Portal de Periódicos da CAPES

One Ought Too Many

2012; Wiley; Volume: 89; Issue: 1 Linguagem: Inglês

10.1111/j.1933-1592.2012.00646.x

1933-1592

Stephen Finlay, Justin Snedegar,

Syntax, Semantics, Linguistic Variation

Normative branches of philosophy take special interest in claims expressed by sentences of the form 'S ought to φ', where S is some agent and φ-ing is some action (broadly conceived). We'll call sentences satisfying this description agential. These claims raise many important philosophical issues, such as: Which ones are true? What makes them true? How can we know if they are true? What follows from them? Some philosophers have hoped that light would be cast on these issues if we could get clearer on the exact meaning of this word 'ought'. Once we start looking to language for philosophical clues we should notice that the word 'ought' is also employed in different kinds of sentences. It appears in normative sentences with different grammatical forms, like 'It ought to be that Charlie was born a girl', and 'There ought to be less suffering in the world'. For convenience we'll call sentences like these nonagential.1 (It also has a non-normative epistemic use meaning approximately what is probable, as in 'The sky ought to get dark soon'.) In these other sentences, 'ought' appears to represent a propositional operator, O(p). Are these other kinds of sentences clues to the meaning of 'ought' in the cases of special interest to normative philosophy? This will be the case if 'ought' has just one linguistic meaning, a unified semantics and logical form (syntax) as many linguists hold. Many philosophers have also embraced this view, which we'll call the Uniformity Thesis.2 On this view agential sentences like 'Vince ought to stop driving drunk' are semantically and logically equivalent to nonagential sentences like 'It ought to be that Vince stops driving drunk', with 'ought' representing a propositional operator in both. This view is rejected by other philosophers who maintain that at least sometimes the 'ought' in agential sentences instead represents a normative relation between the agent S and the action of φ-ing, with the logical form O(S, φ) and meaning roughly is obligated to or has most reason to.3 According to this Ambiguity Thesis 'ought' is semantically and logically ambiguous, so it is a mistake to try to draw philosophical lessons about the agential 'ought' claims of interest to normative philosophy from the other uses of the word. In this paper we defend the Uniformity Thesis against the central objection that it cannot account for all the linguistic data from ordinary usage of 'ought'. Versions of this objection have been raised by a number of philosophers including Gilbert Harman, Peter Geach, Anthony Price, and most recently and rigorously, Mark Schroeder. After clarifying which ambiguity is at issue and explaining the Overgeneration problem it faces, we introduce a contrastivist semantics for 'ought' and show how it provides resources for accommodating and explaining the problematic data without attributing any semantic or logical ambiguity to the word 'ought'. Officially we remain neutral about whether contrastivism and the Uniformity Thesis itself are correct. But since contrastivism is independently well-motivated and the Uniformity Thesis has the virtue of parsimony, we believe this view should enjoy the presumption of truth unless and until better objections are found. The nonagential sentences (1) and (2) say that certain states of affairs ought to be the case (the "ought-to-be"). Here 'ought' is commonly glossed as meaning it is best that… , so we can call these readings evaluative.4,5 From some state of affairs being best nothing directly follows about how any agent has most reason to act, and so these sentences seem to have at most an indirect bearing on agents' deliberations. Sentence (2) entails neither that somebody has most reason to rig the lottery so Larry wins, for example, nor even that Larry has most reason to buy a lottery ticket (the odds of his winning would after all be extremely long). By contrast, the agential sentences (3) and (4) on a natural reading do seem to entail claims about the agents' reasons to act, and so to bear directly on their deliberations (the "ought-to-do"). They do not, however, entail that any particular state of affairs would be best. We can call these readings deliberative. Mark Schroeder identifies five hallmarks that distinguish deliberative readings from evaluative readings. Unlike the evaluative 'ought', the deliberative 'ought' (i) matters directly for advice; (ii) functions to close deliberation; (iii) is characteristically tied to assessments of agents' accountability; (iv) implies 'can';6 and (v) is closely related to obligation.7 We agree that these differences between readings exist, but reject the claim that the Uniformity Thesis cannot accommodate and explain them. How can ambiguities in 'ought' sentences coherently be admitted without accepting the Ambiguity Thesis? The issue between the Uniformity and Ambiguity Theses more precisely is whether any of these sentential ambiguities are due to the word 'ought' itself having more than one meaning. Other explanations of different readings of 'ought' sentences are possible that are compatible with a uniform semantics for the word 'ought' itself. One possibility is an ambiguity in some other part of the sentence. A solution of this kind is the agency-in-the-prejacent theory of Nuel Belnap and John Horty. This theory explains the difference between evaluative and deliberative readings of 'ought' sentences by postulating an implicit agency operator, stit ('sees to it that'), present in deliberative readings and absent in evaluative readings.8 While Schroeder discusses and rejects the agency-in-the-prejacent theory, other explanations of sentential ambiguities are available to uniformity theorists that to our knowledge have not previously been explored. 'Ought' sentences could be context-sensitive, in either a semantic or a pragmatic way. (i) If 'ought' were indexical like 'now' and 'I' then it would contribute differently to the proposition expressed by a sentence, as determined by a uniform semantic rule that selects a value from the context of use. The word 'now' does not have a different semantic meaning at each different time of use, for example, though it is used to refer to different times. Alternatively, (ii) if 'ought' is semantically incomplete in having one or more implicit places filled by some value intended by the speaker—as, for example, 'poisonous' has an argument place for a type of individual: poisonous to whom or what?—then a uniform semantics for 'ought' would allow indefinitely many different readings of the same 'ought' sentences. Interpreters would identify the intended proposition through pragmatic cues arising from the speakers' choice to use those particular sentences in those particular contexts.9 Our solution in section 4 is a form of this second strategy, involving semantic incompleteness resolved by pragmatics. Ambiguity theorists believe that the evaluative/deliberative ambiguity provides an especially and perhaps even uniquely strong case for an ambiguity in 'ought' itself, because it appears to be not merely semantic but also logical (syntactic).10 Grammatically, this means that these two senses of 'ought' would obey different rules in composing meaningful sentences with other words. Whereas the evaluative 'ought' takes a sentence as its object and logically expresses an operator on propositions, the deliberative 'ought' takes two objects, a noun referring to an agent and an infinitival verb phrase referring to an action, and logically expresses a relation between the agent and the action. Consequently the Ambiguity Thesis yields a testable kind of prediction: that 'ought' sentences with certain syntax will only admit one kind of reading and not the other. have natural evaluative readings as equivalent to (1) and (2) respectively, lacking Schroeder's five hallmarks of the deliberative 'ought'. The linguistic explanation of this is that 'ought' is a raising verb, like other sentential operators in English. An idiosyncrasy of English grammar is the requirement that every sentence have a grammatical subject (noun or noun-phrase). For example, although the word 'might' takes a whole sentence (like 'Today is Sunday') as its object, it is ungrammatical to say 'Might be that today is Sunday'. One way of complying with the rule is by employing an expletive or non-referring subject term like 'there' or 'it', as in 'It might be that today is Sunday'. This is found in the nonagential 'ought' sentences like (1) and (2) with the constructions 'There ought to be…' and 'It ought to be that…'. A different way of complying with the rule is by "raising" the noun from the subordinate or prejacent sentence (e.g. 'Today is Sunday') to the subject position before the verb (e.g. 'Today might be Sunday')—hence the name "raising verb". English grammar therefore allows the nonagential sentence (2) to be transformed via subject-raising into the agential sentence (2a) without change in meaning or logical form. The upshot is that if the Ambiguity Thesis is correct then agential 'ought' sentences will be logically ambiguous between (i) evaluative readings where 'ought' is a raising verb like 'might', and (ii) deliberative readings where 'ought' is rather a control verb like 'wants' and 'tries'. (A control verb has a semantic subject for its verb argument that is also the semantic subject of the control verb itself. For example, in 'Bill wants to kiss Lucy', Bill is the subject of both the control verb 'want' and its verb argument 'kiss'—i.e. the wanting and the kissing at issue are both Bill's.) Therefore it is impossible to confirm the Ambiguity Thesis by finding that agential 'ought' sentences do not admit evaluative readings, as it concedes to the Uniformity Thesis that evaluative readings of these sentences will always be available. However, the behavior of raising and control verbs do differ in some significant ways, and it is to these differences that ambiguity theorists must appeal for evidence that deliberative readings can only arise from a control sense of 'ought'. We turn now to examine this evidence. The central problem for the Uniformity Thesis is overgeneration. Semantic and logical uniformity seems to predict the availability of deliberative readings of 'ought' sentences which in fact only admit evaluative readings.11 There are two major pieces of data here. The first comes from nonagential 'ought' sentences, with the grammatical constructions 'It ought to be that…' and 'There ought to be…'. Only raising verbs can occur in these constructions. Control verbs cannot because they require genuinely referring nouns in subject position; hence 'It wants to be the case that…' and 'There wants to be…' are impermissible. The Ambiguity Thesis therefore predicts that nonagential 'ought' sentences should never admit deliberative readings, as they have only non-referring nouns in subject position. By contrast, the Uniformity Thesis entails that every agential 'ought' sentence can be grammatically transformed into a semantically and logically equivalent nonagential sentence. So it seems to predict rather that nonagential 'ought' sentences should admit all the same readings as their agential transformations, including the deliberative readings of those sentences. Our opponents claim that the data confirms the prediction of the Ambiguity Thesis against the prediction of the Uniformity Thesis: nonagential 'ought' sentences like (1) and (2) admit only evaluative and never deliberative readings. This is the first piece of the case for semantic and logical ambiguity in 'ought'. For convenience we will refer to 'ought' sentences like (5b) as passive.14 Here too the data is claimed to confirm the prediction of the Ambiguity Thesis and not that of the Uniformity Thesis. Whereas (5a) has a deliberative reading, "on which it is appropriate for advice for Bill, on which it settles Bill's deliberative question of what to do, on which Bill is accountable if he does not kiss Lucy, which implies that Bill is able to kiss Lucy, and which is on a par with the claim that Bill has an obligation to kiss Lucy",15 it is maintained that (5b) allows no deliberative reading. This is the second major piece of the case for semantic ambiguity in 'ought'. At first glance this seems a strong case for ambiguity.16 Our primary response will be to demonstrate that the Uniformity Thesis can accommodate and explain these data. However, we pause briefly here to voice some doubts about the data.17 While we agree that nonagential and passive 'ought' sentences do not naturally encourage deliberative readings, it seems to us that these readings can sometimes be available, especially with a suitable context. Consider the scenario of a dating game show where the host is about to reveal which of the contestants is obliged to give Lucy a kiss. In this context we find it quite natural to interpret his utterance of (5b), 'Lucy ought to be kissed by…Bill!' as having all five of Schroeder's hallmarks for deliberative 'ought's. The parallel point holds, we think, of 'It ought to be that…Bill kisses Lucy!' A reader might be tempted to suppose that the difficulty of obtaining these deliberative readings itself supports the Ambiguity Thesis over the Uniformity Thesis. This characteristic of nonagential and passive 'ought' sentences is certainly something that a defender of the Uniformity Thesis is obliged to explain; we try to discharge this burden below in section 4. But if our intuitions about the data are correct this presents a more serious problem for the Ambiguity Thesis, which predicts that deliberative readings of these nonagential and passive sentences should not be possible at all, just as there is no permissible reading of (7b) as equivalent to (7a). For the sake of argument however we will proceed by assuming that the ambiguity theorists' description of the data is correct. Only one strategy for explaining the difference between evaluative and deliberative readings of 'ought' sentences with a uniform, raising semantics for 'ought' has been directly considered by ambiguity theorists. On that strategy, the agency-in-the-prejacent theory, evaluative readings result from interpreting 'ought' sentences as containing an implicit agency operator, stit ('sees to it that') in the prejacent. A distinction can then be drawn between It ought to be that Bill kisses Lucy (evaluative) and It ought to be that Bill sees to it that he kisses Lucy (deliberative), as two possible readings of 'Bill ought to kiss Lucy'. Schroeder argues that this theory falls prey to the Overgeneration problem. Since it locates the ambiguity in the prejacent sentence, 'Bill kisses Lucy', it apparently predicts that this ambiguity would also occur in nonagential and passive 'ought' sentences, which are formed by applying 'ought' to the same prejacent. Even if this objection against the agency-in-the-prejacent theory is successful,18 there are other ways of explaining the ambiguity with a uniform, raising 'ought', which do not locate the ambiguity in the prejacent sentence. A contrastivist semantics for 'ought' provides the alternative that we wish to explore. According to contrastivism 'ought' claims are always relativized to a contrast class. For example, to claim that it ought to be the case that Larry wins the lottery is to claim of some set of alternative propositions {a, b, c, …} salient in the context that it ought to be the case that Larry wins the lottery rather than that any other of a, b, c, … obtains.19 Different contrast classes generate different readings for 'ought' claims; for example, saying that it ought to be that Larry wins the lottery rather than that a or b is to make a different claim than saying that it ought to be that Larry wins the lottery rather than that b or c. Contrastivism about 'ought' is independently well-motivated, having been embraced by numerous philosophers and for a variety of reasons.20 If the normative 'ought' is correctly glossed as meaning best, for example, then it would inherit its contrastive character from the contrastive character of better than.21 In (8a) the focus is on 'Bill', naturally suggesting a contrast with a set of alternative propositions which each involve some person other than Bill kissing Lucy; e.g. {Tom kisses Lucy, Jack kisses Lucy, Mary kisses Lucy, …}. In (8b) the focus is on 'kisses', naturally suggesting a contrast with a set of alternatives in which Bill acts in some different way towards Lucy; e.g. {Bill shakes hands with Lucy, Bill says goodnight to Lucy, Bill hugs Lucy, …}. In (8c) the focus is on 'Lucy', naturally suggesting contrast with a set of alternatives in which Bill kisses someone other than Lucy; e.g. {Bill kisses Mary, Bill kisses Jill, Bill kisses Tom…}. In (8d) the focus is on 'kisses Lucy', suggesting contrast with a set of alternatives in which Bill does other things, perhaps with other people. In the case of (8) where no part of the sentence is emphasized, possible contrasts include sets of alternatives in which something completely different happens, perhaps involving different people and actions. For example, A: "What happens next in your play?" B: "Bill kisses Lucy". In each case the focus makes it natural to read the 'ought' sentence as saying that the state of affairs that Bill kisses Lucy is better (in some sense) than any of the alternatives in the contrast class suggested by that focus. So, for example, we read (9a) as saying that it is better that Bill be the person who kisses Lucy (than that it be Tom, or Jack, or Mary, etc. who does), while we read (9b) as saying that it is better that what Bill does to Lucy is kiss her (than that he shake her hand, tell her goodnight, hug her, etc.). These natural readings of (9a)–(9d) all differ significantly from each other, showing that when supplemented with contrastivism, the Uniformity Thesis can accommodate major ambiguities in 'ought' sentences. These contrastivist ambiguities are plausibly relevant to the difference between evaluative and deliberative readings of 'ought' sentences. We hypothesize that to get a deliberative reading of an 'ought' claim for an agent S is approximately to interpret it as having a content that S could (if she justifiedly believed it) appropriately use to settle her deliberations about what to do. Deliberation aims at settling on one course of action out of a set of options lying within the agent's power. Hence, deliberating agents try to identify the best of the options within their power. We can then define the notion of a deliberative set for S: a contrast set of propositions involving alternative options S is able to choose.24 So the kind of contrast set naturally suggested by the focus on (9a), for example, i.e. {Tom kisses Lucy, Jack kisses Lucy, Mary kisses Lucy, …}, will not be a deliberative set for Bill, or presumably for any other agent.25,26 But each of (9b)–(9d) can naturally be read as suggesting a deliberative set for Bill. He may deliberate over whether to kiss Lucy, rather than shake her hand, tell her good night, or hug her; here (9b) will be at issue. Or he may deliberate over whether to kiss Lucy, rather than Mary, Jill, or Tom; here (9c) will be at issue. Our contrastivist hypothesis is that to get a reading of an 'ought' sentence as deliberative for S, it is necessary that it be interpreted as relativized to a deliberative set for S. (We do not claim that a deliberative set interpretation is also sufficient; there may be other necessary conditions).27 The availability of deliberative set readings can be expected to depend on a variety of other, pragmatic factors in addition to focus. Some 'ought' sentences, like 'Larry ought to win the lottery', will resist deliberative set readings because they involve events that aren't sufficiently in the agent's power. Some contexts of utterance will strongly suggest deliberative set readings as what is required for a cooperative contribution to the conversation. Whispering to Bill 'You ought to kiss Lucy!' while he is saying goodnight to her on her doorstep, for example, will strongly suggest a deliberative set reading like that made salient by (9b), since the alternative nondeliberative set readings—like telling him that it is better that he kisses Lucy, than that Tom, Jack, or Mary does—will not ordinarily be so cooperatively communicated in that context. Plausibly, an 'ought' sentence will satisfy Schroeder's five hallmarks for the deliberative 'ought' if (typically), and only if it is given a deliberative set reading, as follows. (i) Deliberation: as we've already observed, an agent deliberates to identify the best out of courses of action within her power.28 (ii) Advice: similarly, a cooperative advisor aims to inform an agent about which is best out of courses of action within her power. (iii) Implies 'Can': it is trivially true that 'ought' implies 'can' if relativized to a deliberative set, since we have defined this as a set of options within the agent's power. (iv) Accountability: plausibly, agents are accountable for choosing the best course of action within their power, and not accountable for what is not within their power. (v) Obligation: plausibly, an agent has an obligation-like relationship to an action (e.g. having most reason to perform it) if and only if it is the best action in her power. Contrastivism therefore offers a promising way to accommodate and explain a difference between evaluative and deliberative readings of 'ought' sentences without positing any semantic or logical ambiguity in 'ought' itself. Can this contrastivist theory escape the Overgeneration problem, the central objection to the Uniformity Thesis? Schroeder diagnoses the problem as an inevitable result of locating the ambiguity in the prejacent instead of in 'ought'. Since the contrastivist explanation locates it rather in an implicit argument place of the 'ought' might we thereby evade the objection? But the overgeneration data teach a broader lesson: that the 'ought' sentences resisting deliberative readings (i.e. passive and nonagential sentences) are precisely those whose grammar dictates that they can only involve a raising 'ought' and not a control 'ought'. It makes no difference whether the uniformity theorist locates the ambiguity in the prejacent or somewhere else; either way he faces the challenge of giving an explanation of this correlation that is not ad hoc. Ambiguity theorists maintain that this challenge cannot be met, and therefore that the only viable treatments of deliberative readings are those appealing to a distinct sense of 'ought' as a control verb. At first glance contrastivism about 'ought' seems to offer nothing that addresses the problem. If the ambiguity in 'ought' sentences arises neither from 'ought' itself nor from the prejacent sentence but rather from differences in a contextually salient set of alternatives, then one might reasonably suppose that mere grammatical differences in how 'ought' and the prejacent sentence are composed shouldn't make any difference for the availability of particular readings. Deliberative-set readings should be no less available for nonagential sentences like 'It ought to be that Bill kisses Lucy' and passive sentences like 'Lucy ought to be kissed by Bill' than for agential 'ought' sentences like 'Bill ought to kiss Lucy'. Semantically each member of these pairs of sentences does not have significantly different truth conditions. But the lesser efficiency of the second member of each pair indicates that a speaker who uses it has some special reason for selecting it over the alternative. For (10b), perhaps the speaker intends to signal that this particular plane is less likely to be late than is typical, or that the possibility it will be late is too negligible to merit comment. For (11b), the speaker may intend to signal that Bill had to stop the car in an unusual way because the brakes failed.30 We shall now argue that a uniform contrastivist semantics for 'ought' can explain the overgeneration data by appealing to Efficiency. Consider first the case of a passive 'ought' sentence like 'Lucy ought to be kissed by Bill'. This sentence is less efficient than its active counterpart 'Bill ought to kiss Lucy', and so Efficiency tells us to expect a special reason for its use. Hence, the ambiguity theorists' claim that the choice of active or passive voice makes no difference to our interpretation of a sentence is simply mistaken. The passive voice differs from the active voice in where it places focus. In particular, passive voice provides a means of de-emphasizing the agent by eliminating reference to her or him; compare 'The cookies were stolen' with 'Someone stole the cookies'.31 This cannot be done with the active voice because of the requirement that every sentence have a grammatical subject, which in active-voice sentences must be the agent-noun. We have already observed that shifting focus is a natural way of controlling the salience of different contrast sets. A natural hypothesis is therefore that passive 'ought' sentences resist deliberative readings because the use of passive voice has the effect of placing focus on the prejacent in a way that strongly raises the salience of nondeliberative contrast sets. So, for example, 'Lucy ought to be kissed by Bill' places focus on the prejacent in a way that suggests an evaluative contrast set like {Tom kisses Lucy, Jack kisses Lucy, Mary kisses Lucy…}, rather than a deliberative contrast class like {Bill kisses Mary, Bill kisses Jill, Bill kisses Tom…}. This hypothesis receives support if we consider which readings are natural when a contrast is made explicit. For example, 'Lucy ought to be kissed by Bill rather than J' is more naturally read as meaning that Bill rather than J ought to kiss Lucy, and not as meaning that Bill ought to kiss Lucy rather than kiss J. We can increase the salience of the latter contrast set by adding intonational focus; e.g. 'Lucy ought to be kissed by Bill rather than J'. Can the use of passive sentences be predicted to have this effect on the basis of the contrastivist account of the deliberative/evaluative ambiguity? We argued in the last section that focus determines the dimension of contrast; e.g. that putting the focus on the agent (e.g. 'Bill ought to kiss Lucy') suggests an evaluative set in which the alternatives differ by agent, like {Tom kisses Lucy, Jack kisses Lucy, …}, while putting the focus on the patient (e.g. 'Bill ought to kiss Lucy') can suggest a deliberative set in which the alternatives differ by patient, like {Bill kisses Lucy, Bill kisses Mary, …}. What we need to explain is that passive 'ought' sentences suggest an evaluative set. So we should expect to find that passive sentences place focus on the agent rather than on the patient or elsewhere. This looks like a problem: haven't we found that passive voice functions rather to shift focus away from the agent? But this problem is easily solved. We observed above that the passive voice has this function because it allows elimination of any reference to an agent. The corresponding passivization of (5a) is therefore 'Lucy ought to be kissed'.32 But significantly, the passive 'ought' sentences like (5b) we are considering are more verbose: 'Lucy ought to be kissed by Bill' (compare 'The cookies were stolen by somebody'). Here the speaker has gone to the extra length of attaching a complement clause ('by Bill') that is not grammatically required in order to explicitly refer to the agent. Efficiency tells us she must have a special reason for doing so. Any special reason for referring to an agent places focus on that agent. Hence passivization can also be a device for putting extra focus on the agent-noun moved out of subject position.33 By contrast, explicit reference to an agent in an active voice sentence like an agential 'ought' sentence cannot by itself place focus on the agent, since for the active voice this reference is necessary to supply a grammatical subject. We therefore have a pragmatic explanation for why passive sentences but not agential sentences will resist deliberative readings if a uniform contrastivist semantics for 'ought' is correct. Passive sentences like (5b) place focus on the agent by virtue of the inefficiency of their reference to the agent, thereby indicating an evaluative contrast set. By contrast, because agential sentences like (5a) 'Bill ought to kiss Lucy' are maximally efficient, their grammar does not itself place focus on any word, permitting both evaluative- and deliberative-set readings. This solution for passive 'ought' sentences does not directly extend to the other half of the Overgeneration problem, involving nonagential sentences like 'It ought to be that Bill kisses Lucy'. Here grammar does not seem to place focus on any particular part of the prejacent 'Bill kisses Lucy', so we do not yet have an explanation why nonagential sentences would resist deliberative readings if contrastivism is correct. However, these sentences are significantly more verbose and awkward than their agential counterparts, so Efficiency can be applied here too. When interpreting the use of a nonagential sentence, we can assume that the speaker has a special reason for using the less efficient sentence. Why doesn't she rather exploit the ability of 'ought' to raise a subject from the prejacent? While the solution to this seems less obvious to us than in the case of passive sentences, there are some plausible answers we can explore. An obvious reason for using the nonagential sentence would be that the speaker intends to prevent a reading that would be salient if she used the most efficient, agential sentence instead. This is to invoke a pragmatic maxim to avoid potentially misleading ambiguity.34 Since nonagential sentences clearly resist deliberative readings while (active voice) agential sentences at least often encourage deliberative readings (as evidenced by our ability to set up the contrast between evaluative and deliberative readings in section 1 by presenting examples of these two kinds of sentence), this suggests that use of an inefficient nonagential sentence must indicate that the speaker intends a nondeliberative set reading. Since the agential sentence is more efficient, speakers can be expected to use it for merely evaluative 'ought' claims so long as there are enough other cues to make the intended evaluative contrast set overridingly salient; for example if the proposition expressed by the prejacent isn't a deliberative option for the agent. This is what we find with (1a), 'Every election ought to be free and fair', and (2a), 'Larry ought to win the lottery', which strongly encourage an evaluative reading due to the nature of their content. Use of a nonagential sentence would therefore be warranted just in case an unwanted deliberative reading would otherwise be salient. This tendency of nonagential sentences to indicate evaluative readings is something that contrastivism needs to explain, however, and not something it can simply help itself to. Since agential sentences are ambiguous between deliberative and evaluative readings an explanation is needed for why nonagential sentences would be used to block one kind of reading and not the other. We propose two different but complementary answers, one based on characteristics of agential sentences, the other on characteristics of nonagential sentences. First, we hypothesize that use of agential sentences carries a default presumption of relativization to a deliberative set. This could be a generalized conversational implicature based on the normal use of 'ought' for practical purposes like giving advice, expressing reproach, and concluding deliberation.35 Or perhaps raising an agent-noun to subject-position functions to place focus on the action, generating a deliberative set for the agent like {(Bill) kisses Mary, (Bill) hugs Lucy, (Bill) greets Tom, …}. Or alternatively, perhaps raising an agent-noun is simply a semi-conventionalized communicative device for indicating a deliberative set for that agent. This cue would of course be highly defeasible, given how easily we can use agential sentences to express evaluative claims, especially by use of focus. But in situations where it is not defeated such a presumption would provide a reason to use a less efficient nonagential sentence if an evaluative set is intended instead. These may include situations where the prejacent sentence naturally lends itself to a deliberative reading (e.g. 'It ought to be that Bill kisses Lucy'), or where focus is more difficult to utilize as in written communication (since italics allow less nuance than intonation). Second, we can look at the intrinsic grammatical features of nonagential sentences themselves. One significant feature is that they provide the only natural way of forming an 'ought' sentence that keeps its prejacent sentence grammatically intact; i.e. 'Bill kisses Lucy' appears intact in 'It ought to be that Bill kisses Lucy' (nonagential), but not in 'Bill ought to kiss Lucy' (agential) or in 'Lucy ought to be kissed by Bill' (passive). We can therefore reason from the speaker's choice to use an inefficient nonagential sentence rather than grammatically split the prejacent, to her having a special reason for keeping the prejacent together. This places focus on the prejacent sentence as a whole rather than on any proper part of it, which makes nonagential sentences uniquely well suited for indicating evaluative contrast sets where the whole prejacent rather than any proper part is the dimension of contrast. Neither passivization nor intonational emphasis can generate this kind of evaluative contrast class. For example, suppose two playwrights are discussing what event should occur next in their new play, and the salient set of mutually exclusive options is {Tom shoots Jack, there is an earthquake, Bill kisses Lucy}. Here the nonagential sentence 'It ought to be that Bill kisses Lucy' seems a more natural choice than 'Bill ought to kiss Lucy' (with any kind of emphasis), or 'Lucy ought to be kissed by Bill'.36 We conclude that by adopting a contrastivist semantics the champion of the Uniformity Thesis can provide a pragmatic explanation of the difficulty of obtaining deliberative readings of nonagential and passive 'ought' sentences, resolving the Overgeneration problem. However, we have shown only that such an explanation is possible and not that it is correct. How can we tell which is superior: our contrastivist and pragmatic explanation on behalf of the Uniformity Thesis, or the control verb theory of the Ambiguity Thesis? The two theories do generate some different predictions. According to the Ambiguity Thesis the difference between evaluative and deliberative readings is fully determined by the semantics—two different meanings for the word 'ought'—and therefore deliberative readings of nonagential and passive 'ought' sentences will be not merely difficult to obtain but impossible. But on our contrastivist supplement to the Uniformity Thesis the difference is partly pragmatic, and therefore depends upon particular circumstances of use. It therefore predicts that in some atypical circumstances it should be possible to get deliberative readings of nonagential and passive 'ought' sentences. If Schroeder's claims about the data are correct this would favor the Ambiguity Thesis over the Uniformity Thesis. However if our challenge to these claims at the end of section 2 was correct (the dating game show case, for example), this would give the advantage to the Uniformity Thesis.37 In our judgment the Uniformity Thesis fares better than the Ambiguity Thesis here. We have argued that the central objection raised by ambiguity theorists against the Uniformity Thesis, that it overgenerates by making false predictions about the availability of deliberative readings of 'ought' sentences, fails. The alleged evidence can be accommodated and explained by appeal to intuitive and well-established pragmatic principles if we adopt a uniform contrastivist semantics for 'ought' that is already popular and independently well-motivated. While we leave open the possibility that competing, noncontrastivist semantics for 'ought' might be able to offer parallel pragmatic defenses of the Uniformity Thesis, the solution we have offered here exploits resources that seem proprietary to a contrastivist semantics. It depends crucially on the idea that the effect of focus on the interpretation of 'ought' sentences is to control the salience of contrast classes, in keeping with the standard view of focus. In addition to defending the Uniformity Thesis this paper therefore provides a novel argument in favor of contrastivism about 'ought'.38 The failure of one objection to the Uniformity Thesis does not prove it correct, or that 'ought' lacks a control sense. But parsimony requires that senses of a word not be multiplied beyond necessity, so the burden is on the ambiguity theorists to provide other, better arguments for a control sense.39 We do not believe this burden is met. If our suggestion above is also correct that the linguistic evidence more strongly supports the pragmatic, contrastivist explanation of the deliberative-evaluative distinction than the hypothesis of a distinct control sense of 'ought', and that the control hypothesis is also subject to significant linguistic objections (as argued in Chrisman forthcoming), then it would be safe to conclude that the alleged control sense of 'ought' representing a relation between an agent and an action is one 'ought' too many. Mark Schroeder has suggested that this debate over the semantics and logical form of 'ought' has important bearing for normative branches of philosophy. While we agree that there may be important philosophical lessons to be learned from the correct analysis of 'ought', we are skeptical that any such lesson follows directly from settling whether or not 'ought' has a control sense representing a relation between agents and actions in addition to its raising sense representing a propositional operator. Much depends on the details of the semantic and logical analyses provided for these senses. 'Ought' sentences may still harbor significant ambiguities even if the Uniformity Thesis is correct, including an ambiguity between deliberative and merely evaluative readings as we've demonstrated. These different readings may also be indicated by grammatical differences so that there is no license to pass from 'Bill ought to kiss Lucy' (as naturally interpreted in a particular context) to 'It ought to be that Bill kisses Lucy' or 'Lucy ought to be kissed by Bill' (as naturally interpreted in the same context), or vice versa. And despite our skepticism about a control sense of 'ought' we agree with ambiguity theorists that on a deliberative reading, utterance of 'ought' sentences communicates that the agent stands in a special normative relation with the action; roughly the relation of having most reason to perform it. On our view this is simply because this is how agents are related to the action φ such that it is better that they φ than that they do any relevant alternative in their power.40 Hence this relation need not be semantically represented in order to be communicated by an 'ought' claim. (Just as, for example, telling someone that they ought to φ communicates that they have the ability to φ without that being part of what is said.) What if the Ambiguity Thesis were correct, despite our arguments? Schroeder suggests this would jeopardize certain metaethical theories about the nature of normativity. Some arguments for these theories would indeed require modification. But the plasticity of natural language—the ease with which words of one syntactic category can be transformed into related words of a different category—shouldn't be overlooked.41 For example, the creation of new verbs out of nouns (and vice versa) has a simple template. If there is a noun 'N', and some behavior of φ-ing is saliently characteristic of Ns, then a verb 'to N' can spontaneously come into use as meaning to φ (e.g. 'to snake', 'to hammer'). Similarly, if there is a raising verb/sentential operator 'Vraising', and some relation R characteristically holds between an agent S and an action of φ-ing when 'It is Vraising that S φs' is true, then a cognate control verb 'Vcontrol' can naturally come into use for the relation R.42 As we observed there is such a relation closely associated with at least agential 'ought' sentences. So although we've challenged the alleged evidence that a control sense of 'ought' exists in ordinary usage, it would be a mistake to insist this couldn't happen. But little of philosophical importance seems to depend on whether it does.