+ All Categories
Home > Documents > The problem of NOT...UNTIL

The problem of NOT...UNTIL

Date post: 03-Dec-2023
Category:
Upload: kuleuven
View: 0 times
Download: 0 times
Share this document with a friend
48
Abstract The problem of not. ..unti/* RENAAT DECLERCK This artiele addresses the uestion why sentences like John didn't wake up until nine are grammatica while theirpositive counterparts are not: *John woke up unti1 nine. It ad uces evidence against the two existing analyses (the one claiming that not is a durative predicate, the other that there is a negative polarity item unt 1 that collocates with punctual predicates). The analysis that is offered heI rests on the claim that not unti1 is an instanee of integrated lexicalizati n, whereby two different semantic elements (roughly, the meaning of le~mporal at and that of an exclusive focuser like German erst or on1y) are 'fused into a single (but complex) lexical item. The author adduces a greai many pieces of evidence supporting this analysis and points out some intere:ting theoretical consequences. 1. Introduetion There have been a lot of rtic1es written about not...until since Fil1more (1963) introduced the s bject of "Neg-raising" into 1inguistics. The reason is that the use of ~not)...until has often been discussed as one of the key arguments in suPPlortofthe existence of Neg-raising as a syntactic transformation. As appeats from (la)-( lc), until can combine with dura- tive predicates but not .Jrith punctua1 (nondurative) ones (on a single- event reading), un1ess the sentence is negative: (1) a. John slept unti nine. b. *John woke up nti1 nine. c. John didn't wjke up unti1 nine. However, the combinatiof of wake up and until nine is sometimes a1so possib1e when the negation does not appear in the same c1ause but in a higher one: I don't think ohn woke up until nine. The argument in favor Linguistics 33 (1995), 51-98 0024- 3949/95/0033-0051 © Walter de Gruyter
Transcript

Abstract

The problem of not. ..unti/*

RENAAT DECLERCK

This artiele addresses the uestion why sentences like John didn't wake upuntil nine are grammatica while their positive counterparts are not: *Johnwoke up unti1 nine. It ad uces evidence against the two existing analyses(the one claiming that not is a durative predicate, the other that there is anegative polarity item unt 1 that collocates with punctual predicates). Theanalysis that is offered heI rests on the claim that not unti1 is an instaneeof integrated lexicalizati n, whereby two different semantic elements(roughly, the meaning of le~mporalat and that of an exclusive focuser likeGerman erst or on1y) are 'fused into a single (but complex) lexical item.The author adduces a greai many pieces of evidence supporting this analysisand points out some intere:ting theoretical consequences.

1. Introduetion

There have been a lot of rtic1eswritten about not ... until since Fil1more(1963) introduced the s bject of "Neg-raising" into 1inguistics. Thereason is that the use of ~not)... until has often been discussed as one ofthe key arguments in suPPlort ofthe existence of Neg-raising as a syntactictransformation. As appeats from (la)-( lc), until can combine with dura­tive predicates but not .Jrith punctua1 (nondurative) ones (on a single­event reading), un1ess the sentence is negative:

(1) a. John slept unti nine.b. *John woke up nti1 nine.c. John didn't wjke up unti1 nine.

However, the combinatiof of wake up and until nine is sometimes a1sopossib1e when the negation does not appear in the same c1ause but in ahigher one: I don't think ohn woke up until nine. The argument in favor

Linguistics 33 (1995), 51-98 0024- 3949/95/0033-0051© Walter de Gruyter

52 R. Declerck

of Neg-raising was that ins1tanceslike these could not be accounted forunless it was assumed that r,0t originated in the subclause and was thenraised into the head clause.

In this paper I will have a new look at not ... until, not because of itspossible relevanee to Neg-rJiSing (whose status as a syntactic transforma­tion is rather controversi ) but because it constitutes an interestingproblem in itself. Clearly, the fact that until, which basically impliesduration, can combine with punctual predicates in negative contexts callsfor an explanation. In the lr'ngUisticliterature two different explanationshave been suggested so far. One is that absence of a situation counts asa durative situation. That iSrnegation is claimed to function as a durativepredicate, so that any negative sentence is compatible with a durationadverbial. This analysis n9t only explains (la)-(1c) but also accountsfor sentences like the fOllOiing:(2) a. *The row will en for two hours.

b. The row will no end for two hours.(3) a. *The row win en as long as neither of them win give in.

b. The row win no, end as long as neither of them win give in.

The other analysis that hasl been proposed rests on the assumption thatthere are two different lexidal items until, viz. a durative one, which canonly combine with a dura~ive predicate, and a punctual one, which isused to locate punctual situations in time. The latter until is claimed tobe an NPI (negative POlari~&item). On this analysis *John woke up untilnine is ungrammatical bec se John woke up is compatible with neitherthe durative until (since wa e up is a punctual predicate) nor the punctualone (since the necessary ner.ative context is lacking). The fact that thispunctual until can only sh w up in negative contexts (and not e.g. inquestions: *Did John wake up until nine?) is handled by treating it as a"strict NPI," that is, one t at occurs only in well-defined contexts.In what follows I will rFfer to these two analyses as the "durative

analysis" and the "NPI analysis." The former was fust proposed byKlima (1964: 289) and lateridefended by Smith (1974), Heinämäki (1978:84), Mittwoch (1977), an~ many others. As a matter of fact, its basicclaim has become quite P9pular.l The idea that a "negative clause isalways durative" (Quirk etl al. 1985: 1081) has become something of acommonplace in the lingUtstic literature. At the same time the riyalanalysis has been very persistent too. The idea that there is a punctualNPI until is to be found not only in a lot of publications of the late 1960sand the 1970s, such as Lak10ff(1969: 141), Lindholm (1969), Stockwellet al. (1973: 253), Karttunef (1974), Seuren (1974a), Hom (1978a: 155,1978b: 174), etc., but alsa in more recent works such as Linebarger

The problem of not. ..until 53

(1981: 14), Edmondson (1981), McCawley (1988: 562), Horn (1989:313), and König (1991: 1 9). In what follows I will have a doser lookat the two views and point out that both of them are unsatisfactory. Oneof the purposes of this pa er is to offer an alternative analysis, in whichthere is only one kind of ntil and the special behavior of not ... until istraeed back to other factfS. This analysis will be proposed without aprior evaluation of the riv Ianalyses. After it has been expounded, I willgo into a great many data oncerning the not ... until construction, notingeach time which of the t ree analyses can accommodate the data inquestion and which cannot. In this way the respective merits and flawsof the three analyses will radually become dear.

2. The analysis

An adequate analysis of n t ... until not only has to account for the factthat until collocates with a: punctual predicate and that this use of untilis only possible if until is within the scope of a negator. It also needs toexplain how it is possible that, in spite of the negation, the situationexpressed in a sentence lik (Ic), John didn't wake up until nine, appearsto be asserted rather than denied. This is one of the basic facts aboutthe construction: (1c) is in erpreted as 'John only woke up at nine', thatis, 'John woke up as late anine'. The primary idea in this is 'John wokeup at nine'; the idea 'John didn't wake up before nine' is also present -it accounts for the 'as lat as' element - but it does not appear to bethe more salient of the two This peculiarity, which I will henceforth referto as the sense of actuali ation of (Ic), has often been noted in theliterature. For example, K rttunen (1974: 288) states that, unlike whatwe observe in sentences li e John was there until jive, the until phrase in(Ic) "does not mark the e dpoint of any interval. Instead it designatesthe time when the event in question took place."

Before offering an anal sis that accounts for the above data, it isnecessary to make dear Wh~Chterminological conventions will be adheredto. Sentences will be treat d as linguistic descriptions of situations. Theterm "situation" will thus be used as a cover term for actions, events,states, and the various oth r "things" that can be expressed in asentence.Situations will be said to lactualize." This is a cover term for notionslike 'happen', 'be performe " 'occur', 'hold', which cannot be felicitouslyapplied to every one of the various kinds of situation. Syntactically, mostsentences consist of a subj ct and a predicate constituent. The predicateconstituent comprises the Pand the optional adverbial adjuncts. TheVP comprises the V, the c' mplement NPs (functioning as direct object,

54 R. Dec/erck

indirect object, etc.), and he necessary adverbials (such as on the bed inI lay it on the bed). Sem ntically, such a sentence can be analyzed interms of a proposition in! olving a predicate, one or more arguments,and possibly adjuncts. The process oflexicalization resulting in a sentence(utterance) is the process f specifying lexical values for the variables inthe open proposition.The first step toward an adequate analysis of not ... until is the observa­

tion that lexicalization mat involve the use of negative lexical items. Oneclear example of this is t~e use of not many in sentences like Not manystudents werepresent: inste~adof selecting the single wordfew, the speakeruses a phrase that is the n gation of the antonym of few. The same typeof lexicalization leads to t e use of phrases like not infrequently (= often) ,not all (= some) , etc. Thi kind of negative lexicalization requires thatthe combination of not an the lexical item following it be conventional­ized (stereotyped, idiomat~c). We can speak of not many (of them), butnot of *not few (of them]; not al! is a stereotyped unit (e.g. Not all ofthem gat paid), but *not taCh is not (e.g. *Not each of them gat paid).(That it is idiomaticity tha is at stake here and not logical incompatibilityappears from the fact tha there are languages in which the forms corre­sponding to not many anti *not few are both acceptable: for example,German nicht viele vs. nidft wenige, Dutch niet veel vs. niet weinig. Thelatter form each time mans something like 'a considerable numberjamount of'.)When such a negative lèxical item functions as an argument NP other

than the subject, the nega ive partiele not is as a rule placed in the AUXconstituent rather than in the relevant lexical item itself: We say I didn'tsee many of them rather t an *1saw not many of them. The effect of thisis that the scope of the n gation extends to the whole VP, even thoughnot negates only the qua tifier and does not express sentential negation(cf. Payne 1985: 202; Horp 1989: 507).The hypothesis I wil! defend in conneetion with not ... until hinges on

the following claims:

A. The negation in sentfnces like (Ic), John didn't wake up until nine,is an instanee of negative lexicalization. That is, not until is used as astereotyped unit (similar tb not many, not infrequently, etc.) and not untily is an instance of negative specification of one of the variables in theopen proposition 'x woke up [at] y'.

B. The variabie specified!in this way is the time variable, That is, not ...until does not specify t le duration of the situation but the time ofactualization of the situat Oll.

The problem of not. ..until 55

c. Apart from this tem~oral meaning, not ... until also lexicalizes themeaning of the restrictive focusing adjunct only in the sense of 'as lateas', as in German erst, Du eh pas, etc. It is thus an instanee of integratedlexicalization, whereby twf' different aspects of meaning (one temporal,the other restrictive) are m pped onto a single (but complex) lexical item.This double aspect of maning leads us to interpret not ... until x assomething like 'only at x', 'as late as x'.

D. As noted by König (1~91: 55), "restrictive particles trigger a presup­position that corresponds to the relevant sentence in the scope of theparticle." This means that like the semantically equivalent sentence Johnonly woke up at nine, the s ntence John didn't wake up until nine presup­poses 'John woke up at ni e'.

E. Needless to say, it fo lows that John didn't wake up until nine alsopresupposes 'John woke p'. This analysis thus naturally accounts forthe "sense of actualizatio "referred to above.?

F. As noted by König (1 91), a focusing partiele has both a focus anda scope. In John only WOk{up at nine the focus of only is nine, whereasits scope is the sentence a a whole. The same is true if 'only at nine' islexicalized as not until nine This means that not has scope over the entireVP. It is in keeping withlthis that not is typically placed in the AUXconstituent and triggers in+rsion when the not until constituent is fronted:Not until nine did John wqke up. Klima (1964: 307) notes that when anegative phrase does not rave the VP in its scope, the negation is notplaced in the AUX consti uent (e.g. He had an accident not many yearsago) and fronting does no trigger inversion (e.g. Not many years ago hehad an accident).

On this hypothesis, w~ich I will henceforth refer to as the "dualanalysis" (because it reststn the claim that not ... untillexicalizes a dualaspect of meaning), (1c) is to be analyzed as 'John woke up [not untilnine]'. The temporal aspec of meaning conveyed by this is paraphrasableas 'John woke up at time t (t does not lie in the timespan indicated byuntil nine)'.

Since the analysis hinge on the claim that not until nine is semanticallyequivalent to only at nine it is necessary to have a closer look at thesemantics of only. The anrlysis of only that will be summarized here isthat which is propoundei in Declerck (1994), which is based on thetreatments offered by Löbfer (1989) and Vandeweghe (1992). Declerck(1994) argues that the restrictive focuser only realizes one of the two

56 R. Declerck

perspectives represented b Figures 1 and 2. The interpretation of onlyis based on a configurati n consisting of two scales and a particularperspective. A scale is an rdered set of values with a particular orienta­tion. In Figures 1 and 2 t~ two scales (and their orientation) are repre­sented by the two vertica arrows. For ease of reference the scales arelabelled "a-scale" and "b scale." The restrictive focuser only expressesan evaluation that is base on one of two perspectives. The first possi­bility, represented by Fi re 1, is that the speaker evaluates a value onthe b-scale from a point ,f view assumed on the a-scale. The value inquestion will be called th "actual value." It is evaluated in the sensethat it is contrasted with a set of "alternative values" that are higher upthe scale. The second po sible perspective, represented by Figure 2, isthat in which the point of [ieW is a value on the b-scale. The actual valueis an a-scale value, and it is contrasted with a set of alternative valuesthat rank below it on the a-scale. In both figures the relation betweenthe point of view and the ctual value is indicated by a horizontal arrow,

actual val e {~-'--.---~_-~-~~-~.---~-~~-~-~~?!~point ofview

alternativevalu s .--~-~

point of view

Figure 1. The a-scale perspectiv of only

Figure 2. The b-scale perspectiv of only

a b

»:} alternative values.,.,/,.,.~~

~~.--~~ actual value

ba

while the relation between he point of view and the alternative values isindicated by a dotted line.

It is shown in Declerck 1994) th at certain types of clause constituentscan only indicate values on ne ofthe two scales. For example, premodify­ing quantifiers are normall associated with the b-scale, but constituentswith a conditional meani! are always treated as a-scale values. Thus, inYou can buy it for only T 0 dollars the actual value (two) is contrastedwith a set of alternative ~rlues ranking above it (only two= 'not morethan two'), which means ~hat the values in question lie on the b-scale.By contrast, in You can on1r buy it for TWO dollars (whereJor two dollarsexpresses a necessary condrtion) the actual value is contrasted with a setof alternative values ranking below it (only for two dollars implies 'notfor less than two dollars') which means that the values in question lieon the a-scale.Important for our purpi se is that time adverbials are always treated

as a-scale values. In John nly woke up at NINE the a-scale is the timescale and the b-scale is th scale of events. Both scales are tempo rallyordered and their orientati n is such that earlier times/events are orderedbefore later ones. ("Highe up the scale" is therefore equivalent to "laterin time.") The evaluation ffected by only is as shown in Figure 3. Theinterpretation of John onl)! woke up at nine is then as follows. From aparticular point of view ori the event scale the speaker evaluates a stageon the time scale. The st~ge functioning as point of view is the stagedescribed by the open prOjOSition 'John woke up at t'. The stage on thetime scale that is evaluated (i.e. the actual value) is the value satisfyingthe unbound time variable in the open proposition, viz. nine. This actualvalue is contrasted with a set of alternative values that are lower downthe scale (i.e. earlier times).What I am arguing is t1at exactly the same analysis applies to John

t e

The problem oJnot ... until 57

9 { _---- John woke up at t.............•.,....~

earlier times --------

Figure 3. The eva/uation effecte by only in John only woke up at nine

58 R. Declerck

didn't wake up until nine. This sentence effects the same kind of evaluation,involving the same scalesl the same perspective, and the same actualvalue and alternative valuès, The only difference lies in the lexicalizationof the restrictive (exclusiv I) evaluation. The fact that not until nine quali­fies as an alternative to on y at nine follows from the fact that the a-scalevalues are times and that (within certain pragmatic restrictions) all thetimes preceding the actual value belong to the set of excluded alternativevalues. This means that the times functioning as alternative valuestogether form a zone (pha e) leading up to the time functioning as actualvalue. Since the expression until nine is an obvious label for this zone,and since the effect of the 6valuation is to exclude this zone of alternativevalues (i.e. to make clea~Jhat none of these values satisfies the variabiet in the proposition 'Jorry woke up at t'), not until nine is a perfectlylogicallexicalization for t,e evaluation represented by Figure 3.

It should be noted th~f this analysis (which, as noted above, I willrefer to as the "dual anal sis") does not invoke any principles that arenot required on indepen ent grounds. It is weIl known that negativelexicalization is possible. We have already referred to two typicalinstances: the type not i requently, in which there are two negatorsinvolved, and the type no manylallloften, in which the negative partielenegates the quantifier but as scope over the entire VP (unless the negativelexical item belongs to an djunct outside the VP, as in He died not manyyears ago). The analysis t at 1 propose for not ... until treats this complexlexical item as similar to t e latter type. The use of not manylalljoften tolexicalize a single variabie in the open proposition has a double effect onthe sentence: on the one ~and, the negation usually takes scope over theVP, so that the whole se~nnce is felt to be negative; on the other hand,there is also an implicatio of actualization: Not many people know thatimplies 'Some people kno that'. What we observe in (Ic) is quite similar.The negative way of SpeCfYingthe time variabie renders the entire sen­tence negative, but at the s me time it is understood that the time variabiedoes receive a value (viz. 'at nine'), which means that the situation isrepresented as actualizingThe dual analysis acco+nts for the three basic data referred to in the

opening sentence of this s ction. It explains the fact that (I c), John didn' twake up until nine, implies not only that John didn't wake up before ninebut also that he did wake up at nine: the analysis refers not only to theset of excluded alternativ values but also to the actual value satisfyingthe time variable. It also accounts for the fact that this use of until isonly possible if until is wit in the scope of a negator: the special interpreta­tion of the sentence (witf until nine referring to the zone of excludedalternative values) is crucially dependent on the presence of a negator,

The problem of not. ..until 59

since it is the negator tha triggers the exclusive interpretation, that is,excludes the alternative v lues. Finally, the dual analysis explains whyan inherently durative w rd like until can collocate with a punctualpredicate: until x does not indicate the duration of the situation, but thezone of excluded alternati evalues, that is, the timespan during whichthe situation does not act alize.The dual analysis also ~ccounts for another characteristic of not until

sentences: if the predicate ~sdurative, it receives an inchoative interpreta­tion. The sentence John didn': sleep until nine (on the relevant interpreta­tion) means that John beg n sleeping at nine. This follows from the factthat not until nine locates the situation at nine. As is weIl known, anysentence that tempo rally 1 cates a durative situation at a point of timeis interpreted inchoatively This is true of John only slept at nine, andhence also of John didn t sleep until nine, which receives the sameinterpretation."

Because it explicitly refers to the two aspects of meaning ('atx' +restrictive focusing), t re dual analysis accounts for the fact that not ...until nine corresponds wit a construction consisting of a time adverbialand a restrictive focusing partiele (interpreted as 'as late as') in manyother languages. We have already referred to the German constructionerst um neun Uhr and its utch equivalent pas om negen uur. French usesa construction with the di continuous focusing expression ne que: Jeanne partira qu'à cinq heure 'John will not leave until five'. Ne que is atwo-word focusing partiel that realizes one of the perspectives shown inFigures 1 and 2. The scal to which the actual value and the alternativevalues belong may be the ime scale, as in the above example, but it mayalso be a scale of a differe t kind, that is, ne... que can be used wheneverEnglish can use only, as i Jean n'a marqué que deux buts cette année'John has scored only togoals this year'. The fact that not untiltranslates as ne... que sup orts our claim that, like the latter, not untilshould be analyzed as aiscontinuous lexical item. This is not to say,however, that the parallel sm between the two is perfect. Because ne...que can be applied to ma y kinds of scales, the temporal interpretationof Jean ne partira qu'à inq heures comes exclusively from the timeadverbial à cinq heures. T e sole contribution of ne... que is to contrastthis time with earlier time , that is, to induce the 'as late as' element ofinterpretation. In English, he same semantic effect is achieved by differentmeans. Since until is inher ntly temporal, not ... until can only be appliedto the time scale. The un il constituent indicates both the actual valueand the zone of alternati evalues, and not makes clear that the latterare to be excluded.In sum, whereas ne... qu is similar to only in that it can be applied to

60 R Declerck

many different scales, not ..until is inherently temporal." The fact thatEnglish has developed th specifically temporal not ... until constructionis probably due to the fa t that the English equivalent of ne... que, viz.only, does not only apply to temporal scales: it has a much wider fieldof application. Apparentl English speakers have felt the need for afocuser with a more speci c meaning.Apart from not ... until, here is another candidate for being analyzed

as a discontinuous focuse I in English, viz. not ... butjexcept. This can beused in some of the cases in which French uses ne... que, but not whenthe reference is to a temporal scale. Thus, we can say They did not doanything butlexcept compl. in but not *John did not arrive butlexcept atjive 0'clock.

3. The data

In this section I will prese t a large number of data in conneetion withthe not ... until constructio . It will be shown that each of them is incom­patible with either the du ative analysis or the NPI analysis, or both,whereas the dual analysis an account for all of them.

3.1. Not. ..until as astere typed unit

The following observation make c1ear that not until x forms a stereotypedunit and may even occur s a single syntactic constituent.

3.1.1. Fronting.

(4) a. Not until (it w s) midnight did he leave.b. *Not at midnigh did he leave.c. *Not after (it w~ls)midnight did he leave.d. *Not while it wa dark did he leave.

The difference of gramm icality between these sentences is accountedfor by the dual analysis, aceording to which not ... until x is the discontinu­ous lexicalization of a conttituent of the type 'focuser +time indication'.As we have seen, this kl· d of lexicalization means that not ... until istreated as a stereotyped uit, that is, as an idiom. One consequence ofthis is that not until x ca surface as a single constituent, as long as itprecedes the AUX constit ent (otherwise not is inserted into the AUXconstituent). This is the ca e when not until x is fronted, as in (4a). Sincethere are na similar idilS 'nol al, 'nol after,etc., sentenceslik,

The problem of not. ..until 61

(4b )-( 4d) are ungrammatical. The difference between (4a) and (4b)-( 4d)is similar to that betweeb Not many people were killed and *Not fewpeople were killed, which ~~due to the fact that, unlike not many, not fewis not a stereotyped unit. 51

Since neither the durati1veanalysis nor the NPI analysis treats not untilas an idiom, neither of th m can explain the difference between (4a) and(4b )-( 4d).

3.1.2. Clefts. Another observation showing that not until nine canfunction as a single syntajtic constituent is that it can be the highlightedconstituent of a cleft:

(5) It was not until ninj that John slept.

This sentence yields two i~terpretations. One reading is 'Until nine is notthe correct specification f the duration of John's sleeping' (i.e. Johnslept until a time other th nnine). On this reading not is an instanee offocusing (speCificational)!negation:6 the sentence is used to reject thevalue nine for the variabl x in the open proposition 'John slept until x'.In doing so it calls up a set of alternatives, one of which is taken tosatisfy the variabie x: (5) implies 'John slept until some time other thannine o'dock'. The other i terpretation of (5) is 'The time that John fellasleep was: not until ninr". On this reading not belongs to the lexicalvalue that is specified for the time variable, The difference between thetwo readings appears fro various observations. First, the former inter­pretation implies a contra t between until nine and another until phrase,whereas the latter interp etation does not. Second, the latter readingremains possible when th predicate is nondurative (as in It was not untilnine that John got up), hile the former reading (in which until ninespecifies duration) is the no longer available. Third, only the formerreading survives when the egation is "raised": (6) can only be interpretedas 'I think that until nine 1snot a correct specification of the duration ofJohn's sleeping'.

(6) I don't think it was ntil nine that John slept.

It follows that sentences Jike (6) are ungrammatical if the original sen­tence does not involve a d rative predicate, because the reading on whichnot rejects a value for the uration varia bIe is then absent (since the ideaof a duration variabie das es with the punctual nature of the predicate):

(7) *1 don't think it was until nine that John woke up.

It is also in keeping w' h the analysis we are arguing that (5) is no

62 R. Declerck

longer ambiguous if we SUl bstitute another temporal preposition (e.g.before) for until:

(8) It was not before ninr that John slept.

This sentence can only b$interpreted as 'Before nine is not a correctspecification of the time en John feIl asleep'. There is no alternativereading 'The time that Joh feIl asleep was: not before nine' because notbefore is not stereotyped t e way not until is.? (At least, not before is notstereotyped in the sense 0 not until. It may occasionally be found as aunit, but then in the sens of after, as in The hooligans jinally left thestadium, but not befare the had smashed a number of windows.)The ambiguity of (5) is problem both for the durative analysis and

for the NPI analysis. Both can explain the reading on which until jive issaid not to be the correct s ecification of the duration of John's sleeping,but neither can account fo the reading on which not untiljive is specifiedas the correct value for t time when John began sleeping. The latterreading requires that not uTtil jive be treated as a single constituent. (Onthis reading we do not have the structure 'It was not êJ. that ... ' but thestructure 'It was êJ. that ...} where êJ. is not until jive. The negation musttherefore belong to the c' nstituent that is extracted from the soureesentence by the cleaving op ration. However, neither the durative analysisnor the NPI analysis treat not ... untiljive as a single constituent.)

3.1.3. Until outside the Jope ofnot. The dual analysis correctly pre­dicts that the sense of actullization arises only if until is within the scopeof not. The following sen{nce does not yield the reading 'It was ninebefore John woke up':

(9) Until nine, John did 't wake up.

The only possible reading I f (9) is 'Until nine it was the case that Johndidn't wake up'. Note that this sentence too suggests that John woke upat nine. However, this sJggestion is no more than a conversationalimplicature (due to the fac that the choice of the expression until nine isonly maximally informat,' e and relevant if John's not-waking-up cameto an end at nine). This i plicature can be cancelled by the context, forexample by the addition 0 I don't know about later, for Ileft at nine orIn fact, I heard later that e did not wake up af ter nine either? In (Ic),by contrast, the sense of ctualization cannot be cancelled, because itbelongs to the semantics o~the construction. Sequences like the followingare semantically anomalofs: John didn't wake up until nine. In fact, Iheard later that he didn't wake up at allo

The following is anoth r type of sentence in which until escapes fromthe scope of the negation

(10) He didn't run away, until the police came.

The special intonation, inlwhich the until c1ause is isolated from the headc1ause, means that the uhtil c1ause is not within the scope of not (cf.Lasnik 1976: 27). As pr Idicted, (10) does not assert actualization butonly implicates it. This i c1ear from examples like (11), in which theimplicature is cancelled:

The problem of not. ..until 63

(11) He didn't run aw y, until the police came and it was too lateto escape.

These observations are not accommodated by the durative analysis.Though the analysis cor ectly predicts the grammaticality of (9) and(10), since in both cases here is a negative context licensing the use ofuntil, it is deficient in t at it fails to explain why (9)-(10) lack theactualization reading that is typical of sentences like (1c). (The durativeanalysis assumes that untj is immediate1y outside the scope of the nega­tion not only in [9]-[lj] but also in [Ic] [cf. Mittwoch 1977: 410;McCawley 1988: 564]. T e fact that [Ic], but not [9]-[10], has the senseof actualization is therefo e an insurmountable problem.) The NPI analy­sis, by contrast, is conso ant with the facts, as it predicts that an untilthat is not within the sco e of negation, as in (9)-( 10), is not the NPIuntil that is used to locat punctual situations in time.

3.1.4. Tags with even. The dual analysis is in keeping with thefollowing set of data:

(12) a. John waited long time, even until five.b. *John woke u , even until five.c. John didn't ~ait, not even until five.d. *John didn't ake up, not even until five.

As appears from (12a)-(1 b), a tag like even untiljive can only be addedto a sentence expressing durative situation. It seems logical to assumethat the same is true of ags like not even until jive, which we add tonegative sentences. Since t e durative analysis treats any negative sentenceas expressing a durative si uation, this analysis predicts that not even untiljive can be added to any f.egative sentence. However, (12d) makes c1earthat this prediction is not[borne out. The fact that not even until jive canbe added to didn't wait ut not to didn't wake up means that only theformer counts as a durat ve predicate. This is in keeping with the dual

64 R. Dec/erckI

analysis and the NPI anal~iS, which hold that John didn't wake up untilnine locates a pnnctual sit ltion in time.

3.2. The negation in not... ntil

The following are a numbe of observations making clear that the not ofnot ... until negates only the ntil constituent and does not deny actualiza­tion of the situation. The e observations are in keeping with the dualanalysis, according to whi h the until constituent denotes the zone ofalternative values that ar excluded, while the existence of an actualvalue, and hence of actuali ation, is presupposed.

3.2.1. External negation. The dual analysis correctly predicts that the

negation. Bolinger (1977: 4 ) defines this kind ofnegation as follows (seealso Kroch 1974; Givón 1 78: 78-88; Hom 1985, 1989: 363):

In external negation the spe ker denies something that has supposedly beenaftirmed. One sign of this is t e use of some rather than any; another is a specialintonation, or punctuation, t I show that the thing is quoted:

[72] You ate some mushroo s - I did not 'eat some mushrooms'.

Bolinger also notes that etemal negation can be effected by the use ofna words:

(13) At least they foundjome of them. - They found none of them!(1977: 48)

However, the not of not un il cannot be used in this way. This is pointedout by Linebarger (1981: 3 ), who notes that sentences like the following"are unacceptable when s oken with a rising intonation on did not, i.e.when they are construed a denials":

(14) *We DID NOT get p untilI2:00.

In my opinion, there is a ouble reason for the unacceptability of suchsentences. One reason is t at extemal negation is wide-scope negationthat denies actualization, hereas the not of not ... until is negation insidea constituent that does n~t deny (but presupposes) actualization. Theother is that extemal nega ion is "echo negation": it echoes a statementalready made and denies i . For this to be possible, the statement that isechoed must be grammatical, and this is not true of *We gat up until12:00. The dual analysis tecognizes that the use of not ... until impliesactualizatiou and that the rITeSPonding positive constructiou is ungram-

matical (if the predicate is punctual). 11therefore predicts the ungram­maticality of (14).The ungrammaticality d( (14) runs counter to the durative analysis,

which assumes that nonftualization counts as a durative situation.Nonactualization is asserte in (14), but until cannot be used. (Note thaton this analysis the fact th t the rejected statement *Wegot up unti112:00is ungrammatical does not explain the ungrammaticality of [14]: unlikethe former assertion, [14] sserts nonactualization and should thereforebe compatible with until.) By contrast, the NPI analysis, which treatsuntil as a an NPI, is comp tible with the ungrammaticality of (14), sincein cases of metalinguistic egation the PPIs from the rejected sentenceare retained rather than eplaced by NPIs (cf. Bolinger's example).?Advocates of the NPI anall sis can therefore trace back the ungrammati­cality of (14) to the ungr aticality of the rejected sentence *We gotup until 12:00.

The problem of not. ..until 65

3.2.2. Irrealis. Further s pport for the view that not forms a unit withuntil in (Ic) can be derive from sentences like the following:

(15) *1wish that John h d woken up until nine.

This sentence is ungrammafical in spite of the fact that there is a negativeinterpretation due to the frct that had woken up expresses irrealis. Thisis predictabie from the dua analysis, since the negation implied by irrealisis not the kind of negatio that licenses the use of until with a punctualverb: the negation in ques ion must be the result of lexicalization. Thatirrealis does not produce his kind of negation is also clear from thefollowing:

(16) I wish that many of s had been there (but in fact none ofus was).

The negative proposition ~mpliedby this is not 'Not many of us werethere' but '11 is not the case that many of us were there'. In the sameway (15) does not imply '~hn didn't wake up until nine' but '*It is notthe case that John woke p until nine'. Like metalinguistic negation,negation resulting from irr· alis is wide-scope negation, paraphrasable as'11 is not the case that S . Sentence (15) is therefore ungrammaticalbecause the correspondin realis sentence S is ungrammatical: *Johnwoke up until nine.The above observations re confirmed by the following:

(17) a.b.

I wish he hadn' woken up before nine.I wish he hadn' woken up until nine.

Sentence (17a) implies 'It is not true that he did not wake up before

66 R. Dec/erck

nine'. Since both negation have the predicate in their scope, they canceleach other, so that we can drop both and end up with a grammaticalsentence: (It is true that) 1e woke up before nine. The same is not true of(17b). Since one negatiori (viz. the one following from irrealis) deniesactualization, whereas the Fther belongs to the not ... until constituent anddoes not deny actualizati' on, the two do not cancel each other. Thisexplains why 'It is not tr e that he didn't wake up until nine' cannotsurface as * (It is true tha ~ he woke up until nine.It is not quite clear ~ether the ungrammaticality of (15) is also

compatible with the durati e analysis. Sentence (15) implies nonactualiza­tion of the situation, and the durative analysis hinges on the claim thatnonactualization makes fpr a durative situation. 50 it seems as if theanalysis cannot explain thfungrammatiCality of (15). On the other hand,the idea of nonactualizati n is implied rather than asserted in (15), andit could be that the duratl ve analysis holds only for sentences assertingnonactualization. t

The ungrammaticality f (15) certainly poses a problem for the NPIanalysis, which assumes t e existence of an NPI until. As is well known,irrealis contexts license the use of NPIs, so the analysis would have toassume an ad hoc rule preventing the use of this NPI in irrealis contexts.(This solution is the mor9 problematic because negative contexts are thestrongest possible NPI trJ' gers [cf. Edmondson 1981].)

3.2.3. Focusing negation. The dual analysis is also in keeping with theobservation that the not I f not ... until in sentences like (1c), John didn'twake up until nine, cannot be interpreted as an instanee of focusing(specificational) negation with nine as the focus of not.lO That is, (Ic) isnot interpreted as rejecti g the value nine for a variabie x in an openproposition '!John woke p until x', and hence does not imply '!Johnwoke up until a time oth r than nine'.

This observation can p obably be accounted for by the durative analy­sis: the interpretation tha~ is not available presupposes actualization ofthe situation, whereas the ~urative analysis hinges on the idea of nonactu­alization. On the other hfnd, the NPI analysis, which treats until as anNPI, cannot account for t e data, since NPIs do occur in the presupposedpart of negative specific. tional sentences, even when actualization ispresupposed: I

(18) a. JOHN didn't 10 anything wrong. (It was Tim.)b. It was not JOElI N who did anything wrong (but Tim).

The NPI analysis would therefore have to assume some ad hoc rulepreventing the use of r NP! until from occurring ;0 negative-

The problern of not. ..until 67

presupposed.The dual analysis expf ins the data because it holds that what is

negated in (Ic) is not just nine but the entire until constituent.For elarity, it should ~e emphasized that the negation of the until

constituent does not have fhe same focusing (specificational) effect as insentences like It wasn't I John who did it or JOHN didn't do it.Specificational sentences li~e the latter are uttered for the specific purposeof rejecting a value for a p esupposed varia bie. This is not what happensin (Ic), nor in the corres onding eleft It was not until nine that Johnwoke up (cf. section 3.1.2 . In It wasn't John who did it the negationbelongs neither to the reje ted value nor to the presupposition: it deniesa relationship between the two. By contrast, in It was not until nine thatJohn woke up the negation belongs to the same constituent as until: not ...until is a discontinuous lel ical item. In spite of the fact that the untilconstituent is the focus of not in the sense that it is negated by not, thissentence is not an instanee f "focusing negation" in the sense of "specifi­cational negation."

However, this does not af.terthe fact that there is an element offocusingin the semantics of not ... fntil. As we have seen, one of the aspects ofmeaning of the constructio is that which in other languages is lexicalizedin the form of an exelusi e (restrictive) focusing partiele (e.g. Germanerst, Dutch pas, French n ... que). Because of this focusing effect, thereis no difference of meanin between John didn't wake up until nine andthe eleft It wasn't until nin that John woke up. The only difference is oneof emphasis. (As is well k own, a eleft typically has an emphasizing asweIl as a focusing functio .) The effect of the covert focuser is that theassignment of the value to t in the open proposition 'John woke up [at1t' calls up a set of alter atives (viz. earlier times), all of which areexeluded.As noted by König (199 : 120-121), the focusing partiele erst in John

wachte erst urn neun Uhr .uf 'John only woke up at nine' takes scopeover the entire open propo ition (i.e. wide scope). The same thing is trueof the covert focuser in Jo n didn't wake up until nine. This explains thepossible use of NPIs and f· ver~ion in such se~tences (e.g..Not un~ilninedid John wake up anybod ), smce an exclusive focuser IS negatrve byassertion (cf. Hom 1989: 9).

3.2.4. Focusing negation gain. In the preceding section we have seenthat the not of not ... until tS not an instanee of specificational negationfocusing the until phrase. We can now add to this that it cannot focusany other constituent eith r. Sentence (19a) is interpreted as (19b) (in

68 R. Declerck

which not forms part of t e presupposition attached to the variable: 'xsuch that x didn't wake u until nine'). It cannot be interpreted as (19c)because we cannot concei e a variable of the kind 'x such that x wokeup until nine'. l(19) a. JOHN didn't ake up until nine.

b. It was John wto didn't wake up until nine.c. *It was not Jo n who woke up until nine.

The above observation is also consonant with the durative analysis:the interpretation corresp~nding to (19c) presupposes actualization ofthe situation, whereas the urative analysis hinges on the idea of nonactu­alization. By contrast, the I PI analysis faces the same problem as referredto in section 3.2.3. It cann t explain why the use of the NPI until prevents(19a) from being interpret das (19c), since NPIs do occur in the presup­posed part of negative SP1eCificationalsentences, even when actualizationis presupposed (as in JO N didn't do anything wrong. It was Tim.).

3.2.5. The sense of actu lization. The dual analysis offers a naturalexplanation for what we ave called the "sense of actualization" that istypical of sentences like (~C): the predominant idea in John didn't wakeup until nine is 'John wo e up at nine'. That is, the until phrase "doesnot mark the endpoint 0 any interval. Instead it designates the timewhen the event in questio took place" (Karttunen 1974: 288). It is inkeeping with this that w~an find sentences like the following:

(20) He had threatened that he wouldn't open the door until eleven,and eleven it was 1hen he opened it for us.

On the dual analysis, this ense of actualization follows from the generalprinciple that "restrictive particles trigger a presupposition that corres­ponds to the relevant sentt nee in the scope of the particle" (König 1991:55): like John only woke u at nine, the sentence John didn't wake up untilnine presupposes the trut of John woke up at nine.

The NPI analysis atte pts to account for this sense of actualizationby assuming that there is speciallexical item until, which is a strict NPI(i.e. it can only be used in explicitly negative contexts) and which is usedto locate punctual situations in time. The durative analysis offers noexplanation at all and is ctually disproved by the very existence of thesense of actualization. Th sanalysis rests on the claim that negation is adurative predicate, which behaves like any other durative predicate inthat it can combine with a duration adverbial. As noted by Karttunen(1974: 290), this predicts hat there is no essential difference of meaningbetween (2Ia) and (2Ib):

The problem of not ... until 69

(21) a. Nancy didn't g t married until 1978.b. Nancy remaine a spinster until 1978.

According to the durativ analysis, both sentences describe a durativesituation that lasted until I978. The fact that both sentences suggest thatNancy got married in 197~must then be traeed back to the presence ofuntil, which refers to an 1dPoint and implies a reversal of the situationat that point. The pro bie with this analysis is that it does not recognizethat the idea of actualizatil n (i.e. the suggestion that Nancy got marriedin 1978) is stronger in (2 a) than in (21b). In (21b) actualization is nomore than conversationall implicated (cf. section 3.1.3 above); in (21a)it forms part of the assert on that is made. This becomes clear when wereplace until1978 by until she died (cf. Karttunen 1974: 290):

(22) a.b.

Nancy remain d a spinster until she died.!Nancy didn't let married until she died.

The fint sentence no longe suggests that Nancy got married: the implica­ture of actualization that arises from the use of until is cancelled by(until) she died. Sentence 22b), on the other hand, still has the sense ofactualization: it is seman ically anomalous because of this.!' We musttherefore conclude that t~e durative negation hypothesis does not offera satisfactory explanation ,f the sense of actualization conveyed in senten­ces like (Ic).

3.2.6. From ... until. Th dual analysis explains why the sense of actual­ization does not arise if w replace until by from ... until:

(23) ?John didn't wake p from eight until nine.

This sentence does not ex~ress actualization because there is no stereo­typed combination *not.. .from ... until that can be used in the same wayas not ... until to negatively specify a time variable. (In [23] the adverbialexpresses duration, not ti e.) Similarly, it is not possible to replace untilby up to or to in (Ic):

(24) a. *?John didn't ake up up to nine.b. *John didn't {ke up to rune.

These sentences cannot b used to express the actualization sense con­veyed by (Ic). If accePtab~l at aH, (24a) can only be interpreted in termsof the other reading of ( c) (viz. 'Until nine John didn't wake Up'),12whereas (24b) is simply u rammatical because to nine can only indicatethe duration of a situation that does actualize.Needless to say, the dUTive analysis is disproved by the ungrammati­

cality of (24a)-(24b), bOthlof which express nonactuaJization and should

70 R. Declerck

therefore count as sentenc s referring to a durative situation. The NPIanalysis can account for ( 4a)-(24b) on the grounds that up to and tocannot be used as NPIs t e way until can. However, this analysis willfind it difficult to explain he absence of the actualization sense in (23),where the speaker does us~ until. Proponents of the analysis will have toaccept an ad hoc rule bloc ing the use of the NPI until in the context ofa preceding from.

3.2.7. Adverbials preventi g the use ofnot. ..until. It is in keeping withthe dual analysis that the ctualization interpretation is ruled out by thepresence of a frequency a verbial:

(25) John didn't wake ulthe house once/twice until nine.

There may be different rea~ons why the actualization sense is not availablehere. If once is used, the f~ct that once is within the scope of not entailsthat actualization is denie[: John didn't wake up the house once assertsnonactualization, hence it is impossible to add a time-specifying adjunctlike at nine or not until nin . An additional problem is that the not of notonce cannot simultaneous y be the not of not until. Put differently, notcannot simultaneously ne~ate once and form part of the integral lexicalitem not ... until. The latter problem also arises when twice is used. Inaddition, the not of not tW1ice is an instanee of metalinguistic negation('not twice but x times'), which is not the kind of not that forms aconstituent with until (cf. eetion 3.2.1 above).

Adverbia1s like any mor or any langer similarly prevent the actualiza­tion reading:

(26) a.b.

He didn't sleep any more until morning.He did not lov her any longer until his death.

In sentences like these the negation applies directly to the situation itselfand not to the until con tituent. This foUows from the fact that thenegation is automatically ssociated with any morejany langer and is nolonger avai1able for association with the until constituent.The above observationsirun counter to the durative ana1ysis. Since He

didn't wake up oncelany jare implies nonactualization in the same wayas He didn't wake up does, the durative analysis (which claims thatnonactualization makes ~IIr a durative situation) wrongly predicts thatthe addition of an until ph ase to the former sentence wiUbe grammaticaland have the same seman ic effect as in the latter.The NPI analysis, by fontrast, is compatible with the observation

insofar as it also assumes hat sentences like He didn't wake up until nineassert actualization rather than nonactualization.

j The problem ofnot. ..until 71

3.2.8. Negative attractio again. In the preceding section we saw thata not that is associated with an adverb cannot also be associated with anuntil further down the sentence, We can now add that some adverbs"attract" (Jespersen 1924: 330) the negation even when they are placedafter the until constituent. Compare:(27) a. John had not oken up until nine.

b. *John had not woken up until nine yet.(28) a. No one had oken up until nine.

b. *No one had efer woken up until nine before.The reason why (27b) and (28b) are ungrammatical is that the negationcannot at the same time be associated with until and with yetiever ...before. If it is associated ith until, the sentences assert actualization andare therefore incompatib e with NPIs like yet or ever before. If thenegation is associated wit I the latter, the sentences express nonactualiza­tion. That is, the use of ot ... yetiever ... before denies actualization in aperiod leading up to som time of orientation: (27b) is interpreted as '!Ithad never happened yet t at John woke up until nine'. Not ... yetiever ...before thus effects senten ial negation of the kind that is paraphrasableas 'It is not the case that ',that is, the kind of negation that is similarto metalinguistic negatio and irrealis. As shown above, this is not thekind of negation that ca~ combine with until. It follows that a secondnegation is necessary to llcense the use of until: Never before had Johnnot woken up until nine.The above observation is in keeping with the dual analysis. It is also

compatible with the NPI nalysis in so far as the latter assumes that theNPI until can only be ins rted into negative sentences that do not denyactualization. The durati e analysis, however, cannot account for thedata, for it wrongly predi ts that (27b) and (28b) (which are as negativeas [27a] and [28a]) shoul be grammatical.

3.2.9. Double negation. Since the analysis assumes that not until speci­fles the time variabie of factual situation, it accounts for the fact thatnot until cannot be insert d into a sentence that already has its predicatenegated, that is, a senten e asserting nonactualization:(29) a. *Not until nin did it not happen.

b. *Not until nin did nobody know anything.The only exception allo ed by the analysis is the possibility that morethan one variabie receives a negative lexicalization in a sentence that stillyields an actualization re ding:(30) a. Not many of t em didn't leave until nine.

b. Not often did hey not leave until nine.

72 R. Declerck

The ungrammatieality 9f (29a) poses no problem for the durativeanalysis, whieh holds that luntil ean be added to any sentenee referringto a durative situation. In his analysis, (29a) is ungrammatieal beeausethe eorresponding sentenc without until nine (*It didn't not happen) isalso ungrammatical. Howe er, this explanation does not extend to (29b):sinee Nobody didn't know a ything is grammatical, (29b) should be gram­matical too. (Note that it cannot be the placement of not before untilthat renders [29b] ungra atieal: He didn't arrive until nine can bereworded as Not until nine did he arrive.)The NPI analysis can robably account for the data, since it also

assumes that not ... untilloc tes factual situations in time and is thereforeincompatible with sentence denying actualization.

3.2.10. Until nine or longel' Karttunen (1974: 287) notes the differencebetween the following sent nces:(31) a. The princess sl pt until nine or longer/more,

b. *The prineess dirn't wake up until nine or Jonger/more.(32) a. The prineess slfJt until nine and beyond.

b. *The princess dl' n't wake up until nine and beyond.Phrases like or langer, or m re, and beyond can be added to an indicationof the duration of a situati n. They express that the duration in questionmay be even longer than rs initially said. Since in (3lb) and (32b) notuntil indieates the time ofl the situation rather than its duration, it islogical that these additio I s are not allowed. The same observationexplains the differenee bet een the following:(33) a. The princess sl pt until nine, if not longer.

b. *The princess di n't wake up until nine, if not longer.(34) a. The princess sl pt longer than until nine.

b. *The princess di n't wake up longer than until nine.By contrast, the following s acceptable:(35) The princess didn't ake up until nine, if not later.Whereas langer refers to du ation, later refers to the time of actualization.

The above observations, whieh show that not ... until nine is an indiea­tion of time rather than d ration, are eompatible with the dual analysisand the NPI analysis but .sprove the durative analysis.

3.3. Restrictive focusing

The following observation have to do with the fact that part of themeaning of not ... until is th t of a restrictive focuser.

The problem of not ... until 73

3.3.1. Presumption of lateness. Astrong piece of evidence in supportof the dual analysis is th~t this analysis offers a natural explanation forthe "presumption of late ess" that forms part of the interpretation ofsentences such as (Ic):

It seems to me that the esse ce of the punctual until is lateness. In saying (28)[The princess didn't wake up ntil nine] I presume that the princess woke up andI let it be understood that t is could have (or perhaps should have) happenedearlier (Karttunen 1974: 292).

Since the durative anal sis fails to account for the actualization senseof (Ic), it also fails to explain the presumption of lateness. Karttunen'sNPI analysis does not re lly explain this either, though Karttunen sug­gests that lateness could bd traeed back to a special pragmatic presupposi­tion that he claims to be I art of the meaning of the NPI until.13 In thedual analysis, however, t e sense of lateness follows naturally from thefact that one of the two e ements of meaning lexicalized by not ... until isthat of a covert restrictiv focuser. As shown in Figure 3, the effect ofthis focuser is that the ac ual value (nine) satisfying the variabie tin theproposition 'John woke p at t' is contrasted with a set of alternativevalues ranking below nin on the time scale (i.e. earlier times). The factthat the speaker deerns ~t necessary to establish this explicit contrastbetween the time of actualization and alternative earlier times can onlymean that the speaker ha! expected the situation to actualize earlier (orhad at least reckoned withlthiS possibility), which is exactly what is meantby the "presumption of lteness."

3.3.2. Only. The dual a alysis also explains the difference in grammati­cality between the followi g sentences:

(36) a. The princess ~ept only until nine.b. *The princess iidn't wake u~ only until ~in~. .

Karttunen (1974: 288) potnts out that In (36a) "until nine designates theend point of an interval'! and only makes clear that this interval "inprinciple could have lasted longer." The fact that only cannot be used in(36b) "seems to indicate that there until 9 does not mark the endpointof any interval. Instead itJ(iesignates the time when the event in questiontook place." This obser ration is correct but it does not explain theungrammaticality of (36b {, as we have seen that, like German erst, onlycan be added to sentencef locating a punctual situation in time, as inThe princess only woke UAat nine. It is therefore wrong to assume thatonly can only restriet dur. tion. The reason why only cannot be used in(36b) is that its meaning is already part of the meaning of not ... until.

74 R. Declerck

Since the dual analysis is t e only analysis making this assumption, it isthe only analysis that can ccount for the data.

3.3.3. Until nine at the latest/earliest. The dual analysis is also inkeeping with the following set of data, which Karttunen (1974: 287-288)refers to as "puzzling":

(37) a. The princess sle t until nine at the latestj*earliest.b. The princess di n't wake up until nine at the earliestj*latest.

A speaker can use the phr ses at the earliest and at the latest when he isconcerned with locating a durative situation in time. When relating thesituation to the time scale, he can use these phrases to specify the loweror upper bound (i.e. the initial or terminal point) of the situation. Whenspecifying the lower boun on the scale he is concerned with the questionhow far he should go do n the scale to find the correct value; whenspecifying the upper boun he is concerned with the question how far hemust go up the scale. Sine earlier times are lower on the temporal scalethan later times, it is in eeping with this that the initial point (lowerbound) is compatible wit~ at the earliest but not with at the latest,whereas it is the other waYlaround for the terminal point (upper bound):

(38) a. John slept fro~ two o'clock at the earliestj*latest until teno'clock at the I testj*earliest.

b. She has been sl eping since nine at the earliest/tlatest.

The use of at the latest . (37a) thus follows from the fact that thespeaker is concerned wit~ the location of the terminal point of thedurative situation. _I_

Sentence (37b) differs frfm (37a) and (38a)-(38b) in that the speakeris not concerned with the iritial or terminal point of a durative situationbut rather with the point i~ time at which a punctual situation actualizes.As we have seen, not ... u1til represents this point as relatively late byestablishing a contrast be~een this point and earlier times. Now, para­doxical as it may seem, a ense of lateness is in keeping with the use ofat the earliest rather than ith that of at the latest: if a situation happenslater than expected (viz. at), it is natural to say that it happens at t 'atthe earliest' (or that it hap ens 'no earlier than' t).14 The rationale behindthis is that the times exclu ed by not ... until x are the times earlier than x.The NPI analysis is also compatible with the data, insofar as it recog­

nizes that (37b) is interpr ted as 'The princess woke up as late as nine'.(However, as we have see ,it does not offer an explanation for the senseof lateness.) The durative ,nalysis, by contrast, is again disproved by the[acts.This analysis treats rtb The princess slept and Th, princess didn't

The problem ofnot ... until 75

wake up as expressing sit ations that are durative and because of thatcompatible with an until phrase. It follows that this analysis cannotaccount for the different istribution of at the latest and at the earliestin (37a)-(37b).

3.3.4. Negation of not. .until. The dual analysis also provides ananswer to the following p zzle:

Our problem with the pune ual until eomes down to this: how ean we explainthat (28) [The princess didn' wake up until 9] eommits the speaker to the viewthat the prineess woke up I\. 9 (or shortly thereafter) while its negation [(I wishthat) the princess hadn't wok n up until 9] would eommit him to to the view thatshe woke up BEFORE 9? (Kar tunen 1974: 291).

The answer is that not unti nine not only specifiesthe time of actualizationbut also excludes earlier imes, that is, asserts nonactualization during

of wide-scope negation su h as effected by irrealis in the above example,the result is an interpret tion in terms of 'It is not the case that therewas no actualization in th period before nine', which is logically equiva­lent to 'There was actuall' ation before nine'. In other words, the effectof wide-scope negation i to deny that nine and no earlier time is thecorrect time of actualizati n. This denial boils down to an assertion thatthe actual time of actuali ation lies before nine. (It is in this way thatwe also interpret It is not true that he only woke up at nine and It is nottrue that he didn't wake Uiuntil nine.)The above observatio does not pose a problem for tbe durative

analysis either. On thi analysis (39a)-(39b) are equivalent to(39a')-(39b'), respectiVel~tAs appears from (39b'), sentence (39b) basthe required implication t at the princess woke up before nine.

(39) a. The princess d dn't wake up until nine.b. I wish the pri~cess hadn't woken up until nine.a'. The not-waki g-up of the princess lasted until nine.b'. I wish the not- aking-up ofthe princess had lasted until nine.

However, tbe above pro lem noted by Karttunen is not automaticallysolved by the NPI arialysis. Karttunen himself tries to account for thedata by assuming that the e is a special pragmatic presupposition attend­ing the use of the NPI un il (cf. note 13). For lack of space I will not gointo this argument here. uffice it to say that the NPI analysis requiressome ad hoc solution to t e above-mentioned problem, whereas tbe dualanalysis does not.

76 R. Declerck

3.3.5. Scales. It has bee, argued by Ladusaw (1980) and many othersin his wake that the cont9xts licensing the use of NPIs are those thatlicense "downward entai1~ents" on a scale. Such a sca1e is "a list ofpropositions formed by su stituting va1ues into a propositional schema(an open proposition) in a ertain order so that each proposition derivedentails the propositions pr~eiOUSlYderived" (Ladusaw 1980: 139; see alsoFauconnier 1976). An ex ression is downward-entailing if it creates asemantic context in which it is the propositions further down the scalethat are pragmatically ent iled. An upward-entailing context is one inwhich the inferences run p the scale. Quantifying expressions may beeither downard-entailing ~monotone decreasing) or upward-entailing(monotone increasing) or either (nonmonotonej." We can check thisby changing the VP in the scope of the quantifying expression in such away that its meaning chan~es from more general to more specific (i.e. byreplacing the VP by anott,er VP whose denotation is a subset of thedenotation of the original ~P). If the sentence with the more general VPentails the corresponding se tence with the more specificVP, the quantify­ing expression is downwar ' -entailing. If the entailment works the otherway around, the expression is upward-entailing. If neither sentence entailsthe other, the quantifyin expression is neither upward-entailing nordownward-entailing. For ~tample, Same of them ran does not entail butis entailed by Same of the"r ran quickly. Hence some of them is upward­entailing. By contrast, Nare of them ran entails but is not entailed byNone of them ran quickly. Hence none of them is downward-entailing.(This means that the addit on of a negation reverses the direction of theentailments.) Finally, Exa tly three of them ran does not entail and isnot entailed by Exactly thr e of them ran quickly. Hence exactly three ofthem is neither upward-ent iling nor downward-entailing.It should be noted that e addition of a time-specifying adjunct does

not affect these entailment . Thus, Same of them ran at jive 0'clock doesnot entail but is entailed by Same of them ran quickly at jive. The sentenceNone of them ran at jive e tails but is not entailed by None of them ranquickly at jive. And Exactl three of them ran at jive does not entail andis not entailed by Exactly hree of them ran quickly at jive.

On the other hand, the ddition of only may change the nature of theentailments. Whereas Som of them ran quickly entails Some of them ran(but not vice versa), Only ome of them ran quickly does not entail andis not entailed by Only so e of them ran.ï" This means that the additionof only renders the quanti ing expression nonmonotone.!? The same istrue in sentences where on focuses a time indication: John only ran atFIVE does not entail John only ran quickly at FIVE or vice versa. Sinceonly at jive is semantically quivalent to not until jive, we may expect the

The problem of not. ..until 77

same thing to be true of Jo n didn't run (quickly) untiljive. This predictionis actuaUy borne out: on terelevant interpretation, John didn't run untiljive does not pragmaticaUy entail John didn't run quickly untiljive or viceversa. (This is especiaUy cl ar when we consider the versions with front­ing: Not until jive did Joh run and Not until jive did John run quicklyobviously do not entail ea hother.)Once again, the above d ta support the dual analysis and disprove its

rivals. The durative analys s treats John didn't run (quickly) until jive assemanticaUy equivalent t Until jive John didn't run (quickly), thuswrongly predicting that bo hare downward-entailing (since negation hasa downward-entailing effe tand until jive does not affect this). In actualfact, only the latter senten eis downward-entailing: Untiljive John didn'trun entails Until jive John idn't run quickly. This does not accord withthe NPI analysis either, s NPIs do not normally have the effect ofcanceUing downward entai ments. (On the contrary, it is widely acceptednowadays that it is precis ly in downward-entailing contexts that NPIsare found.r'" The fact that 0 downward entailments are aUowed in Johndidn't run untiljive is due t the covert restrictive focuser, whose existenceis only recognized by the ual analysis.

3.4. NPI contexts other than negation

The dual analysis offers an explanation for the fact (noted e.g. byEdmondson 1981: 38) tha until (with the sense of actualization) occursin negative contexts, but ot in any other context licensing the use ofNPIs (such as questions, i clauses, etc.). Compare:

(40) a. Did he wake u at any time at al! during the night?b. *Did he wake u (at al!) until nine?

(41) a. You would be punished ifyou said anything to anybody.b. *You would be punished if you woke up until nine.

(42) a. John is the las) person to do anything about these problems.b. *John is the las person to wake up until nine.

The observation that until is excluded from aU contexts licensing NPIsexcept negative ones casts lUbt on the NPI analysis. Those who maintainthat it is an NPI have to . efine it as a "strict NPI" (Hom 1989: 313),that is, an NPI that is stri tly confined to a particular kind of context.Moreover, we have seen t at not any kind of negative context licensesthe use of until. The not of not ... until cannot come from irrealis (cf.section 3.2.2), and neither an it be an instance ofmetalinguistic negation(cf. section 3.2.1) or focusi g (specificationa1) negation (cf. section 3.2.3).

78 R. Declerck

Our until cannot be fOU~d either in negative sentences containing anadverbiallike any more, y t, before, etc. (cf. sections 3.2.7 and 3.2.8). Insum, it cannot be used in any case where the negation has the effect ofdenying actualization. Ne1edlessto say, this renders it very difficult tobelieve in the NPI status of until, for NPIs are as a rule incompatiblewith the assertion of ac lualization: they typically show up in thosecontexts where actualizati n is either denied or questioned or treated ashypothetical, doubtful, et .

3.5. Conclusion

In section 3 we have pres nted 20 observations that are in keeping withthe dual analysis. Many 0 these run counter to the alternative analyses:

A. The durative analysi does not account for the fact that not ... untilis treated as a stereotyped unit and that not until nine behaves as a singlesyntactic constituent when it is fronted (section 3.1.1) or is the highlightedconstituent of a cleft (3 1.2). It does not explain why the sense ofactualization is absent w~en until is not within the scope of not (3.1.3).The claim that John didi't wake up is a durative sentence runs counterto the observation that e cannot add the tag not even until jive to it(3.1.4). The durative anal sis is also disproved by the fact that until (onthe relevant actualization eading) cannot be found in negative sentenceswhere the negation is metalinguistic negation (3.2.1) or focusing negation(3.2.3-3.2.4) or sterns fr Im irrealis (3.2.2). It also fails to account forthe sense of actualization (3.2.5), for the fact that this reading of not ...until cannot be expressed by not .. .from...until (3.2.6), for the fact thatthe actualization reading is prevented by the presence of adverbialsattracting the negation (3.. 7-3.2.8), and for the fact that not ... until ninecannot be followed by or longer or ifnot longer (3.2.10). Moreover, thedurative analysis fails to xplain the sense of lateness conveyed by not .until (3.3.1), the impossiilility of inserting only in a sentence with not .until (3.3.2), and the diff rence between The princess slept until nine atthe latestf+earliest and Ihe princess didn't wake up until nine at theearliestf*latest (3.3.3). Furthermore, the durative analysis wrongly pre­dicts that John didn't ruj until jive is downward-entailing rather thannonmonotone (3.3.5).

B. The NPI analysis d es not account for the fact that not ... until istreated as a stereotyped nit and that not until nine behaves as a singlesyntactic constituent whe it is fronted (section 3.1.1) or is the highlighted

The problem of not. ..until 79

constituent of a c1eft (3.1. ). It also runs counter to the observation thatuntil (on the relevant actu lization reading) cannot be found in negativesentences where the negat~' nis focusing negation (3.2.3-3.2.4) or sternsfrom irrealis (3.2.2). Mor over, it fails to account for the fact that not .from ... until does not expr ss the sense of actualization typical of not .until (3.2.6), as well as for he impossibility of inserting only in a sentencewith not ... until (3.3.2). Th NPI analysis does not offer a natural explana­tion either for the sense 0 lateness conveyed by not ... until (3.3.1), norfor the fact that a denial f John didn't wake up until nine implies thatJohn woke up before nine 3.3.4). Furthermore, the NPI analysis wronglypredicts that John didn't r nuntiijive is downward-entailing rather thannonmonotone (3.3.5). Finally, the fact that until (on the relevant inter­pretation) occurs only in s me kinds of negative context and not in othercontexts generally licensi g the use of NPIs renders it hard to believethat until is an NPI (3.4).

4. Further remarks

4.1. (Non) actualization ithin a timespan

The dual analysis assume that not until nine specifies nine as value forthe time variable in the 0 en proposition and contrasts that value withearlier times. In doing so 't represents the situation both as actualizingat nine and as not actualizl ng in the timespan reaching up to nine. Now,the assumption that not .. until can indicate a timespan within which asituation is said not to ac ualize may not be self-evident, given the factthat we cannot use *John woke up until nine to express the idea 'Johnwoke up within the timesp n reaching up to nine'. It is therefore necessaryto prove the plausibility 0 this assumption.The best example of a preposition/conjunction that can indicate a

timespan within which a s tuation is located is since. Consider:

John has lived ere since he got married.John hasn't do te the washing-up since he got married.John has done the washing-up only three times since he gotmarried.John has done he washing-up since he got married.lassure you I AVE done the washing-up since I got married.Several times.

In these examples since m rks the beginning of a period that reaches upto the time of speech. The e are two possibilities as regards the relation

(43) a.b.c.

d.e.

80 R. Declerck

between this timespan and the situation(s) referred to. Either the speakerrefers to a single situation hat lasts throughout the timespan (as in [43aDor he says how many times a situation has actualized within it. We canspeak of a "universal" an~ an "existential" interpretation, respectively.The existential interpretaJon is the marked one: it requires an explicitindication of the number I f times the situation has actualized (which iszero in [43b1 and three ti es in [43c]). If there is no such indication, weautomatically select a univ1ersalreading (as in [43d], which is interpretedas referring to a habit co tinuing up to the time of speech), unless theuniversal reading is ruled 'ut by the context (as in [43eD.

Since the period referre to in (43a)-( 43e) reaches up to the time ofspeech, we can also use u til now to refer to it:

(44) a.b.c.d.

John has lived in London until now.John hasn't done the washing-up until now.Until now Jo has done the washing-up only three times.John has don the washing-up until now.*1 assure you I HAVE done the washing-up until now.Several times.

e.

Sentences (44a)-(44d) ar9 interpreted in the same way as (43a)-(43d).However, unlike (43e), (41e) is ungrammatical. This is an indication thatthe existential interpretati~n is even more strongly marked in the case ofuntil than in the case of 4ince. Still, it remains possible to get it, as in(44b)-( 44c), even when we replace now by another indication of time:

(45) a. Until yesterdaY[JOhn had done the washing-up three times.b. Until yesterday John had not done the washing-up (once).

It is c1ear, then, that until an indicate a timespan in which a situation issaid to actualize a specificlnumber of times, or once, or not at all. Now,if the situation does not actualize within an until period, this need notmean that it does not actualize at all. As a matter of fact, the use of untilimplicates that the situattn does actualize at the end of the period (cf.note 8). Thus, (45b) implicates that John did the washing-up yesterday.It follows that not ... until an be used as a negative indication of time inconneetion with a factual situation. This is exactly what is assumed bythe hypothesis that we h ve been arguing. It is also confirmed by thepossibility of sentences lik the following:

(46) a. When do you hink he will come back? - Not until he'sfinished that jo in Naples.

b. Pennsylvania's hew abortion law almost certainly will pro videa Supreme cor test, but probablynot until next ycar.

The problem of not. ..until 81

It is also interesting to ote that, because since refers to the beginningof a period and until to th end, either preposition can be used to expressnonactualization within ~ period: we can say either I haven't seen himuntil now or I haven't seenlhim since then. This means that not only not ...until but also not ... since eould in principle have become idiomatic as ameans of expressing act~alization at the end of a period. This is notactua11ythe case, howeve1' although not ... since seems to have developedjust a little in that direction. A possible sign of this is that a sentencebeginning with not since is more acceptable than one beginning with *notafterfbeforejwhile ... : sentences like Not since I was young have I tastedplum pudding are not ungrammatical and strongly suggest actualizationat the terminal point of thl'period. Particularly interesting in this contextare examples like the fo11,wing:

(47) But the next morni g, drying off after his shower, Philip stood onachair to examine,~.s torso in the mirror above the handbasin ....Forty was admitte y a rather advanced age to begin worrying onthis score, but it as only recently that he had acquired anystandards of comp rison. Not since he was at school, probably, hadhe taken a good 100 at another male organ until he came to Euphoria(Lodge 1975: 167- 68).

This sentence seems to be a contamination of Not since he was at schoolhad he taken a good look ... and He had not taken a good look ... until hecame to Euphoria. The fact that the not of not since here also qualifies asthe not of not ... until supports the view that the two are related construc­tions. In section 3.2.7 we saw that punctual until cannot cooccur withphrases like not ...yetlbefcs elany more because these involve a differentkind of not. The same thi gis apparently not true of the not of not since.

4.2. Other cases of integri ted lexicalization?

Apart from until and sine , there are a couple of other prepositions thatcan be used to denote a fespan within which a situation is said not toactualize. For example, if John will wake up at five o'clock, then a11ofthe fo11owingare true at tree:

(48) a. John wi11not ,ake up until five.b. John will not ,ake up for two hours.c. John will not wake up as long as it is not five.

At first sight, examples li e these would seem to support the view that

82 R. Declerck

not is a durative predicate. owever, there are other data that run counterto this (cf. Karttunen 1974 286-287):

(49) a.b.

(50) a.b.

(51) a.b.

The princess sl ,pt to 9 o'clock.*The princess di~n't wake up to 9 o'clock.How long did ~?e princess sleep?*How long did ~he princess not wake up?I washed the dishes while you were not in the kitchen.*1washed the d shes while you didn't wake up.

Karttunen (1974: 287) con bludes from this that "there is no clear factualsupport for the position th t negation is a durative predicate." However,he does not point out that the data also present a problem for his ownNPI analysis, which disres ards examples like (48b) and (48c). If notonly until but also Jor a9d as long as can combine with a punctualpredicate in negative sente~.ces, it is inconsistent to assume there to bean NPI until without maki g similar assumptions for Jor and as long as.19The problem we are fac d with is that of accounting for the fact that

Jor and as long as, but not l 0, while, and how long?, resembie until in thatthey can be added to negJtive punctual predicates in spite of the factthat they express duration. To solve this problem we must recognize thedifferent effects of the va ious durative phrases. In (48a) the not untilconstituent specifies the time of actualization. In none of the other

Isentences do we find anyth~ng similar, because the prepositions introduc-ing the relevant phrases do not form a stereotyped unit with not. In spiteof this, sentences (48b )-( 8c) are grammatical because Jor and as longas can be used to indicate the length of time during which a particularsituation is said not to actualize. The problem with (49b) is that tocannot be used in the samé way: as noted in section 3.2.6, it can only beused to indicate the term{nal point of a situation that does actualize.Sentence (50b) is ungrarru:patical for a similar reason: How long? can beused to inquire about the f.ength of a situation but not to inquire aboutthe length of a timespan w~hin which a situation does not actualize. Forthe latter purpose we use for how long? (as in For how long haven't youbeen to the dentist?).2oSentence (51b), finally, is ungrammatical becausewhile must be followed b~a durative predicate."As noted above, Jor a d as long as resembie until in that they can

indicate a period within hich a situation is said not to actualize. As amatter of fact, these three~xpreSSions are more or less in complementarydistribution. While until äenotes a period by referring to its terminalpoint,for (+NP) and as I ng as (+clause) denote a period directly.P Itfollows that not until x ha 'Pensimplies not as long as x does not happen.However, while the latter rase only expresses nonactualization through-

l The problem of not. ..until 83

out a period, the former lso expresses actualization at the end of theperiod.Like not since, not for ppears to be stereotyped to a certain degree.

It is possible to front the hrase as a whole (e.g. Not for three weeks hadhe had a decent meal) an to use it as the focus of a cleft (It wasn't forthree weeks that he had h d a decent meal). However, not for does notautomatically yield the se se of actualization that is typical of not until.

4.3. Negative contexts lic nsing thepunctual use ofuntil

Although the evidence i favor of the dual analysis is overwhelming,there is one observation t at may seem to run counter to it: this use ofuntil is possible not only fter a preceding not but also in other negativecontexts, even in contexts that are only implicitly negative. However, itcan be shown that not unt1behaves in every respect similarly to not manyand not all, whose status as an instanee of negative lexicalization of aquantifier is widely accep I d (cf. Payne 1985: 202, Hom 1989: 507).23Let us begin by listinJ the possibilities. Both until and many can

"attract" a negation of a y of the following types:

A. The standard examp es are those in which the negation is not. Inthat case the negation need not preeede untillmanylall directly: when thelatter follows the verb, n t is inserted into the AUX constituent.ê" (Asfar as many is concerned, this is the rule unless many forms part of anadjunct, as in He died no many years ago.)25The result is that not hasthe entire VP in its scope. (This accords with the fact that not until andnot manylall trigger inver ion when fronted.)

B. The negation that is associated with untillmanylall can come fromone of the argument NPs

(52) a. None of us has seen all/many ofthem. (='All of us have notseen all/many ! f them.)

b. None of them woke up until nine. (= 'All of them did notwake up until ine')

C. The negation may co e from a lexical item with incorporated nega­tion (e.g. impossible) or f om an "inherently negative" word (e.g.forbid,dissuade, too, be afraid). Thus, all of the following sentences yield aninterpretation in terms of 'not manyjall':

(53) a. They were to poor to buy rnany/all of the clothes, ('Notmanyjall of th clothes were bought by them.')

84 R. Declerck

b. It was impossib e for us to see manyjall of the birds. ('Notmanyjall of the irds could be seen by us.')

c. He forbade he I to invite manyjall of her friends. ('Notmanyjall of her friends were to be invited by her.')

d. They 1eft the a iaries without having seen manyjall of thebirds. ('Not ma yjall of the birds were seen by them.')

In exactly the same way, t e following sentences can all be interpretedin terms of 'not until', tha is, they all yie1d an actualization reading:

(54) a. They were too tred to wake up until after eleven.b. It was impOSSib] for us to leave until nine.c. He forbade her 0 leave until the police arrived (Smith 1974:

72).d. He slept throug the night without waking up until it was nine.

D. The negation attractJ by untillmanylall may come from a negativeimplicatum of another Phrrse:

(55) a. Not manyjall ol them left until nine. ('Some of them did notleave until nine. )

b. Not all of thel have many friends. ('Some of them do nothave many frie:rs.,)

Phrases like not al! or not rany piek out a subset from a set. When thesubset in question is the re~erent of an argument NP, this means that thepredicate applies to this subset and does not apply to the excluded subset.It follows that Not many trt implies 'Many did not leave' and Not al! ofthem left implies 'Some of them did not leave'. As appears from (55a),the negation from such an implicatum can be associated with until.The above explanation •ssumes that it is not the negation preceding

manylall in (55a) that is li11kedup with until. This is in keeping with ourabove observation (see sec ions 3.2.7 and 3.2.8) that the same negationcannot be attracted by rore than one quantifier at the same time.Consider:

(56) a. John didn't wÎe up once until nine.b. *John doesn't 0 ten wake up until nine.

These sentences do not haLe the sense of actualization typical of not ...until. (For this reason unf nine can be fronted without any semanticeffect (apart from the change of sentence topic).) The reason is that notonce and not aften do not produce a negative implicatum, because theyare not felt to piek out a s bset from a set.

The problem of not. ..until 85

The following is anothe~ example that can be accounted for in termsof a negative implicatum:

(57) Not all of the time ombs will explode until tomorrow.

Jackendoff (1971: 296) pOfts out that he finds this sentence acceptableon the reading 'The last ti e bomb will not explode until tomorrow'. Inour analysis this possibilit is due to the fact that not al! implies 'some ...not' and hence 'at least o~e ... not': (57) is interpreted in terms of thisnegative implicatum.

The idea of a negative i plieatum is not a novel one in the linguisticliterature. Baker (1970) a cl Linebarger (1981) have argued that the useof an NPI may be trigger~ by a negative proposition that is implied bythe sentenee in which it ap ears. Thus, I was surprised that anyone votedfor Mondale implies 'I ex ected that no one would vote for Mondale'.The use of anyone is rende ed possible by the negativity of this proposi­tion. What I have argued neonneetion with negative implieata is quitesimilar. (However, there a e many more implieitly negative contexts thattrigger the use of NPIs t an there are that allow the use of until incombination with a punct al predicate; witness the ungrammaticality of*I was surprised that John rJt until nine. I will suggest an explanation forthis below.) I

E. As has often been obsîrved, the negation that is assoeiated with untilmay eome from a higher c ause (though there are many restrictions andspeakers' acceptability jud ements may vary considerablyjr"

(58) a.b.

I don't suppose John will wake up until nine.It is not likely t at John will wake up until nine.I don't suppose it is likely that John will wake up until nine.c.

Sentences like these are tra itionally analyzed in terms of "Neg-raising."As a matter of fact, virtu lly all the literature on not ... until has beenproduced with a view to ither proving or disproving the existence ofNeg-raising as a syntactic ransformation. The grammaticality of exam­pIes like (58a)-(58c) has b en the key argument adduced by those whoinsist that there is a syntaq tic transformation of this kind. However, itshould be clear from the p eceding paragraphs that (58a)-( 58c) do notreally warrant this conclu .on. There is no evidence for the claim thatthe negation that renders th relevant use of until possible should originatein the same clause as the atter. AH that is required is that there is anegation in the preceding Fontext that ean be associated with until bythe relevant rule of interpreration. In many cases the negation in questionis implied rather than asserd; it comes from a negative implicatum. In

86 R. Declerck

sentences like (58a)-(58c) the negator is explicitly present, but not in thesame clause as until. This presents no problem because (58a)-(58c) areinterpreted in much the s me way as (59a)-(59c):

(59) a.b.

I suppose John will not wake up until nine.It is likely that John will not wake up until nine.I suppose it is l'kely that John will not wake up until nine.c.

As noted by Lakoff (19 0: 158) (who attributes the observation toBolinger), the only differe ce of interpretation between (58a)-(58c) and(59a)-( 59c) is that the rmer imply uncertainty on the part of thespeaker while the latter do not. (This follows from the fact that[58a]-[58c] present what he speaker thinks in a form suggesting that itis what he does not thin .) Since (58a)-( 58c) yield basically the sameinterpretation as (59a)-( 9c), not is associated with until without anyproblem.Now that we have list d the various negative contexts licensing the

punctual use of unül." w can address the problem referred to above inconneetion with negative implicata. In fact, it is a double problem:(a) Since the theory of negative implicata has been advanced byLinebarger (1981) to eXPfin the use of NPIs, can we invoke the sametheory while maintaining that until is not an NPI? (b) How can weaccount for the fact that ere are more negative implicata licensing theuse of NPIs than there ar licensing the punctual use of until?

The answer to both qUïstions lies in a specification of the conditionsunder which an implied ne ation can be associated with until. The overallcondition appears to be hat the sentence in which until is used mustyield a negative paraphra e in which until and not are clausemates anduntil is in the direct sco e of noti" For a full understanding of thiscondition it is necessary t go into the various elements.

A. The paraphrase must be a paraphrase of the sentence itself, not ofsome entailment or implic ture of this sentence. This means, among otherthings, that the predicate nd the argument NPs in the paraphrase mustbe the same as in the sente ce paraphrased. This condition is not satisfied,for example, in the case 0 Only John woke up. This sentence does implythe truth of The others idn't wake up, but the latter sentence is noparaphrase of the former: it is merely a paraphrase of an entailment ofthe former. For this reaso the grammaticality of The others didn't wakeup until nine does not war ant the use of *Only John woke up until nine.Similarly, although Have ou done it? implies 'Either you have done itor you have not done it', the not in this 'paraphrase' does not warrantthe use of until because th negative clause 'You have not done it' is not

The problem of not. ..until 87

about Ij you wake up, be uiet, which is paraphrasable as 'You do notwake up or you are quiet: the clause 'You do not wake up' spelIs outone of the alternatives licensed by Ij you wake up, but it is not aparaphrase of the latter, hich is not felt to be a negative clause andhence does not allow the ddition of until nine.)

B. The paraphrase must be such that not and unti! are clausemates.This condition explains th differencebetween the following:

(60) a.b.

I do not think that John woke up until nine.*1do not claim that John woke up until nine.

Whereas (60a) is conventi nally interpreted as 'I think that John didn'twake up until nine', (60b) snot interpretable as 'I claim that John didn'twake up until nine'. Unlik (60a), (60b) thus does not satisfy the require­ment that there be a para hrase in which not and unti! are clausemates.The same principle also e lains the differencebetween Nobody woke upunti! nine and *There was obody who woke up until nine.

C. The paraphrase must e such that unti! is in the immediate scope ofnot.29 This condition, whi h is in keeping with what we have observedin section 3.1.3 (viz. that t e actualization reading does not arise if untilprecedes not), explainswh ,unlike (61a), (61b) does not yield the actual­ization reading that is typ cal of not until:

(61) a. I don't think e will arrive until Friday. ('I think he will notarrive until Fr day.')

b. *I don't think hat until Friday he will arrive. ('I think thatuntil Friday h will not arrive.')

D. In sections 3.2.1 and 3.2.2 we have seen that the negation licensingthe relevant use of unti!m st not be wide-scopenegation denying actual­ization. This rules out bot metalinguistic (extemal) negation and nega­tion resulting from irrealis. Self-evidently,the same restrietion holds fornegative paraphrases in hich not and until are clausemates. Thus,although Ij he had woken 'P... implies 'He did not wake up', we cannotsay *Ij he had woken up u ti! nine... because the not of the implied 'Hedid not wake up' denies ctualization. Similarly, *It is not true that hewoke up until nine is not grammatical alternative to It is true that hedidn't wake up unti! nine.

E. The implied negatio~, which must not deny actualization at theterminal point of the unti period, must deny actualization before that

88 R. Declerck

terminal point. This restrietion ru1es out sentences whose VP is modifiedby a "near-negative" (restrictive) adverb such as hardly or scarcely,because the negation that is implied is no absolute negation: hardly]scarcely means 'almost nor'. Thus John hardly woke up until nine impliesthat John did wake up be~ore nine, but not often. Clearly, this is not thesame sense of actualizatio~ as the one we are investigating (which requiresthat there be no actualization at all until the time indicated).

F. The requirement that there be a negative paraphrase that is a para­phrase of the sentence itsel entails that it is not sufficient that the sentencehas a negative implicatu : the negative implicatum must actually be aparaphrase of the sentence. This means that the sentence must be felt tobe a negative sentence. It rollows that in those cases where the negativeimplicatum is due to the lse of a particu1a.r lexical item, the lexical itemin question must be feIt 0 be negative in meaning. If it involves nonegative morpheme, it m 'st belong to the well-known c1ass of "covertnegatives" (e.g. doubt, witl out, refuse, etc.). Thus, the predicate strangecannot trigger the punctu I use of until, though it does have a negativeimplicatum:

(62) a. It is strange t at John woke up.b. I had expecte that John would not have woken up.

(63) a. *It is strange tliat John woke up until nine.b. I had expectedlthat John would not have woken up until nine.

Although (62b) is a nega~ve implicatum of (62a), (63a) cannot be usedto express the meaning of (63b). The reason is that strange is not felt tobe a negative predicate. S ntence (62a) does not conceive John's wakingup as nonactualizing. By 5ontrast, a sentence like I doubt that John wokeup does conceive it this w1Y. (It follows from this difference that we cansay It is strange that som~one hasn't already woken up but not *1 doubtthat someone hasn't alreaJdiwoken up. In the former case the that c1ausehas retained the form of e positive expectation 'Someone has alreadywoken up', and metalin uistic not has been added to represent thisexpectation as false [cf. adusaw 1980: 174]. In the latter case therecannot be such a remnam of a positive proposition because the use ofdoubt entails that the e ectation is anyhow conceived in terms ofnonactualization.)

In sum, if not is not to be found in the same c1ause as until, it mustbe "in the air" as a result of the use of a lexical item or expression witha c1earlynegative meaning This condition rules out many predicates thatdo trigger the use of NPIs.3o Compare:

The problem ofnot. ..until 89

(64) a. I'm sorry that anything happened.b. *I'm sorry that he woke up until nine.

(65) a. He was stupid to do anything i11egal.b. *He was stupid to wake her up until nine.

(66) a. He was asham d to take any more money (Klima 1964: 314).b. *He was asham d to wake up until nine.

(67) a. He would die ~efore he lent her any money.b. ?He would die before he woke her up until nine.

On the other hand we havt (68):

(68) a. 1'11be damned i[ 1'11hire you until you shave off your beard.b. Why get married until you absolutely have to?

As noted by Hom (1989: ~48), these sentences are "conventiona11y usedto express [al negative profosition which licenses until," viz.:

(68) a'. I won't hire yot until you shave off your beard.b'. You shouldn't .et married until you absolutely have to.

Sentences like Til be da ned ij J'll hire you show a conventional usagethat is different from that f sentences like It is strange that John wokeup. The latter type of sent nee is used to express a negative evaluationof a situation (viz. John's aking up) that is not negative in itself and isnot represented as nonact1aliZing. The former type of sentence, by con­trast, is specifica11yused to epresent the subclause situation as nonactual­izing: Til be damned ij is simply a conventional means of expressingnonactualization. The fact hat J'll be damned ij J'll hire you is interpretedin terms of a single proposition ('I won't hire you') means that until canbe added without there b1ing any danger that it wi11be associated withthe positive proposition 'I' I hire you'.

Conventionalization is f crucial importance in this. An expressionlike I wil! die before ... is ~ry similar in interpretation to Til be damnedij ... but is apparently n t conventionalized to the same degree as ameans of expressing nonaètualization. The sentence ?TIl die before Tilhire you until you shave 01your beard has a low degree of acceptabilitybecause the until clause s lill tends to be associated with the positiveproposition '1'11hire you'. This means that Til die before ... implicatesnonactualization but is not conventionalized in this usage to the samedegree as Til be damned ijl...is. (A syntactic sign of this is that J'll diebefore ... can be fo11owedby any kind of subject NP, whereas Til bedamned ij ... can only be followed by I.)

G. It is clear, then, that t enegation licensing the use of punctual untilmust be part of the meani g of the sentence that is paraphrased. That

90 R. Declerck

is, it must be a negation tpat is either entailed or conventionally impli­cated, not one that is meret:' conversationally implicated. (This is logical:since the negation is a nerssary condition for the punctual use of until,it must invariably be pr sent, hence not dependent on a cancellableimplicature. )An example of a sente~~e that conversationally implicates a negative

proposition is I had inte~Fed to wake her up. This sentence implicatesthat I no longer have this intention, hence that the proposition 'I willnot wake her up' is now 1alid. This implicature can easily be cancelled:I had intended to wake her up, and by God, I wil!! Because of this, theimplicated negation does ot warrant the use of punctual unti!: *1 hadintended to wake her up u ti! nine.By contrast, a not that is conventionally implicated (i.e. part of the

conventional meaning of an expression) may warrant the use of unti!.TypicaJ instances of this a e the cases of Neg-raising.:" Though I do notthink they wil/ hire you ms y not mean exactly the same thing as I thinkthey wil/ not hire you, the former is conventionally accepted as a meansof expressing the same thi g as the latter, albeit it in a somewhat attenu­ated way (cf. Epstein 19 6: 160; Carlson 1983: 120-121; Hom 1989:322).32Hence the gramm ticality of I do not think they wil! hire you unti!you shave off your beard. As noted by Hom (1989: 347), this sentence"is not rendered acceptab~e by being co-derived with I think they wil! nothire you until you shave F,ff your beard, nor by sharing any semanticrepresentation with [the lltter sentence], but rather because it is conven­tionally used to implicate [the Jatter sentence]." In other words, in casesof Neg-raising "the main clause negative belongs to one clause - itsown - both syntacticall and semantically, while being conventionallyassociated with another" Hom 1989: 347).In the above paragraph~ I have investigated the various conditions for

the association of a negatI'on from the preceding context with until. I canconclude by pointing out that the conditions in question also hold forphrases like nothing but, .nobody except, not ... anyth ing but, etc., whichhave an exclusive meanin similar to that of only (cf. section 2):

(69) a. I doubt that h~ did anything but study.b. He predicted no problems except in L.A. (cf. 'He predicted

no rain until +dday.')c. It was impOSSifle to do anything but pray.d. They were too tired to do anything but pray.e. We forbade th m to go anywhere but in the garden.f. They sat there without doing anything except praying.g. I don't 'UPPol they will do anything but pray,

The problem of not. ..until 91

Phrases like not ... anythin but resembIe not ... until in that they are alsoinstances of discontinuous lexicalization. The rules we are assuming fornot ... until are also necessa y to account for the use of these phrases.

5. Implications

In the preceding sections e have reached a couple of interesting conclu­sions: (a) sentences with not ... until and a punctual verb provide noevidence that negation re ders a punctual predicate durative; (b) thereis no NPI until; (c) the dat in conneetion with not ... until do not supportthe claim that Neg-raising is a syntactic rule; (d) we must recognize thepossibility of integrated le icalization (which fuses two different elementsof meaning into a single, lbeit complex, lexical item). We can now addsome further conclusions r this list.

A. If it is correct that nrt ... until nine is a lexically specified unit, theobservation that not need not be overtly present but may be inferrablefrom the preceding conte t means that interpretive rules must alreadyoperate on the level wher lexicalization takes place. As pointed out byRay Jackendoff (personal communication), this may prove to be a nailin the coffin of theories ( ike GB) that hold that lexicalization happensat a particular level, suc as the syntactic level or the level of logicalform. What has been arg ed here calls for a very flexible theory, withrules of interpretation alr ady operative when lexicalization takes place.This may be an unfamilia idea, but not, I would argue, an implausibleone. The point is simply t at a speaker who has the idea 'not until nine'in mind when starting the lexicalization process does not need to eneodethis idea in the form of a explicit not ... until nine. He mayalso embedthe until phrase in a lingui tic context in which the idea of 'not' is implied,knowing that the hearer ill associate this inferred negation with theuntil phrase, especially :tn the predicate is such that it is compatiblewith 'not until' but inco_Tatible with 'until' on its own.

B. In section 3.3.5 we hIe seen that not ... until differs from until in thatit cancels the upward or d· wnward entailments of the sentences to whichit is added. For example,

(70) a. John ran.b. Until/before fi e, John ran.c. John didn't ru .d. Until/before fi e, John didn't run.e. John didn't ru until five.

92 R. Declerck

Sentence (70a) is upward- ntailing: it entails John moved (with a moregeneral VP), but not Joh ran quickly (with a more specificVP). Theaddition of untiljbefore jiv in (70b) does not aIter this. By contrast, theaddition of not in (70c) r verses the direction of the entailments: (70c)entails John didn't run qui kly, but not John didn't move. This sentenceis therefore downward-ent iling. The addition of untiljbefore jive in (70d)again does not change thi . However, if we add not ... until to (70a), theresuIt is neither upward-e tailing nor downward-entailing: (70e) entailsneither John didn't move u til jive (for [70e] is true if John walked untilfive and then started runn ng) nor John didn't run quickly until jive (for[70e], but not the latter se tence, is true if John started running slowlyat fivej."These observations are n t without theoretical significance.They reveal

a use of not (with scope over the entire VP) that is not downward­entailing and hence does Jot reverse the direction of the entailments onthe scale. This runs count r to the widespread view that negation thatdoes not have narrow sco e must have this effect.

C. The above Observati~n also runs counter to the widespread viewthat (apart from interroga ive contexts) NPIs occur exclusivelyin down­ward-entailing contexts (cf Ladusaw 1980;König 1991:106). A sentencelike Not until nine did Joh see anybody is not downward-entailing, butthe use of not until nine d es license the use of the NPI anybody. Thismeans that NPIs are lice sed by any kind of negation that has scopeover the VP, irrespective 0 the entailments"

Katholieke Universiteit LeuvenReceived 17January 1994Revised version received3 September 1994

Notes

* Correspondence address: U iversitaire Campus, B-8500 Kortrijk, Belgium.1. Prince (1976: 405) proposes a variant of this theory: "untilmust operate on (co-occur

with) imperfective clauses" nd negation is one of the elements that can "imperfectiv­ize" asentence.

2. The claim that actualization is presupposed accords with the observation that a denialof John didn't wake up until nine still implies that John woke up, but at some earliertime (cf. section 3.3.4 below)

3. As pointed out by Karttune (1974: 289), there are verbs (e.g. last, remain) that cannotreceive an inchoative interpr tation because of their presuppositional properties. Suchverbs are compatible neithe with not until nine (on the relevant interpretation) norwith at nine:

The problem of not. ..until 93

(i) The party did not la! until nine (no inchoative reading).(ii) *The part lastedjdid ot last at nine.

4. König's (1991: 168) tentati e suggestion that not ... until can be considered as a discon­tinuous focus partiele "jUlt like ne... que in French" is therefore not quite correct:focusing is only one of the wo aspects of meaning conveyed by not ... until.

5. In section 4.1 below it wil! be pointed out that sentences beginning with not since arerelatively acceptable and al 0 strongly suggest actualization at the terminal point of theperiod referred to. This me~ns that, like not until, not since is also becoming idiomat­icized. Even not while seems to be developing in that direction, witness the followingexample cited by König (1~88):

(i) Not while I've got mytits about me, and remain officially sane, do I consult anypsychiatrist.

It is no coincidence that sin e and while, but not at, after, before, etc., should show thisresemblance to until. The ~repOSitions until, since, and while are the only ones thatimply (non)actualization th oughout a period.

6. "Specificational sentence" i the term used by, for example, Akmajian (1979), Higgins(1976), Declerck (1988), a d Keizer (1990) to denote the type of sentence in which avalue is either specified oril ejected for a presupposed variabie. For example, both ltwas John who ran away an JOHN ran away specify the value 'John' for the variabie'the x who ran away' (i.e. t evariable in the open proposition 'x ran away'). In bothcases the value that is spec~ed is focused, either intonationally or structurally (throughthe use of a cleft, which is a~onstruction that is typically focusing and specificational).In negative sentences like lt was not John who ran away and JOHN didn't run away (onthe same interpretation), th value 'John' is rejected as the value satisfying the presup­posed variabie. Since John s again focused, the negation in such a negative specifica­tional sentence is often callel "focusing negation."

7. That it is not until that is ste eotyped rather than not before seems to be the result of anaccidental historical develo ment. In Dutch, which is probably the cognate languagethat is closest to English, it is niet voor(dat) 'not before' that is stereotyped and thatcan be used in the same w y as English not until. In Dutch there is no stereotypedcombination *niet tot( dat) 'not until'. It is also worth noting that this use of nietvoor(dat) is easily recogniz bie because niet always precedes voor(dat), whereas thenegation otherwise precedes the final verb form. Compare:

(i) Hij is niet voor negen rur opgestaan.(ii) Hij is voor negen uur iet opgestaan.

The former is equivalent t English He did not get up until nine, whereas the lattermeans 'It is not the case tha he got up before nine'.

8. This accords with Heinämä i's (1978: 82) claim that whether or not the main clausesituation ceases to be true t the time indicated by the until phrasejclause "is not amatter ofthe logical structu e of until." See also Brée (1985: 28), who argues that thereis often, but not always, "a bonversational implicature that the main proposition is nolonger true after the time in icated by the until-clause."

9. Hom (1989: 397) points ou that the fact that metalinguistic negation does not triggerthe use of NPIs is one of the diagnostic tests one can use to recognize this kindof negation.

10. For a definition of "specific3 ional negation" (focusing negation), see note 6.11. As noted by Karttunen (197 : 290) and Hu (1988: 174), it is this sense of actualization

94 R. Declerck

20.

that distinguishes between ot ... until and not ... before: there is nothing wrong withNancy didn't get married befi re she died. Similarly,

(i) They did not give in btfo.re/*until they ran out of ammunition, and in fact theynever did.

12. Apart from the reading 'Joh only woke up as late as nine', (Ic) also has the reading'John's not-waking-up laster until nine'. However, unless until nine is fronted, thisreading is extremely unlikely to be either intended by the speaker or selected by thehearer. This must be duet ti some kind of "garden path" rule stipulating that theidiomatic reading must be co. structed first.

13. Karttunen assumes that not ..until t is semantically equivalent to. not ... before t but,unlike the latter, has a prag atic presupposition that is responsible for the sense ofactualization at t. The presufPosition in question is defined as follows: "A sentence ofthe form 'NOT (A until TI' where until corresponds to BEFORE in logical formpragmatically presupposes ' BEFORE T or A WHEN T'" (1974: 292). The logicalform 'NOT (A until T)' a d this pragmatic presupposition 'A BEFORE T or AWHEN T' function like the remises of a disjunctive syllogism whose conslusion is 'AWHENT'.

14. This is confirmed by the use of earlier in sentences such as John will come as late as 10p.m., or at any rate not earlier/*later than 9.30.

15. The terms monotone increafing, monotone decreasing, and nonmonotone are due toBarwise and Cooper (1981: 184ff).

16. That is, Only some of them:fan quickly may be true even if all of them ran (providedthe others did not run quic y) and Only some ofthem ranmay be true even ifnone ofthem ran quickly.

17. Hom (1989: 248-249) notes hat some people consider only to.be monotone decreasingrather than nonmonotone.1[llls does not appear to be correct for the kind of examplesunder discussion here. How~ver, nothing crucial depends on this, as the point I wish tomake is that not ... until nin has the same effect as only nine. Since not ... until ninecannot be added to negati le sentences, only positive sentences are relevant for ourpurpose. And the data regarding these are dear: if these sentences have upward entail-ments, the addition of eithertnot ... until or only will cancel them.

18. It is not surprising, therefor , that Ladusaw (1980: 145) points out that his theory ofdownward-entailing context does not account for the "NPI" until.

19. Hom (1989: 313) treats not bnly until but also for as a "strict NP!."This is also c1earfrom the fO~IOWing:

(i) He hasn't been here 19n9 (=the situation ofhis being here started not long ago).(ii) He hasn't been here lor a long time (= the period within which he hasn't been

here is quite long).

Mittwoch (1977: 413) rejec s the relevanee of examples like (5Ia)-(51b), saying that"since while-c1ausesinvo.lV~ stretch of time they are often classed with durationals; infact they are a special sub-cl ss of time adverbials." However, this remark is beside thepoint. The claim made in co neetion with (51a)-(51b) is that the predicate of a while­c1ausemust be durative, an that didn't wake up apparently does no.t qualify as such apredicate. The (correct) re ark that a while-c1ausefunctions as a temporal adverbialrather than as a durational ne is immaterial to this: it does not alter the fact that whilemust be followed by a durat ve predicate.For this reason we do not n rmally use until before a time indication like in two weeks'

21.

22.

The problem of not. ..until 95

25.

time, which measures the le gth of the period. Instead of *Iwill not see him until in twoweeks' time we normally sa I will not see himfor two weeks.As noted by Hom (1989: ~54), the idea 'all not' can lexicalize as no, but there is noone-word lexicalization of' ot all'.Jespersen (1924: 330) notes that there is astrong preferenee for negation that surfacesas not to be placed in the UX constituent: "The general tendency is to use a nexalnegative, even in some case where a special negative would be more apposite": insteadof We are here not to talk nonsense but to act (in which not X but Y forms a singleconstituent ~ cf. Hom 19 9: 403) we tend to use We aren't here to talk nonsense butto act.In drawing a parallel betwen not many and not until I will disregard this use of notmany in adjuncts.Hom (1978b) offers an ex ellent analysis, in which some of the restrictions receive anatural explanation.For ease of reference I wi I henceforth refer to the use of until in not ... until as the"punctual use" of until.Note that the same restricti n does not hold for NPIs. Compare:

(i) Tom didn't say tha he would lift a finger to help us (McCawley 1988: 575).(ii) The director didn' say that your performance was all that bad (McCawley

1988: 575).(iii) *1 didn't say that ill would arrive until 10:00 tomorrow (McCawley 1988:

576).29. McCawley (1988: 560) st tes that "sentences with not many or not al! allow only

interpretations in which th many or al! is in the immediate scope of not, i.e. nothingmay intervene between the two in logical structure." There is a similar restrietion onnot ... until and also on focu ing partic1es like even, only, etc.

30. This condition also rules 0 t sentences that are not negative, except "trivially" as theresult of a double negation cf. Ladusaw 1980: 135): although John woke up is logicallyequivalent to *John didn't nrt wake up, it is not feIt to be a negative sentence, hence theimpossibility of adding untI' nine.

31. Hom (1989: 349) distingui hes between conventional implicatures and conversationalimplicatures that are conve tionalized via a "short-circuited implicature." For simplic­ity, 1williump the two tog ther.

32. Evidence ofthis conventio~lization is that some verbs (e.g. suppose) allow Neg-raisingwhile other, semantically si ilar ones (e.g. surmise) do not, and that the sets of verbsthat allow (or sometimes re uire) it are slightly different in different languages.

33. Note that (70e) entails Joh didn't run beforefive and that this sentence is downward­entailing, like (70d). This confirms our conclusions that not ... until entails not ... beforebut is not semantically equiralent to it.

34. Of course, it is hard to decide whether the NPI is licensed by the presence of not or bythe restrictive (exclusive) focuser that is covert in not ... until. The two are inextricablybound up with each othe . As a matter of fact, it is presumably because "only isnegative by assertion" (H m 1989: 249) that the lexicalization of 'only at nine' cantake the form ofnot ... until ine.

23.

24.

26.

27.

28.

References

Akmajian, A. (1979). Aspects 0 the Grammar of Focus in English. New York: Garland.Baker, C. L. (1970). Double neg tives. Linguistic lnquiry I, 169-186.

96 R. Declerck I

Barwise, J.; and Cooper, R. ( 981). Generalized quantifiers and natural language.Linguistics and Philosophy 4, 15 -219.

Beach, W. A.; Fox, S. E.; and P ilosoph, S. (eds.) (1977). Papersfrom the 13th RegionalMeeting of the Chicago Linguist c Society. Chicago: Chicago Linguistic Society.

Binnick, R. 1.;Davison, A.; Gree1G. M.; and Morgan, J. L. (eds.) (1969). Papersfrom the5th Regional Meeting of the hicago Linguistic Society. Chicago: Chicago LinguisticSociety.

Bolinger, D. (1977). Meaning and orm. London: Longman.Brée, D. S. (1985). The durativ temporal subordinating conjunctions since and until.Journalof Semantics 4, 1-46.

Carlson, L. (1983). Dialogue Gam s. Dordrecht: Reidel.Cole, P. (ed.) (1978). Syntax and S mantics, vol. 14:Pragmatics. NewYork: Academie Press.Declerck, R. (1988). Studies on C ular Sentences. Clefts and Pseudo-clefts. Leuven: Leuven

University Press; Dordrecht: Fo is.-(1994). The onlylalready pu le: a question of perspective. Cognitive Linguistics 5,307-350.

Edmondson, J. A. (1981). Affect vity and gradient scope. Papers from the 17th RegionalMeeting ofthe Chicago Linguistic Society, R. A. Hendrick et al. (eds.). Chicago: ChicagoLinguistic Society.

Epstein, S. (1976). Investigation in pragmatic theory. Unpublished Ph.D. dissertation,University of Califomia, San Djego.

Fauconnier, G. (1976). Etude de c rtains aspects logiques et grammaticaux de la quantifica­tion et de l'anaphore en françai et en anglais. Doctorat d'Etat, Université de Paris VII.

Fillmore, C. (1963). The positio of embedding transformations in a grammar. Word 19,208-231.

Fodor, J.; and Katz, J. (eds.) (196 ). The Structure of Language: Readings in the Philosophyof Language. Englewood Cliffs, J: Prentice Hall.

Givón, T. (1978). Negation in I nguage: pragmatics, function, ontology. In Pragmatics,P. Cole (ed.). New York: Acade ic Press.

Greenberg, J. (ed.) (1978). Uni ersals of Human Language, vol. 4: Syntax. Stanford:Stanford University Press.

Heinämäki, O. (1978). Semantics of English Temporal Connectives. Bloomington: IndianaUniversity Linguistics Club.

Hendrick, R. A.; Masek, C. S.; and Miller, M. F. (eds.) (1981). Papers from the 17thRegional Meeting of the Chicag Linguistic Society. Chicago: Chicago Linguistic Society.

Higgins, F. R. (1976). The Pse do-cleft Construction in English. Bloomington: IndianaUniversity Linguistics Club.

Hom, L. (1978a). Some aspects of negation. In Syntax, J. Greenberg (ed.). Stanford:Stanford University Press.

-(1978b). Remarks on neg-rais g. In Pragmatics, P. Cole (ed.). New York: AcademiePress.

-( 1985). Metalinguistic negation and pragmatic ambiguity. Language 61, 121-174.-( 1989). A Natural History of M gation. Chicago: University of Chicago Press.Hu, J. (1988). Temporal inter retation in English. Unpublished Ph.D. dissertation,University of Arizona.

Jackendoff, R. (1971). On some uestionable arguments about quantifiers and negation,Language 47,282-297.

Jacobs, R.; and Rosenbaum, P. S. (eds.) (1970). Readings in English TransformationalGrammar.Waltham, MA: Ginn

Jespersen, O. (1924). The Philoso y of Grammar. London: Allen and Unwin.

The problem of not. ..until 97

Karttunen, L. (1974). Until. Pap rsfrom the lOth Regional Meeting ofthe ChicagoLinguisticSociety, M. W. La Galy et al. ( ds.). Chicago: Chicago Linguistic Society.

Keizer, M. E. (1990). A typoio of copular sentences. Linguistics 28, 1047-1060.Klegraf, J.; and Nehls, D. (ed .) (1988). Essays on the English Language and AppliedLinguistics on the Occasion of erhard Nickel's 60th Birthday, Heidelberg: Groos.

Klima, E. (1964). Negation in nglish. In The Structure of Language: Readings in thePhilosophy of Language, J. Fo or and J. Katz (eds.). Englewood Cliffs, NJ: Prentice Hall.

König, E. (1988). Subject-operat rinversion after negative expressions in English: semanticaspects of a notorious syntacti problem. In Essays on the English Language and AppliedLinguistics on the Occasion of Gerhard Nickel's 60th Birthday, J. Klegraf and D. Nehls(eds.). Heidelberg: Groos.

-(1991). The Meaning of Focus articles. London: Routledge.Kroch, A. (1974). The semantics of scope in English. Unpublished Ph.D. dissertation, MIT.Ladusaw, W. A. (1980). Polarity Sensitivity as Inherent Scope Relations. New York:Garland.

La Galy, M. W.; Fox, R. A.; a~d Bruck, A. (eds.) (1974). Papers from the IOth RegionalMeeting ofthe Chicago LingUiS[icSociety. Chicago: Chicago Linguistic Society.

Lakoff, G. (1969). A syntactic arrment for negative transportation. In Papersfrom the 5thRegional Meeting of the Chicafo Linguistic Society, R. I. Binnick et al. (eds.). Chicago:Chicago Linguistic Society. 1_

-( 1970). Pronominalization, ne ation, and the analysis of adverbs. In Readings in EnglisbTransformational Grammar, R. Jacobs and P. Rosenbaum (eds.). Waltham, MA: Ginn.

Lasnik, H. (1976). Analyses 0 Negation in English. Bloomington: Indiana UniversityLinguistics Club.

Lindholm, J. (1969). Negative-r ising and sentence pronominalization. In Papers from the5th Regional Meeting ofthe Ch cago Linguistic Society. R.1. Binnick et al. (eds.). Chicago:Chicago Linguistic Society.

Linebarger, M. (1981). The Gra mar of Negative Polarity. Bloomington: Indiana UniversityLinguistics Club.

Löbner, S. (1989). German s hon-erst-noch: an integrated analysis. Linguistics andPhilosophy 12, 167-212.

Lodge, David (1975). Changing laces. London: Penguin.McCawley, J. D. (1988). The Sy tactic Phenomena of English, vol. 2. Chicago: University ofChicago Press.

Mittwoch, A. (1977). Negative sentences with until. In Papers from the 13th RegionalMeeting of the Chicago Lingu tic Society, W. A. Beach et al. (eds.). Chicago: ChicagoLinguistic Society.

Payne, J. R. (1985). Negation. I Clause Structure, T. Shopen (ed.). Cambridge: CambridgeUniversity Press.

Prince, E. (1976). The syntax a d semantics of neg-raising, with evidence from French.Language 52, 404-426.

Quirk, R.; Leech, G.; Greenbau ,S.; and Svartvik, J. (1985). A Comprehensive Grammar ofthe English Language. London Longman.

Seuren, P. (1974a). Negative's raveis. In Semantic Structure, P. Seuren (ed.). Oxford:Oxford University Press.

-(ed.) (1974b). Semantic Struct re. Oxford: Oxford University Press.Shopen, T. (ed.) (1985). Lang age Typology and Syntactic Description, vol. 1: ClauseStructure, Cambridge: Cambrit ge University Press.

Smith, S. B. (1974). Meaning an Negation. The Hague: Mouton.

98 R. Declerck

StockweIl, R. P.; Schachter, P.; a d Hall Partee, B. (1973). The Major Syntactic Structuresof English. New York: Holt, Ri ehart and Winston.

Vandeweghe, W. (1992). Perspe tivische Evaluatie in het Nederlands: de Partikels van deALjNOGjPAS-groep. Gent: Koninklijke Academie voor Nederlandse Taal- enLetterkunde.


Recommended