1
Introduction
If you’ve given it any serious thought, you know that there can be a big difference between what you say and what you actually mean by it. To take a simple example, people rarely state requests forthrightly; instead, they hedge their requests in a variety of ways:
(1) 
Instead of stating the request outright, as in (1a), speakers will frequently ask about the hearer’s ability to fulfill the request (1b) or how they’d feel about doing so (1c), or they’ll comment on their own need for the request to be fulfilled (1d) or how they’d feel if it were (1e). But a moment’s thought will show that only (1a) is literally a request for the book. It’s a curious situation: We’ve developed a wide range of ways to make someone understand us as having asked them for a book when we haven’t literally done so at all. What on earth is going on?
What’s going on, simply stated, is pragmatics. Pragmatics is the field of linguistics that studies meaning in context——specifically, how a hearer understands another person’s intended meaning based on what they’ve said and the context in which they’ve said it, and how speakers craft their utterances with that in mind. In short, pragmatics is the study of the relationship between what is said and what is meant, and between what is meant and what is understood.
As Reddy (1979) observes, the English language has a pervasive metaphor in which we simply ‘put our meaning into words’, then ‘convey it’ to the hearer, who ‘gets it’ (or doesn’t get it, or maybe it ‘goes right past them’ or goes ‘over their head’). Reddy calls this the Conduit Metaphor. But, he points out, the metaphor is misleading: My meaning is never conveyed out of my head and into yours; instead, communication is a complicated and collaborative process by which a speaker encodes meaning into a series of sounds (or signs, in signed languages, or written symbols), which in turn serve essentially as instructions to the hearer for building a corresponding set of ideas in their own mind. And because the hearer’s mental world is inevitably different from the speaker’s mental world, the meaning that gets constructed will inevitably differ slightly as well. (Imagine I tell you I have a cat. The cat you imagine will differ in innumerable ways from the cat being imagined by any other person reading this book, and from the cat that I have in mind.)
As we have noted, pragmatics is a field of linguistics. Linguistics, in turn, is the scientific study of language. The ‘scientific’ part is important: Because it’s scientific, linguistics is descriptive—which means that linguists try to describe the rules that govern our language use. You may be familiar with rules like ‘don’t use a double negative’ and ‘don’t end a sentence with a preposition’, but those rules are prescriptive, not descriptive; they prescribe what someone thinks you should do. A descriptive rule describes what you in fact do. A descriptive rule of English might say, for example, that a determiner like the goes in front of a noun like cat to result in a phrase like the cat. No English speaker would ever say *cat the. (The asterisk indicates that it’s ungrammatical in the descriptive sense that speakers don’t do it—not in the prescriptive sense that teachers tell you not to do it.) Linguists describe the workings of language and its parts in much the same way that botanists describe the workings of plants and flowers, and geologists describe the workings of minerals and tectonic plates.
Linguistics has various subfields: syntax, for example, is the study of sentence structure and explains facts such as why you can’t say *cat the in English. Semantics, which we’ll talk about briefly in this chapter, is the study of literal, conventional meaning—for example, the meaning of the word cat. It’s the aspect of meaning that speakers of a language share, more or less; for example, although we’ll all picture a slightly different cat, English speakers in general agree on what the word cat means. That meaning is conventional; we share it by tacitly agreed-upon convention. Pragmatics, on the other hand, covers the vast amount of meaning that goes above and beyond semantics. It’s what gets us from Would you mind giving me that book? to the interpretation that the speaker is requesting the book. It’s what takes us from what the speaker has said to what we think they actually meant by saying it right here, right now, in this shared context between these people. It’s the difference between convention and intention.
For starters: Some basic terminology
One concept that is crucial to the linguistic study of meaning is truth. To state the obvious, something is true if it accurately describes the world. Is ‘all dogs have tails’ true? Well, check the world. If there’s a dog without a tail, then it’s not true that all dogs have tails. But that means ‘truth’ is relative to a world. And while we live in a pretty great world, there are certainly other ways the world could have been. It would have been possible to have a world in which all dogs have tails; a world that’s exactly like ours except that all dogs have tails is a possible world. Another possible world is one in which dogs normally have two heads. In fact, just about any fact about this world could be changed, and as long as it doesn’t create a logical inconsistency (say, a world in which I’m both a linguist and not a linguist), that’s a possible world. There are an infinite number of possible worlds, and we happen to live in one of them. (The world we live in is not only possible, but actual. Lucky us!) So a statement like ‘all dogs have tails’ will be true in some possible worlds but not in others. Something that can be either true or false in a given world is called a proposition. It’s not the same as a sentence, since the two sentences in (2) express the same proposition.
(2) 
How do you know they express the same proposition? Because they’re both true in the exact same set of worlds. (This holds generally, but not entirely; sentences that are true in all possible worlds, like All blue dogs are blue and All cats are feline, can’t reasonably be said to express the same proposition, and similarly for sentences that are false in all possible worlds.) So to the extent that the semantic meaning of a sentence corresponds to the proposition it expresses, it’s reasonable to say that the meaning of a sentence is a function from possible worlds to truth-values—which is a fancy way of saying that the meaning of a sentence is what tells us whether it’s true or false in some world. That sounds a bit abstract, but it makes sense: Presumably, the one property that is shared by all and only the worlds in which All dogs have tails is true is that, well, all of the dogs in them have tails.
Whether or not a proposition is true in a given world is called its truth-value in that world. The conditions under which a proposition counts as true in a given world are its truth-conditions. (So for ‘all dogs have tails’ to be true in a given world, all the dogs in that world must have tails.) And a truth-conditional theory of semantics is one in which semantic meaning is defined as any aspect of a sentence that affects its truth-conditions. The semantic meaning of the word ‘taller’ in (2a) obviously affects its truth-conditions because if you replace it with ‘thinner’ it will be true in a different set of worlds. In such a theory, I need that book is true if and only if I need that book, regardless of whether I’m saying it as a request. So my need for the book is part of the semantic meaning of the sentence, whereas the fact that I’m using it as a request isn’t, since that doesn’t affect its truth-conditions. If your head is spinning a bit, don’t worry; we’ll return to this issue later (and often).
As two or more people converse, each of them builds up a mental inventory of what things have been talked about, what other things all of the participants in the conversation are assumed to also know about (like, say, the existence of the moon), and what properties have been assigned to those objects. This mental inventory, or model, is called a discourse model. So once I’ve told you Kristy is taller than Jamal, we can assume that our shared model of the discourse contains Kristy, Jamal, and the fact that Kristy is the taller of the two. This is an idealization, of course; we’ve already seen that there’s not really a shared model. You have your model of our discourse, which includes what you believe about my model, and I have my model of it, including what I believe about your model. And yes, what you believe about my model inevitably includes what you believe I believe about your model, and also what you believe I believe about what you believe I believe about your model, and so on, in a theoretically infinite spiral. Nonetheless, we manage to leap nimbly over that infinitely deep chasm and communicate reasonably well, just as though our discourse model were in fact shared. So as long as we keep in mind that it’s an idealization, the shared discourse model is a handy way to think about the building-up of a discourse. In reality, our discourse models are distinct, but usually—hopefully—similar enough to enable the conversation to proceed.
Semantics
If by ‘semantics’ we mean any aspect of meaning that contributes to the truth-conditions of a sentence, then the meaning of a word like cat is those aspects of its meaning that could affect whether a sentence such as (3) is true in a given world:
(3) 
If Sammy is canine rather than feline, (3) is not true; therefore, ‘feline’ is an aspect of the semantics of cat. However, having a tail isn’t a necessary aspect of being a cat, since there are cats without tails; therefore, ‘tail’ is not an aspect of the semantics of cat. In this sense, we could say that the meaning of cat is precisely that set of properties required for something to be a cat.
The problem is that, while this works fairly straightforwardly for a word like cat, there are plenty of words for which it doesn’t work nearly as well. There has been a surprising amount of argumentation lately over what does and doesn’t constitute a sandwich—which is to say, over what the word sandwich means. A Massachusetts judge in 2006 had to rule on the question of whether a burrito counted as a sandwich, because a stipulation in the lease of a Panera café stated that no other sandwich shop could open in the same strip mall; the question at issue was whether a Qdoba outlet sold sandwiches, in the form of burritos. For the record, the judge ruled that a burrito is not a sandwich—but then what about a hot dog? Or a hamburger? Or a gyro? The Atlantic ran an article that purported to decide the question once and for all by offering four criteria necessary for sandwich-hood: A sandwich must have (a) two exterior pieces which are (b) carbohydrate-based, and the whole object must have a (c) primarily horizontal orientation and be (d) portable. But if you present these criteria to a group of English speakers, you’ll immediately get push-back; for example, while criterion (c) excludes hot dogs (which pleases some people but displeases others), it also excludes Italian beef sandwiches (displeasing most).
A theory that looks for a clear set of criterial features by which a word is defined falters in many cases because there simply isn’t always a clear set of such features. Instead, Prototype Theory (Rosch 1973, 1975) argues that many categories are defined in terms of a prototype, the central member of a gradient set with unclear or ‘fuzzy’ boundaries (cf. Zadeh 1965). There is a prototypical member of the set (so, for a sandwich, the prototype might be two slices of bread with meat or cheese in between), but there are also objects whose membership in the set is less clear. The uncomfortable result of adopting Prototype Theory within a truth-conditional semantics, of course, is that it raises the question of what happens to the truth-conditions of a sentence like (4) in a context in which Brianna ate a hot dog:
(4) 
If Brianna ate a hot dog, then the extent to which (4) is true presumably corresponds precisely to the extent to which a hot dog is a sandwich (which in turn raises the tricky problem of how to deal with gradations of truth).
The study of word meaning is lexical semantics, whereas the study of sentence meaning is sentential semantics. Much of truth-conditional sentential semantics has its basis in formal logic. For example, the meaning of a simple word like and is taken to be a function from the truth of two propositions to the truth of a complex proposition combining them with the word and. Yes, that sounds unnecessarily complicated. But it’s easy to see with an example:
(5) 
Sentence (5a) consists of (5b) and (5c) conjoined by the word and. And (5a) is true in a given world precisely when (5b) and (5c) are both true in that world. If either (5b) or (5c)—or both—was false, then (5a) would be false. So the truth-values of (5b) and (5c) are the inputs into a function that and performs, and that function returns another truth-value. If both of the inputs are true, the function returns ‘true’. If either or both of the inputs are false, the function returns ‘false’.
Formal logic defines a set of logical operators that serve as ‘functions’ in this way, and they correspond roughly to certain words or expressions in English. They’re given here with the symbols typically used for them:
You may feel a bit uneasy about these descriptions, on the grounds that they don’t always match the way we use the corresponding terms in English. For example, p˅q (‘p or q’) is true when p and q are both true (what’s called inclusive or), but in natural language we often use or as though it’s false if both of the component propositions are true (what’s called exclusive or). For example, consider (6):
(6) 
Here, the speaker is making a threat: If you’re not in this play, you’ll get a clout and I’ll speak to The Parents—but presumably if you ARE in this play, those things won’t happen. That is, either ‘you’ll be in this play’ or ‘you’ll get a clout etc.’ but not both. Likewise, when I say I’ll buy bread today or tomorrow, I usually mean that I’ll buy bread either today or tomorrow, but not both. This ‘not both’ reading is the exclusive reading. As we’ll see in the next chapter, pragmatics offers a way of getting from the inclusive-or reading to the exclusive-or reading without having to abandon the inclusive, formal-logic interpretation of or. Other discrepancies of this sort have arisen with respect to the other logical operators, and again pragmatics will offer a way of understanding the difference between their formal-logic interpretation and the natural-language use of the corresponding words.
Our discussion of formal logic and truth tables so far has focused on propositional logic, which is to say, logical relationships among propositions. Our symbols p and q represent full propositions, regardless of what those propositions are. Another piece of semantic machinery that will be important for our discussion of pragmatics is predicate logic, which offers a way of representing the internal structure of a proposition.
A proposition is made up of a predicate and one or more arguments. The arguments represent entities, and the predicate represents properties, actions, or attributes of those entities:
(7) 
You can use the tools of propositional logic and predicate logic together:
(8) 
The last bit of formal machinery you need to know about is quantifiers. The two we’ll worry about are the existential quantifier (∃), meaning essentially ‘at least one’ and usually read as ‘there exists…such that’, and the universal quantifier (∀), meaning essentially ‘all’ and usually read as ‘for all…’ Quantifiers are used with variables. Unlike a constant argument representing a specific entity (like those representing Brianna, Sam, etc., in (7)), what a variable represents can vary. We typically use x, y, and z as variables. To see how all this works, consider the examples in (9):
(9) 
The reason for all this notational machinery is to give us an unambiguous metalanguage for talking about language. With a clear way of representing semantic meanings, we can compare them with pragmatic meanings, and we can also use them as a reference point for talking about which meanings in language are semantic and which are pragmatic.
Since this system has its roots in formal logic, it will come as no surprise that it can be used to represent logical arguments and conclusions. For example, if (10a) is true, then (10b) is necessarily also true:
(10) 
This is a relationship of entailment: (10a) entails (10b), which technically means that any world in which (10a) is true is also a world in which (10b) is true. Stated in plain English: If it’s true that Glenda is tall and Wilma is a doctor, then it’s necessarily true that Glenda is tall.
Conclusion
There’s a lot more that could be said about semantics, but this is just enough to give us the necessary background for our study of pragmatics. In Chapter 3, we’ll see how H. P. Grice’s seminal Cooperative Principle arose out of the discrepancies we have mentioned between the semantic treatment of the logical operators and the way they’re used in natural language. This principle will give us a way of seeing how a hearer infers the speaker’s intended meaning based on what has been literally (i.e., semantically) stated and the context in which it has been uttered. We’ll also see how later researchers have attempted to improve on or streamline the Cooperative Principle, but all of these efforts are aimed at answering this same question of how we decide what a speaker is likely to have meant by what they said. With these tools in hand, we will go on in later chapters to examine such phenomena as indirect speech acts (e.g., requests that aren’t phrased as requests), definiteness, word order, and others, all with an eye toward solving the puzzle of how speakers and hearers are able to understand each other when so much of what we mean is left literally unsaid.