Frank Jackson - On Assertion and Indicative Conditionals

7/27/2019 Frank Jackson - On Assertion and Indicative Conditionals

1/26

Philosophical Review

On Assertion and Indicative ConditionalsAuthor(s): Frank JacksonReviewed work(s):Source: The Philosophical Review, Vol. 88, No. 4 (Oct., 1979), pp. 565-589Published by: Duke University Press on behalf of Philosophical ReviewStable URL: http://www.jstor.org/stable/2184845 .

Accessed: 23/02/2012 13:56

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .http://www.jstor.org/page/info/about/policies/terms.jsp

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of

content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms

of scholarship. For more information about JSTOR, please contact [email protected].

Duke University Press and Philosophical Review are collaborating with JSTOR to digitize, preserve and extend

access to The Philosophical Review.

http://www.jstor.org
http://www.jstor.org/action/showPublisher?publisherCode=dukehttp://www.jstor.org/action/showPublisher?publisherCode=philreviewhttp://www.jstor.org/stable/2184845?origin=JSTOR-pdfhttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/stable/2184845?origin=JSTOR-pdfhttp://www.jstor.org/action/showPublisher?publisherCode=philreviewhttp://www.jstor.org/action/showPublisher?publisherCode=duke


2/26

The PhilosophicalReview,LXXXVIII, No. 4 (October 1979).

ON ASSERTION AND INDICATIVECONDITIONALSFrankJacksonIntroduction

T he circumstancesn which t snatural o assert heordinaryindicative onditional IfP thenQ" are those nwhich t snatural o assert EithernotP or,P and Q." and conversely.orinstance, hecircumstancesn which t is naturalto assert If itrains, hematchwillbe cancelled"are precisely hose nwhichit is natural to assert "Either it won't rain or, it will and thematch will be cancelled." Similarly,the circumstances n whichit is natural to assert "Not both P and Q" are precisely those inwhich it is natural to assert "Either not P or not Q." We explainthe latter coincidence of assertion conditions by a coincidenceoftruthconditions.Why not do the same in the case of the con-ditional? Why not, that is, hold that "If P then Q" has the sametruth conditions as "Either not P or, P and Q"?This hypothesis-given the standard and widely acceptedtruthfunctional treatments f "not," "or,"5and "and"-amountsto theEquivalence thesis: the thesis that P Q) is equivalent to(P D Q). (I will use "-*" for he indicative conditional, reserving"0-*" for he subjunctiveorcounterfactualconditional.) In thispaper I defend a version of the Equivalence thesis.As a rule, our intuitive udgements of assertabilitymatch upwith our intuitive udgements ofprobability, hat s,S is assertableto the extentthat thas high subjectiveprobabilityfor ts assertor.Now it has been widelynoted that when (P D Q) is highly prob-able but both -P and Q are not highly probable, it is proper toassert (P - Q).1 The problem for the Equivalence thesis is toexplain away the putative counter examples to "-P e (P-Q)J"and "Q H (P -> Q)," the only too familiarcases where despitethe high probabilityof-P or ofQ. and so of (P D Q), (P -* Q) is

' Though the point is commonly put in terms of evidence, see e.g., CharlesL. Stevenson, "If-iculties," Philosophy f Science,37, 1 (March 1970), 27-49,and G. H. Von Wright, "On Conditionals" in Logical Studies,Routledge &Kegan Paul: London, 1957, see p. 139.565


3/26

FRANKJACKSONnot highly assertable.I will start n section 1 by considering the usual way of tryingto explain away these counterexamples and argue that it fails.An obvious reaction to this failure would be (is) to abandon theEquivalence thesis; but I argue in section 2 that another ispossible, namely,that thegeneral thoughtbehind the usual wayof explaining away the paradoxes of material implication ismistaken.This leads in section 3 to the versionof the Equivalencethesis wish to defend. n section 4 1 pointout some of theadvan-tages of this account of indicative conditionals, and in section5 I reply to possible objections.1.THE USUAL WAY OF EXPLAINING AWAY THE COUNTEREXAMPLES.

Suppose S, is logically stronger han S2: S, entails S2 but notconversely.And suppose S, is nearly as highly probable as S2.(It cannot, of course, be quite as probable, except in very specialcases.) Why then assertS2 instead ofS,? There are many possiblereasons: S2 might read or sound better, , might be unduly bluntor obscene, and so on. But ifwe concentrate on epistemic andsemantic considerationswidely construed, and put aside moreparticular, highly contextual ones like those ust mentioned, itseems that therewould be no reason to assert S2 instead of S.There is no significant oss of probability in assertingS, and,by the transitivity f entailment,S, must yield everything ndmore than S2 does. Therefore,S, is to be asserted rather thanS2, ceterisaribus.This line of thought, which I will tag, "Assert the strongerinstead of the weaker (when probabilities are close)," has beenprominent n defensesof theEquivalence thesisthat the ordinaryindicative conditional, (P - Q), is equivalent to the materialconditional, (P D Q).2 The Equivalence theoristexplains awaythe improprietyof asserting (P D Q) when one of -P or Q ishighly probable by saying that in such a case you should come

2 Particularly in discussion, but see R. C. Jeffrey,ormalLogic, McGraw-Hill:New York, 1967, ch. 3; David Lewis, "Probabilities of Conditionals andConditional Probabilities," PhilosophicalReview,LXXV, 3 (July 1976), 297-315, and for support for the general idea and other arguments for the Equiva-lence thesissee Michael Clark, "Ifs and Hooks," Analysis, 2, 2 (1971), 33-39.

566


4/26

ASSERTION AND CONDITIONALSright out and assert the logically strongerstatement,namely,either-P or Q as the case may be.

The same idea can be put in terms of evidence instead ofprobability.3 f your evidence favors (P D Q) by favoringoneof -P or Q you should simply assert -P or Q. whichever it is,and not the needlesslyweak conditional.' But I will concentratein the main on the probabilistic formulationwhen presentingmy objections.My first bjection is that a conditional like "If the sun goesout ofexistence in ten minutes time, the earth will be plungedinto darkness in about eighteen minutes time" is highlyassert-able. However, the probabilityof the material conditional andthe probability of the negation of its antecedent are both veryclose (if not equal) to one, and so at most the probabilityof theconditional is only marginally the greater.Hence this is a casewheretheweaker is assertable despite the absence of any appreci-able gain in probability, contrary to the maxim "Assert thestronger nstead of the weaker."The second objection is that conditionals whose high prob-ability is almost entirelydue to that of their consequents maybe highlyassertable. Suppose we are convinced that Carter willbe reelectedwhetheror notReagan runs.We sayboth "If Reaganruns, Carter will be reelected" and "If Reagan does not run,

3 I understand that this was the emphasis in H. P. Grice's influential,unpublished William James Lectures, see L. Jonathan Cohen, "Some Remarkson Grice's Views about the Logical Particles of Natural Language," in Prag-maticsofNaturalLanguages, d. Y. Bar-Hillel, Reidel: Holland, 1971; MichaelClark, ibid., and particularly "Ifs and Hooks: A Rejoinder," Analysis,34,3 (1974), 77-83; A. J. Ayer, Probability nd Evidence, MacMillan: London,1972; Stevenson, op. cit.; and J. L. Mackie, Truth,Probabilitynd Paradox,Oxford University Press: London, 1973.4 In theirpresentationof Grice's (tentative) views Cohen et. al. sometimesuseformulationsthat are ambiguous about whether it is all or part of your evi-dence that is meant. If it is all, things are as above; but if it is part, the viewbeing reported is that (P - Q) is assertable if part of your evidence favors(P D Q) without favoring one of -P or Q. even if your total evidence favorsone of them. There is immediate trouble for such a view. Suppose I knowthat Fred and Bill both live in Oak Street.Even though my evidence stronglyfavors the material conditional, it would normally be wrong to assert "IfFred lives in Elm Street,Bill lives in Elm Street" in such a case; neverthelesspartofwhat I know, namely, that they live in the same street, favorsthe materialconditional without favoring ts consequent and without favoring the negation

of its antecedent.567


5/26

FRANKJACKSONCarter will be reelected." The high subjective probability canonly be due to that of the common consequent, yet the con-sequent is allegedly logically stronger nd so by the maxim theconditionals ought not be assertable.Moreover, such cases cannot be handled by a conventionalexemption' fromthe maxim in the case of conditionals withvery probable consequents. Both the followingconditionals arehighly unassertable,ut have very probable consequents: "Ifthe history ooksare wrong,Caesar defeatedPompey in 48 B.C.,""If the sun goes out of existence in ten minutes time, the earthwillnotbe plunged into darkness n eighteenminutestime."The thirdobjection is that there s a thirdparadox ofmaterialimplication. If the Equivalence thesis is true, then ( (P - Q) v(Q -u R) ) is a logical truth.But evidently it is not in generalhighlyassertable. Of course logical truths re as logically weakas you can get, but nevertheless"Assert the stronger nsteadof the weaker" is of no assistance in explaining away the thirdparadox. Whatever you thinkabout this maxim in general, itdoes not apply universallyto logical truths."If that's the wayit is, then that's the way it is," "George must eitherbe here ornot here," "The part is not greaterthan the whole" and so on,are all highlyassertable.The fourth objection is that "Assert the stronger insteadof the weaker" is, of necessity,silent about divergences in as-sertability among logical equivalents simply because logicalequivalents do notdiffern strength. ut theequivalence theoristmustacknowledge some marked divergencesamong equivalents.According to him, ( (-P & (P -> R) ) and (-P & (P -* S) ) arelogically equivalent, both being equivalent to -P. But theirassertability an differ harply."The sun will come up tomorrowbut if it doesn't, it won't matter" is highlyunassertable, while"The sun will come up tomorrowbut if it doesn't, that willbe the end of the world" is highlyassertable.My final objection is that if the standard way of tryingtoexplain away the paradoxes is right, or" and "-a" are on a par.It would, for instance, be just as wrong, and just as right,toassert "P or Q" merely on the basis of knowing P as to assert5 Pace what appears to be Lewis's suggestion, op. cit., p. 308.

568


6/26

ASSERTION AND CONDITIONALS(P D Q) merely on the basis of knowing not P. And, more gen-erally, P F (P orQ)" and "Q F (P or Q)" should strike sas just as much a problem forthe thesis that "P orQ" s equiva-lent to (P v Q) as do the paradoxes of material implication forthe Equivalence thesis. It is a plain fact that theydo not. Thethesisthat "P or Q" s equivalent to (P v Q) is relativelynon-controversial, he thesis that (P -* Q) is equivalent to (P D Q) ishighly controversial.This objection, of course, applies not just to attempts toexplain away the paradoxes in termsof "Assert the stronger,"but to any attempt which appeals simply to considerations ofconversational propriety. t leaves it a mysterywhy we-whoare after all reasonably normal language users-find it so easyto swallow one thesis and so hard to swallow the other.Should we respond to these objections by abandoning theEquivalence thesis or by looking for a differentway of ex-plaining away the paradoxes? An argument forthe latter is thatthethoughtbehind "Assertthe stronger ather than the weaker"contains a serious lacuna, as I now argue.

2. A REASON FOR SOMETIMES ASSERTING THE WEAKER.Suppose, as before,that S, is logically strongerthan S2 andthatS, s probability sonly marginally ower thanS2's.Consistentwith thisit may be that the impact of new information, , on S,

is verydifferent romthe impact ofI on S2; in particular it mayhappen thatI reduces the probabilityofS, substantiallywithoutreducing S2's to any significantextent (indeed S2's may rise). Iwill describe such a situation as one whereS2 but not S, is robustwithrespectto I. If we accept Conditionalization, the plausiblethesisthat the impact of new information s givenbytherelevantconditional probability,then "P is robust withrespectto 1' willbe true ust when both Pr(P) and Pr(P/I) are close and high.6(Obviously a more general account would simply require that

6 See, e.g., R. C. Jeffrey, ogic of Decision, McGraw-Hill: N.Y., 1965; and F.P. Ramsay, Foundations f Mathematics,Routledge & Kegan Paul: London,1931, ch. 7. Robustness is a notion I firstheard about some years ago fromManfred Von Thun in the context of weight n J. M. Keynes's sense, Treatise nProbability,Macmillan: London, 1921. Brian Skyrms uses "resilience" for asimilar notion, see his "Physical Laws and Philosophical Reduction," in569


7/26

FRANKJACKSONPr(P) and Pr(P/I) be close, but throughoutwe will be concernedonly with cases where the probabilities are high enough towarrantassertion,other things being equal.)We can now see the lacuna in the line of thought yingbehind"Assert the stronger instead of the weaker." Despite S, andS2 both being highly probable and S, entailing everythingS2does, theremay be a good reason forassertingS2 either insteadofor as well as S,. It may be desirable thatwhat you say shouldremain highlyprobable should I turn out to be the case, andfurthert may be that Pr(S2/I) is high while Pr(S,/I) is low. Inshort,robustnesswith respect to I may be desirable, and (con-sistentwithS, entailing S2)S2mayhave itwhile S, lacks it.Examples bear thisout. Robustness is an important ngredientin assertability.Here are two examples taken fromthose whichmightbe (are) thoughtto be nothingmore than illustrationsof"Assert the stronger nstead of the weaker."Suppose I read in the paper that Hyperion won the 4.15.George asks me who won the 4.15. I say "Either Hyperion orHydrogen won." Everyone agrees that I have done the wrongthing. Although the disjunction is highly probable, it is nothighly assertable. Why? The standard explanation is in termsof "Assert the stronger nstead of the weaker."' But is this thewhole story?Consider the followingmodification to our case.What I read is that H-- won. The name is too blurred forme to do morethan pickout the initial letter.However I happento know that Hyperion and Hydrogen are the only two horsesin the 4.15 whose names begin with "H," and in addition Iknow that Hydrogen is a no-hoper from the bush. Clearly it isstill the case that "Hyperion won" is highly probable and itwould be quite properforme to say so. But it would alsobe quiteproper for me to say "Hyperion or Hydrogen won," despiteits being weaker and only marginally more probable. Indeedthe natural thingto do would be to say somethinglike "EitherInduction,Probability,nd Confirmation, innesota Studies in Philosophy ofScience, vol. VI, ed. G. Maxwell and R. M. Anderson,Jr., UniversityofMin-nesota Press: Minneapolis, 1975. Neither should be held responsible formyuse of the notion in what follows.

7 "Standard" in that it is offeredby nonequivalence theorists as well asequivalence theorists,see, e.g., J. L. Mackie, op. cit., p. 76.570


8/26

ASSERTION AND CONDITIONALSHyperion or Hydrogen won. It can't have been Hydrogen-he's a no-hoper. So it must have been Hyperion."

The obvious explanation for the marked change in the as-sertabilityof the disjunction is that in the original case it wasnot robust with respect to the negation of both its disjunctstakenseparately, while in the modified case it is. In the originalcase, were I to learn that Hyperion was not the winner I wouldhave to abandon the disjunction. In the modified case I wouldnot, thoughI would have to abandon my low opinion of Hydro-gen. Therefore, n themodifiedcase there s point to asserting hatHyperion or Hydrogen won instead of simply that Hyperion won,even ifthe probabilities are veryclose. This disjunction possessesa relevant robustness that its left disjunct lacks.Indeed surely there are many cases where disjunctions arehighlyassertable even though they have probabilities for theirassertorsonly marginally greaterthan that of one of their dis-juncts. Consider "Either Oswald killed Kennedy or the WarrenCommission was incompetent." This is highly assertable evenfor someone convinced that the Warren Commission was notincompetent. Yet they are in a position to assert the stronger"Oswald killed Kennedy." The disjunction is neverthelesshighly assertable for them, because it would still be probablewere information o come to hand that refutedone or the otherdisjunct. The disjunction is robust with respect to the negationof either of its disjuncts taken separately-and just this maymake it pointful to assert it. Because it makes it acceptable toa possiblehearer who denies one orother of the disjuncts.Moreover, we can have highly probable disjunctions whichare, unlike the two ust considered, significantlymore probablethan either of their disjuncts and yet which are not highlyassertable. Suppose I propose to toss a fair coin five times insuch a way that the tosses re probabilistically ndependent; then"At least one ofthe fivetosseswill be a head" is probable enough(- 97%) towarrantassertion.Consequently so is the equivalentdisjunction "Either at least one of the firstthree tosses or atleast one of the last two tosseswill be a head," and moreovereach disjunct is significantlyess probable than the disjunction.But it would be highly misleading to assert the disjunctionin preferenceto the equivalent sentence. For it would create

571


9/26

FRANKJACKSONin hearers the mistaken expectation that should the first hreetossesfail to yield a head, they can be sure that at least one ofthe last two will.The second example is one of David Lewis's.We are gatheringmushrooms; I say to you "You won't eat that one and live."A dirtytrick: I thought that one was safe and especially delicious, I wantedit myself, o I hoped to dissuade you fromtaking it without actually lying.I thoughtit highlyprobable that mytrickwould work,that you would not eatthe mushroom,and therefore hat I would turnout to have told the truth.Butthough what I said had a high subjective probability of truth, it had a lowassertability and it was a misdeed to assert it. Its assertability goes not justby probability but by the resultant of that and a correction term to takeaccount of the pointlessness and misleadingness of denying a conjunctionwhen one believes it falsepredominantlybecause of disbelieving one conjunct.8

But thisexplanation faces two difficulties. irst,suppose I amnot that onfidentthat my trickwill work. I am pretty ure butnot certain enough to warrant outrightassertion. And furthersuppose thatI am also pretty ertainthatyouwill die forreasonsunconnected with mushrooms.The two factors ombined bringtheprobabilityof "You won't eat thatone and live" up to a levelsufficient o warrant assertion. In this case the probability offalsityof each conjunct contributes significantly o the prob-ability that the negated conjunction is true,but nevertheless twould stillbe a misdeed to assertit. Second, suppose the mush-room really is dangerous and I say "You won't eat that oneand live" while crushingit under my footfor safety's ake. Thedifference in probability between the negated conjunctionand "You won't eat that one" will then be miniscule. But thenegatedconjunction is neverthelesshighly ssertablein thiscase.It seems to me, therefore, hat a better explanation is onein termsofrobustness.You take me to be providinginformationrelevant to mushroom-eating pleasures, and so construct foryourself he followingpiece of practical reasoning. I won't eatthat one and live. (Premise supplied by me.) I eat that one.(Premise you can make true.) Therefore, I won't live. Theconclusion is undesirable, hence you are led to refrainfrommaking the second premise true.Why were you tricked?The argument is valid, the premise Isupplied does have a high probability,and you are able to give8 op. cit.,pp. 306-7.

572


10/26

ASSERTION AND CONDITIONALSthe second premise a high probability. But in order to inferthe conclusion of a valid argument all premises need to be highlyprobable together;nd if you were to make the second premisehighly probable, the firstpremise (supplied by me) would nolonger be highly probable. In the circumstances you were en-titled to take it that not only was "You won't eat that one andlive" highly probable, it was also robust with respect to "Youeat that one." My misdeed lay in asserting something lackingappropriate robustness.

The upshot, then, is simply that when considering proprietyofassertionwe should take account of robustnessas well as highprobability, relevance, informativeness, nd so on.3. THE APPLICATION OF ROBUSTNESS TO CONDITIONALS.

Robustness is a relative affair.A highlyprobable sentencemaybe veryrobust relative to one possible piece of information9 ndthe opposite relative to another. Often the possible informationrelativeto which robustness s desirable is given by the context.In the mushroom gathering story twas obvious that the hearerexpected sentences that were robust relative to his eating themushroom. That is how he was tricked. But context will notalways be enough. It makes sensethat we should have syntacticalconstructionswhich signal the possible information relativeto which we take what we are saying to be robust.Their role would be akin to that of "but" in signalling orindicating a contrast without the obtaining of this contrastbeing a necessarycondition forspeaking truly. 0Thus the truthconditions for "P but Q" re the same as those for "P & Q."In familiar argons"1 their literal content is the same, but the

The possible informationmay be actual. Obviously we are often nterestedin robustness relative to what we might, but don't to date, know. But thisis not part of the definition of robustness. If it was, P would automaticallybecome nonrobust with respect to I on learning I! When I is known at t,ourdefinitionmakesP robust withrespectto I ifand onlyifP ishighly probable at t.See, e.g., M. Dummett, Frege, Duckworth: London, 1973, pp. 85-6.See, e.g., Dummett, op. cit., and various of H. P. Grice's papers, including"Logic and Conversation," in Syntax nd Semantics, ol. 3, ed. Peter Cole andJerryL. Morgan, Academic Press: N.Y., 1975, and "Further Notes on Logicand Conversation," in Syntax nd Semantics,Vol. 9, ed. Peter Cole, AcademicPress: N.Y., 1978.

573


11/26

FRANK JACKSONuse of the first arries a conventional (not conversational) im-plicature that the second does not; or they differn tone but arealike in sense. I will however talk mainly of signalling and in-dicating rather than implicature or tone. What follows doesnot depend crucially on the preciseway such a distinction houldbe drawn.It is, of course, vital that we allow the possibilityof distin-guishingsignallingor indicating an attitude towards a sentencefrom making that attitude part of the truth-conditions, enseor literal content of what we say. There is a great differencebetween producing a sentence S as something accepted andthereby sserted, nd producing it as an example or as somethinggranted for the sake of argument. It is thus important that wecan signal this-perhaps by using Frege's assertion sign-andsuch a signal cannot be taken simply as part of the content ofwhat is said. Because "S and I accept (or assert)S" may as easilyas S itselfbe produced as an example or grantedforthe sake ofargument,rather than being asserted."2I am suggesting, hen, that whenwe asserta sentence itmakessense that we should have ways of indicating that as well asobeying the base rule that requires that S be highly probable,we also take it that,for ome I, S is robust withrespect toLOne way of doing this is to put your sentence in disjunctiveform when it would be shorter and simpler not to. SupposeI am asked what color Harry's car is. It is perfectly cceptableforme to reply simply that it is blue, even ifmy groundforbeingconfident that it is blue is that it is light blue. Unless there isreason to think that the precise shade matters,near enough isgood enough here. Suppose however that I replied that Harry'scar is either light blue or dark blue. This reply is not accept-able in the circumstances even if the precise shade does notmatter, despite the fact that (ignoring borderline cases) it isequivalent to the acceptable one (and so incidentally the dif-ference in assertability cannot be explained by reference to"Assert the stronger nstead of the weaker"). The reason thatthe second reply s notacceptable is that n putting tin explicitlydisjunctiveformyou signal robustnesswith respectto the nega-

12Cf.,Dummett, p. cit. p. 316.574


12/26

ASSERTION AND CONDITIONALStion of each disjunct taken separately. The reply would beproper only if both the (subjective) probability of its beinglight blue or dark blue given it is not light blue and the prob-ability of ts being light blue or dark blue given it is not dark bluewere high. And in our case the former s low.In general we are happiest asserting disjunctions which aretwo-sidedly robust. We most happily assert "P or Q"when Pr(P v Q), Pr(P v Q/-P), nd Pr(Pv Q/-Q)re all high. (Thus, theoftnoted "exclusive feel" about the inclusive"or.") Accordingly,when we are not in a position to so assert,we should expectto have a way ofsignallingmerelyone-sided robustness n orderto avoid misleading our hearers into assuming two-sided ro-bustness.And it seems that we do.Consider the following,common enough kind of case. Youare prettysure that George lives in Boston but not quite sureenough to warrant outright assertion. You are, though, sureenough that he lives somewhere in New England. You say "Helives n Bostonoranywaysomewhere n New England." Likewisewe say things like "He is a fascist communist) or anyhow onthefarright left)" and "Caesar defeatedPompey in 48 B.C., orat least that's what George told me." We use the "P or anywayQ" onstructionto indicate that "P v Q" s robust with respectto -P, but not with respectto -Q. hould you learn against yourexpectation that George does not live in Boston, the disjunctionwill still be highly probable foryou due to its right disjunct"George lives in New England" stillbeing so; but obviously youwill have to abandon the disjunction altogether should youlearn that George does not live in New England at all.A consequence of this asymmetry s that commutation cangive strange sounding results. "He lives in Boston or anywaysomewhere in New England" is a happy saying,whereas "Helives somewhere in New England or anyway in Boston" is not.Nevertheless commutation is valid; for the truth-conditionsof"P oranywayQ" re ust thoseof"P v Q.""George ives nBoston or anyway somewhere in New England" is true if andonly if either "George lives in Boston" is true or "George livessomewherein New England" is true. "Caesar defeated Pompeyin48 B.C., or at least thatiswhatGeorge told me" is true feither

575


13/26

FRANKJACKSONdisjunct is true and false if neither is; and so on and so forth.Signalling robustnessdoes not invade truth-conditions.

Before I apply these ideas to indicative conditionals, let mereview the course of the argument. High probability is an im-portant ingredientin assertability.Everyone accepts that. Butso is robustness. Commonly, cases cited to illustrate "Assertthe stronger nstead of the weaker" really illustratethe impor-tance ofrobustness.The relevantrobustness,however, s relativeto statementsother than the one being asserted. (Every highlyprobable statement s trivially obustwithrespectto itself.)Thuswe need devices and conventions to signal which statementsour assertionsare robust relative to. We have just been lookingat some of these devices and have noted that their presencedoes- not alter truth-conditions.Accordingly, I suggest thattheindicativeconditional construction s such a device. It signalsrobustnesswith respect to its antecedent. Hence it is properto assert (P -* Q) when (P D Q) is highlyprobable and robustwith respectto P, that is, when Pr(P D Q/P) is also high. But,by analogy with explicit disjunction and "-or anyway-," thetruth onditionsof P - Q) are thoseof P D Q). It is like"Never-thelessP" in this regard. The use of "nevertheless" signals therobustnessof P with respect to what has gone before,but thewhole sentence is true if and only ifP is.At first lance it may appear that thisversionof the Equiva-lence thesis is totally opposed to those theories which assignconditionalsassertion nd acceptance, butnottruth, onditions.3But in factit is a half-wayhouse. Consider again "NeverthelessP." Although the whole is true ifand only ifP is true; a part-"nevertheless"-contributes to assertion conditions withoutaffectingtruth conditions. We can give the conditions underwhich it is proper to use "nevertheless," but not those underwhich using it is saying something true. Likewise with thesignalling role of the indicative conditional construction.Ourtheory s thusa supplemented Equivalence theory. n thewidestsenseof"meaning," (P - Q) and (P D Q) do not mean thesame.But theirtruthconditions are the same-they agree in sense or

'3 For recentexamples see ErnestW. Adams, TheLogicofConditionals,eidel:Holland, 1975, and J. L. Mackie, op. cit.576


14/26

ASSERTION AND CONDITIONALSliteral content.The extra element is that in using (P Q) youexplicitly signal the robustness of (P D Q) with respect to P,and this element affects ssertion conditions without affectingtruthconditions.We could have gone further and made the robustness of(P D Q) with respect to P a necessary condition for the truthof (P -* Q). But this seems, as a simple fact of linguistic usage,too strong. For, first,we allow that a person may speak trulyin the conditional mode without deserving o do so. Supposeit is highly probable that it will rain tomorrowand in conse-quence that the matchwill be cancelled. But, with the intentionofmisleading Fred, I say thatif trains,thematch will go ahead.In thiscase "It rains D the match will go ahead" isneitherproba-ble nor robust withrespect to "It rains." Furthersuppose that itdoes indeed rain, but against the odds, the match goes ahead.We allow that I have spoken trulywithout of course deservingto do so. And, secondly,we allow that onemember of the setof conditionals of the form "If I write down the number --,I will write down the number of molecules in this room" istrue. Yet none s robustwith respect to its antecedent.What is the point of signalling the robustness of (P D Q)with respect to P? The answer lies in the importance of beingable to use ModusPonens.Although (P D Q), P, . . Q s certainlyvalid, there is a difficulty bout using it in practice. Supposemy evidence makes (P D Q) highly probable but that I haveno evidence concerningP. Q s of interestto me, so I set aboutfinding evidence forP if I can. The difficulty s that findingevidence that makes P highly probable is not enough in itselffor me to conclude Q by Modus Ponens.For the evidence thatmakes P probable may make (P D Q) improbable. Indeed it iseasy to prove from the calculus that, except in special casesof extreme probability, Pr(P D Q/P) -c Pr(P D Q); that is,normallyon learningP I must owerheprobability give (P D Q)so endangeringthe inference oQ. t is thusofparticular interestwhether or not (P D Q)'s high probability would be undulydiminished by learning P; that is, it is importantwhether ornot (P D Q) is robust with respect to P. In sum, we must dis-tinguishthe validityfModusPonens rom tsutilityn a situation

577


15/26

FRANKJACKSONwhere I know (P D Q) but do not know P."4 The robustnessof (P D Q) relative to P is what is needed to ensure the utilityofModusPonens n such situations.It does not, though,ensurethe utilityof ModusTollens. r(P DQ/-P)can be high when Pr(P D Q/-Q) s low. And this is nowthingsshould be. You may properlyassert (P -* Q) when youwould not nfer P on learning-Q. Suppose you say "If he doesn'tlive in Boston, then he lives somewhere in New England" or"If he works,he will still fail," you will-despite the validityofModus Tollens-neither nfer hat he lives in Boston on learning(to your surprise)that he doesn't live in New England nor inferthathe didn't work on learning (to yoursurprise)thathe passed.Rather on learning either you would .abandon the originalconditional as mistaken. 5Of course it is not only the robustnessofmaterial conditionalswithrespectto their antecedents that is important.Accordinglyif our approach is along the right lines we should expect alinguisticdevice to signal the robustnessof Q with respect to P,notmerelyof P D Q) with respecttoP. But ifthe SupplementedEquivalence thesis s right, he latter s sufficient orthe former.Consider (Q & (P -* Q) ). According to the Equivalence thesisit is equivalent to Q. and according to our supplementation theright-hand conjunct signals that Pr(P D Q/P) is high. ButPr(PD Q/P)simplifies oPr(Q/P).Hence asserting Q & (P-*QJ)is equivalent to asserting Q and also signals the robustness ofQ with respect to P-just what we are looking for. When weassert both Q and (P -* Q) we commonly use a "still" con-struction:"The match will be played, and it will still be playedif it rains," "Carter will be reelected, and if the Camp Davidtalks fail,he will still be reelected." And often we don't botherto repeat the common element, Q. Context makes it clear thatwe thinkthat the match will be played or that Carter will bereelected, and we simply say "The match will still be played if

14 In my view the objection to Disjunctive yllogismn A. R. Anderson andN. D. Belnap, Entailment,Princeton University Press: New Jersey, 1977,conflates these two questions. Note particularly the top paragraph of theirp. 177.15 I here dissent from W. E. Johnson's illuminating remarks in ch. 3 ofLogic, Pt. I, Dover: N.Y., 1964. (Incidentally, saying P only ifQ does seem tosignal robustness of P D Q with respect to -Q.)

578


16/26

ASSERTION AND CONDITIONALSit rains" or "(Even) if the Camp David talks fail, Carter willstill be reelected." A strongerposition is that "If P, then stillQ"entailsQ.16 ut consider ne who makes, ll in one breath,the followingperfectly cceptable remark. "If it rains lightly,the match will still be played. But if it rains heavily, as it wellmay, the match will be cancelled." Surely he is not assertinginter lia that the match will be played.

4. DEFENSE.I take forgranted one negative argument for urSupplementedEquivalence thesis, namely that all its competitors face well-known objections. One obvious positive argument for it wouldconsist in assembling a large number of examples of indicativeconditionals and testing our intuitions concerning assertionagainst the results our theory predicts. Fortunately this is notnecessary. Ernest Adams has provided a simple formula gov-erning our intuitions, and the Supplemented Equivalencetheoryexplains this formula.Adams has shown that the (intuitively ustified) assertabilityof (P-iQ) is given by Pr(Q/P) = df Pr(PQ).17 Thus I assentPr(P)to "If itrains, the matchwill be cancelled" to the extent thatmysubjective probability of the match being cancelled given itrains is high.

16 Some have held the similar position that "Q even ifP" entails Q. See, e.g.,Mackie, op. cit., p. 72, and Pollock, op. cit., p. 29. I would advance a similarobjection to this position.7Adams, op. cit., and his earlier papers "The Logic of Conditionals,"Inquiry, , 2 (1965), 166-197 and "Probability and the Logic of Conditionals,"in Aspects of Inductive ogic, ed. J. Hintikka and P. Suppes, North-Holland,1966, pp. 265-316. Strong evidence that he is essentially right is the numberof authors of very differentphilosophical persuasions who have found thisgeneralkind of thesiscongenial. E.g., Brian Ellis, "An Epistemological Conceptof Truth," in Contemporaryhilosophy n Australia,ed. R. Brown and C. D.Rollins, Allen and Unwin: London, 1969; Richard Jeffrey, If," Journal ofPhilosophy, 1, 21 (Nov. 1964), 702-3; Robert Stalnaker, "Probability and Con-ditionals," Philosophy fScience,37, 1 (March 1970) 64-80, and David Lewisop. cit. Adams' formula does not of course take into account the kind of"local" sources of unassertability set to one side in section 1, like obscenity,rudeness, and long-windedness.

579


17/26

FRANK JACKSONWe explain Adams' thesis as follows. On our theory, theassertabilityof (P -* Q) will be the product of two factors:theextent to which Pr(P D Q) is high and the extent to which(P D Q) is robustwithrespectto P. Butwe have from hecalculusthat Pr(P D Q/P) = Pr(Q/P), and that Pr(P D Q > Pr(Q/P).Consequently both conditions are satisfied to the extent thatPr(Q/P) is high. Q.E.D.An important recent result of Lewis's highlights the sig-nificanceofthisderivation.He proves hattheobvious alternative

explanation of Adams' thesis fails. He proves (by a reductionargument) that (P -* Q) does not differ n truth conditionsfrom P D Q) in such a way as to make Pr(P D Q) = Pr(Q/P). 18Consequently, we can explain why (P -* Q) and (P -* -Q)are not assertable togetherwhen P is consistent.Pr(Q/P) andPr(-Q/P) cannot (fromthe calculus) both be high.19Or, moreprecisely,they cannot both be high relative to the same bodyof evidence. Robustness, like probability in general, is relativeto evidence, and of course Pr(Q/P & R) and Pr(-Q/P & S) canboth be high.Accordinglyour theorypredictsthatwe should behappy to assertboth (P-*Q) and (P-*-Q) when it is explicit hatthe relevant bodies of evidence are appropriately different.Exactly thishappens. Harry and George are discussingwhetherFred went to the rock concert. Harry says "If Fred went, hemust have gone by car, because there was a transportstrikeat the time." George says "But Fred regards the private caras exploitative and never goes anywhere by car on principle;so ifhe went, it cannot have been by car." They conclude "Inthatcase, obviouslyFred did notgo to therock concert." Insteadof regarding their statements as mutually inconsistent,Harryand George draw from them the conclusion that Fred did notgo.20

Our theory, then, makes highly assertable just those con-ditionals intuition judges to be highly assertable. But whatof our intuitive judgements of validity? I am committed to18 Op. cit., pp. 300-2.19When P is inconsistent,Pr(P) = 0, and Pr(Q/P) is undefined; accordinglywe need a ruling about the assertabilityof (P -) Q in such cases. The ruling Iwill follow is that all such conditionals are assertable. Others are possible.20 For otherexamples of this kind see Clark, "Ifs and Hooks: A Rejoinder,"

op. cit.580


18/26

ASSER TION AND CONDITIONALStaking -P .. P- Q and Q . . P - Qto be valid, and theynotoriously ack intuitive appeal. But this lack of appeal seemsto derive fromour reluctance to assert (P -* Q) merelybecausewe are confident that -P and our (less marked) reluctance toassert P -* Q) merelybecause -we re confident thatQ, and ourtheory can explain these easily enough. Neither the fact thatPr(-P) is high nor the fact that Pr(Q) is high is sufficient orPr(P D Q/P) being high. The reason our reluctance is less markedin the case of asserting (P -* Q) on the basis of our certaintythat Q. is that Pr(Q) being high togetherwith P and Q beingprobabilistically independent is sufficient for Pr(P D Q/P)being high.Similarly, what I referredto earlier as the third paradox-that ( (P -* Q) v (Q * R) ) is a logical truthand yet is farfrominvariably highly assertable-is not a decisive objection to oursupplementedversionof theEquivalence thesis,because thepres-ence of signals can make logical truthsunassertable. We havealready noted the plausibility of giving "Nevertheless P" thesame truth-conditions as P. Consequently "NeverthelessP or neverthelessnot P" is a logical truth, but it is not highlyassertable.What of StrengtheningheAntecedent,ypotheticalyllogism,ndContraposition,ll of which are of course valid on our theory.Take Contrapositionsimilar points apply to all three). The prob-lem is not that it seems invalid stated in symbols; exactly thereverse is the case, as is evinced by its appearance in NaturalDeduction systems.The problem is rather a certain class ofapparent counterexamples like: "If George workshard, he will(still) fail; therefore,fhe passes, he won't have worked hard,"and "If Carter is reelected, it won't be by a large margin;therefore f Carter is reelected by a large margin, he won't bereelected."But these apparent counterexamples are paralleled by onesagainst the commutativityof "- or anyway -": for instance,"It won't rain or anyway not heavily; therefore, t won't rainheavily or anyway it won't rain." Despite this,we noted that itseemed clearly rightto give the same truth-conditions o "P oranyway Q" as are standardly givento "P or Q." The explanationforthe counterintuitivefeel must therefore ie not in the failure

581


19/26

FRANKJACKSONofCommutationut in the failureofwhat is signalled by "anyway"to "commute." SimilarlyAdditions hardly appealing when ap-plied to "- or at least -." Consider "Harry said that Caesardefeated Pompey in 48 B.C.; thereforeHarry said that CaesardefeatedPompey in48 B.C. orat least Caesar defeatedPompey in48 B.C."It seems thereforenot unreasonable to attribute the counter-intuitive feel of certain instances of Contrapositiono the failureofwhat is signalled by the indicativeconditional constructionto"contrapose" (and likewiseforHypotheticalyllogism,nd so on).And itmaybe confirmedby inspectionthat theputativecounter-examples to Contrapositionre all ones where Pr(P D Q/P) =Pr(Q/P) is high,and Pr(-Q D -P/-Q) = Pr(-P/-Q) is low. For ex-ample, the probabilityofCarter not being reelected by a largemargingiven he is reelectedmay be highwhen theprobabilityofCarternotbeing reelectedgivenhe is reelected by a largemarginis minimal. Accordingly,we can explain our reluctance to assert"If Carter is reelectedby a largemargin, thenCarterwill not bereelected" even when we are happy to assert "If Carter is re-elected,thenit will not be bya large margin" in terms,not ofthefirst eing falseand the second true,but in terms of what is sig-nalled by saying the firstbeing false and what is signalled bysaying the second being true.

5. ON THREE OBJECTIONS.(i) It may be objected that the account offered bove is cir-cular. The Equivalence thesis tself s notcircular,obviously,butthe supplemented thesis involves a storyabout the role of theindicative conditional construction s signallingrobustness, nditmight be objected thatrobustness an only be elucidated via aconditional construction.P is robust forperson S relative to I

just ifP's high probability forS would not be substantially re-duced if were o acquire theinformation hat . One replywouldbe to urge that we simply define robustness in terms of con-ditional probability, Pr(P/I) = df Pr(PI) No conditionalsPr()there.This reply is open to challenge. For the defense ofso definingrobustness must involve a defense of Conditionalization, the582


20/26

ASSERTION AND CONDITIONALSthesisthat theimpact ofnew information sgivenbytherelevantconditional probability,and it might be urged that talk of theimpact of new informationcan best be understood as talk ofwhatone's probabilitieswould orshould be ifone weretoacquirethe new information.But another reply is possible. The condi-tionals involved here are essentially subjunctive and counter-factual in character,and as such are importantlydistinctfromindicative conditionals. It is not, therefore, selesslycircular toappeal to the former n one's storyabout the latter.

I take the case forseparating out the problem of indicativeconditionals fromthe problem ofsubjunctive conditionals to befamiliar.21 t derives frompairs like "If Carter is bald, no oneknows t" and "If Carterwere bald, no one would know it," and"If Oswald did not shoot Kennedy, someone else did" and "IfOswald had not shot Kennedy, someone else would have." Foreach pair we assent to the first nd dissent fromthe second, andour dissent from hesecond member ofeach pair is accompaniedby assent to, respectively, If Carterwere bald, everyonewouldknow it" and "If Oswald had not shot Kennedy, no one wouldhave."It sometimes seems to be thoughtthat the contrastbetweenindicative and subjunctive or counterfactualconditionals onlyshowsup when (i) theconsequents are knownne way or the otherand (ii) they are not about the future.22 ut here is a pair in-volving a future,doubtful event. I have been told that Fred'sbirthday and George's birthday fall next week. But I cannotremember the exact days, only that Fred's is the day beforeGeorge's. I say "If Fred's is nextTuesday, George's will be nextWednesday," and I don't say "If Fred's were next Tuesday,George's would be next Wednesday" (unless of course I knowthat the hospital specially arrangedGeorge's birthone day afterFred's, or somethingof thatkind). Likewise,suppose thatstationB is half-waybetweenA and C. A personat B observing a train

21From, e.g., David Lewis, Counterfactuals,lackwell: Oxford, 1973, andErnest Adams, "Subjunctive and Indicative Conditionals," FoundationsofLanguage, 6 (1970), 89-94.22 See, e.g., Brian Ellis, "A Unified Theory of Conditionals," JournalofPhilosophicalLogic, 7, 2 (May 1978), 107-124, and (with reservations) Adams,

The Logic of Conditionals,p. cit., ch. 4.583


21/26

FRANKJACKSONpassing throughontimemay well affirm If the train had beenlate leavingA, it would be late pulling intoC" while denying"Ifthe train was late leaving A, it will be late pulling into C."It is nevertheless undeniable that the contrast between in-dicative and counterfactualconditionals is less marked in thecase of conditionals pertaining to the future.For example, beforethe assassination of Kennedy we would say both "If Oswaldwerenot to shoot Kennedy, no one would" and "If Oswald doesnot shootKennedy, no one will." It is only now, after the event,that we say "If Oswald did not shoot Kennedy, someone elsedid." But this is a point in favor of our theory, or t can explainwhy the contrast is less marked in the case of the future.It is a fact that we know more about the past than about thefuture. know moreabout who won lastyear's election than I doabout who will win next year's. In particular,our beliefs aboutthe futureby and large depend on relatively tenuous beliefsabout what presentand past conditions will give rise to, whileour beliefs about the past are frequently ndependent of ourbeliefs about how the past came about. My beliefsabout nextyear'selection winner reston mybeliefs bout presentconditionsand their effect n electoral popularity,while my beliefs aboutlast year's winner are byand large independent ofmy views as towhat led to her success. Predicting election winners calls foratheory f what makes forelectoral popularity,retrodicting hemonly calls for an ability to read the newspapers. Consequentlythe probabilities we assign future events depend on our viewsabout whatwould lead towhat. But by our theory t is theseveryprobabilities thatsettlethe indicative conditionals we assertandit is these veryviews about what would lead to what that areexpressed in subjunctive and counterfactual conditionals. 3Hence the general match between the two in the case of the fu-ture.On the otherhand, the probabilities we assign past eventsmay be largely ndependent ofour views about what would leadtowhat. When Q is past we may givePr(Q/P) a highvalue inde-pendentlyof what we believe gave rise to it, and so may assert(P -* Q) largelyindependently of our stance on counterfactuals

3 I argue this n detail in "A Causal Theory of Counterfactuals," AustralasianJournalof Philosophy, 5, 1 (May 1977), 3-21; but the general idea is widelyaccepted.

584


22/26

ASSERTION AND CONDITIONALSoftheform(- - 9Q),ncludingnparticularPEI-*.Q).Itmight lsobe thoughthat he ontrastetweenndicativeandsubjunctiveonditionalsssimplyue tothedifferentoleofwhat sbeing aken or ranted,resupposedrregardedscommon nowledgenthe ontext.24henweconsiderndica-tive onditionals e"holdon to"common nowledge,henweconsiderubjunctiveonditionalseneednot.25t is taken orgrantedhat omeonehotKennedy, ence venunder he n-dicativelyxpresseduppositionhatOswald did not shootKennedy, e holdthat omeoneelse)did.But under he ub-junctivelyxpresseduppositionhatOswaldhadnot,wemayabandon hispresuppositon.ikewise ith heother xamplesgiven. t iscommon nowledgehatCarter snotbaldand also(we weresupposing)hatFred'sbirthdays theday beforeGeorge'snd that hetrainwason time t B, and these actswerewhatwasbeing etained hen he ndicativeserenques-tion ndbeing bandonedwhen he ubjunctiveserenques-tion.However,he everseappens ith therxamples. ouandhavebeentaking hedateofCaesar's efeat fPompey s com-monknowledge;nd it is ustthisweholdon to in assertingthe ubjunctiveIfthehistoriansadreportedhedateofCaesar'svictorys 50 B.C., theywouldhave beenwrong"notevenhistoriansan change hepast), nd it is just thiswe aban-don nassertinghe ndicativeIfthehistorianso reporthedate s50B.C.,then amwrongngivingt s 48 B.C."Perhapsit willbe objected hatwhatwas taken s common nowledgewasthat hehistoriansave hedateright,ather han hedate

24 I am indebted here in particular to the refereefordrawing my attentionto Robert Stalnaker, "Indicative Conditionals," in Language in Focus, ed. A.Kasher, Reidel: Holland, 1976.25 Stalnaker, ibid., pp. 182-7, shows how to expressthisin terms of the famil-iar possible-worlds approach to counterfactuals (due to him, "A Theory ofConditionals," in Studies n Logical Theory, d. N. Rescher, Blackwell: Oxford,1968, and Lewis, Counterfactuals,p. cit.). According to Stalnaker thisapproachworksforboth indicative and subjunctive conditionals, the difference etweenthe two being due to the fact that in the case of the formerbut not the latterthe similarityrelation is constrained by the need to preserve common knowl-edge. When we consider (P - Q) at world i we are to look for the closest P-world which hareswith whatis being aken o becommonnowledgen thecontextfassertionnd ask whether t is a Q-world.But see thecounterexamples below.

585


23/26

FRANKJACKSONitself.But then my point can be made with a differentpair.We assert If Caesar's victorywas in 50 B.C., the historians havethedate wrong,"while denyingIf Caesar's victoryhad been in 50B.C., the historianswould have got the date wrong." Likewise,it is common ground that the declared winner of a presidentialelection is the person with the most votes. Yet it is just this weare prepared to abandon when we consider indicative condi-tionalsstarting If Ford got more votesthanCarter,. . . ", and justthis we hold on to when considering subjunctive conditionalsstarting"If Ford had got more votes than Carter, . . .(ii) Thus farI have focussed on explaining assent patternstoconditionals in terms of the Supplemented Equivalence theory.What of dissent? Standardly you dissent from an assertion ustwhen its subjective probability of falsity s high (neglecting, asbefore,highly contextual factors ike obscenity). The probablefalsity f what may be signalled by the assertion is by and largeirrelevant.26You dissent from He is poor but happy" just whenit is probable that he is either not poor or not happy, not whenyou dissent fromthe signalled contrast.Dissent is typicallydis-sent from what is literally said. But it is clear and generallyacknowledged that we dissent from conditionals in circum-stances other than thosewhere it is probable thatthe antecedentis true and the consequent false. Even anti-Warrenitesdissentfrom If Oswald killed Kennedy, then the Warren Commissiongot thekiller's dentitywrong;" but theydo not regard"Oswaldkilled Kennedy and the Warren Commission got the killer'sidentity ight"as highlyprobable. They thinkratherthatOswalddid not kill Kennedy and that the Warren Commission got thekiller'sidentitywrong.There are, however, exceptions to the rule that we dissentjust when it isprobable thatwhat is literally aid isfalse.SupposeI say in a serioustone ofvoice "I believe that it will rain tomor-row."There are two ircumstances nwhich younaturallydissent.One is whenyou think am lying, hatis,when theprobabilityoffalsity fwhat is literallysaid is high (by yourlights).The otheris when you think t will not rain. In thiscase your dissentis notfromwhat I literally ay but fromwhat I signal by saying it in aserioustone ofvoice,namely,that tsraining shighlyprobable.26 Cf. Cohen's report of Grice's views, op. cit.

586


24/26

ASSERTION AND CONDITIONALSAnother example is when I say "The winner of the electionfor lub presidentwill come fromTom, Dick and Harry." What

I say counts as trueifanyone of these three wins. But you won'tdissentonlyifyou thinkthis mprobable. You may grant tprob-able because of the excellent chance Tom has of winning butneverthelessdissent because I leftout George, and in your viewGeorge has the best chance afterTom. In otherwords, you willdissent when whatyou would assert is that the winner will comefrom Tom and George, and not only when you are prepared tosay that none of Tom, Dick and Harry has a chance.The explanation for these two cases being exceptions to therule that dissent is prompted by low probabilityof literal truth,appears to lie in thepeculiarlyintimaterelationshipthat obtainsin them between what is said and what is signaled. In the secondexample what is signalled is sufficient or the high probabilityofwhat is said. In saying that the election is out of Tom, Dickand Harry, signal that the highprobabilityforme ofthe tripledisjunctionis robustwithrespectto the conjunction of the nega-tions of any two of the disjuncts (forexample, that it is highlyprobable that Dick will be elected given that Harry won't andTom won't). This is sufficientby the calculus) for he high prob-ability of the disjunction. In the first xample what is signalledis arguably sufficient or the truth of what is said. If I say "Ibelieve it will rain tomorrow" in an appropriately serious toneit is arguable that I signal that I do indeed believe it will raintomorrow,and I am not, say, producing the sentence merelyas a handy example of a belief-sentence. At any rate, it canhardlybe denied that there is a close connection between whatis said and what is signalled in this case.According to our account, conditionals are yet another ex-ample of thesame generalkind. What issignalled by the assertionof (P--Q) amounts to Pr(Q/P) being high. This is sufficient orPr(P D Q) being high. So what is signalled is sufficient or thehigh probabilityof what is literallysaid. Hence, drawing on themoral of the two examples just discussed, dissent from P ->Qmay be prompted by the dissentergiving a low value to Pr(Q/P)as much as by his giving a low value to Pr(P D Q). Moreover,the latter s sufficient orthe formerby the calculus, so all casesof dissent from P--Q) are ones where Pr(Q/P) is low. This re-

587


25/26


26/26

ASSERTION AND CONDITIONALSdistinguish ruth onditionsfrom ssertionconditions.The truthconditions for (P -- Q) are those of (P D Q). There are goodand well-known rgumentsforthis.'And the assertionconditionfor(P->Q) is that Pr(Q/P) be high. There are good and well-known argumentsforthis. End of story.My reason is that conjoining is not explaining. The problemis to explain one in termsof the other. And given the widelyaccepted view that the best approach to meaning and analysis isvia truth onditions,we should hope for theorywhich explainsthe assertionconditions in termsofthe truthconditions. This isessentially what I have attempted. I have tried to show how aplausible thesis about (P -a Q)'s truthconditions, namely theEquivalence thesis, an, in the lightof the importanceofrobust-nessfor ssertability, xplain theplausible thesisabout (P - Q)'sassertioncondition, namely Adam's thesis.In my view this puts a verydifferent omplexion on certainputativecounterexamplesto the Equivalence thesis.We saw, forinstance, how granting the validity of Contrapositionan forcethe equivalence theorist nto holding that "If Carter is reelectedby a largemargin, thenhe will notbe reelected" is true.But whatis itthat s immediatelyvidentbout thisputativecounterexample?Surely that it has verylow assertability.But the supplementedEquivalence theory xplains his, nd what a theorywell explainscannot be an objection to that theory.27MonashUniversity

27 I am conscious ofa more than usually large debt to many discussions withBrian Ellis, Lloyd Humberstone, and Robert Pargetter; and also to commentsfromthe referee.

Frank Jackson - On Assertion and Indicative Conditionals

Documents

Transcript of Frank Jackson - On Assertion and Indicative Conditionals