1 Introduction

An orthodox view in philosophy holds that there is a distinctive variety of thought, identified variously as thoughts’ being de se, being about oneself as oneself, or as ‘I’, which is psychologically and semantically special. They are held to be psychologically special because of their role in coordinating information, motivation, and practical procedure in the explanation of action, and they are held to be semantically special because of how they pick out the thinker through context-dependent or token-reflexive semantic rules. Such rules pick out the object the thoughts are about, which in the case of de se thoughts is the thinker herself, by picking out whoever is the thinker of the thought. The general motivation behind the orthodox view is that, without this way of picking out the thinker, there is no way to explain de se thought’s distinctive role, and without the distinctive role, we cannot account for how action happens. In Perry’s phrase, indexicals are essential.

There has been a persistent strand of skepticism about this orthodox view concerning de se thought. Such skepticism comes in two forms. Radical de se skepticism denies the need for either psychologically or semantically special self-referential thoughts, which requires an alternative way of capturing the sorts of cases that seem to favour the orthodox view. For example, Cappelen and Dever argue that the explanation of action might simply involve intentions having actionable contents, in that they happen to concern outcomes that the thinker can bring about. Thoughts with actionable contents are not a distinct psychological or semantic kind.Footnote 1 A more moderate form of de se skepticism grants that there is a need for a psychologically distinctive kind of thought, and so concurs with orthodoxy to that extent, but denies that they are semantically special. As Ruth Millikan, an advocate of this latter view, puts the point, ‘so-called “essential indexicals” in thought are indeed essential, but they are not indexical.’Footnote 2

Morgan (2021) argues that, while skepticism about the orthodox view is by now well motivated, ultimately it fails. Morgan’s argument for the psychological distinctiveness of de se thoughts is that ordinary thoughts which just happen to be about the thinker result in actions that exhibit a different psychological and physiological form to those that could be thought of as being distinctively de se. Morgan’s argument for the semantic distinctiveness of de se thought is that these psychologically special thoughts bear a special connection to indexical terms in language. Morgan calls this relationship direct expression. De se thoughts are directly expressed by uses of sentences employing the first-person pronouns (‘I’, ‘me’, ‘mine’, ‘my’ in English), which have their meanings given by indexical semantic rules. This (alleged) fact about direct expression needs an explanation. Morgan claims that the best explanation is that the psychologically special thoughts about oneself also employ a concept, or guise, that encodes the same kind of semantic rule as the first-person pronoun, which is an indexical rule. The psychologically special thoughts are therefore also semantically special, contra the moderate de se skeptic.

My aim in this paper is to defend moderate de se skepticism against the argument from direct expression by focusing on a problem with the terms of the argument as stated. The problem is that the test of direct expression Morgan provides—that the use of a term does not involve what he calls ‘identity knowledge’—cannot be the one he wants, since it generates the wrong result when applied to uses of proper names that users know are their own, and more alarmingly, it generates the result that competent uses of ‘I’ do not directly express de se thoughts at all. I will show that a more natural analysis of the special connection is available: accepting an identity involving that guise and the referent of the term is a condition on competent use. But when the explanatory argument is applied using this test, it fails, since the specific content of the semantic rule encoded by the de se guise is irrelevant to the special connection it has to the first-person pronoun. As far as what is represented in thought is concerned, all that matters is that the referent is the right one. This tallies with the idea behind Millikan’s moderate de se skepticism: the psychological role that de se thoughts play is not determined by their semantics.

Section 2 motivates the moderate de se skeptic’s position, and lays out Morgan’s explanatory argument from direct expression. Section 3 lays out the problem with the notion of direct expression, and explains why the alternative analysis, in terms of competent use, is to be preferred. Section 4 shows that the way in which self-referential thoughts are determined as self-referential by semantic rules is irrelevant to the special connection between ‘I’ and the de se guise. Section 5 concludes.

2 The argument from direct expression

As I put it above, the orthodox view makes claims concerning the distinctiveness of a particular kind of thought, along two related dimensions:

Essential Role I: there is a kind of thought (de se thought) that (i) is about/refers to the thinker of the thought and (ii) has a psychological role with characteristics that are essential for the explanation of action.

Indexical Semantics: The characteristics of the thoughts that are essential for the explanation of action is that they refer to the thinker via a context-dependent or token-reflexive semantic rule.

Essential Role I concerns action in a broad sense, and explanation of action in the sense of rational, intentional explanation, one that appeals to the mental states of the agent. Agents know things, and want things, and have ways of achieving what they want in light of what they know. These states need to be coordinated in a way that seems not to apply to thoughts about someone who might happen to be the agent herself, but which have the characteristics of third-personal thought. For present purposes we can define third-personal thought negatively as thoughts that don’t refer by going via a contextual or token-reflexive semantic rule. The basic idea behind Indexical Semantics is that the way in which de se thoughts get to be about the thinker must be different from the way that thoughts of this other kind (typically about other people but in peculiar cases, involving mirrors and amnesiacs, perhaps also about the thinker herself) get to be about whatever it is they are about.

The radical de se skeptic denies this, arguing that third-personal thoughts can play the action explanation role, and also that these thoughts don’t have anything special about them psychologically either. For example, Cappelen and Dever state:

First, it is not necessary for an indexical element to enter into the rationalization [of action]. Second, on our view the agent doesn’t even need to be represented in a non-indexical way in an adequate action rationalization.Footnote 3

They propose that action can be explained simply by the production of intentions with a particular kind of content, one that concerns actions that the agent can carry out, such as intending that she move her arm, as opposed to un-actionable contents, such as that someone else move their arm. Thus, these thoughts are distinctive as to topic, not as to their function in the psychological set-up of an agent. It follows that Essential Role I is a false claim about the psychology of action, and since Indexical Semantics presupposes that claim, it is false also.Footnote 4

Morgan presents a response to this line of thought along the following lines:Footnote 5 when we compare cases where an agent is responding to a thought in forming an intention, we find different characteristics that must go beyond the actionability of the contents. The typical behavioral results of thoughts that are typically third-personal, which lack any special connection to self-knowledge and motivation, are importantly different from those which do have that connection. So, for example, someone who is fleeing a bear when they think of themselves as being threatened by a bear under their de se guise will be in a certain physiological state (their nervous system ‘is disposed to be highly aroused’, as Morgan puts it), with intentions having a particular kind of content (a verbal content, e.g., to run away, rather than a state content, e.g., that someone is safe from the bear), which are executed in particular ways (e.g., seeing a bear is followed by running, as opposed to initiating a series of events that has the indirect consequence that someone is safe from the bear). Morgan captures this as a state being ‘immediately motivationally potent’,Footnote 6 the idea being that not all self-referential states have this character. There is therefore a division to be made between actions that cannot be captured merely by a categorization into those that are actionable for an agent and those that are not.

Morgan concludes that there is a feature of thoughts that are essential for action explanation, the connection between self-knowledge and motivation, that goes beyond their content. It is worth being clear, however, on what exactly Morgan’s argument is meant to show. Special Role as stated claims that that de se thoughts are essential for action explanation in general. Morgan does not defend this. Rather, he defends the claim that there are certain actions for which de se thoughts are essential.Footnote 7 He therefore defends a weaker thesis than Essential Role I, as follows:

Essential Role II: there is a kind of thought (de se thought) that (i) is about the thinker and (ii) has a psychological role with characteristics that are essential in the explanation of a significant sub-set of possible actions.

The moderate de se skeptic can accept the motivation for this claim, or anyway what it motivates. Our question is whether Essential Role II can be coherently combined with a denial of Indexical Semantics. But why would one be moved to do that?

The worry that motivates Millikan’s more moderate skepticism about orthodoxy can be put like this:Footnote 8 Indexical Semantics is mistaken about what needs to be the case for a de se thought to be about the thinker. The key to the semantics of terms like ‘I’ lies in how they are interpreted, in what one needs to know in order to know what was communicated by a given use. Uses of such terms are interpreted using, and can only be properly interpreted by knowing, facts about the context, for example, who the speaker is. The semantic rule specifies a way of interpreting the use given facts about the context. But thoughts are not interpretable in any analogous way. They aren’t issuances in a speech context; they are not produced with communicative intent; they aren’t publicly accessible, datable, or locatable tokens; they do not bring about changes to conversational common ground; and so on. Indexical terms are like this, and that, plausibly, is part of the story about why they are apt to have a context-sensitive semantics. But if all the facts that make indexical words suitable for being indexical do not apply, what reason would there be to bring in guises involving indexical semantic rules in the case of thoughts? If there is no need for indexicality in order for those thoughts to represent what they do, then our best theory ought not attribute to them indexical semantic properties.

A stronger conclusion (which Millikan has drawnFootnote 9) is that the inapplicability of the sorts of interpretation facts that apply to indexical words means it is impossible for thoughts to be indexical. But the weaker point, that there appears to be no basis for taking thoughts to refer via context, is enough to motivate skepticism about Indexical Semantics—to motivate it, mind, not to demonstrate its truth. But the burden, the moderate de se skeptic insists, is on advocates of orthodoxy to demonstrate some need for an indexical semantics, even granting that de se thoughts do have a special psychological role.

The question at issue between the moderate skeptic and the advocate of orthodoxy comes down to whether, granting Essential Role II, there are reasons also to accept Indexical Semantics that outweigh the fact that thoughts are, unlike sentences/utterances, not apt to be interpreted in a context.

The proponent of orthodoxy might claim an explanatory advantage for her view. The skeptic would need to say that the semantic rule that picks out the referent of a thinker’s de se guise is just: it refers to A (de re, not specifically under a given description), where A is the person who happens to be the thinker. Call this kind of rule ‘Millian’. Arguably, the Millian rule for a de se guise does not itself explain why A is the referent of the guise, only that she is. In contrast, an indexical semantic rule does explain this: the thinker is the referent because she satisfies the description that figures in the indexical semantic rule. So, the claim is, there is an explanation offered by the orthodox view that the skeptic cannot as easily give.

One could respond by offering an account of why a Millian de se guise refers to its thinker, perhaps in the form of a general theory of mental reference. While not ruled out in advance, this would certainly be an unwelcome burden. However, the appearance of an explanatory advantage is misleading, and this is not a burden the moderate de se skeptic should accept, so far as the debate with orthodoxy is concerned.Footnote 10 Both the Millian rule and the indexical rule are facts at the same explanatory level: they determine the referent of the de se guise. To the extent that there is a question about why the de se guise refers to the thinker, they supply answers of equally good standing: they both identify the thinker as the referent, albeit in different ways. There will be, for both rules, a further question as to why the rule holds. It is not likely to be the case that the indexical rule is somehow a bedrock fact, any more than the Millian rule might be. There will need to be a metasemantic account (in Kaplan’s sense) of why the semantic rule is the rule for that guise. I stress that this demand applies equally to both rules. Both rules always pick out the right object, but in neither case is this a self-standing fact. So there is no explanatory advantage to be had by opting for orthodoxy on that score.

Morgan offers a more promising argument, in terms of what he calls the de se guise, and the direct expression of a guise. Morgan gives a minimal definition of a guise in terms of Frege’s criterion of individuation for senses: ‘If a rational subject can simultaneously believe “a is F” and disbelieve “b is F”, then “a” and “b” are associated with different guises, even if they have the same reference’.Footnote 11 It is therefore something that psychologically distinguishes co-referential thoughts, such that they can have different roles (in action explanation, reasoning, emotional valence, and so on). I will follow Morgan in leaving it open what, in the final analysis, a guise might be. The de se guise is the guise that gives de se thoughts their particular character, meant to distinguish them from other thoughts which are also about their thinkers. The notion of direct expression is Morgan’s formulation of a special connection ‘I’ stands in to the de se guise. The idea is that the de se guise is expressed by ‘I’ (and the other first person pronouns—I’ll ignore this from now on) in a particular way, one in which it is not expressed by other co-referential terms, i.e., non-indexical names. Morgan provides a test for direct expression which I will set out and raise a worry about in the next section.

The argument Morgan sketches is this: if you grant that there is a de se guise, and that ‘I’ is its direct expression, then we need an explanation of the latter fact. A simple explanation is that the de se guise and ‘I’ have something in common. What they plausibly have in common is that the de se guise encodes the same—or same sort of—indexical rule as the rule that governs the meaning of ‘I’.Footnote 12

Morgan rightly points out that Millikan accepts that there is a special connection, and she herself uses this phrase, but less convincingly he asserts that she also accepts that ‘I’ is the direct expression of the de se guise.Footnote 13 It is important to emphasize the point that direct expression is a technical notion introduced by Morgan; the claim that there is a special expressive relationship between ‘I’ and the de se guise does not entail the analysis of that relationship in terms of direct expression. I underscore this non-equivalence, as the objection I raise below turns on the unsuitability of that analysis, though I do not reject the special connection claim.

Morgan does not propose an answer to the question of how an indexical semantics for the de se guise would explain the (alleged) direct expression fact; he holds this out as a question guiding what he terms the ‘essential indexical research programme’.Footnote 14 The conclusion I want to focus on here is the claim that the special connection between first-person pronouns and the de se guise gives us a reason to think that the de se guise, the one that has an essential role in explanation of some actions, also has an indexical semantics. Although Morgan’s argument is only briefly stated, it articulates a strong current behind the orthodox view, and one reason perhaps that the moderate skeptical position has not received more attention in the literature. It is therefore worth considering in some detail. My objection to the argument, in short, is that there is no path to indexicality from the special connection once the latter is properly analysed, and so no reason to overturn moderate de se skepticism.

3 A problem with direct expression

The main problem with Morgan’s argument from direct expression is knowing what exactly direct expression is supposed to be. This is Morgan’s statement of ‘the test’:

The test for what word directly expresses a guise [a thinker] uses in thought is what word would be used by him on an occasion when he expresses a state involving that guise and his choice of language does not exploit any identity knowledge he has concerning the thing the guise refers to.

We do not get much more by way of discussion or motivation for this as an analysis of the special connection. Direct expression is spelled out in terms of knowing an identity, the idea being that a word directly expresses a guise if it (a) expresses the guise and (b) does so without the user of the word relying on knowledge of an identity involving the guise. How exactly to apply these conditions is not spelled out. On a natural interpretation of the latter condition, however, it turns out that the test has the consequence that proper names (i.e., non-indexical terms) can directly express the guise corresponding to de se thoughts, and that first-person pronouns cannot directly express it, which is the opposite of what Morgan requires from direct expression.

The statement of the test is preceded by a case involving the famously pompous Charles de Gaulle, who (in philosophical lore) only ever self-referred using his own name.Footnote 15 De Gaulle’s use of his own name to self-refer may nevertheless be an expression of the guise that is directly expressed by ‘I’, because ‘he may be stably disposed to avoid the word that directly expresses the guise of him some of his states involve [i.e., his de se guise]’.Footnote 16 The suggestion, not explicitly stated by Morgan, is that de Gaulle exploits identity knowledge in his uses of his own name to express these states. This bears some examination.

De Gaulle utters ‘de Gaulle is hungry’. Let’s assume that the mental state that de Gaulle is expressing, the belief he has that he wants to get others to recognize as the one he is trying to express, employs the de se guise. So it’s the kind of thought we who are prepared to use the first person pronoun, as de Gaulle is not, might express by saying ‘I am hungry’. Morgan’s idea must be that, in order to express this thought using the non-indexical name ‘de Gaulle’, de Gaulle relies on an identity that relates the thought that employs the de se guise to the non-indexical name. What is the identity? There are two possibilities. One is

I am de Gaulle.

This by itself does not explain the use of ‘de Gaulle’ to self-refer, as it would also need to be the case that de Gaulle thinks de Gaulle is called ‘de Gaulle’. Then the belief state would be

I am de Gaulle and (therefore) called ‘de Gaulle’.

Alternatively, it might be a directly metalinguistic identity, such as

I am the referent of ‘de Gaulle’.

Here, ‘the referent of “de Gaulle”’ is a unique description, and as in the first possibility, ‘am’ is the ‘is’ of identity (symmetrical, transitive). So the idea would be that the test returns a negative verdict in this case because de Gaulle’s needs to exploit an identity belief involving de Gaulle’s de se guise and some metalinguistic information about the name used. Perhaps that is not exactly the way of putting it Morgan would have in mind. But I submit that whatever the identity knowledge in the de Gaulle case is supposed to be, it would need to be something of this form.

Generalizing from the case, Morgan’s idea must be that, if a speaker uses a non-indexical name ‘NN’ to express a thought that employs the de se guise, she needs to believe something like I am the referent of ‘NN’ or I am NN and therefore called ‘NN’ in order to use the name to express the thought, for example, that the thinker herself is hungry. Is this right?

Compare the de Gaulle case with a case where a speaker uses a name to refer to someone they know (or anyway believe) to be someone other than the speaker. De Gaulle says: ‘Churchill is lazy’. If the claim about de Gaulle’s use of ‘de Gaulle’ to refer to himself is on the right lines, then we would assume something similar is true of his use of ‘Churchill’ to refer to someone else, in this case, to Churchill. That is, we would expect him to decide to use ‘Churchill’ at least partly on the basis of believing

Churchill = the referent of ‘Churchill’.

But no such identity needs to figure in de Gaulle’s reasoning in choosing ‘Churchill’ to refer to Churchill. All de Gaulle needs to know is that Churchill is called ‘Churchill’, and that if he uses ‘Churchill’ to refer to ‘Churchill’, then his intention to refer to Churchill will be reliably recognized. Being called or known by a name is not an identity fact; it is a property that relates a person to a name. In fact, de Gaulle need not have any notion at all about the reference of the name. He only needs to know a particular social fact about Churchill, namely, that he is known by the name ‘Churchill’. Given his knowledge of that fact, De Gaulle’s utterance of ‘Churchill is lazy’ is properly explained without his needing to entertain an identity concerning the referent of the name.

It is therefore false to say that someone needs to entertain an identity in order to refer to someone else using a name. One needs to know social facts about naming, not metaphysical facts about identity. By this token, in the specific case of someone using a non-indexical name to refer to themselves in expressing a thought that employs the de se guise, all one needs to know about oneself is a fact of the same kind, i.e., I am called ‘NN’. No identity knowledge necessary.

Morgan phrases the test of direct expression in terms of what word a speaker would choose to use to express a guise, and the knowledge a speaker would exploit in the choice to use it. It is not entirely obvious how to evaluate this when applied to cases of self-reference using the referrer’s own name without relying on an identity. But it would seem to follow that non-indexical names can directly express the de se guise, by Morgan’s test, and it would seem that de Gaulle’s is not a case of indirect expression of the de se guise, but one of direct expression. The conclusion that names can directly express de se guises makes the claim that direct expression is the right way to capture the special connection between ‘I’ and the de se guise look dubious. But it could be argued that there is nevertheless a particular explanatory reason for the relation obtaining in the case of ‘I’, one that does not obtain for names. So, while the above point suggests something is amiss, it could perhaps be accommodated.

The problem becomes deeper when we consider the case of someone using ‘I’ to self-refer. Unlike non-indexical names, which are involved in social facts about naming, ‘I’ is not so involved. No one has the first-person pronoun as their name. In fact, no one could be called ‘I’ where ‘I’ is the first-person pronoun. Someone who was called ‘I’ would have a homonym of the first-person pronoun as their name. For ‘I’ to be a name, it would need to be non-indexical, and by assumption ‘I’ is indexical. Given this point, it would be very surprising if a speaker’s choice to use ‘I’ was because of a belief about what someone is called. Instead, it would likely go via knowing the meaning of ‘I’, which means knowing, if only implicitly, something like

‘I’ in a context refers to the speaker in the context.

If one chooses to use ‘I’ to self-refer, and one knows the meaning of the term, one’s choice will likely be guided by the knowledge that I am (shortly to be) the speaker. Again, ‘the speaker’ is a description and this is the ‘is’ of identity.Footnote 17 And it is an identity which employs the first-person guise. Morgan’s test tells us that a speaker’s choice of a term directly expresses a guise if the choice ‘does not exploit any identity knowledge he has concerning the thing the guise refers to’. But that is exactly what the speaker’s choice of ‘I’ would exploit in this case. So ‘I’, in such a case, does not directly express the de se guise.

Perhaps there could be someone who does in fact believe I am called ‘I’ and uses ‘I’ to refer to themselves for that reason. In this person’s usage, according to the test, ‘I’ would directly express the de se guise for the same reason as names used to self-refer. Does this show that there is no problem with the test? No, because such a case would be an example of someone who has misunderstood the meaning of the term ‘I’. In framing the explanatory argument from direct expression, we can assume that only competent use is relevant when testing for direct expression. Direct expression has to do with meaning, and competent use of a term is use that is suitably responsive to the meaning of the term. So it would be a mistake to conclude that ‘I’ directly expresses (or can directly express) the de se guise based on someone’s misunderstanding of the meaning of ‘I’.

If that’s the case, then the test of direct expression ought to be restricted in scope to choices of a term that are based on knowing the meaning of the term, not any old choices based on anything. Given what was said above about ‘I’, we might assume, which anyway seems plausible, that in fact one needs to use ‘I’ based on identity knowledge in order to use it competently. That is, someone who uses ‘I’ knowing its meaning but who also for whatever reason rejects the identity I am the speaker has failed to use ‘I’ competently. I will give an argument for why that might be true in the next section. If we do accept this assumption, then we can say, not only that someone could use ‘I’ because of identity knowledge involving the de se guise, but also that competent users of ‘I’ are those who do accept the identity I am the speaker and use ‘I’ for that reason. From which it follows that ‘I’ does not directly express the de se guise by Morgan’s test.

Conclusion: on the reasonable assumption that only competent uses are relevant for Morgan’s argument, not only is it the case that self-referrers can, by Morgan’s test, directly express the de se guise using non-indexical names, but they never directly express their de se thoughts using ‘I’, at least never when that use is competent. This is exactly the opposite of what the explanatory argument requires.

4 Identifying the speaker

Why does Morgan’s test for direct expression give the wrong result? The opposite of direct expression is, presumably, indirect expression, where expression is indirect if it is based on identity knowledge. The test is meant to generate the result that non-indexical names can only indirectly express the de se guise. That would be the case if using a name to self-refer required identifying the referent of one’s self-conception (the one that employs the de se guise) with the referent of another conception. Such a conception would be of someone impersonally conceived, employing a conception that could be expressed by anyone using a non-indexical name. The de Gaulle case shows what is wrong about this idea. De Gaulle’s self-absorption might be such that he does not have two conceptions of himself, a first-personal one and an impersonal one; it might simply never have occurred to him that there was a way of thinking about himself that is available to others. Even if he does, his use of his own name to self-refer need not go via a conception that is somehow separated off from his own self-conception. As we saw, he just knows a fact about himself, that he is called ‘de Gaulle’.

The problem with Morgan’s test, then, is that it is tied to the (faulty) idea that non-indexical names bear a special connection to impersonal conceptions. This sets Morgan’s explanatory argument off on the wrong foot. The special connection between the de se guise and an indexical term needs to limit competent uses of the term to those that express thoughts that employ the guise, and that connection doesn’t bear on how non-indexical names are used. Direct versus indirect expression is beside the point.

We should, therefore, flip Morgan’s test on its head. Instead of making the special connection turn on a (likely erroneous) claim about how non-indexical names can be used to self-refer, we can think of it as obtaining when a user of a term must accept a particular identity. Rather than the special link between a guise and a word obtaining because an identity belief doesn’t figure in the competent choice of a word, it obtains because one who uses that word must identify its referent with the referent of a given guise, and must use the word on an occasion because one accepts that identity, in order for a use of the word to be competent.

On this analysis, a word bears the relevant sort of special connection to a guise if accepting the identity of the referent of the guise with the referent of the term is a condition on competent use. In contrast to Morgan’s test, this analysis of the special connection generates the result that there is a special connection between ‘I’ and the de se guise, of the kind his explanatory argument requires. No competent use of ‘de Gaulle’ to refer depends on identity knowledge employing the de se guise, including de Gaulle’s use to self-refer, because it is never necessary to identify the referent of the name with a self-conception to use it competently. By contrast, when Churchill utters ‘I am hungover’ intending to self-refer, he needs to have chosen to use the term ‘I’ because he knows I am the speaker and ‘I’ refers to the speaker. Assuming what was said above about competent use of ‘I’ is right, then this is true in general, in virtue of the meaning of ‘I’ and the nature of the de se guise.

But why accept this assumption? Here is a general principle concerning competent uses of a word to convey information:

A speaker is competent in using a referential term t to convey information about an object only if she knows which object the audience would understand to be in question in virtue of the semantic properties of t.

The principle gets its bite from the fact that a speaker might attempt to communicate information but fail to do this in virtue of the meanings of the words she uses. We can articulate this using the idea that a speaker in a speech context is attempting to update the conversational common ground. In attempting to convey information, a speaker is putting forward a proposition that, if accepted by the audience, would be mutually accepted: each party accepts it, and also accepts that each accepts it, and accepts that each accepts that each accepts it, and so on. Were someone to utter a sentence and misidentify what the sentence in her utterance of it expresses, then she has to that extent failed to update the common ground. There is a proposition that she thinks is in the common ground that isn’t; and the audience thinks there is a different proposition that is in the common ground that isn’t. So the principle captures the fact that competence requires understanding what is expressed in using a term.

The principle as applied to ‘I’ has the result that, if a speaker does not accept the identity of the referent of her token of ‘I’ with the referent of her de se guise, then she fails to be competent in her use of ‘I’ on that occasion by the above principle, as I will now show.

Consider a case where A utters the sentence ‘I am bored’: one of the things she has expressed is that the speaker is bored. Since A is the speaker, she has also expressed the proposition that A is bored. If A rejects the identity of the referent of the token of ‘I’ she utters with the referent of her de se guise, then this can only be because either

  1. (1)

    she accepts that the referent of the de se guise is the speaker (she accepts I am the speaker), but thinks the referent of ‘I’ is determined by a different semantic rule, which she thinks picks out someone else;

or

  1. (2)

    she thinks that the referent of the token of ‘I’ is determined by the rule that picks out the speaker, but she does not accept that the speaker is identical with the referent of her de se guise; that is, she identifies the speaker of the token of ‘I’ that she utters with someone else.

(1) is a straightforward case of misunderstanding one’s own words because of not knowing their meaning. A has put forward the proposition that A is F, but she thinks that she has not put forward the proposition that A is F. She has misunderstood the meaning of the term, and as a result has failed to communicate successfully. Her audience would think she accepts a proposition that she does not accept.

(2) is trickier. For one thing, it would take a fairly contorted sequence of events for someone to utter a token of ‘I’ (or any word for that matter) and yet identify the speaker of that token with someone other than themselves thought of under their de se guise. Speaking is an intentional action, and typically agents know that they are engaging in an action in virtue of doing so (even if they do not know that their intention has been brought off, they know that they are trying). Moreover, the kind of self-knowledge of intentional action will be associated with one’s de se guise, since it is that guise that is involved in the intention and the generation of the action through the intention. Were it necessarily the case that someone who uttered ‘I’ knows that they have done so under their de se guise, then type-(2) cases would be ruled out and we could rest content with type-(1) cases. But I am not sure we can rule this out.Footnote 18 So we need to say something about why this is a case of incompetent use.

Unlike (1), the problem does not come from the fact that A has misunderstood the meaning of ‘I’. She has understood it but has not associated the speaker in the context with her de se guise. As a result, there is nonetheless a mismatch between what the world is like according to A and what her utterance of the sentence tells her audience she thinks the world is like, and that makes (2) similar to (1). Here is why:

In (2), A’s understanding of the situation is that there is someone distinct from herself who is uttering/has uttered the sentence; she utters “I am bored” but thinks that someone else is the speaker. Her audience, B, understands the sentence uttered in accordance with the meaning of the terms used and accepts the proposition expressed. As a result, B will rule out the possibility that the individual who uttered the token ‘I’ is not bored, and takes A to be ruling it out also. A will also rule this out: she knows the meanings of the words she uttered, and that is why she uttered ‘I am bored’. So far, the common ground has been updated successfully. However, B will exclude the possibility that A did not utter the token of ‘I’. B knows that A uttered the token of ‘I’, and so after accepting what A expressed, B rules out that A is not bored.Footnote 19 In contrast, A does not exclude this possibility. It is compatible with what she accepts that someone who is not identical with herself uttered ‘I’. So it is compatible with the possibilities A rules out that A is not bored. Therefore, in uttering ‘I am bored’ while rejecting the identity of herself with the speaker, A has misidentified what she has expressed.

(1)-type cases are cases of pure semantic incompetence; (2) is a case of mixed semantic-pragmatic incompetence; her knowledge of the semantics of the term combined with her ignorance about what the context is like results in a misunderstanding of her speech act. In both cases, a speaker rejecting the identity of the referent of her token of ‘I’ with the referent of her de se guise results in a lack of competence with the term, because of the failure to understand what one is conveying in virtue of the semantics of ‘I’.

5 Which, not how

The previous section proposed a natural replacement for Morgan’s analysis of the special connection between the de se guise and ‘I’ in terms of direct expression. There is an identity involving the de se guise that someone needs to accept in order to use ‘I’ competently. This section looks at the explanatory argument with this amended analysis of the special connection between ‘I’ and the de se guise in place. I will argue that the explanatory argument is a non-starter. The means through which the de se guise picks out the thinker is irrelevant to why the psychological role creates this restriction on competent use.

There were two components of the amended test as applied to ‘I’: one is that a speaker who competently uses it in a speech context must accept the identity I am the speaker, where ‘I’ stands for the speaker thought about under the de se guise. The other is that this must be part of the reason for using ‘I’ in that context. The explanatory argument Morgan sketches is that the semantic rule encoded by the de se guise would, if suitably related to the semantic rule encoded by ‘I’, explain why ‘I’ directly expresses the de se guise. On the amended test, this means the semantic rule encoded by the de se guise would need to explain why knowing the identity I am the speaker is a condition on competent use. But this is not the case.

That the thinker knows the identity and takes it as her reason for uttering ‘I’ only depends on and is supplied by her self-knowledge and her knowledge of the meaning of the term ‘I’, plus the fact that it is the speaker/thinker who is thought about and referred to. This is independent of the content of the rule that picks the speaker/thinker out in the thought expressed. The content of the semantic rule encoded by the de se guise (assuming there is such a thing, whether indexical or not) is not relevant to either of the components of competent use. It matters which object is thought about, but not the means by virtue of which it is thought about.

The factors that were driving the semantic and pragmatic considerations that resulted in incompetent use of ‘I’ in the case that the identity is rejected only had to do with the meaning of the term and the problems in conveying information arising from the speaker either misunderstanding the term or misunderstanding the speech context. The requirement that the speaker must know that the referent of a token of ‘I’ is herself, thought of under her de se guise, then follows from these points:

  1. (i)

    The semantic contribution that any token of ‘I’ in a context makes to the sentence in which it occurs is given by the rule that it refers to the speaker in that context.

  2. (ii)

    In uttering ‘I’ as part of a sentence, the speaker is engaging in an act of communication; the speaker is attempting to convey information to an audience.

  3. (iii)

    A speech act that employs a term takes place against the background of, and is intended as an indication of what the speaker proposes to add to, a conversational common ground that should, if the use of ‘I’ is to be competent, come to include the proposition that the speaker is the one who uttered the term.

  4. (iv)

    Someone who does not identify the referent of their de se guise with the referent of ‘I’ thereby must have some understanding of the speech situation (potentially including the conventional meanings of the term she is using) that means she fails to successfully indicate what is to be added in the common ground.

We can therefore explain why the special connection between ‘I’ and the de se guise obtains by adverting to how the semantic function of the term in conveying information relates to the psychological role of the de se guise, specifically with how it coordinates self-knowledge and intentional activity, of which speech activity is a particular case.

The unique expressive relationship is a result of the interaction between the psychological role of the de se guise with the expressive function of ‘I’ as determined by its semantics—not the semantic rule that governs the de se guise. That makes no contribution other than to supply which object is identified with the speaker in the speaker’s conception of the speech situation. How this is determined makes no difference. And that is exactly what the moderate de se skeptic wants, since it undermines the argument that the special connection needs Indexical Semantics to be true.

It is important to emphasize again that the moderate de se skeptic accepts that there is a special psychological role played by the de se guise, and that this psychological role is what coordinates knowledge of what one is doing in uttering a sentence with one’s knowledge of the terms in the sentence. The moderate de se skeptic resists the move to the more radical position on which there is no such psychological role. So it is perfectly coherent for the moderate de se skeptic to appeal to this role in explaining the restriction on competent use, without further explanatory inference to a special semantics for the de se guise.

My argument has centred on two rival analyses of the special connection between de se thought and the first-person pronoun ‘I’. I have argued that (i) the analysis in terms of direct expression fails; (ii) the analysis in terms of a required identity succeeds; and (iii) the required identity analysis does not support the explanatory argument sketched by Morgan. A worry one might have at this point is that these two options are not exhaustive; why not a third? An expressive relationship that might fit the bill has been discussed in recent literature under the heading of de jure coreference. This is coreference between referential items which differs from circumstantial or de facto coreference by virtue of being demanded by the semantic properties of referential items (anaphoric pronouns and their head nouns provide a paradigm example). One might think that tokens of ‘I’ produced by A are de jure coreferential with A’s de se guise; this would count as a special connection between them, and clearly one that turns on their semantic properties. It could perhaps be material for Morgan’s explanatory argument, and so worth considering.

But what exactly is de jure coreference? Recanati (2016) proposes the following as the ‘base relation’ between de jure coreferential terms M and N:

∀x (Ref(M, x) ≡ Ref(N, x))Footnote 20

In other words, the reference of the two terms is strictly the same. Taking ‘A’ as a name, gA for A’s de se guise, and IA for A’s token uses of ‘I’, this would give:

DJC: ∀x (Ref(gA, x) ≡ Ref(IA, x)),

Appeal to DJC is consonant with Morgan’s analysis of the special connection in terms of direct expression, since it makes no mention of an identity; rather, the expressive relation between ‘I’ and the de se guise is given by the biconditional.

DJC by itself cannot support the kind of explanatory argument Morgan sketches, for reasons close to those that ruled it out on the required identity analysis. It is crucial that DJC is restricted to tokens of ‘I’ produced by a given individual. since otherwise it would be plainly false (it is not even the case that ‘I’ as a type is de facto co-referential with A’s de se guise, for example). But then the analysis is neutral between the Millian rule and the indexical rule, since both pick out A wherever they pick out anyone. The semantic properties of each item, on either the orthodox or the skeptical view, mean that the two items are strictly co-referential. It bears emphasizing that, while the condition doesn’t itself impose the requirement of an identity to figure in the production or consumption of the referential items in question, it does not rule it out either.

If de jure coreference is semantically required coreference, this is not captured purely by the DJC condition, as it is not simply a matter of coreference strictly holding, but of this being guaranteed by the semantics of the two terms alone. De jure co-reference between two terms might be thought to be a relation that precisely does rule out the required identity analysis. If co-reference is semantically required, then a correct understanding of the two items would not require adducing an additional identity, since coreference is already a condition on their referring at all. One job that this notion has been given is to offer an explanation of how arguments and inferences can ‘trade on identity’, that is, involve multiple premises that refer to an object without needing an explicit identity premise in order to be valid. So if A’s tokens of ‘I’ and A’s de se guise are de jure coreferential in this sense, then it might seem that the required identity analysis must be false.

However, the proposed analysis cannot support the explanatory argument either, because it is also inconsistent with the de se guise having an indexical semantics. Suppose that the rule for the de se guise is an indexical rule. Clearly it cannot be the same rule as the rule for ‘I’; ‘I’ is a word, tokens of which are produced on an occasion and refer to an object that fulfils a role in the context of that occasion: being the speaker, as we have been putting it. The indexical rule for the de se guise would need also to supply a role, but it cannot be the same role as for ‘I’, because the guise and the word function entirely differently. It would instead be something like: being the thinker of a thought that employs this guise. The point is that the two rules cannot identify the same object in the same way, given the different functions the word and the guise play. That being so, it must be an extra-semantic fact that the referent picked out by the two rules is the same.

On the assumption that the rule for the de se guise is either indexical or Millian, then, the semantically required coreference analysis fails (even if the ‘base relation’ DJC does obtain). Discussions of de jure coreference typically take it to be a relation that either obtains between words, which can be interpreted, or between thoughts/concepts/guises, as they figure in inferences, and not (to my knowledge) between both words and thoughts/concepts/guises. One issue with the proposed analysis is that the motivation for positing de jure coreference due to trading on identity does not obviously apply, and we can live without it. On either view, reference will strictly be the same. On neither view, however, do the semantic facts alone guarantee that co-reference obtains. This is, instead, a product of the special psychological character of the de se guise and how it relates to speech as intentional action, as this section has argued.

6 Conclusion

I have argued that Morgan’s test for which terms directly express the de se guise generates a faulty analysis of the special connection that guise bears to the first-person pronoun. I proposed an alternative analysis, on which only someone who accepts that they themselves (under their de se guise) are identical with the speaker can competently use the first-person pronoun. But no appeal to a special semantics for the de se guise is needed to account for this: what determines conditions on competent use of ‘I’ and the fulfilment of those conditions is that it is the thinker/speaker which is being picked out in the thought being expressed, not the semantic mechanism that achieves this. The special connection is fully accounted for by the expressive role of the first-person pronoun and how it relates to the integration of self-knowledge and action that figures in the psychological role of the de se guise. Morgan’s explanatory argument against moderate de se skepticism can only work, therefore, if it involves a prior demonstration that an indexical semantic rule is essential for the de se guise to play its psychological role. But then the question would already have been settled.