though not in the constructionthat we have been focusing on). Howeve even though am in a way adoptingTsoulas? insight, the arguments that he actually used do not argue for his conclusion.

Tsoulas notes that in French the Wh-island is much weaker when the embeddedclause is infinitival or subjunctive than when it is indicative.ndicative:1. *Que te demandes-tu [a qui Suzy a donnd]?what you you wonder [to rvho Suzy has given]2. *Que te demandes-tu [qui a dit qu' Alex a vu]?what you you wonder [who said that lex saw]nfinitive:3. qui te demandes-tu [quoi donner]?to whom you you wonder [vhat to give]4. Que te demandes-tu [a qui donner]?vhat you you wonder [to vho to give]5. Que te demandes-tu [qui a decide voir]?what you you wonder [who decided to see]Subjunctive:6. Que te demandes-tu [qui a voulu que Sophie voie]?What you you wonder [who rvanted that Sophie see.SUBJ]7. Que te demandes-tu [qui a exigd que Sophie ecrive]?what you you wonder [who required that Sophie write.SUBJ]Tsoulas talks about similar patterns in Greek. Greek has the expected Wh-is-land effect with indicative clauses but he claims that with na-clauses, the effect ismuch weaker:8. Ti - anarotiese [se pion na dosis]?vhat wonder.2sg [to who N give]9. Se pion anarotiese [ti na dosis]?to whom wonder.2sg [what N give]10. Ti anarotithikes [pios apfasise na di]?vhat wonder.2sg [who decided NA see]Tsoulas reminds the reader of extraction facts out of DPs: extraction out ofdefinites or specific indefinites is much ,worse than out of indefinites. This hasbeen noted for English, here are Tsoulas' French examples:

7/28/2019 p.11

28/31

38 Sabine latrdou11. De qui veux-tu voir une Photo?of who you want see a/one photo12. De qui veux-tu voir des photos?of rvho you want see some photos13. *De qui veux-tu voir une certaine photo?

of who you want see a certain Photo14. *De qui veux-tu voir la photo?of vho you want see the photo15. *De qui veux-tu voir ces photos?of vho you want see these photos16. ??De qui veux-tu voir la photo?of who you want see the PhotoTsoulas takes the position that the indicative has a definite feature (which canappear on C or on ), whereas the infinitive and subjunctive have an indefinitefeature. ence the extraction out of indicatives is bad but extraction out of sub-junctives is good. He does not address why indefiniteness is compatible rvith ex-traction while definiteness is not.one important problem with Tsoulas' account is the following. e talks abouta parallelism in extraction, but of course the parallelism, if there is one, is not be-tween definite Ds/indicative clauses on the one hand and indefinite DPs/subjunc-tive/infinitives on the other. n general, extraction out of indicatives is permitted,rhereas extraction out of definite DPs is not permitted. Extraction difficultieswith the indicative arise only once we combine indicative with an island-inducingenvironment like an embedded question. n other words, the definiteness of theindicative is not in itself sufficient to block extraction, an island is necessary. Thismeans that we cannot argue for a parallelism between indicatives and definites \based merely on extraction: definites always block extraction, indicatives block

extraction only if there is an island.Tsoulas' second argument is as follows: There are environments where the in-dicative, as a definite, causes Definiteness Effect violations, whereas the subjunc-tive and infinitive do not, rhich rgues, according to soulas, that they are in-definites: .17. faut que [Pierre partel *part]t is necessary that ierre leave.SUBJ/ *ND18. l faut trouver Sophie \t is necessary to find Sophie19. arrive que Sophie tarde tropl *vient vite]t happens that Sophie is late.SUBJ a|otl comes quickly.NDTsoulas does not provide any arguments to the effect that the ungramatical

7/28/2019 p.11

29/31

A free-choice item hidden in verba morphoogy 39expansions of the above sentences are ungrimmatica because of Definiteness Ef-fects violations. Moreove his claim cannot be that indicative clauses cannot becoindexed with expletives as there are plenty that can (though it's unclear t mervhat Tsoulas' proposal could say about this):

z0. l semble que Marie est malade't seems that Marie is sick'He discusses only the environments vhere "...a) [clausal constituents] alter-nate with DPs and b) the relevant factor governing the distribution of DPs is pre-cisely the Definite vs ndefinite distinction''. Ho/eve the relevant parallels withDPs he shows are only:

l arrive [plusieurs personnes]'There arrived many people'*l arrive SophieFirst of all, it is far from clear whether the uses of arriver in (19) and (21) arealike. s for faoir (17) _ (18), Tsoulas does not give examples with this verbtaking an NP complement. oweve faoir can take NP complements and thereis no problem with these being definite:23. me faut celun stylo' need this/one pen'As for Greek modal verbs that embed na-clauses, he gives only one exampleQtrepi 'must') but this does not take DP complements so v/e cannot test the par-allel with definite DPs. oweve if we look at the modal chriazome 'need', which

can take either na-clauses or NP complements, the parallel is again not support-ed, as the DP-complement can be definite:24. chnazete na figume/ * oti fevgume/ (tha) figumeneeds.3sg NA leave.lpl/ that leave.lpl/ (will) leave.lpl't iS necessary for us to leave'

21.22.

25. chriazome need.lsg ena/aftoto vivlioa/this the book' need a/this book'n other words, we cannot use Tsoulas' second argument for the parallelism un-der discussion; we ony have the parallelism in extraction facts, with the questionsthat arose earlier.To summarize: Tsoulas' discussion of subjunctive and indefinites and in partic-

7/28/2019 p.11

30/31

40 Sabine atridouular Greek na-clauses does not support his conclusion that na-clauses are indefi-nites and therefore we cannot rely on it for independent evidence that na-clausescan be indefinite descriptions of events.

REFERENCESlexiadou, A. & nagnostopouloU, E. 1999: "Raising without infinitives and thenature of agreement". Proceedngs of WCCFL 1'8' 14-26.Bake M. & Travis, L. |997: "Mood as verbal definiteness in a lan-guage". Natural Language Semantics 5, 2 13-269.Beghelli, F. 1998: "Mood and the interpretation of indefinites". The LinguisticReviev 15'277-3.Chomsky, N. 2000: "Minimalist inquiries: the framework". n D. Michaels, R.Martin & J. Uriagereka (eds.), Step by Step: Essays on Minimaist Syntax inHonor of Howard Lasnik. Cambridge, MA: MT Press, 89-156.Dayal, V. 1998: "Any as inherently moda|''. Linguistics and Phiosophyz|,433-476.Dowty, D. 1,979 Word Meaning and Montague Grammar. Dordrecht: Kuwer.Fauconnie G. 1974: "Polarity and the scale principle''. Chicago Linguistic Soci-ety t3,188-199.Giannakidou, . 1997: The Landscape of Poarity tems. Doctoral Dissertation.Groningen Dissertations in Linguistics 18. ;iGlasbey, S. t992: "Distinguishing between events and times: some evidence fromthe semantics of then''. Natura Language Semantics t'285-312.awkins, J. 1978: Definiteness and ndefiniteness: a Study in Reference andGrammaticaity Prediction London: Croom Helm.Heim, 1. t991,: "Artikel und Definitheit". n . von Stechow & D. Wunderlich

(eds.), Semantics. An nternationa Handbook of Contemporary Research.Berlin: Mouton de Grue 487-535.atridou, S. 1988: "on nominative case assignment and a few related things". MTWorking Papers in Linguistics |9, 75-|96.atridou, S. 2003: "A little bit more on the English Perfect". n . Alexiadou, M.Rathert & A. von Stechow (eds.), Perfect xporations. Berlin: Mouton deGruyte 133-151.atridou, S., nagnostopoulou, E. & zvorski, R. 1: "observations about theform and meaning of the perfect;'. n M. Kenstowicz (ed.), en Hae: a Life in \Language. Cambridge, M: MT ress, 189-238.atridou, S. & Embick, D. |997:. " propos prd'. Language 73,58-78.Jaeggli, o. & Safi . 1989: "The null subject parineter and parametric theory".n o. Jaeggli & . Safir (eds.), The Nu Subject Parameter. Dordrecht: Klure-44.

7/28/2019 p.11

31/31

A free-choice item hidden in verbat morphoogy 4|Kayne, R. 1993: "Towards a modular theory of auxiliary selection". Studia Lin-guistica 47 ,3-31.lein, W . 1992: "The present perfect plzz|e'' . Language 68' 525-552.Ladusaw, W. |977: "Some problems lvith tense in PTW''. TeaS Linguistic Forum6,89-102.Linebarge M. L987 "Negative polarity and grammatical representation''. Lin-guistics and Phiosophy 1o' 325 -387 .McCoard, R. 1978: The ngish Perfect: Tense Choice and Pragmatic nference.Amsterdam: Noh-olland ublishing Company.Melvold, J. t99L: "Factivity and definiteness". MIT Working Papers in Linguis-tics 15,97-117.Nathan, Lz}O/.lms.: "Temporal existentials and the amount Perfect".Partee, . t973: "Some transformational extensions of Montague Grammar''.ourna of Phiosophica Logic 2, 509-534.Philippaki-Warburton, . & Catsimali, G. 1999: "on control in Greek". n .Alexiadou, G. Horrocks & M. Stavrou (eds.), Studies in Greek Syntax. Dor-drecht: luwe 22] -24o.Portne P. 1995: "Quantification, events and gerunds''. n E. Bach, E. Jelinek, A.ratzer & B. artee (eds.), Quantification in Natural Language. Dordrecht:luwe 6L9-659.Terzi, A. 1993: PRo in Finite Causes: a Study of the nflectiona Heads of the akan Languages. Doctoral Disseation. CUNY.Tsoulas, G. 1994: "ndefinite clauses: some notes on the Syntax and semantics ofsubjunctives and indefinites". Proceedings of WCCFL 13, 515-530.Varlokosta, S. t994: /ssues in Modern Greek Sententia Compementation. Dc-toral Dissertation. University of Maryland, Colege Park.von Fintel, . 1999: "NP licensing, Strawson entailment and context dependen-cy''. Journa of Semantics |6,97-148.von Fintel, K. 2000: "Whatever". Proceedings from the Tenth Conference on Se-mantics and Linguistic Theory 6ALT 10),27-40.von Fintel, K. & atridou, S. 25 (to appear in Linguistic nquiry): "Anatomy ofa Modal".