We will suppose that at the time of utterance both parties are alive .sx What is required for the truth of this conditional ?sx Following Stalnaker's idea , it is true iff in the situation as close as possible to ours but in which John dies before Joan , Joan inherits the lot .sx Thus if , for example , John's will actually leaves everything to Joan , a situation as close to ours as possible is one in which his will is unchanged , and hence a situation in which , given the normal flow of events , Joan will inherit the lot .sx ( If John is going to die before Joan , then the situation to consider is the actual future course of events :sx nothing is more similar to the actual world than itself .sx ) In this case , Stalnaker's theory seems to give precisely the right result .sx It also delivers the right result in the case of ( 1.21 ) , .sx If someone broke in , they must have repaired the damage before they left , .sx which , arguably , has to be treated as undefined by the probability theory .sx Consider a world in which someone did break in , but which is otherwise as similar as possible to ours .sx In ours , there is no trace of violent entry .sx How could this be , unless the intruder had repaired the damage behind him ?sx So the closest ( that is , most similar ) world in which the antecedent is true is one in which the consequent is true , so the conditional is true ( and hence can be used to infer that no one broke in , given that no intruder would stop to repair the damage behind him) .sx Stalnaker explicitly considers , and rejects , the theory mentioned in chapter 2.4 that a conditional requires for its truth some connection between what would make its antecedent true and what would make its consequent true .sx He offers the following counterexample :sx suppose that at the time of the Vietnam conflict you are convinced that the United States will use nuclear weapons , come what may .sx You therefore also think that the Chinese entering the conflict would have no effect whatsoever on the US decision to use nuclear weapons .sx Yet you will confidently affirm the conditional :sx 1 ) If the Chinese enter the Vietnam conflict , the United States will use nuclear weapons .sx Notice that on Stalnaker's theory this conditional is , if your political views are correct , true , because any smallest revision of the actual world which verifies the antecedent will of course verify the consequent .sx But this is not to say that the truth of the consequent is in general sufficient for the truth of the conditional .sx On the contrary , this will not always hold :sx not , for example , in conditionals of the form 'if not- A then A' .sx What impact does Stalnaker's account of the truth conditions of conditionals have upon validity ?sx It turns out that his account of truth conditions validates ( in the standard sense ) just the arguments that are valid in the probabilistic sense .sx He agrees with the probabilistic theory in finding ( 2.8.1 ) invalid , and he offers a sophisticated explanation of the appearance of validity ( Stalnaker [1980]) .sx His idea is that ( 2.8.1 ) corresponds to a 'reasonable' inference :sx one such that , in any context in which the premises are acceptable , so is the conclusion .sx But reasonable inferences can be invalid , and ( 2.8.1 ) is an example .sx If we think it is valid , it is because we mistake reasonableness for validity .sx Everything hinges , of course , on the account of reasonableness .sx Without giving Stalnaker's whole theory , let us simply see how it is intended to apply to the example .sx A context in which the premise of the argument is acceptable is one which leaves open whether it was the butler or the gardener who did it .sx ( If you knew who did it , you shouldn't assert the premise , but simply 'the gardener did it' , or 'the butler did it' , as the case may be .sx ) When you have asserted or accepted the premise , the context includes the premise :sx it is now part of what is taken for granted .sx Hence when you evaluate the conclusion , you should consider a world in which the gardener didn't do it , but in which the premise holds true .sx Such a world must be one in which the butler did it .sx Reasonableness puts restrictions on what alternative worlds to consider in evaluating a conditional .sx Even if reasonable , the argument is invalid , by Stalnaker's account .sx For there is no guarantee whatsoever that in the world most like ours , assuming that ours is one which verifies the premise , but in which the gardener did not do it , the butler did .sx One possibility is that our world is one in which the butler and the gardener were in league :sx they both did it .sx The most similar world to this verifying 'The gardener did not do it' could well be one in which the victim was not murdered at all :sx a world in which , say , the victim fell ( rather than being pushed) .sx The relevant world cannot be exactly like our world , since in our world ( we are supposing ) the butler and gardener both did it .sx A world in which the butler-gardener conspiracy broke down might well be less similar to ours than one in which the conspiracy is in force , but the conspirators were thwarted at the last moment by their victim's accidental death .sx The truth of the premise of the argument thus does not guarantee the truth of the conclusion , upon Stalnaker's semantics .sx Stalnaker intends his account to apply to subjunctive as well as indicative conditionals .sx But how will it specify the difference between matched pairs of these different kinds of conditional ?sx For example .sx 2.4.34 ) If Oswald didn't shoot Kennedy , someone else did .sx 2.4.35 ) If Oswald hadn't shot Kennedy , someone else would have .sx Let us apply Stalnaker's recipe to these successively , retaining the background assumptions we relied upon in chapter 2.4 to justify the reasonableness of accepting the first while rejecting the second .sx For ( 2.4.34 ) we consider a world in which Oswald didn't shoot Kennedy .sx Remember that Oswald's guilt was a disputed question , but Kennedy's death from shooting was not .sx So a similar world will be one in which Kennedy died from a shooting , while not having been shot by Oswald .sx Such a world , obviously , verifies the consequent .sx Hence Stalnaker's account has it that ( 2.4.34 ) is true , and this is as it should be .sx What should we consider for ( 2.4.35)? Although Stalnaker is not explicit , he clearly intends us to consider a world in which Oswald did not shoot Kennedy .sx ( In other words , we revise the antecedent from 'hadn't' to 'didn't' .sx ) So it would appear that we have to consider just the same world as we did before .sx And just as we must revise the antecedent , to apply Stalnaker's recipe , so we have to revise the consequent .sx What we need to consider , then , is whether in a world in which Oswald did not shoot Kennedy , but which is otherwise as like as possible to this , someone else shot Kennedy .sx In short , we seem to have the same question before us as we did in the case of ( 2.4.34 ) , in which case we should give the same answer .sx But this answer , that ( 2.3.35 ) is true , conflicts with our original intuition .sx ( We imagined that we thought Oswald acted alone , there was no conspiracy , so if he hadn't shot Kennedy no one else would have .sx ) .sx Stalnaker recognizes that there will be considerable context-sensitivity in point of what counts as the world most similar to ours in which the antecedent holds , and he suggests that this can account for the difference between ( 2.4.34 ) and ( ) .sx What remains unclear , however , is the relationship , which presumably ought to be systematic , between the contexts we envisage for ( 2.4.34 ) and ( 2.4.35 ) , and the different dimensions of similarity relevant to the truth conditions .sx For ( 2.4.35 ) we want to count a world in which Oswald did not shoot Kennedy and Kennedy was not shot as the most similar to ours ; for ( 2.4.34 ) we want to count a world in which Oswald did not shoot Kennedy , and he was shot , as the most similar to ours .sx But what features of the context make for this difference ?sx Clearly a vital difference is that in ( 2.4.34 ) it is taken for granted , as part of the background , that Kennedy was shot .sx In ( 2.4.35 ) this background assumption somehow gets suspended in selecting the relevant world .sx But what is the mechanism ?sx It is certainly relevant that the envisaged use of ( 2.4.35 ) , unlike ( 2.4.34 ) , takes for granted that Oswald in fact did shoot Kennedy , yet any world which will satisfy Stalnaker's test will be one in which the presupposition is abandoned .sx Perhaps abandoning taken-for-granted facts induces much more relaxed standards of similarity .sx A world as similar as possible to this one but in which Oswald didn't shoot Kennedy is certainly not this world , in the context of ( 2.4.35 ) , but is not certainly not this world in the context of ( ) .sx The more obvious account of 'most similar' in this case brings out both the indicative and the subjunctive as true .sx This is not invariable .sx Gibbard [1980] has suggested the following case , designed to show that Stalnaker's account works best for subjunctive conditionals , and the probability theory for indicative ones .sx Jack and Stone are playing poker and Stone has just bet the limit in the final round .sx Zack has seen Stone's hand , seen that it is a good hand , and has signalled its contents to Jack .sx Zack remains unaware of the contents of Jack's hand .sx Stone , suspecting mischief , orders the room to be cleared .sx Not knowing the outcome of the game , Zack can confidently assert later :sx 2 ) If Jack called , he won .sx His grounds are that Jack is an experienced player , wants to win , and , knowing Stone's hand , knows whether or not he will win by calling .sx However , Zack can reasonably be doubtful about .sx 3 ) If Jack had called , he would have won .sx After all , Stone's hand was good , and it's not likely that Jack had a better .sx Let us first confirm Gibbard's claim that the probability theory gives an acceptable account of ( 2 ) , and then see how Stalnaker's account deals with the two sentences .sx On the probability theory , ( 2 ) is assertible iff it has high probability , which by ( 1.4 ) , means that Pr ( Jack won|Jack called ) is high .sx The result is that ( 2 ) is ruled highly assertible , as it seems to be , since , given that Jack called , it is extremely likely that he won .sx Turning now to ( 3 ) , Gibbard argues that our intuitions harmonize with Stalnaker's account .sx We in fact regard the truth or falsehood of ( 3 ) as determined by whether or not Jack's hand was better than Stone's .sx On Stalnaker's account , we must find a world as similar as possible to ours , but in which Jack called .sx According to Gibbard , such a world is determined by the following criteria :sx it is exactly like the actual world until it is time for Jack to call or fold ; then it is like the actual world apart from whatever it is that constitutes Jack's decision to call or to fold , and from then on it develops in accordance with natural laws .sx ( pp .sx 227-8 :sx I have changed Gibbard's Pete to Jack ) .sx However , it seems that a world in which Jack calls , and which is maximally similar to ours , is a world in which Jack knows Stone's hand and calls .sx Preserving similarity with our world , this can only be because Jack knows that he has a winning hand .sx In this world , therefore , the consequent of the conditional is true , rather than false , and so , contrary to Gibbard's claim , Stalnaker's account seems not to deliver the desired result for ( 3) .sx Gibbard envisages Zack's response to ( 3 ) as occurring when Zack is still ignorant of what Stone did .sx The most natural setting for ( 3 ) , however , is one in which it is known that Jack folded .sx Given that he folded because he knew Stone had the better hand , Stalnaker's account clearly works well .sx The alternative world is like our world up to the time of Jack's decision .sx Before then , he knows he has a losing hand .sx