5
Reference
What does it mean to refer to something? It seems as though this, at least, should be a simple question to answer: We see a thing in the world, we want to talk about it, so we use a phrase that will allow our hearer to know what thing it is that we’re talking about. We say the daisy in the vase, and we confidently assume our hearer will know exactly what real-world object we’re talking about because we have referred to it appropriately.
But as with, well, almost everything, it’s not nearly that simple. One of the trickiest aspects of the simple noun phrase the daisy in the vase is the use of those two instances of the definite article the: What daisy? What vase? How can my hearer pick out which daisy and vase I mean? And worse yet, how do I know whether they’ll get it right—and even worse, how can they know whether I know whether they’ll get it right? This is a rabbit hole we’ll delicately leap over for the moment, since it’s the subject of the next chapter; but even setting that matter aside doesn’t mean our task has gotten easy. As we’ll see, we can refer to things that don’t exist, we can successfully refer to things using inaccurate descriptions, and we can refer to fictional entities. If I can refer to something that doesn’t exist (e.g., The Abominable Snowman), are the things we refer to really in the world, or are they just in our minds? Just what do we refer to when we refer?
The discourse model
As two (or more) people talk with each other, they build up a mental list of what they’ve talked about. This is obvious; it accounts for the simple fact of being able to interpret she in (1), for example.
(1) 
Here, the only reason the hearer understands who she is supposed to refer to is because they remember the just-mentioned grandmother. We say that the hearer has a model of the discourse (the discourse model, introduced in Chapter 1), and when the speaker mentions a grandmother, the hearer adds the grandmother to their model. But more than that, the hearer knows that the grandmother is in the speaker’s discourse model as well. This discourse model is a model of what constitutes shared knowledge in the discourse, and linguists often talk about the discourse model as though it, too, is shared. But that’s not quite true: The speaker’s model and the hearer’s model can differ. Sometimes this difference will show up through the discovery of a miscommunication:
(2) 
At the end of Bea’s utterance, the interlocutors’ discourse models differ, in a fairly straightforward way: Ann’s model contains an entity with, among other things, the properties of being her maternal grandmother, being a well-known children’s author, and having never met Bea. Her model also has a separate entity with, among other things, the properties of being her paternal grandmother and having met Bea. Bea’s model, on the other hand, at that point has an entity with, among other things, the properties of being Ann’s grandmother, being a well-known children’s author, and having met Bea. At the end of Bea’s utterance it becomes clear to Ann that their models don’t match; Ann’s second utterance then brings the two models into alignment. Notice, though, that it’s perfectly possible to have a minor misunderstanding like this and for nobody ever to realize it. Imagine, for example, that Bea had simply replied, Wow; that’s cool, and was thinking, rather than saying, ‘I wish I’d known that when I met her; I’d have asked about it’. The two of them might have happily continued without ever discovering the error.
So Ann and Bea carry on their conversation as though they share a model, but in actuality, as briefly mentioned in Chapter 1, the two models are distinct. And Ann’s model includes her beliefs about Bea’s beliefs, just as Bea’s model contains her beliefs about Ann’s beliefs—which is to say that Ann has a mini-model of Bea’s model, and vice versa. And of course Bea’s reference to her in (2) assumes not only that Ann’s model contains the same grandmother as Bea’s model does but also that Ann believes that Bea’s model contains this same grandmother (that is, Bea’s reference to her requires that Bea’s model of Ann’s model of Bea’s model has this same grandmother), and Ann needs to realize that Bea’s reference requires this (so Bea’s reference requires that Ann’s model of Bea’s model of Ann’s model of Bea’s model has this)…well, you see that in a sense we require an infinite number of beliefs about each other’s beliefs before anyone can utter anything. This is the Mutual Knowledge Paradox (Clark and Marshall 1981): How can we ever say anything at all? Mutual comprehension would seem to be impossible.1
But of course we do say things, and we do (mostly) understand each other. Clark and Marshall propose that we have various copresence heuristics that help us bridge the chasm: For example, if the speaker and hearer were physically copresent at the time an event occurred, they can both assume that that event constitutes shared knowledge, and similarly for cultural copresence (being members of the same culture, which entitles them to assume that they share the sorts of knowledge that all members of that culture in general share). The crucial thing to keep in mind, though, is the fact that our assumptions about what someone else knows are just that—assumptions. And those assumptions can be wrong. This is the crux of miscommunication. I think it’s fair to say that most miscommunication is due not to malice on anybody’s part, but rather to miscalculations about the extent to which our discourse models are alike, and the extent to which our interlocutor’s understanding of our utterance matches our own intent. And these two things are closely related: To the extent that a hearer’s understanding fails to match a speaker’s intent, the resulting discourse models will fail to match as well.
The reason this is all so important gets back to Reddy and the Conduit Metaphor, from Chapter 1: We tend to assume that we simply convey our meaning to our listener, and that they therefore have ‘understood’ us. We assume that communication is effortless. But the entire insight of pragmatics is that it isn’t effortless at all, and that a great deal of inference is involved in comprehending an utterance. And inference is by its nature an imperfect art. We’ve seen this with respect to implicature, and we’ve seen it with respect to speech acts. And now we will see it specifically with respect to reference.
Referents
So let’s say I have referred to my grandmother in a sentence like I wish I were as good a writer as my grandmother. The first and apparently simplest question is, what have I referred to? Your likely response would be that I’ve referred to an individual in the world. And suppose I then point out that I have no currently living grandmother. At that point you might say, okay, I’ve referred to a past individual. Or you might, after a moment’s thought (and having read the previous section), decide that what I’ve referred to is an entity in my discourse model—what we call (not surprisingly) a discourse entity or (also not surprisingly) a discourse referent. These two perspectives, in fact, represent two opposing sides in the question of what I refer to when I refer: Have I referred to a thing in the world, or have I referred to a thing in my discourse model?
Let’s suppose that when I use a referring expression, I refer to something in the real world. This is known as the referentialist approach, and it’s certainly the most intuitively appealing point of view. After all, why do we refer at all if not to refer to something in the world? And it makes sense that if I say That tree is tall, I’m attributing tallness to a specific actual tree in the actual world. And it might also seem intuitive that if there’s no tree at all, I haven’t referred successfully. But there are problems with this approach. For one thing, we refer to nonexistent things all the time. If I tell you, I think Frodo should have just left the ring alone altogether and you take the bait and disagree with me, we can have a rousing argument about all sorts of fictional hobbits and what they should or shouldn’t do with an equally fictional ring; we have no problem at all referring to nonexistent hobbits. But you might reasonably complain that I’m splitting hairs here—that we may not have actual hobbits, but we certainly have hobbits in a story we both know. But those stories are just ink on a page, or lights on a screen; it’s our brains that turn those things into a story, and the only place that story actually resides is in our minds.
All right, though; let’s set aside fictional worlds for the moment, and stick to cases in which we intend to refer to an entity in the actual world. There are several sorts of cases in which we could refer to something that is, or may be, nonexistent. The first type is when the speaker and hearer disagree about whether the entity exists. Suppose I tell you my house is haunted, and that the ghost makes loud noises at night. And suppose you don’t believe in ghosts. I’ve referred to an entity that exists in my discourse model and not yours—but whether it exists in reality (and therefore whether I’ve referred to anything at all, on a referentialist view) is a matter of disagreement. Nonetheless, we have no trouble referring to it.
Another case is when the speaker and hearer both believe that some entity exists, but disagree on what its attributes are. So consider a conversation like the one in (3):
(3) 
Eyesight issues aside, it’s perfectly possible for two people to disagree on an entity’s attributes, and when we make reference by means of those attributes (the young man playing the horn in the subway this morning) it raises the question of whether the referent is whatever in the world I intended to refer to, or whether it’s whatever in the world the semantics of my utterance actually pick out. Suppose Bea is right about the attributes of the person she and Ann mutually encountered in the subway, so Ann’s reference is all wrong—and now suppose that a couple of hours earlier, a young man actually was playing the horn in the subway (and neither of them saw him or knows about him). Does it make any sense to say that he is the actual referent of Ann’s noun phrase the young man playing the horn in the subway this morning? He’s certainly the entity that the utterance corresponds to most closely in terms of its semantics, but he’s not at all the entity she intended to refer to. In this case it does seem as though she’s referred to the woman playing an oboe, despite the semantics of her utterance, because that’s who she intended to refer to.
Can we reasonably say, then, that my reference is fixed entirely by my intent? It seems as though there must be some limit to the role of intention in reference. I can’t, for example, refer to a truck driver as a ballerina, or to a senator as a grilled cheese sandwich. Except, of course, that I actually can:
(4) 
(5) 
We do this sort of thing all the time. (See Nunberg 1995 and Ward 2004.) So although, all things being equal, I can’t generally refer to my grandmother as a strawberry shortcake, all things are seldom equal, and I bet you’re already thinking of a context in which I could make that very reference.
Kripke (1977) makes a useful distinction between semantic reference and speaker reference; that is, what the semantic meaning of a phrase picks out in the world (the semantic reference) may be different from what a speaker intends to pick out in the world (the speaker reference). Semantically, the phrase my grandmother refers to the mother of one of the speaker’s parents, but pragmatically, the speaker can use it to refer to other entities (e.g., a beloved older friend). But the question here goes deeper: We can refer to things that we believe to exist but don’t (as when someone calls the police to report on an intruder based on the sound of a branch scraping the house) as well as things that are fictional (Harry Potter) and even combinations of the two, e.g., things that a speaker believes exist in fiction but don’t (the ring that Harry Potter carried to Mordor).
So one problem with the referentialist view is that we can refer to things that don’t actually exist in the world, and another is that the semantics of our utterance don’t have to accurately reflect the referent. Another problem is more philosophical, and involves the well-known thought experiment of Theseus’ Ship. Suppose Theseus has a ship. And over the years, individual boards and other components of the ship suffer wear and tear and need to be replaced. This goes on long enough that eventually every single board (etc.) has been replaced. Is it still Theseus’ ship? That is, is it the same ship, despite not having a single molecule in common with the original ship? Worse, suppose another person has systematically collected all of these discarded parts, and uses them to build a ship that is identical in every way to the original ship. Which of the two is Theseus’ ship? I don’t mean in terms of ownership; we know which is the ship that Theseus owns. But which ship is the ship we started with? And if it’s not the one that Theseus has continued to use and maintain, at what point did it change?
We see that there’s a serious question of identity here. And there’s a serious question of what it means to know something’s identity. Suppose that both Ann and Bea are mistaken about the oboe player: In reality it is a middle-aged woman playing an oboe in the afternoon, but they both mistakenly believe it to be a young man playing a horn in the morning. So Ann makes her remark in (3)—The young man playing the horn in the subway this morning was great—and Bea responds He sure was, and their lives go on and they never mention it again. Presumably this was a successful reference despite the semantic difference between the description uttered and the actual reality, and the fact that nobody would ever know the difference.
So maybe reference has less to do with reality than with our mental construction of reality. This is the view taken by mentalists (or cognitivists)—i.e., that a referent isn’t actually an object in the world but rather an object in our mental model of the world (and therefore also, potentially, in our mental model of a discourse). This is the viewpoint taken by famed linguist Noam Chomsky, who argues that the defining feature of a referent is psychic continuity—i.e., the degree to which we think of object A as being the same as object B from the past. So even though every water molecule currently flowing through the Mississippi River may be distinct from those that flowed through it a year ago, those two sets of water molecules still count as the same river because of their continuity as a single entity in the psyche of those of us who know about and think about and talk about the Mississippi River.
Needless to say, this view isn’t free of difficulties either. Consider the beech tree, in an example that comes to us from philosopher Hilary Putnam. Putnam noticed that he didn’t know the difference between a beech and an elm: He knew that they were both trees, but he didn’t know of a single property by which he could distinguish the two. Therefore, they had identical representations in his mental model. For Putnam, this meant that the meaning of the term beech isn’t a mental concept, but rather a set of real-world entities. The words beech and elm don’t pick out two different mental constructs (for Putnam, at least), but they do pick out two different sets of entities in the world (each set being called the extension of the term).
Another argument against the mentalist view is that speakers typically don’t think of themselves as referring to mental constructs, and when they attribute properties to referents, they mean to attribute those properties to real-world (or fictional-world, or hypothetical-world) objects, not to objects in their mental model. To make this more concrete, when Ann in (3) says The young man playing the horn in the subway this morning was great, she means to praise the musical ability of an actual real-world person; she certainly doesn’t mean to say that a construct in her head is musically talented.
There’s no simple answer to this dilemma, but for the purposes of a book on pragmatics, it makes sense to emphasize the importance of the discourse model, because pragmatics is all about intention and belief, and the discourse model is where our intentions and beliefs about the discourse are constructed and stored. So while Ann intends to talk about a real-world young man playing the horn in the subway, it’s important to realize that even if there isn’t an actual real-world young man who was playing the horn in the subway, that’s not a problem for the reference; all that’s needed for the reference to go through is for her discourse model to contain such a person, and for Bea’s discourse model to contain an entity whose attributes are sufficiently close to Ann’s description to allow Bea to decide that the two count as the same thing—again, regardless of the state of reality. For the purposes of negotiating our lives in the real world, it certainly helps if our mental models correspond fairly closely to reality, but it’s not at all necessary for successful communication.
Sense and reference
In the previous section, I talked rather loosely about the difference between a linguistic description and the thing being described. Even if we locate that thing within the discourse model rather than within the world, there remains a difference between the description and the referent. To see what I mean, consider Frege’s famous distinction between the morning star and the evening star. Semantically, the morning star would seem to be the star that we see in the morning, whereas the evening star is the star that we see in the evening. We can make this judgment based on the sense of each of these phrases—that is, what they mean semantically. And in terms of their semantic sense, it would seem odd to say (6):
(6) 
But from the perspective of reference, (6) is a perfectly reasonable thing to say, because the phrase the morning star and the phrase the evening star both refer to the same thing—the planet Venus. This points up the difference between the sense of a phrase and its reference, between what the phrase means linguistically and what a speaker uses that phrase to pick out in the world (or in the discourse model, depending on your view).
The sense of a phrase is a matter of semantics, while its reference brings in pragmatics; thus, we can say that sense is a matter of convention, whereas reference is a matter of both convention and intention. It’s not a matter of intention alone, because we’re not free to use absolutely any phrase to refer to absolutely anything, unless we’ve prearranged it with our hearer (creating a nonce convention between the two of us) or the context makes it clear (as with the reference to a truck driver as the ballerina in (4)).
Because sense is a matter of semantics, it’s very slow to change (although obviously the meanings of words do change over the course of history): The word oboist will mean the same thing next week that it means today. Reference, on the other hand, can change from day to day, hour to hour, or moment to moment: I can say the oboist this morning in reference to someone playing the oboe in the subway, and again this evening in reference to someone playing the oboe in a symphony. The phrase has remained the same, and its sense has remained the same; in either case, the oboist means ‘the person playing (or who can or does play) the oboe’. But which person playing the oboe is being referred to has changed from morning to evening. As we’ll see in the next chapter, the contribution of the word the to determining the referent of the phrase is a complicated matter, but it’s a safe bet that its meaning hasn’t changed much from morning to evening, and neither has that of the word oboist. What’s changed is what those meanings combine to pick out in the context. So what the oboist ‘means’ in terms of its semantics and what it ‘means’ in terms of its reference are two different matters. This means that the meaning of the word meaning is ambiguous between, at the very least, sense and reference (and notice that the second word of this sentence uses the word means in yet another sense). Grasping the meaning of meaning is no mean feat.
Indexicals and deixis
As we’ve just seen, it’s quite possible for the referent of a linguistic expression to change from one context to another. But there are some expressions whose interpretation is necessarily dependent on the context; we could in fact say that the context is an important component of their sense. Consider the word you. Its sense is something like ‘the person or persons to whom the speaker is speaking’. The definition incorporates the contextual matter of who’s being addressed. The sense of a word like cat requires no such mention of the context of utterance. The adjective current as in current affairs is likewise indexical, in that its meaning is something like ‘going on at the time of utterance’—so it is ‘indexed’ to the context of utterance in the same way that the word you is. Tense is indexical as well: The past tense marker means essentially ‘prior to the time of this utterance’.
We’ll talk about anaphoric pronouns like she and it in detail in the next chapter, but they too are indexical in that their meaning makes reference to the context: The referent of she can be either a person previously mentioned or evoked in the discourse or a person being verbally ‘pointed to’ in the physical context. Compare the examples in (7):
(7) 
In (7a), the referent of she is set by the earlier phrase Emily Dickinson. This is a case of anaphora, or reference back to a previously mentioned entity, and the pronoun is said to be anaphoric to the earlier mention of the referent. In (7b), there is no such prior mention of the referent; instead, the linguistic expression picks out some referent in the physical context of the utterance. This is a case of deixis, or verbal ‘pointing’ to something in the context. Both are indexical, in that they both require reference to the context for their interpretation, but anaphora requires reference to the verbal context—specifically, reference to some other phrase that tells the hearer what the referent is—whereas deixis points directly to the intended referent, with no need to pick up the reference through a prior mention.2
Because deixis requires reference to the context for its interpretation, without that context the meaning of the utterance cannot be determined. Imagine you find someone’s cell phone on a coffee-shop table, with this text message visible on the screen:
(8) 
If you don’t know whose phone it is or when the text was sent or by whom, there is no way to know who is asking to stop where, or whom they’re inviting to come, or where ‘here’ is, or which day constitutes ‘tomorrow’. This is the nature of deixis; without the context, the referential meaning is lost.
There are at least three and possibly four types of deixis:
Personal deixis makes reference to a person; examples are I, you, and the other pronouns. I and you in most uses are inherently deictic, since they are used in reference to the speaker or addressee (except for nonreferential cases like you can’t win ’em all), but the other pronouns—he, she, it, they, etc.—have both deictic and anaphoric uses, as described earlier.
Spatial deixis makes reference to space, and includes words like here, there, come, go, arrive, leave, this, and that. In English, spatial deictics are generally organized into pairs of proximal and distal terms, where the proximal member of each pair indicates closeness or proximity to the speaker and the distal member indicates distance from the speaker. Here (proximal) is wherever the speaker is, while there (distal) can be anywhere the speaker isn’t. This (proximal) can indicate any salient object that’s relatively close to the speaker, but it definitely should be closer to the speaker than something being referred to as that (distal). The word come is proximal, indicating increasing proximity to the speaker, while the word go is distal, indicating increasing distance from the speaker. Interestingly, for some of these words, the point of reference needn’t always be the speaker; we call this point of reference the deictic center, and it can vary. So consider (9):
(9) 
Here, the deictic center for the second clause shifts to the addressee; will Stacy come along? doesn’t mean ‘will Stacy move toward the speaker?’ but rather ‘will Stacy move with the addressee?’. Judgments vary on deictic-center shifts; for some people (10) is perfectly fine (with Carl as the deictic center), but for others the use of bring with a deictic center other than the speaker is odd:
(10) 
Temporal deixis makes reference to a time; examples are words and phrases like today, tomorrow, last week, a week from tomorrow, now, then, and so on. Here again we encounter the proximal/distal distinction, with now being proximal and then being distal. The word then has the interesting property that it is generally anaphoric; some prior mention needs to supply the time in question (I’m free tomorrow morning, so I’ll call you then). As we’ve seen, this isn’t true for deictics, because the ‘pointing’ function of deixis makes the reference clear: In the case of personal deictics, I and you need no further explanation; they’re the speaker and the addressee, respectively. In the case of spatial deictics, words like come and go can suffice to indicate motion toward or away from the speaker, while words like this and that can get their reference from contextually available information (salient objects, perhaps helped along with a gesture to indicate which one is meant). But gesturing toward a point in time isn’t possible (unless you’re looking at, say, a timeline in a book). And while now takes its reference directly from the time of utterance (or some other salient ‘now’, as in Jim knew that now he would have to apologize), there are innumerable possible times that can be referred to as then, and there’s no obvious way to ‘gesture’ to the intended time, so some prior mention or previously salient time must be available to provide the reference. Thus, now is generally deictic, but then is generally anaphoric.
Finally, discourse deixis is, as I have hinted, not always included in the list. Consider (11):
(11) 
Here, the question is whether the word that is deictic or anaphoric in the sentence That was a lie you just told. And here the intuitions are subtle. Since I defined deixis as essentially a case of verbal ‘pointing’ to something in the context, if we take ‘context’ to mean ‘nonverbal context’, then obviously (11) is out on the grounds that the word that points to something in the verbal context. But clearly the prior discourse is part of the context, and distinguishing in this way between verbal and nonverbal context may seem arbitrary. Instead, one can argue that there’s a more reasonable distinction to be made between a pronoun that is coreferential with a prior linguistic element and one that the speaker uses to refer to that prior linguistic element itself. Suppose I utter (7a), repeated here as (12):
(12) 
Here, in using the word she, the speaker doesn’t refer to the name Emily Dickinson; instead, both the name Emily Dickinson and the pronoun she are used to refer to the same discourse entity. Since the name Emily Dickinson provides sufficient information for the hearer to identify the referent, whereas the word she on its own does not, interpretation of she is essentially a two-step process: First the hearer must find the antecedent of she, i.e., the name Emily Dickinson; then they must identify the referent of that name. So while Emily Dickinson is the antecedent of she, and the two are coreferential, it is not the case that the name Emily Dickinson is the referent of she.
Now compare that situation with the one in (11). Here, the word that in the sentence That was a lie you just told isn’t coreferential with the earlier utterance We have a good set of fingerprints; the two do not jointly refer to some third entity. Instead, We have a good set of fingerprints is itself the referent of the pronoun that. So there seems to be a clear distinction between an earlier utterance being coreferential with a later pronoun (as in 12) and the earlier utterance being itself the referent of the later pronoun (as in 11)—which seems to argue for that in (11) being deictic rather than anaphoric.
As so often happens, it’s easy for us to get tangled up in the fact that the language we’re using in our analysis is also the language that we’re analyzing—and in the same way, the language we use to talk about the world is also a part of the world we’re talking about. The reason that discourse deixis is sometimes puzzling is that pragmatics is all about context, and the language that enacts the pragmatics is itself a part of the context.