If you ought to do something, does it follow that you are able to do it? The Kantian thesis that ought-implies-can seems intuitive and is widely accepted.1 It is neatly expressed in the Latin motto: Ad impossibilia nemo tenetur: No-one is bound to do the impossible. Nevertheless, there are several powerful purported counterexamples. In this paper I will apply an independently motivated contextualism about ‘ought’ to the debate. The idea is that, as ‘ought’ has different meanings depending on the context in which it is uttered, there are contexts where ‘ought-implies-can’ is true and contexts where it is false. Specifically, we can distinguish an ideal sense of ‘ought’, a deliberative (or guidance) sense of ‘ought’, and a hypological (involving praise or blame) sense of ‘ought’. I will argue that ought-implies-can is true in the deliberative sense, false in the ideal sense, and false in the hypological sense unless the agent is blameworthy for the inability.

It is surprising that despite the large literature on ought-implies-can and the large literature on the context-sensitivity of normative terms there has been almost no discussion of how they might interact. The only paper I know of which develops the connection is Berg (2018). However, Berg doesn’t explain how contextualism interacts with the arguments for and against ought-implies-can. I will defend a theory which makes the connections explicit, and diverge from Berg on the treatment of blame.

In §2 I review the main arguments for ought-implies-can, and in §3 the main arguments against. In §4 I explain the contextualist framework and three hypotheses for how the context determines the meaning of ‘ought’. In §5 I apply the contextualist framework to the arguments of §§ 2 and 3, and show how our intuitions are predicted by the three hypotheses. In § 6 I discuss the problem of disagreement for contextualism, in §7 some dialectical consequences, and in §8 I compare my account to alternative ways of weakening ought-implies-can.

Let me make five quick clarifications. First, I will follow the literature in assuming that ‘ought to’ and ‘obligated to’ are synonymous.

Second, I will use contextualism for concreteness, but there are alternatives to contextualism regarding the linguistic implementation of the semantic variability.2 Arguments that contextualism is the correct linguistic theory are to be found in the linguistics literature and I have nothing to add to them. The power of contextualism lies in its ability to explain our competing intuitions. I will argue that contextualism explains our intuitions, and I take this to be a powerful indirect argument in favour of it.

Third, I distinguish the view that ‘ought’ is context-sensitive from the view that ‘can’ is context-sensitive, and focus on the former. While I agree that ‘can’ is context-sensitive, and that this is relevant to whether ought-implies-can,3 I think that the context-sensitivity of ‘ought’ is crucial and under-discussed, hence the focus of this paper.

Fourth, many non-contextualists agree that ought-implies-can is ambiguous, and comes in different degrees of logical strength.4 The logically stronger readings can be used to explain cases where ought-implies-can seems false and logically weaker readings can be used to explain cases where ought-implies-can seems true. On my view, the context explains these intuitions, so my use of ought-implies-can can be understood in the logically stronger sense. I hold that in contexts where it is true that ought-implies-can, a logically strong reading is still true (and in contexts where it is false that ought-implies-can, a logically strong reading will be all the more obviously false).5

Finally, many of my uses of ‘ought’ are mentions rather than uses, and really belong in inverted commas. For the sake of readability, I will often omit the inverted comments (see Lewis 1996: 566–7 for the same strategy).

In this section I will briefly explain two arguments for ought-implies-can.6

In this section I briefly explain four arguments against ought-implies-can.

The first objection is a direct response to the argument from excuses. Recall the argument from excuses said that inability to perform an action provides an excuse. The objection is that this makes excuses too easy to come by. Specifically, the argument from excuses suggests that agents are excused by making themselves unable to perform the action. Peter Vranas, although a proponent of ought-implies-can, opens his paper with the following dialogue:

We are invited to think that the student cannot get off the hook so easily, that inability does not provide an excuse, and that ought does not imply can. Deniers of ought-implies-can hold that the worlds the agent ought to make true are not a sub-set of the worlds the agent can make true.

Richard Feldman introduces the concept of role-oughts:

It seems that agents with these roles ought to perform certain actions even when unable to do so.

Another class of cases in which obligations seem to survive the inability to perform the action involve compulsive behaviours. Vranas writes:9

A related style of counter-example involves feelings rather than actions. It can be natural to say that someone ought to feel a certain way, even if they are unable to do so. Vranas writes:

These four arguments suggest that there can be actions which an agent ought to perform but cannot. Again, I do not intend to provide an exhaustive list of arguments against ought-implies-can. But I think these arguments cover the main points and will suffice to show how the contextualist approach works.10

In this section I explain the core ideas of normative contextualism and state three hypotheses. It is a familiar thought that whether someone is correctly described as tall depends on the details of the conversation. For example, Michael Jordan, at 1.98m, is tall for an ordinary person, but not tall for a basketball player. So, the truth of ‘Michael Jordan is tall’ depends on the conversational context. It is true given a context in which ordinary people are the being discussed, but false given a context in which basketball players are being discussed.

A popular theory in linguistics is that normative terms like ‘ought’ are context-sensitive in a similar way.11 Applied to ought-implies-can we get:

The traditional Kratzerian semantics posits two parameters – a set of live possible worlds,12 and an ordering of how good they are. ‘S ought to Φ’ means that in the best live possible worlds, S Φ’s.13 And ought-implies-can means that the worlds the agent ought to make true are a sub-set of the worlds the agent can make true:

Putting these together:

To give this theory substance we need to say something about the contexts in which ought implies can, and the contexts in which it does not.

Starting with the first parameter, what determines the live possible worlds? I suggest that part of what determines the live possible worlds is the aim of the conversation.14 We can distinguish (at least) three aims people might have when using ‘ought’15:

To see how (i) and (ii) can come apart, consider the student who, at 8:57am, has not written their paper, which is due at 9am. The Professor, knowing this, might say

We can understand the Professor’s utterance as (i) expressing an ideal. In this sense, the utterance is true and relevant. However, it makes no sense to understand the utterance as (ii) offering guidance, given that the Professor knows that the paper cannot be written in 3 minutes.

To see how (i) and (iii) can come apart, suppose that the student has no paper because they were in a coma for the last month, through no fault of their own. Knowing this, the Professor might reasonably make the same utterance to (i) express an ideal, but not to (iii) ascribe blame.

To see how (ii) and (iii) can come apart, notice that the Professor might reasonably make the utterance to (iii) ascribe blame without intending to (ii) offer guidance, e.g. in Late Paper, where the Professor knows the student went to the movies.

Let’s make these aims explicit using ‘ideal-ought’, ‘deliberative-ought’, and ‘hypological-ought’. I suggest that which ‘ought’ is used partially determines which possibilities are live. Ideal-ought is associated with the most expansive set, deliberative-ought is associated with the least expansive set, and hypological-ought is associated with an intermediate set. Specifically, I offer three hypotheses which I will defend in the next section by showing how they predict the intuitive verdicts in all the examples:

The main options for the second parameter, the ordering of possible worlds, are either moral, practical, or epistemic standards.18 For example, a moral standard will rank worlds in order of how morally good they are, while a practical standard might rank worlds in order of how good they are relative to the interests of some agent. I mention this for completeness, but I won’t make use of any shifts to this parameter. I think the conflicting intuitions about cases can be explained purely by the possible worlds parameter.

I’ll now show how contextualism plus the three hypotheses predicts all the opposing intuitions.

A familiar objection is that contextualism wrongly predicts that there is no genuine disagreement. For example, according to contextualism the Professor and the student are using different parameters. Once the parameters are made explicit, each would accept what the other is saying. Disagreement disappears. And that looks like the wrong verdict, as there seems to be genuine disagreement. There are four well-known responses and I think there is some truth in all four, so I will briefly run through them. (I won’t take a line on which response best applies to which cases).

A first response is that in many cases there isn’t any disagreement. It is common for people to appear to disagree, only to find out that they have been using language slightly differently. Chalmers (2011) argues that this is typical even in philosophical debates. And if it happens in philosophical debates, it is all the more plausible that it happens in everyday conversations where speakers are far less reflective about the concepts they use.

A second response is that there is disagreement about preferences over what to do (Björnsson & Finlay 2010; Finlay 2014: ch. 8). The hidden goal of the conversation is to establish what to do. The Professor’s preference is that students turn in their papers on time, and, failing that, that a just outcome is reached. The student’s preference is that they get the highest mark with the minimum effort (we can assume). Thus there is no disagreement over which propositions are true, but disagreement over what to do.22

A third response is that disagreement involving contextualist terms should be understood as disagreement about which parameters to use. To motivate this view, note that contextualism says that the parameters are determined by the conversational context, and which parameters are operative might be in flux in a conversation. For example, consider the following conversation when all parties know that Michael Jordan is 1.98m tall and Sun Mingming is 2.45m tall:

It is plausible that this conversation is best understood as an implicit negotiation about the standards for the word ‘tall.’

Lewis (1979; 1996) suggests that conversations should conform to the principle of accommodation, which says that when a speaker makes an utterance involving a context-sensitive term, the parameters shift to make the sentence true. In Lewis’s terms, the utterance changes the “conversational score.” Thus B is attempting to shift the conversational score to one in which the relevant sense of “tall” is “tall-for-a-basketball player.”

But what if the hearer does not want to change the conversational score? Then we get what Plunkett and Sundell (2013) call metalinguistic negotiation, in which, to continue the metaphor, each person tries to get their favoured values of parameters onto the scoresheet. Thus the apparent disagreement is really an instance of metalinguistic negotiation.23 There is no guarantee that there will be a successful conclusion to the metalinguistic negotiation, resulting in an unresolved disagreement. This fits nicely with the sense of unresolved disagreement in the Professor and student case. The Professor is attempting to put a hypological context on the scoreboard, the student is attempting to put a deliberative context on the scoreboard, and neither is giving way to the other.

Someone might object that understanding a disagreement as a metalinguistic negotiation trivializes the disagreement, making it about parameters and concepts rather than about important matters like what we ought to do. But there is nothing trivial about this disagreement. As Bolinger writes:

More concretely, if you are being blamed and facing punishment for wrongdoing, then the question of which possibilities are live is anything but trivial.24

A fourth response is that there is a privileged ‘ought’, or an ‘ought’ with authority. The contextualist could add that speakers try to co-ordinate on the privileged ‘ought’ i.e. the correct values of parameters.25 Even if speakers agree that the privileged ought is relevant, there can still be disagreement about which ‘ought’ that is. That is, there can still be disagreement about which are the correct parameters. In fact all that’s needed for genuine disagreement is that the speakers have differing beliefs about which ‘ought’ is privileged (even if none of them are actually privileged).

To be clear, it’s not plausible that there is a privileged value of the aim parameter (express ideals, deliberate/guide, praise/blame) or the standard parameter (e.g. moral standard, practical standard). Those are up to the participants. But it is plausible that there is a privileged or correct set of live possible worlds given an aim and standard. For example, given a conversation about blame and the moral standard, it plausibly follows that some particular action (e.g. turning the paper in on time) is in the privileged or correct set of live possible worlds. So there is room for disagreement, and indeed for one party being right and another wrong, when it comes to the set of live possible worlds. This predicts that, once it is established that the conversation is about blame, the student cannot insist that there is no live possibility in which they hand in the paper.

I shall argue that the foregoing entails that ought-implies-can should not be used the way it often is in the literature.

Let’s take a quick look at three influential examples. Howard-Snyder (1997) moves from ought-implies-can to the rejection of objective consequentialism.26 The idea is that objective consequentialism requires us to perform the action with the best consequences, we often do not know what the consequences of our actions would be, so we cannot perform the action with best consequences. Assuming ought-implies-can, it follows that it’s not the case that we ought to perform the action with best consequences.

Widerker (1991) argues that ought-implies-can, plus causal determinism, leads to the rejection of deontic claims such as “John’s action was wrong.”27 The idea is that causal determinism entails that John could not have done anything else. As John could not have done anything else, ought-implies-can entails that it is not the case that he ought to have done something else. And if it is not the case that he ought to have done something else, then John’s action is not wrong (nor is it right).

Alston (1988) argues from ought-implies-can to the rejection of deontic claims about beliefs such as “Bill’s belief is unjustified.”28 The idea is that we do not have voluntary control over our beliefs, so John could not have believed something else. As Bill could not have believed something else, ought-implies-can entails that it is not the case that he ought to have believed something else. And if it is not the case that he ought to have believed something else then John’s belief is not unjustified (nor is it justified).

These authors all identify a sense in which something cannot be done, then invoke ought-implies-can to infer that it is not the case that it ought to be done. But the context-sensitivity of ‘ought’ prevents such a quick inference.

In response to Howard-Snyder (1997), the fact that we don’t know which action has the best consequences does not prevent objective consequentialism from being a theory of the ideal-ought. In response to Widerker (1991) the fact that John could not have done something else is compatible with that fact that he ideally-ought to have done something else. And in response to Alston (1988) the fact that Bill could not have believed anything else is compatible with that fact that he ideally-ought to have believed something else. The authors would need to argue that ought-implies-can is true in all relevant contexts. As utterances of the sentences the authors discuss could occur in many different contexts, it seems that no general argument could be given.

I do not say that no ought-implies-can principle is usable. But attention must be paid to the context to make sure there is no illicit shift of the values of parameters.

I have defended a logical weakening of ought-implies-can: in some (not all) contexts, ought-implies-can. It might be useful to consider two other ways of weakening ought-implies-can.

Bassford (2022: 781) suggests making the consequent of ought-implies-can logically weaker:

This does a nice job of dealing with cases like Late Paper. The student could have written the paper, so we retain the intuitive verdict that they ought to have done so.

But it cannot handle the arguments from role-oughts, addictive behaviours or fixed feelings. In cases where the agent cannot, nor could have, performed the action, it would follow that there is no obligation. But it seems that there is.

King (2019: 71) weakens ought-implies-can by making the antecedent logically stronger:

Very roughly, the natural oughts are those we have just by being human, while non-natural oughts are those we can only acquire voluntarily, e.g. by making a promise. King suggests that only the latter kind of obligations survive one’s inability to perform the action.

But both principles are questionable. The student’s obligation to hand in a paper is non-natural, but seems to survive his inability to do it. So that seems to be a case in which non-natural ought implies can. For a counterexample to natural-ought-implies-can, consider Aurelia’s utterance “I ought to save him but can’t get past the sharks.” Suppose Aurelia has not willingly acquired any relevant obligations, e.g. she is not a lifeguard, she is just a passer-by. So we have a natural ought. Nevertheless, we can understand the utterance in a way that makes it true. Thus we seem to have a counter-example to natural-ought-implies-can.

There are powerful arguments both for and against the thesis that ought-implies-can. Rather than trying to explain away any intuitions it is better to make sense of the debate by explaining how opposing intuitions can be correct. One advantage of contextualism is that it does exactly that. Another advantage is that contextualism is independently motivated by work in linguistics. I have argued that contextualism can explain the intuition that ought-implies-can, while also predicting that in many contexts it is false that ought-implies-can. Specifically, I have defended three hypotheses:

The main upshot is that philosophers cannot appeal to ought-implies-can in their arguments without paying attention to the context. However, it does follow that in deliberative contexts philosophers can use ought-implies-can, and that inability excuses (unless the agent is blameworthy for the inability).