some truths and nontruths about 'even if

Some truths and nontruths about even jf'

Abstract

RENAAT DECLERCKand SUSAN REED

The authors describe the arious aspects of interpretation of even if conditionals. The following a e their main findings: (a) it is (part of) theinvariant meaning of eve that this focusing adverb evokes a sense of"expected incompatibility '; (b) in "implicative conditionals" (= thoseinterpreted as 'P leads to ') this sense of expected incompatibility manifests itself as a "sense of unexpectedness"; (c) even if conditionals thathave the sense of unexpec edness implicate an expectation understanding(' Onemight expect P topr clude Q') and entail a nonpreclusiveunderstanding ('P does not preclude '); (d) in implicative even if conditionals, evenevokes an even if scale, w ich is a scale of improbability and whose valuesare if conditionals, ordere from low (= least unlikely) to high (= mostunlikely); (e) because th values on the even if scale are conditionalsentences, implicative ev if conditionals are truly conditionals: theyexpress a conditional rela ion between Pand Q; (f) in quite a few casesthis relation does not hol between Pand the overt Q-clause but ratherbetween Pand an implicit Q. In that case the overt Q-clause is interpretedas an adverbial clause of reason; (g) the range of values on the even ifscale may be determined b the range of propositional values on an ancillaryscale of P-propositions, w ich may itself depend on another scale; (h) theordering of the values on he P-scale, and the concomitant interpretation,may be determined by the positive or negative meaning of the verb in theQ-clause; (i) there are go d reasons to include polar even if conditionalsin the class of scalar eve if conditionals; (i) even blocks the necessityimplicature ("conditional erfection"} that is very often invited by implicative conditionals; (k) no additional apparatus is needed to analyze thevarious kinds of even if onditionals with a nonassertoric Q-clause; (I)nonimplicative even if con itionals still have a sense of expected incompatibility but laek the sense 0 unexpectedness (yielding the expeetation understanding and the nonpree sive understanding) and the scalarity that are

Linguistics 39-2 (2001),203-255 0024-3949/01/0039-0203© Walter de Gruyter

204 R. Declerck and S. leed

typical of implicative even if conditionals; (m) we can distinguish twoclasses of nonimplicative e n if conditionals: those in which the P-clauseis purely concessive and tho e in which it expresses some kind of commenton the contents, truth, pres -positions, etc., of the Q-clause.

1. Introduetion

As appears from the list f references at the end of this article, manypeople have written about ven if. Together they have made virtually anyclaim that could be made bout even ij? but since the claims are oftencontradictory, it is time to eparate the correct claims from the incorrectones. This is what we will at empt to do in this article. To put it metaphorically, most of the pieces of the jigsaw puzzle have been offered, but theyhave been mixed with piece that do not fit in, and some of the necessarypieces are still missing. It i our purpose to identify the right pieces andto provide the missing on s, so that the jigsaw puzzle can finally becompleted.

Most of the authors ha e treated even ij, not as a single compoundconjunction, but as a com ination of the focusing partiele even and theij of a conditional. We s bscribe to this analysis. What needs to beestablished, then, is what t e effect is of putting even before the ij clauseof a conditional. As regard this, different claims have been made in theliterature. One of the aims of the present artiele is to show that theseclaims are often incorrect eneralizations, and (related to this) that therange of possible interpret tions of even ij is wider than has generallybeen assumed. We will also suggest explanations for these data and willattempt to show that they 11follow from a unitary meaning of even.

2. Preliminaries

2.1. We will use situation as a cover term for anything that can beexpressed in a sentence (i.e. ctions, events, processes, states - cf. Lyons1977). The verb "actualize"fill be used as a cover term for the predicatesthat are typically associate I with one of these categories. Thus, ratherthan saying that an actiVidl is performed, an event happens, a processtakes place, or a state hol s, we will simply say that the situation inquestion actualizes.

2.2. According to some li guists (see section 4.4.2), even ij sentencesare not conditionals. Howe er, we will argue in section 4.4 that at least

Some truths and nontruths about even if 205

those even ij sentences th t invite or require a scalar analysis must beconditionals, because the alues on the relevant scale are conditionals.Because of this, we will use the term "even ij conditional" more frequentlythan the semantically emp ier label "even ij sentence."

2.3. Thinking ofthe logi al representation 'IfP, (then) Q', we will referto the conditional clause as the "P-clause" and to the head clause (whetherit is itself subordinate or n t) as the "Q-clause." The situations expressedin these clauses will be called the "P-situation" and the "Q-situation."Because both the protasis P-clause) and the apodosis (Q-clause) may bepositive or negative, we s all refer to the two propositions expressed inthe P-clause and the Q-cla se as, respectively, "P" and "Q," rather than"p" and "q." This allows s to refer to affirmative propositions as [+p]and [+q], and to negative propositions as [-p] and [-q]. The symbols"P" and "Q" then stand or a proposition that is ultimately expressedby a tensed modalized cla se, with a positive or negative polarity value.As is clear from the par phrase 'if P, Q', ij does not form part of the

proposition P. Thus, in I' I be happy ij she comes, P is not ij she comesbut she comes. This is log cal, since only the latter has a truth value. Ifshe comes resembles a (no rhetorical) question in that it does not havea truth value; it is therefo e not a proposition.

2.4. As argued in Decle ck and Reed (forthcoming), there are manydifferent ways of classifyi g conditionals. One of the distinctions madethere is between "implic ive" and "nonimplicative" conditionals. By"implicative" we mean the conditionals in which Pinduces or implies Q,th at is, in which there is kind of causal, resultative, or licensing linkbetween Pand Q. Under t is heading we ean bring together "inferentialeonditionals" and "actuali ation conditionals." It is typical of these thatthey can generally be para hrased in terms of'IfP, thenQ'. The followingillustrate the two possibili ies:

(1) a. If it rains, the atch will be cancelled. (= actualization conditional: the actu lization of the P-situation will result in theactualization of he Q-situation)

b. If Watt didn't i vent the telephone, (then) someone else musthave done. (= i [erential conditional: the truth of P leads to theconclusion Q)

The following are exampl s of nonimplicative conditionals:"

(2) a. If John is not s art, at least he is reliable.b. He is one of th best students, if not the best, that we have

ever had.

206 R. Declerck and S. R ed

c. If I may say so, don't think he is at all to be trusted.d. I couldn't do ac ime even if I wanted. (TRC 102)

As we will see below, one of the factors determining the interpretationof even ij" may be whether the if conditional to which even is added isimplicative or not. This is interesting, because the examples that havebee.n discus~ed in the lite fatur.e h~ve nearly. ~lways been examples inwhich even IS added to an plicative if conditional. 4

2.5. We will argue in sect on 4.2 below that when even is added to animplicative ilconditional, e en anticipates, assumes, or evokes an expectation that the truth of the -proposition willjwouldjdoesjetc. cancel thevalidity of the Q-propositi n," and at the same time denies the truth ofthat expectation. In other ords, in this type of even if conditionals, 'evenifP, Q' denies the validity of the evoked expectation 'if P, not-Q'. Thismeans that there is both n "expectation understanding" ('you mightexpect P to preclude Q') and a "nonpreclusive understanding" C'Pwilljwouldjdoesjetc. not pr clude Q'). Thus, Even if I supported it, theproposal wouldfail is unders ood as 'You might expect that my supportingthe proposal would preclud the proposal's failing (= expectation understanding) but in fact my su porting the proposal would not preclude itsfailing (= nonpreclusive un erstanding)'.

2.6. Since even is added t an if conditional, and since 'even if P, Q'evokes an expectation unde standing that is expressed by an if conditionalof the type 'if P, not-Q', t is potentially very confusing to speak of"implicative even if conditi~nals." This could be understood as meaningeither that the even if conditional as a whole is implicative, or that onlythe if conditional is, or that the if conditional expressing the expectationunderstanding is. It is ther1eore important to note that in this artiele weare going to use "implicativ even if conditional" in this latter sense, thatis, as indicating an even if onditional whose expectation understandingis expressed by means of •n implicative conditional. Thus, Even if itrains, the match wil! not be I ancelled will be called an implicative even ifconditional because the rej cted expectation 'I[ it rains, the match willbe cancelled' is implicative.

2.7. In spite of being if c nditionals, some even if conditionals do notshow a direct "if" relation etween the even if clause and its head clause(i.e. the overt Q-clause). I, fact, there are two cases in which such adirect relation is lacking. T first possibility is that the relevant Q-clauseis not the overt Q-clause b t a that clause embedded into it, as in Evenif it rains, it's been decide that the match wil! not be cancelled. This


sentence can be analyzed s a syntactic variant of It's been decided thatthe match wil! not be canc lied even ij it rains. Second, and more interestingly, there is the possib lity that the "if" relation holds between theP-clause and an implicit -clause that must be reconstructed from thecontext. To make the difDrence c1ear, we wi11first consider an examplein which there is no impli it Q-clause:

(3) Even if the weather .s awful, the fête will take place.

This is interpreted as 'On might expect that the fête wi11not take placeif the weather is awful, b t (in fact) awful weather will not preclude thefête taking place'. The p ssibility of the "if" relation holding betweenthe even ij clause and an i plicit Q-clause is illustrated by the following:

(4) [She's got the best m tive of any of them, and she'd probably havethe nerve as well.] B t she couldn't have planned it all, surely, evenif somehow she had the opportunity (to commit the murder) thatnight, [say, after she got back from Stratford]. (DOe 240)

What is going on here is hat the speaker is going through a check-listof what would be require to make a good suspect (viz. motive, means,opportunity). He comes t the conclusion that the referent of she wouldnot make a good suspect, even if she had had the opportunity (which ishardly likely), because sh could not have planned it alloSo the messageconveyed in the even ij c nditional is, 'The view that she's not a goodsuspect wouldn't be unde mined (precluded) by it transpiring (which isunlikely) that she had the 'pportunity, because she couldn't have plannedit all'. That is, the "even i " relation holds between the implicit Q-clause(which we shall call Qd s e isn't a good suspect and the P-clause she hadthe opportunity. The appa ent Q-clause, she couldn't have planned it all,surely (which we shall cal Qz), is not the actual Q-clause but gives anexplanation for the assert on of 'Ql even if P'. We can read (4) as 'Butshe couldn't have planned it a11,surely, so she's (still) not a good suspect,even if somehow she had he opportunity'.If the interpretation 'Ev n P does not preclude Q' concerns an implicit

Q, the overt Q-clause as a mie expresses the reason why the speakerconveys this message: 'Cm believing that) Qz causes (me to say that) Ql,even if P' (where Qz is the overt Q-clause and Ql the implicit one), whichis equivalent to '(my sa ing that ) "Ql, even if P" results from (mybelieving that) Qz'. It folows that we can paraphrase such a sentencewith either '[Even if P, Q ] because Qz' or '[Even if P, Qz], so that Q/.Thus, (4) can be paraphr sed as 'Even if she had the opportunity, shecannot have committed th murder, because she could not have planned


it a11',or, alternatively, as 'Even if she had the opportunity, she couldnot have planned it a11,so he cannot have committed the murder'.The fo11owingis another xample in which P is related to an implicit Q:

(5) [So next time when y u sort through your wardrobe and wonderif Oxfam can se11w rn or old fashioned garments, remember]Wastesavers wi11put hem to good use even if our shops cannot.(COB-W)

This does not mean 'You ight expect that if Oxfam shops can't put thegarments to good use, Was esaverswon't, but that expectation is wrong'.The scenario evoked is tha of someone choosing between 'Oxfam shopscan se11these, so I'11give em to Oxfam' and 'Oxfam shops can't sellthese, so I'll throw them a ay'. That is, (5) evokes an expectation thatif Oxfam shops cannot pu the garments to good use, no one can, butthen denies this on the b is that Wastesavers can (and will) put thegarments to good use. In 0 her words, (5) expresses that the P-situation('Oxfam shops cannot put them to good use') does not preclude theIMPLICIT Q-situation 'some ne can put them to good use' because it doesnot preclude the overt Q-c ause situation 'Wastesavers wi11put them togood use'. (H is, of cours, possible to interpret the missing Q-clauseslightly differently, e.g. as ' here is no point in sending the garments toOxfam' or 'Oxfam cannot benefit from them', but this obviously doesnot affect the point we are aking here.)The observation that eve ij conditionals may have to be interpreted

in terms of an implicit Q-cl use has, to the best of our knowiedge, beenhinted at only once before Dancygier (1988: 118) gives the fo11owingexamples:?

(6) a.b.

(Even) if he attac s me, I've got a gun.(Even) if she ca11d yesterday, I was out at the time.

She comments that (a) such xamples "are interesting in th at the relevaneeof their apodoses can only be explained through elements of meaningwhich are not expressed on the surface" (1988: 118), (b) "in such (veryspecific) cases the scope of even is not the sentence as a whole, but thesurface protasis with its nderlying continuation" (1988: 118), and(c) such sentences "cannot unction in the same way if the order of theirclauses is reversed - app rently because their apodoses can only beconsidered relevant in relati n to what comes in the scope of even" (1988:119). However, claims (b) d (c) seem to be disproved by examples likeour (4) above, in which th even ij clause fo11owsthe overt Q-clause sothat, by Dancygier's criteri n of word order, neither the overt Q-clausenor the implicit Q-clause is ithin the scope of even.

3. The semantics of even


3.1. With the possible ex eption of Kay (1990: 83~84), everyone dealingwith the subject of even i has claimed that even represents its focus as"unexpected" or "unlikel ," hence that it is (or at least forms part of)the basic meaning of ev n in even ij that it represents the truth oractualization of 'ijP, Q' a contrary to expectation. This analysis is putforward, for example, by Jespersen (1940: 21.66), Fraser (1969: 67),Hom (1969: 106), Quirk e al. (1985: 1099), Kjellmer (1989: 257). Lycan(1991: 115~116) states th t "everyone knows that the main function,probably the only functio of even is to carry that expectation-contravening connotation." The lat st confirmation is to be found in Dancygier(1998: 162): "what even eems to share with although is the 'negativeexpectation.' "

In this artiele we will .gue a slightly different analysis. Our view isthat the basic meaning of ven is "expected (or expectable) incompatibility," which can manifest it elf in more ways than "negative expectation."Compare the following:

Even John faile the test.John's always p tient with me, even when he's in a hurry.Even if I get n help from anyone, I will go through withthe scheme. II enjoyed the w~lk, even if - as I found out afterwards - therest of the grou I didn't.

In (7a), the speaker had clearly expected there to be incompatibilitybetween the propositions Some people may fail the test' and 'John mayfail the test': in his opini n, one would have expected John to be theleast likely person, or one of the least likely persons, to belong to the setof people that might fai the test. In other words, there is expectedincompatibility between t e two propositions, in the sense that one wouldhave expected that if 'So e people may fail the test' turned out to betrue, 'John may fail the t st' would turn out to be false.In (7b), there is expecte~ incompatibility between John's being patient

with me and his being e a hurry. In (7c), even is added to an ijconditional. In the next sectien we will see that, at least in implicativeconditionals like (7c ),8 th combination of ijwith even's sense of expectedincompatibility leads to a 'sense of unexpectedness," which includes bothan "expectation underst ding" (viz. one might expect me not to gothrough with the scheme if I get no help) and a "nonpreclusive understanding" (viz. this expe tation is wrong: even if I get no help fromanyone, this will not pre ent me from going through with the scheme).

(7) a.b.c.

d.


However, there is no such sense of unexpectedness in (7d), where evenis added to a nonimplica ive conditional, because there is neither anexpectation understanding ('One would have expected me not to enjoythe walk if the others did n t enjoy it') nor a nonpreclusive understanding('The fact that the others di not enjoy the walk did not prevent me fromenjoying it'). Sentence (7d also does not assert that it was unexpectedfor the others not to enjo their walk. (It may actually have been thecase that the general expect tion was that the walk would not be enjoyabie, and that, except in case, that expectation was borne out.) In(7d), even just expresses t t my appreciation of the walk was differentfrom that of the others, a d that I find this rather surprising, in otherwords, that I had expected here to be incompatibility between a positiveappreciation of the walk myself and a negative appreciation of thewalk by the others. Even th s conveys a sense of expected incompatibility;it does not express "unex ectedness" or "negative expectation" in thesense that [-p] ('The res of the group didn't enjoy the walk') wasunexpected. (As noted ab ve, it may have been the case that it wasgenerally EXPECTED that tewalk would not be enjoyable, and that,except in my case, th at exp ctation was borne out.)In sum, the general mea ing of even is that it expresses expected (or

expectable) incompatibilit between two propositions. When even isadded to an implicative co ditional, as in (7c), this produces a "sense ofunexpectedness," which co sists of a combination of an "expectationunderstanding" ('One migh expect me not to go through with the schemeif I get no help') and a "no preclusive understanding" ('This expectationis wrong: even if I get no h lp from anyone, I WILL go through with thescheme'). When even is add d to a nonimplicative conditional, as in (7d),there is no "sense of unexpe tedness," because there is neither an "expectation understanding" ('0 would have expected me not to enjoy thewalk if the others did not njoy it') nor a nonpreclusive understanding('The fact that the others di not enjoy the walk did not prevent me fromenjoying it').

3.2. Apart from its expec ed incompatibility sense, even is commonlyrecognized to have scalarit as part of its semantics. We will not go intothis aspect here but defer t e discussion to section 4.4.

4. Implicative even ijcondi ,iona)s

Our analysis of even ij is i keeping with some claims that are familiarin the literature but is at v rianee with some others. First, we disagree


with the claim that even ij onditiona1s a1waysconvey a sense of unexpectedness. We will argue th at it is only in implicative conditionals (i.e. thosewhose expectation underst nding is expressed by means of an implicativeconditional - see section .6) that the sense of expected incompatibilityinduced byeven is equiva ent to a sense of unexpectedness. Second, wedo not accept Lycan' s (19 1: 115-116) suggestion that the expression ofunexpectedness is "the 0 ly function of even" in the combination evenij, even though this view, e pressed in terms of a scale of unexpectedness,is widespread in the lingu stic literature. With the possible exception ofSweetser (1990), no one h s argued that what we call the "nonpreclusive"understanding (= 'Even P does not preclude Q') is another aspect of themeaning of even ij. In fac , for many authors, a major feature of even ijsentences is th at P is IRRE EVANT to Q.9 This means that previous treatments generally focus on he fact that P does not affect the truth conditions for Q, rather than q estioning what the relation between Pand Qactually IS. The trouble ith this approach is that, while it naturallyinvolves the claim that eve ij sentences are not conditionals at all - seefor example Fraser (1969: 71), Dancygier (1998: 164) - its proponentsstill relate the meaning 0 even ij to unexpectedness (or, as Dancygier[1998] calls it, "negative e pectation"). Since unexpectedness clearly hasto do with the relation be een Pand Q (see Dancygier 1998: 162), it isimpossible to understand he claim about unexpectedness without someidea of WRAT the relation between Pand Q is: what IS the relation thatis unexpected?Third, we will argue th t, even if we restriet ourselves to even's sense

of unexpectedness in imp icative conditionals, we need to define thatsense more clearly than h s generally been done in the literature. Thus,if we consider Even ij the weather is awful, the fête wil! take place, weneed to say more than th t this sentence implies that it is contrary toexpectations that the fête ill take place if the weather is awful. In ouropinion, the interpretatio of the sentence involves two understandings,viz. an "expectation under tanding" ('We might expect awful weather topreclude the fête taking lace') and a "nonpreclusive understanding"('Awful weather will not p eclude the fête taking place'). More generally,there is an expectation un erstanding 'We might expect P to preclude Q'(or 'We might expect P to result in not-Q') and a nonpreclusive understanding 'P does NOT pree ude Q'. This characterization of the meaningof even ijis similar to Swe tser's (1990) analysis ("even P is sufficient forQ" or "even P is insufficie t for not-Q"), except that Sweetser's analysisseems to refer to the no preclusive understanding only and does notdistinguish between implic tive conditionals (which have the two understandings) and nonimplic tive ones (which, as we will see, do not). It

follows that Sweetser's ace unt involves scalarity in the interpretation ofall even ij conditionals, wh reas we shall argue that it has a role only inthe interpretation of impli ative ones.

Fourth, we will argue at, in implicative even ij conditionals, theexpectation understanding iffers from the nonpreclusive understandingin that it is only a cancella le implicature, while the nonpreclusive understanding forms part of the semantics (invariant meaning) of implicativeeven ij conditionals.

4.1. The sense of unexpect dness

Consider the following:

(8) a.b.

Even if John hel sus, the job will not be finished today.Even if John doe not help us, the job will be finished today.

Because (8a)-(8b) are imp icative even ij conditionals, the combinationof ij and the sense of expected incompatibility induced byeven pro ducesa sense of unexpectedness. his means that (8a)-( 8b) yield both an expectation understanding and nonpreclusive understanding. That is, theyassert that the expectation hat P will/would/does/etc. preclude Q is notcorrect; in doing so they nat rally imply th at what is asserted is unexpected(surprising). Thus, (8a) imp ies that it is surprising (because it goes againstthe expectation to the cont ary) that the job will not be finished today ifJohn helps us, and (8b) i lies that it is surprising that the job will befinished today if John does ot help us.A question that naturall arises is whether the sense of unexpectedness

conveyed by implicative ev ij conditionals concerns P only or concernsthe entire conditio naI(i.e. t e P-Q relation). To see which of these optionsis the correct one, consider he following:

(9) Even if it snows, the atch will not be cancelled.

If the sense of unexpected ess concerned P only, then (9) would implyonly 'It is unlikely to snow', that is, 'Snow is not expected.' However, thisunderstanding is at best an i plicature of (9), as it can easily be cancelled:Even ij it snows, which is qut e likely, the match wil! not be cancelled. Sinceat least the nonpreclusive nderstanding in the sense of unexpectednessforms an inherent (noncanc llable) part ofthe meaning ofimplicative evenij, we must conclude that t e second theoretical possibility is the correctone: the sense ofunexpecte ess concerns the conditional as a whole, thatis, the implicative relation btween Pand Q. This means that in (9) it canbe paraphrased as follows:

( 10) It is surprising that + p] will lead to [- q] because wewould expect[+p] to preclude [-q].


4.2. The nonpreclusive un erstanding of implicative even if conditionals

Consider the following ag in:

(11) a. Even if John h lps us, the job win not be finished today.b. Even if John d es not help us, the job win be finished today.

It is clear that (lla)-( 11b) have not only an expectation understandingbut also a nonpreclusive understanding: they can be paraphrased as(12a)-( 12b), respectively:

(12) a. One might exp ct that the job would be finished today if Johnhelped, but in act the job will not be finished today if Johnhelps. (= One ight expect that P would preclude Q - wherePis [+p] and Q is [-q] - but P will not preclude Q.)

b. One might exp ct that the job would not be finished today ifJohn did not ~elp, but in fact the job win be finished todayif John does lot help. (= One might expect that P wouldpreclude Q - here P is [-p] and Q is [+q] - but P willnot preclude Q.

Needless to say, the nonp eclusive understanding ('P does not precludeQ') is not an implicature ut a logical entailment of what is asserted byan implicative even ij con itional. It is an aspect of meaning that cannotbe cancelled. The followin is anomalous, because contradictory:

(13) *Even if John help us, the job will not be finished today, but infact John's helpin us win prevent the job not being finishedtoday.

4.3. The expectation und rstanding ofimplicative even if conditionals

The expectation understa ding 'One might expect P to preclude Q' follows as astrong implicat re from the combination of the nonpreclusiveunderstanding of the impli ative even ij conditional ('P does not precludeQ') and the sense of une pectedness ('it is surprising that P does notpreclude Q'). Naturally, 't is unexpected that "ij P, Q'" suggests 'it isto be expected that "if P, ot-Q"', because to say that something is notexpected automatically e okes a contrast with something else that IS

214 R: Declerck and S. R ed

expected. That "somethin else" is usually the opposite of what is notexpected.

It is in keeping with this hat (14a) is interpreted as (14b) and stronglysuggests (14c):

(14) a.b.

Even if it rains, the garden party will go through.Contrary to ex ectations, the garden party will go throughif it rams. (sense of unexpectedness +nonpreclusiveunderstanding)One might expe t that rain would preclude the garden partygoing through. expectation understanding)

More generally, (l5a) expr sses (l5b), which strongly suggests (15c):

c.

Even if P, Q.Contrary to exp ctations, P doesjdidjwilljetc. not preclude Q.(sense of unexp ctedness +nonpreclusive understanding)If P, then P is e pected to preclude Q (or P is likely to resultin not-Q). (exp ctation understanding)

(14b) naturally suggests (1 c) because the idea of "contrary to expectations" in (l4b) contrasts ith "one might expect" in (l4c), and (l5b)naturally suggests (15c) be ause to say that nonpreclusion is contrary toexpectations implies that p eclusion is in accordance with expectations.However, the expectation nderstanding (l5c) is only an implicature of(15b), because it can be cancelled by the context without semanticanomaly.'? Thus, the expe tation understanding 'You might expect thishorse to react if a bomb go s off' that is evoked by (16a) is cancelled bythe added sentence in (l6b) without (16b ) as a whole being semanticallyanomalous (contradictory):

(15) a.b.

c.

Even if a bomb ~oes off, this horse won't react.Even if a bomb goes off, this horse won't react - as is to beexpected from a police horse.

Moreover, in all implicativ even ij conditionals, the expectation understanding 'you might expect to preclude Q' is cancelled by the Q-clauseitself, which says that P do SNOT prec1ude Q.The fact that the expect tion understanding may be cancelled by the

context does not alter the f ct that it always arises in implicative even ijconditionals: there is an nderstanding that the speaker anticipates,assumes, or evokes an expe[ation (on the part of the addressee, someoneelse, or "people in general ') th at the truth of the P-proposition willjwouldjdoes etc. prec1ude th validity of the Q-proposition - an understanding th at is at the same time refuted by the inevitable nonprec1usive

(16) a.b.

understand!ng.1.1 The cla~ that even ij invariably calls up an ~~pectationunderstandmg IS confirm d by the fact that even ij conditionals areunacceptable if they lack is sense. Compare the following:


If Iris Murdo h didn't write The End of the Affair, it musthave been Gr ham Greene.*Even if Iris urdoch didn't write The End of the Affair, itmust have be n Graham Greene.If it rains a 10 , we won't need to water the garden.*Even if it rainf a lot, we won't need to water the garden.Even if it rainf a lot, we'll (still) need to water the garden.Even if it do sn't rain a lot, we won't need to water thegarden.

Unlike (17a), (17b) is una ceptable because even intro duces the pragmatically unacceptable expecta ion understanding 'one might expect GrahamGreene not to have writte The End of the Affair if Iris Murdoch didn'twrite it'. The (un)accepta ility of (18b)-(18d) similarly has to do withwhether or not the expecta ion understanding is pragmatically acceptable:in (18b) it is not, but a eh nge ofpolarity (positive vs. negative) in eitherthe P-clause or the Q-cla se entails that the expectation understandingmakes perfect sense in (18 ) and (18d).

As pointed out in sect on 2.7, the Q-proposition of an implicativeconditional may correspo d with the overt Q-clause or may be implicit.

inside, which is interprete as 'You might expect th at if it freezes we wi11stay inside, but in fact we i11NOT stay inside if it freezes' (= 'You mightexpect P to preclude Q - where Q is 'we not stay inside' - but in factP wiUnot preclude Q'). In other words, 'even iJP, Q' denies the truth ofthe evoked expectation 'i, P, not-Q'. The fo11owing sentence furtherillustrates this use of even . :

(17) a.

b.

(18) a.b.c.d.

(19) Teil her 1'11be ther , even if they have to wheel me in. (Dexter1994: 149) (= It mi ht be supposed that my having to be wheeledin wouldpreclude m being there, but this supposition would befalse:I WILL be there.)

However, there are also c ses in which it is not the validity of the overtQ-clause itself that is cance led but the validity of a Q-clause implicit in it:

(20) [Double glazing an draught insulation can cut lost heat by halfand save about f60 year. Insulated cavity walls are also effective,saving up to no a y ar. However, they can be expensive to insta11,costing up f400.] ven if you already have basic energy-saving


measures in place th re are usually between 7 and lOother stepsyou could take to m ke even more savings. (COB-W)

This does not mean 'Contr ry to expectations, your already having basicenergy-saving measures in lace does not prec1ude there being between7 and lOother steps that ou could take to make even more savings',but something like 'Contra y to expectations, your already having basicenergy-saving measures in place does not prec1ude your thinking stillfurther about saving energ , because there are between 7 and lOothersteps you could take to ma e even more savings'. Similarly,

(21) [AHABTA tour ope ators are obliged to go to arbitration and toaccept the final deci ion. You can get full details of the schemefrom ABTA or a I cal consumer adviser. Note:] Even if yourholiday was not arra ged through an ABTA member, anyone whoprovides a service h s alegal duty to do so with reasonable careand skill, in areaso able time, and for a reasonable charge. [Afirm must observe tese obligations.] (COB-W)

This does not mean 'You ight think that if your holiday has not beenarranged by an ABTA m ber, whoever arranged it does not have alegal duty to arrange it wit reasonable care (etc.), but that assumptionis wrong: your holiday's no having been arranged by an ABTA memberdoes not prec1ude any serv ce provider's having alegal duty to arrangeit with reasonable care (e c.)'. The real interpretation is 'Contrary towhat you might expect, yo r holiday's not having been arranged by anABTA member does not p ec1udeyour having a right to redress ij thingsgo wrong, because anyone ho provides a service has alegal duty to doso with reasonable care (et .).'The following are furth r examples of even ij conditionals with an

implicit Q:

(22) a. Even if we had undertaken a hunt, it would not have beensuccessful. (= ven ij we had undertaken a hunt, that wouldn'thave changed anything, because it wouldn't have beensuccessful.)

b. [She won't let e alone.] Even if I go upstairs to work, shebrings me six c ps of coffee in two hours. (BOF 83) (= Evenij I go upstairs 0 work, she does not leave me alone, becauseshe brings me si cups of coffee in two hours.)

c. Even if there w re a sheriff in this town, he wouldn't be ableto stop the ban its. (= Even ij there were a sheriff in this town,that wouldn't so ve the problem, because he would be unable tostop the bandits.


In sum, implicative even i conditionals are interpreted in terms of anexpectation understanding and a nonpreclusive understanding. The interpretation resulting from hese understandings can be paraphrased as'Contrary to what might ,e (or might have been) expected, P does notresult in not-Q', or 'Cont ary to what might be (or might have been)expected, P does not precl de Q'. It needs stressing, however, th at the Qin question is not always e pressed by the overt Q-clause but may be animplicit Q.

4.4. Scalarity

4.4.1. Since (at least) H rn (1969), all authors on even have addedanother element to the "c ntrary to expectation" part of even's meaning,viz. that of scalarity. Eve is interpreted as scalar because it places theconstituent it focuses as e tremely high or low on a particular scale. Forexample, Even Mary man ged to do it implies that (of those who did it)Mary was the least likely person (or one of the least likely persons) tobe able to do itY Sweetsel (1990) puts it as follows:

al! ... examples of even ij ... are more than simply concessive: they express notonly opposition between the two clauses but the further idea that the protasisrepresents a relatively extre e possibility from among the possible conditionswhich can be expected to oe ur in opposition to the truth or the fulfil!ment ofthe apodosis (1990: 136)

Dancygier (1998) also as erts that scalarity IS a central feature of themeaning of even:

What al! accounts share is th claim that even introduces a scale of unlikelihood,or negative expectation, the ighest position on which is occupied by the referentin the scope of even.... Th s what even seems to share with although is the"negative expectation" - he ,ce its concessive use - but its most salient meaningis that ofscalarity .... the str cture in the scope of even ijranks high on the scaleof "negative expectation" (1 98: 161-162).

4.4.2. Since, as noted in eetion 4.1, the sense of unexpectedness of evenij concerns the condition I as a whole (i.e. the relation between PandQ), the scale that author attribute to the use of even ij must apply tothe RELATION between Pad Q. (This is often not clear from the literature.) Thus, Teil her Til b there, even ij' they have to wheel me in can beinterpreted as meaning t at their having to wheel me in is the leastexpected (or at least a hi hly unlikelyj'" circumstance in which I wouldbe there, and as implying contrast with other circumstances in which I

218 R. Declerck and S.

would be there, not just ith other "less extreme" circumstances.l" Inother words, in implicative even ij conditionals each value on the relevantscale is an if conditional.P In the above example, the values (ordered interms of increasing unlikeli ood) might be something like I'll be there ijlam driven there, Til be th re ij I have to drive there myself, Til be thereij I have to go there by bik , I'll be there ij I have to go there on foot, I'llbe there ij they have to wh el me in, etc., or, possibly, J'll be there ij Tmweil, Til be there ij Tm stil not fully recovered, Til be there ij Tm still ill,I'll be there ij (Tm so ill th t) they have to wheel me in, etc.

It is worth pointing out hat the view that (at least in implicative evenij conditionals) even ij refers to a scale whose values are ij conditionalsruns counter to the wide pread claim that even ij sentences are notconditionals. The latter cla m is made explicitly by Fraser (1969: 66):

concessive conditionals [= ev n if conditionals] ... while they have the generalappearance of conditional se tences, ... have no conditional force; the if ispreceded in each case by an e en which neutralizes the hypothetical force of thefollowing if clause. The result is a concessional clause, a clause which plays norole in determining the truth onditions for the main clause of the sentence.

The claim that even ij sentejces are not conditionals is also made explicitlyby Athanasiadou and Dir en (1997: 85), Dancygier (1998: 164), andLycan (1991: 123). The lat er puts it as follows:

Although Q even if P has the uperficial aspect of a conditional and although itseems granunatically to be si ply the result of applying even to an ordinaryconditional, it does not seem i tuitively to be conditional in meaning. A speakerwho asserts Q even if Pis typic lly feIt to have asserted that Q - unconditionally.

The above view is equivalent to the view that in even ij conditionals P is"irrelevant" to Q. This view was first voiced by Fraser (1969: 68), whowrites that in Mary willieaje even ij John stays there is an inference that"Mary will leave no matter what happens." However, Fraser goes on tocomment that there is (Sim[ltaneouslY) an interpretation that "it wouldbe very unlikely for Mary 0 leave if John stays" (1969: 68). Now, ifthere is an interpretation th t it would be very unlikely for Mary to leaveif John stays, then there is~an expectation of a conneetion between thetwo situations, which mean that the even ij conditional can hardly implythat P is seen as IRRELEVAN to Q.The view that even ij sentlces are not conditionals is further disproved

by the fact that there are clauses that seem fairly uncontroversiallyconditional and seem to rel te to the Q-clause in a way similar to thatof even ij clauses. Thus, in ( 3b) the even ij clause relates to the Q-clausein the same way as the ij cl use does in (23a):


(23) a. [Wear this bul et-proof jacket.] That way you will be OK ifyou are shot a .

b. [Wear this bull t-proof jacket.] That way you wil! be OK evenif you are shot at.

In both (23a) and (23b), he ij clause specifies the salient case in whichQ is true (or in which the -situation actualizes). The difference betweenthe two examples is that he addition of even in (23b) emphasizes thatcompatibility between the P-situation and the Q-situation is not usuallyexpected, or expectable.

4.4.3. It is clear from ou above discussion of Tel! her Tl! be there, evenij they have to wheel me in that implicative even ij conditionals lendthemselves to a scalar an lysis, and that the values on the even ij scaleare ij conditionals. It is int resting to note that the position of the relevantij conditional (i.e. the on combining with even) on the even ij scale isoften determined by the p sition of its P-clause on an "ancillary" (subsidiary) propositional scale P-scale), and that this position may itself bedetermined by the positio of a constituent of the P-clause on anotherancillary scale. The consti uent in question is then strongly accented andis felt to be the real focus f even. For example, evenmay focus a numeralin the even ij clause. In t at case the numeral is heavily accented andinterpreted as representin a (relatively) extreme value on the numericalscale (which is a "natural ' scale). This numeri cal scale functions as anancillary scale to the P-sc le (whose values are P-propositions) becausethe position of the numera on the numerical scale determines the positionof the P-clause on the P scale. The latter scale itself functions as anancillary scale to the even ij' scale (whose values are if conditionals) inthe sense th at the positio of the P-clause on the P-scale determines theposition of the if conditi nal on the even ij scale. However, since (inimplicative even ij conditio als) even always puts the relevant conditionalhigh on the scale ofunexp ctedness, the ordering ofthe values (conditionais) on the even ij scale m y be the reverse of the ordering of the values(propositions) on the anci lary P-scale. For example,

(24) a. Even if you ha e written SIX papers, you will not pass thisexam rif you d not know the coursebook by heart]. (In thiscontext, "six pa ers" is an unusual!y high value on the numericalscale of "numb r of papers compatible with failing this exam. "This numerical scale functions as an ancillary scale to theP-scale whose alues are the P-propositions 'You have writtenO/I/2/3/4/5/6/et . papers'. The high position oj'You have written six papers' n this ancillary P-scale, plus the use of "even,"


determines the h 'gh position of 'You wil! not pass ij you havewritten six pape s' on the "even ij" scale on unexpectedness.)

b. [Afterwards he ept on saying that it was me who'd agreedto do it, me w~o'd started it all not him. Mum! He was awicked liar, but' even if it was just one per cent me you'vegot to forgive e. (DOe 89) ("lust" represents one per centas a low value n the numerical scale. This low position of"one'' on this ancillary scale determines the low position of 'Itwas one per een me' on the P-scale. However, "even" makesclear that the c nditional 'You've got to forgive me ij it wasone per cent me is a very high value on the "even ij" scale,which is a scale 0 unlikelihood and whose valuesare conditionalsofthe type Tou' e got toforgive me if it was 1/2/3/4/5/etc. percent me'. This is an example, then, of how "even" reverses thedirection (orien tion) of the ancillary scales underlying the"even ij" scale, ecause on the latter scale the least likely valueranks highest.) 16

c. Even if you hav written NO papers at all, you will pass [ifyou know the c ursebook by heart]. ("No" is the lower limitof the numerica ancillary scale, hence 'You have written nopapers' is the 10 est value on the P-scale. However, the valuesare ordered in th reverse way on the "even ij" scale: 'You wil!pass ij you hav written no papers' is the highest (i.e. leastlikely) value on this scale. The lower values on the scale areTou wil!pass ij ou have written {I/2/3/etc.} papers'.)

Apart from the numerical cale, there are other natural quantificationalscales:(25) Even if EVERYBO Y helped, we wouldn't be able to move

that boulder. ("Ever body" represents the highest possible value onthe quantificational ale of "number of people." It follows that'Everybody helps' is t e highest value on the P-scale, which is ancillary to the "even ij" scale. It follows that 'We will not be able tomove that boulder ij e erybody helps' ranks highest on the "even ij"scale, the lower valu s being 'We wil! not be able to move thatboulder ij {most ofu / some of us / not many of us / etc.} help'.)

Apart from a qantifier, ev can also focus another constituent of theij clause:

(26) Even if JOHN is late the coach will leave on time.

This should be analyzed a ong similar lines to (24a)-(24c) and (25).However, it also allows s to make an interesting observation: the


meaning of (26) changes rasticaUy if we move even into the ij clause,as (28a) demonstrates. Co pare (26) with (27):

(27) If even JOHN is lat, there is no discipline in the team any more.

In (26) the ij clause expre ses an "open" condition (i.e. a condition thespeaker thinks mayor ma not be fulfilled). By contrast, the ij clause of(27) expresses either a "f ctual" condition or a "closed" one. (In theformer case, the P-situati n is represented as BEING a fact in the actualworld, in the latter it IS AS UMED to be a fact.) This difference of meaningentails that the foUowing re unacceptable on the intended readings:

(28) a. *If even JOH is late, the coach willieave on time. (intendedreading: open pondition)

b. *Even if JOHN is late, there is no discipline in the team anymore. (intendld reading: closed or factual condition)

The different interpretatio s of (26) and (27) are due to the fact that thescale determining the inte pretation is not the same in (27) as in (26).In (26), where even prece es the ~lclause, the relevant scale is an even ijscale, whose values are CO;ditionals, and the least likely conditional (i.e.the conditional in which e P-situation is least likely to be compatiblewith the Q-situation) ran s highest, whereas in (27), where even occursinside the ij clause, the r levant scale is a scale of people ordered inaccordance with a partic lar criterion.'? In (26) the range of values(conditionals) on the ev n ij scale is determined by the range ofP-propositions on the anci lary P-scale. The range of values on the P-scaleis determined by the rang of persons on an anciUary scale whose valuesare persons, such as Mike, Jack, Gordon, John. In this case this ancillaryscale is evoked by the spe ificational focus on John, which calls up a setof alternative values. (A ' specificational" sentence specifies a value, viz.the focus, for an implicit ariabie. Thus, JOHN is late specifies John asthe value satisfying the p esupposed variabie 'the x who is late', in thesame way as It is John W 0 is late does - see Declerck (1988). One ofthe characteristics of a spe ificational sentence is that a contrast is evokedbetween the value that is selected and a set of other poten tial values thatare not selected. It is the sum total of these values that constitutes theset of values on the ancil ary scale of persons. The position of "John"on that scale determines t e position of John is late on the P-scale, andconsequently the position of the conditional 'The coach will leave ontime if John is late' on th even ij scale.)!"The difference between (26) and (27) follows from the fact that only

the former involves an ev n ij scale. This has to do with the fact that in(27) there is only one focu ing speech act (with even as focuser and 'John


is late' as highest value on he scale of persons that might be late, whichis a scale of unexpectednes or unlikelihood), whereas in (26) there aretwo: there is focusing by ven as wel! as specificational focusing. (Thelatter results in [26] bein equivalent to 'Even if it is JOHN who islate .. .'.) Even ij implies a propositional scale (P-scale) and creates aneven ij scale (which is also an improbability scale). The specificationalfocusing creates the range of values of the P-scale (viz. 'Mike is late','Jack is late', etc.). This in ts turn determines the range of values of theeven ij scale (viz. 'The coa h wilileave on time if MIKE is late', etc.).Even then places the condi ional 'The coach will leave on time if JOHNis late' high on the even ij s ale, thus representing its Q-situation as moreunlikely than that of 'The oach wil!leave on time if MIKE is late', etc.

If the focus of even is nei lher ijnor a quantifier, nor a specificationallyfocused constituent of the i clause, the hearer himself has to identify theancillary scale that determi es the even ij scale:

(29) We'll find the girl, ev n ifshe's at the OTHER end ofthe WORLD!

This sentence suggests that one of the least likely cases in which the girlwill be found is when she i at the other end of the world. In this casethe first ancillary scale is he scale of places where the girl might be,ordered according to the egree of difficulty they entail of finding thegirl: the more difficult the pace renders it to find the girl, the higher thisplace ranks on this scale. N turally, 'at the other end of the world' is anextremely high value on t is ancillary scale. This scale determines theancillary P-scale, which is a scale with as values propositions of the type'The girl is at X'. In accor ance with the ordering on the first ancillaryscale, the proposition 'The girl is at the other end of the world' is anextremely high value on thi ancillary P-scale. Because of even, the conditional 'We'll find the girl if she's at the other end of the world' isinterpreted as lying near the upper limit ofthe even ijscale (improbabilityscale), whose values are co ditionals of the type 'We'l! find the girl ifshe's at X'. (Remember t at on the even ij scale the least likely ij"conditional is the highest v lue.)

There are also cases in w ich even focuses the ij clause as a whole. Inthat case there is no ancil ary P-scale, and therefore no special (contrastive) accentuation patter : the nuclear accent ofthe ijclause is weakerand is in its unmarked positi n, viz. on the last open-class word conveyingnew information. For exam le,

(30) a. Teachers are be ng urged by trade union activists not tointervene in vio ent incidents even if a pupil is in danger.(COB-W)


b. Should a pare t put their children's needs above their own,even if it mean sacrificing their own happiness? (COB-W)

c. When looking or a partner we often go for someone like oneof our parent , even if it's not very comfortable for us.(COB-W)

In each of these example the ij conditional represents a value that isparticularly high (possibly the highest) on the even ij scale. In (30a) thevalues constituting the sc le are conditionals whose ij clauses representthe various cases in whic it might be thought that it is to a certaindegree necessary for teac ers to intervene in violent incidents. In (30b)the values are conditionals whose P-clauses describe situations that differin the degree to which the require parents to put their children's needsabove their own. In (30c) the values are conditionals referrring to situations in which it is possible that wewould choose a partner who resemblesone of our parents. In all three examples, the position of the 'if P, Q'values on the even ij scale s determined by the degree to which 'P resultsin (or is compatible with) 'is seen as unlikely to be true.

4.4.4. Another thing that has not been noted yet in the literature is thatthe ordering of the values n the P-scale, and the concomitant interpretation, may be determined y the polarity (in meaning) of the verb of theQ-clause. (By "polarity in meaning" wemean that the speaker may viewa situation as either positi e [pleasant, satisfactory, desirabie] or negative[unpleasant, unsatisfactor , undesirable].) Compare the following:

(31) a. Even if you ans er TWO of the three questions correctly, you{fail / don't pa s} this exam. ("fail" or "not pass" is a negativeidea, hence "tw " is interpreted as "as many as two"}

b. Even if you ans er TWO of the three questions correctly, you{pass / don't fa1} this exam. ("pass" or "notfail" is apositiveidea, hence "tw r s is interpreted as "only two")

As suggested by the para hrases of two, the orientation of the ancillaryP-scale is different in (31 ) and (31b): you answer TWO of the threequestions correctly is seen s a high value on the P-scale of (31a), but asa low value on the P-scal underlying (31b). However, because of thedifferent semantic polarity of the Q-clauses of the ij conditionals in (31a)and (31b), both condition Is are interpreted as ranking high on the eveniJscale: (31a) expresses th t it is highly unexpected (unlikely) to fail theexam if you answer two f the three questions correct1y, while (31b)expresses that under the ame circumstances it is highly unexpected topass the exam.


4.4.5. The various cases f scalar even ij conditionals that have beenreferred to have the folloying characteristics in common. The even ijscale is a scale determined Py the degree of improbability that P leads to(or is compatible with) Q. he values on the scale are implicative conditional sentences, that is, e pressions of 'P leads to Q' (which minimallyimply 'P is compatible wit Q') .19 These ij conditionals are ordered fromlow to high according to t e degree of improbability that P leads to Q,or, if you prefer, the degre of expeetation (probability) that P leads tonot-Q (i.e. that P precIude Q). Thus, Even ij it snows, the game wil! notbe cancel/ed evokes a scale ,f improbability whose values are eonditionalsof the form 'if P, Q' (wh re Q is [- q]). Possible values on this scale(ordered from low to high) may be

a. The game will not be caneelled if there is a lot of wind.b. The game will not be cancelled if there is some drizzle.c. The game will not be cancelled if it rains.d. The game will not be caneelled if it snows.The ordering of these values on the improbability scale is from

(a) (= lowest value) to (d) (= highest value). Thus, (d) represents thecase in which the expectation that P will lead to not-Q (i.e. that P willprecIude Q) is strongest: ac ording to the speaker, snow is the P-situationthat is most likely to evo e the expectation 'P is sufficient for not-Q'(where Q is [-q] and not- is therefore [+q]). In other words, it is if itsnows that the expectation that the game will be cancelled is highest. Itfollows th at the even ijscale ean be referred to not only as an "improbability scale" but also as an " xpectation scale."20 The values on the scaleare eonditionals of the ty e 'P leads to Q', each with its concomitantexpeetation understanding 'P will lead to not-Q'. Since (in implieativeeven if eonditionals) even i plies that the expectation is denied, that is,that P will not preclude Q i, spite of the expectation to the contrary, theeven ijconditionals corresp ,nding to (a)-(d) have not only an expeetationunderstanding but also a n nprecIusive understanding.

4.4.6. It should be cIear n w that a sealar analysis of (implieative) evenij conditionals refers to an ven ij seale, which is a scale of improbability(with the least probable va ue ranking highest). The scalar values are ijconditionals, that is, expres ions of eompatibility between Pand Q. Thisscalar analysis crucially in lves the following understandings:a. the "extreme value u derstanding": the even ij conditional repre

sents a value that ranks ver high (sometimes highest) on the even ijscale.b. the "plural set under~tanding": by focusing a particular ij condi

tional, even evokes a set or other ij conditionals, which represent lessextreme (Iower) values on lhe even ij seale.


c. the "contrastive unde standing": the if conditional representing theextreme value is con traste with all the other (less extreme) values (ifconditionals) on the even i scale.d. the "ordering unders anding": the if conditionals representing the

values on the even ifscale a e ordered in terms ofthe degree ofimprobability that they are valid. T e highest conditional on the scale is that inwhich P is least likely to b compatible with Q.

e. the "implicative und rstanding": the conditionals on the even ifscale must be implicative onditionals, that is, conditionals of the type'P leads to Q' (the weakesl form of which is 'P is compatible with Q').These five understandin s are essential to a scalar interpretation of

even if conditionals. This means that there is automatically a scalarinterpretation if all the un erstandings are present, and that there is noscalar interpretation if 0 or more of them are lacking. (The latterpossibility has not been ree gnized in the literature: all even if conditionalsare traditionally assigned scalar analysis. However, it will be shown insection 5 that only impli ative even if conditionals can be analyzedthis way.)

4.4.7. Before examining onimplicative even if conditionals, it is necessary to counter a potential objection to the claim that all implicative evenif conditionals require a s alar analysis. This potential objection has todo with the "extreme val e understanding" and the "plural set understanding." It would seem hat there are implicative even if conditionalsth at are nonscalar because they lack these understandings. Por example,

(32) I mean er er these t ings all point to the fact that they're going totry and hold on to t eir, hold on to their positions you knowevenif, even if they're cri·ticized. (COB-S)

At first sight, it seems que tionable whether the utterance of this even ifconditional evokes an extr me value understanding: it is not obvious that"their" being criticized is ~t (or near) an extreme of any scale of P-casesthat are unlikely to be co patible with their trying to hold on to theirpositions. The reason this ould seem an inappropriate reading of (32)is that being criticized doe not seem extremely unlikely to go along withtrying to hold on to one's osition. (Being found guilty of fraud, sexualharassment, or arson, yes, but being criticized?) In fact, (32) just seemsto be a simple case of even . denying or precluding a possible assumptionthat P is sufiicient for not,Q: 'you might expect that they would not tryto hold on to their positi ns if they were criticized, but they will try tohold on even if they are'. Note that "if" should not be read as "iff").This reading lacks not onll the "extreme value understanding" but also

226 R. Declerck and S.

the "plural set understandi g": if there is a scale, the only two values onit are 'they will try to hold on to their positions if they are criticized' and'they win try to hold on to heir positions if they are not criticized'. Sincethese are the only two v lues, they both represent something like anextreme end of the scale, vi . the highest and lowest value, respectively."The crucial question is wh ther we can still speak of a scale if there areonly two values, which are each other's polar opposites.

It should be noted that ven (33), which is the prototypical examplein the literature on even i .conditionals and which is always treated asscalar, allows a polar inter retation:(33) Even if it rains, the atch will not be cancelled.With a contrastive accent n rains, we can readily assume a number ofalternative conditions (e.g. 'if only ten members turn up', 'if it is notsunny', etc.) that might be onsidered less likely to bring about cancellation of the match than 'if 't rains'. But if there is no contrastive accent,there is at least a possibilit of reading (33) as expressing little more thana polar contrast between ' f it rains' and 'If it doesn't rain', that is, thepossibility of reading (33) s 'Even if it DOES rain, the match will notbe cancelled' .22 For exampl , I could say Even ij it rains, the match won'tbe cancelled to somebody nfamiliar with the sport of rugby (and thevery wet conditions under which it is sometimes played), and it wouldbe very unlikely that I w uld have in mind a scale of situations thatmight (wrongly) be expect d to cause cancellation, of which rain is themost likely. I would more likely simply be explaining th at the personshould understand that, de pite the looming black clouds, we will still begoing to the rugby match. In other words, (33) can simply be a denialor precluding of an assum tion of incompatibility between rain and thematch not being cancelled In this interpretation, the two contrastingconditionals involved (viz. 'The match will not be cancelled if it rains'and 'The match will not be cancelled if it does not rain') would not seemto be on a real SCALE of im robability but simply relate one to the other.Effectively, they are the eq ivalents of the alternative answers to apolarquestion - see also Haima (1974: 342) and König (1986: 231).In sum, apart from even ij conditionals that have a clear plural set

understanding, there are al~~ even ij conditionals that just evoke a singlealternative with which the ~_conditional is contrasted. The alternative isa conditional with a P-da se that is the polar opposite of the P-dauseof the even ij conditional. he following is an example that can only beinterpreted this way:(34) [I hope John win ag ee. But] even if he doesn't, we'll carry out

this decision.

Same truths and nontruths about even if 227

In this case the only cont ast there is is between 'if John agrees' and 'ifJohn doesn't agree', whic are each other's polar opposites. It is typicalof sueh polar even if con itionals that they can be paraphrased by 'Q,both if Pand if not-P', Ot 'Q, whether or not p': 'We'Il carry out thisdecision, whether John ag ees with it or not'.It is open to debate wether in these cases we can still speak of a

SCALE. It seems reasonab to argue that the plural set understandingthat is typical of a scala interpretation requires a number of valueshigher than two. Howeve , there are a number of arguments that pleadin favor of assigning a (s ewhat special) scalar analysis even to polareven if conditionals. To b gin with, a set of two (one of them being theif conditional focused by ven) may be a small set, but it IS a plural set.Second, like prototypical s alar even if conditionals, (34) is an implicativeeonditional with an expeet tion understanding as weIl as a nonpreclusiveunderstanding. Third, eve though there are only two conditionals thatqualify as scalar values, the eis an ordering understanding and an extremevalue understanding in th sense that the if conditional fol1owing evenranks highest on the scale f improbability: it is naturally less likely thatwe will carry out a decisio if someone involved disagrees than it is likelythat we will carry it out f the person in question agrees. Fourth, therequirement of a contrasti e understanding - the idea that the if conditional representing the ext eme value is contrasted with all the other (lessextreme) values (if conditi nals) on the scale - is satisfied, even thoughthere are only two value : being each other's polar opposites is thestrongest possible form 0 contrast. Fifth (and finally), (34) seems torequire a scalar analysis ,ecause there is an ancillary P-scale with thepropositions 'John will no agree' and 'John will agree' as values, as wellas an even if scale with th conditionals 'We'll carry out the decision ifJohn doesn't agree' and' e'll carry out the decision if John agrees' asvalues. For all these reaso s it seems warranted to consider polar even ifconditionals as scalar. Ho ever, unlike the even if scale in "prototypical"scalar even if conditionals the even if scale in question is the extremetype in which the plural s of values is restricted to two.

A few more examples ay help to justify the suggestion that apolaranalysis is sometimes nece sary. Consider the following:

(35) [You have the New and Style Guide. - Yeah. Yeah. - I have... brought that ye h.] - Erm it is the just I think the one listthat is worth chee ing even if it doesn't appear in the newcitations. - [Mm. orth checking against the original because it,it's one that newsr om certainly will go straight to to see what,what we've said.] ( OB-S)


The meaning of the even cIause here is not 'it is the one list that isworth checking even in th ja circumstance that is mostjvery unlikely towarrant checking it' or 'it .s the one list that is worth checking even intheja circumstance in whic you would least expect it to be worth checking'. What it does mean is' hether or not it appears in the new citations,it is the one list that is orth checking'. As is clear from the use of'whether or not' in this par phrase, this is apolar rather than prototypicalscalar interpretation.

(36) [Unfortunately, ther 's no way we can avoid going up to Newcastlewithout giving offen e. We will of course have to be introduced toyoung Sam but] eve if he hadn't arrived lan's Dad and Doreenexpect to see us [ it's difficult enough to get away with onlygoing up there for a day or two]. (COB-W)

There is probably an impli it Q-clause in this sentence so th at it shouldbe read as 'even if he had 't arrived, Ian's Dad and Doreen expect tosee us AND SO WE WILL HA E TO GO' or alternatively (and equivalently),'even if he hadn't arrived, WE WOULD STILL HAVE TO GO BECAUSE Ian'sDad and Doreen expect t see us'. However, (36) does not mean thatyoung Sam's not having ar ived would have been the least likely circumstance (for Ian's Dad and oreen to have expected to see us, and) forus to have to go. lt means hat whether he had arrived or not, we wouldhave to go, because lan' Dad and Doreen expect to see us. (Thisinterpretation arises via 'if he hadn't arrived, we would still, CONTRARYTO THE EXPECTATION THAT MAYHAVE EVOKED IN SAYING THAT WE HAVETO BE INTRODUCED TO HIM, have had to go'.)

(37) [She told me that sh wiUbe doing one more course after this andthen be looking for job. 1 hope she finds one.] Even if she does,though, she will still have to move to a cheaper house. (COB-W)

It is not likely that the maning here is that 'her' finding a job is highon a scale of "negative expe tation" (improbability), "the highest positionon which is occupied by tereferent in the scope of even" (Dancygier1998: 161-162), that is, th t her finding a job is the most unlikely causeof her having to move to cheaper house (or, more generally, the mostunlikely circumstance in w ich she will have to move to a cheaper house).

(38) [Your choice will be overned by your priorities - your dedicationand practice will be uch more quickly apparent from the sevenballs route! Neverth less] you will probably find many of his tipson practising extremFly useful, even if you do not wish to followthe same directions s he has. (COB-W)

ISome truths and nontruths about even if 229

On a prototypical scalar nterpretation, this means something like: notwishing to fo11owthe same directions as he has (fo11owed)is the unlikeliest(or a very unlikely) circu stance in which you would find his tips onpracticing useful (and yo wi11still find them useful), so you wi11findthem useful IN ALL LESS U LIKELY CASES. On apolar interpretation, (38)expresses 'you might expect finding his tips useful to be inextricablylinked to wishing to fo11w the direction he has followed, but yourexpectation would be wro g: you wi11find the tips useful whether or notyou wish to fo11owthe di eetion he has fo11owed'.It is worth noting that 0 e way oftriggering apolar alternative analysis

is to put a strong accent m ij - see also Barker (1994: 254-258):

(39) a. Even IF th at i true, you cannot draw any conclusions fromit yet.

b. Even IF my lient was there, that wouldn't mean he hadanything to do with the theft.

In such sentences, the eve ij scale is determined by an ancillary P-scaleconsisting of only two pol r alternatives: P vs. not-P. Because 'iJP' rankshigher on the even ij sca e (where the highest value is the least likelyone), there is an emphasi on the suppositional (nonfactual) nature ofthe proposition expresse in the ij clause. In other words, even IFexpresses that of the two lternatives - fulfilment and nonfulfilment ofthe condition - fulfilmen is the less likely.

It should be noted that the existence of polar even ij conditionals hasbeen noted by some aut ors (e.g. Haiman 1974; König 1986; Barker1994), but that these ofte tend to classify too many even ij conditionalsas belonging to this clas . Haiman (1974) seems to treat a11even ijconditionals as polar. B rker (1994: 254) applies the notion to suchexamples as the following

(40) a. The conferene was good, even if most of the papers went ontoo long and Dw pleased the crowd.

b. She spoke to im clearly even if somewhat bluntly.

In later sections we wi11c aim that (40a) is a "purely concessive" even ijconditional and that (40b is a conditional with a "commenting" even ijclause. In our opinion, ither of them is amenable to an analysis interms of 'Q, whether or n t P'. As a matter of fact, even ij conditionalsnever receive a polar int~[?retation if they do not have an expectationunderstanding and a no~reclusive understanding. This becomes clearwhen we compare the fo11wing sentences:

(41) a. Even if it does rain, the match wi11not be cance11ed.b. Even if she ha noticed my presence, she showed no sign of it.


Sentence (41a) denies (i.e. recludes the truth of) the expectation 'If itrains, the match will be c ncelled'. Since this conditional expresses 'Pleads to Q', (41a) yields a olar interpretation ('The match will not becancelled, whether it rains r not'). By contrast, sentence (41b), whichis a nonimplicative even ij onditional - see below, does not deny thetruth of the expectation 'If he had noticed my presence, she would haveshown sign of it'. It follow that (41b) does not yield the polar reading'Had she noticed? If she ha , she showed no sign, and if she hadn't, shedidn't show any sign either'. (In fact, 'If she hadn't, she didn't show anysign' is nonsensical.f"A polar reading is also e cluded if the even ij clause is specificational,

as in I will do it {even ij J HN is there / even if it is John who is there}.The appropriate paraphras here is not 'I will do it if John is there, andI will do it if John is not here' but rather 'I will do it if (it is) John(who) is there, and I will d it if SOMEONE ELSE (e.g. Bill, Jack, William,Derek, etc.) is there'. (In 0 her words, there is reference here to a scaleof improbability whose va ues are determined by an ancillary P-scaleresulting from specification 1 focusing.)

4.5. Even if precludes the i iplicature that P is necessaryfor Q

It is well known in the ling istic literature that, failing any indication tothe contrary, 'ij P, Q' tend to be read as 'Q if and only if P' (= 'Q iffP'). This tendency (implic ture) is known as "conditional perfection"(Geis and Zwicky 1971): t e pragmatic interpretation of ij is often suchthat a one-way conditional ('if P, Q') is "perfected" into a biconditional('if P, Q' + 'if not-P, not-Q' . In other words, 'if P, Q' is often interpretedas 'iff P, Q' (= 'Q only if P' . For example, Ij the weather isfine, we'll goto the seaside is convention lly interpreted as having ' ... but we won't goif the weather isn't fine' a part of its meaning. In other words, theunmarked pragmatic interp etation is 'If and only if the weather is finewill we go... ', in which th weather being fine is seen as a NECESSARY

condition for our going to the seaside. Like all conversational implicatures, this implicature can e cancelled by the context, for example bythe addition of In fact we'[ go to the seaside anyhow, even ij it rains -see Haegeman (1984). (Th implicature can also simply fail to arise, asin Mother wil! be happy ij ou tel! her that, which does not suggest thatMother will not be happy i she is not told the news in question.)Van der Auwera (1997) and Hom (2000) have suggested different

explanations for the origin f the conditional perfection implicature. Our


own explanation for the j plicature, which we will henceforth refer toas the "necessity implicat re," is as follows.ê" Consider (42):

(42) If it doesn't rain ne t Sunday, John'll drive to the seaside.

The eventual truth (in th actual world) of the assertion Next SundayJohn' II drive to the seaside depends on a variety of pragmatic conditions:John is still alive next Su day, he is not seriously ill, he has a car at hisdisposal, the seaside is wit in driving distance, John can drive a car, etc.When uttering (42), all tese conditions are pragmatically presupposedto be fulfilled. The condit on referred to in the P-clause is just a furtherpragmatic condition. How ver, unlike the other conditions, th is conditionneeds mentioning becaus it is not pragmatically presupposed to befulfilled. This means that, although strictly speaking the P-condition isnot sufficient for Q in tha there are other conditions to be satisfied, it isinterpreted as the only rel vant condition, that is, as the only pragmaticcondition that is not self-e ident, since all the other conditions are pragmatically presupposed to be fulfilled. In accordance with the Griceanmaxims of quantity and r[levance, the P-condition is the only one thatneeds mentioning, because it is the only one that presents necessary (new)and relevant information. It follows that it is interpreted as a sufticientand necessary condition, s~nceall the other conditions are disregarded.

In sum, in the actual w~rld, absence of rain is not the only pragmaticcondition for driving to the seaside and is neither a sufficient nor anecessary pragmatic condition for doing so. However, the fact that it IS

mentioned as a condition, and that it is the ONLY condition mentioned,implicates that this conditi n can be seen as both sufficient and necessary:if only one condition is ~entioned, the hearer has a right to concludethat there is in fact only on1e(relevant) condition, hence.that the conditionmentioned is the only ne ,essary and sufficient condition for Q. In theabsence of any indication 0 the contrary, (42) will therefore be interpreted as meaning that absencb of rain on Sunday is a necessary and sufficientcondition for the proposition 'We drive to the seaside' to become true inthe actual world.Perhaps it is worth noti g that we are arguing here th at the sufficiency

understanding is not part of the semantics of 'if P, Q', as is universallyaccepted in the literature, ut just an implicature. Our claim is corroborated by the fact that the s ciency understanding is cancellable:

(43) If it rains, we'll sta inside - provided the headmaster agrees, ofcourse. (Rain is no the only condition here that must be satisfied;it is therefore not a ufficient condition.)


Another point worth ma ing is that the necessity and sufficiency implicatures only play a part in t e interpretation of implicative ij conditionals.Neither of the implicatures arises in nonimplicative ij conditionals, suchas the following (already re erred to as [2d]):

(44) Icouldn't do a crime if Iwanted. (This sentence does not implicatethat my wanting to do a crime is a sujjicient orjand necessarycondition for my bein able to do one.)

Moreover, even in implicat ve conditionals, the implicatures may fail toarise because they are bloc ed by pragmatic knowIedge:

(45) [What shall we give Susan for her birthday?] - She would behappy if she was giv n a necklace.

Since it is natural that there are many different circumstances in whichpeople may be happy, the ecessity implicature 'Susan would be happyonly if she was given a ne klace' (= 'Susan would not be happy if shewas not given a necklace') oes not arise: it is blocked by our pragmaticknowledge of the world.Returning to our main line of argument, we can point out - with

König (1986: 236) and Deygier (1998: 163) - that the addition ofeven to an implicative ij c nditional blocks the necessity implicature.Even has this effect irresp ctive of whether Pand Q are positive ornegative. For example,

(46) a. Jane will help y u if she has the time.b. ?Jane will help y u even if she has the time.

(47) a. Jane won't hel you if she doesn't have the time.b. ?Jane won't hel you even if she doesn't have the time.

(48) a. Jane will help y u if you don't treat her badly.b. ?Jane will help y u even if you don't treat her badly.

(49) a. Jane will not h lp you if you treat her badly.b. ?Jane will not h lp you even if you treat her badly.

In each of the (a) examples, ijis by implicature interpreted as "iff." Ineach of the (b) sentences, th addition of even blocks this reading. (Theseeven ij sentences are odd b cause the sense of unexpectedness inducedbyeven is hardly compati le with the contents of the conditional towhich even is added: our kn wledge of the world tells us that having thetime is not a very unlikely condition for helping someone but in factvirtually a necessary conditi n for doing so; similarly, we do not normallyhelp a person unless that p rson does not treat us badly.)


Dancygier's (1998: 163) explanation for this blocking effect of even isas follows. According to er, the necessity implicature typically ariseswith causal conditionals: ince (in her opinion) even ij conditionals arenever causal (i.e. they me n 'Q despite P' rather than 'Q because of P'),they do not yield the nec ssity implicature. However, we have arguedthat implicative even ij co itionals do have a kind of causative meaning,since they necessarily have the nonpreclusive understanding 'even P doesnot prevent Q'. The folIo ing examples show clearly that it is wrong toassume that even ij condit onals are never causal:

(50) a. [I'llieave not nly if Bill comes but also if Jim comes.] 1'11even leave if G rdon comes. (adapted from Barker 1994: 251)

b. Even if you dri k (just) a drop, you'll get fired.

(SOa) yields a causal inte pretation ('even Gordon's coming will causeme to leave') but there is 0 necessity implicature (i.e. no understandingthat I willieave ONLY IF ordon comes.) Similarly, in (50b) drinking adrop clearly causes the Q- ituation, but the implicit contrast of "a drop"with other quantities that are more obvious (i.e. expectable, probable)candidates to satisfy the v riabie X in 'if you drink quantity X you'll getfired' rules out the necess ty implicature. Moreover, the claim that thenecessity implicature requ res a causal reading is disproved by exampleslike If any of these childre is wearing an Air Force badge, he or she hasa parent in the Air Force. he 'Q because of P' interpretation is certainlyruled out here (at least if he sentence is to be interpreted as true in theactual world), but the nee ssity implicature is not: the sentence could beread as 'It is only if these children are wearing an Air Force badge thatthey have a parent in th Air Force' (which may actually suggest 'Pbecause of Q' rather than the other way round).25

In our opinion, König's 1986) explanation is the correct one: "concessive conditionals ( ... ) ex lude conditional perfection as an admissibleinference pattern by entai ing or presupposing that the conditional relationship holds for a whole series of antecedents" (1986: 236) In ourterminology, this means t at the "if and only if" interpretation of ij ineven ij conditionals is inco patible with the plural set understanding thatforms part of the scalar eaning: 'even ij P1, Q' implies that there areone (in polar even ij co ditionals) or more (in prototypical even ijconditionals) values of th type 'iJP, Q' on the even ij scale. This meansthat P1 is not the only co dition for Q, hence that it is not a necessarycondition. This explanati n is in keeping with the observation that thenecessity implicature neve arises in nonimplicative conditionals, whichdo not allow a scalar anal sis.

234 R Dec/enk and S. Rfed

4.6. Nonassertoric implica ive even if conditionals

So far we have only consid red even ijconditionals in which the Q-clauseis a (positive or negative) ssertion. However, the Q-clause of an implicative even ij conditional ay also be interrogative or directive. Thefollowing illustrate a first ossibility:

Win I really be ~unished even if I do not do anything wrong?[Cleaning-up tit:Help prevent spins by placing the mixingbowl on a folde , damp tea towel, ... Put a plastic sheet underthe table and s eep food up before it gets trodden aroundthe house. And even if you hate washing up, don't assumethat your child Ivvin- [make clearing up afterwards part ofthe fun.] (COB-W)

(5Ia) is a wide-scope polar question, which is interpreted as 'Is it reallytrue that I win be punished even if I do not do anything wrong?' This isju st the interrogative versi n of an assertoric implicative even ij conditional: it has both an exp ctation understanding ('I expect not to bepunished ifI do not do anyt ing wrong') and a nonpreclusive understanding (Will my not doing a ything wrong not preclude my being punished?'). This kind of interr gative even ijconditional does not pose anyparticular problems. The s me is true of (5Ib), which differs from anassertoric implicative even ij conditional only in that the Q-clause is adirective: (51b) has both an expectation understanding ('You mightexpect that you can assum that if you hate washing up, your child windo too') and a nonpreclusi understanding ('That expectation is wrong:don't assume that your chi d win hate washing up if you do').The following sentences . lustrate a second possibility:

(51) a.b.

(52) a. (speaking about a two-party state) Even if party A becomesmore popular in the city, party B will win the elections.

b. Even if party A ecomes more popular in the city, win partyB win the electi ns?

c. Even if party A becomes more popular in the city, vote forparty B!

(52a) is an implicative even i conditional with an expectation understanding ('You might expect par y A to win the elections if it becomes morepopular in the city') and a ns npreclusive understanding 'That expectationis wrong: party B win win the elections'). Sentence (52b), which is apolar question with narr w scope, is the interrogative counterpartof (52a): it has the same xpectation understanding and a nonpreclusive (expectation-cancellin ) understanding ('I am questioning this


expectation'). Sentence (5tc) is the directive counterpart of (52a): it hasthe same expectation un erstanding and a nonpreclusive (expectationcancelling) understandin ('Act counter to the expectation: vote forparty B').The following illustrate a third possibility:

(53) a. Even if I take his apple and eat it, who shall suffer?b. [That damned fool boy!] What does he mean by trying to

come home nOf' even if he did promise? (LOB)

These sentences are rheto .cal questions, which are interpreted as if theywere assertions. They the efore present no problem at all, because, asusual, they are interpreted in terms of an expectation understanding anda kind of nonpreclusive u derstanding: 'Surely, my eating this apple willnot preclude (= alter the fact) that nobody will suffer, although youmight expect the contrary, and 'You might expect that I'd like him tocome home if (= now tha ) he promised, but that expectation is wrong:I don't want him trying t come home now'.The following illustrate a fourth possibility:

(54) a. ["Order! Orde !" - "Don't interrupt me, damn you! Youknow he's my usband and that I'm his wife.] Even if he's tobe unknown as usual- do I get my pension?" (TUS 71)

b. [The older you are the less strenuous the exercises should beoBut] even if y 're over seventy, do try to get your musclesmoving. (LOB)

These even ifsentences ca~ be analyzed as speech-act conditionals" withboth an expectation und~l.standing and a nonpreclusive understanding.Thus, (54a) is interpreted j~'You might expect me not to ask the questionDo I gel my pension? if my husband is to be unknown, but the [act thathe is to be unknown actu .lly does not prevent me from asking if I willget it'. The directive (54b) can be analyzed similarly: 'You might expectme not to tell you to tr to get your muscles moving if you're overseventy, but in fact your eing over seventy does not prevent me tellingyou to do so'. These sen ences show that nonassertoric Q-clauses areunproblematic if 'preclude Q' can be interpreted as 'preclude mysayingjaskingjordering th It Q'.27A final possibility of an i plicative even if conditional with an interrog

ative Q-clause is illustrated by the following:

(55) ["She could already edead, couldn't she?" - "You have to thinkpositively, Nicole." - "And even if she is still alive, what willhappen to her if De- nison's thugs get to her first?" (COB-W)

236 R Declerck and S. R ed

This is the only type of n nassertoric even ij conditional that appearsdifficult to account for in rms of an expectation understanding and anonpreclusive understandi g. The sentence as a whole would not seemto be a speech-act even ij conditional. The Q-clause looks like a true(nonrhetorical) wh questi n. There are only two polar values on theancillary P-scale (and thus, n the even ij scale): 'If she is still alive' and'If she is no longer alive'. owever, there is no ensuing polar interpretation of the kind we have bserved in (34) (= [J hope John wil! agree.But) even ij he doesn't, Wf'll carry out this decision) because there isapparently no expectation understanding and no nonpreclusive understanding. Sentences like (5 ) would therefore seem to defy our analysis.However, the solution to tl e problem becomes clear when we compare(55) with its assertoric courerpart:

(56) And even if she is st 11alive, something (bad) will happen to herif Dennison's thugs et to her first.

Here too the expectation nderstanding and the nonpreclusive understanding would seem to be acking, but in fact they are not, because (56)has an implicit Q-clause:

(57) And even if she is s ill alive, she is hardly better of! than ij she'sdead, because somet ing (bad) wi11happen to her if Dennison'sthugs get to her first.

In the same way, (55) is hort for

(58) And even ifshe is stil alive, wil! she be better of!than if she's dead?Because something ( ad) will happen to her if Dennison's thugsget to her first.

The Q-question here prese ts no problem because it can be analyzedeither as a rhetorical qu stion or as the Q-clause of a speech-actconditional like (54a).

5. Nonimplicative even ifc nditionals

Nonimplicative even ij contions, P does not precludefollowing are some typical

itionals do not express 'contrary to expecta, and do not allow a scalar analysis." Thexamples:

(59) a. But they don't rally look like musicians, even if, occasionally,they sound like i . (COB-W) (This isnot interpreted as meaningthat the conditio al 'They don't really look like musicians ij,


occasionally, th y sound like it' is a value which contrasts withother condition Is on an "even ij" scale.)

b. But he is cleve and a talented researcher even if, in parts ofthis passionatel angry book, the talent is ill-used. (COB-W)

c. However one oes appreciate that the job of prime ministeris not one with a huge amount of long-term security, even if,once they have been removed from office, very few of themare actually to be seen queuing for Salvation Army soup.(COB-W)

d. [The appearan e of three new books on the subject ( ... )proves that.] e n if philosophers want no part of the action,publishers kno a good thing when they see it. (TLS)

These even ij conditional can be subdivided into two groups: even ijconditionals with a "purel concessive" P-clause and even ij conditionalsin which the P-clause expr sses some kind of comment on the Q-clause.

5.1. Purely concessive eve if clauses

According to Quirk et al. (1985), "concessive clauses indicate that thesituation in the matrix is ontrary to expectation in the light of what issaid in the concessive clau e.... Often they also imply contrast betweenthe situations described b the two clauses" (1985: 1098) It is clear thatthis definition covers all e en ij conditionals, since (as we have seen) asense of expected incomp tibility (and hence of contrast) is an inherentpart of the semantics of ven. In implicative even ij conditionals, thissense appears as a sense f unexpectedness (combining an expectationunderstanding and a nonp eclusive understanding). Nonimplicative evenij conditionals do not yiel this sense of unexpectedness: the only aspectof meaning expressed byeven is now the general sense of expectedincompatibility. However, hen if is used in a contest triggering a concessive reading, as in If she's i l-mannered, at least she's honest, the combination with even, with its s nse of expected incompatibility, results in a"purely concessive" noni plicative even ij conditional. For example,

(60) Even if politicians ere not interested in this subject, the mediawould still be payin a lot of attention to it, [because they knowthe general public is interested in that kind of subject].

This counterfactual even i sentence, which is similar to (59d), yields apurely concessive interpre ation, which involves a sense of contrast butwhich is clearly not scala and does not involve the expectation andnonpreclusive understandi Igs: (60) cannot be paraphrased as 'You might

238 R Declerck and S. J ed

"peet that if politicians1"not interested in thc subject, that wouldpreclude the media paying a lot of attention to it, but in fact the mediawould still be paying a lot of attention to it'. Such an interpretation isblocked by our pragmatic nowledge of the world, as everybody knowsthat what the media are i erested in is in no way determined by whatpoliticians are interested in 29

Apart from counterfact al clauses, even ij can also intro duce purelyconcessive clauses with nother kind of possible-world meaning.Consider, for example, the following even ij conditional:

(61) Even if we grant All Janik's point that Weininger was an intense,sineere young man n search of a truly "moral science," clearlysomething went bad wrong with his thought-experiment. (TLS)

As argued in Declerck and Reed (forthcoming), this kind of conditionalhas a "neutral" meaning i that it assumes nothing in conneetion withfulfilment of the conditio in the real world. In this respect it is likeComrie's (1986: 90) exam~.le If the butler had done it, we would havefound just the c/ues that we did in fact find. It is also a clear example ofa purely concessive even ij onditional. A suitable paraphrase is 'Clearly,something went badly w~ong with Weininger's thought-experiment,although we might grant I, llan Janik's point that Weininger was anintense, sineere young ma in search of a truly "moral science." (Assaid before, the absence 0 a sense of unexpectedness, and hence of anexpectation understanding, does not alter the fact that even does inducea sense of expected inco patibility: the fact that Weininger was anintense, sineere young ma in search of a truly "moral science" seems[but, the author insists, is ot] incompatible with the fact that he madea faulty thought-experimen .)In the following exampl , the even ij clause is interpreted as "open"

(i.e. the condition is seen a one that mayor may not be fulfilled):

(62) [If by chance he does not kill me I shall be crippled for life.] Evenif he fails to kill or ~ripple me I shall be permanently disfigured.(LOB)

Neediess to say, this is no interpreted as 'You might expect that if hefails to kill or cripple me shall not be permanently disfigured, but infact I shall be permanently disfigured'. By contrast, the sentence can beread as 'I may not be killed r crippled by him, but I shall be permanentlydisfigured'. This interpretat on is not implicative but purely concessive.In the following examplei of purely concessive even ij conditionals the

P-clause is "tentative," tha is, the truth of P, or the actualization of theP-situation, is represented snot impossible, but rather unlikely.


(63) a. [He's in prison sol he couldn't come to our party even if hewanted to.

b. [Edward Augu tus Freeman ( ... ) told Englishmen that TheAnglo-Saxon ( r, as he preferred, "Old English") Chroniclewas "the book you should reverence next after your Biblesand Homer."] e would probably be satisfied th at its readership has stood up at least as well as either of theirs; even ifhe would not h ve been so pleased that a great deal more ofit is in New En land than in Old. (TLS)

Although in these sentenc s there is a sense of expected incompatibilitybetween Pand Q (as there always is when even is used), neither of themhas the expectation unders anding 'You might expect P to result in not-Q'(in other words: 'You mig t expect P to preclude Q'). Thus, the followingdo not appear to be part f the interpretation of (63a)-(63b):

(64) a. You might exp ct th at his wanting to come to our party wouldresult in his bejng able to do so. (Q is [-q}J

b. You might exp ct that the fact that he would not have beenpleased that ost of the readership of the Old EnglishChronicle is in few England would preclude his being satisfiedthat its readerF:SiPhad stood up as least as well as that of theBibles and H er, but this fact would actually not haveprecluded his eing satisfied that this was the case. (Thissentence is al! right in itself, but it is not an appropriateparaphrase of [ 3b IJ

There are also purely con essive even ij sentences in which the P-c1ausereceives a "factual" inter retation, that is, in which P is taken to beunconditionally true in th actual world:

(65) But they don't rea ly look like musicians, even if, occasionally,they sound like it. (I OB-W)

There is no expectation u derstanding 'You might expect that it wouldfollow from the fact that they occasionally sound like musicians thatthey look like musicians', 3 nor a nonpreclusive understanding 'soundinglike musicians actually d es not prevent them from not looking likemusicians'. There is also 0 scalar reading that sounding like musiciansis the most unlikely circu stance in which "they" would not look likemusicians (nor a polar rea ing 'they do not look like musicians, whetheror not they sound like m sicians'j." Sentence (65) simply means 'Theymay (occasionally) sound ike musicians, but they do not really look likemusicians', which is a pur ly concessive reading.


The following are simil r purely eoneessive examples with a faetualeondition:

(66) a. [I did it beeau e I think] I had the sheer gumption to getthrough and enjoy my life, even if it has taken me all thistime to finally e me to terms with the parts of myself I wasn'ttoo keen on. (C B-W)

b. [Roehdale FC a e not an anaehronism, though. They are thesame as Liverp ol only in miniature.] We have a new stand,exeeutive boxes an eleetronie seoreboard (even if it doesn'twork). [We eve have our own fiag.] (COB-W)

Let's close off this seetion n purely eoneessive even ij eonditionals withthree further remarks. Firs , purely eoneessive even ij eonditionals differfrom implieative ones in th t the relation between Pand Q eannot be arelation between Pand an i plieit Q rather than between Pand the overtQ-clause. Consider the foll wing:

(67) [The new vicarage w sa three-bedroomed affair, built in the loealstone, but in a style best described as anonymous. "It just suitsus," Walter Primp oid everyone.] And even if Thyrza Primpregretted their two-p as-in-a-pod grandeur in the old viearage, shehad by then gained n unassailable position in the town, [and hernatural parsimony r conciled her to her redueed state]. (adaptedfrom COB-W)

At first sight this even ij co I ditional would seem to be purely concessive,as it invites the paraphrase 'Thyrza Primp might have regretted their ...grandeur in the oid vicarag , but she had by then gained an unassailableposition in the town'. In fa t, however, (67) is more Iikely to be interpreted as au even ij condition 1with an implicit Q-clause:

(68) Even if Thyrza Prim regretted their two-peas-in-a-pod grandeurin the old vicarage, he had become reconciled to living in a morehumble vicarage bec use she had by then gained an unassailable position in the town, [and because her natural parsimonyreconciled her to her reduced state].

This interpretation is neces arily implicative, not purely concessive.Second, it is interesting 0 see that a purely coneessive even ij clause

may have the effect of ean elling (or blocking) an implicature that theQ-clause yields when used ir isolation. Consider (69):

(69) Her betrothed wouid have learned her reason one day, she says,even if it proved to bb on Judgement Day. (TLS) (= IJ things hadnot fallen out as the did, her betrothed would (still) have learned


her reason one day, even ij the learning (had) proved to be onJudgement Day.)

Here the even ij clause ge s rid of the reasonable assumption that "oneday" means at its broades "one day within his life." This implicaturecancelling effect of even ij' s also apparent in the following:

(70) [But in his meeting ith council ( ... ) the Iraqi ambassador refusedto commit to the r solutions. He called them extraneous issues,and he said] even i Iraq wanted to comply with them it couldtake many years to do so. [For instance, Resolution 674 calls forIraq to pay war rep rations to Kuwait. But with its "war-ravagedeconomy," he sug ested, "Iraq is in no position to do so."](COB-W)

Clearly, the ambassador ants to say that the resolutions are unreasonable, but he does not wan this to suggest that Iraq has any intention ofcomplying with the resolut ons - a suggestion that the Q-clause (it couldtake [Iraq) many years to omply with the resolutions) on its own actuallydoes make.FA third, and final, rema

sive even ij and even thouconcerns the relation between purely conces. Consider the following:

(71) a. [It was the Ion liness of the neighbourhood, they supposed,that kept the h use next to theirs empty. ( ... ) The house stoodtwo hundred y rds from the Bartlebys' and] A. liked lookingout of the win ow now and then and seeing it, even if it wasempty. (quotat on from aPatricia Highsmith noveI, cited byKönig 1986: 24 )

b. "Kubla Khan,' even if originally scribbled down in a frenzyof recollection, was carefully revised for publication. (TLS)

There is little difference 01 interpretation between these even ij sentencesand the following:

(n) a. A. liked lookinf out of the window now and then and seeingit [the house], ven though it was empty.

b. "Kubla Khan, ' even though originally scribbled down in afrenzy of reCOliction, was carefully revised for publication.

There is a minute differen e between (71a) and (na): while (71a) is fromthe point of view of A., ( 2a) is from the point of view of the narratoror, rather, is authorial. (P rhaps [na] COULD be read as from A's pointof view, but that is not thtj obvious reading, and it still differs from [71a]in that [71a] CANNOT be anthorial). This seems to be to do with the factthat while the though clauJe in (na) is a statement of a fact, the ij clause


in (71a) is not: it expresse a closed condition, that is, a condition thatis assumed to be fulfilled. And closed conditions, which are typicallyechoic (cf. Akatsuka 1991 ,33 are discourse-oriented - they cruciallyinvolve an addressee, even if the addressee is the self, as in (71a), whichwe understand as A. thinki g to himself "The house is empty, but I likelooking out of the window and seeing it, even if it IS empty." Similarly,we can paraphrase (71b) ith a closed conditional, which strongly sug-

scribbled down in a frenzy ( ... ) but it was carefully revised for publication', or 'I agree that K was originally scribbled ( ... ), but still, itwas ( ... )'. Sentence (nb), by contrast is purely authorial: there is noindication of echoic qualitijS.

In sum, while (na)-(nb are formally identical to (71a)-(71b), exceptthat even though is substit ted for even ij, their interpretations are notquite identical. This mean that substituting even though for even ij inpurely concessive even ij onditionals is not often possible without achange (albeit a minute 01e) of interpretation. The reason is that eventhough can only introduce ~lauses with a factual meaning. By contrast,even ij introduces an ij con itional, and, as noted above, ij conditionalsmay be factual, as in (65) nd (66a)-(66b), but they more often have anonfactual (i.e. closed, opel' tentative, neutral, or counterfactual) meaning. In the latter cases even though is not a valid alternative to even ij:

(73) a. [He's in prison, sol he couldn't come to our party even if hewanted to. (*ev though)

b. Her betrothed ould have learned her reason one day, shesays, even if it proved to be on Judgement Day. (TLS)(*even though)

5.2. Commenting even if c uses

According to Barker (1991:: 254), even ij conditionals can receive a'whether-or-not-P' interpre ation (in our terms: a polar reading) evenwhen the P-clause is factu 1, as in the following examples (= Barker's[8a]-[8c], 1994: 254): t(74) a. The conference as good, even if most of the papers went on

too long and fe pleased the crowd.b. Women have ah~ays kissed women, even ifmore cynical males

describe the pr~cess as being reminiscent of prize fightersshaking hands. (\TAM 12)

c. She spoke to hi clearly even if somewhat bluntly.


Barker claims not only th t these sentences mean 'Q whether or not P,though I think that P' bu also that this paraphrase is equivalent to 'Qalthough P' (1994: 254). I our opinion, neither of these claims is correct.Thus, (74a) does not recei e the polar interpretation 'Either way, whethermost of the papers went n too long or not, the conference was good',and the 1atter paraphras is not equivalent to 'Although most of thepapers went on too long, the conference was good' (which paraphrasedoes seem closer to the i terpretation of [74a]). Similarly, Barker attributes a polar reading to (74b), but the reading 'Women have alwayskissed women, both if more cynical males describe the process as beingreminiscent of prize fighte s shaking hands and if they do not' is bizarre,and it is certainly not equ valent to 'Women have a1ways kissed womenalthough more cynical ma es describe the process as being reminiscent ofprize fighters shaking han s'. Similar remarks apply to (74c).

In our opinion, (74a)- 74c) are not implicative (as they would be ifthey were polar). Sentence (74a) might be considered as purely concessive,but we are not really eert in that even if is interpreted as despite the factthat. Sentences (74b )-(74 ) are definitely neither implicative nor purelyconcessive. As a matter 0 fact, they do not belong to any of the typeswe have distinguished so ar. They illustrate a type (not discussed in theliterature) in which the ev n if c1ause has a "commenting" function, thatis, represents the speaker' comment on what is said in the Q-c1ause.

Let's consider (74c) firs . This sentence expresses little more than

(75) She spoke to him c early. She did so somewhat bluntly.

except that the use of even n (74c) adds the idea of expected incompatibility. This means that P rep esents some sort of (at least potential) detraction from the validity of (or of uttering Q). This aspect of meaningexplains why the coordina ion of two manner adverbials (e.g. clearly andbluntly) by means of even ifïs only acceptable if the two ideas expressedcan be expected to be inc mpatible with each other:

(76) a. She spoke to I im c1early {and / *even if} at just the rightvolume. ("eve if" is odd because there is no easily imaginableincompatibility between clarity of speech and speech at just theright volume)

b. She spoke to h m clearly {and / even if} somewhat {loudly /quietly}. ("eve if " is OK ifwe interpret "loudly / quietly" as"too loudly / uietly," which detracts a bit from the positiveevaluation in " he spoke clearly" r4

c. She spoke to im slowly, even if somewhat quietly. (This isOK if the slow ess is a positive factor and the quietness is a


negative factor t at nevertheless does not cancel out the positiveone, for exampl ij she is speaking to someone in a languagehe finds difficult to understand, in which case slowness wouldmake it easier vr him to understand her, but quietness mightmake it a little arder to understand her.)

d. He spoke brut lly, even if {*kindly / *slowly / ??clearly /honestly}. ("br tally" and "kindly" cannot be seen as compatibie; "slowly" an "clearly" simply cannot be made relevant asmanners of spee h that might be seen (potentially) to detractfrom the brutalit of someone's speech; but "honestly" is something positive th t can be set against the negative "brutally. ")

As for (74b), this sentence is similar to (74c) in that the even ij clausehas a commenting functio . It is therefore equivalent to

(77) Women have always kissed women. More cynical males describethe process as being eminiscent of prize fighters shaking hands.

except that even in (74b) a ds the idea of expected incompatibility. Thismeans that (74b) differs fro (77) in that it expresses a logical conneetionbetween the two clauses, vit. that P might be thought by some to detractfrom (the validity, felicity, ,tc., of) Q but in the speaker's view does not.In the following example where the even ij clause again has a comment

ing function) the connectio is of a different kind:

(78) The King Bolo lyri, s have been known to Eliot scholars andenthusiasts for some years now, even if only from hearsay. (TLS)

In this case the following s ems a reasonably suitable paraphrase:

(79) The King Bolo lyri s have been known to Eliot scholars andenthusiasts for som9 years now, but sometimes they have onlybeen known from he rsay.

The meaning of but in the paraphrase (79) is not so much adversativeas "downtoning": the com enting P-clause weakens (an element of) theinterpretation of the Q-cl use by denying one of the implications orimplicatures that could be r ad from the Q-clause. When used in isolation,the sentence The King Bolo lyrics have been known to Eliot scholars andenthusiasts for some years n w suggests that Eliot scholars and enthusiastshave had first-hand access 0 the text of the King Bolo lyrics for someyears. The commenting eve ij clause says th at for at least some scholarsand enthusiasts this has no been the case, and thus has a downtoningeffect on the interpretatio I of the Q-clause. (The but clause in [79]actually has a slightly diffe ent downtoning effect. While [78] means 'Q,DESPITE some small downto ing', [79] means 'Q, but not quite as fully Q


as you might have thoug t from just "Q" '. However, this just showshow difficult it is to find paraphrase that receives EXACTLY the sameinterpretation as the sente ce it is supposed to paraphrase.)A similar downtoning a alysis might be suitable for (74b): in isolation,

the sentence Women have lways kissed women suggests that women havealways shown genuine sig s of affection for each other, because that iswhat kissing is generally ssumed to be. The addition of even ij morecynical males describe the process as being reminiscent of prize fightersshaking hands questions t is implicature by suggesting that the kissing isseen by some as only don perfunctorily.In the following exampl , the commenting even ij c1ause is downtoning

in that it adds a somewha negative qualification to a positive idea.

(80) She's one ofthe best students, even ifnot the BEST, we've ever had.

Interestingly, this sentenc is downtoning ('I can certainly say she's oneof the best, even though I m not sure I can say she's the best'), whereasthe following (without eve , with the nuc1ear accent on the, and withoutthe heavy "parenthesis" intonation of the even ij clause of [80]) isupgrading ('She might not just be one of the best, she might be THEbest'):

(81) She's one of the be t students, if not THE best, we've ever had.

A negative commenting e en ij c1ause mayalso serve to block the possibility of an excessive inte pretation of a negative idea expressed in thehead c1ause. In that case the even ij clause is upgrading rather thandowntoning:

(82) At this time of ye ,Detroit is cold, even if it is not downrightfreezing.

In isolation, Detroit is col does not rule out (even if it does not invite)the interpretation that it is reezing in Detroit. The even ij c1auseprecludesthis reading by saying tha the temperature is higher than that.

Conversely, even ijhas "downgrading" connotation in commentingeven ij clauses that are eant to soften the speaker's immodesty inuttering a self-congratulat ry Q-clause:

(83) ["Because we are a small company, everything is overseen by meor my beloved hus and," she writes,] "and it all runs very efficiently, even ifI sa so myself!" (COB-W)

The even if c1ause here is not downtoning, because it does not weakenthe validity of Q, but it is owngrading in that it renders the Q-utterancemore tentative.


Another type of comme ting even ij clause consists of those that wewin refer to as "evaluating " We can distinguish three subtypes: "truthevaluating," "content-eval ating," and "presupposition-evaluating."Truth-evaluating even ij lauses express a comment that concerns the

degree of truth of the Q-cl. use, or the authority behind it. This kind ofevaluating even ij clause h s a downgrading connotation.

(84) The murderer has ap arently been arrested, even if the news hasn'tbeen confirmed yet y the authorities.

A second subclass of eva uating even ij clause is "content-evaluating."It is typical of these that they express the speaker's comment on thesemantic content of the Q- lause:

(85) [You know, that kni e's somewhere, isn't it? The knife that someone stuck into McClure. The knife that BROOKS stuck intoMcClure. That's thelinfuriating thing for me.] Knowing that thebloody thing's SOME*HERE, even if it's at the bottom of the canal.(DOC 133)

Here the Q-clause says th t the knife must be somewhere. The even ijclause adds to this the com ent that "at the bottom of the canal" is theleast likely, or the least ace ssible, place for the knife to be, and in doingso makes it clear that "so I ewhere" in this instanee cannot just includeobvious and easily accessib e places. This may imply that there is somenotion of scale involved in the interpretation of the sentence, but this isnot the normal improbabil ty scale that is typical of implicative even ijconditionals, as is clear fror a comparison with (86):

(86) We'll find the blOOdYIthing, even if it's at the bottom of the canal.

This sentence has the usual ,xpectation understanding ('You might expectthat the bloody thing win nEt be found if it's at the bottom of the canal')and the nonpreclusive understanding ('lts being at the bottom of thecanal win not preclude us fi$ding the bloody thing'). There are no similarunderstandings in (85), whifh implies neither 'You might expect that thebloody thing isn't anywhere if it's at the bottom of the canal' nor 'ltsbeing at the bottom of the danal win not preclude the bloody thing beingsomewhere'. In other word, if there is a scale in (85), it is not an evenij scale whose values are c nditionals expressing various kinds of P-Qrelation. Rather, the scale i one of (in)accessibility of places included inthe reference of somewhere, as opposed to being a scale of P-situationsthat win pro duce the Q-sit ation.The following is anothe example with a content-evaluating even ij

clause:


(87) I think about religi n ( ... ) as the exercise of law as applied byeach man to hims f, even if that law be anarchy, negation ordespair. (WOS 72)

There might be a case f r arguing that what the even ij clauses in(86)-(87) evaluate are th contents of the words somewhere and law,respectively. This would ean that the even ij clauses express a metalinguistic comment. However we will not commit ourselves to this analysis,because it may weIl be tha in (87) it is not the way(s) in which the wordlaw can be used th at is at i sue, but the types of law that can be imagined.Similarly, in (86) there m ght not really be a query about the meaningof the word somewhere ut rather a concern that the strategies fordelimiting its reference on this occasion are deployed in the right way.Apart from truth-evalu ting and content-evaluating commenting even

ij clauses, there is a third ubtype of evaluating even ij clause, which wewill refer to as "presuppo ition-evaluating." It is typical of this subtypethat the comment express d in the even ij clause does not concern thesemantic content of Q as a hole, but only the justifiedness of a pragmaticpresupposition of the Q-p oposition: the P-clause questions, cancels, orotherwise comments on this pragmatic understanding. For example,

(88) ["How could I hav closed the door? It wasn't open." - "I don'tcare.] You should h ve closed it even if it wasn't open."

The phrase close a door 0 ly makes sense if (and therefore pragmaticaUypresupposes that) the door in question is open. In (88), where the speakeris being unreasonable (pr bably in order to avoid admitting that he isin the wrong), the even ij lause runs counter to this presupposition.

In this conneetion we sh uld also refer to the observation made in conneetion with (69)-(70) ab ve that even ij clauses may cancel or block apragmatic implicature. Sud even ijclauses combine a concessive meaningwith a commenting one. he foUowing is a further example:

(89) John did a good job clearing away the glasses after the party, evenif those on the win owsill up there escaped his attention.

As argued in Declerck (19 7), if John cleared away the glasses is used inisolation, the is by implic ure interpreted as al! the. In (89), the even ijclause cancels this implica ure of inclusive reference.A final note on comme ting even ij clauses is that purely concessive

even ij clauses that receive a factual interpretation might be argued to beeven ij clauses commentin on the speech act of uttering Q. This pointof view is actually defende in Sweetser's (1990: 138) "speech-act analysis" of (what we caU) pu ely concessive even ij, according to which a


purely concessive even ij se tence means "I say (insist ) that X, even if Iadmit that Y":

(90) These animals are n t mammaIs, even if they are viviparous.

However, since a speech- ct comment on the Q-clause is necessarilyconcessive, it seems to b an academie question whether we shouldconsider the factual even i clause of (90) as purely concessive or as acommenting even ij clause expressing a concessive speech-act commenton Q). This does not alter he fact, though, that the distinction betweenthese two types of nonim licative even ij conditional is often relevant.The sentence [He's inpriso , so] he couldn't come to our party even ij hewanted to does not mean 'I say that he couldn't come to our party, evenif I admit that he wanted t ' (= Sweetser's "I say [insist] that X, even ifI admit that Y"). In fact we have already noted that many purelyconcessive even ijcondition Is resist Sweetser's speech-act analysis - seenote 31.

6. Conclusion

In this artiele we have at empted to describe the various aspects ofinterpretation of even ij co ditionals. The following have been our mainfindings: (a) because the alues on the even ij scale are conditionalsentences, implicative even ij conditionals are truly conditionals: theyexpress a conditional relati n between Pand Q; (b) in quite a few casesthis relation does not hold between Pand the overt Q-clause but ratherbetween Pand an implicit . In that case the overt Q-clause is interpretedas an adverbial clause of re son; (c) it is (part of) the invariant meaningof even that this focusing a verb evokes a sense of expected incompatibility; (d) it is only in implic tive conditionals that this sense of expectedincompatibility manifests i self as a sense of unexpectedness; (e) even ijconditionals that have the ense of unexpectedness implicate an expectation understanding ('One light expect P to preclude Q') and entail anonpreclusive understandin~ ('P does not preclude Q'); (f) in implicativeeven ij conditionals, even I~vokes an even ij scale that is a scale ofimprobability and whose v~ues are ij conditionals, ordered from low (=least unlikely) to high (= ost unlikely); (g) the range of values on theeven ij scale may be deter ined by the range of propositional values onan ancillary P-scale, which may itself depend on another scale; (h) theordering of the values on he P-scale, and the concomitant interpretation, may be determined 'y the positive or negative meaning of theverb in the Q-clause; (i) th re are good reasons to include polar even ij


conditionals in the class of scalar even if conditionals; (j) even blocks thenecessity implicature that i very often invited by implicative conditionals;(k) no additional apparatfs is needed to explain the various kinds ofeven if conditionals with la nonassertoric Q-clause; (l) nonimplicativeeven if conditionals still have a sense of expected incompatibility but lackthe sense of unexpectedne s (yielding the expectation understanding andthe nonpreclusive underst nding) and the scalarity that are typical ofimplicative even if conditi nals; (m) we can distinguish two classes ofnonimplicative even if con itionals: those in which the P-clause is pure1yconcessive and those in w ich it expresses some kind of comment on thecontents, truth, presupposifions, etc., ofthe Q-c1ause. (The two functions,however, are not mutua1l1 exclusive.)

Some of the above fi~dings, as well as some minor ones not yetmentioned here, are corr~tions of, or additions to, the literature. To ourknowledge, it has never be n explicitly argued (a) that the unitary meaningof even is "expected inco atibility," (b) that not all even if conditionalshave a sense of unexpecte ness, (c) that the sense of unexpectedness, ifpresent, involves an eXPfctation understanding plus a nonprec1usiveunderstanding, (d) that t~ expectation understanding differs from thenonpreclusive understandi g in that it is only a cancellable implicature,(e) that only implicative even if conditionals allow a scalar analysis,(f) that the range of val s on the even if scale may be determined bythe range of proposition 1 values on an ancillary P-scale, which mayitself depend on another sfale, (g) that the ordering of the values on theP-scale, and the concomi~ant interpretation, may be determined by thepolarity (in meaning) of the verb of the Q-clause - see section 4.4.4,(h) that the sufficiency nderstanding of 'if P, Q' is not part of thesemantics of conditionals but just an implicature, (i) that nonassertoricimplicative even if conditi I nals should be analyzed exactly like assertoricones, (j) that the P-c1aus of a nonimplicative even if conditional maybe purely concessive or/a d commenting, (k) that many seemingly problematic even if condition Is can be naturally explained by positing animplicit Q-clause, but (l) that this is only possible as far as implicativeeven if conditionals are c ncerned.

Received 30 August 2000Revised version received15 February 2001

Notes

Catholic University of Leuven

1. This articJe was written within a project sponsored both by the Fonds voorWetenschappelijk Onderzo k - Vlaanderen and by the Onderzoeksfonds of the

250

University of Leuven (K.U .. ). We also wish to thank the two anonymous referees,who made useful remarks 0 the first draft of this article. Correspondence address:Renaat Declerck, KULAK, niversitaire Campus, É. Sabbelaan 53, B-8500 Kortrijk,Belgium. E-mail: [email protected].

2. Dancygier (1998) is particu arly rich in this respect. If we criticize her more thananyone else, this is largely ecause she has said far more, and more explicit andperceptive things about even than other writers.

3. Many of the examples addu ed below are attested examples. See the list of Soureesfollowing the References for the abbreviations used, which are foUowed by the pagenumber where appropriate.

4. The only exceptions are Swe tser (1990), who recognizes the existence of "speech-act"even ij conditionals, and Dan~Cygier(1988, 1998), who speaks .of "conversational" evenij conditionals, The kind of xample Dancygier has m mind IS Mary is already on herway here, even if you don't wa t to hear about it (Dancygier 1998: 164).

5. We are using the term "valid~'ty" here because not all Q-clauses have a truth value. InDrink a glass of water befor breakfast, even ij you're not thirsty, the even ij clausedescribes a situation that mi ht be thought to make the injunction expressed in theQ-clause invalid, but injuncti ns do not have a truth value.

6. The same is true of the (i probable) sentence Even ij it rains, the match will becancel/ed, because the expect tion 'The match wiUnot be caneeUed if it rains' is read as'It is not the case that the ma eh will be canceUed if it rains', i.e. the negation has widescope and the conditional in i s scope is implicative.

7. Barker (1991: 23-25) discuss s a type of "anomalous cases" (1991: 25) of counterfactual even ij conditionals, whic ,in our analysis, are explained in terms of an implicit Q:

(i) (said aftel' losing abattl) Even if the reinforcements had come, they would havebeen too tired after their [ourney to help us. (1991: 24) (Durparaphrase is: Even ijthe reinforcements had cote, we would (still) have lost [hebattle because they wouldhave been toa tired after heirjourney to help us.)

However, Barker does not s gest the possibility of accounting for such examples bypositing an implieit Q-propo ition, In fact, he is only interested in finding an explanation for his conclusion that t ere are even ij eonditionals sueh as (i) that "have falseconsequents," which runs co nter to the claim that "even-ijs should never have falseconsequents" (1991: 24). Th explanation he proposes is based on "conversationalconsiderations" (1991: 24) an is quite different from ours.

8. As we will see in section 5, tJ sense of expected incompatibility of even can manifestitself in still different ways wh n even is added to an ij conditional.

9. Actually, in a way, Sweetse (1990) too (whether intentionally or not) treats P asirrelevant in her definition of he meaning of even ij: the even ij c1auseseems to be justa way of getting to the stateJnt that Q "wilI almost certainly occur": "Suppose I saythat certain extremely unfavo able circumstances will nonetheless be sufficiently favorable to ensure Y (or sufficie Iy unfavorable to aUow ~ Y). I have produced a verystrong statement that Y wiU0 cur whatever happens; since nearly aHcircumstances aremore favorable to Y than X is, Y wiU almost certainly oeeur. Such, I claim is thecorrect interpretation of even V-sentences" (1990: 132).

10. It is important to see that "ear~elling" an implicature means that the implicature doesarise but is then denied or qu tioned by the subsequent context. If a usual implicaturedoes not arise at all, we say i is "blocked." Thus, the implicature that the should beread as all the - see Declerek (1987) - is cancelled in The glasses were braken, at least


most of them, but blocked ( y pragmatic knowIedge) in There are cracks in the pavingstones.

11. It might seem odd to claim that the nonpreclusive understanding (which is an entailment of the assertion mad in the conditional) evokes an expectation understandingthat runs counter to it. Ho ever, it is by no means unusual for an assertion to evokean expectation to the contlry. This happens, for example, in John DID do it, whichimplicates (because of the ricean maxim of relevance) that there exists a claim to thecontrary. In fact, any con rastive assertion automatically evokes an expectation orclaim to the contrary, beca se there can only be such a sense of contrast if there is acontrasting proposition.

12. It does not really matter hether we speak of a "probability scale" (with the value"most likely" at the uppertnd and the value "least likely" at the lower end) or of a"scale of unlikelihood" (on which the values are ordered in the reverse way). In whatfo11owswe will adopt the la ter convention.

13. According to some people e.g. Fauconnier (1976), König (1986), Van der Auwera(1986) - even marks the e treme of a scale. However, this claim is too strong, as hasbeen pointed out by (amo g others) Fi11imore(1965), Hom (1969), Anscombre andDucrot (1983), Dancygier 1988), Kay (1990), and Lycan (1991).

14. Similarly for simple even, ay (1990: 90) notes in conneetion with The test was so hardthat even Charlesfailed it, 'we do not necessarily, as some have thought, presupposethat Charles was thc least ikcly pcrson to fail the test of those who took it. But we

the test Charles was the lea t likely to do so."IS. This runs counter to Köni's (1986: 232) and Dancygier's (1998: 162) view that the

values are conditions: "the scale implied byeven is understood as a scale of possibleconditions, one ofwhich aJ:1ears in the protasis" (Dancygier 1998: 162).

16. The observation that the ncillary numerical scale ultimately determines the even il'scale is in keeping with the fact that, in sentences where "even ij ... just" makes sense,even can be put either be ore the constituent focused by just or before ij with noapparent difference ofmea ing:

(i) I would blame her ifi was evenjust one per cent her fault.(ii) I would blame her evienif it was just one per cent her fault.

17. There is a similar difference between the fo11owing:

(i) 1'11be angry even if Jeturns up.(ii) 1'11be angry if Jane e lenturns up.

In (i) even refers to a scale "(hose values are conditionals, with J'1l be angry ijJane turnsup representing a (relativel~ extreme value on that scale. In (ii), which is interpreted as'1'11be angry if Jane does s much as turn up', even refers to a scale whose values arenot conditionals but are th various things that Jane can do that are sufficient to makeme angry.

18. In (26), the values on the ancillary scale of persons are ordered in accordance withtheir importance, i.e. the d gree to which it is necessary that the person be present onthe coach or/and its destin tion; what the P-scale is measuring is the expected likelihood of a person's being I te to prevent the coach leaving on time; what the even ijscale is measuring is the lik lihood of the coach leaving on time. In (27), the values onthe scale of persons are ordered in accordance with the probability of a person beinglate; for lack of even ijtherJ is neither an even ijscale nor an ancillary P-scale.

252

19. This definition of "implicativ ifconditional" has been elaborated on in section 2.4 andshould be distinguished from the definition of "irnplicative even i{conditional" givenin section 2.6. (The latter is a even if conditional whose expectation understanding isexpressed by means of an imp icative if conditional.)Dancygier (1998: 162), in a di cussion of Haiman (1974), speaks ofa "scale ofnegativeexpectation." This notion is not unpacked in detail, but it is likely that it is notdissimilar to the scale just d scribed here, except for one important difference: forDancygier, the scale is (despi e her view that even if sentences are not conditionals) ascale of "possible conditions, ' not a scale of conditionals, although she says that thescale "relates specifically to t~e p/q relation."The polar option that is NOTI chosen in the even ij' conditional is normally feit to belower on the improbability sc le (i.e. more likely) than the one that is chosen. That is,ifI say Ij John comes, the par y wil/ be good - oh, and of course, it'll be good even i/hedoesn't come, John's coming s interpreted as less unlikely to go with the party beinggood (i.e. more likely to brin about a good party) than his not coming.As noted by Barker (1994: 52), a contrastive accent on the auxiliary do induces apolar interpretation.Dancygier (1998: 166) seems 0 argue that even if conditionals like (41b) are scalar. Wesee no justification for this an lysis.For lack of space, we cannotl go into Van der Auwera's (1997) and Horn's (2000)explanations here.Surprisingly, Dancygier (19918) seems to agree with this: "There is some confusionamong inferential conditonalf,' as some seem to pass the test [for inviting the conditional perfection inference], ihi,e others do not. Thus if the protasis is accepted (eventhough only tentatively) as lhe only premise licensing the conclusion, perfection isapplicable, as in the case of he was wearing a purple jacket, then it was him I saw atthe party" (1998: 163).Speech-act conditionals are c ' nditionals in which the if clause gives the reason why thespeaker makes the speech actlof uttering the Q-clause. The best known example in theliterature is If you're hungry, Jhere'sa pie in thefridge.According to Sweetser (1991,speech-act even if statements amount to 'I say (insist)that X, even ifl admit that '( 1990: 138). However, it will be pointed out below thatthis is an overgeneralization see note 31.Dancygier's (1998) discussio~ of the meaning of even if conditionals assumes that theyare always implicative, e.g. "4ven ifsentences are always related to the speaker's beliefsabout causal links which can be expected to hold" (1998: 163-164). However,Dancygier does mention th~1existence of "conversational" (and epistemic) even ifsentences, saying that "Just a in the case of predictives, these sentences refer negativelyto relations between pand q t at might be expected to hold" (1998: 164).Apart from the purely conces ive interpretation, it is also possible to read (60) in termsof an implicit Q-clause: 'Eve~ if politicians were not interested in this subject, it wouldstill be a current topic, becausf the media would still be paying a lot of attention to it'.On this reading the sentence i~scalar and implicative.Since Q is [-q] here, 'cause not-Q' means 'cause [+q]'.Note that there is also no ECESSARY implication of 'even if I admit that', whichdisproves Sweetser's (1990: 1; 8) "speech-act analysis" of (what we call) purely concessive even ij, according to whio a purely concessive even ifsentence means 'I say (insist)that X, even if I admit that V'. (We must admit, however, that Sweetser does notexplicitly claim that al! purel concessive even if sentences are speech-act conditionals.It is just that al! the "pure co lcessives" - she does not use this term herself - that she

20.

21.

22.

23.

24.

25.

26.

27.

28.

29.

30.31.


analyzes [1990: 138], she an Iyzes as speech-act conditionals.) There are further problems with this analysis. Swe tser's paraphrase does not explicitly teil us how the scalarmeaning that she attributes 0 even ij takes a part in the interpretation, but there areindications that it should b read as something like 'admitting that P is an extremelyunfavorable circumstance f r the performance of the speech act in Q, it is nonethelesssufficiently favorable to ens rejallow the performance of the speech act in Q, so let usconsider that 1am performi g the speech act of insisting that Q'. This kind of paraphrase is not appropriate fo (65), and neither is it for examples like the following:

(i) Even if Bruno were a~ive,he could not publicly respond [because it would be abreach of privacy]. (T S)["When I see headlin s like 'Oscar, Depardieu, Rapist,' it's bad, bad for myfamily, my wife E'isab1eth,my children. !t's bad because] even if it's not true andit's stupid - it stays i people's minds. [It dirties me," Depardieu told the Frenchpress.] (COB-W)"Kubla Khan," even f originally scribbled down in a frenzy of recollection, wascarefully revised for p blication. (TLS)

(ii)

(iii)

33.

Clearly, (i) is not interpret d as 'Admitting that Bruno's (hypothetical) being alivewould be extremely unfavor bie to my asserting that he could not respond publicly, itwould nevertheless be suffi iently favorable, so let us consider that I assert that hecould not respond publicly'. Similarly, (ii) cannot be analyzed in terms of 'I insist thatit stays in people's minds, ev n if 1admit that it's not true and it's stupid'. And it wouldbe downright silly to claim hat (iii) means 'admitting that Kubla Khan was originallyscribbled down in a frenzy of recollection is highly unfavorable to the speech act ofinsistingjasserting that it wa carefully revised for publication' .

32. Perhaps (68) could be seen as implicative rather than purely concessive. This wouldmean that implicative even ij conditionals mayalso have an implicature-cancellingeffect.The following conditionaIs, [' hich are clearly echoic, typically have a closed condition:

(i) ["He's coming here htmself tomorrow." - "Right.] If he's coming here himselftomorrow, I no long r need to telephone him today."

(ii) If, as you say, the wat r will rise as high as this, we'd better move our furniture tothe first fioor.

34. The version with and may s em a bit odd, but if it is interpreted as purely descriptive,say as an answer to a poli eman wanting the exact description of how she spoke, itis OK.

References

Akatsuka, Noriko (1991). Dracu a conditionals in discourse. In Jnterdisciplinary Approachesto Language: Essays in Honor >.l S- Y. Kuroda, Carol Georgopoulos et al. (eds.), 25-37.Dordrecht, Boston, and Lond n: Kluwer Academie.

Anscombre, Jean-Claude; and ucrot, Oswald (1983): L'Argumentation dans la Langue.Brussels: Mardaga.

Athanasiadou, Angeliki; and Dir en, René (1997). Conditionality, hypotheticality, counterfactuality. In On Conditiona's Again, Angeliki Athanasiadou et al. (eds.), 61-96.Amsterdam and Philadelphia: enjamins.

254

Barker, Stephen J. (1991) Even, till and counterfactuals. Linguistics and Philosophy 14,1-38.

-(1994). The consequent-entail ent problem for even if. Linguistics and Philosophy 17,249-260.

Comrie, Bernard (1986). Conditi nals: a typology. In On Conditionals, Elizabeth ClossTraugott et al. (eds.), 77-99. Ca bridge: Cambridge University Press.

Dancygier, Barbara (1988). Condi ionals and concessives. Papers and Studies in ContrastiveLinguistics 24, 111-121.

-(1998). Conditionals and Pred ction: Time, Knowiedge, and Causation in ConditionalConstructions. Cambridge: Cam ridge University Press.

Declerck, Renaat (1987). Definite ess and inclusive reference. Joumal of Literary Semantics16,12-29.

-( 1988). Studies on Copular Sent nces, Clefts and Pseudo-clefts. Leuven: Leuven UniversityPress; Dordrecht: Foris.

-; and Reed, Susan (forthcomi g). Conditionals: A Comprehensive Empirical Analysis.Berlin: Mouton de Gruyter.

Fauconnier, Gilles (1976). Étude e certains aspects logiques et grammaticaux de la quantification et de I'anaphore en rançais et en anglais. Unpublished Doctorat d'État,Université de Paris VII.

Fillmore, Charles (1965). Some EI tailment Rules of English. P.O.L.A. 10. Columbus: OhioState University.

Fraser, Bruce (1969). An analYls of concessive conditionals. In Papers from the FifthRegional Meeting of the Chicag Linguistic Society, Robert I. Binnick et al. (eds.), 66-73.Chicago: Chicago Linguistic So iety.

Geis, Michael L.; and Zwicky, AIOld M. (1971). On invited inferences. Linguistic lnquiry2,561-566.

Haegeman, Liliane (1984). Pragm tic conditionals in English. Folio Linguistica 18,485-502.Haiman, John (1974). Concessi es, conditionals, and verbs of volition. Foundations ofLanguage 11, 341-359.

Hom, Laurence (1969). A presu positional analysis of only and even. In Papers from theFifth Regional Meeting of the hicago Linguist Society, Robert I. Binnick et al. (eds.),98-107. Chicago: Chicago Ling istic Society.

-(2000). From if to iff: condit onal perfection as pragmatic strengthening. Joumal ofPragmatics 32, 289-326.

Jespersen, Otto (1940). A Modern English Grammar on Historical Principles, Part 5: Syntax(vol. 4). London: Allen and Unrin; Copenhagen: Ejnar Munksgaard.

Kay, Paul (1990). Even. Linguisti~s and Philosophy 13, 59-111.Kjellmer, Göran (1989). Even ifa d even though. English Studies 70,256-269.König, Ekkehard (1986). Condi ionals, concessive conditionals and concessives. In On

Conditionals, Elizabeth Closs raugott et al. (eds.), 229-246. Cambridge: CambridgeUniversity Press.

Lycan, W. G. (1991). Even andev n if. Linguisticsand Philosophy 14, 115-150.Lyons, John (1977). Semantics. C mbridge: Cambridge University Press.Quirk, Randolph; Greenbaum, idney; Leeeh, Geoffrey; and Svartvik, Jan (1985). A

Comprehensive Grammar Ofthe{ngliSh Language. London and New York: Longman.Sweetser, Eve (1990). From Etym logy to Pragmatics: Metaphorical and Cultural Aspects ofSemantic Structure. Cambridge: Cambridge University Press.

Van der Auwera, Johan (1986). S eech act conditionals. In On Conditionals, Elizabeth ClossTraugott et al. (eds.), 197-214. ambridge: Cambridge University Press.


-(1997). Conditional perfectio . In On Conditionals Again, Angeliki Athanasiadou et al.(eds.), 169-190. Amsterdam a d Philadelphia: Benjamins.

Sourees

BOF

COB-S

COB-W

DOCLOB·

TAMTLS

TRC

TUS

WOS

O'Neill, Michael; a d Seabrook, Jeremy (1971). The bosom of the family.Plays and Players 19 I.The COBUILD-Cor us of English: UK, Spoken English (n.d.).Birmingham: Harper ollins (available on-line).The COBUILD-Cor us of English: UK. Written English (n.d.). Birmingham:HarperCollins (avail bie on-line).Dextcr, Colin (1994). The Daughters of Cain. London: Macmillan.(1991). Lancaster-O Ic/Bergen corpus of British English. In The ICAMECollection of English anguage Corpora. Bergen: Norwegian Computing Centerfor the Humanities.The Australian Mag zine (1993). December.Collection of artiele (on CD-ROM) from the Times Literary Supplement(1997). London: Th Times.Barnes, Peter (197). The Ruling Class: A Baroque Comedy. London:Heinemann.Ustinov, Peter (196 ). The Unknown Soldier and his Wife: Two Acts of WarSeparated by a Truc . London: Heinemann.Osborne, John (1971). West of Suez. London: Faber and Faber.

some truths and nontruths about 'even if

Documents