A Flexible Contextualist Account of ‘Ought’[1]
Draft at 8/30/09
J.L. Dowell
Introduction:
A complete metaethical theory has at least two parts. One part is linguistic: To provide a descriptive semantics, i.e. an assignment of semantic values, to our moral or, more broadly, practical expressions, and to provide a pragmatics for the use of expressions with the identified semantics. A second part is ontological and metaphysical: To take a stand on whether there are genuinely moral or normative properties and relations and, when that stand is affirmative, to take a further stand on their nature. There are many different contending linguistic theories, for example, Cornell realism,[2] emotivism[3] and its expressivist cousins,[4] Hare’s imperatival theory,[5] and various hybrid combinations of these views.[6] Some of these theories are perhaps best understood as revisionary attempts to reform our use of our moral expressions, rather than straightforwardly descriptive theories. But even revisionaries must acknowledge the importance of beginning with our evaluative expressions as we find them; after all, it’s these that they must show are in need of revision for revisions to have a point.
This paper is a contribution to the first of these two metaethical projects. In the literature in linguistics and philosophy of language, the canonical descriptive semantics for modal expressions (like “might”, “may”, “ought”, “necessarily”, and “must”) is that they are quantifiers over possibilities, where the domains of quantification are contextually restricted. My primary aim here is to defend my favored version of a contextualist semantics for modal expressions against several related objections to any contextualist account of their normative use. A secondary aim here is to raise the level of debate about the semantics of ordinary, moral discourse by relying on methodological principles that are standard among linguists and philosophers of language doing descriptive semantics, but, surprisingly, are rarely followed by metaethicists engaged in the same project.[7]
1. Contextualism
about Modals: The Canon
On the standard view, “a modal is an expression (like ‘necessarily’ or ‘possibly’) that is used to qualify the truth of a judgment”[8] (p.1). On the canonical view, modal expressions like “may”, “might”, “ought”, and “must” are quantifiers over possibilities. Their basic form is
MODAL(B)f,g( f)
where B is the domain of quantification or modal base, f, the prejacent, is a proposition that gets evaluated at the possibilities in B, and MODAL determines (roughly) which or how many of the possibilities in B are possibilities in which f must come out true for MODAL(B) f,g (f) to come out true. f and g each determine a restriction on the modal base, with g, when necessary, ranking the f-worlds.
In their standard uses, “may” and “must” require, respectively, that some or all of the possibilities in B are possibilities in which f comes out true. In some uses, “ought” is interchangeable with “must”. In its standard use, though, “ought” functions as a comparative[9] and is so not so interchangeable. In these cases, the value for g must rank the f-worlds.
In some cases, the use of an explicit, restrictor phrase sets the values for f or g. Here are some examples:
In view of the local traffic laws, visitors must stop at stop signs.
In view of Sally’s preferences, she may take the turnpike to get to Boston.
In view of what Holmes knows, the gardener must be the culprit.
Given the climate and soil conditions, hydrangeas can grow here.
In other cases, though, the restrictions are supplied by the context.[10] Following von Fintel and Gillies, I’ll call such modal statements ‘bare’.[11] On the canonical view, bare normative modal statements (BNMs) get restricted by the context of utterance in two ways. First, the set of all possible worlds is restricted to those compatible with a set of circumstances or (and here I offer a common[12] amendment to Kratzer’s canonical view) with a body of information. This sets the value for f. Second, the remaining worlds are restricted by a contextually selected standard.[13] This sets the value for g. The modals contained in both explicitly and contextually restricted statements are themselves neutral, i.e., have a single meaning across all contexts in which they appear.[14] ALL contexts really means all, not just the normative ones—the “must” in “water must be H2O”, “one must not kill the innocent” and “the butler must have done it” each make the very same contribution to the determination of a proposition. This is one of the theory’s singular advantages; in contrast to views according to which “must” or “ought” in English is ambiguous,[15] it provides a simple, highly unified explanation of a great variety of linguistic behavior, including a speaker’s ability to generate and understand new modal statements.
2. Puzzles for Contextualism about BNMs
2.1 A Puzzle about Practical Advice
Recently, several philosophers have offered a variety of objections to contextualist treatments of BNMs which may seem to suggest that the canon is in need of revision. One such puzzle concerns the possibility of receiving practical advice. Imagine a doctor deliberating about which of three drugs, X, Y, or Z to prescribe a patient to relieve the symptoms of her skin irritation. Doctor’s limited information suggests that X and Y each will provide complete relief, while drug Z will provide only partial relief. In light of this, she asserts:
Doctor: (X) “I ought to prescribe either X or Y.”
Suppose a consulting physician has more information about the patient’s medical history and each of the drugs than Doctor does. In particular, Consultant knows that while it is true that, absent any interfering drug already in the patient’s system, either X or Y would provide complete relief, it is also true that Patient is already taking drug W which together with exactly one of either X or Y is certain to kill the patient. He doesn’t know which drug would have which effect; each is equally likely to be the lethal drug. Z, in contrast is certain to provide some relief and certain not to have any negative side-effects. Given this, Consultant replies
Consultant: (Z) “No, it is not the case that you ought to prescribe either X or Y; you ought to prescribe drug Z”.
We can sum up the expected outcomes, given the information of the Doctor and Consultant, respectively, this way:
Drug X Drug Y Drug Z
Doctor 1.x complete cure 1.x complete cure 1.x partial relief Consultant .5 x Death + .5 x complete cure .5 x Death + .5 x complete cure 1.x partial relief[16]
The first question is: How can a contextualist capture the sense we have that Doctor and Consultant are each giving different and incompatible answers to a common question? If each of (X) and (Z) is relativized to the speaker’s information, then Doctor and Consultant are not disagreeing and Consultant’s assertion is not an answer to the same question as Doctor’s. But since Consultant’s “no” is felicitous, Doctor and Consultant do seem to be disagreeing. Moreover, Consultant seems to be advising Doctor by aiming to give a superior answer to the very question Doctor aims to answer with (X).[17]
A second question is: How can a contextualist capture our intuition that, in this case, we morally ought to do what we know to be suboptimal? Capturing this intuition in not the contextualist’s burden alone, but rather a burden for any theory of BNMs.
2.2 A Puzzle about
Attitude-Ascriptions and Normative Disagreement
The first puzzle concerns the contextual selection of an appropriate value for the information/circumstances parameter for uses of ‘ought’ that are pretheoretically relative to some body of information. A second puzzle for contextualism concerns the selection of an appropriate value for the standard. Several authors have offered cases that, they argue, pose problems for a contextualist account of attitude ascriptions with BNMs complements. One such puzzle arises because it is difficult to see which contextually determined information state or set of circumstances and standard could make sense of ordinary inferences involving belief attributions with such complements. Here’s a modified version of the puzzle discussed by Bjornsson and Finlay[18]:
1. Jefferson Davis believes that one must not aid fugitive slaves.
2. It is not the case that one must not aid fugitive slaves.
3. Therefore, Jefferson Davis believes something that is not true.
The inference looks sound. The puzzle is over how the contextualist can capture this. If #1 is to be true, the complement must characterize Davis’ state of mind. If it does, that complement will get its content determined by Davis’ moral standard, a standard that prohibits aiding fugitive slaves. But if context selects that standard, then the second premise is no longer true, since relative to that standard, one must not aid fugitive slaves. Suppose instead that context selects our standard (the ‘speaker’ in #2). In that case, #2 is true, since our standard permits aiding fugitive slaves. But this standard makes the first premise false—Davis doesn’t believe that one must not aid fugitive slaves relative to that standard. Perhaps, though, the inference involves a subtle context-shift: In the first premise, context selects Davis’ standard, but context has shifted by the time we arrive at the second premise; now it selects our standard. This has the virtue of making both premises come out true, but at the price of rendering the argument invalid.
Notice that a similar puzzle about the values for the information/circumstances parameter can be generated if we consider the how Consultant might reason to himself after Doctor has asserted (X). Consider the following representation of Consultant’s reasoning:
1. Doctor believes that she ought to prescribe either X or Y.
2. It is not the case that Doctor ought to prescribe either X or Y.
3. Therefore, Doctor believes something that is false.
Here, too, the reasoning looks sound. The first premise comes out true, if the complement is relative to Doctor’s information at the time of her utterance, but that value for the information parameter in #2 makes it come out false. #2 comes out true if it is read as relative to Consultant’s information, but that value makes #1 come out false; intuitively, Doctor doesn’t believe that she ought to prescribe either X or Y, relative to that information. And if the first and second premises are relative to Doctor’s and Consultant’s information respectively, then the inference isn’t valid.[19]
Weatherson discusses a related puzzle. The contextualist, he argues, must treat ascriptions involving modals differently from contextualist treatments of other context-sensitive expressions. For other context-sensitive expressions embedded in complements of knowledge and belief ascriptions, the way they function in those complements patterns in the same way. (For example, “I” used in either complement always denotes the speaker.) But, he argues, context-sensitive modals wouldn’t. We can see this by contrasting the difference between our willingness to assert “Jefferson Davis believes that one must not aid fugitive slaves” with our unwillingness to assert “Jefferson Davis knows that one must not aid fugitive slaves”. A contextualist, he argues, must hold that the contextually selected standard in complement of the belief-ascription is Davis’, but, our unwillingness to assert the second means that she must hold that the standard selected in the knowledge-ascription is the speaker’s.[20]
We may again use the Doctor/Consultant example to get a parallel case that targets the value for the information/circumstances parameter. Consultant is perfectly prepared to assert the first premise in the above inference, “Doctor believes that she ought to prescribe either X or Y”. But intuitively, we think he should be unwilling to assert “Doctor knows that she ought to prescribe either X or Y”. This difference between how context selects values in each of the two pairs of cases (apparently selecting the attributee’s standard/information in the first in each pair, the attributor’s in the second) means that, on a contextualist treatment, ‘must’ and ‘ought’ (used normatively) function unlike other context-sensitive terms. While it isn’t impossible that this is so, it would be surprising if it were.
3. A Canonical, Flexible
Contextualist Account of Modal Expressions, its Inspiration and Advantages
3.1 Inspiration
Each of these puzzles arises from particular assumptions about what values for each of the parameters contexts selects in the respective cases. Adequate responses require both the identification of rival parameter values that avoid the puzzling consequences and—importantly--a general story about why context selects those values. Provision of the latter is the provision of a general account of how context selects a restriction for the modal. The account that I favor takes Kaplan’s contextualist account of denotation determination for demonstratives as its inspiration. On that account, the speaker’s intentions select a demonstrative’s denotation in a context. The key is to follow Kaplan’s understanding of the content of a speaker’s intention by distinguishing (in the quantificational case) between the restriction that a speaker’s intention plus context selects and the speaker’s fallible beliefs about which restriction is selected.
This distinction is easily illustrated with an example of a contextually-determined domain restriction with another sort of quantificational expression, that of quantifiers over individuals. Suppose that I have a large lecture course. In conversation at the course’s beginning, I explain my policy against failing students. I say,
(D) “Every student will get a D or better”.
From conversational context, it is clear that ‘every student’ has its domain restricted to the students in my class. On a plausible, Gricean/Kaplanian account, it was my contextually manifestable intentions that did the restricting. But what was the content of my intention such that I managed to do that? Here’s something it wasn’t: my mental enumeration of every individual student in my large course. I may have a view about which individuals are students in my course, but that view is fallible. Before the end of the add/drop period, it’s indeterminate which students will be in my course at its end. My clear intention here isn’t to quantify over the set of students I believe to be students in my course at the time of my utterance. Rather, it’s to quantify over a set of individuals each of whom answers to the description “student in my class” at its end. Context works to manifest this intention to my audience and these together determine the restriction.
3.2 The Account
Modal expressions are quantifiers over possibilities. When a bare modal expression is used, its ‘flavor’, as bouletic, epistemic, legal, or deontic, etc. is determined by a speaker’s referential intentions in a context of use. A referential intention is a speaker’s intention for the addressee to recognize some feature of the context as settling extension in that context. In most uses, the domain of the modal quantifier is restricted and when it is, it is the speaker’s referential intentions that do the restricting. If we take this general account of modals and apply it to BNMs in particular, we will get the following hypotheses:
H1: The proposition expressed by the use of a BNM is determined by either a contextually selected body of information or a set of circumstances and a contextually selected standard.
H2: Which of each is selected is determined by a speaker’s ‘referential’ intentions, i.e. her intentions for an audience to recognize a salient feature of the context as determining a body of information or circumstances and a standard.
H3: Referential intentions need not be explicitly formulated by a speaker to herself at the time of her utterance. Often their contents are discovered empirically, by the speaker’s recognizing a possible parameter value as partly determining the restriction she intends.[21]
H4: There are at least as many different kinds of normative modal propositions as there are publicly manifestable intentions to set each of the relevant parameters.
To this core account, I add the following:
H5: So-called ‘subjective’ “ought” statements are those to which the information/fact parameter is set by a body of information.
H6: So-called ‘objective’ “ought” statements are those to which the information/fact parameter is set by a set of circumstances.
H7: So-called ‘rational’ “ought” statements are those to which the standard parameter is set by the requirements of rationality, if there are any.
H8: So-called ‘moral’ “ought” statements are those to which the standard parameter is set by the absolute, moral standard, if there is one.[22]
H9: For a sincere speaker S’s assertion of a BNM to be linguistically appropriate, S must believe the proposition expressed by her use of her BNM on the H1 and H2 account of which proposition its use expresses. (This is a norm of semantic competence.)
H10: For a sincere speaker S’s assertion of a BNM to be epistemically warranted, S must be justified in believing the proposition expressed by a BNM on the H1 and H2 account of which proposition its use expresses.
H1 expresses the core of the account’s contextualist semantic proposal, H2, its metasemantic core. H3 is an independent hypothesis about how we may identify the contents of a speaker’s intention on an occasion of use, one that, I believe, fits with our ordinary practice of identifying ‘what someone meant by what they said’. H4 I take to be a consequence of the semantic and metasemantic proposals. H5 and H6 represent how the present account captures the common thought that there are both subjective and objective ‘ought’s (without positing ambiguity), and H7 and H8, the idea that claims about the requirements of rationality and morality are distinct.
Finally, the purpose of H9 and H10 is methodological. Semantic proposals are tested by their fit with ordinary speaker’s sense of when uses of the targeted term are appropriate or inappropriate. It is sometimes assumed in the modals literature that, to be plausible, a theory’s predictions must make uses that ordinary speakers’ deem “fine” or “good” come out as warranted.[23] But this requirement is clearly too strong. A descriptive semantic theory aims to assign meanings to expressions on the basis of speakers’ competent uses of those expressions. Unfortunately, ordinary speakers often make claims for which they lack adequate evidence—but the lack of evidence does not thereby undermine their ability to have made those very claims. The stronger requirement has the effect of making it impossible for a speaker to make a claim for which she lacks warrant by throwing out her usage as data for the theory as soon as she tries to. The result is that ordinary speakers often are deprived of the ability to make the very claims they take themselves to. As an empirical claim about the world, this is simply implausible.[24] However, having marked this distinction, I will for the most part aim to hold my theory’s predictions to the stronger standard.
3.3 Advantages:
One clear advantage of the present account is its fit with a simple,
unified account of modal expressions in general. In particular, it promises a unified
explanation of a great variety of uses of ‘must’ and ‘ought’, including, at
least, legal, moral, teleological, rational, bouletic, epistemic, and so-called
‘subjective’ and ‘objective’ ones.[25]
A second advantage is the account’s generality. The hallmark of a contextualist account of a term is the idea that there is a general, context-independent story to be told about how context works to select the values for the parameters that it does. To avoid being ad hoc and to generate predictions, a full defense of contextualism must provide just such a story. Without one, any defense of contextualism risks warranting no more than a new type of ambiguity theory.[26] On the present account, what binds together all of the different, successful uses of our modal expressions is the presence of a speaker’s publicly manifested, domain-restricting intention.
A third advantage is its fit with a plausible account of our other quantificational expressions. It is widely accepted among philosophers of language that quantifiers over individuals are expressions whose domains are contextually restricted.[27] On one plausible account, it is speakers’ referential intentions that, together with context, do the restricting.
Fourth, H8 fits with widely-held views about the contents of our moral concepts and about our ordinary understanding of moral discourse as presupposing the existence of an absolute or categorical moral standard.[28] Fifth, H6 and H8 together capture the sense we have that the ‘ought’s in so-called ‘Jackson cases’ are both fully subjective AND moral. (Jackson cases are characterized by the widespread intuition that the agent in the scenario morally must do something she knows to be suboptimal. For example, in the Doctor case, our intuition is that the Doctor morally must prescribe Z, though she knows it to be suboptimal.)
Finally, a less obvious advantage of the present account is its ability to provide a satisfactory explanation of what’s going on in two apparent puzzles about practical inferences. The first such puzzle has lead some to advocate wide-scope accounts of the logical form of certain normative requirements, while the second has lead others to give up modus ponens for inferences involving practical conditionals. The present account makes available simple solutions to both puzzles that neither depend upon wide-scoping nor on giving up on modus ponens. I’ll discuss this issue in full after showing how flexible contextualism can resolve the two apparent puzzles, discussed above.
4. Flexible Contextualist Solutions
The puzzles for contextualism discussed above arise because the cases seem to involve normative disagreement across contexts. Let’s examine this appearance more closely. One question is about each puzzle’s set-up: Certain assumptions are made about which values the contextualist must hold context selects. But how were these values selected? In order to test the claim that context must select these values, we need a general account of how context selects values for BNMs. The puzzles don’t provide such an account; they begin with an assumption of the values selected.
So let’s back up and ask what the account provided here should say about which values context selects in each of the puzzle cases, beginning with the Doctor example. Here I propose to test my flexible contextualism not only against the original case, but also against a couple other cases for which my solution generates predictions. In the original case, Doctor and Consultant possess different information about the efficacy of the three drugs and about the patient’s medical history. Let t3 (as a first approximation) be the latest time at which Doctor can effectively prescribe a drug.[29] Let t1 be the time at which Doctor asserts (X) and t2 the time Consultant asserts (Z). In asserting each, Doctor and Consultant seem to be disagreeing about which drug Doctor ought to prescribe. This is in part because they each seem to be addressing a common question: What ought Doctor to do? The challenge for the contextualist is to capture both of these intuitions within a framework that treats bodies of information as relevant for determining propositions expressed. Here too the puzzle rests on an assumption about which value a contextualist account holds that context selects, namely, that context selects for each of (X) and (Z) the information the speaker possesses at the time of utterance. If that’s right, then Doctor and Consultant are neither disagreeing, nor addressing a common question.
What should a flexible contextualist say about this case? The contextualist treatment for this case will be more complicated than that for the Davis case, since, unlike in the Davis case, the value for its standard parameter ranks the worlds in the modal’s domain. So, let’s back up a bit first and identify a general recipe for determining the proposition expressed by the comparative use of an ‘ought’ in a context of deliberation on the present account.
Since Doctor is deliberating and using ‘ought’ comparatively, this is the recipe we’ll need to use to determine which propositions (X) and (Z) express. The challenge for the contextualist in this case is to identify the body of information context selects and to provide a generally applicable story of why context selects the information identified. The beginning of wisdom is to note exactly what is puzzling about the case, given the puzzle’s presumed contextualist treatment. On that treatment, a speaker is asking herself “which act performed at t3 is best given what I know now?” Interest in that question would be surprising, because Doctor finds herself in a context of deliberation and, as Finlay and Bjornsson note, deliberators tend to be news-sensitive.[31] As a first approximation, we can think of news-sensitivity dispositionally, as a readiness to revise one’s judgment about what one ought to do in light of new information. Given that deliberators, at least when they are rational, are typically news-sensitive, it would be odd if Doctor were interested in which act would be best at t3, given what she knows at t1, only to abandon her interest in that question, should new information come to light prior to t3. If Doctor is news-sensitive, it is better to understand her as intending to speak (also to a first approximation) to the question “what ought I to do at t3 given the information I’ll have then?”
There is a complication here. How exactly is this informational parameter set so that it gets intuitively the right results? How, for example, do we capture the idea that there may be some information that is relevant to answering a practical question, but that an agent may not have at the time of acting because she is negligent in gathering it? This is genuinely a complication and not a difficulty. It requires correctly identifying what it is for an agent to have a referential intention in the relevant sense. As noted above, a referential intention is not something whose content a speaker must explicitly formulate prior to her utterance. Often, we may discover the contents of our intentions empirically, by noting our reactions to challenges. To see this, consider again the case of quantification over individuals. Suppose I tell you that
(S) “Every student was at the meeting.”
Knowing that some students are away for the semester, you’re surprised and wonder whether we teleconferenced with them. So, you reply,
(R) “What? Even those not is residence?”
to which I reply
(M) “No, I meant every student in residence.”
Here, I clarify for you what restriction I intended in my original assertion. But it is plausible that I didn’t explicitly formulate to myself the restriction “student in residence” prior to my assertion. Your prompt and my reaction help me to learn something about the content of my original, intended restriction.[32]
The case of modal restrictions, I’m suggesting, is precisely parallel. We can discover the content of the restriction we have in mind by noting our reactions to learning of new evidence, for example, Doctor’s reaction to the new information provided by Consultant. The most natural thing for Doctor to reply to Consultant is something like
(Z’) “You’re right; I ought to prescribe Z.”
I’m proposing that the information parameter is set in a context by the speaker’s domain-restricting intentions in that context, where the contents of those intentions are discovered by observing the speaker’s disposition to recognize some piece of information as in or out of her intended restriction.[33] If Doctor responds in the natural way, with (Z’), she is best understood as intending to include at t1 the sort of information that Consultant gives her at t2 in the body of information that restricts the domain of the modal in (X). Given that information, the proposition expressed by (X) is false and Doctor rightly corrects herself in (Z’). If Consultant’s information is unexpected, Doctor is nonetheless warranted in saying what she does with (X), as she is at t1 reasonable to assume that the information she’ll have at t3 is the same as the information she already possesses at t1.
(Notice that Doctor can also revise her view about which time is the time at which she’ll need to act in the same way. She picks out the time as ‘the time at which I’ll need to act’ or ‘the time at which I’ll need to act, in light of the information available’[34] and has a substantive opinion about when that is, which she can revise. To be relevant for determining the denotation of ‘the time at which I’ll need to act’ or ‘the time at which I’ll need to act, in light of the information available’, the information moving the time must come before t3. Whether it is information or circumstances that fixes the relevant time is, of course, determined by context and speaker’s intentions.)
Using the above recipe and assuming that Doctor has also uttered (Z’), here’s how we find the proposition she expresses with (X):
Drug X Drug Y Drug Z Nothing
.5 x Death + .5 x complete cure, .5 x Death + .5 x complete cure, 1.x partial relief, 1. x no change
Using the same procedure, we can see that (Z) and (Z’) are both true. This fits with our intuitions about these utterances, while explaining how it is that (X) and (Z) express rival answers to a common question. That they are rival answers is shown by their truth placing incompatible requirements on the relative rankings, by a single standard, of the actions available to Doctor, where what is ranked is the same set of subsets of possible worlds.
What about a case in which Doctor expects new information from consultant prior to t3? The present proposal predicts that (X) is infelicitous and unwarranted in that case; instead, Doctor should assert,
(J) “I don’t know which drug I ought to prescribe; I need to hear from Consultant first.”
And that fits with our intuitions about the case. If, at t1, Doctor knows that Consultant will provide her with new, relevant information priori to t3, then our intuitions are that asserting (X) is infelicitous and unwarranted and that Doctor should instead assert (K).
The hypothesis that deliberators typically have news-sensitive domain-restricting intentions also suggests that hindsight, third-party evaluators with greater information should not criticize deliberators for choices that were suboptimal, given their enhanced information state, if those choices were best, given the information available to deliberators at the time of action. Is this prediction supported by our intuitions in cases of hindsight evaluation? Consider such a case. Suppose that you are a member of an audience watching a dramatic reenactment of the exchange between Doctor and Consultant, many years after the actual event. Medical knowledge has progressed since the actual exchange and it is now known that drug X is certain to kill patients taking drug W and drug Y certain to cure them. Watching the reenactment, you hear Consultant assert (Z) and then Doctor assert (Z’). What would be an appropriate audience reaction? Here are some options:
(Dp) “I disagree. Doctor ought to have prescribed Y.”
(Fp) “That’s false. Doctor ought to have prescribed Y.”
(Gp) “That’s a shame. Doctor ought to have prescribed Y.”
(Sp) “That’s a shame. It’s too bad Doctor didn’t know that Y would have completely cured Patient. But she did the right thing; she ought to have prescribed Z.”
The responses I’ve gotten to this case from both philosophers and ordinary English-speakers fit a pattern. A strong majority found either Sp to be the best response or tied with Gp. Among the remaining group, there was a majority preference either for Gp or the view that all except Dp were acceptable. No one ranked Dp or Fp as better than both Sp and Gp. An even larger majority found Dp unacceptable and, most of those found Fp also unacceptable.
These responses are just what we would expect in this case were the present hypothesis correct. Dp and Fp are critical responses and, as such, are unwarranted on the present hypothesis. “That’s a shame” is not critical of the deliberators’ choice; it’s an expression of regret. As such, it makes room for a possible context-shift. The BNM in Gp is then heard as making a claim about what choice would have been best, given the information the speaker currently has. But “that’s a shame” needn’t induce such a context shift. Whether the BNMs in Gp and Sp have the same domains as those in Z is a matter of speaker’s intentions. The surrounding material in Sp makes it clear that the domain for the BNM in that utterance is the same as that in Z.
Interestingly, many who stuck up for either Dp or Fp did so by insisting that they are semantically appropriate, though either silly or false. This is also the right thing to say on my account, which, in H9 and H10, distinguishes between norms of assertion that govern linguistic appropriateness and those that govern warrant.[36]
An additional advantage of the present account with respect to this case is its ability to accommodate our substantive, moral intuitions. Our intuitions in the original doctor/consultant case are that Doctor morally ought to prescribe Z and that prescribing either X or Y is morally impermissible. This can initially seem odd, since Doctor knows that prescribing Z is suboptimal. (It’s this feature that makes the present case a so-called ‘Jackson’-case.) What is morally best is prescribing the drug that provides complete relief to Patient without killing her. How can we make sense of the idea that someone can be morally required to do what she knows to be suboptimal?
On the present account, context selects a subjective, moral domain for the ‘ought’s in Z and Z’. In this sense, they are subjective, moral oughts without ‘ought’ being ambiguous. On H4, a subjective ‘ought’ is one whose domain is f-restricted by a body of information. On H7, a moral ‘ought’ is one whose domain is g-restricted by an absolute moral standard. Z and Z’ are intuitively both warranted and this can only be so on the assumption that the domains are restricted by a body of information. If they were restricted by circumstances, then they should seem unwarranted and false. But they don’t. To see that the ‘ought’s are moral, imagine that Doctor most wants to find out which of X or Y would kill the patient. She could then reply to Consultant’s information about the presence of W in Patient’s system with “oh, excellent! Now I can discover which is lethal in combination with W! So, I ought to prescribe either X or Y”. For most of us, our intuitive reaction to this is Doctor ought not prescribe either X or Y because doing so has a 50% chance of killing Patient. The expected outcome of that prescription is not in dispute between Doctor and us. What’s in dispute is what she morally ought to do.
What about the second puzzle about practical inference and attitude attribution? Here too the issue apparently involves disagreement across contexts. Consider first the Jefferson Davis case. What does the present account say about the contribution the complement makes to determining the whole proposition expressed by
(K) Jefferson Davis believes that one must not aid fugitive slaves.
in the first premise of the problematic inference? To answer this, we need to know how the speaker is trying to characterize Davis’ state of mind. For example, is the object of Davis’ belief held to be a proposition about moral norms or about local, prevailing legal norms? One clue to the speaker’s intentions is provided by an examination of her evidence. Suppose her evidence is the text of a speech, relayed in a book on Davis that provides information about the context in which the speech was given. Suppose the speaker’s grounds are Davis’ assertion
(F) “One must not aid fugitive slaves”.
The speaker’s understanding of (F) will itself depend upon clues that she has about Davis’ grounds for making it. Suppose it is in the context of a speech in which Davis rejects the authority of “man’s law” for a “higher” one. Here Davis is best understood as intending to select an absolute moral standard.[37] This intention is not an intention to select whichever standard would make his assertion true. Rather, he is expressing a substantive view about what the standard he intends to select prohibits. Or suppose instead that Davis appeals to the prevailing laws of the Confederacy at the time he asserted (F). Then Davis would be best understood as selecting local, prevailing legal norms to determine a restriction. Which of these is the case will determine which proposition it is best for the speaker to select as the complement of her belief report.
Imagine, as seems plausible, that Davis did not believe merely that aiding fugitive slaves was prohibited by the laws of the Confederacy. Suppose that he believed that it was prohibited by a higher, moral law and that this was suggested by the material in the book about Davis’ speech. In that case, the belief our speaker should attribute to Davis is the belief that, given the circumstances, the moral law prohibits aiding fugitive slaves. That is the very proposition our speaker rejects in the second premise of the problematic inference, namely,
(N) “It is not the case that one must not aid fugitive slaves.”
In using (N), our speaker is not denying that Confederate law prohibited aiding fugitive slaves. She’s rejecting the thought that moral norms generate such a requirement. As with Davis, she does not intend to select whichever standard would make the proposition she expresses with (N) true. She is intending to select an absolute moral standard[38] and express a substantive opinion about what permissions it generates. So, (N) is the denial of the very proposition (J) ascribes to Davis belief in. The conclusion
(C) “Jefferson Davis believes something that is not true”
follows, then, from (J) and (N) as needed.
What we now have is a recipe for a puzzle and a recipe for its solution. To block the solution, the anti-contextualist must find a case that involves 1) clear normative disagreement and 2) a clear difference in contextually-selected standard, according to the present account of how those standards are selected. The basic Jefferson Davis case looked like a good candidate. Jefferson Davis and we seem to differ sufficiently in our moral outlooks that it is plausible to suppose that we occupy relevantly different contexts. Yet there looks to be enough broad overlap in our situations, culturally and politically, that it is plausible to suppose that we are genuinely disagreeing instead of deploying different normative concepts. Nonetheless, the case fails to pose a genuine problem for the flexible contextualism defended here. The challenge for the anti-contextualist is to find a case with the relevant features that is not susceptible to the sort of solution available in the Davis case.
With the solutions to the puzzle about practical advice and the Jefferson Davis inference on the table, it is easy to see what to say about the Doctor/Consultant inference. Here it is again:
1. Doctor believes that she ought to prescribe either X or Y.
2. It is not the case that Doctor ought to prescribe either X or Y.
3. Therefore, Doctor believes something that is false.
For the reasons already given, Doctor’s utterance of (X) is relative to a body of information that includes Consultant’s. But Doctor doesn’t know that she doesn’t have all of the information she intends her ‘ought’ to be relative to. She thinks that she already knows all of the information she intends to include. Since, relative to the incomplete information she has, she ought to prescribe either X or Y, she believes that this is what she ought to do, given the information she intends (X) to be relative to. But, because her assertion IS relative to a larger body of information that includes the consultant’s, both what she says with (X) and the belief (X) expresses are false. Moreover, since deliberators can generally be assumed to be news-sensitive, Consultant is reasonable to attribute this belief to Doctor with #1. Since the second premise here expresses the content of Consultant’s belief, which is relative to the same body of information as Doctor’s, that premise is the negation of the proposition the first premise attributes to the Doctor belief in. The argument, then, is sound.
Weatherson’s observation that our differential willingness to assert “Jefferson Davis believes that one must not aid fugitive slaves” and “Jefferson Davis knows that one must not aid fugitive slaves” is also easily explained within the framework of the contextualist account I’m defending. We need only appeal to a general difference between knowledge-ascriptions and belief-ascriptions, namely, that since “knows” is a factive and “believes” isn’t, the former ascriptions carry the presupposition that the complement is true, while the latter don’t. Given this, the difference can be explained this way: In the case of “Jefferson Davis knows that one must not aid fugitive slaves” we are not willing to presuppose the truth of the complement since that would require a willingness to presuppose that in every morally ideal world that contains slaves, no one is aiding them. That’s a presupposition we’re just not willing to accommodate.
The parallel phenomena in the Doctor/Consultant case is straightforwardly explained in the same way. Since Doctor is in a context in which she is deliberating with Consultant, Consultant has every reason to believe that Doctor is sensitive to the kind of information about Patient’s medical history that Consultant has. So, when Doctor asserts (X), Consultant has reason to believe that the value for the information parameter that Doctor intends includes his information. Given Doctor’s assertion of (X), Consultant may then reasonably assert, “Doctor believes that she ought to prescribe either X or Y” and intend to attribute to Doctor belief in the false proposition that, given the information they have together, Doctor ought to prescribe X or Y. But since Consultant knows that the proposition he’s attributing to Doctor belief in is false, he should be unwilling to assert “Doctor knows that she ought to prescribe either X or Y”.
5.A Final Advantage: A Natural Solution to
Two Puzzles about Practical Inferences
[THIS SECTION UNDER CONSTRUCTION; DO NOT
CITE]
5.1 First Puzzle
Consider two cases involving practical inferences, the first completely mundane, the second, of a kind that is unusual and the subject of much philosophical discussion and controversy.
HARLEM:
R1. You want to go to Harlem.
R2. If you want to go to Harlem, you ought to take the A-train.
R3. Therefore, you ought to take the A-train.[39]
We can easily imagine deliberative contexts in which two speakers’ assertions may be represented by the premises and conclusion in HARLEM. Here’s one: You’re visiting your friend in New York. You tell her that you’d like to visit the famous neighborhood in which Langston Hughes once lived and wrote and where Jacob Lawrence created some of his most famous paintings.
Your friend: “Ah, I see. You want to go to Harlem. Well, if you want to go to Harlem, you ought to take the A-train.”
You take out your subway map and point:
You: “So, I ought to take the A-train?”
Your friend: “Yes, that’s right.”
Notice that in this context, HARLEM seems to be a perfectly good way of representing the speakers’ exchange and the inference seems sound. This is in part because we most naturally hear the conclusion as making a claim about what to do, on the background assumption that you’ve got certain desires. This exactly fits with the prediction given by flexible contextualism, once it’s conjoined with Kratzer’s account of conditionals.[40] On Kratzer’s account, indicative conditionals are implicitly modalized. The propositions they express are restricted to the set of worlds in which their antecedents are true or in which the ‘standards’ expressed in their antecedents are met. The conditional comes out true just in case all the worlds in which the antecedent is true (or…), so is the consequent. This means that indicative conditionals are not material conditionals.[41] If we think of propositions as sets of possible worlds, it is easy to see that material conditionals express different propositions from Kratzer’s indicative conditionals, since they are true at different sets of worlds. The former are true at worlds in which the antecedent is false, while the later are modals that do not include those worlds in their domains of quantification.
We can now put together Kratzer’s account of conditionals together with flexible contextualism about modals to get a prediction about what proposition is expressed by “if you want to go to Harlem, you ought to take the A-train”; it’s true just in case the best worlds in which your desire to go to Harlem is satisfied are worlds in which you get there by taking the A-train.[42] HARLEM is valid, precisely because the context indicates that the restricted domain over which the modal in the second premise quantifies is the same contextually-restricted domain over which the conclusion does. So we get just what we want: An explanation for why, given that the premises are true, the conclusion is as well.
Consider now the difference between HARLEM and another, apparently similar practical inference.
MURDER:
M1. You want to murder messily.
M2. If you want to murder messily, you ought to use a chainsaw.[43]
M3. Therefore, you ought to murder with a chainsaw.
Notice that, unlike the first inference, which is completely ordinary, this case is really weird. It’s a bit difficult to find a natural context for this inference understood as conveying practical advice. Unlike in the Harlem case, most of us would refuse to advise someone who announced their desire to murder messily, even if we were in a position to offer helpful suggestions. This will be important later in explaining our inclination to reject the conclusion of MURDER, but not of HARLEM. First, though, we need to find a natural context for MURDER. Imagine two mafiosos, Jack and Jill, deliberating about how best to prevent snitching. Jill has some experience in deterring snitches. Her advice is to make a vivid example of the local snitch.
“Whatcha wanna do” she says “is kill ‘em real messily.”
Jack asks “what? you mean like with a chainsaw or something?”
To which Jill replies “Yeah. You oughtta kill ‘em with a chainsaw.”
Let MURDER represent their reasoning. John Broome has suggested that we explain our unwillingness to accept the conclusion in inferences like MURDER by giving the modal in its second premise wide-scope and interpreting the conditional as a material conditional.[44] If we do that, the conclusion won’t follow from the truth of the premises. But this claim, that the conditional in the second premise takes wide-scope, is surprising. What is the difference between MURDER and HARLEM? There appears to be nothing more than a difference in desires and means to their satisfaction, which is to say, there appears to be no semantic difference, no formal difference, and hence there should be no difference in the validity between the two.
So, why do we feel strongly inclined to reject the conclusion of MURDER, but to accept that in HARLEM? There seem to be two reasons. First, the contexts in which we most naturally expect to find a sentence like M3 are moral contexts and it is just false that in all morally ideal worlds, Jack is murdering a snitch with a chainsaw. In other words, when we hear an utterance of a sentence like M3, we don’t hear it as expressing the proposition that it in fact expresses in the mafiosos’ context, but hear a different, more familiar proposition that would be clearly false. So we reject M3. If we force ourselves to focus carefully on the context in which the conclusion of MURDER is actually used, we intuitively hear it at least as less bad than when we look at the inference outside of any context.
But this does not guarantee that all of us will hear M3 as fine, even in its context. It seems to me that this is not best explained by the hypothesis that what Jill said in her context is false. An equally plausible explanation of our inclination to reject M3 rests on our commitment not to aide and abett murders. This is manifested in our complete disinclination to advise either mafioso on how best to prevent snitching. In particular, we wouldn’t ourselves use what Jill says in the second premise to advise Jack. Since we certainly wouldn’t provide such advice, we also wouldn’t assert the conclusion of MURDER.[45]
5.2 Invalid, Apparently Valid, Practical
Inferences
James Dreier discusses an interesting, different puzzle about practical inferences, (etc.)[46]
BATTLE:
B1. Either we will win or we will lose.
B2. If we win, it is better to have few soldiers than to have many.[47]
B3. If we lose, it is better to have few soldiers than to have many.
B4. Therefore, it is better to have few.
The inference is clearly invalid. Where does it go wrong? Dreier argues that the mistake is reliance on modus ponens as an inference rule. I’ll argue instead that the difficulty lies in the interaction of certain applications of or-elimination with comparative BNMs. In showing this, I’ll consider several other cases for comparison with BATTLE and argue that my flexible contextualism correctly predicts the validity or invalidity of each. Here’s the first case:
Modus Ponens (MP-) BATTLE:
MP-B1. We will win.
MP-B2. If we win, it will be better to have had few.
MP-B3. Therefore, it will be better to have had few.
This inference is clearly valid.
My flexible contextualism correctly predicts this. On that account, MP-B2 expresses a proposition that is true just in case the winning worlds in which few soldiers fight are ranked more highly overall than the winning worlds in which many fought. (Suppose, following Dreier’s suggestion, we rank winning worlds in terms of the degree of glory realized.) Suppose this is true and that we are in a winning world (i.e. that MP-B1 is true). Then MP-B3 is true; it is better that the actual world is one in which a few fought, i.e. it is among the worlds in which the soldiers have most glory just in case only a few fought. So, my account predicts and explains why MP-BATTLE is valid.
So, why is BATTLE invalid? Notice first that BATTLE contains MP-BATTLE as a subproof, as well as the valid
MP-BATTLE’:
MP-B1’. We will lose.
MP-B2’. If we lose, it will be better to have had few.
MP-B3’. Therefore, it will be better to have had few.
This suggests that the source of the invalidity in BATTLE is not modus ponens. Does this mean that or-elimination is its source or the combination of or-elimination with a BNM? I think something a bit more complicated is going on. To see this, consider another example:
GARDEN PARTY:
GP1. Either the garden party is before luncheon or it is after.
GP2. If it’s before luncheon, it’s better to wear your hat (than not).
GP3. If it’s after luncheon, it’s better to wear your hat (than not).
GP4. Therefore, it’s better to wear your hat (than not).
This argument is clearly valid, as my account also predicts.[48] But what, then, is wrong with BATTLE? Dreier identifies one source of the difficulty early on in his discussion. When he discusses BATTLE with his students in his decision theory course, he says that he tells them that the argument overlooks the difference having few or many can make to the outcome of the battle. This is exactly right. A second feature of the argument is that it may be presumed that both speaker and addressee are not indifferent to which of the two disjuncts is actual. An advantage of my account is that it explains how both features of the argument make it invalid.
The first step in the explanation requires identifying what sort of standard, on the present account, is contextually selected in each of B2-B4. The answer for B2 and B3 are to be found in Dreier’s discussion. The reason winning with a few is better is that there is more glory in winning with a few. The reason losing with a few is better than losing with many is that it is more shameful to lose with many. For each premise, its grounds rest on a presumed interest of your average solider. So, the modals are bouleic. On the present account, B2 is true just in case, among the winning worlds, more of the soldiers’ glory-related desires are satisfied when they fight with a few than are satisfied when they fight with many.[49] B3 is true just in case, among the losing worlds, the worlds in which more of the soldiers’ desires to avoid shame are satisfied when they fight with a few than are satisfied when they fight with many.[50] Suppose that, so read, B2 and B3 are both true. Suppose also that winning or losing exhaust all the possibilities for the actual world’s future. It just doesn’t follow that it is better to have few.
Why not? Because the standards contextually selected in B4 are not the same as those in either B2 or B3. In offering BATTLE, the speaker is aiming to appeal to interests soldiers can be presumed to have. What is more important to a soldier than winning? This means that the most natural way of hearing the conclusion is as selecting a standard that ranks winning above glory or avoiding shame.[51] Heard this way, B4 is true just in case, among the worlds that are ways the actual world could go in the future, more of the worlds in which we win are worlds in which we have a few and there are more worlds with many among those in which we lose. Given that having many makes winning more likely, this is surely false.
In other words, BATTLE is invalid because it involves a context shift that induces a shift in content between premises and conclusion. It’s invalid for the same reasons NOW, below is.
NOW1. It’s raining now.
NOW2. If it’s raining now, then it’s wet now.
NOW3. Therefore, it’s wet now.
where NOW3 is uttered at some time considerably later than NOW1 and NOW2. In other words, the difficulty in BATTLE is the presumption in context of a preference between the two states that are the disjuncts of the first premise, together with a probabilistic connection between the obtaining of the states mentioned in the consequent of each conditional and which of those two disjuncts will be realized. This hypothesis about the source of BATTLE’s invalidity is easily tested as it’s easy to find other invalid inferences with just these features. Consider, for example, this one:
SMOKING:
S1. Either you’ll die early or you won’t.
S2. If you die early, it’s better to have smoked (than not).
S3. If you won’t, it’s better to have smoked (than not).
S4. It’s better to smoke (than not).
The argument is clearly invalid. Why? A natural context in which to find an inference such as SMOKING would be deliberation about whether to continue smoking or give it up. When the preference for smoking is the only preference against which we rank the worlds in which the antecedents are true, then S2 and S3 come out true. But this does not guarantee that a smoker’s preferences in all the worlds (compatible with the circumstances or information) are most satisfied in smoking worlds, since it can’t generally be presumed that a smoker is indifferent between the choices of more and less life.[52]
6. Conclusion
In the philosophy of language and linguistics literature, contextualism about modals, such as “must”, “ought”, and “may”, is the canonical view. It’s great advantage is it’s ability, if correct, to provide a simple and highly unified explanation of a great variety of language use. Some philosophers have recently raised several puzzles for contextualism about ‘ought’. Some (Bjornsson and Finlay and Weatherson) have argued that contextualists are unable to provide a fully satisfying explanation of the interaction of modals and attitude-ascriptions in the context of moral disagreement. Others (Bjornsson and Finlay) have discussed cases in which it seems that the contextualist is unable to capture the relevance of practical advice and deliberation. I hope to have shown that both cases rest on an assumption about which values for the parameters a contextualist must accept are contextually selected, values that my flexible contextualism rightly rejects.
A further advantage of my account is that it provides a non-ad hoc, prediction-generating, explanatory, and general story about how contexts selects intuitively plausible values for both parameters. That account, in relying upon publicly manifestable, referential intentions, fits perfectly with well-worked out, Gricean accounts of language use. The ability of the account to generate predictions and explanations rests on its well-developed procedure for identifying the contents of such intentions, through an identification of the nature of the disposition to use and evaluate utterances that having such an intention confers.[53] A surprising feature of the recent debate about modals, both normative and epistemic, is that, so far, no one has provided such a prediction-generating account of how a modal’s domain gets restricted. Without such an account, it is hard to see how the rivals of the flexible contextualism defended here could fully count as testable, empirical theories.
The account also fits with a plausible, Kaplanian account of how contexts determines domain restrictions in the case of quantification over individuals. In the literature on quantification over individuals, it is agreed on all sides that such quantifiers are context-sensitive.[54] So, it is an important advantage of my account that modals, as quantifier expressions,[55] receive a precisely parallel treatment.
Finally, the account has the advantage of providing satisfactory, solutions to two different puzzles about practical inferences; solutions that require neither giving up on modus ponens nor the ad hoc positing of a non-obvious logical form for some conditionals, but not others (as in HARLEM versus. MURDER).
The weight of these advantages, taken together, strikes me as more than the sum of its parts. The aim of a descriptive semantics for any term is to provide the simplest, most unified, and explanatory theory of how that term and related terms are used. A satisfactory semantics for a term, in turn, must fit with our best theories of language use overall. In relying mostly on familiar tools in the philosopher of language’s toolkit, the present account is guaranteed to fit with well-worked out and widely accepted overall theories of language use. As a contextualist theory, it has the advantage of simplicity, unity, and fit with the canon among linguists on its side. This makes the theory overall more simple and unified than any of its serious rivals, ambiguity or relativist theories. The only reason to give up on such a powerful theory would be its inability to account for some data in the form of puzzle cases. Here I hope to have shown that so far, no such puzzle case has been identified, at least not for BNMs. That, in addition, it provides simple and plausible solutions to outstanding puzzles about practical inferences is icing on the cake that places further burdens on its rivals to provide equally simple and plausible solutions.
References
Barker, Stephen. 2006. “Truth and the Expressing in Expressivism.” In Metaethics After Moore, ed. T. Horgan. Oxford: Clarendon Press, 298-317.
Bianchi, Claudia. 2006. “’Nobody Loves Me’: Quantification and
Context.” Philosophical Studies 130: 377–397.
Bjornsson, Gunnar. and
Stephen Finlay. Ms. “Metaethical
Contextualism Defended”.
Blackburn, Simon. 1984. Spreading the Word. Oxford: Clarendon
Press.
Boisvert, Daniel. 2008. “Expressive-Assertivism.” Pacific
Philosophical Quarterly, 89(2).
Boyd, Richard. 1988.
“How to be a Moral Realist.” In Essays on
Moral Realism, ed. G. Sayre-McCord.
Ithaca: Cornell University, 181-228.
Copp, David. 2001.
“Realist-Expressivism.” Social Philosophy and Policy, 18(2),
1-43.
Darwall, Stephen.
1983. Impartial Reason. Ithaca.
Cornell University Press.
Dowell, J.L. ms. “A Defense of Canonical Contextualism about Epistemic Modals.” (http://www.unl.edu/philosop/people/faculty/dowell/dowell.shtml )
Dowell, J. L. 2008. “Empirical Metaphysics: the Role of
Intuitions about Possible Cases in Philosophy.” Philosophical Studies,
BSPC 2007 volume, Volume 140, Issue 1, pp.19-46.
Dreier, James. 2009. “Practical Conditionals” in Reasons for Action, eds. D. Sobel and S.
Wall. Cambridge. Cambridge University
Press.
Dreier, James. “Moral Relativism and Moral Nihilism” Ethics Handbook
Dreier, James. 1990. “Internalism and Speaker Relativism.” Ethics 101(1), 6-25.
Finlay, Stephen. 2008. “The Error in Error Theory.” Australasian Journal of Philosophy 86(3), 347-369.
Finlay, Stephen. 2005. “Value and Implicature.” Philosopher’s Imprint 5(4), 1-20.
Finlay, Stephen. 2004. “The Conversational Practicality of Value Judgment.” The Journal of Ethics 8(3), 205-223.
von Fintel,
Kai & Anthony S. Gillies.
Forthcoming. “Might Made Right”
(Ms. April 2008, URL http://mit.edu/fintel/fintel-gillies-2008-mmr.pdf
). In Epistemic Modality, ed.s B. Weatherson and A. Egan. Oxford. Oxford University Press.
von Fintel, Kai & Anthony S.
Gillies. 2008. “CIA leaks.” Philosophical
Review 117: 77-98.
von Fintel, Kai and
Sabine Iatridou. Ms. “What to Do If You
Want to Go to Harlem: Anankastic Conditionals and Related Matters.” (http://mit.edu/fintel/www/harlem-rutgers.pdf
)
Garson, James. 2009.
“Modal Logic.” In the Stanford
Encyclopedia of Philosophy, Fall 2009 edition, ed. Edward Zalta, http://plato.stanford.edu/archives/fall2009/entries/logic-modal/.
Gibbard, Allan.
2003. Thinking How to Live.
Cambridge: Harvard University Press.
Gibbard, Allan.
1990. Wise Choices, Apt Feelings.
Cambridge: Harvard University Press.
Hare, Richard. 1991. The
Language of Morals. New York: Oxford
University Press.
Horgan, Terry and
Mark Timmons. 2000. “Nondescriptivist Cognitivism.” Philosophical Papers, 29(2), 121-153.
Jackson, Frank. 1991. “Decision-Theoretic
Consequentialism and the Nearest and Dearest Objection.” Ethics.
Jeffrey, Richard.
1983. The Logic of Decision. Chicago: Chicago University Press.
Joyce, Richard. 2001. The Myth of Morality. Cambridge: Cambridge University Press.
Kaplan, David. 1989a. “Demonstratives. An Essay on the Semantics, Logic,
Metaphysics, and Epistemology of Demonstratives and Other
Indexicals.”
In Themes from Kaplan, ed. J.
Almog, J. Perry, and H. Wettstein, 481–563.
Oxford: Oxford University Press.
Kaplan, David. 1989b. “Afterthoughts.” In Themes from Kaplan, ed. J. Almog,
J. Perry, and H. Wettstein, 565–614. Oxford: Oxford University Press.
Kolodny, N. and John
MacFarlane. Ms. “Ifs and Oughts.” (http://johnmacfarlane.net/ifs-and-oughts.pdf
)
Kratzer, Angelika. 1977. “What Must and
Can must and can mean.” Linguistics and
Philosophy 1: 337-355.
Kratzer, Angelika. 1981. “The notional
category of modality.” In Words, Worlds, and Contexts: New Approaches
in Word Semantics, ed. H. J. Eikmeyer and H. Rieser, 38-74. Berlin: de Gruyter.
Kratzer, Angelika. 1986. “Conditionals.” Chicago Linguistics Society 22:
1-15.
Kratzer, Angelika. 1991. “Modality.” In
Semantics: An International Handbook of
Contemporary Research, ed. A. von Stechow and D. Wunderlich, 639-650. Berlin: de Gruyter.
MacFarlane, John. Forthcoming.
“Epistemic modals are assessment-sensitive” (Ms of May 9, 2008, URL http://johnmacfarlane.net/epistmod.pdf.). In Epistemic Modality, ed. B. Weatherson and A. Egan.
Oxford. Oxford University Press.
Railton, Peter. 1989.
“Naturalism and Prescriptivity.“ Social
Philosophy and Policy 7(1), 151-174.
Ridge, Michael. 2007.
“Ecumenical Expressivism.“ In Oxford
Studies in Metaethics, Volume 2, ed. R. Shaffer-Landau. Oxford: Oxford University Press.
Schroeder, Mark.
2009. “Hybrid Expressivism.“ Ethics
119(2), 257-309.
Schroeder, Mark. Ms. “Do ought’s take propositions?“
Sloman,
Stanley, Jason and Szabo, Zoltan. 2000. “On Quantifier Domain
Restriction.”
Mind
and Language 15(2), 219–261.
Stevenson, Charles. 1948. “Meaning: Descriptive and Emotive.” Philosophical Review 57(2), 127-144.
Swanson,
Eric 2006. Interactions with Context. Ph.D dissertation, MIT.
Weatherson,
Brian 2008. “Attitudes and Relativism.” Philosophical Perspectives, 22
(2008): 527-544.
[1] Many thanks to Campbell Brown, Stephen Darwall, James Dreier and David Sobel for comments on earlier drafts and to John Broome and Stephen Finlay for discussion. Thanks also to audiences at Emotions, Evolution, and Metaethics Conference at the University of Sydney and at ANU.
[2] Boyd [1988], Railton [1989.
[3] Stevenson [1948].
[4] Gibbard [1990] and [2003], Blackburn [1984].
[5] Hare [1991].
[6] Finlay [2004] and [2005], Boisvert [2008], Copp [2001], Ridge [2007], Horgan and Timmons [2000], and Barker [2006].
[7] Kolodny and MacFarlane [ms] and Finlay and Bjornsson [ms] are rare exceptions.
[8]Garson [2009] p.1. See also Kratzer [1991].
[9] A comparative modal is so-called because it requires the comparison or ranking of worlds or sets of worlds. A clear example of a normative comparative expression is “….is better than…”.
[10] This summary of the canonical view owes much to the clear and concise presentation in von Fintel and Gillies forthcoming.
[11] von Fintel and Gillies [2008].
[12] See Finlay and Bjornsson [ms] and Kolodny and MacFarlane [ms].
[13] Where a ranking of the f-worlds is required, the selected standard orders them as more to less ideal. Here I use ‘standard’ in a catholic sense to include restrictions or rankings set by the satisfaction of some set of preferences (These are bouletic ‘ought’s. )
[14] See Kratzer [1991] and Kratzer [forthcoming].
[15] See, for example, Gibbard [1990] and Schroeder [ms].
[16] This case is a version of the case that Jackson made famous in his [1991].
[17] Bjornsson and Finlay discuss a structurally similar puzzle in their [ms]. There they offer a different, contextualist solution to the one I will defend below.
[18] [Ms]. Their example is inspired by Schroeder’s [2009] discussion. Here I replace their sample belief-ascription with one from Weatherson [2008].
[19] Thanks to Jonathan Schaffer for discussion here.
[20] See Weatherson [2008]. Weatherson’s original example uses ‘ought’, not ‘must’, but to keep clear of the unnecessary complications that arise with ordering sources on the contextualist account, I’ve here substituted the latter for the former. This substitution is harmless, though, since whatever force Weatherson’s argument has isn’t diminished by it.
[21] For an elaboration and defense of this way of understanding referential intentions, see Dowell [2008].
[22] For some explanation and defense of H8, see footnotes 37 and 38. Notice that H8 means that the present account isn’t contextualist or relativist in the sense some have used to characterize metaethical contextualism or relativism, i.e. the view that modals used morally are themselves context-sensitive, selecting different moral standards in different contexts.
[23] See MacFarlane [forthcoming] and von Fintel and Gillies, [forthcoming], for two places in the literature where this assumption is implicitly relied upon.
[24] For more on the distinction between norms of semantic competence and those of warrant, as well as their appropriate methodological role, see Dowell [ms].
[25] For an application of the present contextualist account to epistemic modals, see Dowell [ms].
[26] This is a drawback of Bjornsson and Finlay’s rival contextualist account [ms], which provides no prediction-generating general account of how context selects values for a modal’s parameters.
[27] Stanley and Szabo [2000], Bianchi [2006].
[28] See footnotes 37 and 38 for an explanation and defense of this claim.
[29] This characterization of t3 will need to be revised later to handle certain complications, but it will be simpler to ignore them initially.
[30] What about cases in which there is a slightly favored, but no clearly favored action? My theory predicts that our intuitions about the truth-value of “[Deliberator] ought to O” will be correspondingly weaker and this is just what we seem to get in such cases. (This recipe is drawn, in part, from Kratzer’s [1991] discussion, Sloman [xxxx], and, of course, Jeffrey [1983].)
[31] Their usage of that phrase differs from mine. For discussion, see their [ms].
[32] For a further defense of this account of referential intentions and the discovery of their contents, see Dowell [2008].
[33] For more on the relationship between referential intentions and dispositions, see Dowell [2008].
[34] ‘Availability’, on the present proposal, is also to be cashed out in terms of intentions, where the contents of those intentions will at least often be discovered by a speaker’s recognition of some piece of information as ‘available’ in her intended sense. This is an advantage of the present proposal over the Bjornsson and Finlay contextualist proposal, which relies on an unanalyzed notion of ‘available information’. For a discussion of the challenges in spelling out the notion of ‘availability’ or ‘epistemic reach’ without the reliance on intentions and dispositions, see von Fintel and Gillies [2008] and MacFarlane [forthcoming].
[35] For an explanation of the present notion of ‘availability’, see footnote 35.
[36] One possible, rival view, solipsistic relativism, incorrectly predicts that Dp is not only appropriate, but also warranted and true. So, it conflicts with this response pattern.
[37] Here I rely on the default view in metaethics that ordinary moral discourse presupposes the existence of a moral standard that is absolute or categorical in the sense of generating genuine requirements independently of the contents of anyone’s beliefs about which actions they require, permit, or prohibit, or of the inclinations of those bound by them. For a defense of the default view, see Joyce [2001].
[38] This solution depends upon the claim that ordinary speakers, when making moral claims using “must” intend a modal restriction that selects absolutely morally ideal. Of course, there may not be any such worlds (because there isn’t an absolute moral standard, for example) in which case, those intentions fail to determine propositions. The present view presupposes realism about the semantics of ordinary moral discourse, not about the metaphysics, so it is compatible with a kind of error theory. That said, the present solution does not wholly avoid controversy. Some have rejected the claim that ordinary speakers intend for their moral claims to be measured against an absolute standard. (See, for example, Finlay [2008].) I take it to be a drawback of such views that, rather than attributing error at the level of the truth of what ordinary folks say when they use simple, declarative sentences containing moral vocabulary, they are forced instead to attribute error at the level of their understanding of what they’ve said. That, it seems to me, is a less plausible error to attribute to the folk.
[39] The example is adapted from one in von Fintel and Iatridou [ms].
[40] Kratzer 1991b (“Conditionals”)
[41] This is Kratzer’s conclusion. Here I leave open the possibility that “if…then….” in English has a use as a material conditional.
[42] This is a bit rough. For a more careful discussion of the interaction between modals and conditionals, see von Fintel and Iatridou [ms].
[43] The example is of course similar to Darwall’s in his (1983).
[44] See Broome, “Reasons”, etc.
[45] Following Darwall (1983), we might instead give the advice “if you want to kill snitches, you ought to see a psychiatrist”.
[46] Dreier [2009].
[47] In Dreier’s example, the second and third premises have the consequent “it is better to have few”. In order to bring out the comparative nature of these modals, however, I’ve chosen to put them in their more explicit form.
[48] GP2 is true, on my account, just in case, of the worlds in which the garden party is before luncheon, the worlds in which you wear your hat are better than the ones in which you don’t. GP3 is true just in case the same is true of the worlds in which the garden party is after luncheon. GP1 is true just in case the garden party’s being before or after luncheon exhaust all the options. GP4 is true just in case, in all the worlds in the context set, its better for you to wear your hat than not. This is clearly true whenever the premises are. So, the argument is valid.
[49] An alternative is that the worlds are straightforwardly ranked in terms of glory achieved rather than the satisfaction of a desire for glory. Nothing in my argument hangs on this difference.
[50] Alternatively, we could give B3’s truth conditions in terms of avoidance of the shameful without effect on my argument.
[51] This claim seems plausible to me, but notice that my argument doesn’t depend upon anything so strong. I need only that it is not obvious that the conclusion should be read as having the same standard as either premise. Indeed, to make the argument valid, we’d also need both B2 and B3 to select the same standard.
[52] Having said this, I do think that there are contexts, unusual ones, in which a smoker is indifferent to the difference between more and less life, in which the argument is valid. But this also is predicted by my account.
[53] The account of referential intentions and the discovery of their contents is only sketched here. For further development and discussion, see Dowell [2008] and Dowell, ms.
[54] See, for example, Bianchi and Stanley and Szabo.
[55] Although there is much dispute about how truth-values for modals are determined, it is almost universally accepted among linguists and philosophers of language that modal expressions are quantifiers over possibilities.