Dispositions and Generics

Dispositions and Generics Ryan Wasserman Western Washington University Quine defined dispositions twice over. In Word and Object, he endorses a version of the conditional analysis: The subjunctive conditional is seen at its most respectable in the disposition terms. To say that an object a is (water-) soluble at time t is to say that if a were in water at t, a would dissolve at t. To say that a is fragile at t is to say that if a were struck smartly at t, a would break at t. (1960: 222–3)1 In “Natural Kinds”, he advocates a different approach: Intuitively, what qualifies a thing as soluble though it never gets into water is that it is of the same kind as the things that… dissolve. (1969: 16) This passage suggests a generic analysis of dispositions, as it appeals to natural kinds or genera and to general truths about the behavior of those kinds.2 This paper explores Quine’s second idea as an alternative to his first. I begin by reviewing some familiar problems for the conditional view (section 1). I then discuss a simple version of the generic analysis and show how it avoids those problems (section 2). Finally, I consider some *The published version of this paper appears in Philosophical Perspectives 25: 425-453. The common version of the conditional approach is also endorsed by Goodman (1954) and Ryle (1949), although Ryle’s endorsement comes with immediate qualifications. 2 Quine goes on to argue that there is an important link between natural kinds and subjunctive conditionals (1969: 17), in which case his two approaches may be closely related. I suggest a similar conclusion in section 6. See fn. 57. 1 1 more sophisticated versions of the view (sections 3-5) and assess the overall prospects for the approach (section 6). 1. The Conditional Analysis The guiding idea behind the conditional approach is that disposition ascriptions are to be analyzed in terms of conditional truths. But there are many different ways of linking dispositions to conditionals, and thus many different versions of the conditional approach.3 Quine begins with the claim that something is water-soluble just in case it would dissolve in water. More carefully: something is disposed to dissolve in water (at a particular time) if and only if it would dissolve, were it placed in water (at that time).4 Generalizing on this idea, we have the simplest version of the conditional approach: The Simple Conditional Analysis: N is disposed to M when C iff N would M if C.5 In this schema, N is to be replaced with a noun phrase, M is to be replaced with a verb phrase (one which picks out a manifestation), and C is to be replaced with a sentence or—more often—a verb phrase (one which specifies a stimulus condition). So, for example, the simple account says that the vase is disposed to break when dropped if and only if the vase would break if dropped.6 3 For a more detailed discussion of the various possibilities, see Manley and Wasserman (2008). 4 Henceforth, references to times (and worlds) will remain implicit. 5 Here, I follow the established convention of formulating the analysis in terms of a biconditional, which is assumed to hold of necessity under every assignment. But I also note that the convention is mistaken, since an analysis involves more than a modal claim. The idea behind this analysis is that facts about dispositions are grounded in or obtains in virtue of the corresponding conditional fact. This dependence claim may imply a modal thesis but it obviously goes beyond that—after all, one could agree that there is a necessary connection between dispositionality and conditionality, but insist that this is so because counterfactual truths are grounded in facts about dispositions, rather than the other way around. 6 Following Lewis (1999), we can distinguish between what we might call ‘explicit dispositional predicates’ and ‘implicit dispositional predicates’. Explicit dispositional predi- 2 The simple conditional analysis gets many cases right, but there are also some famous counterexamples, like the ones described by C.B. Martin (1994). Here is a variation on one of his cases7: suppose that a sorcerer takes a liking to my favorite crystal vase. The sorcerer wishes to guard against destruction, so he forms the following intention: if the vase is dropped, he will magically strengthen its intrinsic structure, so that the vase will bounce harmlessly off the ground. In that case, (1) is true and (2) is false. (1) The vase is disposed to break when dropped. (2) The vase would break if dropped. David Lewis refers to this problem as the problem of finkish dispositions.8 Lewis’s response to this problem is to revise the view. He begins by pointing out that, in the imagined case, the sorcerer would prevent the vase from breaking by removing the intrinsic causal basis of the vase’s fragility. We can therefore avoid the problem by requiring that the causal basis remains intact. Thus: Lewis’s Conditional Analysis: N is disposed to M when C iff N has some intrinsic property B such that, if it were to retain B when C, it would thereby M.9 cates include the term ‘disposed’ (or some variant of that term) whereas implicit ones do not. The reduction of implicit ascriptions to conditionals takes place in two steps. First, we connect the implicit ascription to an explicit counterpart (e.g. something is fragile if and only if it is disposed to break when dropped). Second, we connect the explicit dispositional ascription to a conditional statement (e.g. something is disposed to break when dropped if and only if it would break if dropped). 7 The variation is due to David Lewis (1999: 138). 8 The name is doubly-appropriate. When used as an intransitive verb, “fink” is ambiguous between “To inform against another person” and “To withhold promised support or participation” (The American Heritage Dictionary of the English Language, Fourth Edition). In Martin’s original example, a dead wire is attached to a device that detects when the wire is about to touch a conductor. At that point, the device alters the intrinsic structure of the wire so that it will conduct electricity. This is a fink in the first sense—it is an “informant” about the upcoming contact. The wire, on the other hand, is a fink in the second sense—owing to the presence of the electro-fink, the dead wire withholds the “promised” behavior and acts in the opposite way (by conducting electricity). 9 This is a slight simplification of Lewis’s proposal. For his official statement, see Lewis (1999: 149). 3 In this schema, B is to be the causal basis of the relevant disposition. For example, if my vase were to retain its delicate crystalline structure when dropped, it would thereby break. Hence, Lewis’s conditional analysis says that the vase is disposed to break when dropped, even when it is in the presence of a protective sorcerer. But now suppose that the sorcerer alters his intentions. If the vase is dropped, its intrinsic structure will go untouched, but the meddling magician will alter the environment by turning the hard floor into a soft pillow. In that case, (1) is still true but (3) is now false: (1) The vase is disposed to break when dropped. (3) The vase has some intrinsic property B such that, if it were to retain B when dropped, it would thereby break. In the imagined case, the vase would not break when dropped, even if it retained all of its intrinsic properties—including its disposition to break when dropped. Rather than removing this disposition (as in the case of finkish dispositions), the sorcerer prevents that disposition from revealing itself. Mark Johnston (1992) calls this the problem of masked dispositions.10 One might try to avoid this problem by revising the revised account. That is certainly an option. There are more versions of the conditional analysis to consider, and more counterexamples to uncover.11 But there are also more general worries about the conditional approach. I will briefly mention three. First, there is the problem of degrees.12 Dispositional predicates take degree modifiers, as in (4). (4) a. The vase is very fragile. b. The professor is somewhat irritable. 10 See also Smith (1977) and Bird (1998). Johnston, for example, suggests a modification intended to avoid the problem of masks. Manley and Wasserman (2008) discuss this proposal and review several other suggestions. For a detailed discussion of the problem of masked dispositions, see Fara (2005). 12 This problem and the next are developed in Manley and Wasserman (2007) and Manley and Wasserman (2008). 11 4 c. The vase is highly disposed to break when dropped. The problem is that whether or not a given counterfactual is true is not a matter of degree. It might be true that the vase would break if dropped, but it is not as if that counterfactually is very true, as opposed to just true. So, on the face of it, the conditional approach lacks the resources to make sense of statements like (4a), (4b), and (4c). Second, there is the problem of comparatives. In addition to taking degree modifiers, dispositional predicates take the comparative form, as in (5): (5) a. The vase is more fragile than the mug. b. The professor is less irritable than her husband. c. The vase is more disposed to break when dropped than the mug. Suppose that the vase and the mug are both fragile and that both of the corresponding conditionals are true (both the vase and the mug would break if dropped). It is not the case that one conditional is somehow more true than the other—both are simply true. But, in that case, it is unclear how the conditional approach can make sense of statements like (5a), (5b), and (5c). The problem of degrees and the problem of comparatives are closely related. On the standard semantics for gradable adjectives like ‘rich’, one begins with the comparative form: ‘is richer than’.13 That predicate provides a partial ordering of individuals with respect to their comparative wealth. That ordering gives us a scale, so that we can say that an individual is rich just in case she or he meets a certain threshold on that scale (where this threshold is largely determined by context). Moreover, we can understand degree-modifiers like ‘somewhat’ or ‘highly’ as adjusting the threshold that must be met for the positive predicate to apply. If we apply the standard approach to the current case, we will begin with a comparative predicate like ‘is more fragile than’ or ‘is more disposed to break than’. That predicate will predicate will provide an ordering, and thus a scale, so that we can say that something is fragile or dis13 On the standard account, see Klein (1989), Schwarzschild and Wilkinson (2002), and von Stechow (1984). 5 posed to break just in case it meets a certain threshold on the scale. In this way, the positive predicate is analyzed in terms of the comparative form. The problem is that, on this way of looking at things, the conditional approach is in trouble right from the start, since the analysis has the wrong analysandum. What we need is an analysis of what it is for one thing to be more disposed than another. And it is not clear how any single conditional could do that. The third general problem for the conditional approach is the problem of absent stimulus conditions. The conditional account, again, focuses on statements like (1): (1) The vase is disposed to break when dropped. But what about statements like (6)? (6) The vase is disposed to break. The vase is disposed to break when dropped or crushed or hit by a hammer… but it’s also just disposed to break. (6) does not specify any stimulus conditions—at least not explicitly—in which case there is nothing to put into the antecedent of a conditional. And if there is no conditional, there is no conditional analysis.14 At this point, I have reviewed five problems for the conditional approach. None of these objections are conclusive—there are potential replies to 14 The problem generalizes. Consider the property of being introspective. The introspective person is one who is disposed to introspection, but it doesn’t seem as if there are any specific stimulus conditions associated with this disposition. That is, there is no condition φ such that someone is introspective if and only if they are disposed to introspection when φ. Different people might be disposed to introspect on different occasions for different reasons, and still be equally introspective. The same seems true of many other dispositions: affable people have a kind disposition, irascible people are disposed to anger, and loquacious people are disposed to talk, but it doesn’t seem as if there is a universal trigger for each of these dispositions. The obvious reply is to this objection is to say that the stimulus conditions are somehow fixed implicitly. See fn.26. 6 each of the problems, as well as replies to those replies.15 But the worries are at least somewhat worrisome, and that provides us with at least some motivation to look elsewhere. 2. The Simple Generic Analysis Generic sentences are generally divided into two categories.16 The first category consists of sentences with kind-referring noun phrases (NPs): (7) a. Lions are dangerous. b. A grape is a berry that grows on vines of the genus Vitis. c. The automobile was first mass-produced in 1901. The NPs in (7) do not refer to particular lions, grapes, or automobiles, but to kinds that subsume such particulars. (7a), for example, does not say that some specific lion or lions are dangerous, but that lions—general speaking—are dangerous.17 Since these sentences concern genera, they are one source of genericity. The second category of generics consists of characterizing sentences18: (8) a. Ann trains lions. b. Bill drinks juice with breakfast. c. My car gets good gas mileage. 15 And replies to the replies to the replies! See, for example, the exchange between Manley and Wasserman (2008), Bonevac, Devers, and Sosa (forthcoming), and Manley and Wasserman (forthcoming). 16 I use ‘generic sentence’ to cover both characterizing sentences and sentences which feature kind-referring NPs. Others use the term to cover characterizing sentences only. See, e.g., Krifka et.al. (1995: 3). Pelletier and Asher (1997: 1131) and Schubert and Pelletier (1988: 259-60) draw the same distinction with different terminology (they distinguish between what they call “episodic generics”—which assert a regularity about events—and “nonepisodic generics”—which assert a regularity about the members of some kind). 17 (7a) is also a characterizing sentence (see below) and so it involves genericity twice over. (7a) says that lions (generally) are (generally) dangerous. Henceforth, I will use the term ‘particular characterizing sentences’ for characterizing sentences that lack kindreferring NPs. 18 Characterizing sentences are also called ‘habituals’; as Lyons (1977: 716) points out, this term is potentially misleading. 7 The statements in (8) do not record particular events, processes, or states. Rather, they report general patterns that involve such particulars. (8b), for example, does not concern a specific meal or drink; it says instead that occasions on which Bill eats breakfast are—generally speaking—occasions on which he drinks juice. Since characterizing sentences concern general patterns, they are a second source of genericity. One of the characteristic features of generic sentences is that they tolerate exceptions.19 Take, for example, the following pair of claims: (9) Crows are black. (10) Bill drinks wine with dinner. Clearly, it would be a mistake to equate these claims with their universally quantified counterparts: (11) All crows are black. (12) It is always the case that Bill drinks wine with dinner. Suppose, for example, that Bill normally drinks wine with dinner but that there are some occasions on which he has only water. That would be enough to falsify (12), but not (10). So too, the existence of a single albino crow is enough to falsity (11), but not (9). More generally: generic sentences report on what is typical, and what is typical may not be universal.20 This kind of tolerance inspires the generic approach—the hope is that dispositions can be analyzed in terms of generics and that problematic cases like finks and masks can be treated as tolerable “exceptions” to the general rule.21 There are many different ways that this analysis might go, but we can illustrate the potential benefits by focusing on a simple proposal: 19 This is itself a characterizing sentence, and thus admits of exceptions. For example: ‘Squares have four-sides’ is a generic sentence, but it does not admit of exceptions. 20 For overviews of genericity, see Krifka et.al. (1995), Pelletier and Asher (1997), and Cohen (1999b). 21 See, especially, Fara (2005, 2006, section 2.4) and Lowe (1982, 1989: 170-1, and 2006: 125-8). Many linguists have also postulated a tight connection between genericity and dispositions. See, for example, Krifka, et al. (1995: 7). 8 The Simple Generic Analysis: N is disposed to M when C iff N Ms when C.22 On this account, explicit23 dispositional ascriptions are analyzed in terms of particular24 characterizing sentences. Bill is disposed to drink wine with dinner because he drinks wine with dinner. Salt is disposed to dissolve in water because it dissolves in water. And my vase is disposed to break when dropped because it breaks when dropped. We will consider some worries for this simple idea in a moment. But let us begin with the good news. First, the simple generic analysis avoids the problem of finkish dispositions. According the simple conditional analysis, (13) is analyzed in terms of (14): (13) Ann is disposed to smoke when nervous. (14) Ann would smoke if she were nervous. On the simple generic analysis, (13) is analyzed in terms of (15): (15) Ann smokes when nervous. Now: suppose that a sorcerer has taken a temporary interest in Ann. If Ann were to now get nervous, the sorcerer would cast a spell that would alter Ann’s internal brain states, so that he would choose not to smoke. In that case, (14) is false—it is false that Ann would smoke if nervous. But (15) would still be true—it is still be true that, generally speaking, Ann smokes when nervous. Hence, the simple generic analysis delivers the right results—(13) is true, even when the disposition is finkish. The same solution applies to our original example. The fragile vase breaks when dropped.25 That is what fragile things do. So the vase has the disposition to break when dropped, even when it is being watched over by a guardian sorcerer. 22 23 24 25 This analysis is considered (and rejected) by Michael Fara. See Fara (2006: 69). See fn.6. See fn.17. Or does it? See section 4 for discussion. 9 The simple generic analysis also avoids the problem of masked dispositions. Suppose once again that Ann is disposed to smoke when nervous, and that a protective sorcerer is present. If Ann was to become nervous, the sorcerer would alter her environment by magically removing all cigarettes. In that case, (14) is false—Ann would not smoke if nervous. But (15) is true—Ann typically smokes when nervous, even though she wouldn’t smoke in this unusual case. Once again, the same solution applies to the original case. Perhaps the sorcerer would turn the hard ground into a soft pillow if you dropped my vase at this particular moment. In that case, the vase would not break if dropped. But it might still be true that the vase breaks when dropped. Third, the simple generic analysis avoids the problem of degrees, since characterizing sentences can include appropriate adverbs of quantification. For example, (16) a. Ann is somewhat disposed to smoke when nervous. b. Ann is very disposed to smoke when nervous. can be analyzed in terms of (17) a. Ann sometimes smokes when nervous. b. Ann very often smokes when nervous. Fourth, the simple generic analysis avoids the problem of comparatives since characterizing sentences can feature comparisons. For example, (18) Ann is more disposed to smoke when nervous than Bill. can be analyzed in terms of (19) Ann smokes when nervous more than Bill. Fifth, the analysis avoids the problem of absent stimulus conditions. The problem for the conditional account concerned statements like (20): 10 (20) Ann is disposed to smoke. This sentence does not specify any conditions that would trigger the relevant disposition, in which case there is nothing to put into the antecedent of the relevant counterfactual. But that is no problem for the generic analysis, since it does not make use of conditionals. The proponent of this approach can instead analyze (20) in terms of (21): (21) Ann smokes.26 To sum up: the simple generic analysis seems to avoid all of the problems facing the conditional approach. That is the good news. The bad news is that there are new problems for the new approach. The general problem with the generic analysis is that many characterizing sentences are tied to actual events in a way that disposition ascriptions are not. This general problem manifests itself in several different ways. The first is the problem of accidental patterns: in some cases, an appropriate pattern of events will suffice for the truth of a characterizing sentence, but not for the presence of a disposition. Michael Fara offers the following example: …what does the claim that Tristram is disposed to tremble when he’s angry add to the bare claim that he trembles when he’s angry? …it seems that adding ‘‘disposed’’ serves to rule out accidents of a certain kind. If, by unfortunate coincidence, Tristram happens to be standing near an earthquake each time he’s angry, and so is caused to tremble, it would be true to say that he trembles when he’s angry, but not that he’s disposed to do so. (2005: 69) Fara makes two claims about this case. The first is that Tristram is not disposed to tremble when angry. That judgment is clearly correct. But Fara also claims that Tristram trembles when angry. This is more con26 (21) does not explicitly identify the conditions in which Ann characteristically smokes, but those conditions may be implicitly expressed, relative to a context of utterance. For details, see Spears (1974) and Schubert and Pelletier (1989). 11 troversial.27 In fact, this second claim runs contrary to the linguistic orthodoxy, which says that: …true characterizing sentences cannot capture a mere accidental generalization. For example it might be true that every child ever born in Rainbow Lake, Alberta, was right-handed. Still the generic sentence Children born in Rainbow Lake are right-handed and A child born in Rainbow Lake is right-handed are not true— unless there is something special about Rainbow Lake that causes right-handedness… Characterizing sentences express “principled” generalizations over the entities of a class, and do not capture mere “accidental” facts about them. (Krifka, et al. 1995: 54) 28 The example in this passage involves a kind-referring term (Children born in Rainbow Lake), but the same point applies to particular characterizing sentences like the one about Tristram. It might be true that every situation in which Tristram is angry is one in which he trembles, but it’s not true that Tristram trembles when angry29—not unless that generalization is appropriately non-accidental. One might try to resist the standard view,30 but we can avoid the issue by imagining a less accidental version of the accidental problem. Suppose that Bill is deeply cynical about love and that, as a result, (22) is false: (22) Bill is disposed to cry at weddings. 27 As Cohen (1999a) points out, speaker intuitions about characterizing sentences are often divided. See section 5 for further discussion. 28 The quote gives the summary opinion of “the Generic Group”—a collection of linguists including Manfred Krifa, Greg Carlson, and Francis Jeffrey Pelletier. The same opinion is expressed by Dahl (1975), Lyons (1977: 195), Carlson (1995: 233), and Cohen (1999a: 223-5), among others. (For more on the Generic Group, see the preface to Carlson and Pelletier 1995.) 29 Not, that is, on the generic reading of this sentence. There are also noncharacterizing readings of this claim, but they would be irrelevant to the simple analysis. 30 For example, one might argue that the relevant sentence is true, but suggests something false—viz., that the anger causes the trembling. And one could try to support this hypothesis with the cancellation test: “Tristram trembles when he’s angry, but it’s not the anger that does it—it’s those earthquakes that keep happening!” 12 However, Bill is watched by a sorcerer who disapproves of his cynicism. Whenever Bill attends a wedding, the sorcerer casts a spell that causes Bill to cry.31 In that case, (23) is true: (23) Bill cries at weddings. 32 This behavior may be “accidental” to Bill in the sense that it is does not flow from his own intrinsic nature. But the pattern of weddings and tears is no coincidence—it is explained by the sorcerer’s activity, and his intentions are counterfactual-supporting (If Bill were at a wedding right now, he’d be crying). For these reasons, (22) is a “principled” generalization. And for this reason, it is true.33 We thus have a counterexample to the simple view: Bill cries at wedding, but is not disposed to do so.34 4. Fara’s Generic Analysis The generic approach to dispositions is promising, but the simple analysis is mistaken. The good news is that the problem of accidental patterns 31 Let’s suppose this is a reverse-mask, rather than a reverse-fink, so that Bill never actually gains the disposition to cry at weddings. 32 After all, the sorcerer casts a spell when Bill attends a wedding. That’s characteristic of the sorcerer. And Bill cries when the spell is cast. That’s characteristic of the spell. So Bill cries when he attends weddings. That’s characteristic of him. This argument involves the following form of inference: φ when ψ, θ when φ ∴ θ when ψ. If characterizing sentences operate like conditionals, (Burton-Roberts 1977, Thrane 1980, Kratzer 1986), this is just a hypothetical syllogism. 33 Of course, we would not normally say that weddings make Bill cry, but his crying is caused by the spells and weddings are a cause of the spells being cast. So, by transitivity, weddings are a cause of his crying. 34 Some linguists may resist this conclusion. For example, Schubert and Pelletier (1988) claim that a characterizing sentence requires more than a “principled” or “nonaccidental” pattern—they claim that the predicated property must be “essential” and “inherent” to the object. (1988: 364) The property of being one who cries at weddings is neither essential nor intrinsic to Bill. So perhaps Schubert and Pelletier will say that (23) is false. My own view is that this builds too much into the content of a characterizing sentence. As Cohen (1999b: 26) notes, it is true that Cheetahs run faster than any other animal and that Spices are affordable, but the properties in these cases are neither essential nor intrinsic. See section 5 for further discussion of the accidental pattern problem. 13 admits of a simple fix. The example of Bill and the sorcerer showed that a certain kind of behavior could be characteristic of an individual, even when that behavior is not grounded in the individual’s own character. We can therefore avoid accidents by requiring such a ground. This is the motivating idea behind Michael Fara’s version of the generic analysis. He puts his proposal as follows: Fara’s Generic Analysis: N is disposed to M when C iff N has an intrinsic property in virtue of which it Ms when C.35 Following the lead of David Lewis, we can think of the relevant intrinsic property as the causal basis for the corresponding disposition. Salt, for example, has a particular chemical structure. This is an intrinsic feature of salt, and it is in virtue of having this feature that salt dissolves in water. Hence, Fara’s generic analysis yields the correct result that salt is disposed to dissolve in water. So too, there is something about the intrinsic structure of Ann’s brain in virtue of which she smokes when nervous, and there is something about Bill’s internal constitution that explains his pattern of drinking wine with dinner. Hence, Fara’s generic analysis tells us that Ann is disposed to smoke when nervous and that Bill is disposed to drink wine with dinner. Now consider the case of Tristram and his shaking. Suppose for the moment that the relevant characterizing sentence is true: Tristram trembles when angry. Still, there is no intrinsic property in virtue of which Tristram trembles when angry, so he is not disposed to tremble when angry. The same point applies in the case of the reverse-masking magician. Bill does cry during weddings, but this is not because of his intrinsic properties. Bill’s tears are caused by an external agent, so (22) is false— Bill is not disposed to cry at weddings. In this way, Fara’s account avoids the problem of accidental patterns. But now consider a related worry: the problem of absent patterns. It is widely recognized that disposition ascriptions can be true, even if the relevant disposition is rarely manifested. In fact, a disposition ascription 35 Fara refers to this as “The Habitual Account” (2005: 70). 14 can be true even if the relevant disposition is never manifested. Suppose, for example, that my vase is never dropped and never breaks. That is clearly consistent with (1)—even if the vase does not drop or break, it is still disposed to break when dropped. Things are very different when it comes to characterizing statements. If a disposition is not manifested, there will be no pattern of behavior. And if there is no pattern of behavior, the relevant characterizing sentence will be improper. For example, it seems wrong to say that the vase breaks when dropped when it never drops or breaks. Of course, it would be hasty to reject the proposal on the basis of this one case. As Fara notes, there are other cases in which it seems perfectly appropriate to assert a characterizing sentence in the absence of any characteristic behavior.36 Suppose, for example, that I own an orange crusher that is never actually used. Even if the machine is never turned on, and even if it never crushes any oranges, there is nothing odd about: (24) This machine crushes oranges when it is turned on. It might be true that the machine is disposed to crush oranges when turned on, even if that disposition is never manifested. But now consider a third case.37 Suppose that Ann learns, through genetic screening, that he has the disposition to get violent when drunk. As a result, she completely abstains from drinking and goes on to live a perfectly peaceful life. Now consider (25) and (26): (25) Ann is disposed to get violent when drunk. (26) Ann is disposed to get violent when drunk, but she never gets violent or drunk. (25) will typically suggest that Ann has actually been drunk; after all, evidence for this kind of claim will typically come from repeated observation. However, (26) illustrates that this implication is cancellable, which suggests that it is pragmatic, rather than semantic. In that case, the truth of (25) is consistent with the details of the case. Now compare (25) to (27): 36 37 See Fara (2005: 66-8). See also Carlson (1995: 231). Thanks to David Manley for suggesting this example. 15 (27) Ann gets violent when drunk. In this case, cancellation strikes me as contradictory: (28) Ann gets violent when drunk, but she never gets violent or drunk. The same point can be made even more strongly in the case of finkish dispositions. Suppose once again that Ann is disposed to smoke when nervous, but that she is watched over by a protective sorcerer who temporarily removes this disposition whenever she becomes nervous. Now suppose that Ann is placed in stressful situations everyday throughout her entire life—each time she becomes nervous and each time the sorcerer stops her from smoking. In that case, (29) is true:38 (29) Ann is disposed to smoke when nervous, but she doesn’t smoke when nervous (because the sorcerer stops her). According to Fara, the first conjunct in (29) implies that Ann smokes when nervous. But substituting this statement for the disposition ascription yields a clear contradiction: (30) Ann smokes when nervous, but she doesn’t smoke when nervous (because the sorcerer stops her). Once again, the upshot seems to be that characterizing statements and disposition ascriptions make different demands on the world—at least some characterizing statements require a characteristic pattern of events, whereas dispositional ascriptions do not. That is the problem of absent patterns. Fara actually anticipates this objection. In response, he asks us to imagine a case in which 38 Assume that (29) is asserted at a time when Ann is not nervous (a time at which she has the relevant disposition). 16 …packers have carefully protected [a] delicate glass cup with a support structure, ensuring that situations in which it is struck are ones in which it doesn’t break. But imagine also that a scrupulous vandal has got her hands on the cup. The vandal spends weeks hitting the cup with various large objects, attempting to break it. She could, of course, simply remove the packers’ structure and easily smash the thing; but that would violate her scruples. Instead, she is intent on getting one up on the packers by smashing the cup while it continues to be protected. Her attempts are in vain, however, and in the end she gives up. (2005: 74-5) Fara concedes that, in this case, the vandal can truthfully assert: (31) The cup doesn’t break when struck. Moreover, he accepts that the packers can truthfully assert: (32) The cup is disposed to break when dropped. (“That,” they point out, “is why we had to protect it in the first place!”) However, Fara denies that this is a problem for his account: What it shows, rather, is that disposition ascriptions, like the habitual sentences they embed, are context-dependent to a certain extent. The disposition ascription [32] expresses a falsehood in the context of the vandal’s utterance of [31], while in the context of the museum specialists’ utterance… it expresses a truth. This context-dependence of disposition ascriptions runs in exact parallel to that of the corresponding habitual sentences, and so we have… an explanation of the context-dependence of disposition ascriptions—it is entirely parasitic on that of their embedded habitual sentences. (75) That is Fara’s response to the problem of the uncommon event. In some contexts, (31) expresses a truth and (32) expresses a falsehood; in other contexts, (31) expresses a falsehood and (32) expresses a truth. But there 17 are no contexts in which (31) and (32) both express truths (or both express falsehoods), so there is no context in which Fara’s generic analysis gives the wrong results. There are at least three worries for this response. The main problem is that (29) clearly shows that disposition ascriptions and characterizing statements can come apart, even if we hold the context fixed. The same is true in Fara’s example, since the packers can truthfully assert: (33) The cup is disposed to break when dropped, but it doesn’t break when dropped (because of our protective packing). Since both conjuncts express truths relative to a single context, we have a clear counterexample to Fara’s account. The second worry for Fara concerns the problem of absent stimulus conditions. As noted earlier, it is possible to attribute a disposition without specifying a stimulus: (20) Ann is disposed to smoke. This was puzzling on the conditional analysis, since no antecedent is provided (at least not explicitly). The generic approach avoided this issue by analyzing (20) in terms of (21): (21) Ann smokes. Fara explicitly endorses a variant of this view: N is disposed to M if and only if N has some intrinsic property in virtue of which N Ms. (2005: 70) This would solve the absent stimulus problem. But the solution may be more problematic than the problem. Suppose that a sorcerer creates Ann, as an adult, ex nihilo. This is an evil sorcerer, so he creates Ann with an uncontrollable nicotine addiction and then places her on a desert island with nothing to smoke. As a result, she lives out her life, alone on the island, without ever smoking. Clearly, this is case in which Ann’s intrinsic states dispose her to smoke—that’s how her brain is wired. But those same states do not make it the case that she smokes. Obviously, (21) is only true if Ann actually 18 smokes.39 She does not do that, so (21) is false. Moreover, an appeal to context-dependence will be of no help, since there is no reasonable context in which (21) expresses a truth. A final worry for Fara’s has to do with comparatives. We noted that, on Fara’s view, N is disposed to M just in case N has some intrinsic property in virtue of which N Ms. One might try to extend this idea to cover comparatives as follows: N is more disposed to M than N* just in case N has some intrinsic property in virtue of which N Ms more than N*. (34), for example, would be analyzed in terms of (35): (34) Ann is more disposed to smoke than Bill. (35) Ann has some intrinsic property in virtue of which she smokes more than Bill. But this account is unsatisfactory. Once again, suppose that (34) is true, but that Ann is stranded on a deserted island. Ann longs for a cigarette every day, but never actually smokes. Bill, in contrast, has easy access to cigarettes and smokes a few times a week. Thus: Ann is more disposed to smoke, but Bill ends up smoking more. Since Bill smokes more than Ann, it is false that she smokes more than Bill. Hence, (35) is false. As before, an absent pattern provides a counterexample to the proposed analysis. The important point about this case is that an appeal to contextdependence will be of no help, since comparative claims are largely context-invariant. Context may play a role in determining whether or not someone counts as ‘disposed to smoke’ or whether it’s true to say that a given individual ‘smokes’. But context is largely irrelevant to the question of whether one individual smokes more than another or whether one is more disposed to do so. Thus, an appeal to context-dependence cannot bridge the divide between (34) and (35). 4. Quine’s Generic Analysis 39 See Krifka, et al. (1995: 54). 19 I have argued against the simple analysis, as well as the simple amendment suggested by Fara. Perhaps it is time to go back to where we began. Recall Quine’s original claim about solubility: “what qualifies a thing as soluble… is that it is of the same kind as the things that… dissolve.” (1969: 16) So, for example, a particular piece of salt is disposed to dissolve in water because it is salt and salt dissolves in water. More generally, an object has the dispositions that it does in virtue of the behavior that is characteristic of its kind: Quine’s Generic Analysis: N is disposed to M when C iff there is some kind K such that (i) N is a K and (ii) Ks M when C. 40 The distinctive feature of this proposal is that it is doubly generic. To say that Ks M when C is to say, roughly, that typical Ks typically M when C. For example, to say that cats purr when happy is to say that typical cats typically purr when happy. There may be some unusual cats that don’t purr at all (even when happy). And there may be some times when even normal cats take a break from purring (despite being happy). But it is still the case that, generally speaking, cats generally purr when happy. This feature of Quine’s account helps to avoid all of the problem cases discussed so far. First, the analysis avoids the problem of finkish dispositions. Take the case of the crystal vase that is watched over by a guardian sorcerer (so that it will not break if dropped). The vase is still crystal and it is still true that crystal breaks when dropped. Hence, the particular crystal vase is still disposed to break when dropped. Second, the analysis avoids the problem of masked dispositions, for exactly the same reasons. Suppose that the sorcerer will turn the hard floor to a soft pillow if the vase is dropped. That does not change what kind of object the vase is; nor does it alter the behavior that is character- 40 For the record, it is not clear whether Quine would endorse the analysis as formulated here. Quine’s comments on dispositions and kinds are brief, and he does not provide a general statement of his thesis. In addition, he often uses universal claims, rather than characterizing statements, when discussing specific examples. But the Quinean proposal is at least inspired by Quine and, in that way, earns its name. 20 istic of that kind. It is still true that crystal breaks when dropped, so it is still true that this particular crystal vase is disposed to break when dropped. Third, the analysis avoids the problem of absent patterns. Suppose that the fragile vase is routinely dropped and, on each occasion, is protected by a sorcerer. In that case, the vase does not break when dropped, because of the magical protection. But it is still disposed to break when dropped, since it is still true that crystal breaks when dropped—the protected vase has the disposition because of what happens to its unprotected kind-mates. The same point applies to the masking variant involving DesertIsland Ann. Ann does not smoke, since she has no cigarettes. But Ann does have the addiction and it is generally the case that smoking addicts smoke. Hence, Ann is disposed to smoke, even though she never has the opportunity. Finally, Quine’s analysis avoids the problem of accidental patterns. Suppose that an earthquake randomly occurs every time that Tristram is actually angry. And suppose (with Fara) that makes it true that Tristram trembles when angry. Does this mean that Tristram is disposed to tremble when angry? That depends on how we think about kinds, and this is a difficult issue (see section 5). But it is certainly not true that humans tremble when angry. Nor is it true that humans-that-areintrinsically-like-Tristram tremble when angry. So the Quinean might conclude that Tristram lacks the disposition. The same point also applies to the magical variation of the accidental pattern problem. Suppose that the sorcerer makes Bill cry at weddings. Still, it is not generally the case that humans—or at least humans-thatare-intrinsically-like-Bill—cry at weddings. So, perhaps the Quinean can conclude that Bill is not disposed to do so. To sum up, Quine’s account seems to avoid all of the problem cases. But what if we alter some of those cases? Imagine a sorcerer that keeps careful watch over all cats. Whenever a cat scratches its ear, the sorcerer makes it sneeze. In that case, one might think that (36) is true: (36) Cats sneeze when they scratch their ears.41 41 Of course, it is uncontroversial that there are some contexts in which (36) expresses a truth, since there are some contexts in which the NP is interpreted non-generically. In 21 The argument for (36) would be just like the earlier argument for (23): Cats sneeze when the spell is cast (that’s what the spell does) and the spell is cast when cats scratch (that’s what the sorcerer does), so cats sneeze when they scratch (that’s what they do). We therefore have an apparent counterexample to the Quinean proposal—cats sneeze when they scratch, but they are not disposed to do so. The problem of the absent pattern can also be reintroduced in much the same way. Imagine a world populated by duplicates of Desert-Island Ann. Every person alive is addicted to nicotine. Everyone is disposed to smoke. But no one has the opportunity. Since nobody actually smokes, (37) is clearly false (at that world): (37) People smoke. Yet each individual person is disposed to smoke. This gives us a second counterexample to the proposed account. 5. Lowe’s Generic Analysis The general problem for Quine was the same problem from earlier— characterizing statements are more closely tied to actual events than are disposition ascriptions (even if those characterizing sentences include kind-referring NPs). There are various ways that one might try to resist this criticism.42 particular, one can hear (36) as saying that some (particular) cats sneeze when they scratch their ears. And that is clearly true. But it seems to me that (36) is also true when the NP is read as a generic (cats), rather than particular (some cats). 42 One possibility is to try and mix elements from Fara and Quine. For example, one might try to say that N is disposed to M when C if and only if there is some kind K such that N is a K and Ks M when C in virtue of their intrinsic properties. That would avoid the problem of the accidental pattern, including the variant of that problem that involves the sneeze-inducing sorcerer. But it would not help solve the absent pattern problem, since objects that don’t do something don’t do it in virtue of their intrinsic properties. For example, it’s not true that people smoke in the desert-Ann example. So it’s not true that they smoke in virtue of their intrinsic properties. A better idea is to mix elements from Fara, Quine, and Lowe. See below. 22 One possibility is to appeal to natural laws. This variation on the Quinean proposal is defended by E.J. Lowe. He writes: When we say that a particular piece of rubber ‘is elastic’… what we are saying, according to my account, is that this object belongs to a kind, K, such that the law obtains that K is elastic. In this case of course the kind in question is rubber. Similarly when we say that a particular object is ‘water-soluble’, what we are saying is that it belong to a kind, K, such that the law obtains that water dissolves K... on this view, individual objects possess their various natural ‘powers’ in virtue of belonging to substantial kinds which are subject to appropriate laws. (2006: 107) Generalizing on these examples, we have Lowe’s generic analysis: Lowe’s Generic Analysis: N is disposed to M when C iff there is some kind K such that (i) N is a K and (ii) it is a law that Ks M when C. The advantage of introducing laws is two-fold. First, it is possible to have a universal pattern without a law (e.g., All gold spheres are less than a mile in diameter). Second, it is possible to have a law without a corresponding pattern (e.g., Newton’s first law of motion). For these reasons, one might hope that Lowe’s analysis can avoid the problems posed by accidental and absent patterns. Begin with the case of the sneezing cats. In one sense, the pattern of scratching and sneezing is non-accidental—it is explained by the intentional activity of the quirky magician. But the magical activity is itself an accidental feature of our world, in the sense that it is not necessitated by nature—it is not a law of nature that cats sneeze when they scratch their ears. Hence: Cats are not disposed to sneeze when they scratch their ear. Now consider the universal version of the absent pattern problem: all crystal vases everywhere are protected by guardian sorcerers, so that they don’t break when dropped. This is no problem for Lowe, since (one could argue) it is still a law that crystal vases break when dropped, even 23 if they don’t end up breaking when dropped. This would be a case of a ceteris paribus law in which ceteris is never actually paribus. Lowe’s analysis may be an advance over Quine’s, but problems remain for both. One worry is that, on kind-based accounts, every member of a given kind will have the same dispositions. That might be the right result when it comes to chemical kinds like salt or water, but it seems incorrect when it comes to biological kinds like cat or lion. Suppose, for example, that Fluffy is an unusual cat—she is allergic to ear scratching and is therefore disposed to sneeze when she scratches her ear. Now consider the following pair of statements: (38) Fluffy is disposed to sneeze when she scratches her ear. (39) Fluffy is a cat and it is a law that cats sneeze when they scratch their ears. Fluffy is a cat, but it is not a law that those kinds of creatures sneeze when they scratch. So, if cat is the relevant kind, Lowe’s account would say that (38) is false. That verdict would be incorrect, in which case we would have a counterexample to the account. The same problem also arises in the opposite direction. Suppose, for example, that Pedro is a lion that is born and raised in captivity and thus lacks some of the dispositions shared by his wild brethren. In particular, suppose that Pedro is not disposed to hunt when hungry. Still, he is a lion. And it is a (ceteris paribus) law that lions hunt when hungry. Hence, (40) is true, but (41) is false. (40) Pedro is disposed to hunt when hungry. (41) Pedro is a lion and it is a law that lions hunt when hungry.43 Matters seem even worse when we turn from biological kinds to functional kinds. According to Quine and Lowe’s accounts, objects have the dispositions that they do in virtue of the kinds to which they belong. But in the functional case, the reverse seems to be true—whether or not an 43 It might be pointed out that Pedro is a domesticated lion and that domesticated lions do not hunt when hungry. That is true, but irrelevant. Pedro is a domesticated lion, but he’s also a lion. Hence, there is some kind K such that Pedro is a K and it is a law that Ks hunt when hungry. 24 object falls under a given kind is typically determined by its capacities and dispositions. To take an earlier example: What qualifies a particular machine as an orange crusher is its ability to serve a certain function— namely, crush oranges. That’s just to say that something is an orange crusher in virtue of the fact that it is disposed to crush oranges in the right circumstances. So too, coffee grinders are coffee grinders in virtue of their ability to grind coffee. And the same can be said about calculators, coffee grinders, and a host of other objects. The problem, of course, is that if facts about kind-membership are grounded in facts about dispositions or abilities, then the later facts cannot also be grounded in the former. The preceding problems threaten the current proposal. But they also point toward a final solution. Begin with the question: Do the laws dictate that creatures like Fluffy sneeze when they scratch their ears? That depends. If “creatures like Fluffy” refers to cats, then the answer is no; but if it means things that are intrinsically just like Fluffy, then the answer is yes. The laws dictate that that kind of creature does sneeze when it scratches. Second: Do the laws dictate that creatures like Pedro hunt when hungry? Again, it depends. If “creatures like Pedro” refers to lions, then the answer is yes; but if it means things that are intrinsically like Pedro, then the answer is no. The laws dictate that that kind of creature doesn’t hunt when hungry, so (40) is false. These observations suggest that we should eliminate talk of kinds and focus instead on intrinsicness. Doing so will allow us to avoid the problems posed by Pedro and Fluffy. It will also avoid any threat of circularity, since it would avoid any appeal to functional kinds.44 More generally, these observations indicate that we should aim for a final analysis of dispositions that incorporates ideas from Quine (genericity), Fara (intrsincality) and Lowe (lawhood). It is to this project that we now turn. 44 It would also provide a nice explanation of why the Quine-Lowe approach works in the cases where it works and fails in the cases where it fails. Kind-based accounts work well when it comes to the kinds of physics and chemistry; they fair poorly when it comes to biological kinds and functional kinds. That is because, in the domains of chemistry and physics, co-membership in a kind guarantees intrinsic similarity. This is not the case when it comes to less basic kinds, and that is why the Quine-Lowe account fails. 25 6. The Final Generic Analysis Linguists take generics sentences to have a tripartite structure, consisting of a binary operator (GEN), a restrictor (φ), and a matrix (ψ): GEN (φ, ψ)45 GEN is typically interpreted as an adverb of quantification, along the lines of Lewis (1975)—that is, as an unselective operator binding all free variables: GEN x1…xn (φ x1…xn, ψ x1…xn) To take an earlier example: (7a) Lions are dangerous. GEN x (x is a lion, x is dangerous) A similar treatment is suggested for particular characterizing sentences, where the quantifiers are taken to range over situations and the whenclauses are treated as quantifier restrictors. To take another example from earlier: (15) Ann smokes when nervous. GEN x (Mary is nervous in x, Mary smokes in x) (Generally, situations in which Mary is nervous are situations in which she smokes.) Almost all linguists agree about the logical form of generic statements,46 but there is considerable disagreement about the semantics of 45 See, for example, Heim (1982), Kratzer (1981), and Krifka, et al. (1995). Leslie (2008: 6) suggests that all theorists agree on this point, but that is too strong. For a dissenting opinion, see Liebesman (forthcoming). 46 26 the generic operator. We will limit our attention to one of the most popular approaches, on which GEN receive a modal interpretation.47 The standard framework for the modal analysis is inherited from Stalnaker (1968), Lewis (1973), and Kratzer (1981). Kratzer’s formulation involves three different components. The first component is the modal relation, which determines the strength of a modal claim (e.g., necessity, possibility, probability). If we think of modal operators as quantifiers over possible situations,48 then strength is determined by the kind of quantifier being used (e.g., ‘every’, ‘some’, ‘most’). The second component of Kratzer’s framework is the modal base, which determines the flavor of a modal claim (e.g., epistemic, deontic, teleological).49 Formally, the modal base is a function from one situation (a situation considered as actual) to a set of other possible situations (the situations in the domain of quantification). Informally, we can think of a modal base as a set of propositions in view of which the modal claim is made. For example, if the claim is epistemic, the modal base might consist of our evidence or the body of known truths. The final component of Kratzer’s framework is the ordering source which provides a partial ordering of the possibilities consistent with the base. Lewis’s (1973, 1979) similarity metric for counterfactuals would be one example of such an ordering, though other metrics may be appropriate for other kinds of modal claims. With this framework in place, we can define various modal notions like ‘must’, ‘probable’, and ‘good chance’: Definition 1: It must be that φ is true relative to background situation s, modal base ƒ, and ordering source ≤ iff: Every x (x ∈ ƒ(s) , φ is true in x)50 ≤ 47 See Dahl (1975), Nunberg and Pan (1975), Heim (1982), Krifka, et al. (1995: 49-58), and Schubert and Pelletier (1988, 1989). 48 It is often noted that, in analyzing generics, it is most natural to quantify over possible situations, rather than (entire) possible worlds. I will follow this pattern here, though I will omit some of the formalism. See Barwise and Perry (1983) for a formal presentation of situation semantics. For an application to generics, see Krifka, et al. (1995: 57-8). 49 This summary ignores the different roles that premise sets and conversational background play in determining the modal base. See Kratzer (1977, 1981) for a more careful statement of the view. See also Kratzer (forthcoming). 50 I treat ‘every’ and ‘some’ as binary quantifiers to match ‘most’ and ‘many’ (see below). 27 (where x∈ƒ(s) iff x is among the closest situations to s in ƒ(s), under ≤) ≤ Definition 2: It is probable that φ is true relative to <s, ƒ, ≤ > iff: Most x (x ∈ ƒ(s) , φ is true in x) ≤ (where Most x (φ x, ψ x) is true iff |{x:φ x ∧ ψ x}|÷ {x:φ x}|> ½) Definition 3: There is a good chance that φ is true relative to <s, ƒ, ≤ > iff: Many x (x ∈ ƒ(s) , φ is true in x) ≤ (where many functions much like most, but the minimal value is allowed to vary according to context)51 We can also analyze conditional statements, on the assumption that the consequent of a conditional includes an implicit modal. Intuitively, what φ → ψ says is that, if φ is true, it must be that ψ. So, given definition 1, we can characterize the conditional as follows: Definition 4: φ → ψ is true relative to <s, ƒ, ≤ > iff: Every x (x ∈ƒ(s) ∧ φ is true in x, ψ is true in x) ≤ On this analysis, the antecedent restricts the modal base to those possibilities at which φ is true and must ψ is then evaluated with respect to that base.52 Next: it often noted in the linguistics literature that generic statements are roughly equivalent to conditionals.53 For example: to say that lions are dangerous is to say—roughly—that if something’s a lion… it’s dangerous. The crucial question is: Which modal operator is to replace the ellipsis? Dahl (1975) and Krifka et al. (1995: 52) assume must. Schubert and Pelletier (1988, 1989) prefer probable. I suggest a third option— to say that lions are dangerous is to say that if something’s a lion, there’s a good chance it’s dangerous. More generally: 51 For a recent discussion of ‘many’, see Lappin (2000). If we interpret ‘→’ as the counterfactual connective and ‘≤’ as Lewis’s (1973) similarity metric, then this definition is equivalent to the standard treatment of counterfactuals. 53 See Burton-Roberts (1977), Thrane (1980), and Krifka, et al. (1995: 52). 52 28 Definition 5: GEN x (φ x, ψ x) is true relative to <s, ƒ, ≤> iff: Many x (x ∈ ƒ(s) ∧ φ x, ψ x) ≤ For example: (38) Ann smokes when nervous. GEN x (Ann is nervous in x, Ann smokes in x) is true relative to <s, ƒ, ≤ > iff: Many x (x∈ ƒ(s) ∧ Ann is nervous in x, Ann smokes in x) ≤ On this analysis, the when-clause of an episodic generic restricts the modal base, just like the if-clause of a conditional. The analysis tells us to restrict our attention to the closest possibilities in which Ann is nervous, and check to see whether Ann smokes in those situations. If she smokes in sufficiently many, then that behavior is characteristic, in which case the characterizing statement is true. What proportion of nervousness-situations must be smokingsituations in order for it to be the case that Ann smokes when nervous? How many is many? The answer, of course, is it depends. Recall the standard treatment of gradable predicates like ‘tall’: the comparative form—‘taller than’—provides an ordering, which gives us a scale, which allows us to say that the positive predicate applies, relative to a context, just in case the relevant individual meets the contextually-determined threshold on that scale. The same idea can be adapted to gradable quantifiers like ‘many’ and ‘few’. In the current case, the comparative— ‘smokes more than’—provides an ordering, which gives us a scale. Context fixes a threshold on that scale, and ‘smokes’ applies to Ann just in case he she meets that threshold. This brings us back to the case of dispositions. We have seen throughout that there is a close connection between generic statements and disposition ascriptions. One might therefore be tempted to posit a covert dispositional operator—DISP—which functions just like GEN.54 In that case, one could represent a disposition ascription as follows: 54 See Fara (1995: 61-3). 29 (13) Ann is disposed to smoke when nervous. DISP x (Ann is nervous in x, Ann smokes in x) This may not be a plausible representation of (13)’s logical form, but it does suggest a promising account of its truth-conditions. The idea is to treat (13) like (15), so that disposition ascriptions are analyzed in terms of (restricted) quantification over possible situations. Thus: Definition 6: DISP x (N is C in x, N Ms in x) is true relative to <s, ƒ, ≤ > iff: Many x (x ∈ ƒ(s) ∧ N is C in x, N Ms in x) ≤ In order to turn this definition into a proper analysis of dispositions, we must say more about modal background against which disposition ascriptions are typically made—in other words, we must identify the relevant modal base and ordering source. Our earlier discussion highlighted the importance of two elements in our thinking about dispositions: (i) the laws of nature and (ii) the intrinsic properties of the subject. So, at a minimum, our modal base and ordering source should preserve these two things. We can achieve this by taking the relevant base (ƒ*) to be a (minimal) set of propositions describing (i) and (ii), and by taking the relevant ordering source to be empty (≤ ).55 We can then put the final proposal as follows: ∅ The Final Analysis: N is disposed to M when C (relative to a background situation s) iff N Ms in many C-situations. (where x is a C-situation iff x ∈ƒ*(s) and N is C in x) Restricting our attention to the actual world, this proposal says that N is disposed to M when C just in case N Ms in most of the possible situations where (i) the actual laws obtain, (ii) N has its actual intrinsic properties, and (iii) N is subject to the stimulus condition C. Take, for example, our original example: ≤∅ (1) The vase is disposed to break when dropped. 55 So that, for any x and w, x∈ƒ(w) ≡ x∈ƒ(w). ≤∅ 30 On the final approach, we are to begin with the set of all nomically possible situations. We then restrict our attention to those cases in which the vase exists with its actual intrinsic structure. We then restrict our attention still further to those situations in which the vase is dropped. And now, focusing on just those cases, we ask: does it break? Of course, the vase does not break in all of these cases, since some of those cases involve finks or masks or other unusual events. But most of the time, it breaks. Since most is normally enough for many, (1) is normally true. The final analysis has much to recommend it. As just noted, finks and masks pose no problems. The explanation for this is that counterfactual claims and disposition ascriptions employ different ordering sources.56 The crystal vase that is protected by a sorcerer does not break in any counterfactually-close possibility. That is why (2) is false. (2) The vase would break if dropped. But the ordering source for disposition ascriptions—≤ —leaves many more dropping-situations in play. In most of those the cases the fink is absent, so there are many possibilities in which it breaks. That is why (1) is true, even in the finkish case. The same point applies to masks. Absences and accidents are equally unproblematic. For example, Desert-Island Ann never actually smokes, since she never has the opportunity. But in many possibilities she does, and in most of those she smokes when nervous. The opposite point applies to Cynical Bill. Every time that Bill goes to a wedding, he cries. The same is true at counterfactually-close weddings. Both of these patterns are explained by the presence of the eccentric sorcerer at the actual world. However, that sorcerer’s presence (and intentions and abilities) are accidental to our world. In the vast majority of nomologically possible situations, Bill is left alone at weddings. And in those cases, he does not cry. Similar remarks apply to the cases of Tristram (the trembling boy), Fluffy (the sneezing cat), and Pedro (the domesticated lion). None of these cases pose problems for the final view. In addition, the gradable quantifier in the final analysis provides a straightforward treatment of gradable dispositions and thus avoids the ∅ 56 As well as different covert quantifiers. 31 problem of comparatives and the problem of degree-modifiers (from section 1). In fact, the most serious problem with the final “generic” analysis might be its name.57 The current proposal was inspired by the modal treatment of generics, but there is no genericity in the finished product. In fact, the modal treatment helps to explain—and thus underscore—the main difference between generics and disposition ascriptions. We have noted that, in many cases, characterizing statements are closely tied to actual events. For example, Desert-Island Ann is disposed to smoke, but it is not true that she smokes. And Bill cries at weddings, but is not disposed to do so. On the modal approach, we can explain these discrepancies by positing different modal bases for the two kinds of claims. In the case of disposition ascriptions, we generally hold fixed the laws and the intrinsic properties of the subject.58 That is all. The same facts are held fixed when making characterizing statements. But that is not all. What other propositions are built into the relevant modal base? What other facts are held fixed when evaluating characterizing sentences? I have no general answer to these questions, but the facts are fairly clear when it comes to Ann and Bill. In Ann’s case, one holds fixed the fact that she has no access to cigarettes. In that case, none of the possibilities determined by the base are ones in which she smokes. So, she does not smoke in many of those cases. So, it is false that Ann smokes (relative to the typical modal base for generics). In Bill’s case, the salient facts involve the intentions and abilities of the sorcerer. The relevant modal base must preserve these facts, in which case all (or at least many) of the closest possibilities in which Bill attends weddings are ones in which he cries. So, it is true that Bill cries at weddings (relative to the typical modal base for generics). To sum up: I have suggested that generic statements and disposition ascriptions can both be analyzed in terms of proportional quantification 57 A second problem is that the current proposal is not terribly original. Ironically, the final “generic” analysis is almost equivalent to the version of the conditional analysis defended in Manley and Wasserman (2007, 2008). For more on the relationship between these two formulations, see Vetter (forthcoming) and Manley and Wasserman (forthcoming). 58 At least in typical cases. There may be some contexts in which one holds certain extrinsic facts fixed. For possible examples, see Fara’s discussion of “entrenched” finkishness. (1995: 75-8) 32 over possible situations, but the two kinds of claims are made relative to different bases. If correct, this would explain both the obvious similarities, and the subtle differences, between dispositions and generics.59 References Barwise, and John Perry. 1983. Situations and Attitudes. Cambridge: MIT Press. Bonevac, Daniel, Josh Dever, and David Sosa. forthcoming. The Counterexample Fallacy. Mind. Bird, Alexander. 1998. Dispositions and Antidotes. The Philosophical Quarterly 48: 227–34. Burton-Roberts, Noel. 1977. Generic Sentence san Analyticity. Studies in Language 1: 155-96. Carlson 1995. Truth Conditionals of Generic Sentences: Two Contrasting Views. In Carlson and Pelletier. Carlson, Gregory N. and Fracis Jeffry Pelletier. 1995. The Generic Book. Chicago: The University of Chicago Press. Cohen, Ariel. 1999a. Generics, Frequency Adverbs, and Probability. Linguistics and Philosophy 22: 221-53. Cohen, Ariel. 1999b. Think Generic! The Meaning and Use of Generic Sentences. Stanford: CSLI Publications. Dahl, Östen. 1975. On Generics. In E. Keenan, ed., Formal Semantics of Natural Language. Cambridge: Cambridge University Press. Fara, Michael. 2005. Dispositions and Habituals. Noûs 39: 43–82. Fara, Michael. 2006. Dispositions. In E.N. Zalta, ed., Stanford Encyclopedia of Philosophy. Goodman, Nelson. 1954. Fact, Fiction, and Forecast. Cambridge: Harvard University Press. 59 An earlier version of this paper was presented at the 2010 Pacific Division Meeting of the American Philosophical Association. I thank my commentator on that occasion, Jason Turner, who helped me draw a connection between the generic and conditional accounts. I also thank Daniel Howard-Snyder, Frances Howard Snyder, Hud Hudson, David Manley, Ned Markosian, and Dennis Whitcomb, all of who provided comments on earlier drafts of this paper. 33 Heim, Irene. 1982. The Semantics of Definite and Indefinite Noun Phrases. Ph.D. dissertation, University of Massachusetts, Amherst. Johnston, Mark. 1992. How to Speak of the Colors. Philosophical Studies 68: 221–63. Klein, E. 1980. A Semantics for Positive and Comparative Adjectives. Linguistics and Philosophy 4: 1–45. Kratzer, Angelica. 1981. The Notional Category of Modality. In H.J. Eikmeyer and H. Rieser, eds., Words, Worlds, and Contexts: New Approaches to Word Semantics. Berlin: de Gruyter. Kratzer, Angelika. 1986. Conditionals. In A.M. Farley, P. Farley, and K.E. McCulloguch, eds., Papers from the Parasession on Pragmatics and Grammatical Theory. Chicago: Chicago Linguistic Society. Kratzer, Angelika. forthcoming. Modals and Conditionals. Oxford: Oxford University Press. Krifka, Manfred, Francis Jeffry Pelletier, Gregory N. Carlson, Alice ter Meulen, Gennaro Chierchia, and Godehard Link. Genericity: An Introduction. In Carlson and Pelletier 1995. Lappin, Shalom. 2000. An Intensional Parametric Semantics for Vague Quantifiers. Linguistics and Philosophy 23: 599-620. Lawler, John. 1973. Studies in English Generics. University of Michigan Papers in Linguistics. Leslie, Sarah-Jane. 2008. Generics: Cognition and Acquisition. Philosophical Review 117: 1-47. Lewis, David. 1973. Counterfactuals. Oxford: Blackwell. Lewis, David. 1975. Adverbs of Quantification. In E. Keenan, ed., Formal Semantics of Natrual Languages. Cambridge: Cambridge University Press. Lewis, David. 1979. Counterfactual Dependence and Time’s Arrow. Nous 13: 455-76. Lewis, David. 1999. Finkish Dispositions. Reprinted in D. Lewis, Papers in Metaphysics and Epistemology. Cambridge: Cambridge University Press. Liebesman, David. forthcoming. Simple Generics. Noûs. Lowe, E.J. 1982. Laws, Dispositions and Sortal Logic. American Philosophical Quarterly 19: 41-50. Lowe, E.J. 1989. Kinds of Being: A Study of Identity, Individuation and the Logic of Sortal Terms. Oxford: Blackwell. 34 Lowe, E.J. 2006. The Four-Category Ontology: A Metaphysical Foundation for Natural Science. Oxford: Oxford University Press. Lyons, John. 1977. Semantics. Cambridge: Cambridge University Press. Manley, David and Ryan Wasserman. 2007. A Gradable Approach to Dispositions. The Philosophical Quarterly 57: 68–75. Manley, David and Ryan Wasserman. 2008. On Linking Dispositions and Conditionals. Mind 117: 59-84. Manley, David and Ryan Wasserman. forthcoming. Dispositions, Conditionals, and Counterexamples. Mind. Martin, C. B. 1994. Dispositions and Conditionals. The Philosophical Quarterly 44: 1–8. Nunberg, G. and N. Pan. 1975. Inferring Quantification in Generic Sentences. In Papers from the 11th Regional Meeting of the Chicago Linguistic Society. Pelletier, Frances Jeffry and Nicholas Asher. 1997. Generics and Defaults. In J. van Benthem and A. ter Meulen, eds., Handbook of Logic and Language. Amsterdam: Elsevier. Quine, W.V.O. 1960. Word and Object. Cambridge, MA: MIT Press. Quine, W.V.O. 1969. Natural Kinds. In N. Rescher, ed., Essays in Honor of Carl G. Hempel: A Tribute on the Occasion of His Sixty-Fifth Birthday. Dordrecht: D. Reidel Publishing Co. Ryle, Gilbert. 1949. The Concept of Mind. London: Hutchinson. Schubert, Lenhart K. and Francis J. Pelletier. 1987. Problems in the Representation of the Logical Form of Generics, Plurals, and Mass Nouns. In E. Lepore, ed., New Directions in Semantics. London: Academic Press. Schubert, Lenhart K. and Francis J. Pelletier. 1988. An Outlook on Generic Statements. In M. Krifka, ed., Genericity in Natural Language: Proceedings of the 1988 Tubingen Conference. Schwarzschild, Roger and Karina Wilkinson. 2002. Quantifiers in Comparatives: A Semantics of Degree based on Intervals. Natural Language Semantics 10: 1–41. Smith, A. D. 1977. Dispositional Properties. Mind 86: 439–45. Spears, A. K. 1974. On the Notion Occasion and the Analysis of Aspect. In Papers from the 10th Regional Meeting, Chicago Linguistic Society. 35 Stalnaker, Robert. 1968. A Theory of Conditionals. In N. Rescher, ed., Studies in Logical Theory. Oxford: Blackwell. Thrane, T. 1980. Referential-Semantic Analysis: Aspects of a Theory of Linguistics. Cambridge: Cambridge University Press. Vetter, Barbara. forthcoming. On Linking Dispositions with Which Conditionals? Mind. von Stechow, Arnim. 1984. Comparing Semantic Theories of Comparison. Journal of Semantics 3: 1–77. 36