The Practitional Approach to Imperatives

Transcription

1 The Practitional Approach to Imperatives 1 Introduction It is a philosophical commonplace that there are two distinct forms of reasoning. One theoretical reasoning has to do with what is the case, whereas the other practical reasoning has to do with how to act. This fundamental two-way distinction between kinds of reasoning is reflected in two of the three universally attested sentence types. Declaratives, at base, seem to be for saying what (one believes) is the case, and imperatives, at base, seem to be for telling people what to do. 1 The distinction here is semantic: Close the door!, for example, clearly does not mean the same thing as You closed the door. A central task on the philosophy of language is to explain this basic semantic difference between declaratives and imperatives. 2 Guided by their previous work on the meaning of declaratives, theorists who have taken up this explanatory project have tended to do so in one of two ways. The first, which takes its inspiration from Frege, starts by dividing meaning into the separable elements of reference, 1 The third universally attested sentence type is the interrogative, and it can concern either the theoretical or the practical. On the theoretical side, it can be used to ask what is the case or for an explanation of what is the case; on the practical side, it can be used to ask what is to be done or why something is to be done. We think this is best explained by analyzing the interrogative in terms of the declarative and imperative, which seem to be explanatorily more basic, but we will not pursue the matter here. 2 Imperatival meaning is addressed in discussions of illocutionary acts performed by using different sentence types, especially in the tradition begun by Austin (1962) and Searle (1969); see also Recanati (1981/1987). This literature, however, tends to focus on the use of imperatives to perform specific speech acts such as to command, request, advise, etc., rather than the semantics of the imperatival form that allows for such use. (We discuss this difference below.) The semantic question is addressed more directly but less systematically in discussions of the force/content distinction in Frege ( /1984); see especially Dummett (1973) and McGinn (1977). Somewhat more systematicity with respect to imperatival meaning specifically is developed by Pendlebury (1986), Hanks (2007, 2015), and Parsons (2012). Several formal semanticists and logicians have been interested in the issue of modeling the semantic contribution of imperatival mood, though only with varying degrees of interest in attached philosophical questions. See especially Stenius (1967), Han (1998), Portner (2004, 2007, 2009), Mastop (2005), Barker (2012), Fox (2012), Charlow (2014), and Starr (unpublished). See also Hare (1949, 1967) and Vranas (2008, 2010, 2011) for related work on the logic of imperatives. We discuss some of this work below. 1

2 sense, and force. 3 All sentences, declarative and imperative alike, are said to have propositions as their senses, constrained in some way by the referents of their terms and how these terms are put together. What distinguishes imperatives from declaratives, then, is their respective forces. Declaratives put forward their propositional contents in the Ur-way, i.e., as true. By contrast and in modification, imperatives put forward their propositional contents as to be made true. Because it marks the distinction between declaratives and imperatives in terms of the force with which they put forward their propositional contents, call this approach the force approach. The second approach starts instead with the idea that truth conditions are the core of meaning. Any other semantic distinctions we might want to make are to be traced to distinctions that can be made with the resources of truth conditions. Although it may be possible for this approach to subsume the force approach (perhaps, e.g., by grounding make-true force in truth conditions), its more common strategy is to identify a special prescriptive element in the truthconditional content attributed to imperatives, thereby distinguishing them from declaratives. For example, the imperative Close the door! has been said to have the truth condition that the speaker commands the hearer to close the door; alternatively, it has been argued that that its truth condition is that the hearer should close the door. Views of this sort, we will say, take the truth-conditional approach to imperatives. The goal of this paper is to suggest a different starting place and thereby motivate a new approach for explaining the semantic difference between declaratives and imperatives. In doing so, we aim to be sensitive to some of the most basic concerns of the traditions in semantics that underwrite each of the two approaches just mentioned; indeed, we take one of the attractions of this paper to be the extent to which it maintains envoys in each camp. We aim, however, to upend the order of explanation common to both, which, as indicated above, moves from a freestanding view of the declarative case to the imperative case by modification. Reflecting our be- 3 To be clear, we are not making any claims about Frege s own view. For a discussion of Frege s views and how they may have changed over his career, see Huntley (1980); see Hanks (2007) for a contrasting view. 2

3 lief (which we will not defend here) that theoretical and practical reasoning are equally fundamental, neither reducible to the other and both crucial to understanding each other, we want to suggest a framework for thinking about the meaning of imperatival sentences that treats declaratives and imperatives as equally fundamental. Motivating this new approach will principally involve two lines of discussion. The first concerns whether the explanantia of imperatival meaning must be, as Portner puts it, imperative-specific tools (forthcoming, sec ). Along with a growing number of theorists, we believe that such imperative-specific tools are required. We part with these allies, however, by insisting that the development of these semantic tools requires paying close attention to the distinctive subject-predicate sentential structure of imperatives. We will argue that this structure has not been adequately examined and that, as a result, even the best contemporary accounts of imperatival meaning require further development. To work up to this claim, we will open by critiquing the force approach and the truth-conditional approach. We are not the first to challenge these approaches. To those familiar with other challenges, our critiques may seem superfluous. 4 On the other hand, committed proponents of either of the two approaches may be less than completely persuaded by our arguments. The goal of these arguments, however, is not to improve upon previous challenges nor to convince committed proponents to change their minds. Rather, our aim is to articulate desiderata for any semantic account of imperatives by highlighting some serious difficulties that the force and truth-conditional approaches face. The second line of discussion concerns our alternative approach. We will argue that imperatival content is sui generis and equally basic to declarative content, much as one might (and we do) think that practical reasoning is sui generis and equally basic to theoretical reasoning. To try to triangulate on an explanation of something we believe to be sui generis, we will first explain how this content can be understood by analogy to more the familiar propositional 4 Indeed, some contemporary philosophers and linguists dismiss the force approach out of hand and expend most of their critical energy attacking the two versions of the truth-conditional approach (see, e.g., Zanuttini, Pak, and Portner (2012: ) and Charlow (2014: )). 3

4 content of declaratives. This is where we want to suggest new resources for the tool-kit of semantics. We will propose that, along with the familiar notion of propositions, we should recognize a relatively unfamiliar notion of practitions. At the most simple logical level, we think of these as structured abstract contents typically comprising an agent-term and an action-term much as propositions are commonly conceived as descriptive contents typically comprising a subject-term and a property-term. However, whereas propositions connect their subject and property terms to form something truth-apt, practitions connect their agent and action terms to form something semantically evaluable in other ways, such as with respect to satisfaction or legitimacy. 5 Of course, there is no uncontroversial conception of propositions in this arena, so just as we cannot address every imaginable force or truth-conditional view, we cannot address every imaginable view of the nature of propositions. This does not prevent us from raising important questions about how practitions compare and fit together with more familiar elements of the tool-kit commonly used to theorize about the meaning of sentences expressing propositions, such as sets of possibilia, contentful attitudes, structured representations, Boolean operations, entailment, models of dynamic discourse functions, etc. The standard semantic tool-kit is already multifarious and complex, so we have no chance here of fully answering questions about how practitions fit with other tools. We nevertheless think we can show how fertile this relatively unexplored ground is by roughing in some initial trenches and planting a preparatory few seeds. 2 The Force Approach As we noted in the introduction, views that take the force approach to explaining the meaning of imperatives separate the propositional content that imperatives supposedly share with correla- 5 Here we draw on Vranas (2011), who argues that there are two distinct candidates for understanding the validity of imperatival inference. He calls these obedience preservation and bindingness preservation and argues that confusion arises from not distinguishing between these and understanding their different roles in understanding the content of imperatives. We discuss this idea in sec. 5, where we make some brief and provisional comments on the logic of imperatives. 4

5 tive declaratives from the force with which this content is put forward. As a first step towards understanding this approach, consider the following pair of sentences: (1) Close the door! (2) The door is closed. If we compare these sentences based on the words they commonly share, we may suppose that each sentence concerns the same object (the door) and the same property (being closed). If that is right, then drawing on a common and widespread view, we may go on to characterize this sameness by saying that these sentences contain the same proposition but differ in the way each relates the proposition to truth: whereas (2) presents the proposition as true, (1) puts it forward as to be made true. The analysis we have just made limits itself to surface features of the sentences. This, however, seems intuitively wrong, for whereas (2) concerns only the door and the property of being closed, (1) also involves someone specifically, the addressee performing the act of closing the door. Once our attention is drawn to this fact, we have reason to think that, if (1) has a propositional corollary, it is not the proposition expressed by (2). Such a corollary needs to specify an agent closing the door if it is to run parallel to (1). Taking a cue from Searle, we might take the following to be properly parallel pair 6 : (1) Close the door! (3) You will close the door. If this is the correct analysis, then the shared propositional core concerns the addressee closing the door in the future, but whereas (3) asserts that the proposition is true, (1) puts it forward as to be made true. The proponent of the force approach might put this by saying that (3) and (1) have a common propositional content, but whereas (3) has is-true force, (1) has make-true force. 6 Although (3) is inspired by Searle, Searle himself would use H to designate the hearer of the sentence where we have instead used you (see Searle (1969: 57-71)). We thank an anonymous referee for pointing out Searle s relevance to the present discussion. 5

6 By distinguishing declarative meaning from imperative meaning in this way, the force approach aims to capture a deep difference between two of the universally attested sentential moods. Since the speech-act analyses by Austin (1962) and Searle (1969), however, one must be careful when one speaks of a sentence s force. As they both point out, one and the same sentence form can be used to perform all manner of speech acts; with imperatives in particular, one can state commands, give advice, make requests, issue warnings, etc. 7 Each of these may be conceived of as involving its own sort of illocutionary force. The force of the force approach to imperatival meaning is of a different order, for it is supposed to be a feature shared by all imperatives, which is why it figures in explanations of the meaning of the imperatival sentential form rather than in explanations of the various uses to which this form can be put. Indeed, one might expect this shared force to play a role in explaining what unifies the diverse speech acts one can perform by uttering an imperative sentence. Moreover, if illocutionary force belongs exclusively to the pragmatics of language use, which are (according to orthodoxy) separable from the semantics, then this shared force of the force approach is not illocutionary force at all, since on this approach it plays an essential role in generating the meaning of imperatives as such. 8 For these and perhaps other reasons, linguists call the force of the force views sentential force, which is to be distinguished from (pragmatic) illocutionary force. 9 Once we have 7 For a detailed list of things one can do with imperatives, see Starr (unpublished, section 5.1). For a careful presentation of the notion of illocutionary force and how it must differ from grammatical mood (or what below we call sentential force), see Green (2009). 8 This point is made by Sosa (1967). 9 See, e.g., Chierchia and McConnell-Ginet (1990), Portner (2004), and Lohnstein (2007). We thank Paul Portner for helping us clarify this. Conceived as an element of the meaning of imperative sentences, sentential force may just be a different name for sentential mood. These ideas are not, of course, unconnected to illocutionary force. Sentential force/mood might be seen as the conventional linguistic encoding of something like the canonical range of illocutionary act potentials carried by a small set of grammatically distinguishable sentence types. Although almost any sentence can in principle be used to perform almost any kind of illocutionary act, imperatives seem to exist for commanding, requesting, advising, inviting, etc. Theories of sentential force/mood attempt to explain what it is about the meaning of imperatives that makes them such. 6

7 labeled this distinction, we might sensibly demand that force theorist explain the relations between sentential force, illocutionary force, and propositional content. 10 Once this demand is in focus, however, the force theorist faces a prior and potentially thorny problem. Part of the appeal of the force approach, we think, is its simplicity: it takes a well-studied theoretical notion that of a proposition and accounts for imperatival meaning by modifying that entity with a simple operator a sentential force. It is not clear, however, how this approach should characterize logically complex sentences that embed an imperative with a declarative under a sentential operator. These complex sentences are not uncommon. This should be obvious in the case of conditionals: (4) If it starts to rain, bring in the laundry! is an unremarkable thing to say. Conjunction and disjunction also allow for mixed embedding consider the following: (5) Get some matches, and I ll start collecting tinder. (6) Bring some peanut brittle, or I ll get some. Any full account of the meaning of imperatives will need to explain these and related cases. This is not an easy task. 11 These cases are challenging for any view, but they seem particularly problematic to the force approach. To see why, consider how the approach can analyze (6). Suppose we treat (6) as having a single, complex proposition as its content (or sense ), whose meaning is determined by the disjunction of its atoms. If force is strictly an operation on propositional sentential con- 10 Recanati (1981/1987) argues that utterances of imperatives differ from utterances of declaratives in having directive force as part of their meaning, which he cashes out in terms of having a representational content combined with a to-be-made-true direction of fit, whose fit is supposed to come about through some action of the hearer caused by the utterance. This comes closer to the notion of semantic force, but note that it is a view about the content of utterances rather than the content of sentences. 11 In section 5.2 we say more about the logic of imperatives. See fn. 46 for a list of some prominent attempts to address the challenges posed by this logic. For more on the challenges themselves, see Starr (unpublished: 3.1) and Charlow (2014). 7

8 tents, then the force theorist will need to assign exactly one force to (6). It would clearly be wrong to characterize the whole logically complex sentence as having is-true force, but it also seems wrong to characterize it as having make-true force. The mood of the second disjunct is declarative, and, were it uttered on its own, there would be no temptation to characterize it as an imperative. 12 To proceed this way, then, the force theorist will have to introduce new mixedmood forces. We know of no serious attempt to develop such an account, and we think this is for good reason. It seems that a mixed-mood force would mix its moods according to the logic of the sentence it operates on, which, it would seem, indicates that sentential force is in the end a feature of complex sentences atoms. If this is right, then what we might call the traditional force view, which treats force non-atomically, seems to be a non-starter. This might lead one to develop what we could call a progressive force view, maintaining that force is atomic and thus not extra-propositional in the case of complex sentences. To say this is to say that an imperative s force is not separable from its content but rather is part of that content. We think that something along these lines is the correct thing to say; indeed, this is at the heart of the practitional view we present below. To make this move, however, is to give up on much of the explanatory strategy that motivates the force approach in the first place. Again, the appeal of this approach is to start with propositional content, whose semantic operation within logically complex contexts is familiar and well studied, and then to describe imperatival meaning in terms of an operation on this kind of content. To treat the putative force element of imperatival meaning as internal to the content of a given imperative is to abandon this approach, for to go this route is to forgo being able to draw seamlessly on the logical structures into which propositions fit. To characterize complex sentences, for example, one will have to start from scratch; one will not be able treat conditionals, conjunctions, and disjunctions as (or determining) truth-functions, so one will need a radically new semantics for if-then, and, 12 Sentences of the form ϕ!-or-p often seem to mean ϕ! If you do not comply, p, which might suggest that imperatives never really embed under or. However, as Starr (unpublished) notes, that interpretation is not always plausible, as (6) in a normal context reveals. 8

9 and or. Again, we think careful attention to imperatives does, in fact, require reconsideration of these fundamental issues, and we hope to make a bit of progress on these matters in section 5. Our point here is that to do this is to give up on the force approach s familiar tack. There is another problem for the force approach that, we think, reveals constraints on an acceptable theory of the meaning of imperatives. The problem concerns the way in which imperatives seem, as a function of their meaning and not just how they are typically used, to direct agents to act. One of the most well-developed attempts to capture this within the force approach is by appeal to so-called stit ( see-to-it-that )-operators, for these views explicitly seek to explain the apparent agent-directing feature of imperatival meaning. 13 The appeal to a stitoperator is a popular strategy in deontic logic and the logic of agency for ensuring that some reference to agency is incorporated into the propositional content of an event description or the prejacent of agentive deontic modals. To see how appealing to stit-operators might work, imagine that Anne says (1) to Sarah. Because Anne has directed (1) specifically at Sarah, we could use Sarah as the subject of the stit-proposition: that Sarah sees to it that Sarah closes the door Then, a stit-analysis of the meaning of (1) in this context might be represented as follows: FORCE [make-it-true] + CONTENT [Sarah sees to it that Sarah closes the door]. Perhaps not all imperatives must specify a particular agent as the subject: if in response to a medical crisis one yells, Get a doctor! the subject of the stit-proposition is probably best char- 13 See Kanger (1957/1971: 42), Belnap and Perloff (1993), Belnap, Perloff, and Xu (2001), and Horty (2001: ch. 3). Compare inter alia Sosa, who treats imperatives as an operation Let it be the case on a proposition, which he formalizes as D:α, where the propositional variable α can be replaced only by nominal clauses describing possible states of affairs which are exhausted by someone s making something true (1967: 57-58). He cites Hare s (1952: ch. 2) distinction between neustic and phrastic elements of meaning as inspiration for the view. 9

10 acterized as someone. In our example, however, let us assume it is clear that Sarah is the one who is to see to it that the door is closed. Now, if Anne means what she says, then not only is Sarah to see to it that the door is closed, but Sarah herself must be the one who closes the door. If Sarah pays Julia to close the door, Anne might reasonably complain that Sarah has not satisfied the imperative. Had Anne said to Sarah merely, See to it that the door is closed!, there would be no grounds for such a complaint, for one can see to it that a goal is realized without being the one who realizes the goal. To be sure, Anne can utter (1) to Sarah without caring who closes the door, so Anne may be indifferent if in response to (1) Sarah pays Julia to close it. If Anne is indifferent, however, she would speak more accurately by telling Sarah to see to it that the door is closed. This reveals, we think, that to say (1) is to say something stronger than, See to it that the door is closed! : it is to direct Sarah to be the one who closes the door. The stit-account just sketched attempts to accommodate this point by linking the grammatical subject of sees to it that to the grammatical subject of the embedded verb phrase (e.g., close the door ). 14 Even when such a link is stipulated, however, there can be an important difference between a person s seeing to it that she does something, on the one hand, and her simply doing it, on the other. To appreciate the difference, consider Baier s (1971) complaint against analyzing intentions in terms of a bring-it-about-that relation. Her immediate target is Chisholm s (1971) claim that intentions can always be conceived as intentions to bring it about that p, which construes the objects of intentions as states-of-affairs, which, in turn, can be articulated by propositions. Baier argues that one s intending to ϕ is not always, and indeed is not typically, the same as intending to bring it about that one ϕs: she compares her intention to speak sharply to the boy who has been stealing her raspberries to the intention to bring it about that she speaks sharply to the boy. The goal of the former intention is to confront the boy, whereas the goal of the latter is a sort of self-manipulation. We think this is an important distinction and will discuss its application to imperatives at several points below. 14 On the importance of this linkage to stit-accounts, see Belnap and Perloff (1993: 28-29). 10

11 Return now to the stit-account we have been sketching. It characterizes (1), when directed at Sarah, as telling Sarah to see to it that she (Sarah) closes the door. It would seem that Sarah can see to it that she closes the door by bringing it about that she closes the door; that is, it seems that she can see to it that she closes the door by manipulating herself into being a doorcloser. If Baier s point is correct, such self-manipulation is different from Sarah simply and straightaway closing the door. If this is correct, however, it shows the stit-account under consideration to be insufficiently fine-grained. The application of Baier s point to imperatives implies that (1) is satisfied only if Sarah straightaway closes the door, but the stit-account developed so far counts (1) as satisfied either by a straightaway closing or by an act of self-manipulation. 15 One might respond to this worry by complaining that it is the result of the specific way we have constructed the account. To get around the worry, one might recommend analyzing imperatives of the form S, make it the case that ϕ as corresponding with S, stit-ϕ and distinguishing these from S, let it be the case that S ϕs, which could then be used to analyze (1). The latter is satisfied only if Sarah closes the door straightaway, not if she self-manipulates. This suggests that a way around Baier s point is to analyze ϕ! with Let it be the case that S ϕs! 16 Although this analysis gets around the problem of failing to discriminate imperatives that demand straightforward action from those that allow for self-manipulation, it does so by introducing an analysans that generates its own problems. We can put the matter in the form of a dilemma. Either letting something be the case is an action, or it is not. If it is not, then let it be the case that S ϕs is a poor analysis of ϕ! On this analysis, the meaning of Close the door! does not prescribe the addressee to do anything, for on this analysis, letting is not an action. Suppose, then, that to let something be the case is an action. Going this route, however, 15 To take a phrase from Velleman, our critique here is that this stit-account... fail[s] to cast the agent in his proper role (2009: 123). 16 We thank an anonymous referee for this suggestion. 11

12 opens the analysis back up to the problem that Baier s point poses. On this analysis, letting it be the case that Sarah closes the door will be satisfied just in case Sarah lets it be the case that she closes the door. Notice the structural similarity here to the stit- account: both make explicit that the addressee who is directed to make the fact true/let the fact be true/see to it that the fact is true is also the referent of the content s grammatical subject. In doing so, these analyses must specify the relevant individual twice, both as part of the fact to be made true and as the one to make the fact true. Because of this dual-specification, the relevant fact can always be made true by self-manipulation. If this is a general feature of any plausible force account and we suspect it is, since as soon as one splits force and content one would seem to need to locate the addressee in both of them it provides another reason for considering an alternative approach to explaining imperatival meaning. 3 Truth-Conditional Views We have just raised two objections against force-based accounts of imperatives, each of which, we think, teaches us something important about imperatival prescriptivity. By prescriptivity, we mean to pick out that feature of imperatives in virtue of which they seem to be semantically suited for telling people what to do. It is because of the prescriptivity of the imperatival type that to use an imperative with its literal meaning (to borrow an idiom from Searle) is to direct its addressee to do something. 17 As we see it, the prescriptivity of imperatives is a feature, in the first instance, not of their uses, but rather of their literal meanings. Although non-imperatives can be used prescriptively, imperatives seem to be essentially prescriptive. The first lesson from the previous section is that we need to explain prescriptivity in such a way that holds out the promise of accounting for logically complex mixed-mood sentences. Specifically, an account 17 See Searle (1978) for a discussion of these meanings. We hope this claim is intuitively compelling to native English speakers, for most English imperatives involve a phonologically null you. It might seem that the presence of third person imperatives in other languages pose a threat to our claim here we think they do not. We discuss this later in this section. 12

13 should make it possible to explain how one but not the other of the atomic sentences can be prescriptive. Second, an account should respect Baier s point, viz., that there is a difference between self-manipulation and normal action. Although some imperatives do allow selfmanipulation as a way to satisfy them (e.g., See to it that p! ), we should prefer views whose account of imperatival meaning discriminates this difference to those that do not. One way to respond to the first lesson is to maintain that, contrary to initial appearances, imperatives express propositions that determine truth conditions. As we noted in the introduction, such accounts exist, and they come primarily in two sorts; for reasons that will become obvious, we will call the first sort performative accounts and the second sort deontic modal accounts. 18 Both of these can be understood as locating prescriptivity in the distinctive proposition claimed to be expressed by imperatives. The performative version of the view starts by noting that in the standard case a person uses an imperative to direct its addressee to do something. This view can be articulated in terms of truth: in the standard case, it is true that the use of an imperative directs the addressee to do what the imperative states. Drawing on this fact, Lewis suggests that imperatives ought to be treated as paraphrases of the corresponding performatives, having the same base structure, meaning, intension, and truth-value at an index or on an occasion (1970: 57). 19 As an example, he considers (7) Be late! and suggests that it corresponds to 18 Zanuttini, Pak, and Portner (2012) call the former speech act views and the latter modal views. Charlow (2014) labels the class of views cognitivist and distinguishes between explicit performative views and modal views. 19 This is not the only possible performative view, but it is both plausible and elegantly simple, so we will focus on it. Davidson (1979) also floats a performative view, according to which imperative sentences have dual truth conditions, the first making prospective anaphoric reference to the second and setting its mood. For example, he would treat (1) as having truth conditions along the following lines: My next sentence is imperatival in force. You will close the door. See also Price (1988: ch. 3) and Jackson and Pettit (1998) for views similar to Lewis s. Bohnert (1945) offers still another truth-conditional analysis: he argues that imperatives have disjunctive truth conditions of the form Either you will φ or X will happen. 13

14 (7') I command you to be late. His view is that (7) has the same truth conditions as (7'). One might object here that commanding is too specific to capture the meaning of imperatives, since (7) clearly could be used to perform other prescriptive speech-acts such as recommending, suggesting, etc. However, we think it is open to Lewis to argue that there is some genus of which all of speech-acts an imperative can be conventionally used to make are species. If we call it prescribing, then a more nuanced version of the performative view would hold that (7) has the truth conditions given in (7'') I prescribe you to be late. Then, Lewis might argue that not only does this view allow for straightforward embedding of imperatives under sentential connectives, but it also captures the way imperatives seem, as a matter of their meaning, to prescribe something of an agent by writing prescriptivity into their putative truth-conditional content. The natural skeptical response to treating (7') or (7'') as a specification of the meaning of (7) is that whereas (7') and (7'') appear to have truth-apt contents, it seems odd to say that an imperative sentence such as (7) could be true or false. 20 We will elaborate this worry during our analysis of the deontic modal truth-conditional account. To set the stage for that analysis, we first want to present two other concerns we have about the performative account. Consider the generic terms of affirmation and dissent, Yes and No focus specifically on the latter. On Lewis s view, it seems that using No to reject what is said by an imperative amounts to denying that the prescription has been made. Not only is this clearly false, but it is also an inaccurate 20 We should note here that some (e.g., Austin) suggest that utterances of explicit performative sentences such as (7') or (7'') are not truth-apt (see Bach and Harnish (1979: ch. 10)). Even if this is correct, it will not save the performative analysis of the content of sentence (7). Even if specific utterances of (7') or (7'') are not truth apt, that does not change the fact that those sentences (as distinguished from their utterances) are the sorts of sentences that can embed under is true. Moreover, if utterances of (7') and (7'') are not truth-apt, their similarity to the way (7) is typically used would seem to support rather than undermine our contention that imperatives do not have truth-conditions. We thank an anonymous referee for drawing our attention to this. 14

15 analysis; to say No in response to an imperative is typically to refuse to comply with it, not to deny that it has been issued. This does not seem like a mere quirk of language, so we think Lewisian views are going to fail to characterize properly what it is to affirm or to repudiate an imperative. 21 This is not the only problem that Lewisian views face when subjected to what we might call a dialogical analysis. Consider the range of questions one can ask by saying Why? in response to an imperative. The addressee of (7) might ask Why? to query the motives of the person who has issued the imperative, but the addressee might also, given Lewis s view, sensibly use Why? to ask in virtue of what (7) is true. This is, at very least, bizarre; it is difficult to think of natural contexts in which an addressee might say Why? in response to the issuing of an imperative in order to ask in virtue of what it is true that the imperative has been issued. Lewis might respond by saying that although Why? can ask this particular question, as a practical matter, we rarely have a cause to ask it. The mere fact that his account presents this as a question one might intelligibly ask, however, would seem to tell against the view. For similar reasons, the account seems to be committed to the intelligibility of the following dialogue: A: Be late! B: Why? A: Because I want to hear myself prescribe something. The reason this dialogue is weird is that, when B asks A Why?, A would normally need to explain why B should comply with the imperative A has issued; A s response seems deaf to B s query Mastop (2005: 13ff) makes a similar point. 22 There are other related problems. As Parsons (2012) argues, the performative view has undesirable consequences about the entailment and inconsistency relations imperatives stand in. For example, it would seem to treat (3) as entailing that someone commanded someone to close the door, which it does 15

16 This last point is useful not only as a further argument against the performative view but also for revealing a motivation behind the deontic modal approach. 23 This approach begins by noticing the tight connection between saying φ! and You should φ. We can explicate this connection by noting that Why?, when asked of either, will often produce the same answer. Deontic modal versions of the truth-conditional view exploit this connection to explain the meaning of imperatives in terms of their corresponding should-sentences. To see how such an account might go, imagine that Jonah is working on a bicycle, and Marty tells him: (8) Grease the chain! Insofar as we restrict consideration to Marty s preferences with respect to Jonah, then the following would seem to be true: (9) Jonah should grease the chain. For this reason Kaufmann (2012: ch. 3) has argued that imperatives such as (8) have truth conditions like those of (9). As is well known in the literature on the semantics of modals, there are many readings of a should-sentence such as (9), depending on whether the conversational context in which it occurs explicitly or implicitly encourages the interpretation in terms of moral norms, prudential maxims, someone s preferences, or probabilistic standards. 24 Hence, the flavor of the should is partially determined by the context of use. For this reason, on Kaufmann s version of the deontic modal view, (8) has the same truth conditions as (glossing over a few details): (9') In light of the relevant rules, preferences, or ends, Jonah should grease the chain. not, and it would treat (4) as consistent with the sentence Don t be late!, which it seems not to be. Compare Charlow (2014: ) for further discussion. 23 Examples of this approach include Han (1998), Schwager (2006), Aloni (2007), Kaufmann & Schwager (2009), and Kaufmann (2012). We focus on the last of these because it is the most developed. 24 See Kratzer (1977, 1981). For helpful overview of the vast literature that has grown out of Kratzer s seminal work on the subject, see Portner (2009: 29-45, 47-84) and Hacquard (2011). 16

17 where the relevant rules, preferences, or ends are determined by the context in which the imperative is used. For example, if Marty s use of an imperative comes in the context of employee regulation (imagine that he and Jonah are mechanics at a bicycle shop), it might be interpreted in light of business rules, whereas if it comes in the context of Marty s desire for Jonah to get his hands dirty, it might be interpreted in light of Marty s desires. This helps to narrow the sense of should that featuring in the deontic modal analysis of imperatives, but note that not all should-claims that are relativized to rules, preferences, or ends direct agents to act. For example, in a context where we are outside observers on Larry, who is severely down on his luck, we might say (10) Larry should to win the lottery. 25 We are clearly not telling Larry what to do; indeed, if we were asked to tell him what to do, we might say that he should save his money rather than wasting it on lottery tickets. We can even use a sentence similar to (10) with no agent in mind whatsoever, e.g., (11) The lottery should be won by whoever needs the money most. Because of this, should-sentences such as (10) and (11) have been called evaluative in the sense that they merely evaluate the desirability of some state of affairs rather than tell someone what to do. The states of affairs evaluated may involve an agent per accidens, but this is not necessary. It seems, then, that capturing the prescriptivity of imperatives will require more than giving them truth conditions employing a should relativized to rules, preferences, or ends. It will require conceiving of this should as an agentive rather than evaluative deontic modal The example derives from Schroeder (2011), drawing on Harman (1973) and Geach (1982), who uses something very similar to argue that ought is ambiguous between agentive (his term is deliberative ) and evaluative senses. See also Finlay and Snedegar (2014) for a response to Schroeder, which shows how much of the data he attempts to handle semantically can be handled in terms of Gricean pragmatics. 26 That is, a should used to advise/recommend/direct an agent in doing something rather than merely to evaluate a state of affairs as good in some respect. There are a variety of names for what we call agentive should s ( deliberative, directive, performative, ought-to-do ) and for what we call evaluative should s ( political, propositional, critical, ought-to-be ). 17

18 Having isolated the relevant sense of should, let us return to our dialogical analysis. We noted earlier that asking Why? either of an imperative or its corresponding shouldsentence will often evoke one and the same answer. We are now in a position to see why this is so when it is so. Because it is the agentive should that features in the corresponding deontic modal, to ask Why? of this modal claim is to ask what good there could be in or what reasons there could be for performing the relevant action. When one says You should φ and the should in question is agentive, there is a presumption that there is some good in and/or are reasons for the agent s φ-ing. To ask Why? in response to this deontic modal assertion is to ask about the goodness/reasons the speaker takes there to be in the addressee s φ-ing. Why? can ask about this same goodness or reasons when it is asked in response to an imperative; indeed, an addressee asks the same question Why? if she responds to an imperative with Why should I? The deontic modal view offers an attractive explanation of this; in doing so, it is superior to the performative view. 27 This, we think, helps us learn still more about the distinctive prescriptivity of imperatives. An imperative is semantically suited to direct an agent to do something, which may include exercising her agency to be some way. 28 In general, the addressee can probe the reasons for doing the relevant thing by asking, Why? We have just seen that this same question can be asked of the deontic modal declarative that corresponds with the imperative. In either case, the agent who asks Why? does so to probe the basis of what she is directed to do; in either case, she is thus probing a direction. Whatever an imperatival sentence directs, therefore, can be directed by a declarative should-sentence; the deontic modal account of imperatives exploits this fact to explain the distinctive meaning of imperatives. 27 The analysis here draws on the discussion of Why? in Anscombe (1963). We have sought to remain neutral on whether the goodness and the reasons that the question probes can be state-given. 28 At the outset, we suggested that imperatives typically conjoin agent terms with action terms. The hedge typically was meant to allow for the case of imperatives prescribing someone s being in some way (E.g., Be there by 10! ). Here we use do something loosely to include being in some way; below we speak somewhat more precisely about imperatives mobilizing a concept around which an agent can organize her will. 18

19 This fact also provides a reason for thinking that imperatives have, as a Fregean might put it, fully saturated contents. A content is fully saturated if, at the highest node of its logical form, it involves both a concept and an object brought under the concept. Standardly, the object in question will saturate the concept by taking the form of a grammatical subject. In English, the grammatical subject of an imperative is typically phonologically null but appears to be semantically second-personal: if, e.g., I want to command you to run, I can simply say, Run!, and the fact that it is you who are to run is implicit. If it is not clear who an imperative is directed at, we can use a vocative, e.g., Amar, run!, which makes it clear that Amar is the relevant you who is to run. Now if you respond by asking me, Why?, you are asking me for reasons why you should run; again, you would be probing the same reasons had you asked Why? in response to my saying, You should run. This deontic modal sentence clearly has saturated content, as does the explication of Why?, i.e., Why should I run? The grammatical subject of each of these refers to the same person. The fact that Why? asks the same question of the corresponding imperative is thus best explained, it would seem, by taking the imperative to have fully saturated content, its grammatical subject likewise referring to the same person as the explicated version of Why? One might worry at this point that we are leaning too heavily on the surface grammar of English to make general points about imperatives, for there are languages that involve what we might call third person imperatives. 29 Indeed, even English seems to allow for them. 30 Examples of third person imperatives include the following: (12) Rain! (13) Maître d, someone seat the guests! 29 We thank Paul Portner for raising this point and for pointing us to all of the examples and citations discussed in this paragraph. 30 Potsdam (1998) suggests as much; example (13) below is from him. 19

20 (14) Layke tini baje aav e. (14) is a Bhojpuri sentence that translates roughly as The children come at 3 o'clock! ; it might be said to a teacher with the goal of directing him to ensure that the children arrive at These are interesting cases, but we think that in spite of what their grammar might suggest, the logical form of each is best understood as a fully saturated content articulating the addressee with a concept that the addressee is to realize through action. (12), we think, cannot be understood except as addressed to some imagined agent e.g., the gods, the fates, the personified sky, etc. for how else could it be imagined to be responsive? (13) and (14) both direct their addressee to do something: (13) directs the maître d to see to it that someone seats the guest, and (14) directs the addressee to see to it that the children come at 3. If the maître d does nothing relevant but the guests get seated, he has not satisfied the imperative; the same is true if the addressee of (14) does nothing relevant but the children arrive at 3. In both cases and, we suspect, in any similar case the addressee is called on to see to it that someone else does something. The surface grammar might disguise the logical structure of these imperatives, but in all of them, it would appear, the subject is the addressee, and it saturates a concept of seeing to it that someone else do something. 32 Returning to the main line of argument: we also commend the deontic modal view for satisfying the two desiderata presented at the beginning of the section. 33 Because it accounts for a given imperative s prescriptivity in terms of its propositional content and thus at the atomicsentential level, it is at least prima facie capable of explaining logically complex mixed-mood sentences. The view also respects Baier s point. It does not analyze, e.g., Run! as calling for the agent to see to it that she runs; rather, it connects the agent directly with the act of running, 31 Zanuttini, Pak, & Portner (2012) present example (14) and show that such imperatives are robustly attested in languages other than English. 32 This means that some imperatives are illuminated by a stit- type analysis. It would be surprising if this were not the case since, as we have already pointed out, See to it that p! is obviously imperatival. Our contention above, however, is that this is not an accurate analysis of basic imperatives. 33 One might argue that the performative view also satisfies these desiderata, but it does not reveal the connection between them that we discuss later in this paragraph, so we shall presently set it aside. 20

21 asserting that the act is something the agent, according to the imperative, should do. Moreover, the deontic modal view also reveals that these two desiderata are interconnected in a way that may not have been initially obvious. Both are satisfied by the deontic modal account because the account explains an imperative sentence s meaning in terms of how its addressee is to exercise her agency. In the basic case, this prescriptivity is presented as a way of linking an agent, the grammatical subject, with an action, which is part of the predicate. Because prescriptivity is thus presented as part of an imperative s propositional content, the account holds out the possibility of explaining mixed-embedding within the confines of the semantics and logic of propositional operators; because it immediately links the grammatical subject to the act to be performed, it respects Baier s point. All of this, we think, counts in the favor of the deontic modal version of the truth-conditional view of the meaning of imperatives. Unfortunately, the difficulties of treating imperatives truth-conditionally are, we think, simply insurmountable. Kaufmann seeks to explain away the appearance that imperatives are not truth-apt by pointing out that some deontic modal sentences have a performative rather than descriptive use. For example, in response to the question May I leave now?, someone with authority might say (15) Yes, you may leave now. This is best interpreted not as a description of what is permissible given the relevant rules but rather as the granting of permission. Although (15) embeds under It is true/false that, it would be weird for someone else to say, No, that s false, unless of course they were speaking from a higher authority and were canceling the granted permission. It thus seems that sentences that are grammatically and semantically embeddable in propositional contexts can have nondescriptive performative roles, which, Kaufmann claims, explains the non-truth-conditional appearance of imperatives. This explanatory strategy, however, is limited. To see why, note that many modal sen- 21