Cappelen and Dever challenge the widely accepted idea that some key aspect of intentional action is essentially indexical. They argue that the classical arguments for this coming from Perry are in fact arguments for a different phenomenon: the opacity of explanatory contexts. I agree with Cappelen and Dever that what Perry says about the ineliminability of indexical terms from explanations of intentional action fails to amount to an argument for this indexicality being essential. But this should not lead us to be sceptics. In this paper, I present a different argument for the essential indexicality of intentional actions. The key premise of this argument is that intentions themselves are essentially indexical. I provide evidence for this premise and defend it against potential criticism. I also show how the essential indexicality of intentions can be used to vindicate Perry's original claims about the essential indexicality of certain beliefs and desires.


Consider a case. Suppose Jack trips and ends up spilling coffee on Jill. Jill briefly pauses to take stock of what happened, and after a moment's reflection chooses to respond by spilling coffee on Jack. In such a case, we say Jack's coffee-spilling action was unintentional, while Jill's was intentional. Interestingly, we can explain Jack's action in a purely non-indexical way, that is, without (explicitly or implicitly) having to make use of indexical expressions like ‘I’, ‘you’, and ‘himself’: Jack's action of spilling coffee on Jill was due to Jack tripping and Jill being in the wrong place at the wrong time. But it is more difficult to explain Jill's action non-indexically. We might say it was the result of Jill deciding that she will spill coffee on Jack. This appears to be a non-indexical explanation of Jill's action. But on closer inspection this may not be so. Concealed in the explanation seems to be a suppressed quasi-indicator (to use Castañeda's terminology1). What we really are saying is Jill's action resulted from her deciding she herself will spill coffee on Jack. Thus, the detailed explanation of Jill's action is indexical after all.

Since there is nothing odd about this case, a natural thesis to extract from it is that a distinguishing feature of intentional actions is they are essentially indexical. Something about intentional actions makes them fundamentally indexical, and consequently any account of these actions needs to recognize this indexicality. Call this the Essential Indexicality Thesis.

Essential Indexicality Thesis Intentional actions are essentially indexical, that is, an action is intentional only if something about it is ineliminably indexical.

This thesis is not new. Many have endorsed it in one way or another. Notable amongst this group are Hector-Neri Castañeda and John Perry, but there are others as well.2

But not everyone is sold on the Essential Indexicality Thesis. In particular, Cappelen and Dever (2013) have recently challenged it.3 They argue that the classical arguments in Perry (1979) for the thesis are in fact arguments for a different phenomenon: the opacity of action explanation contexts. Their claim is that the ineliminability of indexicals from action explanations, like the one we gave above about Jill's coffee-spilling action, is merely a run-of-the-mill Frege Puzzle, so our drawing the conclusion that this action is itself in some way essentially indexical is unwarranted. Though this does not prove the thesis is false, it does undercut the reasons most have for accepting it. Cappelen and Dever thus conclude that we should be sceptics.

I agree with Cappelen and Dever (hereafter C&D) that what Perry says about the ineliminability of indexical terms from explanations of certain actions does not amount to an argument for the Essential Indexicality Thesis. But I do not think this forces us to be sceptics. I think there is a better argument for the thesis, one which is often overlooked. The argument in question comes from the work of Hector-Neri Castañeda, from which Perry takes his cue. According to Castañeda, intentions are essentially first-personal and indexical—there is no such thing as an intention that is purely third-personal and non-indexical, though intentions may be reported with third-personal terms (‘Jill intends that she spill coffee on Jack’).4 If this is right, we can resist C&D's scepticism.

My overall aim in this paper is to use Castañeda's insights about intentions to defend the Essential Indexicality Thesis. But I also have a secondary aim of equal importance. That aim is to make clear in what way intentional actions are essentially indexical. C&D seem to think it solely involves the beliefs and desires we have and which enter into the process of deciding what to do. This, I argue, is incorrect. Rather, the essential indexicality of intentional action has everything to do with the nature of intentions. If some beliefs and desires are also essentially indexical, it is only because they have to be if they are to interact with our intentions in the right way to produce actions deserving the name intentional.

In the next section, I lay out C&D's argument against Perry. In Sections III and IV, I discuss strategies for responding, and make some remarks about the terms of the debate. Then, in Sections V–VII, I present the alternative argument for the Essential Indexicality Thesis, which I call the Argument from Intentions. Section VIII responds to an objection and fills in some details. And in Section IX, I show how the Argument from Intentions can be used to vindicate Perry's original claims about the essential indexicality of certain beliefs and desires.


In his ‘The Problem of the Essential Indexical’, Perry presents a series of examples purporting to show that indexicality is essential to some of our thoughts and behaviours. Perhaps his most famous example, the one we will focus on, is the Messy Shopper:

Messy Shopper

Someone in the supermarket has a broken sack of sugar and is walking around making a mess. So I give chase, following the trail of sugar in order to tell this person they are making a mess. But no matter how quickly I run around the aisles of the market, I can't seem to catch up to the messy shopper. After awhile it dawns on me: the reason I can't catch up to the messy shopper, the reason I keep running around in a circle, is because the messy shopper is me. Upon realizing this I stop giving chase, lean over into my shopping cart, and readjust the sugar sack.

According to Perry, my realizing that I am the messy shopper is essential to explaining why I eventually acted the way I did, readjusting my sugar sack: my behaviour cannot be completely explained without indexical reference to myself. Not even a non-indexical, third-personal thought about me being the messy shopper (e.g., the thought that the messy shopper is Chester Fox, where ‘Chester Fox’ refers to me5) seems to suffice for fully explaining my behaviour, since I may be unaware that the object of that thought (the thing referred to by ‘Chester Fox’) is me.

That is the basics of Perry's argument for the essential indexicality of intentional action; indexicals are essential to intentional action because in order to fully explain such actions indexical reference must be made to the agent whose action it is.6 We will have more to say about this argument later, especially about what is going on in the background. For the moment let us assume that is all there is to it.

C&D's objection to Perry's argument is that what Messy Shopper-style examples really show is not that indexical reference is essential to intentional action, but rather that you cannot substitute into explanations of intentional actions salva veritate.

On at least one natural reading of the arguments in Perry (1979), they establish nothing more than that indexicals can't be substituted salva veritate in explanation contexts. The claim that explanation contexts are opaque, however, is not surprising. Those drawn to the thought that agency and the first person are significantly connected want more than that. (Cappelen and Dever 2013: 57)

If this is correct, then even if we cannot eliminate the indexical expressions from Messy Shopper-style cases, the reason for this is not due to anything about indexicality itself. The ineliminability of these expressions is rather due to something else—i.e., the opacity of explanation contexts.

To support this, C&D argue that examples can be constructed having all the relevant features of Messy Shopper-style cases, but which do not obviously or substantively involve indexical thought or language. (They construct similar examples for Perry's other cases as well, such as the amnesiac Rudolph Lingens case, which is why we do not need to discuss these other cases.)

Messy Superman

Superman, who is in the supermarket, has a broken sugar sack and is walking around making a mess. So I give chase, following the trail of sugar in order to tell Superman he is making a mess. But no matter how much I look, I can't seem to find Superman. The only person I find in my endeavor is Clark Kent. After awhile it dawns on me: the reason I can't find Superman, and the reason the only person I can find is Clark Kent, is that Clark Kent is Superman. Upon realizing this I turn back to where I last saw Clark and tell him he is making a mess.

This example appears to be structurally identical to the Messy Shopper example. But the thought eventually leading me to tell Clark he is making a mess—the thought that Clark Kent is Superman—appears not to be an indexical thought; it is a third-personal thought involving non-indexical proper names. Thus, since this example and Perry's Messy Shopper example share all the same relevant features, in particular how a realization of an identity can impact one's behaviour, and since the Messy Superman example does not involve indexical thought, the only reasonable conclusion we can draw is that what is important about these examples is not indexicality or indexical thought. What is important from a philosophical point of view, according to C&D, is that explanation contexts like those involved in both Messy Shopper and Messy Superman are opaque in a general and similar sort of way. Use of indexical terms is merely a way of bringing this opacity out.


There are two strategies for responding to C&D's argument. We might reject their claim that the Messy Shopper case and the Messy Superman case are really similar in all the relevant ways. Or we might accept that they are sufficiently similar, but deny that the thoughts involved in Messy Superman are really non-indexical. For my part, I think the second strategy is the right one. But not for the reasons one might expect. Before we get to that, let us look at some ways we might employ these two strategies.

We might be tempted to respond to C&D by maintaining that the Messy Shopper and Messy Superman cases are in fact not structurally identical. We might want to say: ‘Look, the Messy Shopper case involves indexical expressions embedded under an attitude verb, whereas the Messy Superman case does not. That alone is enough to draw an asymmetry between them.’ In one sense this response is warranted. Indexical expressions are different from non-indexical ones. Semantically, they are context-sensitive, referring to one thing in one context and to something else in a different context. However, this difference is irrelevant. The simple fact that an indexical, whether embedded or not, shifts its reference from one context to the next is not enough to establish the claim that Messy Shopper-style cases essentially involve indexicality. We also need there to be something special about the indexical reference involved in Messy Shopper-style cases. But simply maintaining that the Messy Shopper case is different from the Messy Superman case because the former involves a special kind of reference that the latter lacks is to beg the question.

A better response is to grant that the cases are sufficiently alike, but deny the claim that the Messy Superman case does not involve the relevant sort of indexical reference. One idea is that the thought I come to have and that leads me to do what I do in Messy Superman is not merely that Clark Kent is Superman. It is that Clark Kent himself is Superman. The ‘himself’ here is a third-personal indexical reflexive pronoun that is anaphoric on ‘Clark Kent’. But according to this response, it does not just refer to what ‘Clark Kent’ refers to. Were that the case then the pronoun would be semantically superfluous, contributing no more information than what ‘Clark Kent’ contributes (i.e., reference to Clark Kent). Arguably, however, it is not superfluous. It seems to supply information that plays a substantive role in determining my actions. My thought is more along the lines of that Clark Kent as in that guy I just saw is Superman. And it is this indexico-demonstrative reference to Clark that leads me to chase him down and tell him he is spilling sugar. Without this reference, I simply have a thought of the form x is identical to y, but this alone would not lead me to chase down an individual. I need to first relate this person to myself and my surroundings, and I do this, in thought, through an indexico-demonstrative use of ‘himself’. Thus indexical reference is involved in the Messy Superman case, albeit implicitly.

There is a lot going on in this response. Presumably, C&D are going to deny that explaining my behaviour in Messy Superman requires indexico-demonstrative reference to Clark Kent. According to them, my behaviour in Messy Superman can be explained in terms of an action inventory. An action inventory is a set of actions available to someone in a given situation. In Messy Superman, my action inventory would be something like: Chester looking for Superman by the cash registers, Chester asking an employee where Superman is, Chester giving up the search for Superman, Chester telling Clark Kent he is making a mess, and so on. I, Chester, can engage in any one of these actions, and only I can. But the one I will engage in is determined by the intentions I have, which are in turn determined by my desires and beliefs. C&D describe the inventory as a ‘matching’ system: ‘An agent constantly seeks to match his intentions with his action inventory, and when he finds a match, action occurs’ (Cappelen and Dever 2013: 50).7 Here is how the explanation of Messy Superman will go:

I intend to tell Superman he is making a mess. At first I don't know Clark Kent is Superman. But after a bit I come to believe that he is; I come to believe that Clark Kent is Superman, which has the form x is identical to y. This belief leads me to form a new intention (or modify an old one). Now I intend to tell Clark Kent he is making a mess. This intention is entered into my action inventory and finds a match, as one of the actions available to me is Chester telling Clark Kent he is making a mess. This match results in me going to tell Clark he is making a mess.

This appears to be a satisfactory explanation, and appears to be so without building any sort of indexical reference into the contents of my thoughts.

Kapitan (2014) raises a number of doubts about C&D's action inventory account of intentional action. First, he asks what happens when we confront ‘competing desire-belief sets, each of which has already found a “match” in the action inventory’. How does the action inventory handle such situations if not by the agent deliberating on these competing actions and then choosing between them? Secondly, what about the contents of one's intentions? It would seem that:

When Margo decides to play tennis, for example, what she intends is not merely an action type indexed to a time, but an undertaking of that action type. Whose undertaking? Her own, it seems. It is strange to say that I could intend to undertake Saul Kripke's action of releasing his unpublished papers. I might desire and intend to do something that would cause Kripke to release his papers, or to shine his shoes, but to perform his actions? (Kapitan 2014)

I think Kapitan's questions are the right ones to ask, and I think his suggestive answers are along the right track. In particular, his answers suggest the place to look for indexicality in Messy Superman-style cases—and other cases of intentional action—is not in one's beliefs or desires. It is rather in one's intentions. If intentions, and thereby intentional actions, are essentially indexical, then one's intention to stop Superman from making a mess is indexical.


Before moving on, we need a better sense of what the debate is about. It might seem that what we care about are explanations of actions, such that we are trying to figure out whether we can explain everyone's actions without having to utilize indexical expressions. But the debate is not merely about whether some action explanations must involve indexical expressions. The debate is about whether actions, and more specifically intentional actions, are essentially indexical. The involvement of indexicality in action explanations is meant to be one way of drawing out the Essential Indexicality Thesis (notice the thesis makes no mention of action explanations). Importantly, one can reject that action explanations must involve indexical expressions without rejecting the Essential Indexicality Thesis.

Consider an analogy. Water is essentially molecular. But suppose we want to explain why water takes the shape of containers. One explanation is that water is a liquid and liquids are such that they are disposed to take the shape of any container they are placed in. Not only is this explanation adequate, but also it is also complete: nothing more needs to be said to explain the behaviour of water in containers. But notice this explanation makes no mention of water's molecularity. We could of course expand on this already satisfactory explanation and add some further detail. In particular, we could explain what it means for something to be a liquid. A liquid is a generally non-ordered arrangement of molecules held together by particular intermolecular bonds.8 It is at this stage of spelling out the notion of a liquid—and thereby expanding on our explanation of water's container-related behaviour—that the essential molecularity of water becomes apparent.

With this in mind we can draw out an important question lurking in the background since we began. Is there something common to all action explanations that might be essentially indexical, something whose indexicality need not be mentioned in order to give a satisfactory and complete explanation of someone's actions? I argue that intentions must be indexical, and because of this indexicality is essentially involved in both Messy Shopper-style and Messy Superman-style cases. Put another way, the problem with C&D's argument is they mistakenly assume that if the behaviour in Messy Superman-style cases can be satisfactorily and completely explained without mention of indexicality, then there is nothing about the causes of the behaviour that is indexical. But there is something that is indexical in the cause of the behaviour, namely, the intentions behind it.

It is worth pausing to address an exegetical matter. In what follows, I will be maintaining that most of what I say has been said by or is in the spirit of Castañeda. This might suggest Perry was remiss, claiming that (some9) intentional actions are essentially indexical without backing it up. For a variety of reasons this would be uncharitable to Perry. First, in a telling footnote, Perry gestures at the work of Castañeda, and indicates that this is where the arguments for the Essential Indexicality Thesis lie.10 Secondly, Perry's aim in the ‘The Problem of the Essential Indexical’ is not to argue for the Essential Indexicality Thesis. Rather, it is to argue against a view he calls the Traditional Doctrine of Propositions. He was assuming the Essential Indexicality Thesis and using this assumption to argue against the Traditional Doctrine of Propositions.

Finally, in his later works on what he terms ‘self-notions’, Perry also provides further independent support for the Essential Indexicality Thesis.11 I will not go into the details of Perry's argument for why self-notions are essential in the mental lives and intentional actions of creatures like us. In short, self-notions, hence indexicality, are essential because without them we would not be able to explain why we rely on the particular information that we do when deciding what to do. Perry calls this person identical information, and it is information we take to be about our persons (i.e., our selves). Our self-notions are those under which we store person identical information, and such a notion is necessary for the possibility of intentional action.

Though I think Perry is right, the reason I will not be discussing this argument is I do not think he does enough to establish the necessity of self-notions for intentional action. Sceptics like C&D can respond by pointing out that third-person notions together with the action inventory model of intentional action can do the work that self-notions are supposed to do: the reason an individual, X, acts on information involving X and not someone else is because only X-information properly hooks up to the actions available to X. For example, if the only actions available to me are Chester going to the store and Chester doing the dishes, then clearly only information that somehow has something to do with Chester (or with the acts of going to the store and doing the dishes) will determine the act that I ultimately perform. If this is right, self-notions are not necessary.

The foregoing response to Perry on behalf of the sceptic, and to some extent Perry himself, presupposes that all there is to explain when it comes to intentional actions is how one's information determines the acts they perform. This presupposition is false. There is more to intentional action than just the information relied on and the acts performed. There are also the intentions behind those acts. Moreover, the real role of the information one has is not in determining what acts they will perform. It is in determining what intentions they will form. So the question we should be asking is: Why do I use the information I do in deciding what acts to perform? I return to give an answer to this in the final section, when I sketch my defence of Perry.


Our argument for the Essential Indexicality Thesis, based on Castañeda's work, concerns the nature of intentions and how they relate to intentional action and, more broadly, agency. It goes like this.

  1. Intentions are essential to intentional actions and agency.

  2. Intentions are essentially indexical.

  3. Therefore, intentional action and agency are essentially indexical.

That is it. Though Castañeda never presents this exact argument, he nevertheless endorses and defends both of its premises.12 Thereby, he and others who endorse the premises should also endorse the conclusion. Moreover, the conclusion of this argument is the thesis C&D think we have no reason to accept. So if this argument succeeds, they are mistaken. Since the argument is valid, everything depends on the plausibility of the premises. In particular, much depends on the second premise, as this is the one sceptics will want to reject. Sections VII and VIII are devoted to making the case for this premise. In the next section, I'll say a few things by way of defence of the first premise.

A clarifying remark. By ‘an intention’ what I mean is a mental state. Sometimes ‘intention’ is instead used to designate the content of these mental states. This was Castañeda's use of the term. In the more recent literature, however, it has become standard to use ‘intention’ in the way I will be using it—i.e., as designating a mental state, and not a mental state's content.


That intentional action requires having intentions is as close to a truism as one can get. How can I be able to act intentionally if I do not have, nor ever had, any intentions? Granted, some people deny that the intention causing the action must be an intention to do that particular action, if the action is to be intentional. According to Bratman (1987), all that is required for intentional action is there be some intention or other, and that intention is a cause of the action. For example, I can intentionally drive my car if I intend to go to the store, believe that the only way to get to the store is by driving, and so drive my car in order to satisfy my intention to go to the store. This is perfectly in line with the claim that intentions are essential to intentional action. If anything, it makes the claim more defensible, since it often seems to be the case that one intentionally performs some act without having a specific intention to perform it.

One might nevertheless want to object on the grounds that this presupposes intentions are mental states. When we say someone acted with an ‘intention’, so this line of objection goes, we are not saying they have some special mental state and this mental state is the cause of their action. According to some followers of Davidson's early views on intentions, ‘the phrase [‘acted with an intention’] does not refer to an event or state of the agent, but is a way of redescribing what he is doing in terms of a “primary reason”.’13 If this is right, then it is not that intentional actions require having intentions, but rather that having intentions requires intentional action.

It is well known that a view like this has difficulty explaining intentions we have, but which have not yet been satisfied or otherwise acted on (sometimes called intentions for the future or distal intentions, though I think both of these are misleading names). Davidson himself recognized this problem in his later works, and as a result rejected his original view in favour of one according to which intentions are particular desire-belief complexes, which is a reductionist version of the view that intentions are mental states.14 Moreover, the view that intentions are mental states is very intuitive and has been thoroughly motivated and defended in the action theory literature.15 I will not rehash these arguments here.

The more interesting claim is that intentions are essential to agency. This does not have the ring of a truism. The basic idea behind it is that agency is the ability16 to make decisions about what to do, and intentions are the output of the decision-making process. In Castañeda's words: ‘Formally, we are at this juncture characterizing agents as creatures who propose to themselves ends and are subjects of mandates and prescriptions’ (Castañeda 1975: 143). Proposing an end to oneself is Castañeda's way of saying practically deliberates, i.e., decides what to do. Suppose that I get into a car wreck and as a result fall into a vegetative state. Prior to the wreck, when I had this ability to make decisions, I was an agent. But I lose this status as an agent after the wreck due to having lost all higher level cognitive abilities, and in particular the ability to make decisions and choices. In this way we can see the tight connection between agency and choice: the latter is essential for the former. This means that if indexicality is essential to intentions, then it is essential to agency as well.


Intentions are non-cognitive attitudes that lead to and thereby explain some of an agent's actions. In most circumstances, one's intention to do A causes one to do A. Circumstances in which this causal link fails include cases of physical or mental impairment or inability, cases of conflicting intentions (i.e., distinct intentions that mutually exclude each other's satisfaction), and cases where the agent does not know how to perform the act they intend to perform.17 Moreover, in those cases in which an intention successfully leads to an action that satisfies it, we say that the intention explains the action. My intention to wake up early this morning explains why I woke up early this morning.18

Although often characterized in terms of acts to perform, intentions involve more than just acts. They also involve a subject—i.e., the individual whose intention it is.19 My intention to go running on Tuesdays does not only involve the act of running on Tuesdays. It also involves me. What I intend is myself to go running on Tuesdays, that I go running on Tuesdays. If we think intentions are best modelled as To-Do Lists (a la Portner 2004), then we would say that the act of running on Tuesdays is on my To-Do List. Not yours, not my neighbour's, not anyone else's but mine. Similarly for your intentions: you intend yourself to do something, and so performing that act is on your To-Do List. One advantage of modelling intentions in this way is it allows us to sidestep the debate over whether the contents of intentions are acts, propositions, or something else.20 Each of these options can be modelled using To-Do Lists.

Notice I have characterized intentions using indexical pronouns: ‘I’, ‘Me’, ‘Myself’, ‘My’, ‘You’, ‘Yourself’, and ‘Your’. The question is whether one's intentions must be spelled out indexically. By this we mean, is it essential to intentions that they be indexical? There is good reason to think so. For one thing, it does not even make sense to characterize someone's intentions non-indexically. For example, while the sentences in (1) and (2) are perfectly sensible, those in (3) and (4), taken as they are, without interpreting them as having unarticulated material (see below), are incomprehensible.

(1) a. I intend that I go running.

   b. You intend that you go running.

   c. Jill intends that she (herself) go running.

(2) a. I intend myself to go running.

   b. You intend yourself to go running.

   c. Jill intends herself to go running.

(3) a. *I intend that Jill go running.

   b. *You intend that Jill go running.

   c. *Jill intends that Jack go running.

(4) a. *I intend Jill to go running.

   b. *You intend Jill to go running.

   c. *Jill intends Jack to go running.

The key difference between the sentences in (1) and (2) and those in (3) and (4) is that the embedded clauses in the former, but not the latter, have indexical subject terms that co-refer with the subject term of the main clauses, which suggests that indexicality, of some form, is not eliminable from reports about intentions.21

It is important to note that the sentences in (3) and (4) will be fine if interpreted as saying that the main clause subjects intend to make it the case that the object clause subjects perform the relevant acts. If (4b), for example, is interpreted as saying that the addressee, who assume is not Jill, intends to make it the case that Jill go running, then it will sound perfectly natural and fine. We might call this the intends-for interpretation: the addressee intends for—has as a goal—Jill to go running. This is not the interpretation we are concerned with here. We are concerned with the straight intends-to interpretation, on which (4b) says that the addressee intends Jill to—has as a planned course of action—go running. It is this interpretation of (4b) and the rest that is unintelligible. I understand how I can have an intention the complex content of which is that I see to it that someone else do something (intends-for), but I do not know how to begin to understand how someone can have an intention the sole content of which is that someone else do something (intends-to). This, I take it, is essentially the point Kapitan was making in his case concerning himself intending Saul Kripke to release his unpublished papers.

Why do the intends-to readings of the ascriptions in (3) and (4) not make sense? They do not because they imply one person has direct authority over someone else's agency, which is incomprehensible. For example, (4c) implies that Jill has direct authority over not just Jack's actions, but over his agency as well. By ‘direct’ authority, what is meant is not that Jill can cause Jack to do certain things. What is meant is Jill literally makes Jack's decisions. She is not merely making his decisions for him; she is making his decisions full stop. Sure, Jill can have indirect control over Jack by issuing directives to Jack, manipulating him in certain ways, and otherwise guiding his actions and decisions. She can even take Jack's agency away, perhaps by hijacking his brain and turning him into an automaton. It is in such cases that we say Jill makes decisions for Jack. But actually controlling Jack's agency—actually making Jack's decisions full stop is simply not possible. In order for Jill to make Jack's decisions full stop, Jill would have to be Jack. That, however, is beyond the reach of possibility.22

If you're like me, you will not know how to even begin thinking about what it would be like for one person to have direct authoritative control over another. Such a situation is so far beyond comprehension that trying to articulate it reduces me to mumbles. This is why I say such situations are not just impossible, but incomprehensible. And it is due to this incomprehensibility that constructions like those in (3) and (4) are incoherent, and in turn why intention ascriptions have to be formulated with indexical expressions, because only one's self can make one's decisions full stop.

It is worth noting that C&D's objection to arguments for the Essential Indexicality Thesis does not apply here. Recall that C&D aim to show that Messy Shopper-style cases are merely cases of opacity: the reason you cannot eliminate the indexical expressions from Messy Shopper-style cases is because action explanations are opaque in a (non-special) way that prevents substitution of the indexical expressions for non-indexical expressions salva veritate. Importantly, since action explanations are merely opaque, substitution of indexical for non-indexical expressions in Messy Shopper-style cases does not render them incoherent. But the same does not go when non-indexical intends-to ascriptions are involved. Because non-indexical intends-to ascriptions are incoherent, any story involving them will likewise be incoherent.


C&D appear to have a way to reply to the foregoing argument. It is to deny that non-indexical intentions are impossible, hence deny that non-indexical intention ascriptions are incoherent. Throughout their book they assume that, like beliefs and desires, intentions can be non-indexical, and seem to think that this is so because beliefs and desires are non-indexical: ‘Every agent has a very wide range of third-person beliefs and desires that give rise to third-person intentions, which in turn rationalize or motivate actions (via their recognition)’ (2013: 49). Clearly this is not enough to establish that intentions can be non-indexical. For it might be the case that intentions are the products of not just our beliefs and desire, but of some third factor as well, which might introduce indexicality into our intentions. So what C&D need is that our intentions can be fully or exclusively determined by our non-indexical beliefs and desires. A claim as strong as this, however, needs support.

Why is it not that our desires and beliefs only ever partly determine our intentions, where the remaining part is determined by something else entering into the decision-making process? To answer this we need an account of practical decision-making. What goes into practical decision-making are reasons, which are largely determined by—or just are—an agent's beliefs and desires. For example, my belief that chocolate contains fibre gives me reason to eat chocolate. And my desire to eat chocolate gives me another reason. But my belief that chocolate causes bloating gives me reason not to eat it. When it comes time to decide whether to eat chocolate, all these reasons enter into my decision-making process. Assuming I am rational, if the reasons-for outweigh the reasons-against, I will form an intention to eat chocolate, which in turn will cause me to do so.23

For those like Castañeda this process of deciding what to do, or practical reasoning, always takes place within a first-personal, subjective perspective, which is ‘personal, ephemeral, confrontational, and executive’ (1989: 126). The point Castañeda is pushing is that there is no such thing as deciding what to do, or reasoning, under a third-person, objective perspective, as this would be mere computation or calculation, as a non-agential computer would do. We, however, are not mere computers. Similarly, Burge argues that ‘fully understanding the concept of reason, and engaging in reasoning in the most reflective and articulated way, require having the I concept and being able to apply it for this purpose’ (2000: 259). Though Burge's point is weaker than Castañeda's (he is only saying that one cannot understand the concept of reason fully unless one can conceptualize oneself in a first-person way), it still serves to mark an important relation between reason and the first-person.

That we are not mere computers can be drawn out by considering cases of conflicting intentions. Suppose that on Monday Jill forms an intention to phone her mother at 4 pm on Thursday. Now it is Wednesday and Jill is trying to schedule a meeting with her boss to discuss a problem at work. The only time her boss is available is 4 pm on Thursday. Not thinking about the intention she formed earlier, Jill agrees to meet then and thereby forms an intention to do so. Soon afterwards, she realizes her mistake. She now has two intentions that cannot both be satisfied. If Jill were a mere computer, this conflict would cause a system crash.24 But this is not what happens. Rather, Jill (qua herself) recognizes the conflict and makes an executive decision to abandon one of the intentions and keep the other (or abandon both). It is here, in deciding between the conflicting intentions, that we see Jill is more than a computer, that decision-making requires an agent herself to be part of the process.

This brings out a core problem with C&D's action inventory account of intentional action, according to which an intentional action is one that results from an intention finding a ‘match’ in an inventory of those actions available for an agent to perform. What happens when two competing intentions simultaneously find matches in the inventory? According to our account of decision-making, in such cases the agent herself will choose which intention to give up (or choose to give up both). This is part of the executive function of agency. However, such an account of what happens when two (or more) intentions conflict is not open to C&D. Nor can C&D say that in such cases the action-production process simply crashes or halts. If it simply crashed or halted, then neither action should result—but we know this is rarely the case.

The point being driven at is no matter how C&D spell out cases of conflicting intentions, they are ultimately going to have to appeal to the volitional capacities of the agent. They are going to have to appeal to some process where the agent adopts—or to use Burge's (2000) terminology, conceptualizes using—a first-person perspective and within that perspective makes a choice as to what to do. This is what makes us agents and distinguishes us from non-agents.25 So even here, in the agential decision-making process, we find indexicality to be indispensible.

We can now return to the question we began with—might anything else besides our desires and beliefs contribute to the structure and content of our intentions? Yes. Not only do our desires and beliefs make contributions (by giving rise to reasons that go into the process of forming intentions), but it is possible that the first-person perspective through which intention formation unfolds also makes a contribution. Roughly, it contributes itself. Such a view paves the way for a deeper explanation of why intentions are essentially first-personal: they are because they are formed by an agent deliberating within their own first-person perspective, and the reason no one can directly control anyone else's agency is because to do so would require deliberating from within their (the other person's) first-person perspective, which is impossible.


I have argued that both intentions themselves and the process by which they are formed (i.e., practical deliberation) are essentially indexical. Since intentions and practical deliberation are essential to intentional action and agency, this means that intentional action and agency are essentially indexical as well. But what about beliefs (and desires26)? Perry's central claim in ‘The Problem of the Essential Indexical’ is that some beliefs are essentially indexical as well. In the Messy Shopper case, the belief that I am the messy shopper is essentially indexical, according to Perry, because without indexicality I never would have adjusted the sugar sack in my cart. We are now in a position to explain why this is, and thus respond to C&D's objection.

The key feature of the Messy Shopper case that was overlooked in our original presentation of it is my intention to stop the messy shopper. Here is how we should have presented the case:

Messy Shopper Redux

Someone in the supermarket has a broken sack of sugar and is walking around making a mess. Upon noticing the trail of sugar, I form an intention to stop the messy shopper. So I give chase, following the trail of sugar in order to tell this person they are making a mess. But no matter how quickly I run around the aisles of the market, I can't seem to catch up to the messy shopper. After awhile it dawns on me: the reason I can't catch up to the messy shopper, the reason I keep running around in a circle, is because the messy shopper is me. Upon realizing this I stop giving chase, lean over into my shopping cart, and readjust the sugar sack.

Crucially, like all intentions, the one I have here is essentially indexical: I intend myself to stop the messy shopper. It is because of the indexicality of my intention that my later belief about the identity of the messy shopper must also be indexical.

Suppose my belief about the identity of the messy shopper is purely third-personal, such that it is simply of the form x is identical with y. How would such a belief interact with my intention, which is not purely third-personal? In short, it would not. If I conceive of myself in a first-person way in my intentions, then my beliefs about myself must also involve conceiving of myself in a first-person way if those beliefs are to affect my intentions or the way I go about satisfying them. So my belief about the identity of the messy shopper must be first-personal after all. This, I take it, is what Perry was pointing out in his later writings on indexicality and self-notions. My self-notion is inextricably involved in my intentions and deliberations, which is why the information stored under this notion plays a stable role in affecting my actions (Perry 1998).

Carrying these considerations over, the Messy Superman case also has suppressed indexicality. The case in full detail will mention my intention to stop Superman from making a mess. This intention, like all others, is essentially indexical. Thus indexicality is involved in the Messy Superman case after all. It is just that in our original discussion of the case, we were not looking in the right place. The place to look for indexicality, in the first instance, is not in one's beliefs, but rather in one's intentions. The indexicality of the belief involved in the Messy Shopper case, unlike in the Messy Superman case, is due to the way beliefs interact with intentions. That belief is essentially indexical only because it has to be if it is to interact in the right way with my intention to stop the messy shopper. And the reason the belief in the Messy Superman case need not be indexical is that I am not Superman, so there is no need for me to believe that I am if my intention is to be satisfied.27


C&D argue that we should be sceptical about indexicality being essential to intentional action and agency. They do this by arguing that the arguments of Perry, which are often taken to establish the Essential Indexicality Thesis, in fact fall short of doing so. The trouble with C&D's approach is that they overlook a different argument for intentional action being essential indexical. This is the Argument from Intentions: the Essential Indexicality Thesis holds because (1) intentions are essential to intentional action and agency and (2) intentions are essentially indexical. Thus, one reason indexicality is philosophically interesting and ‘deep’ is that without it we cannot hope to give a full explanation of intentions and, thereby, intentional action. Furthermore, the essential indexicality of intentions gives us a way to vindicate Perry's original claim that in some cases (e.g., Messy Shopper) certain beliefs must be indexical as well. So those who wish to rely on what Perry says about these cases are not wrong to do so.28

Castañeda (1967)
E.g., Lewis (1979), Burge (2000), and Bermudez (forthcoming).
Another challenge comes from Millikan (1990).
See especially Castañeda (1975).
Not my real name.
This may not be completely right. Perry never explicitly says that in all cases explaining an intentional action requires indexical reference to the agent. Nevertheless, he should accept it for the reasons given below.
The talk of an ‘agent constantly seeking to match his intentions with his action inventory…’ appears to be metaphorical. The matching process has to be brute and for the most part automatic, or else C&D will be smuggling indexicality in through the back door.
See Egelstaff (1994).
See footnote 6.
In particular, he cites Castañeda (1966, 1967). A notable omission from the works Perry lists is Castañeda (1975).
See in particular Perry (1998).
For the first premise, see chs 12 and 13 of his (1975). For the second premise, see ch. 6 of the same.
Davidson (1963: 5–8)
See Davidson (1978).
See Setiya (2015) for a nice overview and references.
By ‘ability’ I do not mean ‘physically able to’ or even ‘mentally able to’. One can have the ability to swim, say, even if their legs are broken and they are physically unable to swim.
One might deny that you can have an intention to A if you do not know how to A. I think this is obviously too strong. I can intend to fly a plane next year even though I do not now know how to fly a plane. A more plausible claim is that you cannot satisfy an intention to A if you don't know how to A. See Gibbons (2001), Stanley and Williamson (2001), and Setiya (2012) for further discussion of the link between intentions and know-how.
The question remains of how intentions cause the actions they do. There is a sizable literature on this topic, and I will not be able to do it justice here. For those interested, a good place to start is Harman (1976) and (1986), followed by Mele (1992) and Kapitan (1995).
Castañeda argues most clearly for this point in his (1972). But see also Castañeda (1971).
See Castañeda (1972) and Setiya (2015) for discussion.
The subject of the embedded clause can be–and often is–left implicit, as in:
  • I intend to go running.

A common way to account for the subject of apparently subjectless to-clauses is by postulating the presence of an unarticulated pronoun–‘PRO’–that is anaphoric on the subject of the main clause. Thus, the real syntax of (i) is:
  • (ii) I intend PRO to go running.

The reason I do not discuss sentences like these is because ‘PRO’ introduces additional complications we need not be concerned with.
Burge (2000: 255-6) makes similar remarks in the context of a discussion of evaluating the rationality of one's attitudes.
Provided I do not suffer some sort of cognitive or physical defect that would prevent formation of the intention or performance of the act.
This is why AI systems are designed in a way that either precludes the possibility of conflicting intentions, or when they do arise, one intention is selected on the basis of some predetermined, mostly arbitrary value (McDermott and Forgy 1976).
Arguably, some non-human animals can do this as well. Dolphins and certain primates appear to form something like proto-intentions, attitudes that though perhaps not formed via reasoning, are nonetheless formed via a process involving a first-person perspective.
Though my focus in this section is on beliefs, everything I say carries over mutatis mutandis to desires.
This is not to say I could not be confused and think I am Superman. Clearly I could. But that is beside the point.
I owe thanks to Robin Jeshion, Tomis Kapitan, Scott Soames, and Gabriel Uzquiano for their invaluable suggestions and comments on previous versions of this paper. Special thanks goes to Robin Jeshion, as without her encouragement this paper never would have been written.


‘Yes, Essential Indexicals Really Are Essential’
Intention, Plans, and Practical Reason
Cambridge, MA
Harvard University Press
‘Reason and the First Person’
Knowing Our Own Minds
The Inessential Indexical
‘‘He’: A Study in the Logic of Self-Consciousness’
‘Indicators and Quasi-Indicators’
American Philosophical Quarterly
‘Intentions and the Structure of Intending’
Journal of Philosophy
‘Intentions and Intending’
American Philosophical Quarterly
Thinking and Doing
D. Reidel
Thinking, Language, and Experience
Minneapolis, MN
University of Minnesota Press
‘Actions, Reasons, and Causes’
The Journal of Philosophy
Philosophy of History and Action
D. Reidel
P. A.
An Introduction to the Liquid State
Clarendon Press
‘Knowledge in Action’
Philosophy and Phenomenological Research
‘Practical Reasoning’
The Review of Metaphysics
‘Willing and Intending’
R. E.
Philosophical Grounds of Rationality
‘Intentions and Self-Referential Content’
Philosophical Papers
‘Review of The Inessential Indexical’
‘Attitudes De Dicto and De Se’
Philosophical Review
‘Production System Conflict Resolution Strategies’
D. A.
Pattern-Directed Inference Systems
New York, NY
Academic Press
The Springs of Action
R. G.
‘The Myth of the Essential Indexical’
‘The Problem of the Essential Indexical’
‘Myself and I’
Philosophie in Synthetisher Absicht
‘The Semantics of Imperatives within a Theory of Clause Types’
R. B.
Proceedings of Semantics and Linguistic Theory
Ithaca, NY
CLC Publications
‘Knowing How’
Proceedings of the Aristotelian Society
The Stanford Encyclopedia of Philosophy
Summer 2015 Edition http://plato.stanford.edu/archives/sum2015/entries/intention/ accessed 10 Aug 2015
‘Knowing How’
Journal of Philosophy